Review of Operational Practices and Implications for Bangladesh / A Improving Lead Time for Tropical Cyclone Forecasting Review of Operational Practices and Implications for Bangladesh Cover photo: Typhoon Nargis over the Bay of Bengal. Credit: NASA Improving Lead Time for Tropical Cyclone Forecasting Review of Operational Practices and Implications for Bangladesh ©2018 The World Bank The International Bank for Reconstruction and Development 1818 H Street, NW Washington, DC 20433, USA Disclaimer This report is a product of the staff of the World Bank with external contributions. The findings, interpre- tations, and conclusions expressed in this volume do not necessarily reflect the views of the World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. Questions regarding figures used in this report should be directed to persons indicated in the source. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. iii Preface T his report was published in June 2015. In republishing cyclones.” Since this development is highly germane, the the report, an in-depth follow-up study to determine in above statement is no longer correct. The conclusions have detail what may have changed since then has not been been revised accordingly, as follows: “Starting this year, done. However, to the best of the team’s understanding, NHC has the option to issue advisories, track and intensity the descriptions of both the science of forecasting and the forecasts, watches, and warnings for disturbances that are operational practices at the institutions involved remain not yet a tropical cyclone, but which pose the threat of bringing largely accurate, with one substantive exception. tropical storm or hurricane conditions to land areas within 48 hours.” Details are available at https://noaanhc.wordpress. In the original conclusions section, the possibility of com/2017/06/29/potential-tropical-cyclones-fitting-the- cyclone track and intensity forecasts being produced bill-for-more-timely-warnings/. While these products are before cyclogenesis has occurred was discussed. The 2015 issued only in the limited circumstances described—for lead report stated that while such forecasts had been produced times less than 48 hours, not for the 10-15 day lead times by numerical weather prediction centers and a university group, “there is no evidence of a national center that has described in the report—this is clearly a substantive change, actual responsibility for issuing forecasts for any country and indicates that the state of the art as practiced by national actually issuing official forecasts of this type yet, although the centers is evolving. National Hurricane Center (NHC) has begun experimenting with such products internally.” In 2017 the NHC did, in fact, Beyond this, only minor facts that have changed, such as begin issuing such products to the public. These forecast titles or affiliations of individuals mentioned in the report products are issued for what the NHC calls “potential tropical (including some of those listed in the acknowledgments). Dhaka. Photo by Meinzahn/Thinkstock.com v Contents Acknowledgments......................................................................................................................................................... viii Acronyms ................................................................................................................................................................... ix Executive Summary...................................................................................................................................................... xi Objectives............................................................................................................................................................. xiii Approach............................................................................................................................................................... xiii Key findings.......................................................................................................................................................... xiv Forecasting at the BMD....................................................................................................................................... xv Recommendations................................................................................................................................................ xvi Chapter 1 Introduction: Tropical Cyclones in the Bay of Bengal ............................................................... 1 Objectives ................................................................................................................................................................... 2 Approach and methodology .......................................................................................................................................... 3 Organization of the report ............................................................................................................................................ 3 Chapter 2 Forecasting: Analysis of Operational Practices ............................................................................. 5 Tropical cyclone forecast............................................................................................................................................... 5 What is a tropical cyclone forecast?.............................................................................................................................. 6 Forecasting existing storms........................................................................................................................................... 7 Forecasting genesis........................................................................................................................................................ 10 What types of observations are used?........................................................................................................................... 11 What types of models are used?.................................................................................................................................... 12 The forecast process....................................................................................................................................................... 15 Currently available skills and recent developments..................................................................................................... 20 Storm surge forecasting................................................................................................................................................. 23 What types of observations are used?........................................................................................................................... 23 What types of models are used?.................................................................................................................................... 24 The forecast process....................................................................................................................................................... 25 Treatment of uncertainty............................................................................................................................................... 26 The CIFDP Approach...................................................................................................................................................... 27 Big picture ................................................................................................................................................................... 28 vi Chapter 3 Tropical Cyclone and Storm Surge Forecasting in Bangladesh ............................................ 33 Characteristics of Tropical Cyclones in the Bay of Bengal............................................................................................ 34 Regional context for tropical cyclone forecasting in Bangladesh................................................................................ 37 The Bangladesh Meteorological Department................................................................................................................ 37 Data available at BMD.................................................................................................................................................... 37 BMD forecasts................................................................................................................................................................. 38 Factors constraining BMD’s forecast practices............................................................................................................. 40 Other physical systems needs........................................................................................................................................ 41 Human resources needs................................................................................................................................................. 42 Collaboration with the IMD............................................................................................................................................ 43 Collaboration with other organizations......................................................................................................................... 46 Chapter 4 Findings and Recommendations............................................................................................................... 47 Key findings ................................................................................................................................................................... 48 Recommendations ........................................................................................................................................................ 51 Annex 1 List of Stakeholders Consulted ...................................................................................................................... 53 References ................................................................................................................................................................... 55 vii Tables and Figures Table 3.1 The Saffir-Simpson Hurricane Intensity Scale......................................................................................... 34 Table 3.2 The Tropical Cyclone Intensity Scale Used by the India Meteorological Department .......................... 34 Figure 2.1 Sample Forecast Map from the Australian Bureau of Meteorology, March 2006 ................................. 9 Figure 2.2 Individual Tracks for Tropical Cyclone Irene .......................................................................................... 18 Figure 2.3 Strike Probability (%) That Tropical Cyclone Irene Will Pass within 120 Kilometers for the Next 10 Days for the ECMWF Ensemble Prediction System Forecast Starting at 00 UTC on August 22, 2011 .................................................................................................................................. 19 Figure 2.4 Strike Probability (%) for Tropical Cyclone Activity within 300 Kilometers (Systems with Maximum Wind Speed Greater Than 8 Meters per Second) for the Two-Day Period of August 20–22 Based on the ECMWF Ensemble Prediction System Forecast from 00 UTC on August 17, 2011 ................................................................................................................................... 20 Figure 2.5 Average Track Errors, in Nautical Miles, in the U.S. National Hurricane Center’s Forecasts as a Function of the Year in Which the Forecasts Were Made ............................................................... 20 Figure 2.6 Average Track Errors, in Knots, in the U.S. National Hurricane Center’s Forecasts as a Function of the Year in Which the Forecasts Were Made ............................................................... 21 Figure 2.7 Track Errors at 48 Hours Lead Time (Nautical Miles), as a Function of Year ........................................ 21 Figure 2.8 Track Errors at 48 Hours Lead Time (Nautical Miles), as a Function of Year, “Early” Guidance .......... 29 Figure 3.1 Tropical Cyclone Tracks in the Bay of Bengal, 1981–2010 .................................................................... 35 Figure 3.2 Total and Mean Number of Tropical Cyclones Reaching at Least Tropical Storm Intensity in the Bay of Bengal Each Year, 1981–2010 ........................................................................................... 35 Figure 3.3 Histograms of the Time, in Days, between Genesis and the First Landfall of Tropical Cyclones in the Bay of Bengal, Western North Pacific, and North Atlantic Basins, 1980–2011 .......................... 36 Figure 3.4 Warning Issued by Bangladesh Meteorological Department on May 13, 2013...................................... 38 Figure 3.5 Warning Issued by Bangladesh Meteorological Department on May 15, 2013 ..................................... 39 Figure 3.6 Track Forecast Issued by the IMD RSMC on May 12, 2013, for Cyclone Mahasen ................................. 44 Figure 3.7 Historical Track Errors for the RSMC (in Blue) and the Joint Typhoon Warning Center (in Red) for Tropical Cyclones in the North Indian Ocean (including both Bay of Bengal and Arabian Sea) .... 46 viii / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Acknowledgments T his report has been prepared through a collaborative Hurricane Specialist, Eric Blake, Hurricane Specialist, Todd effort between the Bangladesh Meteorology Department Kimberlain, Hurricane Specialist, Chris Landsea, Science and and the World Bank. The core team included Adam Operations Officer, Stacy Stewart, Senior Hurricane Specialist, Sobel, Professor, Department of Applied Physics and and Jamie Rhome, Storm Surge Team Lead, all at the National Applied Mathematics, Department of Earth and Environmental Hurricane Center; Frank Marks, Research Meteorologist and Sciences, Lamont-Doherty Earth Observatory, Columbia Director, S. G. Gopalakrishnan, Research Meteorologist, and University, lead technical author and consultant to the World Sim Aberson, Research Meteorologist, at NOAA’s Hurricane Bank; Shahpar Selim, Environmental Specialist, Jorge Jose Research Division; Tim Neely, Chief, Environment, Science Escurra, Water Resources Consultant; Angie Harney, Team & Technology Affairs, and Dec Ly, Science and Technology Assistant; Marie Florence Elvie, Team Assistant; Erika Vargas, Officer, at the U.S. Embassy, New Delhi; Robert Falvey, Director, Knowledge Management Specialist; and Poonam Pillai, Senior Joint Typhoon Warning Center; James Doyle, Head, Mesoscale Disaster Risk Management Specialist and Task Team Leader, Modeling Section, Naval Research Laboratory; Russell Elsberry, World Bank. Peer reviewers for this report included James Distinguished Research Professor, Naval Postgraduate School; Franklin, Branch Chief, Hurricane Specialist Unit, NOAA/NWS/ National Hurricane Center; Frank Marks, Director, Hurricane Frederic Vitart, Senior Scientist, European Centre for Medium- Research Division, NOAA; Christoph Pusch, earlier Lead Range Weather Forecasting; Don Resio, Professor, University Disaster Risk Management Specialist, GPSURR; and Deepak of North Florida, Coastal Inundation Forecast Project; Andrew Singh, Senior Disaster Risk Management Specialist, GPSURR. Burton, Regional Forecasting Centre Operations Manager, Additional comments were received from Taoyong PENG, Chief, Australian Bureau of Meteorology; Treng-Shi Huang, Taiwan Tropical Cyclone Programme Division, Weather and Disaster Central Weather Bureau; Chun-Chieh Wu, Professor, National Risk Reduction Services Department, World Meteorological Taiwan University; Michela Biasutti, Lamont Research Organization; Peter Webster, Professor, School of Earth and Professor, Suzana Camargo, Lamont Research Professor, and Atmospheric Sciences, Georgia Institute of Technology; Dr. Michael Tippett, Associate Professor, Applied Mathematics, at James Belanger, Research Scientist 11, Georgia Institute of Columbia University. Technology; and Lia Sieghart, Program Leader, World Bank. We thank them all. Our sincere thanks to Qimiao Fan, Country Director, Bangladesh, Nepal and Bhutan, Rajashree Paralkar, Manager, The preparation of the report involved extensive consultation Operations, Bangladesh and Sanjay Srivastava, Program with numerous government officials and technical experts. A Leader, Bangladesh Country Management Unit. A special special thanks to Mr. Shamsuddin Ahmed, Director, Bangladesh thanks to Andras Hovrai, earlier the Country Coordinator Meteorological Department; Md. Shah Alam, previous Director, for Bangladesh at the World Bank, for supporting the team Bangladesh Meteorology Department; Arjumand Habib, in initiating this report. Sincere thanks also to Johannes C. previous Director, Bangladesh Meteorology Department; and M. Zutt, earlier Country Director for Bangladesh, Bhutan, Dr. L. S. Rathore, previous Director General, Indian Meteorology and Nepal, to Christine E. Kimes, earlier Operations Advisor, Department for their support and collaboration in the preparation Bernice K. Van Bronkhorst, earlier Practice Manger, Urban and of this report. We also thank the following individuals for their Disaster Risk Management Unit, AFR 1 and Christoph Pusch, kind assistance in the writing of this report: Taslima Imam, Practice Manager, Disaster Risk Management and Climate Meteorologist, at the Bangladesh Meteorological Department; Change Unit, South Asia. for their support. We are immensely Md. Amirul Hussain, Executive Engineer, Flood Forecasting grateful to Robert J. Saum, Director, Regional Integration, and Warning Centre, Bangladesh Water Development Board; Md. Mobassarul Hasan, Senior Specialist, and Zahir-ul Haque South Asia, Salman Zaheer, earlier Director, Regional Khan, Director, Coast, Port & Estuary Managing Division, at Integration, to Sanjay Kathuria, Lead Economist, GMTSA, and the Institute of Water Modeling, Bangladesh; M. Mohapatra, to Eric Nora, Senior Operations Officer, for their support to Scientist, B. K. Bandyopadhyay, Scientist, and S. K. Roy the South Asia Regional Hydromet Services and Resilience Bhowmik, Scientist, at the India Meteorological Department; Program. Thanks also to the U.K. Department for International U. C. Mohanty, Professor, A. D. Rao, Professor, and S. K. Dube, Development for providing grant funding through the Joint Emeritus, at the Indian Institute of Technology; Koji Kuroiwa, Technical Assistance Program, GFDRR, European Union and former Chief, Tropical Cyclone Programme, and Maryam to the government of Netherlands for funding through the Golnaraghi, Chief, Disaster Risk Reduction Programme, at the Water Partnership Program, without which preparation of World Meteorological Organization; Michael Brennan, Senior this report would not have been possible. Review of Operational Practices and Implications for Bangladesh / ix Acronyms ADT Advanced Dvorak Technique ATCF Automated Tropical Cyclone Forecasting System BoB Bay of Bengal BMD Bangladesh Meteorological Department CIFDP Coastal Inundation Forecasting Demonstration Project CIFDP-B Coastal Inundation Forecasting Demonstration Project for Bangladesh CPP Cyclone Preparedness Program DDM Department of Disaster Management DEM digital elevation model ECMWF European Centre for Medium-Range Forecasting ESCAP Economic and Social Commission for Asia and Pacific (United Nations) FFWC Flood Forecasting and Warning Centre GFS Global Forecast System GTS global telecommunication system HFIP Hurricane Forecast Improvement Program HWRF Hurricane Weather Research and Forecast IIT Indian Institute of Technology IMD India Meteorological Department JICA Japan International Cooperation Agency JMA Japan Meteorological Agency JTWC Joint Typhoon Warning Center MOU Memorandum of Understanding NCEP National Centers for Environmental Prediction NHC National Hurricane Center (U.S.) NOAA National Oceanic and Atmospheric Administration (U.S.) RIMES Regional Integrated Multi-Hazard Early Warning System RSMC Regional Specialized Meteorological Center SLOSH sea, lake, and overland surges from hurricanes UTC universal time clock WIS World Meteorological Organization Information System WMO World Meteorological Organization WRF Weather, Research, and Forecast x / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Review of Operational Practices and Implications for Bangladesh / xi Executive Summary Floodwaters surrounding houses in Dhaka. Photo by Stockbyte xii / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING DHAKA Stretching across part of southwestern Bangladesh, the Sundarbans is the largest remaining tract of mangrove forest in the world. Credit: NASA Bangladesh is historically one of the most vulnerable nations to tropical cyclone–induced storm surge. Unlike previous tropical cyclones, such as in 1970 and 1991, the most recent storms have not caused the enormous numbers of fatalities due to better shelters, better warning, last-mile connectivity, and evacuation procedures. Nonetheless, the large population in the low-lying coastal areas of the country remains vulnerable to flooding in cyclone events, particularly as sea level continues to rise. Sundarbans. Credit: Thinkstock.com Review of Operational Practices and Implications for Bangladesh / xiii A key contributor to improved disaster preparedness in Bangladesh would be improved lead times for tropical cyclone and storm surge forecasts. At present, the Bangladesh Meteorological Department (BMD), the main government agency responsible for forecasts for tropical cyclone and storm surges, issues explicit forecasts of tropical cyclone behavior only three days ahead of time. Forecasts with longer lead times are possible and would bring significant benefits. The Regional Specialized Meteorological Centre (RSMC), operated by the India Meteorological Department (IMD) with responsibility for the South Asia region, has recently increased the lead time for their tropical cyclone forecasts to five days, which is the operational standard in most other basins. Advances in numerical weather prediction—particularly ensemble prediction systems operated at numerical weather prediction centers in several countries (often separate institutions from those with a mandate to produce tropical cyclone forecasts)—may allow probabilistic forecasts with useful skill to be issued at still longer lead times, perhaps as long as 10 or 15 days, to the extent that users can understand and accommodate the large uncertainties at those lead times. Critical to extending the lead times for forecasting is a better understanding of the factors that limit the lead time of current forecasts. This issue is important not just for Bangladesh but for all of the Bay of Bengal countries, including Myanmar, Sri Lanka, and India—all of which are at risk from tropical cyclones and at increasing risk of storm surge–driven flooding as sea level rises. OBJECTIVES The main objectives of this report are to (i) assess the current state-of-the-art tools and operational practices in tropical cyclone and storm surge forecasting, (ii) assess existing operational practices at the BMD and regionally, and (iii) propose recommendations for improvements in the lead times of tropical cyclone forecasts for Bangladesh. APPROACH The paper is based on a review of documentary research and consultations with stakeholders. Extensive discussions were held with officials at the BMD, the IMD, and other weather services. In reviewing international operational practices, considerable focus is placed on the practices of the U.S. National Hurricane Center in Miami, with input from a number of other well-equipped and well-staffed centers, primarily in industrial nations. Further analysis can be undertaken as a follow-up to this report. xiv / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING KEY FINDINGS Review of international and regional operational practices Tropical cyclone forecast lead time: At state-of-the-art source of uncertainty in storm surge forecasts, which national centers, tropical cyclone forecasts are produced limits the lead time, is uncertainty in the tropical with lead times of up to five days. Storm surge forecasts cyclone forecast. are produced with shorter lead times of 72 hours or less. Tropical cyclone forecasts at the IMD: The RSMC While the necessary science and technology to further for tropical cyclones at the IMD identifies and names increase this lead time—and thus to allow forecasts of tropical cyclones in the Bay of Bengal; provides landfall more than five days ahead of time for all storms outlooks, advisories, and warnings to the BMD and in the Bay of Bengal—exist at the level of research and other countries in a panel on tropical cyclones for the development, they have not yet been mainstreamed Bay of Bengal and the Arabian Sea region; and serves into operational forecast practices of most national as the hub for the transmission of meteorological data meteorological services (including those of the most via the global telecommunication system (GTS)/World industrialized nations). Meteorological Organization (WMO) Information System Skill of forecast track has improved faster than skill (WIS).1 Objective tropical cyclone track and intensity of forecast intensity: The skill of tropical cyclone track forecasts have been produced at the IMD since 2003, forecasts has improved dramatically in recent decades, at which time the maximum lead time was 24 hours; the maximum lead time was increased to 72 hours in 2009 in large part due to steady improvements in global and to 120 hours in 2013. Storm surge guidance is also numerical weather prediction models, observations included, based on the Indian Institute of Technology and data assimilation systems used to initiate those Delhi and Indian National Centre for Ocean Information models, and ensemble methods. Ensemble methods Services storm surge model, at shorter lead times. allow improvement over individual model performance and also allow estimation of forecast uncertainty. The IMD has considerably higher capabilities in terms Forecasts of tropical cyclone intensity, on the other of observing system, research and development support hand, have not improved at anywhere near the rate that base, computing power, modelling and skill levels of track forecasts have. The goal of improved intensity available human resource than the BMD does. It has forecasts is currently driving a large scientific research computer workstations dedicated to tropical cyclone effort, particularly in the United States. forecasting. Dedicated software packages, comparable to those at the National Hurricane Center, are available Timing of operational tropical cyclone forecasts: to ingest necessary data and automate some tasks to Forecasts of tropical cyclogenesis—the initial formation streamline and facilitate the forecaster’s job. Dvorak of tropical cyclones—have improved greatly in recent analysis is carried out on these workstations once the years. Genesis can now be forecast with some accuracy necessary satellite data have been ingested. Its forecasts at leads of five days or longer. Operational forecasts of are quantitative and extend to longer lead time than the subsequent tropical cyclone’s track and intensity, the BMD’s. In a few respects, however, IMD forecast however, are not begun until genesis has occurred, technologies lag what is available at some other centers. although ensemble prediction systems already provide For instance, while much numerical model guidance the raw material to enable such forecasts to start before is ingested into the forecast software, the only tropical genesis and although such forecasts have been made, cyclone tracks ingested are those from the deterministic both by academic researchers and by U.S. forecasters, model runs. Any improvement in IMD forecasts has the on an experimental basis. potential to improve BMD forecasts as well. But due to capacity constraints at the BMD, data and information Uncertainty and storm surge forecasting: Storm surge already publicly available and provided regionally are forecasting is being carried out as a part of operational often not used for forecasting at the BMD. tropical cyclone forecasting. Distinct numerical models representing the ocean, separate from those used to Bangladesh is connected to the Global Telecommunication 1 forecast the tropical cyclone itself, are used to predict System (GTS) and is in the process of transitioning to the WMO the surge. The tropical cyclone forecast provides the Information System (WIS). The two terms are used interchange- winds that drive the storm surge forecast. The greatest ably in this report. Review of Operational Practices and Implications for Bangladesh / xv FORECASTING PRACTICES AT THE BMD BMD tropical cyclone forecasts: The BMD produces tropical cyclone forecasting at the BMD. These data tropical cyclone and storm surge forecasts for are not available digitally and even the images are not Bangladesh. The lead times for these forecasts are 72 consulted. Since these data can be valuable for estimates hours or less. It appears that the BMD does not issue—at of tropical cyclone intensity and structure, it would be least not on the Internet—quantitative track or intensity advantageous if they were available and used. Part of the forecasts. A map showing the observed track (up to the issue may be training in the use of these data. present time) is issued, but it does not extend into the Third, the observational networks over land and the future. BMD forecasts of tropical cyclones tend to be adjacent ocean, which are an essential component limited to textual forecasts and warnings. These give of the tropical cyclone forecast process, also need only qualitative information about the storm’s future improvement. For instance, adequate bathymetric data behavior and do not extend far in the future. for coastal Bangladesh are not available for storm surge Factors constraining lead time for BMD forecasts: The and coastal inundation forecasting. Automated tide lead-time and skill of BMD forecasts are limited by a gauges are also needed for measurements of storm surge. number of factors, both material and human. First, the Fourth, the BMD does not have a dedicated staff for agency lacks much of the state-of-the-art hardware and forecasting tropical cyclones. When a tropical cyclone software used elsewhere for tropical cyclone forecasts is present in the Bay of Bengal, forecasts of its track and does not obtain all the globally available and and intensity are prepared at the BMD by the same potentially useful data from observations and numerical forecasters who normally forecast other types of models. For instance, BMD forecasters currently lack weather during the remainder of the year. The education tools to carry out key tasks such as visualization of and training opportunities available to staff are also model forecast guidance and production of graphic track limited, making it more difficult for forecasters to take forecasts based from that guidance, as well as Dvorak advantage of the latest developments in tropical cyclone analysis and other steps associated with assessment of and storm surge forecasting. At present, there is no the storm’s present state. These are made easier and Department of Atmospheric Sciences (or Meteorology, more effective by computer hardware and especially or any equivalent or comparable designation of the by software designed specifically for the purpose. field) anywhere in Bangladesh. There is an atmospheric Further, at present the BMD does not operationally run physics research group in the Physics Department at the models dedicated to tropical cyclone forecasting. The Bangladesh University of Engineering and Technology BMD also faces frequent power outages that disrupt its and scattered individual faculty members in related operations. fields at other Bangladeshi universities, but no group that is equivalent to an entire department or that has Second, a critical limitation for the BMD is the bandwidth strong ties to the BMD. This seriously compromises of its GTS/WIS link. For instance, coastal radar data the government’s ability to forge links with academic from Bangladesh are not sent back to the IMD through institutions on weather- and climate-related research— the GTS and thus are not available for assimilation into linkages and partnerships that are often at the crux of the IMD’s numerical models. Currently the bandwidth of innovation and research-based service delivery. BMD’s GTS link is 64 kilobytes per second, whereas 5 megabytes per second are estimated to be required for Influence of the size of the Bay of Bengal: In addition sharing BMD’s coastal radar observations over the GTS to the above factors, the nature of the Bay of Bengal back to the IMD. Moreover, due to the limited bandwidth also plays a role in influencing the lead time for tropical of the GTS link, many data are available to the BMD only cyclone forecasting in the region. A key limitation on through the Internet in the form of images viewed on a the maximum lead time of forecasts for Bangladesh— Web browser and are not available in a digital form for whether they are produced by the BMD, the IMD, assimilation into models. Some data from polar orbiting or someone else—is the natural physical constraint satellites, such as microwave sounders and imager imposed by the smallness of the Bay of Bengal. Because scatterometers for surface winds are not used at all in Bay storms cannot move far without reaching land, xvi / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING most of them have lifetimes significantly shorter than forecasts are now being produced by some numerical five days, measured from the time of genesis to landfall. weather prediction centers and university groups using As long as track forecasts begin at genesis, landfall ensemble methods and products with the necessary forecasts for such storms cannot be made with more information. The National Hurricane Center in the US than five days of lead time. has started issuing watches and forecasts for potential tropical cyclones, though this is very recent and is being Most national centers are operationally producing 5 done out to 48 hours. If IMD and BMD were to begin day forecasts. If forecasts of landfall are desired with issuing forecasts of genesis, they would be matching lead times greater than five days for all storms, it will be what some of the more sophisticated forecasting essential that the forecasts begin before genesis. Such centers have only recently started to operationalize. RECOMMENDATIONS For the BMD to improve its forecast lead times, a number of actions can be taken. Strengthen BMD hardware, software, and and inundation forecasting, and calibration of coastal infrastructure: In the short term, some relatively basic radars, as well as intercomparison against rain gauges and inexpensive improvements could help improve and installation of buoys for oceanographic and BMD’s capacity for tropical cyclone forecasting. The meteorological observations offshore. BMD should obtain dedicated workstations with Improve training and capacity building: Improved appropriate software to carry out key tasks associated education and training opportunities for BMD staff with tropical cyclone forecasting, such as model are critical in improving the agency’s capacity for analysis, visualization, and so forth. The department improved service delivery. In the short term, the goal should also ensure backup systems for its computers in of improving forecasts requires more training of BMD case of power outage. personnel in existing and developing science and Enhance data and information sharing through technology that specifically address that goal: Dvorak improved network systems: In addition to technique, numerical weather prediction, ensemble improvements in computer hardware and software, the prediction methodologies, radar data analysis, and BMD can have better access to useful data—both from the like. It would be particularly valuable for BMD observations and from numerical weather prediction forecasters to become better acquainted with the models—that are already, in principle, available through capabilities of the modern global model ensemble improvements in network systems. The department prediction systems. In the long term, however, the should make efforts to increase the bandwidth available government of Bangladesh will need to invest in for its GTS link so that it can obtain information and development of a cadre of trained meteorologists and products available regionally and globally. This will also atmospheric scientists by supporting teaching of these allow the BMD to share data and information with the topics at the university level. Another recommendation IMD and other relevant agencies. is to establish a National Meteorological Training and Research Center in Bangladesh to meet the national Strengthen the observation network for tropical requirements. In that case, the existing Meteorological cyclone forecasting: The BMD would benefit from Training Institute of the BMD can be upgraded to improvements and expansion of the current observing contemporary standards. Support should also be system, such as access to automated tide gauges at provided for establishing university level departments the coast for measurements of storm surge, better focusing on meteorology and atmospheric sciences. bathymetric and topographic data for storm surge Review of Operational Practices and Implications for Bangladesh / xvii Improve coordination between the BMD and other Strengthen regional collaboration, including the agencies: For improvements in forecasting to contribute IMD’s role in research, technology transfer, and meaningfully to disaster preparedness and improved training: The RSMC at the IMD is the regional center for early warning systems, close coordination between predicting tropical cyclones and storm surges and its the BMD and other agencies such as the Bangladesh role as a coordinator of research, technology transfer, Water Development Board and the Department of and training should be strengthened for the benefit of Disaster Management is needed. There are already the IMD, the BMD, and other operational agencies in the strong relations between these agencies that can region. be further enhanced. In storm surge forecasting, Study the use of ensemble forecasts for improving WMO’s Coastal Inundation Forecast Demonstration lead times for tropical cyclone forecasting: The Project for Bangladesh aims to improve the state of possibility of producing ensemble forecasts for South the art and is expected to provide important lessons Asia with lead times greater than five days should be on how to improve coastal inundation forecasts in actively studied. The active interest and engagement Bangladesh, and it should be supported. This could of the local agencies—the BMD and the IMD—is include facilitation of better cooperation between essential to this effort, as they bear responsibility for the BMD and other national agencies, particularly the forecasting for Bangladesh for the region respectively. Bangladesh Water Development Board. The critical Important questions in this regard include not just problems of forecast evaluation and verification need those addressed in this report but also the extent to to be addressed by improvements to the observational which the greater uncertainties associated with longer network—in particular, by automated gauges that can lead-time forecasts may be compatible with their use in measure water levels at the coast and inland. emergency management. Tropical Cyclone Mahasen. Credit: www.skymetweather.com xviii / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING 1 Chapter 1 Tropical Cyclones in the Bay of Bengal Bangladesh is one of the most densely populated countries in the world. Owing to its low-lying topography, dense river network, location, and climate, it is exposed to a range of water- and climate-related hazards. Tropical cyclones are among the most severe of these hazards. 2 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Although the Bay of Bengal (BoB) accounts for only a small forecasts can be produced so that landfall can be predicted fraction of the world’s tropical cyclones, 10 of the 14 of these five days or more ahead of time. storms associated with the highest fatalities globally have occurred in the Bay of Bengal, with a large fraction of those Webster (2008, 2012, 2013), in particular, argues that affecting Bangladesh (Webster 2012). Bangladesh is historically probabilistic forecasts with useful skill can be issued at among the most vulnerable nations to tropical cyclone–induced times longer than five days in advance—perhaps even as storm surge. Over 300,000 people were killed by a devastating much as 15 days before landfall. The essential ingredients cyclone in 1970 and 140,000 in 1991. Cyclone Sidr, which of such forecasts already exist in the outputs from ensemble made landfall in Bangladesh in 2007, caused around 3,500 prediction systems such as those produced by the European deaths; this much smaller number, compared with the 1970 Centre for Medium-Range Forecasting (ECMWF). Such and 1991 events, has been attributed to better shelters and extended lead times for tropical cyclone forecasts could, if better warning and evacuation procedures. Nonetheless, the the forecasts were accurate enough to be useful in decision large population in the low-lying coastal areas of the country making, enable significant cost savings and a palpable remains vulnerable to flooding in cyclone events, particularly positive impact on local livelihood and assets. as sea level continues to rise. If lead times of 10–15 days with relatively high accuracy A key focus for improving disaster preparedness and early are possible, as the literature suggests, why is it not being warning systems in Bangladesh is improved lead times for done in Bangladesh and what can be done to improve the tropical cyclone forecasting including the quality and skill forecast lead time?? This question motivated the writing of of the forecast. At present, the lead time for tropical cyclone this paper. Critical to extending the lead times for forecasting forecast used by the Bangladesh Meteorological Department is a better understanding of the factors that limit increasing (BMD), the main government agency responsible for issuing the lead time of current forecasts. This issue is important not forecasts for tropical cyclone and storm surges, is three days. just for Bangladesh but for all of the Bay of Bengal countries, However, based on publicly available data (for example, including Sri Lanka, Myanmar, and India—all of which are at from satellites) and existing weather models, it is possible risk from tropical cyclones and at increasing risk of storm to increase the lead time at which useful tropical cyclone surge–driven flooding as sea level rises. OBJECTIVES The main objectives of this report are to: cyclone forecasts for the South Asia region, its operational practices with respect to tropical cyclone forecasting are also Assess state-of-the-art tools and operational practices considered. The issue of whether the small size of the Bay of in tropical cyclone and storm surge forecasting, Bengal is a limiting factor in increasing lead times for tropical including regional operational practices in South Asia cyclone forecasts is also explored. at the Regional Specialized Meteorological Center (RSMC) located at the India Meteorological Department This report extends previous literature on the topic in several (IMD) in Delhi ways. First, it presents a broad overview of current international operational practices as undertaken by national meteorological Assess current operational practices and identify key centers with responsibility for tropical cyclone forecasting, factors constraining improvement of lead time for which has so far not been undertaken. This analysis provides tropical cyclone forecasting at BMD an important context for understanding operational practices Propose recommendations for improving current in developing countries such as Bangladesh. Second, based operational practices at BMD. on consultations with relevant agencies and stakeholders, the paper also assesses the current practices and capabilities of To the extent that the IMD is designated as the RSMC by the the BMD in some detail, comparing them with current practices World Meteorological Organization (WMO) and issues tropical at other centers. Review of Operational Practices and Implications for Bangladesh / 3 The report has a limited focus as indicated above, and usefulness in disaster preparedness. How certain must a several important questions are not addressed here. One forecast be in order for useful action to be taken based on it? is the relative value of improvements in forecasts versus While these questions are centrally important to the broader improvements in other aspects of disaster preparedness, task of reducing harmful tropical cyclone impacts, they including communications, local preparedness activities, are outside the scope of this report. Finally, the review of and the building of shelters. Much of the large reduction in international and regional operational practices mainly relies fatalities in Sidr in 2007 compared with the 1970 and 1991 on the operational practices of agencies in a few countries due storms in Bangladesh is attributed to these other measures. to limitations of budget. A more detailed analysis based on Another is the extent to which the greater uncertainties the experiences of additional countries should be undertaken associated with longer lead-time forecasts diminish their as a follow-up to this report. APPROACH AND METHODOLOGY The information and points of view in this paper are derived were also consulted. A visit to the WMO provided additional from a literature review and discussions with a range of perspectives on its role in facilitating and organizing tropical officials and experts (see Annex 1). This included consultations cyclone and storm surge prediction activities in South Asia with staff at the U.S. National Hurricane Center (NHC) and the and worldwide. In Bangladesh, extensive consultations were Hurricane Research Division of the U.S. National Oceanic and undertaken with officials at the BMD, the Bangladesh Water Atmospheric Administration (NOAA), both in Miami, Florida, Development Board, and other agencies. These discussions and with officials in several national meteorological agencies, were important particularly to clarify current operational such as the Australian Bureau of Meteorology and regional forecast practices, which are often not documented fully in weather forecasting agencies. Officials at the IMD and the peer-reviewed scientific literature. experts at the Indian Institute of Technology (IIT) in Delhi ORGANIZATION OF THE REPORT Following this description of the background and rationale particular, we examine the extent to which Bangladesh uses for the report, Chapter 2 describes current global operational international and regional operational practices and the practices for forecasting tropical cyclones and storm surges, technical and governance issues that limit their use. Chapter including practices at other national agencies. Chapter 3 4 provides a summary and recommendations. assesses current operational practices in Bangladesh. In 4 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Typhoon Nargis over the Bay of Bengal. Credit: NASA / 5 Chapter 2 Forecasting: Analysis of Operational Practices TROPICAL CYCLONE FORECAST WHAT IS A TROPICAL CYCLONE? The formal definition of a tropical cyclone is not simple. The American Meteorological Society’s Glossary of Meteorology, for example, gives a long and complex definition (see http://glossary.ametsoc. org/wiki/Tropical_cyclone). A shorter definition is that used by the U.S. National Hurricane Center: “A warm- core non-frontal synoptic-scale cyclone, originating over tropical or subtropical waters, with organized deep convection and a closed surface wind circulation about a well-defined center.” 6 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING A tropical cyclone whose wind speeds are below 34 knots (17 cyclone. These criteria are basin-dependent and include meters per second) is a tropical depression. When a tropical thresholds for maximum sustained surface wind speed in cyclone’s wind speeds reach 34 knots, it is labeled a tropical some basins but not others. The U.S. National Hurricane storm in several basins and a cyclonic storm in the North Center, for example, requires no wind speed threshold but Indian Ocean. Definitions of storms with higher intensities begins forecasts when disturbances reach the depression also vary from basin to basin, as discussed further in Chapter stage, as defined by the presence of organized deep convection 3 (see Tables 3.1 and 3.2 and accompanying discussion). and a closed circulation. The IMD, on the other hand, begins forecasting when winds reach 28 knots, while the Australian In practice, there is broad agreement in most important cases Bureau of Meteorology begins 24–48 hours before the storm about whether a given weather disturbance is a tropical is expected to reach 34 knots. cyclone or not, but there can be disagreement, particularly for weak systems or systems with some properties of The process of a tropical cyclone’s coming into existence by extratropical cyclones (cold core, less deep convection, etc.). meeting these criteria for the first time—whatever the criteria The fact that different definitions are used for weak systems are in the basin in question—is known as genesis. Forecasting has ramifications for forecasting. In the operational systems genesis is traditionally distinct from forecasting the behavior used at present by all national forecast centers we are aware of existing storms (although they are done by the same of, tropical cyclone forecasts are begun once a given system forecasters). These are described separately below. meets specific criteria that are used to define a tropical WHAT IS A TROPICAL CYCLONE FORECAST? A tropical cyclone forecast predicts the future behavior of a of interest, such as the position or intensity of the storm. The tropical cyclone. National centers currently produce forecasts error in the forecast is the difference between that value and of tropical cyclone track, intensity, and other parameters for that observed at the verification time. Deterministic forecasts a given tropical cyclone once it already exists—that is, once it are virtually never precisely correct, particularly at longer has met a basin-dependent set of criteria, as just described. lead times, but the goal is for the error to be small, averaged Before discussing specific types of forecasts, it is important to over many forecasts. For example, if the forecast maximum clarify how some terms are defined in this report. wind speed of a tropical cyclone at a given lead time is 50 knots, and the intensity observed at the verification time is 55 The lead time of a forecast is understood as the difference knots, the intensity error is 5 knots. The skill of deterministic between the time at which the forecast is issued and the time forecasts is typically evaluated by averaging these errors over for which the forecast applies, known as the verification time. all storms for which forecasts were issued over some period If a forecast is issued today at 12 noon that predicts what the of time. The averaging must be done separately for each weather will be tomorrow at 12 noon, the verification time is lead time, because errors typically increase with lead time; tomorrow at 12 noon and the lead time is one day. Forecasts 24-hour forecasts should be compared with other 24-hour are typically issued with lead times between zero—a forecasts but not with 48-hour forecasts, for example. statement of the storm’s present properties at the forecast time, known as the “analysis”—and either three or five days, A probabilistic forecast predicts outcomes as probabilities at either 6- or 12-hour intervals. rather than categorically. For example, a probabilistic forecast could state that there is a 20 percent probability Forecasts can be categorized as either deterministic or that a particular location will experience winds of 65 knots probabilistic. or greater during a particular 12-hour time period. Unlike a deterministic forecast, a single probabilistic forecast can A deterministic forecast gives categorical predictions of the never be incorrect (unless the probabilities stated are 0 values of the forecast storm parameters at a given lead time, percent or 100 percent, in which case it is effectively a with no explicit statement of uncertainty. At each lead time, deterministic forecast), nor can the error in a single forecast a deterministic forecast gives a single value for each variable be measured. Review of Operational Practices and Implications for Bangladesh / 7 The skill of probabilistic forecasts can only be measured over departures from climatology indicate that the forecast a large set of such forecasts. The skill can be broken down contains information. Since the climatology is known even if into two components: reliability and resolution. Reliability there is no forecast, a forecast of climatological probabilities, measures whether the forecast outcomes actually occur such as 50 percent chance of being above the median, does with frequencies given in the forecast. Taking the example not add anything.) Evaluating probabilistic forecasts is more above, consider a large set of forecasts predicting a 50 complex than evaluating deterministic forecasts.2 percent probability of 65 knot winds or greater. If winds of 65 knots or greater are observed at the time of verification Probabilistic forecasts offer the potential for a faithful in 50 percent of the cases, these forecasts will be found to expression of the actual knowledge available to the forecaster be reliable. Resolution, on the other hand, measures how at the time of forecast, since there is always some uncertainty different the probabilities are from their climatological in that knowledge. However, understanding the nature of that uncertainty requires some sophistication on the part value— that is, from the average frequency of the given of the user. Deterministic forecasts are easier for the user to outcome over time. A forecast of 50 percent probability of understand, but they implicitly (or explicitly) overstate the being above the median, issued repeatedly, will have low forecaster’s certainty about the future. resolution (even if it is reliable). The ideal is to have good resolution—probabilities that capture significant departures 2 Special thanks to Michael Tippett (Associate Professor, Columbia from the climatology—without losing reliability. (Significant University) for clarifying this issue. FORECASTING EXISTING STORMS Tropical cyclone forecast centers might issue a variety of hour = 0.51 meters per second. (Intensity can also be products to communicate the impending arrival of a cyclone. quantified by the minimum surface pressure of the The intended audience for these products may include storm, in units of hectopascals, although this measure is emergency managers, the media, and the general public. less directly relevant to users than wind and so usually The forecast is usually presented with some combination of is not featured prominently in forecasts for the public.) words, images, and numerical information (for example, in Size: Several measures of storm size may be forecast. tabular form). These describe the present state of the storm Common measures are the radii at which winds as well as predictions for its future. In addition to the forecast decrease below given thresholds (e.g., 34 or 64 knots, itself, special alert messages may be issued when the threat although the IMD also issues forecasts of 28 knot and to populations or other interests is deemed to have reached 50 knot wind radii). Because real storm wind fields are specific pre-determined levels; in the United States and some not circularly symmetric, these radii are often forecast other places, these are watches and warnings, defined below. separately for four different quadrants, or 90-degree wedges, surrounding the storm center. The behavior of a tropical cyclone is represented in a forecast by a small set of parameters that describe the state of the The track, intensity, and size of the storm may be forecast storm at each time for which forecasts are issued. The deterministically or probabilistically. Probabilistic forecasts parameters predicted usually include the following: yield products such as cones of uncertainty. Many forecast centers use such cones, although their precise meaning varies Track: The latitude and longitude of the storm center slightly from center to center. For example, the National with time. Hurricane Center in Miami, Florida, uses the following definition: The cone represents the probable track of the Intensity: The maximum surface wind speed associated center of a tropical cyclone and is formed by enclosing the with the storm that is sustained for a specific averaging time (where the averaging time varies from basin to area swept out by a set of circles along the forecast track (at basin). It is common to measure wind speed in knots 12, 24, 36 hours, and so forth). The size of each circle is set and to predict intensity to within 5 knots. For reference, so that two-thirds of historical official forecast errors over the 1 knot = 1.15 miles per hour = 1.85 kilometers per previous five-year sample fall within the circle. In other words, 8 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING the forecast states that there is a 67 percent probability that 48 hours, and Cyclone Warnings at least 24 hours ahead of the center of the storm will lie within the given circle at each the expected hazard on the coast. lead time. As “possible” and “expected” are qualitative terms, some Because the probabilities used by NHC to define the sizes centers also issue quantitative forecasts of the specific of the circles forming the cone are estimated by the set of probabilities that tropical storm-force and hurricane-force historical errors in these deterministic forecasts over the winds will be experienced. These can be expressed as areas last five years, they are static; throughout a season, at each indicated on a map where the range of probabilities for winds given lead time (for example, 48 hours), the radius of the of a given strength or greater fall within a particular range 67 percent probability circle is the same for each storm. (say, 30–50 percent) are filled in a single color, while those in However, the actual degree of scientific uncertainty about the other ranges are filled in other colors. forecast position of the storm at a particular lead time may be different from one storm to the next. Such situational, dynamic Figure 2.1 shows a sample forecast map from the Australian uncertainties can be estimated from model ensembles. Some Bureau of Meteorology. This was issued for Severe Cyclone centers are beginning to use such ensemble information in Alfred in March 2006. The map indicates the current location defining their cones of uncertainty (Dupont et al. 2011); this and intensity of the cyclone (category 3 in the Australian is discussed further below. system); current and forecast radii of very destructive, destructive, and gale-force winds (here depicted by circular Forecasts of storm track, intensity, and size alone are areas, as different radii are not forecast for different not optimal for many users, as they do not explicitly tell quadrants); the most likely track and range of likely tracks; people what hazards will be experienced at their location. and warning zones for gale-force winds within 24 and 48 Other forecast products are designed to give the user this hours. The range of likely tracks is broadly analogous to the information. Watches, Warnings, and Alerts, in particular, cone of uncertainty used by the U.S. NHC and other centers. express the likelihood of wind hazard at a specific location Here it is not quite a cone, in that the range is not a set of over an interval of time. We discuss these products as defined perfect circles centered on the most likely track with radii in the United States by the NHC and in India by the IMD; fixed to predetermined values at each lead time; the Australian other centers may use different definitions, but typically have forecasters adjust their range of likely tracks manually for broadly analogous categories expressing different levels of each storm. At present, that adjustment may be made based imminence or severity of the threat. on the spread in the ensemble of tracks from one or more forecast models (Andrew Burton, personal communication). The different products differ in the degree of certainty and the closeness in time of the impending hazard. The In addition to the range of quantitative products with fixed NHC defines a Hurricane Watch as “an announcement that formats—track, intensity, and size forecasts, watches and hurricane conditions are possible within the specified warnings, and so forth—forecast centers typically issue area,” while a Hurricane Warning is “an announcement that verbal statements. These allow a greater degree of flexibility hurricane conditions are expected within the specified area.” and nuance. They may explain the nature of the uncertainty, Analogous definitions are used for Tropical Storm (greater the logic of the forecasters’ thinking, or the severity of the than 34 knots) Watches and Warnings. These are issued for potential hazards in ways that the formatted products do specific localities, which may be listed in text form (as a list not allow. As an example, the map shown in Figure 2.1 was of counties or other administrative geographic units, such as accompanied by the following remarks: “Severe tropical towns) or as areas demarcated on a map. Watches are issued cyclone Alfred is expected to continue intensifying and start by NHC 48 hours in advance of the expected hazard, while moving towards the North Kimberley coast today. Tomorrow, warnings are issued 36 hours ahead.3 In India, the IMD issues it should recurve toward the southeast and impact the coast Precyclone Watches at least 72 hours, Cyclone Alerts at least later in the day. It will be weakening, however destructive winds are still expected on the far north coast. Significant NHC issues Hurricane Warnings 36 hours ahead of the expected 3 arrival of tropical storm conditions, because many preparedness rainfall is expected over much of the northern Kimberley actions need to start once winds reach tropical storm strength. causing significant flooding.” Review of Operational Practices and Implications for Bangladesh / 9 While the remarks restate information available in the In particular, they describe the hazard due to rainfall and forecast map, they also contain additional information. consequent flooding, which are not indicated in the map. Figure 2.1 Sample Forecast Map from the Australian Bureau of Meteorology, March 2006 Community Threat Past Cyclone Details Warning Zone – Gales within 24 hours Forecast Location and Intensity Number Warning Zone – Gale from 24 to 48 hours Past Track and Movement Forecast Cyclone Details Current Cyclone Details (at 24 and 48 hours from issue) Current Location and Intensity Number Forecast Location and Intensity Number Very Destructive Winds Vey Destructive Wind Boundary Destructive Winds Destructive Wind Boundary Gale Force Winds Gale Force Wind Boundary Most Likely Future Track Range of Likely Tracks The forecast path shown above is the Bureau’s best estimate of the cyclone’s future movement and intensity. There is always some uncertainty associated with tropical cyclone forecasting and the grey zone indicates the range of likely tracks. Source: From http://www.bom.gov.au/cyclone/about. Note: This map issued for Severe Tropical Cyclone Alfred, March 2006. 10 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING FORECASTING GENESIS Forecasting the genesis of a tropical cyclone is qualitatively across centers globally. In these forecasts, the probabilities different from forecasting the behavior of a tropical cyclone of genesis are forecast in 10 percent increments. These after it exists. In forecasting the genesis, the forecaster forecasts have been produced since 2008, initially at lead is predicting the behavior of a weaker tropical weather times up to 48 hours, then increased to 120 hours in 2013. disturbance and trying to predict whether it will undergo the process of tropical cyclogenesis—becoming a tropical In recent years, the ability of global dynamical models cyclone. to predict genesis accurately has increased significantly, particularly when multiple models are considered as a group The process of tropical cyclogenesis is less well understood (Halperin et al. 2013). Additionally, scientific understanding than the processes that control the track of a mature tropical of large-scale influences on genesis, particularly the Madden- cyclone. Until relatively recently, numerical models of any Julian oscillation and convectively coupled equatorial waves kind were not capable of predicting genesis with any skill. (for example, Zhang 2005; Bessafi and Wheeler 2006; The practice of genesis forecasting has thus traditionally Kiladis et al. 2009; Camargo, Wheeler, and Sobel 2009), has been heavily reliant on the forecaster’s judgment and skill. improved. As a consequence of both of these factors, the NHC This has been rapidly changing in recent years as the ability has now extended their genesis forecasts to 120 hours. The of numerical models to simulate genesis has improved Australian Bureau of Meteorology produces genesis forecasts dramatically. for the public up to three days in advance and may provide some general text on the likelihood of genesis as much as In many basins—the North Atlantic or Western North Pacific, seven days in advance in some circumstances. The Bureau, on for example—genesis often (though by no means always) a commercial basis, also produces specialized forecasts for occurs far from land, so that the time between genesis and industrial clients with lead times as long as 28 days (Andrew the possibility of landfall is relatively large—say, five days Burton, personal communication). Using a dynamical- or longer. Additionally, systems are, by definition, weak statistical model, the IMD produces a forecast of a Genesis immediately after genesis and less threatening than they will Potential Parameter for the Northern Indian Ocean, which become if they intensify. These factors may have historically expresses the probability of tropical cyclogenesis for the next led to less effort being spent on the forecasting of genesis seven days. than on the forecasting of the tracks and intensities of existing storms. Genesis forecasting was not seen as essential to the Even when genesis is forecast with high confidence, however, protection of life and property in most cases. In some basins, it is still kept separate from forecasts of the resulting mature however—such as the North Indian Ocean or the Australian storm. The traditional forecast process does not “cross” region—it is more common for genesis to occur relatively the genesis event. In other words, if forecasts of track and close to land, and such systems may pose dangers despite the intensity are issued out to 120 hours for existing tropical short time available for intensification before landfall (either cyclones, and genesis is believed very likely to occur within because they intensify rapidly or because some hazards, the next 48 hours, in theory it would be possible to issue a particularly rainfall-driven flooding, can be acute even for forecast product that would include both the genesis event nominally weak systems). Existing forecast systems may not itself (however defined, recognizing that this is different handle such cases well. from one basin and forecast center to another) and the first 72 hours of the post-genesis tropical cyclone. Ensemble Across different centers, forecasts of tropical cyclogenesis are prediction systems do, in fact, allow such forecasts to be less uniform in content and format than track and intensity produced in principle (e.g., Belanger et al. 2013). However, forecasts for mature storms. Some centers do not forecast at the time of writing this report most national centers genesis at all, and those that do typically do so at shorter lead with formal responsibility for tropical cyclone forecasting times and with less quantitative precision (at least until very do not issue such forecasts to the public as part of routine recently) than they use when forecasting track and intensity operations, although the NHC (and perhaps other agencies) for existing storms. The NHC produces genesis forecasts that have begun doing so recently at short lead times in a limited are among the more detailed and quantitative of those issued set of circumstances. Review of Operational Practices and Implications for Bangladesh / 11 WHAT TYPES OF OBSERVATIONS ARE USED? The start of any forecast system is data and observations. they are also very valuable, but they have limitations in The following types of observations are relevant to tropical either strong precipitation or very high wind conditions. cyclone analysis and prediction: Weather radar can image the precipitation field, and, if it is a Doppler radar, the wind field in areas Surface observations include surface meteorological of precipitation, at high resolution in both space and stations on land, instrumented buoys at sea, and time. Radars are typically based on land and have a ships at sea. These may measure wind, pressure, temperature, humidity, and sometimes other variables horizontal range of 100–200 kilometers. Radars can such as precipitation or radiation. scan vertically to obtain three-dimensional volumetric observations of radar reflectivity (related to the Radiosondes include weather balloons launched from quantity and type of precipitation-sized particles) specific weather-observing stations. These measure and velocity of the precipitation particles (which is wind, pressure, temperature, and humidity over a normally assumed to be close to the velocity of the large vertical range, typically the entire depth of the air) in the case of Doppler radar, at all altitudes at troposphere, or into the lower stratosphere (above 15 which there are hydrometeors large enough to reflect kilometers). The number of radiosonde stations owned the radar beam. They are valuable for tropical cyclone and operated by national meteorological agencies is observations once a storm gets close enough to land to much smaller than the number of surface meteorological be within range, but they are generally not available observing stations. Radiosonde observations are over the open ocean. Exceptions are the space-borne generally not available over the ocean, apart from a few radars aboard the tropical rainfall measuring mission select islands. and CloudSat satellites, operated by the U.S. National Aeronautics and Space Administration, both of which Geostationary satellites stay in a fixed position relative capture intermittent snapshots of tropical cyclones to Earth and image a large area of Earth’s surface. They over the ocean. produce images from the infrared wavelengths during all hours as well as visible wavelengths during daylight Aircraft specifically instrumented for tropical cyclone hours. Several countries, including the United States, observation may be deployed when available. At Japan, China, India, and the European Union, operate present, such aircraft are deployed routinely only by the a group of geostationary satellites that continuously United States in the North Atlantic and East Pacific and image the entire Earth. They are essential to modern by Taiwan and Hong Kong in the Western North Pacific. tropical cyclone analysis and prediction because they These aircraft take a range of observations, including make it virtually impossible for a tropical cyclone to flight-level measurements of all normal meteorological go undetected. Images are typically available every 30 variables (wind, temperature, humidity). Some aircraft minutes. carry radar, including in some cases Doppler radar. They may have the capability to launch dropsondes, Polar orbiting satellites move relative to Earth’s surface which are equivalent to radiosondes except that they and image only a small area every day. Some sensors fall downward rather than rising upward and thus are on polar orbiting satellites make observations in the available only below the flight level. Some aircraft may microwave portion of the electromagnetic spectrum; have additional remote sensors, such as radiometers these instruments can see through clouds and allow (similar to those deployed on polar orbiting satellites), more clear imaging of the inner structure of a tropical which can measure surface winds over the ocean. cyclone. When available, such images are very useful for characterizing the structure and intensity of tropical A large subset of the available observations over the cyclones. Because polar orbiters only image a small entire globe is incorporated by several numerical weather fraction of Earth’s surface at any time, however, a given storm typically passes through their view only prediction centers into global dynamical models to derive intermittently—perhaps several times a day, with some global meteorological analyses. The process by which the views covering only part of the storm. Polar orbiters observations are incorporated into the models is known as also include scatterometers, which can estimate the data assimilation. This is a process by which all available speed and direction of surface winds; when available, observations are blended with the model itself to produce an 12 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING estimate of the state of the atmosphere (winds, temperatures, (as well as those in other nations, in many cases). Examples pressures, etc.) that is an optimal combination of the of such centers include the National Center for Environmental two.4 Ideally, the resulting analysis should be close to the Prediction (NCEP) in the United States, the European Centre observations wherever high-quality observations of these for Medium-Range Weather Forecasts based in the UK, and meteorological variables exist, while using the model as a the National Center for Medium Range Weather Forecasting substitute where they do not. The meteorological analyses (NCMRWF) in India. The analyses are available to tropical produced by the data assimilation process have the advantage cyclone forecasters. of uniformity, being available on regular three-dimensional grids (although their quality may not be uniform). Such Data assimilation with numerical models has largely replaced analyses are the best estimates of the atmospheric state on a the process of hand analysis, by which a human forecaster large scale and are also initial conditions for global numerical produced a large-scale weather map from raw observations. model forecasts. However, hand analysis is still performed as well at a few forecast centers, including the Bangladesh Meteorological The data assimilation process is not typically carried out at Department and the Darwin office of the Australian Bureau tropical cyclone forecast centers but rather at national or of Meteorology. international numerical weather prediction centers whose products are made available to all national forecast offices 4 China and India do their own data assimilation. WHAT TYPES OF MODELS ARE USED? GLOBAL DYNAMICAL MODELS Global dynamical models are the workhorses of modern Vertical resolutions are much finer—typically hundreds of numerical weather prediction. Global refers to the model meters, perhaps less near the surface—due to the inherently domain, which covers the entire Earth. Dynamical refers to the much smaller vertical scale of the atmosphere compared fact that the models simulate the behavior of the atmosphere with its horizontal scale. A model’s resolution determines the explicitly by solving mathematical equations that express the minimum size of the features it can represent accurately, in laws of physics. much the same way as the pixel size determines the minimum size of an object that can be distinguished in an image from The state of the atmosphere in a dynamical model can a digital camera. be thought of as represented by the numerical values of meteorological fields—such as temperature, wind, pressure, Global dynamical models are run operationally, typically and humidity—on a discrete grid—that is, at a finite set of either two or four times per day every day, by national weather points, usually regularly spaced apart.5 The spacing between prediction centers. Some centers have developed their own grid points is often referred to as the resolution of the model;6 models and run them operationally, such as NCEP (whose typical horizontal resolutions in state-of-the-art global model is known as the Global Forecast System (GFS) model), forecast models at present are in the range 10–50 kilometers. ECMWF, the United Kingdom Meteorological Office, the Japan Meteorological Agency (JMA), Environment Canada, the 5 In practice, many operational global models are spectral, U.S. Navy, the India Meteorological Department and India’s meaning that fields are represented by waves of different NCMRWF, and the China Meteorological Administration. wavelengths rather than grid point values, and the resolution is the smallest wavelength retained. The inherent limitations of finite resolution are the same, and this discussion is phrased in The models are initialized with an analysis. This represents terms of the more intuitive grids in physical space. the best estimate of the atmospheric state at the initial time 6 The use of the word “resolution” to refer to what is actually the (zero lead time) and is produced by assimilating all of the “grid spacing” is technically incorrect, as a model only truly resolves features large enough to span multiple grid points. This available observations into the model, as described earlier. is nonetheless standard usage and is adopted here. An important feature of data assimilation for this purpose Review of Operational Practices and Implications for Bangladesh / 13 is that it reduces inconsistencies between the model and directly from global model output will invariably be biased observations, which may result from errors in either the low for strong storms, although it may be possible for a model or the observations. The results are better suited forecaster (or automated algorithm) to correct for this bias for initializing a model forecast than using an estimate to some extent. based on the observations alone; initialization directly with observations would typically result in an artificial period of All told, the ability of modern global models to simulate rapid change, or “shock,” at early times as the model adjusts tropical cyclones overall—not only the tracks, but also the to the initial conditions. initial formation, or genesis of the storms—is remarkably improved over that of previous generations and allows for Because global models are run on global domains, they can skillful forecasts with rather little human assistance in many be used to inform forecasts of weather occurring anywhere. cases. This is discussed further below. For the most part, forecast data from the models are routinely available to operational centers worldwide, so that a given REGIONAL DYNAMICAL MODELS center has access in real time to the output of global weather A regional dynamical model differs from a global one in that forecast models from most or all of the national centers that it represents only a specific geographic region rather than the produce such output. Thus it is not necessary for each country whole global atmosphere. This means that the model domain to develop its own global modeling system in order to have has lateral boundaries within the atmosphere, at which access to the results from such systems. boundary conditions must be specified. These are taken from a global forecast model. The primary reason for running a The basic outputs consist of physical fields, pressure, regional model is that the smaller domain allows higher temperature, precipitation, and winds on two- or three- spatial resolution, which is highly beneficial for simulating dimensional numerical grids. For the purpose of tropical tropical cyclones. A regional model can have a resolution of a cyclone forecasting, many centers, including NCEP, ECMWF, few kilometers or less. This allows simulation of the strongest and JMA, also run automated tracking programs that identify intensities, allowing at least the potential for more accurate tropical cyclones in the model output. These produce model forecasts of intensity than are possible with global models. tracks—position, intensity, and possibly size information— that can be used directly in forecasting tropical cyclones. This Whereas global models are run operationally every day—for spares the forecaster the task of extracting a model-predicted forecasting of all weather, not just tropical cyclones—regional track from the full three-dimensional meteorological fields prediction models configured specifically for tropical cyclone predicted by the model. forecasting are typically brought into operation only when a center has identified a storm as being of significant concern. The smallest flow features that can be resolved in a dynamical This may happen when a storm has reached sufficient model are several times the grid spacing. Grid spacings of intensity to be named, by whatever criteria are used in tens of kilometers, typical of current global models, are the basin in question, or earlier, when a disturbance has still inadequate for a truly realistic representation of the become a depression and thus presents the possibility of structure of a tropical cyclone, particularly in the inner core intensification to a named storm. The model is often run on a where the eyewall, for example (the ring-shaped region of domain centered on the storm and moving with the storm, to intense precipitation, deep convective cloud, and high winds allow optimal use of computational resources. surrounding the clear eye in a strong tropical cyclone), can have a width of only a few kilometers. As the strongest winds Some older regional models use simplified forms of the typically occur over a small region in or close to this inner governing equations and sets of model choices designed core, a lack of adequate spatial resolution to resolve that core specifically for predicting tropical cyclones. Examples typically leads to the simulation of maximum sustained wind include representing the atmosphere as a single layer with no speeds weaker than those observed. As a result, the current vertical structure, the so-called barotropic model; as only a generation of global weather models is still largely incapable few vertical layers, as in the quasi-Lagrangian model (Mathur of producing model tropical cyclones with intensities at the 1991); or as the beta and advection model, in which the higher end of those observed. Intensity estimates produced storm moves with the large-scale flow plus a deviation due 14 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING to the effect of Earth’s rotation and sphericity (Marks 1992). interpretation of all available information. If aircraft or other The large-scale flow at the boundaries and in the initial specialized observations are available, these may be used conditions can be taken from a global weather prediction to construct more accurately constrained initial conditions. model. An idealized vortex whose properties are based on Recent research has demonstrated positive impacts on both those deduced in real time by forecasters to represent a given track and intensity forecasts from assimilation of aircraft-based tropical cyclone is superimposed on the large-scale initial Doppler radar data, for example in Hurricane Katrina in 2005 conditions. These models were developed in a period when (Weng and Zhang 2012) and Typhoon Jangmi in 2008 (Zhang the more comprehensive dynamical models either could not et al. 2012), although this remains an area of active research. represent tropical cyclones well at all or could not be run quickly enough, given available computational resources, ENSEMBLE METHODS to be available to the forecast process. Although some are Numerical forecast models are run in two different modes: still in operational use, these models have been supplanted deterministic and ensemble. A deterministic forecast run is to a large extent by more comprehensive models that use a single run of the best version of a given model, typically much more complete and sophisticated representations of meaning at the highest spatial resolution currently available, atmospheric physics, as model improvements and greatly starting from the best estimate of the initial state. An increased computer power have made these comprehensive ensemble is a set of runs, usually with lower spatial resolution or “full-physics” models more competitive. than used for the deterministic forecast run. Each run in the ensemble starts from slightly different initial conditions. A widely used full-physics regional model has been NOAA’s Adding a different small perturbation to the best estimate of Geophysical Fluid Dynamics Laboratory hurricane model. This the initial state produces each set of initial conditions. These is run not only for the U.S. NHC to forecast Atlantic and Eastern small perturbations are still consistent with that estimate, Pacific storms; the U.S. Navy’s Fleet Numerical Meteorology within the uncertainties inherent in the observation, analysis, and Oceanography Center also runs a version of this model for and data assimilation process. As the model runs progress, most named tropical cyclones in other basins worldwide. Thus the differences between ensemble members will grow, forecasters in every basin can, at least in principle, have access because the atmosphere is chaotic. This means that small to results from at least one regional high-resolution model. The differences in initial conditions amplify rapidly, leading to Hurricane Weather Research and Forecast System (HWRF), a large differences after a finite time (Lorenz 1963). newer model, is used operationally at the U.S. National Center for Environmental Prediction and the India Meteorology The ensemble spread, or range of solutions among the Department. HWRF output is also provided globally, including ensemble members, can be used as a practical measure of the for the Bay of Bengal, to the Joint Typhoon Warning Center uncertainty in the prediction. The spread almost invariably (JTWC) by the NCEP as an experimental product under the grows with lead time, indicating increasing uncertainty Hurricane Forecast Improvement Program (HFIP).7 at longer lead times. However, the spread also varies from forecast to forecast at the same lead time. This indicates that Initialization of regional weather models is a more difficult the uncertainty at a given lead time is not a unique function problem than initialization of global models because regional of lead time. Rather, it is to some extent situational—some models involve fine spatial scales. While the purpose of a forecasts will have a greater degree of uncertainty than regional weather or tropical cyclone model is to capture the others, independent of the quality of the model or the skill of structure of the storm in fine spatial detail, the available the forecaster who is interpreting it. observations are often inadequate to represent that level of detail at the initial time. This strongly degrades the forecast Tropical cyclone forecast centers typically have access of those structural details at future times. Standard analyses to output from numerical model ensembles produced by from global models are usually too coarse to resolve the inner several national weather forecast centers globally. Each core of the storm. It is common to produce a bogus, or artificial, may contain several to perhaps tens of individual ensemble vortex into the initial conditions, based on the forecasters’ members. These ensemble outputs in many cases include See http://www.emc.ncep.noaa.gov/HWRF/WestPacific/index. 7 tropical cyclone tracks produced by automated tracking html. algorithms applied to the model output fields. In addition Review of Operational Practices and Implications for Bangladesh / 15 to these products from individual forecast centers, the motion at its present speed and direction. Statistical models forecaster has access to consensus forecast aids, which are that are more complex than persistence have been developed produced by taking weighted averages of results from a set to predict both the tracks and intensities of tropical cyclones. of different models. Examples are the Florida State University Because dynamical models have become so skillful at predicting Superensemble described by Krishnamurti et al. (1999) tracks, statistical track models are for the most part no longer and consensus products used at NHC and the Joint Typhoon useful, except as benchmarks against which to measure skill— Warning Center listed by Heming and Goerss (2009). These that is, a forecast that is no better than the statistical models consensus products are typically found, on average, to may be considered not to have useful skill. For the purpose of be superior to any individual model in the ensemble (for predicting intensity, statistical models are still considered to example, Goerss, Sampson, and Gross 2004), an indication be at least the equal of dynamical models, because there has that the differences between models can be considered been relatively little improvement in the ability of dynamical random errors to some extent. models to predict intensity in recent decades. However, statistical models have some well-known limitations, such as STATISTICAL MODELS their inability to predict rapid intensification periods, when the intensity of a tropical cyclone increases by 30 kt or more over a A statistical model is one that predicts an event based on the period shorter than 24 hours. statistics of past events. A set of predictors is chosen that is believed a priori to be relevant; in the case of statistical STATISTICAL-DYNAMICAL MODELS models for tropical cyclone prediction, predictors could be environmental variables such as sea surface temperature, Statistical-dynamical models are, as the name suggests, vertical wind shear (the difference in speed and direction hybrids of the two approaches. These are statistical models of the horizontal wind at different altitudes), or parameters in that tropical cyclone characteristics are predicted as describing the present state of the storm. The predictors specified functions of larger-scale environmental variables, representing the environment are typically averaged over a but dynamical in that the large-scale environmental variables region larger than the storm itself. Empirical relationships are obtained at future times from dynamical model output. are derived between the predictor variables at the initial The statistical component attempts to overcome the time and the behavior of storms at later times, using deficiencies of the dynamical model in representing the statistical methods such as linear regression. These models tropical cyclone, while the dynamical component makes use do not use the equations of motion or any other aspect of of the skill that the dynamical model does have in predicting atmospheric physics, except inasmuch as understanding the the large-scale environment in which the storm is evolving. physics informs the choices of predictors. As a consequence, Important examples are the Statistical Hurricane Intensity statistical models are many orders of magnitude cheaper Prediction Scheme (DeMaria and Kaplan 1994, 1999; DeMaria computationally than dynamical models. et al. 2005), Logistic Growth Equation Model, and the Rapid Intensity Index. These models remain competitive with The simplest statistical model, for example, is persistence: dynamical models in prediction of tropical cyclone intensity the storm will maintain its present intensity and continue its and provide important guidance to forecasters. THE FORECAST PROCESS THE TRADITIONAL PROCESS, WITH A HUMAN FORECASTER the atmosphere everywhere on a global or at least regional As carried out in most national tropical cyclone forecast scale.) The analyst will use all available observations for this centers, the forecast process involves one or more human purpose. Many types of observations are available only at a forecasters. The first component is a tropical cyclone analysis, limited set of times or locations—for example, if the cyclone meaning an estimate of the state of an existing tropical happens to pass over a ship or buoy or when a polar orbiting cyclone at the present time. (This is not to be confused with satellite swath happens to capture it. Even when available, a global or synoptic analysis, which represents the state of surface in situ observations or radiosondes are available at a 16 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING small number of locations and thus would poorly define the 120 hours. The forecaster consults all available guidance—a structure of the storm on their own. Aircraft observations are generic term used here to refer to any external information, valuable but are available only in a subset of basins and then but typically referring to the results from dynamical or for a subset of storms in those basins. Geostationary satellite statistical prediction models. imagery is virtually guaranteed to be available and as a consequence is essential to modern tropical cyclone analysis. Ideally, all guidance is available to the forecaster in real time and in a form that makes the process of interpreting that The Dvorak technique (Dvorak 1984) is the standard method guidance and synthesizing it into a forecast as straightforward for using geostationary imagery to estimate the intensity of as possible. In the more advanced centers, the process is the storm and the location of its center. This involves several done on digital computers. The software environment in steps that require pattern recognition on the part of a human which both observations and model guidance are presented analyst. It is thus partly subjective, and different analysts can to the forecaster is a significant component of the forecast obtain somewhat different results. Besides differences from system. At the NHC and JTWC, the Automated Tropical Cyclone one individual to the next, there are systematic differences in Forecasting System (ATCF) is used to display model guidance, the application of the Dvorak technique from one forecasting while observations are displayed primarily using the NCEP center to another, which can lead to inconsistencies in analyzed Advanced Weather Interactive Processing System (ATCF also storm position and intensity for a given storm in real time. has a limited capability to display satellite data). All model track predictions, for example, are displayed as lines on a Automated implementations of the Dvorak technique have map on the screen in ATCF. The system is also able to correct been developed that are objective. One such implementation, biases due to initial position error in model guidance by the Advanced Dvorak Technique (ADT), has skill comparable relocating the start of each track to the center fix determined to that achieved by human analysts (Olander and Velden by the forecaster. After also consulting three-dimensional 2007). ADT is produced for storms in all basins (including model output fields in order to interpret the model tracks, the the North Indian Ocean) and made publicly available by the forecaster produces a forecast by drawing a track on the same Cooperative Institute for Meteorological Satellite Studies at screen on which the guidance tracks appear. the University of Wisconsin, Madison. Nonetheless, human analysts are still prevalent in operational practice worldwide. Faced with all available guidance, the forecaster makes a judgment about which guidance to trust more than others At the start of the forecast, the forecaster uses his or her or how to average or interpolate between different guidance judgment to assess all available observations, including products. The forecaster may be guided by intuition, physical a Dvorak analysis, to produce an estimate of the current insight, experience with previous storms, or other factors. position and intensity of the storm, as well as wind radii. This It is not impossible that expectations about users’ potential is the analysis, or forecast, at zero lead time. Such an analysis reactions to the forecast may be taken into account in some is produced for any storm that has already been determined circumstances—for example, the risk of forecasting an event to be a tropical cyclone (by the criteria in use in the particular more severe than actually occurs may be perceived differently basin) but it may also be produced for a weather system that than the risk of forecasting one less severe than actually is just approaching that threshold, in order to determine if it occurs—although the goal is forecasts with zero bias. has become a tropical cyclone. There may also be a desire for consistency from one forecast For each tropical cyclone, the forecaster must next produce to the next. This is relevant either because the guidance a forecast for each lead time. The lead times are typically changes rapidly between forecast cycles or simply because spaced 6 or 12 hours apart and extend to a maximum, typically each forecast may be made by a different forecaster or For each tropical cyclone, the forecaster must next produce a forecast for each lead time up to the maximum. The lead times are typically spaced 6 or 12 hours apart and extend to a maximum, typically 120 hours. Review of Operational Practices and Implications for Bangladesh / 17 group of forecasters, and one set may judge the available ensemble prediction systems for tropical cyclone prediction guidance differently than the next one. Even in the presence have emerged in the last several years. Once an ensemble of such impetus for a change in the forecast, the forecaster prediction system of this kind is set up, it can, in principle, be may decide not to alter the forecast as rapidly as he or she entirely automated.8 No human forecaster is needed, strictly otherwise might, in order to maintain consistency. This speaking, to produce the forecast, although a human forecaster is viewed as important to avoid confusing the users with may be involved, in practice, for quality control. The JMA, a forecasts that oscillate or otherwise vary from one cycle to national forecast center and one of the WMO regional centers the next and to average out short-term variations in guidance for predicting tropical cyclones, operates such a system (in that may occur over time. A policy of maintaining continuity parallel with a more traditional forecast process involving over consecutive forecast cycles—apart from any clear reason human forecasters); the ECMWF does as well (although ECMWF for a dramatic change—is a strong constraint on forecasters forecasts are not publicly available but are provided only to at the NHC in Miami, for example (James Franklin, personal certain national centers and other clients). communication). In this process, as it is implemented in many if not most of These systems use numerical models exclusively. A typical the forecast centers, information from numerical guidance system uses a single numerical model that is run at a single is invariably used to inform the forecast, but it may not be center and in ensemble mode and uses objective algorithms explicitly presented as part of the forecast. The forecaster will to produce a forecast directly from the model output. An almost certainly look at the track predictions in the model objective tracking algorithm is used to detect and track ensembles—ensembles constructed from multiple runs of tropical cyclones in each model run. Once one or more a single model with different initial conditions as well as tropical cyclones are detected, special ensemble members multimodel ensembles consisting of runs of independent may be initialized, using a procedure designed to optimize models—and the degree of spread will inform the perception the perturbations in the initial conditions specifically for of the uncertainty. However, many centers do not explicitly the purpose of estimating the uncertainty in the prediction use the ensemble spread to compute their quantitative of those tropical cyclones. The tracking algorithm is applied metrics of uncertainty (for example, the cone of uncertainty) to the resulting model integrations to produce an ensemble or to show the tracks from the ensembles in their forecast of tracks, and then objective algorithms are used to produce products. In the case of the NHC, at least, this is the result of forecast quantities from those tracks. a conscious decision; it is believed that many users are not capable of understanding the ensemble information and that The forecasts produced by ensemble systems lend themselves incorporating it would make the forecast less effective (James to probabilistic interpretation. The probabilities are derived Franklin, personal communication). directly from the ensemble model tracks. For example, the probability of wind speed greater than 65 knots at a given Some formal products do use ensemble information; in the case of NHC, ensemble spread is used to construct the location can be estimated by dividing the number of ensemble wind speed probability product. More broadly, information members in which the wind speed exceeds that value by the about ensemble spread may also enter the forecasts in less total number of ensemble members. These probabilities are formal ways. At the RSMC New Delhi, the cone of uncertainty inherently situational and dynamic; the range of possible is constructed based on the past climatology of errors, as forecast outcomes will depend on the ensemble spread at NHC. In the text bulletin, however, the area of possible (for example, Dupont et al. 2011). Visual inspection of the landfall is indicated taking into consideration the spread in “spaghetti plot”—the set of tracks from individual ensemble deterministic models and ensembles. members—gives a qualitative impression of the uncertainty in the track forecast, as shown in an example from ECMWF in ENSEMBLE PREDICTION SYSTEMS Figure 2.2 (Vitart et al. 2011). As the ability of global dynamical models to simulate tropical 8 It is also possible to develop an automated forecast system cyclones accurately has improved, and as ensemble methods based on single model runs with no ensemble information, al- for using those models in forecasting have advanced, new though we are not aware of such a system currently operational. 18 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Figure 2.2 Individual Tracks for Tropical Cyclone Irene Date 20110822 00 UTC @ECMWF Individual trajectories for IRENE during the next 240 hours tracks: thick solid=OPER; thick dot=CTRL; thin solid=EPS members [coloured] 0-24h 24-48h 48-72h 72-96h 96-120h 120-144h 144-168h 168-192h 192-216h 216-240h Source: Vitart et al. 2011, courtesy F. Vitart, ECMWF. From the ECMWF Ensemble Prediction System forecast starting at 00 UTC on August 22, 2011, using various colors to indicate tracks over 24-hour intervals out to 240 hours. Note: Black lines represent the tracks produced by the model at two different horizontal resolutions, T1279 (solid) and T699 control (dot). Ensemble forecast systems might predict quantities different storms but also for storms that have not yet been named but from those in traditional forecasts. As an example, in addition that some ensemble members predict will form. to producing an ensemble mean track prediction (the average of all tracks for a given storm over the ensemble) for storms Figure 2.3 shows an example of an ECMWF strike probability that have already been named, the JMA and ECMWF’s forecast for a storm (Irene in 2011) that had already been ensemble prediction systems predict a strike probability—the named. Figure 2.4 shows a strike probability forecast issued probability that a tropical cyclone center will pass within a two and four days before the formations of Harvey and Irene, specified distance of a given location—not only for named respectively (both from Vitart et al. 2011). The latter figure Review of Operational Practices and Implications for Bangladesh / 19 Figure 2.3 Strike Probability (%) for Tropical Cyclone Activity within 300 Kilometers (Systems with Maximum Wind Speed Greater Than 8 Meters per Second) for the Two-Day Period of August 20–22 Based on the ECMWF Ensemble Prediction System Forecast from 00 UTC on August 17, 2011 Date 20110822 00 UTC @ECMWF Probability that IRENE will pass within 120km radius during the next 240 hours tracks: solid=OPER; dot=Ens Mean [reported minimum central pressure (hPa) 994 ] n 5-10 n 10-20 n 20-30 n 30-40 n 40-50 n 50-60 n 60-70 n 70-80 n 80-90 n > 90 Source: Vitart et al. 2011, courtesy F. Vitart, ECMWF. Note: Solid lines and open circles represent the observed tracks of tropical storm Harvey (to west) and hurricane Irene (to east) between August 20 and August 24. Harvey formed on August 19, Irene on August 21—two and four days after this forecast was produced. thus is a forecast simultaneously of genesis and the track of in particular, IMD uses a 28-knot threshold, while some the storm (in this case, two storms) after genesis. This product other centers use a 34-knot threshold, and NHC does not use is qualitatively different from any produced by the traditional any wind speed threshold but instead criteria based on the forecast process, in which forecasts of genesis and of storm organization of the system’s convection and circulation— evolution after genesis are kept completely distinct, and the ensemble-based products such as that shown in Figure 2.3 latter are not produced at all until genesis has occurred. do not require any specific criteria to have been reached at While the point at which track and intensity forecasts are the time the product is issued or for any disturbance to exist begun for existing disturbances differs at different centers— at all yet. 20 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Figure 2.4 Strike Probability (%) for Tropical Cyclone Activity within 300 Kilometers (Systems with Maximum Wind Speed Greater Than 8 Meters per Second) for the Two-Day Period of August 20–22 Based on the ECMWF Ensemble Prediction System Forecast from 00 UTC on August 17, 2011 Source: Vitart et al. 2011, courtesy F. Vitart, ECMWF. Note: Solid lines and open circles represent the observed tracks of tropical storm Harvey (to west) and hurri- cane Irene (to east) between August 20 and August 24. Harvey formed on August 19, Irene on August 21—two and four days after this forecast was produced. CURRENTLY AVAILABLE SKILLS AND RECENT DEVELOPMENTS This section briefly summarizes the level of skill that time; note that forecasts for 96 and 120 hours were not is currently available operationally and the trends in made before 2001. skill in recent decades. The U.S. NHC is taken as broadly representative of the state of the art. Figure 2.5 shows It is evident from Figure 2.5 that the accuracy of track average annual track errors in the NHC official forecasts forecasts has improved steadily with time. The accuracy of (in nautical miles; 1 nautical mile = 1.85 kilometers) as a today’s 72-hour track forecast, for example, is comparable function of time over the last two decades, while Figure 2.6 to that of a 24-hour forecast of 20 years ago. At the same shows average annual intensity errors over nearly the same time, Figure 2.6 shows that there has been little improvement period. Different curves are shown separately for each lead in intensity forecasts. The accuracy of intensity forecasts is Figure 2.5 Average Track Errors, in Nautical Miles, in the U.S. National Hurricane Center’s Forecasts as a Function of the Year in Which the Forecasts Were Made Source: From http://www.nhc.noaa.gov/verification. Note: Different curves show different lead times, as indicated in the legend. Review of Operational Practices and Implications for Bangladesh / 21 Figure 2.6 Average Track Errors, in Knots, in the U.S. National Hurricane Center’s Fore- casts as a Function of the Year in Which the Forecasts Were Made Source: From http://www.nhc.noaa.gov/verification. Figure 2.7 Track Errors at 48 Hours Lead Time (Nautical Miles), as a Func- tion of Year Source: From http://www.nhc.noaa.gov/verification. Note: From model guidance. Each symbol represents a different model, as indicated in the legend. indistinguishable now from what it was 15 or 20 years ago, fact that their own ability to improve accuracy is decreasing particularly at lead times of 72 hours or less. to the point that they soon may not be able to add any value to the model tracks (for example, James Franklin, NHC, and Much of the improvement in track forecasts is clearly Andrew Burton, Australian Bureau of Meteorology, personal attributable to the improvement in numerical model guidance. communications). Figure 2.7 shows the trends in track forecast errors at 48 hours lead time since 1970 from various individual numerical Intensity is a different issue. The lack of improvement in models. The downward trend in the forecast errors since intensity forecasts is widely recognized as a major scientific 1990 shown in Figure 2.6, with a decrease from about 200 challenge. The contrast between track and intensity forecasts nautical miles in 1990 to about 100 nautical miles in 2005, is clearly traceable to the contrast between the accuracy of closely matches the downward trend in the models shown the model guidance of track and that of intensity. Numerical in Figure 2.7. Forecasters at modern centers are quite frank model forecasts of intensity have not improved at the about the tremendous improvement in the models and the rate that track forecasts have. In contrast, statistical and 22 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING statistical-dynamical models of intensity remain comparable million a year and the following goals, with the 5-year mark in accuracy to the best dynamical models. set at 2014 and the 10-year mark at 2019 (Gall et al. 2013): The reasons for this are not entirely clear. The quality of the Reduce numerical forecast errors in track and intensity models is one issue; resolution (and thus computing power), by 20 percent in 5 years and by 50 percent in 10 years model physics, and numerical implementation are likely all for forecast days 1–5 important to some degree. Initialization is also important. Extend forecast skill to 7 days, with skill equivalent to Even a very good model will not give an accurate forecast that of the 5-day forecasts when they were introduced if the initial conditions are not correct, and this depends on in 2003 accurate observations of the storm. Whereas track depends mostly on the large-scale wind field, intensity appears to Increase probability of detecting rapid intensification depend to a significant extent on the internal structure of to 90 percent at day 1 and 60 percent at day 5, and smaller-scale features in the storm’s inner core. Defining that decrease the false alarm ratio for rapid intensification structure requires high-quality observations near the storm to 10 percent at day 1 and 30 percent at day 5. center, such as can be obtained by aircraft. Although intensity prediction is not the only target of HFIP, it is clearly the central difficulty—and the one that has At a more fundamental level, however, it is not entirely largely motivated the program. This program is sponsoring clear how predictable tropical cyclone intensity is, even much work in numerical modeling, observations, and data in principle. To the extent that it depends on the detailed assimilation. As one example, Doppler radar observations internal structure of the storm, it can be expected that the taken from aircraft are now being assimilated operationally structure may evolve chaotically, in the sense of sensitive into a high-resolution, limited-area, dynamical model, namely dependence on initial conditions and on time scales shorter the Hurricane Weather Research and Forecast model; these than those on which the large-scale flow does. The large- computations show the promise of substantive improvement scale flow is understood to be potentially predictable out in intensity and structure forecasts (for example, Zhang et al. to time scales of a week or two (Lorenz 1963). It is possible 2011; Aksoy et al. 2012). that at least in some situations, a change in tropical cyclone intensity cannot be predicted—even in principle and even In a recent publication, Gall et al. (2013) state that the HFIP with a perfect model—for more than, say, one day. That may is on target to meet its goals by the stated timelines. Landsea be an overly pessimistic speculation, but the question of and Franklin (2013), in contrast, argue that the improvements potential predictability of tropical cyclone intensity has not in intensity targeted by HFIP may be difficult to reach simply been addressed well enough yet to allow confident statements because of limitations in the accuracy (now and in the near about how well intensity might ever be predicted. future) of observational estimates of storm intensity. In 2010 NOAA started the Hurricane Forecast Improvement Program. This is a major research effort with a budget of $25 STORM SURGE FORECASTING Although intensity prediction is not the only target of HFIP, it is clearly the central difficulty—and the one that has largely motivated the program. This program is sponsoring much work in numerical modeling, observations, and data assimilation. Review of Operational Practices and Implications for Bangladesh / 23 WHAT IS A STORM SURGE FORECAST? Storm surge is the temporary elevation of sea level due to meteorological disturbance. A tropical cyclone can induce storm surge via wind stress on the sea surface, reduction in land moves inland during a surge event. The extent of flooding atmospheric surface pressure, and transfer of momentum on normally dry land, both vertically and horizontally—that from breaking waves into the mean current. Storm surge is, where flooding occurs and how high the water is above is defined as the difference between the mean water level ground at those locations—is known as inundation. over some period of time (at least long enough to average out waves) and what it would otherwise be due only to the A storm surge forecast is a prediction of the storm surge. In astronomical tide. The total mean water level due to tide and practice, a storm surge forecast may be a prediction of the surge combined is the storm tide. surge only along the coast or may include a prediction of inundation. Like a tropical cyclone forecast, a storm surge Storm surge is normally defined by the vertical rise in water forecast may be deterministic or probabilistic. level at the coast—or more properly, at the location where the coast normally is, since the boundary between water and dry WHAT TYPES OF OBSERVATIONS ARE USED? A storm surge forecast is driven primarily by a meteorological mathematical representation of a physical process occurring forecast of wind and pressure fields as functions of time, by dynamically in time, as in the case of a meteorological model observational data defining the solid surface over which water or storm surge model, but simply a static data set in a digital flows, and by data defining the elevation and flow of water format). High-quality topographic data may be obtained, for at the lateral (side) boundaries of the numerical model used example, from aircraft surveys using lidar (laser imaging to produce the forecast. The meteorological forecast comes detection and ranging). either from dynamical or statistical models or from human forecasters’ synthesis of those models, as described earlier. Unlike other observations that enter a tropical cyclone forecast or storm surge forecast, bathymetric and topographic The height of the solid surface over which water flows is data can be considered static over a relatively long period of called bathymetry when the surface is below the water time because the solid surface does not change rapidly, as the in normal conditions and topography when the surface is atmosphere and ocean do. These data can be gathered once normally on dry land. Bathymetric and topographic data are and then reused in all forecasts, although they may need to be essential inputs to storm surge forecasts. Bathymetry has a updated periodically as the landscape evolves over time; the strong role in controlling the height of the storm surge for topographic data for New Orleans, for example, are updated given meteorological conditions. Grossly, a shallower bottom once a year (J. Rhome, personal communication). over a larger shelf leads to a higher surge (see, for example, Dube et al. 2009). Topography has an equally strong role in A specification of the water elevation and currents of the determining which regions will be inundated for a given surge ocean, including those due to the tides, is needed at the lateral along the coast. Not only is the natural topography relevant, but built structures—including flood control structures such boundaries of a storm surge model. This can be taken from as levees, dikes, and polders—are as well. a larger-scale ocean model in which a storm surge model is embedded. Such a larger-scale model may assimilate ocean Bathymetric and topographic data are needed at high spatial observations in addition to being driven by meteorological resolutions—much higher than those at which meteorological forcings. Some more-advanced models, which treat the fields are ever required (or available). Accurate high-resolution potentially inundated land surface explicitly as well as the sea, topographic data in gridded form are often referred to as a may also consider river inflows and precipitation in the domain, digital elevation model (the word model here does not imply a so observations of these quantities may be incorporated. 24 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING WHAT TYPES OF MODELS ARE USED? A storm surge model is a model of the flow of the ocean, given from any direct effect of waves on structures on land). A similar atmospheric and oceanic inputs as described above. Like the distinction holds with tides. If a model does not explicitly dynamical atmospheric models used for weather prediction, include the tides, the total storm tide can be estimated by storm surge models are dynamical models that represent the simply adding the known astronomical tide to the simulated evolution of the fluid by solving the mathematical equations storm surge. This may be accurate in many cases, but there is of motion in an approximate form on a grid. Being specifically also the possibility of nonlinear interactions between the tide designed to predict storm surge, these models make specific and the surge that can change the surge itself. Models that assumptions and approximations that are appropriate for include the tide explicitly can account for such interactions that purpose and different from those used in other sorts of on a physical basis. It is also possible to use such models ocean models. to develop parameterizations of the nonlinear surge-tide interaction so that it can be taken approximately into account Many storm surge models use so-called shallow water in simulations that do not explicitly include the tide (see, for equations, which approximate the ocean as a fluid with example, Lin et al. 2012). a constant density that moves with the same horizontal velocity at all depths. The state of the fluid is represented by Another model choice that may have some quantitative velocity of the water and height of the water surface, both of impact is the bulk drag coefficient used in the computation which are two-dimensional fields because of the assumption of momentum transfer from the air to the sea. For a given of uniformity of the velocity with depth. Some models use the wind speed, this coefficient determines the force exerted on full three-dimensional equations of motion rather than the the ocean. The drag coefficient is a consequence of a range shallow water equations and thus represent the variations of physical processes, including interaction of the wind of current with depth. This increases computational cost with waves and sea spray, and varies as a function the wind substantially and is believed to yield only minor improvements speed itself. There is considerable uncertainty in the value in accuracy. A range of grids and numerical methods are used of the drag coefficient, especially at high wind speeds. This for solving the dynamical equations. Higher resolution is uncertainty translates into uncertainty in the surge. often used closer to the coast, to focus computational power where it is most needed to resolve the surge. Surface winds are specified in two different ways in storm surge models for operational forecasting. First, winds can One important feature of a model for these purposes is the be taken directly from a meteorological forecast model. In nature of the boundary condition at the coast. Some models this case the winds are specified as functions of horizontal treat the coast as though it were a wall, impermeable to the position and time, at the grid spacing of the meteorological surge. Such models are incapable of simulating inundation model. This has the advantage of not imposing a priori explicitly. Other models allow for the wetting of dry surfaces assumptions about the structure of the storm, and it allows and the inland migration of the water line and thus explicitly for an arbitrarily complex wind field. On the other hand, simulate inundation. the surge model must commit to that particular numerical model rather than to the tropical cyclone forecast, which is Another set of distinctions of model types is between those produced (typically by a human forecaster) after considering that explicitly account for waves and those that do not, as multiple models and other information. well as those that explicitly account for tides and those that do not. Waves can transfer momentum to the mean flow Alternatively, winds can be derived from the much more of water and thus influence the surge level. This can be a limited set of tropical cyclone forecast parameters— quantitatively significant effect in some cases (and is distinct including track, intensity, size, and speed of forward Many storm surge models use so-called shallow water equations, which approximate the ocean as a fluid with a constant density that moves with the same horizontal velocity at all depths Review of Operational Practices and Implications for Bangladesh / 25 motion—and then have internal parametric models carried out at a tropical cyclone forecast center, and the determine the distributions of wind from those using computational resources available for predicting storm surge predetermined functional relationships. This has the may be substantially less than those used for running global disadvantage, in principle, of having the parametric models dynamical models at the major numerical weather prediction impose a priori assumptions about the structure of the centers. wind field. An advantage of this method, however, is that the parameters it requires are those that are produced in As a result, some of the more sophisticated models used a typical tropical cyclone forecast, so that the storm surge for storm surge research may not be practical for use in model can take the forecast itself directly as input. Another forecasting because of their higher computational cost. The advantage of the parametric approach is that the number SLOSH (sea, lake, and overland surges from hurricanes) model, of data that must be specified is much smaller than if a for example (Jelesnianski, Chen, and Shafer 1992), used by meteorological forecast model is used. Thus it is simpler the U.S. National Hurricane Center (Glahn et al. 2009), is to perform a wide range of storm surge simulations (either among the simplest available: it does not include waves or in real time, for forecast purposes, or in research mode) to tides, and it has a no-normal-flow lateral boundary condition allow for variations in track, intensity, and so forth, as it is at the coast so that it cannot explicitly simulate inundation. simpler to specify variations in those few parameters than it However, it is computationally light. This allows simulations is to specify variations in the entire wind field. to be completed quickly, so that, for example, an ensemble of simulations can be done during a six-hour forecast cycle Use in operational forecasting requires that computations to generate a probabilistic forecast, as described in the next be completed quickly. The computations are likely to be section. THE FORECAST PROCESS The process of forecasting a tropical cyclone-induced storm low tide), or the level above the mean tide (approximately the surge is closely coupled to the process of forecasting the midpoint between high and low tide). The National Hurricane tropical cyclone itself. The surge forecast requires as input Center forecasts water heights as a range of values above a forecast of tropical cyclone winds, either directly from a ground, presumably referring to ground at the coast. meteorological model or from a forecast of track, intensity, and size parameters. These data are then input to a storm An inundation forecast is a specific prediction of water height surge model, which generates a prediction of surge. If tides above ground at specific locations—for example, expressed as are not included in the model, the total peak storm tide a map. If the forecast is produced by a model that explicitly forecast depends on the time of peak surge relative to the predicts inundation (wetting of dry surfaces), then an high tide, so the timing of the peak surge must either be inundation forecast can, in principle, be produced directly predicted accurately or give a range of possibilities. from model output. Otherwise, an inundation forecast can be produced separately. The NHC has recently developed A forecast of storm surge or storm tide may be, strictly, an inundation forecast, made public on an experimental just a forecast of water rise at the coast (and perhaps also basis in 2014. It is based on an ensemble of SLOSH model at locations seaward of the coast). There is potential for forecasts of water rise along the coast, combined with a high- confusion on the part of users about the reference level above resolution topographic data set or “digital elevation model” which water rise is being forecast. Choices include a forecast (DEM), to generate a map of water elevations above ground, of just the surge (water level above the astronomical tide at based on assumptions broadly similar to but somewhat more any given time) or of total storm tide. If storm tide, the latter sophisticated than assuming constant water surface height may be expressed as an increase above a fixed specific level above a fixed datum (J. Rhome, personal communication). (datum), the mean level of low water (essentially the average 26 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING TREATMENT OF UNCERTAINTY Because the storm surge forecast depends strongly on the forecasts applied to a storm surge model. The ensemble of wind forecast, it inherits errors associated with the latter. wind forecasts can be obtained directly from an ensemble Additional errors result from the uncertainties inherent in the of meteorological model forecasts, if the surge model takes storm surge model and possibly from the other input data, forecast winds directly as input. If (as appears to be more such as bathymetry, topography, or oceanic lateral boundary typical) the surge model takes forecast track, intensity, and conditions. The storm surge forecast can thus generally be wind radii as inputs, then an ensemble of these parameters is expected to be more uncertain than the tropical cyclone needed. This can be generated by perturbing a deterministic forecast itself. The treatment of this uncertainty and its forecast in some way, for example (as done at the NHC) communication to users are critically important. One response using a set of perturbations based on the range of historical to the additional uncertainty in storm surge forecasts may forecast errors. An ensemble of tracks could also be obtained be a reduction in the maximum lead time at which forecasts from a numerical model ensemble after application of a are produced. The NHC, for example, produces storm surge tracking routine. Once an ensemble of surge forecasts has forecasts out to 72 hours, although tropical cyclone forecasts been generated, it can be expressed, for instance, as a set are produced to 120 hours. of probabilities that a given value of surge will be exceeded, as in NHCs Probabilistic Hurricane Storm Surge (p-surge) As with forecasts of tropical cyclones (or any other product.9 Verbal advisories may state a range of possible phenomenon, for that matter), uncertainty in storm surge forecasts may be addressed by producing forecast products values. that are explicitly probabilistic. This is most straightforwardly 9 See http://www.nws.noaa.gov/mdl/psurge2.0/about. done through ensemble methods. An ensemble of surge php?S=Karen2013&Adv=10&Ty=e10&Z=m1&D=agl&Ti=incr&Ms- forecasts can be generated from an ensemble of wind g=1&Help=about An aerial view of damage to villages and infrastructure following Cyclone Sidr. Credit: U.S. Marine Corps photo by Sgt. Ezekiel R. Kitandwe (RELEASED) Review of Operational Practices and Implications for Bangladesh / 27 THE CIFDP APPROACH The Joint WMO-Intergovernmental Oceanic Commission’s system set up for the Hawaiian island of Oahu (Taflanidis et Technical Commission for Oceanography and Marine al. 2012). In real time, the cyclone forecast parameters could Meteorology and the WMO Commission for Hydrology be used to sample or interpolate from this library of results have together initiated the Coastal Inundation Forecasting to obtain a storm surge prediction that corresponds to those Demonstration Project (CIFDP), “in order to meet the parameters. This is then repeated with a range of variations challenges of coastal communities’ safety and socioeconomic in the cyclone forecast parameters to generate a probabilistic sustainability through the development of coastal inundation forecast. Because the library was pre-computed, the storm forecasting and warning systems at the regional scale” surge model itself does not have to be run in real time. (JCOMM 2013). The project recommends a specific set of practices for developing storm surge and inundation forecast The advantage of this approach is that by decoupling the systems and offers technical assistance to countries in intensive computation from the forecast process, it removes implementing those systems. This section describes briefly the requirement that the model be computationally inexpensive some aspects of the approach recommended by CIFDP (D. enough to run in real time. A more sophisticated and realistic Resio, personal communication). model can be used for the surge simulations than would be feasible in real time. Further, the intensive computation to CIFDP does not recommend running a dynamical storm surge produce the library does not have to be done in the same model in real time. Rather, it recommends performing a large place or by the same group of people as those operating the number of simulations with such a model just once, in advance forecast system in real time. Although there is some error of any forecast operations, to produce a library of simulation associated with the use of a surrogate model to interrogate the results that spans the range of conditions believed to be precomputed database for a given scenario, relative to direct possible at the location of interest. A set of 350 simulations, computation with the dynamical surge model for that same with a wide range of tropical cyclone track, intensity, and size scenario, that error is modest and is included in the computed parameters, was judged to be adequate for a demonstration uncertainty in the forecast (Taflanidis et al. 2012). 28 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING BIG PICTURE This section considers the relative merits of different at the International Workshop on Tropical Cyclones held in automated ensemble forecast systems based directly on Manila in 1982 indicated that they had access to guidance global model ensembles versus traditional forecast systems from global models. that use a wider range of guidance and a greater role for the human forecaster. Today it is safe to assume that every country desiring it has some access to guidance from global models. A great deal The practice of tropical cyclone forecasting, in something like of such guidance is available freely on the Internet,10 and more its modern form, dates back to the 1950s, as practiced in the is available through agreements between national centers. United States. During that period, aircraft reconnaissance The quality of this guidance has improved dramatically, began to be routinely done, and the first numerical weather particularly for track and genesis. The global dynamical prediction models came into operation. At least until the models from which this high-quality guidance for tropical 1980s, however, dynamical models were not particularly cyclone track and genesis comes are the same ones used to skillful at tropical cyclone prediction. Global dynamical forecast other types of weather; they are not specialized tools models—in particular, the ones used for general weather developed just for tropical cyclones. forecasting—did not have great skill at forecasting tropical cyclone track and had even less skill at forecasting intensity Automated storm tracking from model output and ensemble or genesis. forecast systems allow sophisticated tropical cyclone forecasts to be generated without any intervention from As a result, tropical cyclone forecasting relied heavily on a human forecaster other than for quality control and a range of specialized models and tools. These were both interpretation. Given access to currently available global dynamical and statistical in nature. None was especially ensemble forecast systems, it takes relatively few resources to accurate by today’s standards, and a high degree of forecaster produce a forecast today whose accuracy—for track, at least, expertise was needed to interpret all this guidance and turn if perhaps not yet intensity—is comparable to that attained by it into a skillful forecast of the track of a tropical cyclone. the best national centers a relatively short time ago—perhaps Intensity forecasting was even more primitive than track a decade, perhaps even less. This is still a new development, forecasting; dynamical models were nearly useless, and however. Virtually all currently operating national centers intensity forecasts were based largely on statistical model came into existence and developed their structures at a time guidance and the forecaster’s judgment. Genesis forecasting when numerical model guidance was much poorer than it is was not seriously attempted; the forecast operation did not today and when the human forecaster’s judgment played a truly begin until a tropical cyclone was clearly in existence. much greater role. After a roughly five-year period in the late 1980s during These national centers strive to do better than the baseline which dynamical models did not improve (DeMaria and Gross offered by global dynamical model ensembles. The best- 2003), a period of steady improvement began around 1990 supported and most sophisticated centers, such as the WMO and has continued up to the present. The role of numerical RSMCs, have access to resources well beyond the global models in the field at that time is indicated by the WMO models and other publicly available data (for example, Global Guide to Tropical Cyclones (WMO 1992), which stated that numerical model output was becoming an increasingly See, for example, http://www.ral.ucar.edu/guidance/realtime/ 10 reliable source of guidance for operational forecasters. This current/, http://mag.ncep.noaa.gov/, and https://www.fnmoc. document reported that 50 percent of the countries surveyed navy.mil/wxmap_cgi/. Many storm surge models use so-called shallow water equations, which approximate the ocean as a fluid with a constant density that moves with the same horizontal velocity at all depths Review of Operational Practices and Implications for Bangladesh / 29 satellite data). These resources may include high-resolution Imagine for a moment that someone has to start a new regional models, highly trained expert forecasters, and tropical cyclone forecast center in a place where none exists. possibly aircraft reconnaissance. With all these assets, If the quality of the forecast is defined solely by technical forecasters at such centers are able, on average, to improve metrics—that is, by the accuracy of the track and intensity on the skill of the global multimodel ensemble. forecasts alone, leaving aside the effectiveness with which that information is communicated to users—it is now possible Figure 2.8 shows track errors at 48-hour lead time from to produce a good tropical cyclone forecast (by quite recent individual models, similar to Figure 2.7; however, it shows historical standards) very cheaply, simply by using automated only “early” models (available to the forecaster at the start of algorithms applied to global ensemble products already the forecast cycle), starts in 1990, and also shows the official produced elsewhere. Some expertise and effort are needed NHC forecasts. The NHC forecasts have similar errors to the to produce actual forecasts from those products, but the “consensus” forecasts, constructed by averaging multiple necessary investment is much smaller than those required to models (the lowest points on the plots, with plus symbols), produce the ensemble products (or any other state-of-the-art indicating that NHC forecasters (in the Atlantic, at 48 hours numerical weather prediction products) in the first place. An lead time) have less skill at track prediction than the average additional advantage of this approach is that the ensemble of the models. Yet the trend toward model improvement prediction systems allow, in principle, a long-range forecast suggests that this may not be true for much longer, if for no that seamlessly crosses the genesis event, facilitating longer- other reason than at some point the models will approach range forecasts without new or special methods (though the the theoretical limits of predictability, beyond which chaos development and evaluation of specific forecast products still theory tells us that further improvement is impossible. requires skill and effort). Global models are still poor at predicting tropical cyclone Alternatively, someone can make a large investment of human, intensity, so there may be much greater room for national financial, computer, and other resources to replicate the centers to beat the global models there; yet improvement capabilities of the best numerical weather prediction centers: in intensity forecasts has been slow in practice. Genesis perform data assimilation, run the models, and produce forecasting is now becoming a reality, at lead times of five ensemble and other numerical weather prediction products. days and longer, yet this again is largely due to the global By doing this sufficiently well, as well as training forecasters models. to use the numerical guidance to produce forecasts— Figure 2.8 Track Errors at 48 Hours Lead Time (Nautical Miles), as a Function of Year Source: From http://www.nhc.noaa.gov/ verification. Note: From model guidance. Each symbol represents a different model, as indicated in the legend. Only “early” models, available to the forecaster at the start of the forecast cycle, are included. The NHC official forecasts are shown by the solid curve, while the CLIPER (climatology and persistence) statistical forecast is shown by the dashed curve. 30 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING essentially by replicating all the combined capabilities of potential to save lives and may be well worth the resources both advanced forecast centers such as NHC and numerical invested in it. It is also possible that the latest research on weather prediction centers such as NCEP or ECMWF—a intensity prediction will lead to new improvements on that forecast could be produced that may be better than what front, widening (at least for a time) the margin between could be achieved using only automated algorithms based on what the best available methods can do at a regional scale existing global model ensembles. But if it is better, that is not (for example, high-resolution regional models assimilating likely to be by a large margin. If resources are limited, this aircraft-based radar observations, all interpreted by expert all-out approach does not make sense. It would make more forecasters) and what the global models can achieve. But it sense just to train a group of forecasters to use existing global is important to recognize that the tremendous improvement ensemble products. in accuracy over the last couple of decades makes it much easier to produce a tropical cyclone forecast with reasonable This is not to devalue the increase in accuracy that does accuracy and also makes it much more difficult (though still come from a substantial investment in improved forecasts. possible) to produce a forecast with better accuracy than the Any improvement in track or intensity prediction has the real global models. The increasing availability of sophisticated meteorological information from many sources, on the Internet and otherwise, leads to greater attention and expectations on the part of users, but this interest may not be accompanied by adequate knowledge and sophistication to allow them to understand the information. Review of Operational Practices and Implications for Bangladesh / 31 In practice, however, a forecast is not defined only by the that determine impacts. It would be premature for decision quantitative predictions of track and intensity. A modern makers to take major action, such as ordering evacuation, forecast incorporates many other elements, including a based on such an uncertain forecast. Such a forecast might, wide range of graphic products, watches and warnings, and however, justify taking some preliminary preparatory actions, verbal advisories. These are critically important to the users’ such as moving supplies or key personnel closer to where ability to understand both the nature of the hazard (since the they might be needed. position of the center and maximum sustained wind speed are very limited metrics) and the uncertainties in the forecast. Given this, the tropical cyclone forecast center has a Communication of uncertainty is particularly important when critical role as an educator as well as a communicator. The the uncertainty is largest, as it normally is for track at long necessary education may target both relevant professionals, lead times and for intensity at all but the shortest lead times. such as emergency managers and government officials, and the public, and it may occur over the long term as well as The increasing availability of sophisticated meteorological within a single forecast. Though outside the scope of this information from many sources, on the Internet and report, these education and communication aspects are at otherwise, leads to greater attention and expectations on least as critical to disaster preparedness as is the science the part of users, but this interest may not be accompanied of forecasting. A prerequisite for the effective education of by adequate knowledge and sophistication to allow them forecast users is that the forecasters themselves must have to understand the information. A forecast at five days may a good understanding of the underlying science, much of have useful accuracy, but large uncertainty. For example, a which—such as that behind ensemble forecast systems—is forecast that landfall will occur in some region five days from still relatively new and rapidly evolving. now may come with a large uncertainty in the exact location of landfall (or perhaps even whether landfall will occur) and even greater uncertainty in the intensity of the storm at landfall, storm surge, precipitation, and other variables 32 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Storm over Dhaka. Credit: NASA 33 Chapter 3 Tropical Cyclone and Storm Surge Forecasting in Bangladesh This chapter assesses tropical cyclone and storm surge prediction as they are practiced in Bangladesh. The focus is on the operational practice of the Bangladesh Meteorological Department. Given that India’s Meteorological Department is the Regional Specialized Meteorological Center, its practices and links with the BMD are also discussed. The overall aim is to see how lead time for tropical cyclone forecasting in Bangladesh can be increased, in light of the operational practices discussed in Chapter 2, and the constraints that need to be addressed. 34 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING CHARACTERISTICS OF TROPICAL CYCLONES IN THE BAY OF BENGAL Storms in the Bay of Bengal are of shorter duration than 3.2, the Saffir-Simpson scale allows comparison of BoB data storms in other basins, and this has important implications more directly with statistics for other basins. (The IMD’s for forecasting. This conclusion is based on “best track” scale has fewer categories at high intensities but adds an data from the U.S. Joint Typhoon Warning Center for the additional category, “deep depression,” at the low end. In BoB and Western North Pacific basins, as well as data from both systems, a “depression” is not strictly defined, except the U.S. National Hurricane Center for the North Atlantic— that its maximum sustained winds must be below the lowest specifically, statistics for the period 1981–2010, a period threshold on the scale—28 knots in the IMD scale, 34 knots in with reliable, consistent data. To allow direct comparison the Saffir-Simpson scale. An additional difference is that the with data for other basins, the Saffir-Simpson scale, shown IMD defines maximum sustained winds by 3-minute averages, in Table 3.1, is used to categorize storms according to their while other countries use either 1 minute (U.S.) or 10 minutes, intensity. Although the BoB countries use the intensity scale which is the WMO standard.) of the India Meteorological Department, shown in Table Table 3.1 The Saffir-Simpson Hurricane Intensity Scale Category Kt Mph m/s Depression Tropical storm 34–63 39–74 17–32 1 64–82 75–95 33–42 2 83–95 96–110 43–49 3 (major) 96–112 111–129 50–58 4 (major) 113–136 130–156 59–69 5 (major) 137 or higher 157 or higher 70 or higher Table 3.2 The Tropical Cyclone Intensity Scale Used by the India Meteorological Department Category Kt Mph m/s Depression Deep depression 28–33 32–38 14–16 Cyclonic storm 34–47 39–54 17–24 Severe cyclonic storm 48-63 55-73 25–32 Very severe cyclonic storm 64–119 74–137 33–61 Super cyclonic storm 120 or higher 138 or higher 62 or higher Figure 3.1 depicts tracks of tropical cyclones (those reaching period 1980–2010, one year had no storms, several had at least tropical storm intensity, maximum wind speed of one, and several had six or seven, as shown in Figure 3.2 34 knots or greater) in the Bay of Bengal for the decades (panel a). The North Indian Ocean is unique among tropical 1981–90, 1991–2000, and 2001–10. This 30-year period cyclone basins in having two distinct active seasons, one in corresponds roughly with the period of satellite observations spring (peaking in May) and another in autumn, as shown in and constitutes the highest-quality historical data on tropical Figure 3.2 (panel b): during July-September, the formation cyclones. The analysis shows that storms tend to move from of tropical cyclones is suppressed by strong vertical wind east to west or south to north, although many recurve toward shear associated with the monsoon. Of the two seasons, the the east as they reach higher latitudes near Bangladesh. later one, occurring primarily in October-December, is the more active. On average, the Bay of Bengal experiences 3.5 storms a year, but there is considerable variability between years: in the Review of Operational Practices and Implications for Bangladesh / 35 Figure 3.1 Tropical Cyclone Tracks in the Bay of Bengal, Bay of Bengal 1981–2010 Storms 1911 – 1990 32 TCs Bay of Bengal 25N Storms 1991 – 2000 39 TCs 20N 15N 10N Bay of Bengal Storms 2001 – 2010 70E 75E 80E 85E 90E 95E 100E 35 TCs Source: Data from the U.S. Joint Typhoon Warning Center. Note: Each panel shows data for one decade of the 30-year period for storms reaching at least tropical storm intensity (maximum sustained surface winds greater than 34 knots on the Saffir-Simpson scale). Number of storms per year, Bay of Bengal Figure 3.2 Total number of 8 tropical cyclones each year (top) and mean number of 6 tropical cyclones per month Number of storms (bottom) reaching at least tropical storm intensity in the 4 Bay of Bengal, 1981-2010. 2 Note: The results in the bottom graph are computed by adding the total number of storms occurring in each 0 month during the period 1981–2010 1980 1985 1990 1995 2000 2005 2010 and dividing by the number of years, hence the fractional values. Number of storms per month, Bay of Bengal 1 8 0.8 6 Number of storms 0.6 4 0.4 2 0.2 0 0 J F M A M J J A S O N D 36 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Tropical cyclones have shorter lifetimes in the Bay of The abscissa in Figure 3.3 shows the duration of this period Bengal, on average, than in most other basins. Figure 3.3 between genesis and landfall, in days, while the ordinate shows histograms of the time between genesis (defined as shows the number of storms with the corresponding duration. first achievement of maximum sustained wind greater than The sum of all of the values in each histogram is the total 34 knots) and first landfall for storms in the Bay of Bengal, number of storms observed in the basin. The median value is Western North Pacific, and North Atlantic, during 1980–2011. around 5 days in both the Pacific and Atlantic, but only half The Western North Pacific and North Atlantic are chosen for that, around 2.5 days, in the Bay of Bengal. Very few storms in comparison, as they have both relatively large numbers of the BoB last longer than 5 days. This is presumably due to the tropical cyclones (the Western North Pacific is the most active physical geography of the Bay, as it is simply a much smaller basin on Earth) and good observational records. They have body of water than the Atlantic or Pacific. It is not possible also been served historically by some of the largest and most for a Bay of Bengal storm to form nearly as far offshore as advanced forecast centers, and so the behavior of storms in in the other basins, so that—given typical speeds of tropical these basins has guided the development of tropical cyclone cyclone motion—these storms tend to reach land sooner after science and forecast practice. formation. Figure 3.3 Histograms of the Time, in Number of storms per month, Bay of Bengal 1980-2011 Days, between Genesis and the First 40 Median = 2.5 Landfall of Tropical Cyclones in the Bay of Number of storms Bengal, Western North Pacific, and North 30 Atlantic Basins, 1980–2011 Note: Genesis is when the maximum sustained sur- 20 face winds first exceed 34 knots. The median values are shown in the upper right corner of each plot. 10 0 0 5 10 15 20 Days Duration of storms, Western North Pacific 1980-2011 150 Median = 5 Number of storms 100 50 0 0 5 10 15 20 Days Duration of storms, Atlantic 1980-2011 100 Median = 4.625 80 Number of storms 60 40 20 0 0 5 10 15 20 Days Review of Operational Practices and Implications for Bangladesh / 37 Current practice—in all national forecast centers and in all landfall forecast is shorter there than in the other two basins. basins—is to produce forecasts of a tropical cyclone’s track The IMD extends its lead time by using a 28-knot threshold to and intensity only after an existing disturbance has reached begin track and intensity forecasts, rather than the 34-knot a specified intensity threshold. Because tropical cyclones in threshold used elsewhere. the BoB have shorter lifetimes, the maximum lead time of a REGIONAL CONTEXT FOR TROPICAL CYCLONE FORECASTING IN BANGLADESH Bangladesh is a member of the WMO/United Nations Economic An RSMC serves each of the regional bodies; for the WMO/ and Social Commission for Asia and Pacific (ESCAP) Panel ESCAP panel, the RSMC is located at the IMD in New Delhi. on Tropical Cyclones for the Bay of Bengal and the Arabian The RSMC issues guidance to the meteorological services of Sea (the other panel members are India, Maldives, Myanmar, other member nations in the form of outlooks and advisories. Oman, Pakistan, Sri Lanka, and Thailand). The WMO/ESCAP The national services of the individual member nations retain panel is one of five regional bodies under the WMO Tropical responsibility for issuing forecasts for their own nations. Cyclone Program, which fosters regional collaboration and Thus the BMD forecasts may differ from the guidance issued capacity building for forecasting of tropical cyclones and by the RSMC, based on the judgment of local forecasters or storm surge. the use of local observations not available to the RSMC, but the RSMC’s advisories and outlooks are available as guidance to the BMD. THE BANGLADESH METEOROLOGICAL DEPARTMENT MANDATE The BMD is responsible for all meteorological activities in upper air radiosonde stations, of which one (in Dhaka) is Bangladesh. Central to its mandate is forecasting weather, made available internationally via the GTS; and five radars, including tropical cyclones, for the nation. The BMD operates three of which are Doppler radars. At its headquarters in an observational meteorological network throughout the Dhaka, the BMD operates the Storm Warning Center, which is country and the adjacent oceans, including 35 synoptic specifically responsible for issuing tropical cyclone warnings observatories making surface weather observations; three and also storm surge warnings. DATA AVAILABLE AT THE BMD In addition to the local observations taken by its own network, are described in a checklist that the forecaster uses to make the BMD obtains a range of meteorological data from outside assessments of a large pre-defined set of quantities in the the country via a GTS link to the RSMC in New Delhi. The GTS satellite images, weather maps, and other data. carries a set of standard meteorological data that are shared between meteorological services worldwide. In addition to As elsewhere, satellite observations are critical to tropical being used to transmit data from the RSMC to the BMD, it cyclone analysis at the BMD. Dvorak analysis is used to estimate also carries observations made by the BMD in Bangladesh tropical cyclone position and intensity from geostationary back to the RSMC and to the global GTS/WIS. As with many infrared and visible satellite imagery. Estimates of position forecast centers, the BMD does not have a dedicated staff and intensity from other centers, including the JTWC in for forecasting tropical cyclones. When a tropical cyclone Hawaii, are consulted in addition to estimates produced in- is present in the Bay of Bengal, forecasts of its track and house and from the RSMC. A range of global and regional intensity are prepared at the BMD by the same forecasters numerical models are consulted for forecast guidance, who normally forecast other types of weather during the including the Weather, Research, and Forecast (WRF) and remainder of the year. The BMD does have a set of special the Hurricane Weather, Research, and Forecast model run practices for tropical cyclone analysis and forecasting. These in India at the IMD, the Indian Institute of Technology in 38 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Delhi, and the National Centre for Medium Range Weather and forecasting model) is run on an experimental basis on Forecasting, as well as the standard set of global models a larger computer cluster supplied recently by the Japan from the United Kingdom, United States, ECMWF, and International Cooperation Agency (JICA). Neither of these elsewhere. Two regional numerical models are run in-house models is dedicated specifically to or designed for tropical at the BMD: the WRF model is run operationally on a modest cyclone forecasting; both are run routinely for all weather workstation, and another regional model (the nonhydrostatic situations. mesoscale model, another version of the weather research BMD FORECASTS practices of both BMD and IMD in most respects. This storm To assess whether the lead time for tropical cyclone and was, however, unusual in one important respect. It formed storm surge forecasts issued by the BMD could be increased more than five days before landfall, allowing five-day to five or more days, it is first necessary to understand what forecasts to be produced without requiring that the forecasts forecasts and warnings are issued by the BMD presently and begin before genesis. how they are produced. The BMD’s forecast products are illustrated using examples obtained from its webpage during It appears that the BMD does not issue—at least not on the the lifetime of Cyclone Mahasen in May 2013. IMD’s forecasts Internet—quantitative track or intensity forecasts. A map of Mahasen are also described. In both cases, forecasts showing the observed track (up to the present time) is issued, issued on the website are taken as indicative of the forecast but it does not extend into the future. The BMD’s forecasts of center’s practices. Forecast information is issued to a variety tropical cyclones appear to be limited to textual forecasts and of users by other means, including by radio and television to the broader population and by direct communication warnings. These give only qualitative information about the with emergency management officials. It is assumed that storm’s future behavior and do not extend far in the future. the character of the information issued through these other Additionally, the BMD employs a system of numerically channels is not radically different from what can be found on ranked “warning signals,” which are issued to maritime the Internet. As described, however, there is some evidence ports, indicating varying degrees of imminence or severity in the case of Mahasen that disaster management officials of a weather threat. Figure 3.4 presents a warning issued on had access to information beyond that available on the BMD May 13, 2013, three days before the Cyclone Mahasen made web page. landfall on May 16, 2013. It gives quantitative estimates of the storm’s current position and intensity, but its explicit As Mahasen was quite recent, the forecasts made in this predictions of the future indicate only that “it is likely to example are also taken here as representative of the current intensify further and move in a northerly direction.” Figure 3.4 Warning Issued by Bangladesh Meteorological Department on May 13, 2013 Source: Downloaded from bmd. gov.bd on that date. Review of Operational Practices and Implications for Bangladesh / 39 At the same time, the warning is noteworthy for the fact that it much more detailed information about regions likely to be does, in a sense, contain a longer-range probabilistic forecast, affected by the storm. Quantitative storm surge information is in that it orders the raising of “Local Warning Signal No. 4 for also given, with a set of locations forecast to be inundated by Chittagong, Cox’s Bazar, and Mongla ports.” This puts these storm surge of 5–7 feet above the astronomical tide. locations on alert that they are under possible danger from the cyclone, without stating that it is imminent. Landfall occurred The Internet is, according to BMD personnel, not the primary on May 16th, approximately three days after this warning mode of disseminating forecasts in Bangladesh. Television was issued; the first warnings were issued by the BMD two and radio are believed to reach more of the population. The days before, on May 11th, when Mahasen was named, and Cyclone Preparedness Programme (CPP) (Habib, Shahidullah, the first local warning signals were raised on May 12th. Thus and Ahmed 2012) under the Government of Bangladesh and while the BMD did not issue explicit long-range forecasts of the Bangladesh Red Crescent Society are among the important the storm’s behavior, it did begin to alert the country to the organizations that respond to cyclone forecasts and warnings. potential threat four and five days before landfall. While this is Press reports at the time indicated that the District Disaster perhaps atypical—Mahasen was a long-lived storm, forming far Management Committee, a local entity under the country’s offshore—it was not the first time that the BMD had issued such Comprehensive Disaster Management Plan (Government of early warnings; the first local warning signals for Cyclone Sidr the People’s Republic of Bangladesh 2010), was meeting in 2007 were also raised five days before landfall (S. Ahmed about the storm as early as May 12th, four days before landfall and T. Imam, personal communication). (“Cyclone Eyeing Chennai,” Daily Star, May 13, 2013). Since no explicit forecast of landfall had been made by the BMD that Figure 3.5 shows a warning issued two days later, on May early, this suggests either that the BMD transmits information 15th, a little less than 24 hours before landfall. At this point to disaster management officials beyond what is available the ports of Chittagong and Cox’s Bazar were under Danger on the Internet or that disaster management officials consult Signal No. 7 and Mongla was under No. 5. The warning has other sources in addition to the BMD. Figure 3.5 Warning Issued by Bangladesh Meteorological Department on May 15, 2013 Source: Downloaded from bmd.gov.bd on that date. 40 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING FACTORS CONSTRAINING BMD’S FORECAST PRACTICES A number of factors constrain BMD’s capacity to improve Many model data themselves are not available for analysis on its tropical cyclone forecast practices. The first limitation BMD computers. pertains to data from external sources. Data are available from sources outside Bangladesh, including both observational Some key data are not used at all. In particular, polar- data and numerical model–produced data, to which the BMD orbiting satellites, such as microwave sounders and imagers either does not have access or does not have access in forms and scatterometers for surface winds, are not used at all that would make them most useful. in tropical cyclone forecasting at the BMD. Although the data are not available digitally at the BMD, the images are Some local data are not on the WIS. Not all observations taken not even consulted. As these data can be quite valuable for in the South Asian region that may be relevant to tropical estimates of tropical cyclone intensity and structure, it would cyclone prediction are shared via the WIS. This is normal; the be advantageous if they were available and used. Part of the WIS protocol does not include all observations taken by any issue may be training in the use of these data. country. Other data may be on the WIS but not obtained by BMD for reasons that are not clear. An important example is The software environment could be improved. In better- data from Indian Ocean buoys; these data are on the WIS but equipped centers, key tasks carried out by tropical cyclone the BMD does not appear to obtain them currently. Similarly, forecasters—particularly visualization of model forecast some BMD observations are not shared with the RSMC; for guidance and production of graphic track forecasts based example, coastal radar data from Bangladesh are not sent from that guidance, as well as Dvorak analysis and other steps back to IMD through the WIS and thus are not available for associated with assessment of the storm’s present state—are assimilation into IMD’s numerical models. Greater sharing of made easier and more effective by computer hardware and, local, non-WIS data would benefit the entire region and could especially, software designed specifically for the purpose. be accomplished by bilateral agreements. The BMD lacks these tools. This places greater burdens on the forecasters and makes it more difficult for them to use A critical limitation for the BMD is the bandwidth of its WIS the available data in the most optimal way. Dvorak analysis, link. Currently this is 64 kilobytes per second, a rather low for example, is best done on a computer workstation that has value. Sharing BMD’s coastal radar observations over the WIS the necessary satellite data loaded into software designed for back to IMD, for example, is estimated to require 5 megabytes the purpose. The digital data are not available at the BMD but per second. While this is 20 times more than the current neither are dedicated workstations with appropriate software, bandwidth, it is not a particularly large number by global (or so that even if the data were available, the information would even, probably, Bangladeshi) standards. It would be highly not be as useful as it should be. Altogether, this renders desirable to increase the bandwidth available to the BMD for the Dvorak analysis procedure there more primitive than at its WIS link. better-equipped centers. Presumably as a consequence of the limited bandwidth of the The BMD does not have access to a software environment WIS link, many data are available to the BMD only through comparable to the Automated Tropical Cyclone Forecast the Internet in the form of images viewed on a Web browser. System (the system used at the NHC and JTWC in the United This means that, while BMD forecasters are in principle able States, described in Chapter 2) for ingesting model track to consult a range of observations and numerical model guidance directly into a unified graphical environment, products similar to those used at other centers (because a allowing the forecaster to directly compare different track wide range of data are available over the Internet), many of guidance and use it to produce a forecast track directly in those data are not available at the BMD directly as digital digital form. Rather, many different forms of guidance are data; they are only available as images. This is a serious consulted separately on Web browsers, and the information limitation. Dvorak analysis, for example, requires identifying must be combined in the forecaster’s mind. Moreover, the areas with temperatures below certain thresholds by infrared tropical cyclone tracks produced from global models by the brightness; this requires digital data. Similarly, much model various centers using their own automated tracking programs output is available at the BMD only as images of model fields. are not available at the BMD at all; only images of the model Review of Operational Practices and Implications for Bangladesh / 41 fields are. Altogether, these limitations make the forecaster’s Another limitation pertains to the observational network of task more difficult, as chores that could be done both the BMD itself. The BMD’s own observations, over Bangladesh accurately and automatically by the computer instead must itself and the adjacent ocean, are an essential component of be done by the forecaster. the tropical cyclone forecast process. The BMD would benefit by improving and expanding its current observing system Finally, some data are not used for reasons that are unclear. with the following: For example, BMD forecasters stated that they do not have access to data from the buoys in the Bay of Bengal that are Automated tide gauges at the coast, which would be owned and operated by India. These buoy data could be invaluable for measurements of storm surge quite important for analysis of tropical cyclone intensity and Better bathymetric and topographic data (DEM) for structure in some situations. IMD personnel explained that storm surge and inundation forecasting these buoy data are available on the WIS. It is not clear why the BMD believes it does not have access to these data; it is Calibration of coastal radars as well as intercomparison not a bandwidth issue, as buoy data are sparse and require against rain gauges little bandwidth. Buoys for oceanographic and meteorological observations offshore. OTHER PHYSICAL SYSTEMS NEEDS Better hardware and software systems for tropical cyclone In-house data assimilation of WIS and local observations, analysis and forecasting, such as the Automated Tropical including radar, upper air observations, and surface Cyclone Forecasting System for displaying model guidance, synoptic observations, would also be desirable. Compared NCEP’s Advanced Weather Interactive Processing System with installation of hardware and software systems, the for displaying observational data, or their equivalents costs here are more substantial and the required training would be particularly valuable. As indicated, the BMD does is more significant. Additionally, the impact on forecasts not have systems equivalent to these. Such systems are would not necessarily be immediate; some aspects of these currently available at regional centers such as NHC (and technologies are experimental, and their ability to improve IMD), which synthesize many data sources and automate forecasts has not been fully demonstrated, even at the best- and streamline many of the tasks involved in tropical cyclone equipped centers. Nonetheless, in-house numerical weather analysis and forecasting, including Dvorak analysis of predictions with high-performance computing should at least satellite observations. Installing such systems, acquiring the be a long-term goal. necessary data streams to realize their full capabilities, and training BMD personnel in their use would be a very low-cost The BMD offices are subject to regular electric power outages. improvement that could lead to a significant improvement in While some computers have backup power systems, the BMD operational practices in the short term. outages are nonetheless at best an inconvenience and at worst potentially disruptive to BMD’s operations. Besides BMD would like to have a stronger in-house facility in the need to keep computers and other critical equipment operational numerical weather prediction. This could running, Bangladesh has a warm climate, and the loss of include tools specifically targeted at tropical cyclones, such air conditioning during parts of the year may inhibit the as the Hurricane Weather Research and Forecast model. performance of BMD personnel. 42 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING HUMAN RESOURCES NEEDS Compared with some other agencies—that is, meteorological According to BMD personnel, as well as to scientists at services in other countries of comparable or smaller size— the South Asia Association for Regional Cooperation the opportunities available to BMD personnel for contact Meteorological Research Center, there is no Department of with other institutions, people, and thus scientific trends Atmospheric Sciences (or Meteorology, or any equivalent or and developments in the broader field of meteorology and comparable designation of the field) anywhere in Bangladesh. atmospheric sciences are limited. The goal of improving (There is an atmospheric physics research group in the Physics forecasts requires, in the short term, more training of BMD Department at the Bangladesh University of Engineering personnel in existing and developing science and technology and Technology, and perhaps scattered individual faculty that specifically address that goal: Dvorak technique, members in related fields at other Bangladeshi universities, numerical weather prediction, ensemble prediction but no group that is equivalent to an entire department or methodologies, radar data analysis, and the like. While a that has strong ties to the BMD.) narrow focus on these methods might be adequate to achieve some improvement in forecast practices, these methods do The lack of any university department to which the BMD might not exist in isolation from the broader scientific field out of develop ties puts the department at a great disadvantage which they have emerged. The long-term implementation of compared with comparable institutions elsewhere. Most these technologies at the BMD will be more effective—and important, it limits the scope of the education available to the adoption of future technologies not yet envisioned will BMD personnel. The BMD obtains staff mainly by hiring also be greatly facilitated and accelerated—if they are part graduates in physics and applied mathematics, typically from of a broader effort to improve the foundational education Dhaka University, and then training them in-house for one and training opportunities available to BMD personnel in year before they start work as forecasters. A smaller subset meteorology and atmospheric sciences. of forecasters spends periods of time—typically one year or Women in rural village, southern Bangladesh. Video Still: Stephan Bachenheimer / World Bank Review of Operational Practices and Implications for Bangladesh / 43 less—obtaining training at institutions outside the country intellectual ground and geographic proximity can take many (often in India, but sometimes elsewhere). While these in- forms and bring many benefits. The BMD has no intellectual house and foreign training experiences are critically important partner of this type. and valuable, it is not likely that they are able to cover the foundations and more recent developments in the field to the The BMD does benefit from interactions with institutions extent that would be possible in a full undergraduate degree overseas, including the IMD in India, JICA and the Japan program, let alone a graduate program. Meteorological Agency, the Norwegian Meteorological Service, and perhaps others. Nonetheless, a stronger Additionally, the value of relationships between universities partnership with an institution in the region, if not the country, and government agencies can, in the best cases, extend would be of great value. Of course, such a partnership would well beyond the obvious educational role. A good university likely require BMD personnel to spend significant time at the department will perform research, some of which may be of partner institution and thus away from their duties. This may interest and possibly even operational use, to forecasters. not be possible, as the BMD does not have a sufficiently large Intellectual exchange between different types of institutions— staff to allow many long absences, but in the long term such that is, a government mission agency versus an academic development opportunities for staff would bring considerable university or research institute—that share common benefit. COLLABORATION WITH THE IMD The IMD is the government agency responsible for weather Indian Ocean (BoB and Arabian Sea) have been thoroughly forecasting for India. As part of its responsibilities to evaluated in two recent studies (Mohapatra et al. 2013a,b). the nation, the IMD produces tropical cyclone forecasts. The skill in the track forecasts is somewhat less than that of Additionally, it hosts the Regional Specialized Meteorological the Western North Pacific forecasts by the RMSC in Tokyo or Center for the WMO/ESCAP panel at its central office in the NHC in Miami, but it is increasing more rapidly with time New Delhi, thus providing an important service to the other (Mohapatra et al. 2013a). Skill in the IMD’s intensity forecasts nations on the panel. is not as straightforwardly evaluated as in these other basins (because of the different baseline references against which The RSMC issues a tropical weather outlook once daily, at skill is evaluated; persistence is used by IMD in the Northern six hours universal time (06 UTC), under all circumstances. Indian Ocean whereas the “climatology and persistence When a tropical depression—the embryonic form of a tropical (CLIPER)” model is used as a reference in the Pacific and cyclone—has formed, the RSMC issues special tropical Atlantic; Mohapatra et al. 2013b). Similar to the other basins, weather outlooks twice daily. When the storm becomes a however, it is clear that the IMD’s intensity forecasts are both deep depression (winds greater than 28 knots but less than 35 significantly less skillful than their track forecasts and show knots) these outlooks are issued five times daily, and track and much more gradual improvement with time. intensity forecasts for the next 120 hours are produced. When the maximum sustained winds exceed 34 knots, the system is This section describes some technical aspects of the IMD’s named and the RSMC issues tropical cyclone advisories eight operations in New Delhi as they pertain to Bangladesh via times daily (WMO 2012). Tropical cyclone advisories include IMD’s role as the RSMC. As might be expected, given its role (among other information) track and intensity forecasts. as the RSMC as well as the larger size of India’s population and economy, the IMD has capabilities considerably greater Objective tropical cyclone track and intensity forecasts than those at the BMD. At New Delhi, the IMD has a high- have been produced at IMD since 2003, at which time the performance computing facility with a 14.4 teraflop capacity, maximum lead time was 24 hours; the maximum lead time which was expected to be increased to 110 teraflop by July was increased to 72 hours in 2009 and to 120 hours in 2013, that is used for numerical weather prediction. Smaller 2013. Storm surge guidance is also included, based on the but still substantial computer facilities and numerical Indian Institute of Technology Delhi and the Indian National weather prediction operations are carried out at IMD regional Centre for Ocean Information Services storm surge model, offices in other states. In New Delhi, the WRF regional model at shorter lead times. The IMD’s forecasts for the northern is run routinely under all weather conditions for forecasts 44 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Figure 3.6 Track Forecast Issued by the IMD RSMC on May 12, 2013, for Cyclone Mahasen Note: The observed track is shown in black, the forecast track in red, and the cone of uncertainty in green. The forecast extends to 120 hours from the time at which it was made. out to 72 hours, using an outer domain with 27-kilometer for the purpose are not present. Numerical guidance from a horizontal resolution and a nested inner domain, covering broad range of numerical models is ingested and tracks are India and its immediate surroundings (including the Bay of shown on the screen, allowing the forecaster to visualize the Bengal and some of the Arabian Sea as well as Bangladesh guidance in its most useful form and produce a track forecast and other neighboring countries) with 9-kilometer horizontal graphically on the screen. resolution. The Global Forecast System model and data assimilation system are also run in-house; advantages of this The RSMC produces products—graphical and text-based— local version over the U.S. GFS are that the IMD version is able that are broadly similar to those produced at other regional to incorporate a broader set of local observations (including centers. While the BMD’s Internet forecasts, shown earlier, many that are not on the WIS) and that the model run times were limited to qualitative verbal statements at short lead are better synchronized to the operational needs of IMD and times, the IMD’s are quantitative and extend to longer the region. A similar GFS system is also run at the National lead times. Figure 3.6 shows, for example, a track forecast Center for Medium-Range Weather Forecasting, also in Delhi, issued for Cyclone Mahasen (May 2013). Besides illustrating and those results are made available to the IMD. the format used, including the cone of uncertainty, it is noteworthy that this was a 120-hour forecast; it was issued The RSMC’s facilities specifically for tropical cyclone on 06 UTC on May 12th, and the last forecast time was 06 forecasting are comparable to those at other RSMCs. The UTC on May 17th. This demonstrates unequivocally that the IMD has computer workstations dedicated to tropical cyclone IMD now can and will issue five-day forecasts in situations forecasting. Dedicated software packages, comparable to where it is deemed appropriate. those at the National Hurricane Center described in Chapter 2, are available to ingest necessary data and automate The IMD’s forecasts are available to the BMD (and to the some tasks to streamline and facilitate the forecaster’s job. public). As a regional partner of the RSMC, the BMD could, Dvorak analysis is carried out on these workstations once the for example, use the IMD’s forecasts as the basis for its own, necessary satellite data have been ingested. This is in contrast adjusting them according to its own forecasters’ judgment to the current situation at the BMD, where only satellite and adding detail about the expected local impacts of an imagery is available and dedicated hardware and software event. Review of Operational Practices and Implications for Bangladesh / 45 The skill of the IMD/RSMC track and intensity forecasts are In a few respects, the IMD’s forecast technologies lag what similar to those of other centers. Figure 3.7 compares errors is available at some other centers. While much numerical in forecast storm position produced by the RSMC to those model guidance is ingested into the forecast software, the produced by the U.S. Joint Typhoon Warning Center, which only tropical cyclone tracks ingested are those from the also forecasts for the North Indian Ocean. In both cases, the deterministic model runs. Global ensemble runs—from the errors shown are for the whole basin (including the BoB and THORPEX Interactive Grand Global Ensemble—are available, the Arabian Sea) and are taken from the reports published but only on separate machines and not directly in the forecast by the two centers after the 2011 season (RSMC 2012; JTWC system. The forecaster is thus not able to make the best use of 2011). The top panel shows the errors, in kilometers, for the the ensemble information. In other respects as well the IMD’s 2011 season, as a function of lead time. The two centers forecasts are more deterministic and less probabilistic than clearly have similar accuracy; the differences are likely too they might be. While the IMD does use a cone of uncertainty small to be considered significant, given that these results in its track forecasts—constructed very similarly to that at the are from a single season and thus consist of a small number NHC, with fixed radii at each lead time based on historical of forecasts (particularly at the longer lead times). The only errors—other aspects of the forecast are deterministic. The difference that is clearly significant is that the JTWC produced IMD does not have the technology to produce probabilistic forecasts at lead times out to 120 hours during this period, forecast of wind radii or storm surge, although probabilistic whereas the RSMC produced forecasts at a maximum lead rainfall forecasts are produced. Technology transfer from the time of 72 hours. Panel b of Figure 3.7 gives a somewhat United States or other centers that do produce these products longer perspective, comparing the position error at 24 hours would be desirable in order for the IMD to acquire these lead time during the period 2003–11. The JTWC’s errors capabilities more rapidly. appear to be systematically slightly lower, though not in every year, and the differences are not great.11 At present, the IMD does not routinely issue explicit forecasts of tropical cyclogenesis. The possibility of a A more comprehensive evaluation of the RSMC’s forecasts can 11 be found at http://www.wmo.int/pages/prog/www/tcp/TCM-7- tropical cyclone’s formation may be discussed in the 2012.html, as well as in Mohapatra et al. (2013a,b). RSMC’s daily tropical weather outlooks; these have a time Figure 3.7 Historical Track Errors for the 300 RSMC (in Blue) and the Joint Typhoon 250 Warning Center (in Red) for Tropical Track error (km) 200 Cyclones in the North Indian Ocean RSMC (including both Bay of Bengal and 150 JTWC Arabian Sea) 100 50 Source: From data published by the RSMC and the JTWC. 0 0 20 40 60 80 100 120 Note: Panel a shows the position errors as a function of lead time for the 2011 season. Panel b Lead time (hr) shows the errors at 24 hours lead time as a function 220 of the year in which the forecasts were issued, from 200 2003 to 2011. 24 hr track error (km) 180 RSMC 160 RSMC JTWC 140 JTWC 120 100 80 2003 2004 2005 2006 2007 2008 2009 2010 2011 Year 46 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING horizon of 24 hours. The IMD is currently developing some The goal is for India to acquire instrumented C-130 aircraft new capabilities for tropical cyclone forecasting. It is similar to those used by the U.S. Air Force for tropical developing an in-house capability to run the HWRF, through cyclone reconnaissance in the Atlantic. These would have a Memorandum of Understanding (MOU) with NOAA in the the potential to make a substantial positive difference in the United States and close collaboration with scientists at analysis and forecasting of tropical cyclones by providing NOAA’s Hurricane Research Division. This capability includes accurate observations of storm structure and intensity over data assimilation and other techniques developed under the Bay of Bengal. HFIP results from the Atlantic suggest that the Hurricane Forecast Improvement Program. Another these data would be particularly effective when assimilated major development, part of the same MOU, will be the into the HWRF (see, for example, Gall et al. 2013). acquisition of aircraft for tropical cyclone reconnaissance. COLLABORATION WITH OTHER ORGANIZATIONS Several other organizations in Bangladesh and in the region The Japan International Cooperation Agency has a long history interact with the BMD on issues related to tropical cyclone of support of and collaboration with the BMD. Recent examples and storm surge forecasting. include JICA’s supply and installation of a high-performance computer and configuration of that computer to run the The Regional Integrated Multi-Hazard Early Warning mesoscale model; support for installation of observational System (RIMES) is an international and intergovernmental facilities, including automatic weather systems and coastal organization that generates and applies early warning radars; and development of new forecast products. information for natural disasters. RIMES is based in Thailand but has 13 member states and 18 collaborating countries The BMD collaborates with the Norwegian Meteorological throughout South, Southeast, and East Asia as well as East Institute. The BMD obtains weather data from ECMWF via Africa. RIMES provides technical support to the BMD both in this collaboration, which may include other activities but is cyclone and storm surge forecasting and facilitates for the not specific to tropical cyclones. The Centre for Atmospheric WMO Coastal Inundation Forecasting Demonstration Project Sciences at the India Institute of Technology, Delhi, has a close for Bangladesh (CIFDP-B). relationship with the IMD and (somewhat less directly) with the BMD. The storm surge model used at both the IMD and The Bangladesh Department of Disaster Management (DDM), the BMD was developed at IIT by Professor Emeritus Shishir which includes the Cyclone Preparedness Program, interacts Dube and is currently maintained by Professor A. D. Rao, who with the BMD on cyclone warning and response. The DDM heads an active research group with a significant focus on and the CPP have signed as participants in the CIFDP-B. storm surge and coastal inundation. Professor U. C. Mohanty The Bangladesh Water Development Board includes the Flood maintains an active research program on tropical cyclone Forecasting and Warning Centre (FFWC), which is responsible modeling and prediction, including high-resolution tropical for predicting floods on land. Although the FFWC’s activities cyclone simulation and prediction using the WRF model, historically have focused largely on riverine and precipitation- data assimilation, and other topics, and maintains a close driven floods, their mandate includes all flooding on land. connection to the IMD. 47 Chapter 4 Findings and Recommendations This concluding Chapter summarizes key finds of the report and proposes recommendations. 48 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING KEY FINDINGS TROPICAL CYCLONE FORECAST LEAD TIME has improved greatly in recent years. The accuracy of these IN STATE-OF-THE-ART NATIONAL CENTERS models, combined with improved understanding of the influence of coherent large-scale modes of variability on The review of operational practices suggests that at state- genesis, now allow genesis to be forecast with some accuracy of-the-art national centers, tropical cyclone forecasts are at leads of five days or longer in some basins. produced with lead times of up to five days. Both the U.S. NHC and the IMD (as of 2013), for example, issue five-day UNCERTAINTY IN STORM SURGE FORECASTING forecasts of track and intensity. We are not aware of any national center yet issuing forecasts of track or intensity Storm surge forecasting is carried out as a part of tropical for a specific tropical cyclone with lead times greater than cyclone forecasting. A range of numerical models are used; in five days on an operational basis. Storm surge forecasts are some cases, the models used for forecasting may not be the produced with maximum lead times shorter than those used equivalent of the best research-grade models due to the need for the storm itself, and it seems that no centers produce a for fast computation in real time, typically on computers that storm surge forecast with lead time longer than 72 hours. are much less powerful than those used by global numerical weather prediction centers. Storm surge forecasting may be SKILL OF FORECAST TRACK HAS IMPROVED probabilistic, with a range of simulations done to allow for FASTER THAN SKILL OF FORECAST INTENSITY uncertainties in forecast storm track and intensity. Track forecasts have improved dramatically in recent decades, FORECASTING INUNDATION STILL IN A STATE in large part due to steady improvements in global numerical OF RELATIVE INFANCY weather prediction models, data assimilation systems used to initiate those models, and ensemble methods. Forecasting inundation—the horizontal and vertical extent Ensemble methods allow improvement over individual of flooding on normally dry land (as opposed to storm model performance and also allow estimation of forecast surge, which is the height of water above astronomical tide uncertainty. Forecasters make active use of these capabilities, along the coast)—is still in a state of relative infancy. The and modern cyclone forecasts rely heavily on ensemble and technology exists, but it is only beginning to be introduced consensus products from a range of global models. Formal operationally. Verification of inundation forecasts, once estimates of uncertainty provided in forecasts, such as cones they exist, will be an important challenge. In many places, of uncertainty, do not explicitly rely on ensemble information including Bangladesh, upgrades to the existing observational in most cases, although a minority of centers do base those network will be required for quantitative measurements of estimates on ensemble spread in whole or in part. sufficient quality to evaluate the accuracy of an inundation forecast after the fact. Intensity forecasts have not improved at anywhere nearly as fast as track forecasts have. Intensity prediction is currently TROPICAL CYCLONE FORECASTS AT IMD regarded as a critical problem and is the subject of a major The RSMC for tropical cyclones at the IMD identifies and research program in the United States. High-resolution names tropical cyclones in the Bay of Bengal; provides numerical models and assimilation of observations that outlooks, advisories, and warnings to the BMD and other capture mesoscale processes near the storm center, such countries in a panel on tropical cyclones for the Bay of as Doppler radar observations from aircraft, are the most Bengal and the Arabian Sea region; and serves as the hub promising tools for improvement. These tools show promise for the transmission of meteorological data via the GTS/ in research mode, but they have not yet been clearly realized WMO Information System. Objective tropical cyclone track in operations. and intensity forecasts have been produced at the IMD since 2003, at which time the maximum lead time was 24 hours; TIMING OF OPERATIONAL TROPICAL CYCLONE the maximum lead time was increased to 72 hours in 2009, FORECASTS and beginning with Cyclone Mahasen in May 2013, the IMD The ability of global models to predict tropical cyclogenesis issues forecasts out to 120 hours. Storm surge guidance is Review of Operational Practices and Implications for Bangladesh / 49 also included, based on the Indian Institute of Technology forecasts are 72 hours or less. The BMD issues a smaller range Delhi and Indian National Centre for Ocean Information of products than the IMD, and its forecasts are informed by a Services (INCOIS) storm surge model, at shorter lead times. smaller range of data and resources. The BMD obtains a range of numerical model outputs through the GTS link, but it does The IMD has access to considerably more material and human not appear to make use of them to the extent that it might. resources than the BMD does. It has computer workstations Ensemble methods are not heavily used. The BMD’s forecasts dedicated to tropical cyclone forecasting. Dedicated software are essentially deterministic rather than probabilistic. packages, comparable to those at the National Hurricane Center, are available to ingest necessary data and automate FACTORS CONSTRAINING LEAD TIME FOR BMD some tasks to streamline and facilitate the forecaster’s job. FORECASTS Dvorak analysis is carried out on these workstations once the The lead time and skill of BMD forecasts are limited by a necessary satellite data have been ingested. Its forecasts are number of factors, both material and human. quantitative and extend to longer lead times than the BMD’s. In a few respects, however, IMD forecast technologies lag First, the agency lacks much of the state-of-the-art what is available at some other centers. For instance, while hardware and software used elsewhere for tropical cyclone much numerical model guidance is ingested into the forecast forecasts and does not obtain all the globally available and software, the only tropical cyclone tracks ingested are those potentially useful data from observations and numerical from the deterministic model runs. models. For instance, BMD forecasters currently lack tools to carry out key tasks such as visualization of model The RSMC is a sophisticated facility, using a wide range of forecast guidance and production of graphic track forecasts observations and numerical model products (both produced based from that guidance, as well as Dvorak analysis and in-house and from other national centers) to generate a range other steps associated with assessment of the storm’s of forecast products broadly comparable to those at other present state. These are made easier and more effective by RSMCs. Some specific products found at other centers, such computer hardware and especially by software designed as probabilistic storm surge forecasts, are not yet available, specifically for the purpose. Further, at present the BMD does but some new capabilities are being developed through not operationally run models dedicated to tropical cyclone collaboration with NOAA in the United States. The Hurricane forecasting. The BMD also faces frequent power outages that Weather Research and Forecast model has already been disrupt its operations. implemented under this collaboration, and development of an aircraft reconnaissance capability is planned. The latter Second, a critical limitation for the BMD is the bandwidth could be particularly effective in remedying a relative lack of its GTS/WIS link. For instance, coastal radar data from of in situ observations over the Bay of Bengal. IMD’s role as Bangladesh are not sent back to the IMD through the GTS the RSMC means that IMD provides a range of analysis and and thus are not available for assimilation into the IMD’s forecast products to the other members of the WMO Tropical numerical models. Currently the bandwidth of BMD’s GTS link Cyclone Panel, so improvements in IMD’s capabilities have the is 256 kilobytes per second, whereas 5 megabytes per second immediate potential to help Bangladesh and other countries are estimated to be required for sharing BMD’s coastal radar in the region. Any improvement in IMD’s forecasts has the observations over the GTS back to the IMD. Moreover, due potential to improve BMD’s forecasts as well, to the extent to the limited bandwidth of the GTS link, many data are that BMD uses IMD as one of its sources of guidance. But available to the BMD only through the Internet in the form due to capacity constraints at the BMD, data and information of images viewed on a Web browser and are not available already publicly available and provided regionally are often in a digital form for assimilation into models. Satellite data, not used for forecasting at the BMD. for example, are obtained only in the form of images, which do not allow Dvorak analysis—the global standard method for ANALYSIS OF FORECASTING PRACTICES determining the position and intensity of a tropical cyclone— AT BMD to be carried out in its most optimal form. Some data from Tropical cyclone and storm surge forecasts for Bangladesh polar orbiting satellites, such as microwave sounders and are produced by the BMD in Dhaka. The lead times for these imager scatterometers for surface winds are not used at all 50 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING in tropical cyclone forecasting at the BMD. These data are INFLUENCE OF THE SIZE OF THE BAY OF not available digitally and even the images are not consulted. BENGAL Since these data can be valuable for estimates of tropical One significant reason that typical maximum lead time by cyclone intensity and structure, it would be advantageous either BMD or IMD until recently was 72 hours—as emphasized if they were available and used. Part of the issue may be by both BMD and IMD personnel in discussions—is the natural training in the use of these data. physical constraint imposed by the smallness of the Bay of Bengal. A consequence of this constraint, as shown above, Third, the observational networks over land and the is that most Bay of Bengal storms have lifetimes significantly adjacent ocean, which are an essential component of the shorter than five days, measured from the time of genesis as tropical cyclone forecast process, also need improvement. defined by tropical storm intensity. This factor will continue For instance, adequate bathymetric data for coastal to limit the lead time at which landfall can be forecast as Bangladesh are not available for storm surge and coastal long as track forecasts begin at genesis (even defined using inundation forecasting. Automated tide gauges are also a 28-knot threshold). If genesis occurs less than five days needed for measurements of storm surge. before landfall, then even if a five-day forecast is issued at the moment of genesis, landfall will still occur before the last Fourth, the BMD does not have a dedicated staff with day of that forecast. If forecasts of landfall are desired with requisite skills for forecasting tropical cyclones. When a lead times greater than five days, it will still be essential that tropical cyclone is present in the Bay of Bengal, forecasts of such forecasts begin before genesis. its track and intensity are prepared at the BMD by the same forecasters who normally forecast other types of weather during the remainder of the year. The education and training FORECASTS WITH LONGER THAN FIVE DAY opportunities available to staff are also limited, making it LEAD TIMES USING ENSEMBLE METHODS more difficult for forecasters to take advantage of the latest ISSUED EXPERIMENTALLY developments in tropical cyclone and storm surge forecasting. Extensions in lead time could result from the use of numerical At present, there is no Department of Atmospheric Sciences models producing forecasts without the requirement of an (or Meteorology, or any equivalent or comparable designation existing disturbance (as defined by any threshold), to the of the field) anywhere in Bangladesh. There is an atmospheric extent that the models can forecast the formation of the physics research group in the Physics Department at the disturbance. Such forecasts are now possible using ensemble Bangladesh University of Engineering and Technology and methods, and products with the necessary information are scattered individual faculty members in related fields at other being produced by some numerical weather prediction Bangladeshi universities, but no group that is equivalent to centers (for example, ECMWF) (see Vitart et al. 2011) and an entire department or that has strong ties to the BMD. This at least one university group (under Prof. Peter Webster at seriously compromises the government’s ability to forge links Georgia Institute of Technology (see Belanger et al. 2012)). with academic institutions on weather- and climate-related The National Hurricane Center in the US has started issuing research—linkages and partnerships that are often at the crux watches and forecasts for “potential tropical cyclones”, of innovation and research-based service delivery. though this is very recent and is being done only out to 48 hours when a storm is expected to threaten land shortly after genesis. If IMD and BMD were to begin issuing forecasts of this type, they would be matching what some of the more sophisticated forecasting centers have only recently started to operationalize. Review of Operational Practices and Implications for Bangladesh / 51 RECOMMENDATIONS For BMD to improve its forecast lead times, a number of methodologies, radar data analysis, and the like and also actions can be taken. These include the following: in the underlying fundamentals of atmospheric science. It would be particularly valuable for BMD forecasters to Strengthen BMD hardware, software, and become better acquainted with the capabilities of the infrastructure: In the short term, some relatively basic modern global model ensemble prediction systems. In and inexpensive improvements could help to bring the long term, however, the government of Bangladesh the BMD closer to the current operational state of the will need to invest in development of a cadre of art in tropical cyclone forecasting. BMD should access trained meteorologists and atmospheric scientists by dedicated workstations with appropriate software to supporting teaching of these topics at the university carry out key tasks associated with tropical cyclone level. Another recommendation is to establish a forecasting such as data assimilation, model analysis, National Meteorological Training and Research Center visualization, and so forth. Improved hardware and in Bangladesh to meet the national requirements. In software should include installation of a modern system that case, the existing Meteorological Training Institute for ingesting and analyzing all data relevant to tropical of BMD can be upgraded to contemporary standards. cyclone forecasts, as is available at larger centers such as the NHC and the IMD. BMD should also ensure backup Coordination between the BMD and other agencies: For systems for its computers in case of power outage. improvements in forecasting to contribute meaningfully to disaster preparedness and improved early warning Enhance data and information sharing through systems, close coordination between BMD and other improved network systems: In addition to agencies such as the BWDB and Department of Disaster improvements in computer hardware and software, Management is needed. There are already strong rela- BMD can have better access to useful data—both from tions between these agencies that can be enhanced. In observations and from numerical weather prediction storm surge forecasting, the World Meteorological Orga- models—that are already, in principle, available through nization’s Coastal Inundation Forecast Demonstration improvements in network systems. BMD should make Project for Bangladesh aims to improve the state of the efforts to increase the bandwidth available for its GTS/ art and is expected to provide important lessons on how WIS link so that it can obtain information and products to improve coastal inundation forecasts in Bangladesh, available regionally and globally. This will also enable and it should be supported. It would be most desirable data and information sharing from BMD with IMD and if a way could be found to integrate the surge and inun- other relevant agencies. dation forecast computations performed in Bangladesh into a single model, ideally through a collaborative effort Strengthen observation network for tropical cyclone involving the BMD, the Bangladesh Water Development forecasting: The BMD would benefit from improvements Board, the Bangladesh Inland Water Transport Authority, and expansion of the current observing system, such the Hydrographic Department of the Bangladesh Navy, as access to automated tide gauges at the coast for the Survey of Bangladesh, and any other relevant entities measurements of storm surge, better bathymetric and for bathymetry, water level, river discharge, and other topographic data (DEM) for storm surge and inundation data sharing. It would be particularly desirable to devel- forecasting, and calibration of coastal radars, as well as op a probabilistic (as opposed to a deterministic) fore- intercomparison against rain gauges and installation cast capability for both surge and inundation, accounting of buoys for oceanographic and meteorological observations offshore. for the uncertainties in storm track and intensity. The IMD’s forecasts using the IIT Delhi and INCOIS models Training and capacity building: Improved education can also be generalized to produce a probabilistic fore- and training opportunities for BMD staff are critical in cast of storm surge based on a range of simulations to improving the agency’s capacity for improved service allow for uncertainties in forecasts of track and intensity. delivery. In the short term, the goal of improving Perhaps most important, the critical problems of forecast forecasts requires more training of BMD personnel evaluation and verification need to be addressed by im- in existing and developing science and technology provements to the observational network—in particular, that specifically address that goal: Dvorak technique, automated gauges that can measure water levels both at numerical weather prediction, ensemble prediction the coast and inland. 52 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Strengthen regional and global collaboration The use of ensemble forecasts for improving lead time including IMD’s role in research, technology transfer for tropical cyclone forecasting should be actively and training: The RSMC at IMD is the regional center studied: The possibility of producing ensemble for predicting tropical cyclones and storm surges forecasts for South Asia with lead times greater than five and its role as a coordinator of research, technology days, as recommended by Webster (2008, 2012, 2013) transfer, and training should be strengthened for the and Belanger et al. (2013), should be actively studied. benefit of the IMD, the BMD, and other operational Such long-range forecasts, including both genesis and agencies in the region. The RSMC could play a role subsequent storm behavior, would be of tremendous broadly comparable to that of the NOAA Hurricane value for Bangladesh and South Asia more broadly. Research Division in the United States (in addition to Greater involvement of the international research the role it already plays, which is more analogous to community in this task is important in this, as advocated that of the NHC, the U.S. operational forecast center). by Webster (2013). The active interest and engagement The research activities on tropical cyclones and storm of the local agencies—the BMD and the IMD—is essential surge could be strengthened at Indian universities such to this effort, as they bear responsibility for forecasting as Indian Institute of Technology Delhi, another IIT, the for the nation of Bangladesh. They are the entities Indian Institute of Science Bangalore, or IITM Pune, and responsible for predicting tropical cyclones and storm these could be more coordinated with the operational surges and, ultimately, for protecting life and property activities at IMD. A broader range of basic and applied in the region. Any effort toward the goal of longer- research could be carried out, focused on implementing range forecasts must recognize that forecasts of track the latest developments in prediction of tropical and intensity that are issued before genesis are being cyclones and storm surges and transferring them into issued to the public by even the more sophisticated operations. Regular training could be carried out, and national forecasting centers only recently. If such personnel from the forecast centers could also spend forecasts are to become operational in South Asia in longer periods in residence for more in-depth training any useful way, it will only be with the engagement of and collaborative research. In addition, the BMD should the local agencies, requiring a substantial period of reach bilateral agreements with the IMD and perhaps development and testing. Efforts should be made for other entities to obtain useful data that are not publicly capacity development toward this end through training available on the WIS. Obtaining satellite data in proper of forecasters, knowledge transfer from industrial digital form on workstations with software designed countries, and development of software tools for for Dvorak analysis is important. Obtaining tropical visualization of ensemble products in a graphical user cyclone tracks produced by global model ensembles in interface and geographic information system–based a software environment designed to use them as inputs platform. Important questions in this regard include not to the forecast process is perhaps even more important. just those addressed in this report but also the extent to Regional collaboration in training, as mentioned earlier, which the greater uncertainties associated with longer could support BMD forecasters to become better lead-time forecasts may be compatible with their use in acquainted with the capabilities of the modern global emergency management. model ensembles that produce these tracks. Review of Operational Practices and Implications for Bangladesh / 53 ANNEX 1: LIST OF STAKEHOLDERS CONSULTED Name Title Organization Md. Shah Alam Former Director Bangladesh Meteorology Department Arjuman Habib Former Director Bangladesh Meteorology Department Shamsuddin Ahmed Director Bangladesh Meteorology Department Taslima Imam Meteorologist Bangladesh Meteorology Department Md. Amirul Hussain Executive Engineer Bangladesh Meteorology Department Dr. L. S. Rathore Former Director General India Meteorological Department Md. Mobassarul Hasan Senior Specialist Institute of Water Modeling, Bangladesh Zahir-ul Haque Khan Director Institute of Water Modeling, Bangladesh M. Mohapatra, Scientist India Meteorological Department B. K. Bandyopadhyay Scientist India Meteorological Department S. K. Roy Bhowmik Scientist India Meteorological Department U. C. Mohanty Professor India Institute of Technology A. D. Rao Professor India Institute of Technology S. K. Dube Emeritus India Institute of Technology Koji Kuroiwa Former Chief, Tropical Cyclone Programme World Meteorological Organization Maryam Golnaraghi Chief, Disaster Risk Reduction Programme World Meteorological Organization Michael Brennan Senior Hurricane Specialist National Hurricane Center Eric Blake Hurricane Specialist National Hurricane Center Todd Kimberlain, Hurricane Specialist National Hurricane Center Chris Landsea Science and Operations Officer National Hurricane Center Stacy Stewart Senior Hurricane Specialist National Hurricane Center Jamie Rhome Storm Surge Team Lead National Hurricane Center Frank Marks Research Meteorologist and Director NOAA Hurricane Research Division S. G. Gopalakrishnan Research Meteorologist NOAA Hurricane Research Division Sim Aberson Research Meteorologist NOAA Hurricane Research Division Tim Neely Chief, Environment, Science & Technology Affairs U.S. Embassy, New Delhi Dec Ly Science and Technology Officer U.S. Embassy, New Delhi Robert Falvey Director Joint Typhoon Warning Center James Doyle Head, Mesoscale Modeling Section Naval Research Laboratory Russell Elsberry Distinguished Research Professor Naval Postgraduate School Frederic Vitart Senior Scientist European Centre for Medium-Range Weather Forecasting Don Resio Professor U. North Florida, Coastal Inundation Forecast Project Andrew Burton Regional Forecasting Centre Operations Manager Australian Bureau of Meteorology Treng-Shi Huang Taiwan Central Weather Bureau Chun-Chieh Wu Professor National Taiwan University Michela Biasutti Lamont Associate Research Professor Columbia University Suzana Camargo Lamont Research Professor Columbia University Michael Tippett Associate Professor, Applied Mathematics Columbia University Peter Webster Professor Georgia Tech University Storm over Dhaka. Credit: crazydiva/Thinkstock.com Review of Operational Practices and Implications for Bangladesh / 55 REFERENCES Aksoy, A., S. Lorsolo, T. Vukicevic, K. H. Sellwood, S. D. Dvorak, V. 1984. Tropical Cyclone Intensity Analysis Using Aberson, and F. Zhang. 2012. “The HWRF Hurricane Satellite Data. NOAA Technical Report NESDIS 11. Ensemble Data Assimilation System (HEDAS) for Washington, DC: National Oceanic and Atmospheric High-Resolution Data: The Impact of Airborne Doppler Administration, Satellite and Information Service. Radar Observations in an OSSE.” Monthly Weather Review 140 (6): 1843–62. Gall, R., J. Franklin, F. Marks, E. N. Rappaport, and F. Toepfer. 2013. “The Hurricane Forecast Improvement Project.” Belanger, J. I., P. J. Webster, J. A. Curry, and M. T. Jelinek. Bulletin of the American Meteorological Society 94 (3): 2012. “Extended Prediction of North Indian Ocean 329–43. Tropical Cyclones.” Weather and Forecasting 27: 757–69. Glahn, B., A. Taylor, N. Kurkowski, and W. A. Shafer. 2009. “The Role of the SLOSH Model in National Weather Bessafi, M., and M. C. Wheeler. 2006. “Modulation of South Service Storm Surge Forecasting.” National Weather Indian Ocean Tropical Cyclones by the Madden-Ju- Digest 33 (1): 3–14. lian Oscillation and Convectively Coupled Equatorial Waves.” Monthly Weather Review 134 (2): 638–56. Goerss, J. S., C. R. Sampson, and J. Gross. 2004. “A History of Western North Pacific Tropical Cyclone Track Forecast Camargo, S. J., M. C. Wheeler, and A. H. Sobel. 2009. “Diag- Skill.” Weather and Forecasting 19 (3): 633–38. nosis of the MJO Modulation of Tropical Cyclogenesis Using an Empirical Index.” Journal of the Atmospheric Government of the People’s Republic of Bangladesh. 2010. Sciences 36 (10): 3061–74. National Plan for Disaster Management, 2010–2015. At http://www.dmb.gov.bd/reports/npdm_final.pdf. Demaria, M., and J. M. Kaplan. 1994. “A Statistical Hurricane Intensity Prediction Scheme (SHIPS) for the Atlantic Habib, A., Md. Shahidullah, and D. Ahmed. 2012. “The Basin.” Weather and Forecasting 9 (2): 209–20. Bangladesh Cyclone Preparedness Program. A Vital Component of the Nation’s Multi-Hazard Early Warning Demaria, M., and J. M. Kaplan. 1999. “An Updated Statistical System.” In Institutional Partnerships in Multi-Hazard Hurricane Intensity Prediction Scheme (SHIPS) for the Early Warning Systems, ed. M. Golnaraghi, 29–62. Atlantic and Eastern North Pacific Basins.” Weather Berlin: Springer. and Forecasting 14 (3): 326–37. Halperin, D. J., H. E. Fuelberg, R. E. Hart, J. H. Cossuth, P. DeMaria, M., and J. M. Gross. 2003. “Evolution of Prediction Sura, and R. J. Pasch. 2013. “An Evaluation of Tropical Models.” In Hurricane! Coping with Disaster: Progress Cyclone Genesis Forecasts from Global Numerical Mod- and Challenges since Galveston, 1900, ed. R. Simpson, els.” Weather and Forecasting 28 (6): 1423–45. R. Anthes, and M. Garstang. Washington, DC: American Geophysical Union. Heming, J., and J. Goerss. 2009. “Track and Structure Forecasts of Tropical Cyclones.” Global Perspectives DeMaria, M., M. Mainelli, L. K. Shay, J. A. Knaff, and J. Kaplan. on Tropical Cyclones, ed. J. C. L. Chan and J. Kepert, 2005. “Further Improvements to the Statistical Hurri- 298–323. Singapore: World Scientific. cane Intensity Prediction Scheme (SHIPS).” Weather and Forecasting 20 (4): 531–43. JCOMM (Joint WMO-IOC Technical Commission for Ocean- ography and Marine Meteorology). 2013. “Coastal Dube, S. K., T. S. Murty, J. C. Feyen, R. Cabrera, B. A. Harper, Inundation Forecasting Development Project (CIFDP).” J. D. Bales, and S. Amer. 2009. “Storm Surge Modeling At http://www.jcomm.info/index.php?option=com_ and Applications in Coastal Areas.” In Global Per- oe&task=viewDocumentRecord&docID=8918. spectives on Tropical Cyclones, ed. J. C. L. Chan and J. Kepert, 363–406. Singapore: World Scientific. Jelesnianski, C. P., J. Chen, and W. A. Shafer. 1992. SLOSH: Sea, Lake, and Overland Surges from Hurricanes. Dupont, T., M. Plu, P. Carof, and G. Faure. 2011. “Verification Technical Report. Washington, DC: National Oceanic of Ensemble-Based Uncertainty Circles Around Tropical and Atmospheric Administration, U.S. Department of Cyclone Track Forecasts.” Weather and Forecasting 26 Commerce. (5): 664–76. JTWC (Joint Typhoon Warning Center). 2011. Annual Tropical Cyclone Report, 2011. Technical Report. Pearl Harbor, HI: JTWC. 56 / IMPROVING LEAD TIME FOR TROPICAL CYCLONE FORECASTING Kiladis, G. N., M. C. Wheeler, P. T. Haertel, K. H. Straub, and Taflanidis, A., A. Kennedy, J. Westerink, J. Smith, K. Cheung, P. E. Roundy. 2009. “Convectively Coupled Equatorial M. Hope, and S. Tanaka. 2012. “Rapid Assessment of Waves.” Review of Geophysics 47 (2). Wave and Surge Risk during Landfalling Hurricanes: Probabilistic Approach.” Journal of Waterway, Port, Krishnamurti, T. N., C. M. Kishtawal, T. LaRow, D. Bachiochi, Coastal, Ocean Engineering 139 (3): 171–82. Z. Zhang, C. E. Williford, S. Gadgil, and S. Surendran. 1999. “Improved Weather and Seasonal Climate Fore- Vitart, F., F. Prates, A. Bonet, and C. Sahin. 2011. “New Trop- casts from Multimodel Superensemble.” Science 285 ical Cyclone Products on the Web.” ECMWF Newsletter (5433): 1548–50. 130 (Winter): 17–23. Landsea, C. W., and J. L. Franklin. 2013. “How ‘Good’ Are Webster, P. J. 2008. “Myanmar’s Deadly Daffodil.” Nature the Best Tracks? Estimating Uncertainty in the Atlantic Geoscience 1: 488–90. Hurricane Database.” Monthly Weather Review. In press. doi:10.1175/MWR-D-12-00254.1. –––. 2012. Bay of Bengal Tropical Cyclones and Severe Convective Systems: Predictability, Prediction and the Lin, N., K. Emanuel, M. Oppenheimer, and E. Vanmarcke. Impacts of Climate Change. Washington, DC: World 2012. “Physically Based Assessment of Hurricane Surge Bank commissioned background report for the Bangla- Threat under Climate Change.” Nature Climate Change desh Sundarbans NLTA. 2: 462–67. ———. 2013. “Improve Weather Forecasts for the Developing Lorenz, E. N. 1963. “Deterministic Nonperiodic Fow.” Journal World.” Nature 493 (7430): 17–18. of the Atmospheric Sciences 20 (2): 131–40. Weng, Y., and F. Zhang, 2012. “Assimilating Airborne Dop- Marks, D. 1992. The Beta and Advection Model for Hurricane pler Radar Observations with an Ensemble Kalman Track Forecasting. Technical Report. Washington, DC: Filter for Convection-permitting Hurricane Initializa- National Oceanic and Atmospheric Administration, U.S. tion and Prediction: Katrina (2005).” Monthly Weather Department of Commerce. Review, 140, 841-859. Mathur, M. 1991. “The National Meteorological Center’s WMO (World Meteorological Organization). 1992. WMO Quasl-Lagrangian Model for Hurricane Prediction.” Global Guide to Tropical Cyclones. Geneva: WMO. Monthly Weather Review 119 (6): 1419–47. ———. 2012. Tropical Cyclone Operational Plan for the Bay Mohapatra, M., D. P. Nayak, R. P. Sharma, and B. K. Bandyop- of Bengal and the Arabian Sea. World Meteorological adhyay. 2013. “Evaluation of Official Tropical Cyclone Organization Technical Document WMO/TD 84, and Track Forecast over North Indian Ocean Issued by India Tropical Cyclone Programme Report TCP-21., Secretar- Meteorological Department.” Journal of Earth System iat of the World Meteorological Organization, Geneva, Science 122 (3): 589–601. Switzerland. Mohapatra, M., B. K. Bandyopadhyay, and D. P. Nayak. 2013. Zhang, C. 2005. “Madden-Julian Oscillation.” Review of Geo- “Evaluation of Operational Tropical Cyclone Intensity physics 43 (2). doi:10.1029/2004RG000158. Forecasts over North Indian Ocean Issued by India Meteorological Department.” Natural Hazards 68 (2): Zhang, F., Y. Weng, J. F. Gamache, and F. D. Marks. 2011. 433–51. “Performance of Convection- Permitting Hurricane Initialization and Prediction during 2008–2010 with Olander, T. L., and C. S. Velden. 2007. “The Advanced Dvorak Ensemble Data Assimilation of Inner-Core Airborne Technique: Continued Development of an Objective Doppler Radar Observations.” Geophysical Research Scheme to Estimate Tropical Cyclone Intensity Using Letters 38 (15). L15 810, doi:10.1029/2011GL048 469. Geostationary Infrared Satellite Imagery.” Weather and Forecasting 22 (2): 287–98. Zhang L., Z. Pu, W.-C. Lee, and Q. Zhao, 2012. “The Influence of Airborne Doppler Radar Data Quality on Numerical RSMC (Regional Specialized Meteorological Centre). 2012. Simulations of a Tropical Cyclone.” Weather and Fore- Report on Cyclonic Disturbances over North Indian casting 27: 231–39. Ocean during 2011. RSMC Tropical Cyclone Report 02/2012. New Delhi: India Meteorological Department, RSMC Tropical Cyclones.