Wind Resource Mapping in Zambia MESOSCALE WIND MODELLING REPORT JULY 2015 This report was prepared by DNV GL, under contract to The World Bank. It is one of several outputs from the wind Resource Mapping and Geospatial Planning [Project ID: P145271]. This activity is funded and supported by the Energy Sector Management Assistance Program (ESMAP), a multi-donor trust fund administered by The World Bank, under a global initiative on Renewable Energy Resource Mapping. Further details on the initiative can be obtained from the ESMAP website. This document is an interim output from the above-mentioned project. Users are strongly advised to exercise caution when utilizing the information and data contained, as this has not been subject to full peer review. The final, validated, peer reviewed output from this project will be the Zambia Wind Atlas, which will be published once the project is completed. Copyright © 2015 International Bank for Reconstruction and Development / THE WORLD BANK Washington DC 20433 Telephone: +1-202-473-1000 Internet: www.worldbank.org This work is a product of the consultants listed, and not of World Bank staff. The findings, interpretations, and conclusions expressed in this work 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 and accept no responsibility for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. 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 non-commercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: +1-202-522-2625; e-mail: pubrights@worldbank.org. Furthermore, the ESMAP Program Manager would appreciate receiving a copy of the publication that uses this publication for its source sent in care of the address above, or to esmap@worldbank.org. RENEWABLE ENERGY WIND MAPPING FOR ZAMBIA Mesoscale Wind Modeling Report 1- Interim wind atlas for Zambia The World Bank Document No.: 702833-USSD-R03-D Issue: D, Status: FINAL Date: 3 July 2015 IMPORTANT NOTICE AND DISCLAIMER 1. This document is intended for the sole use of the Customer as detailed on the front page of this document to whom the document is addressed and who has entered into a written agreement with the DNV GL entity issuing this document (“DNV GL”). To the extent permitted by law, neither DNV GL nor any group company (the "Group") assumes any responsibility whether in contract, tort including without limitation negligence, or otherwise howsoever, to third parties (being persons other than the Customer), and no company in the Group other than DNV GL shall be liable for any loss or damage whatsoever suffered by virtue of any act, omission or default (whether arising by negligence or otherwise) by DNV GL, the Group or any of its or their servants, subcontractors or agents. This document must be read in its entirety and is subject to any assumptions and qualifications expressed therein as well as in any other relevant communications in connection with it. This document may contain detailed technical data which is intended for use only by persons possessing requisite expertise in its subject matter. 2. This document is protected by copyright and may only be reproduced and circulated in accordance with the Document Classification and associated conditions stipulated or referred to in this document and/or in DNV GL’s written agreement with the Customer. This document may not be disclosed in any prospectus or stock exchange listing without the express and prior written consent of DNV GL. A Document Classification permitting the Customer to redistribute this document shall not thereby imply that DNV GL has any liability to any recipient other than the Customer. 3. This document may be printed or download provided that: no documents or related graphics in this document are modified in any way; no graphics in the document are used separately from the corresponding text; DNV GL’s copyright and trade mark notices and any permission noted appear in all copies; no attempt is made to interfere with the intended and efficient functioning of the document, and no attempt is made to reverse engineer, disassemble, or decompile the document or any information herein, and no process or procedure is made to derive the source code of any software included herein. 4. This document has been produced from information relating to dates and periods referred to in this document. This document does not imply that any information is not subject to change. Except and to the extent that checking or verification of information or data is expressly agreed within the written scope of its services, DNV GL shall not be responsible in any way in connection with erroneous information or data provided to it by the Customer or any third party, or for the effects of any such erroneous information or data whether or not contained or referred to in this document. 5. Any energy forecasts estimates or predictions are subject to factors not all of which are within the scope of the probability and uncertainties contained or referred to in this document and nothing in this document guarantees any particular wind speed or energy output. KEY TO DOCUMENT CLASSIFICATION For disclosure only to named individuals within the Customer’s Strictly Confidential : organisation. For disclosure only to individuals directly concerned with the Private and Confidential : subject matter of the document within the Customer’s organisation. Commercial in Confidence : Not to be disclosed outside the Customer’s organisation. DNV GL only : Not to be disclosed to non-DNV GL staff Distribution for information only at the discretion of the Customer (subject to the above Important Notice and Disclaimer and the Customer’s Discretion : terms of DNV GL’s written agreement with the Customer). Available for information only to the general public (subject to the Published : above Important Notice and Disclaimer). DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page i www.dnvgl.com Project name: Renewable Energy Wind Mapping for Zambia DNV GL - Energy Report title: Mesoscale Wind Modeling Report 1 Renewables Advisory Customer: The World Bank, 9665 Chesapeake Drive, Suite 435 1818 H Street, N.W. San Diego, CA 92123 Washington, DC 20433 Tel: 703-795-8103 Contact person: Francesca Fusaro Enterprise No.: 94-3402236 Date of issue: 3 July 2015 Project No.: 702833 Document No.: 702833-USSD-R03-D Issue/Status D/FINAL Task and objective: Present the results of the interim mesoscale wind mapping for Zambia, including a detailed summary of assumptions and methodology used. Prepared by: Verified by: Approved by: Daran Rife Craig Houston Richard Whiting Global Head of Mesoscale Modeling Senior Advisor, Strategy & Policy Service Line Leader, Project Development Jessica Ma Global Specialist – Mesoscale Modeling Gemma Ebsworth/Caroline Donhue ☐ Strictly Confidential Keywords: ☐ Private and Confidential World Bank, Zambia, wind, mesoscale, mapping, ☐ Commercial in Confidence methodology, GIS ☐ DNV GL only ☒ Customer’s Discretion ☐ Published © Garrad Hassan America, Inc.. All rights reserved. Reference to part of this report which may lead to misinterpretation is not permissible. Issue Date Reason for Issue Prepared by Verified by Approved by D 3 July 2015 FINAL B 22 December 2014 FINAL C 17 April 2015 FINAL DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page ii www.dnvgl.com Table of contents 1 INTRODUCTION ............................................................................................................................. 0 1.1 Project description ....................................................................................................................... 0 2 METHODOLOGY – DNV GL WIND MAPPING SYSTEM............................................................................ 2 3 RESULTS ....................................................................................................................................... 4 3.1 Preliminary and Unvalidated Wind Speed Map ................................................................................. 4 3.2 Preliminary and Unvalidated Wind Speed Uncertainty Index .............................................................. 7 3.3 Preliminary and Unvalidated Wind Energy Map ...............................................................................10 3.4 Preliminary and Unvalidated Seasonal and Diurnal Trends ...............................................................12 3.5 Preliminary and Unvalidated Estimates of Interannual Variability ......................................................14 3.6 Interim Wind Atlas (*.lib files) .....................................................................................................15 4 REFERENCES ................................................................................................................................17 APPENDIX A – DETAILED METHODOLOGY ...........................................................................................18 APPENDIX B – PRELIMINARY AND UNVALIDATED MESOSCALE OUTPUTS ................................................35 List of tables Table 3-1 Representative energy loss factors applied........................................................................... 10 Table 3-2 “12 x 24” matrix of energy production (% of annual mean energy production) ......................... 12 List of figures Figure 2-1 Mesoscale computational grid configuration used for the DNV GL WMS simulations .................... 2 Figure 3-1 Preliminary and unvalidated mesoscale wind sped map at 100 m AGL, created using the DNV GL Wind Mapping System. ...................................................................................................................... 5 Figure 3-2 The multiphysics ensemble approach ................................................................................... 7 Figure 3-3 Preliminary and unvalidated wind speed uncertainty index created with the DNV GL Wind Mapping System ............................................................................................................................................ 9 Figure 3-4 Preliminary and unvalidated wind energy map created using the DNV GL Wind Mapping System 11 Figure 3-5 Seasonal variation of energy production for a selected grid cell ............................................. 13 Figure 3-6 Seasonal flow pattern over Zambia as simulated by the DNV GL Wind Mapping System ........... 14 Figure 3-7 Preliminary and unvalidated interannual variability index (IAVI) for wind speed ...................... 16 DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 3 www.dnvgl.com 1 INTRODUCTION The results described in this report are derived from interim output and are preliminary and unvalidated, and they have not been subjected to full peer review. DNV GL does not guarantee the accuracy of the maps, data, and visualizations presented in this report, and accepts no responsibility whatsoever for any consequence of their use. Wind speed values shown in tables and maps should not be relied upon in an absolute sense. Rather they should be strictly interpreted as indicative (e.g., elevated windiness near mountaintops and escarpments). Users are strongly urged to exercise caution when using the information and data contained within this report. During Phase 2 of this project, measurements will be collected from a number of representative sites throughout the country over a 24 month period, and these will be used in Phase 3 to develop a final, validated, peer-reviewed suite of outputs from this project, which will be made available at the project’s completion. A new 2-km-resolution mesoscale wind atlas has been generated by DNV GL for the entire country of Zambia, providing for the first time information on the potential resource. It is based on a complete 10-year simulation of the local and regional wind flows, and will serve as the foundation to a broader program of sustainable development of renewable energy in Zambia, while dramatically increasing the awareness of the available resources within the country to both policy makers and potential investors. A unique characteristic of this new 2 km resolution interim wind atlas is the availability of hourly output over the entire 10-year simulation period, which represents the full range of wind and thermal stratification conditions over Zambia. It allows the monthly, seasonal, yearly, and interannual variation in the wind resource to be fully quantified. This unprecedented level of granularity will set the benchmark for a new era in the renewable energy mapping. 1.1 Project description Zambia is still in the early stages of exploring the resource potential for wind power, and to date there are no utility scale wind turbines operating in the country. Furthermore, only a small number of low elevation meteorological masts exist, presenting a significant barrier to policymakers interested in evaluating the potential for wind energy in Zambia. The key goal of this project is to provide Zambian policy makers and stakeholders with accurate and valuable knowledge of the national wind resource which has direct applicability both for formulating energy policy and implementing wind projects. The installation and operation of high quality wind measurements throughout the country will also strengthen local capacity to support future development of wind projects in Zambia. The primary deliverable supporting this goal is a well-validated national mesoscale wind resource atlas for Zambia. This atlas will greatly improve the awareness and understanding of the locations with the greatest potential for wind energy. When used in combination with a Geographic Information System (GIS), this DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 0 www.dnvgl.com yields a highly valuable planning tool to enable policy makers to develop sound energy strategy, and will stimulate commercial wind development by removing the current knowledge barrier. This Mesoscale Wind Modeling Report documents the preliminary mesoscale modeling approach and results, and has been used as a primary input into the identification of suitable measurement locations to support both the validation of the atlas and the Candidate Site Identification Report 1]. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 1 www.dnvgl.com 2 METHODOLOGY – DNV GL WIND MAPPING SYSTEM The new interim wind atlas has been generated using the DNV GL Wind Mapping System (WMS). WMS is a dynamical downscaling system developed to generate high-resolution mesoscale wind maps for any part of the world. Full details for the methods and inputs to the WMS appear in Appendix A – Detailed Methodology. The WMS configuration for this project employs telescoping, one-way interacting computational grids to achieve a resolution of 2 km across all of Zambia (see Fig 2-1). Figure 2-1 Mesoscale computational grid configuration used for the DNV GL WMS simulations A complete hourly time series with the 10 km resolution grid is performed over the entire 15 April 2004-15 April 2014 period is first generated. Next, a nested 2 km grid is used to provide an hourly time series over the last year within the 10 km resolution simulation (15 April 2013 to 15 April 2014). These outputs provide the basis for the high resolution downscaling performed with the DNV GL Analog Ensemble, which is further described in Appendix A, to yield a complete 10-year times series at 2 km resolution over the entire country. A unique characteristic of this new 2 km resolution interim wind atlas is the availability of hourly output over the entire 10-year simulation period, which represents the full range of wind and thermal stratification conditions over Zambia. It will allow the monthly, seasonal, yearly, and interannual variation in the wind DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 2 www.dnvgl.com resource to be quantified. This unprecedented level of granularity sets the benchmark for renewable energy mapping over the coming years. Furthermore, horizontal grid spacing of 2 km allows the topographic features of Zambia to be modelled with much higher fidelity. This leads to more accurate results because it enables the diurnal variation of processes in the planetary boundary layer (PBL), as well as the local forcing, to be well represented. The preliminary wind speed map, and other associated derived outputs are presented and discussed in the following sections. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 3 www.dnvgl.com 3 RESULTS 3.1 Preliminary and Unvalidated Wind Speed Map Figure 3-1 shows the interim atlas for 100 m above ground level (AGL). Again, the atlas is based on a complete 10-year hourly simulation at 2 km resolution. This provides for the first available quantitative map of the long-term wind resource across the country. A number of interesting features of the wind climate in Zambia emerge from this map: Muchinga Escarpment: The best wind resource appears to lie along the entire length of the Muchinga Escarpment to the border with Tanzania in the northeast. High wind speeds extend well beyond the leading edge of the escarpment and across the plateau and extend to the northwest. Eastern province: The Zambia borders with Mozambique and Malawi appear to contain promising wind resource, which lies over relatively smooth terrain. Central province: The central part of Zambia contains good wind resource potential that is positioned along existing infrastructure and load centers. Luangwa river basin and Lake Kariba: These low lying regions have poor exposure to the predominantly easterly winds, and thus have some of the country’s lowest wind speeds. It is clear that the best resources lie mainly along the higher elevation locations, which is perhaps to be expected. While the new atlas represents the best possible estimate of the wind resource within the region using the latest advances in mesoscale modeling, it does have inherent limitations. First, small or localized landscape features are poorly resolved or not resolved at all. For instance, narrow canyons or inlets, isolated mountains and bluffs, or small clearings in forested areas are not resolved by this dataset. As a general rule of thumb, the WRF model is capable of resolving spatial variations in the wind field at scales ranging from about 3-7 grid lengths [2]. In the present case, this equates to features with length scales of about 15 km and larger. Thus, an essential future step is to further downscale these results using a microscale model to account for the impacts of very localized features. Such modeling will be undertaken during Phase 3 of the project, and will incorporate the validation using ground based measurements collected during Phase 2. A full set of wind speed maps at the following heights (AGL) are shown in Appendix A – Detailed Methodology: 10 m, 20 m, 50 m, 80 m, 100 m, and 200 m. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 4 www.dnvgl.com Figure 3-1 Preliminary and unvalidated mesoscale wind sped map at 100 m AGL, created using the DNV GL Wind Mapping System. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 5 www.dnvgl.com DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 6 www.dnvgl.com 3.2 Preliminary and Unvalidated Wind Speed Uncertainty Index It is well known that mesoscale model simulations are subject to a number of unavoidable errors and uncertainties, including the input datasets, the lateral boundary conditions, the numerical approximations used in the dynamical core of the model, and the imperfect representation of the complex physical processes that strongly drive the winds within the boundary layer. DNV GL has employed a Monte Carlo method to mitigate these errors, and to provide statistically defined estimates of the metrological uncertainty in the mesoscale results. The “multiphysics ensemble”, illustrated in Figure 3-2, amounts to constructing several unique versions of the model by selecting different suites of physical process parameterizations, and each model version is then used to perform a separate parallel downscaled simulation. For this project, DNV GL has selected 9 ensemble members for this analysis. Each of the ensemble simulations provides an equally plausible representation of the wind climate, within the likely range of meteorological uncertainty. See Appendix A – Detailed Methodology for more details. The “ensemble spread” or standard deviation among the ensemble of simulations provides an objective estimate of the meteorological uncertainty, allowing us to place a statistically defined error bar on the resource estimate for any point within the numerical wind map/atlas. WRF physics 1 WRF Ensemble physics 2 Ensemble Raw variance MERRA WRF Ensemble physics 3 WRF physics 9 Simulation 1 Simulation 2 Simulation 3 Simulation 4 Simulation 5 Preliminary wind speed uncert. index Simulation 6 Simulation 7 Simulation 8 Simulation 9 Figure 3-2 The multiphysics ensemble approach DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 7 www.dnvgl.com The preliminary wind speed uncertainty index is defined as the standard deviation of the nine multiphysics mesoscale ensemble member solutions, as shown in Figure 3-3. Areas with a high index value, and therefore high standard deviation, indicate where there is a lack of consensus between the nine multiphysics ensemble members. This highlights the apparent difficulty in accurately modeling the wind flows in these areas. During Phase 3 of the project, ground-based measurements and microscale modeling will be used to refine the uncertainty index. Interestingly, high uncertainty index values are not strictly tied to rugged terrain. For example the uncertainty index values are relatively low in the Central, Northern, and Luapula Provinces, where the terrain is generally more rugged. Lower uncertainty index values are also exhibited in the Western and North-Western provinces, where the terrain is smoothly varying. The highest overall uncertainty index values lie along the border regions with Zimbabwe, Mozambique, and Malawi, and may be directly tied to the terrain features in these areas. It is perhaps to be expected that areas of rugged terrain are subject to higher overall uncertainty index values in the mesoscale map, as the increased orographic roughness, eddy shedding, and other similar non-linear affects will lead to stronger overall variability in the wind field, and therefore the wind flows in these areas are more challenging to accurately predict. Still it remains unclear why many of the other mountainous regions within Zambia do not exhibit such high uncertainty index values, particularly the mountains within the border region with Tanzania. This requires further study, and the measurements collected during Phase 2 of the project, together with microscale modeling, will be key to understanding and developing a more refined uncertainty index. One noteworthy aspect of the present analysis is that the strongest overall uncertainty index values lie in the lower Zambezi River where Zambia, Zimbabwe, and Mozambique meet. Examination of the seasonal and annual wind flows in this area (see Section 3.4) indicates a varying degree of inland penetration of the easterly trade winds from the Indian Ocean, which are prevalent throughout the year, but are particularly strong during between June and November. Two of the ensemble members indicate very limited inland penetration, and lower overall wind speeds in this area, while 3 other members indicate a deep penetration of strong winds. There is also a large diversity in the orientation of the winds in this area, with some ensemble members indicating a more southeasterly orientation, while others exhibit a more easterly orientation. All of this leads to generally low overall agreement between the 9 ensemble member solutions in this area, which likely results from the winds being more difficult to predict. The wind speed uncertainty index has been used along with other important parameters to guide the recommendations for the Candidate Identification Report [1]. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 8 www.dnvgl.com Figure 3-3 Preliminary and unvalidated wind speed uncertainty index created with the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 9 www.dnvgl.com 3.3 Preliminary and Unvalidated Wind Energy Map The preliminary wind energy map is presented in Figure 3-4. Here the values represent the approximate net annual energy production (GWh/annum) that would be produced by a single “generic wind turbine”, if it were to experience the long-term mean wind conditions simulated in the 2 km preliminary mesoscale wind map at 100 m. These calculations involve a number of assumptions, which are described in more detail in Appendix A, including a description of the generic wind turbine selected by DNV GL. The details of this turbine are among the many topics that DNV GL is keen to solicit feedback for, both from the Client and key stakeholders as part of the Phase 1 workshop. The methodology used is outlined as follows. For each hourly record from the 10-year simulation, the simulated air density and wind speed value at 100 m AGL are used to create a power curve at each 2 km grid cell. This results in database of (24 hours × 3650 days) hourly power values specific to the speed and density at each time and grid cell location. From this 10-year hourly power time series, the mean annual gross energy in GWh for each 2km grid point for each individual year is calculated. The individual yearly results at each grid point are then combined to give the long-term mean energy output. Finally, the values presented here include consideration of a set of high-level starting assumptions for typical energy losses which may be expected for wind farm development in this region. The list of losses applied to generate the preliminary wind energy map is shown in Table 3-1 below. Table 3-1 Representative energy loss factors applied Loss Factor Value Wake effect 93.0% Availability 92.0% Electrical efficiency 97.0% Turbine Performance 98.5% Environmental 99.0% Curtailments 100.0% Total 80.9% This means that approximately one fifth of the extractable energy content of the calculated wind resource will be lost. While such loses can be lowered through selection of advanced turbine technology, good wind farm design and careful maintenance, they cannot be entirely eliminated on any wind farm. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 10 www.dnvgl.com Figure 3-4 Preliminary and unvalidated wind energy map created using the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 11 www.dnvgl.com 3.4 Preliminary and Unvalidated Seasonal and Diurnal Trends The time series simulation approach used for this project provides for a wealth of additional information, and the following section provides some highlights that are directly relevant to this project. A “12 x 24” matrix for each 2 km grid point was created to estimate the diurnal variation in mean wind speed for each calendar month, as well as a 12 x 24 matrix of wind energy production. These data, over approximately 190,000 grid cells, will be supplied to the Client prior to the Phase 1 workshop. An example of the matrix for energy production is shown below for a single location in Table 3-2. Table 3-2 “12 x 24” matrix of energy production (% of annual mean energy production) Predicted hourly and monthly energy production as a percentage of mean annual energy production Generic 3MW power curve at 100 m AGL Energy production [%] Hour Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Total 00:00 0.25 0.32 0.48 0.57 0.50 0.50 0.56 0.63 0.68 0.74 0.55 0.37 6.15 01:00 0.26 0.34 0.50 0.60 0.53 0.54 0.60 0.66 0.71 0.75 0.55 0.37 6.40 02:00 0.24 0.35 0.52 0.62 0.55 0.56 0.61 0.68 0.71 0.75 0.55 0.36 6.49 03:00 0.25 0.35 0.54 0.64 0.60 0.59 0.64 0.70 0.70 0.72 0.57 0.37 6.66 04:00 0.27 0.38 0.57 0.68 0.63 0.60 0.67 0.72 0.70 0.72 0.56 0.39 6.89 05:00 0.27 0.37 0.56 0.68 0.64 0.61 0.68 0.70 0.67 0.68 0.52 0.35 6.72 06:00 0.19 0.28 0.45 0.61 0.58 0.55 0.65 0.64 0.58 0.54 0.41 0.26 5.74 07:00 0.13 0.20 0.35 0.52 0.51 0.48 0.58 0.52 0.47 0.43 0.31 0.19 4.69 08:00 0.09 0.13 0.25 0.39 0.37 0.38 0.47 0.40 0.37 0.34 0.23 0.14 3.56 09:00 0.07 0.09 0.17 0.29 0.27 0.29 0.37 0.30 0.28 0.25 0.17 0.10 2.65 10:00 0.06 0.07 0.12 0.21 0.19 0.21 0.27 0.23 0.21 0.19 0.13 0.08 1.98 11:00 0.05 0.06 0.10 0.16 0.15 0.17 0.21 0.18 0.18 0.17 0.12 0.07 1.62 12:00 0.05 0.05 0.09 0.13 0.12 0.14 0.18 0.16 0.17 0.16 0.11 0.06 1.42 13:00 0.06 0.06 0.09 0.13 0.12 0.14 0.17 0.15 0.17 0.16 0.12 0.07 1.41 14:00 0.06 0.07 0.10 0.13 0.12 0.13 0.16 0.15 0.17 0.18 0.14 0.09 1.50 15:00 0.07 0.08 0.11 0.13 0.12 0.13 0.16 0.15 0.17 0.18 0.14 0.09 1.53 16:00 0.10 0.12 0.15 0.17 0.16 0.16 0.18 0.18 0.19 0.21 0.17 0.13 1.90 17:00 0.15 0.17 0.21 0.25 0.27 0.25 0.27 0.30 0.28 0.29 0.22 0.18 2.84 18:00 0.18 0.20 0.29 0.34 0.34 0.34 0.36 0.40 0.40 0.43 0.28 0.23 3.80 19:00 0.21 0.25 0.34 0.41 0.40 0.40 0.43 0.47 0.49 0.53 0.37 0.29 4.56 20:00 0.24 0.26 0.38 0.45 0.42 0.43 0.46 0.51 0.54 0.59 0.43 0.32 5.04 21:00 0.22 0.26 0.41 0.49 0.44 0.44 0.47 0.54 0.60 0.65 0.45 0.31 5.27 22:00 0.21 0.28 0.43 0.51 0.45 0.45 0.49 0.58 0.63 0.69 0.49 0.32 5.52 23:00 0.21 0.29 0.43 0.53 0.46 0.47 0.52 0.59 0.66 0.71 0.51 0.32 5.69 Total 3.85 5.05 7.61 9.64 8.93 8.95 10.14 10.51 10.71 11.05 8.08 5.46 100.00 These data can be presented in a number of forms, and access to the underlying data allows the various stakeholders to customise charts and figures for their specific needs. For example, the above matrix can easily be distilled into a monthly profile of energy production as shown in Figure 3-5. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 12 www.dnvgl.com Predicted monthly variation in energy production at 100m agl, ( -15° 21' 18.73", 29° 3' 56.64") 12% Percent annual energy production 10% 8% 6% 4% 2% 0% Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month Figure 3-5 Seasonal variation of energy production for a selected grid cell It is well known that Zambia lies within the influence of the easterly trade winds of the Indian Ocean, which are prevalent throughout the year, but are particularly strong during June to November. For an understanding of this type of regional variation, the seasonal wind flow patterns simulated by the WMS are shown Figure 3-6. It is clear that winds during the height of austral summer (Dec-Feb) are generally weak, as shown in the upper-left hand panel of the figure. However, the situation changes dramatically during austral autumn and winter (Jun-Nov), where the 2-km wind atlas indicates wind speeds of 7-8 m/s over much of Zambia between June and August, with 9-11 m/s wind speeds over the Muchinga Mountains, the Nyika Plateau along the Malawi border in the north, and the high mountains in the Northern Province and along the border of Tanzania. There are also strong winds indicated along the higher elevation areas along the Zambia-Mozambique border. This likely reflects the strengthening of the easterly trades by the strong temperature gradient that points from southeast to northwest, extending from the relatively cool coastal zone of Mozambique, to the warm continental interior of the Democratic Republic of Congo and Angola. These winds flow nearly perpendicular to the Muchinga Mountains in eastern Zambia, and as the air flows up the windward flanks of the mountains, its strength is further enhanced. The countrywide wind speeds weaken somewhat between September and November. The strong wind speeds in the Muchinga Mountains may be particularly attractive areas for wind farm development, as the Great North Road lies along the entire spine of the Muchinga Mountains, and electrical transmission lines parallel this road. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 13 www.dnvgl.com Figure 3-6 Seasonal flow pattern over Zambia as simulated by the DNV GL Wind Mapping System 3.5 Preliminary and Unvalidated Estimates of Interannual Variability An Interannual Variability Index (IAVI) of wind speed and energy is derived from the WMS 10-year simulations and presented in Figure 3-7 for each 2 km grid point. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 14 www.dnvgl.com Care should be taken when using this map, since a robust estimate of interannual variability typically requires at least 30 years of reliable reference data [3]. The IAVI is defined as the standard deviation of the 10 samples available of simulated annual mean wind speed and annual energy production for each 2 km grid point. It is important to note that these preliminary maps of IAVI are not yet validated. Once measurements become available to better understand the mesoscale model performance, the IAVI will be revised and improved. Furthermore, uncertainty models used for wind farm financing have almost exclusively used measurements to derive interannual variability rather than model-derived estimates. 3.6 Interim Wind Atlas (*.lib files) The hourly outputs of the WMS at each grid cell of the 2 km mesoscale domain are used to calculate the generalized wind climate statistics. This is done according to the industry-standard WAsP-format “wind atlas” (“WAsP-lib”) file, according to the European Wind Atlas method (Troen and Petersen, 1989). This was performed using the Danish Technical University (DTU) Generalized Wind Climate software package. When creating the WAsP-lib files there is an option to account for vertical thermal stability, as it is defined by the WRF model (Hahmann et al. 2014). DTU conducted a number of sensitivity and validation tests as part of the Wind Atlas for South Africa, and concluded that accounting for thermal stability in the generalization process resulted in a lower overall wind speed error relative to hub height measurements. DNV GL has performed similar tests at other locations around the world and found overall improvements where night-time stable conditions are prevalent. Although Zambia lacks measurements to estimate the occurrence and frequency of stable conditions, its high terrain elevations (average elevation is 1000 m AMSL), and two distinct dry seasons very likely lead to strong nocturnal radiative cooling, which is conducive to forming vertical temperature inversions. Examination of the WRF-based time series of the Obukhov stability parameter indicates that there is a greater frequency of stable conditions during the dry seasons in Zambia. Thus, we have chosen to account for thermal stability to create the generalized wind climate WAsP-lib files. The full suite of *.lib files will be delivered to the Client in advance of the Phase 1 workshop. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 15 www.dnvgl.com Figure 3-7 Preliminary and unvalidated interannual variability index (IAVI) for wind speed DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 16 www.dnvgl.com 4 REFERENCES [1] Candidate Site Identification Report, DNV GL, 22 December 2014, 702833-USSD-R02-B [2] W. C. Skamarock, 2004: Evaluating Mesoscale NWP Models Using Kinetic Energy Spectra. Mon. Wea. Rev., 132, 3019–3032.doi: http://dx.doi.org/10.1175/MWR2830.1 [3] Raftery A., Tindal J. and Garrad A. (1997), Understanding the risks of financing windfarms, Proc. EWEA Wind Energy Conference, Dublin. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 17 www.dnvgl.com APPENDIX A – DETAILED METHODOLOGY THE DNV GL WMS Model configuration The new interim wind atlas is generated using the DNV GL Wind Mapping System (WMS). WMS is a dynamical downscaling system developed to generate high-resolution mesoscale wind maps for any part of the world. At the heart of WMS is the Weather Research and Forecasting (WRF) model, a state-of-the- science open source mesoscale model developed and maintained by a consortium of more than 150 international agencies, laboratories, and universities (Skamarock et al. 2008). The WRF model has been employed successfully for a spectrum of applications ranging from operational weather forecasting (e.g., Hacker et al. 2011), to climate downscaling (e.g., Leung et al. 2009; Liang et al. 2012; Mearns et al. 2012). And because WRF is one of the most widely used mesoscale models in the renewable energy industry (e.g., Potter et al. 2008; Draxl et al. 2012; Parks et al. 2010), its performance for wind energy applications is well understood. For the application described herein, DNV GL’s WMS utilizes WRF version 3.4.1 to create 2-km hourly analyses on a 2-km horizontal grid, with 36 vertical levels covering the period 15 April 2004-15 April 2014, inclusive. The result is a database of (10 years x 365 days x 24 hours) three-dimensional analyses. The physical process parameterizations are chosen for maximum numerical stability, and for producing the best overall representation of hub height winds within Zambia. These choices are based on DNV GL’s extensive validation studies conducted at more than 500 locations on every major continent, including a number of sites in South Africa. The WMS configuration uses telescoping, one-way interacting computational grids to provide complete coverage for Zambia and the bordering regions (See Figure A-1). Their respective horizontal grid increments are 10 km and 2 km, and both grids use 36 vertically stretched terrain following levels, with approximately 13 levels within the lowest 1.5 km AGL. Figure A-1 Mesoscale computational grid configuration used in the DNV GL WMS simulations DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 18 www.dnvgl.com The 10 km outer grid’s mesh size is 300 × 300, and the nested 2 km grid’s mesh size is 900 × 900. Both grids employ the recommended lateral boundary buffer zone. Model Inputs The U.S. Geological Survey (USGS) Earth Resources Observing System (EROS) 1-km dataset (Loveland et al. 1995) is used to define the vegetation type, which in turn is used to specify the surface aerodynamic roughness length; shown in Figure A-2 and Figure A-3 respectively. There are 5 broadly similar land use classes within Zambia: urban (roughness length of 0.5 m), cropland and pasture (roughness length of 0.1 to 0.15 m), savanna/grassland (roughness length of 0.05 m), forest (roughness length of 0.5 m), and water bodies (roughness length of 0.0001 m). The terrain elevation from the 2 km grid is also shown in Figure A-4. Overall, WRF has a very good representation of the rich diversity of topographic features, including the Muchinga Mountains in the northeast, the Nyika Plateau along the Malawi border in the north, the broad Zambezi River drainage basin, and the Zambezi depression in the south. Even though the 2 km grid provides unprecedented resolution for wind maps within Africa, it is important to note that small or localized landscape features are poorly resolved or not resolved at all. For instance, narrow canyons or inlets, isolated mountains and bluffs, or small clearings in forested areas are not resolved by the model. As a general rule of thumb, the WRF model is capable of resolving spatial variations in the wind field at scales ranging from about 3-7 grid lengths (Skamarock 2004). In the present case, this equates to features with length scales of about 6-14 km and larger. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 19 www.dnvgl.com Figure A-2 Land use types for Zambia as defined by the WRF land cover database Note: The global land cover database used within the WRF model contains 24 distinct categories. There are 5 broadly similar land use classes within Zambia (marked with red dots) DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 20 www.dnvgl.com Figure A-3 Aerodynamic roughness length as defined by WRF DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 21 www.dnvgl.com Figure A-4 Model terrain elevation (m) used in the WRF 2 km grid for Zambia DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 22 www.dnvgl.com Initial land surface conditions for each simulation are based upon the National Aeronautics and Space Administration (NASA) Global Land Data Assimilation System (GLDAS; Rodell et al. 2004) dataset that is defined on a (0.25° × 0.25°) grid. The GLDAS fields used here are substrate soil moisture and temperature, ground skin temperature, and snow water equivalent. These fields are interpolated to all land grid points on each WRF grid. Sea surface temperatures (SSTs) are interpolated to ocean and lake grid points on each WRF model domain using the National Centers for Environmental Prediction (NCEP) global optimum interpolation SST (OISST) dataset (Reynolds et al. 2002) that is defined on a (0.25° × 0.25°) grid, and is updated daily throughout the entire 10-year simulation. These inputs are used in combination with and advanced and robust land surface model to properly represent the land- and sea-surface processes that strongly drive the boundary wind flows. Initial and Lateral Boundary Conditions Individual simulations are initialized using the large-scale (0.5° latitude × 0.67° longitude grid) National Aeronautics and Space Administration’s (NASA) Modern-Era Retrospective Analysis for Research and Applications reanalysis (MERRA; Rienecker et al. 2011). The MERRA analyses also provide the lateral boundary conditions every 6 h during the simulations, which helps preserve consistency between the large- scale features between the WMS and MERRA. MERRA is a state-of-the-science third generation global reanalysis of the satellite era from 1979-present (Rienecker et al. 2011), with assimilation of an extensive set of global measurements. As described in Rife et al. (2013), assimilated data most relevant for wind energy applications are winds from rawinsondes, profilers, NEXRAD radar, land-based stations, aircraft, ship, and QuikSCAT scatterometer winds. An array of satellite-based measurements is also assimilated by MERRA, including data from the entire constellation of NASA Earth Observing System (EOS) satellites. DNV GL has extensively used MERRA for bankable wind resource assessments, and has a thorough understanding of its quality and suitability for long-term reference. It exhibits substantially better correlations to measurements in a broad variety of locations compared to first and second generation reanalysis (such as the NCEP-NCAR reanalysis), making it a superior starting point for mesoscale downscaling. Additionally, the many active evaluations of MERRA appearing in the open peer-reviewed literature demonstrate that its quality is well understood. Steps of the DNV GL WMS process There are 5 distinct steps in the overall WMS modeling process which are detailed below and described in the subsequent sections. • Step 1: Determine period over which MERRA exhibits temporal consistency • Step 2: Perform 10 km resolution simulations for most recent multi-year period, and 2 km resolution nested simulations for the most recent 365-day period • Step 3: Perform analog ensemble (AnEn) downscaling to generate a complete 10-year time series at 2 km resolution • Step 4 : Generalized wind climate modeling for the interim wind atlas DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 23 www.dnvgl.com • Step 5: Preliminary mesoscale wind speed uncertainty index map Step 1: Determine period over which MERRA exhibits temporal consistency The process begins by determining the period over which MERRA exhibits temporal consistency within the client country. For bankable resource prediction applications it is vital that the data set used is consistent. DNV GL has conducted a formal temporal consistency analysis at the global level, using the World Meteorological Organization (WMO) change point analysis technique (Wang, 2008a,b), and finds that all reanalysis datasets are broadly consistent since about January 2000. However, because many developing countries lack dense networks of reliable long-term ground based measurements, the impact of the large changes in the global satellite observing systems over the past 30 years will have a much greater influence on the consistency and accuracy of reanalysis in these regions. Thus, it is important to establish the local and regional consistency of MERRA within the Zambian region. Ideally, comparisons between MERRA and reliable long-term ground based wind datasets would be conducted over the region to determine MERRA’s temporal consistency. As stated in the TOR, Zambia currently lacks measurement data from ground based measurement stations. As an alternative, we investigated the quality of publically available data from WMO airport stations across Zambia and found that the temporal reliability and overall quality of the wind measurements is generally poor, and this prevents us from performing statistically robust and meaningful comparisons to the measurements. As a result, the investigations reported here focus on analyses and statistical tests performed on MERRA itself, as well as comparisons between MERRA and another widely used and thoroughly vetted third generation global analysis: the European Centre for Medium-Range Weather Forecasts Interim Re-Analysis (ERA-Interim; Dee et al, 2011). In the first step, we obtain monthly time series of the MERRA zonal (easterly) and meridional (northerly) wind components at the 925 hPa isobaric level over the full reanalysis period from 1979-present. Each time series represents a spatial average of the northerly and easterly wind components across the Zambia region. The time series were subjected to the WMO change point tests, to identify any statistically significant “climate shifts” or other stepwise changes in the time series over this period. Again, the impact of the large changes in the global satellite observing systems over the past 30 years may have a significant influence on the consistency and accuracy of reanalysis in the Zambia region, given its dearth of ground-based measurement stations. A significant change point is detected in December 1981 at the 95% confidence level in the wind fields, but no other change points were detected. A more detailed analysis is performed at 11 representative sites within Zambia, as shown in Figure A-5. Locations were chosen to provide good geographical coverage of the country, as well as a spectrum of terrain and wind flow complexities. The MERRA hourly time series data at 50 m above ground level (AGL) was obtained for each location from January 2000 to April 2014 (the most recent output available at the start of the project). This is a readily available vertical layer that is closest to the current standard hub heights. The hourly data was averaged to produce a monthly time series, and these were subjected to the WMO change point tests. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 24 www.dnvgl.com Figure A-5 Map of locations (blue dots) used for the detailed change point analysis and tests DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 25 www.dnvgl.com The MERRA time series for all but one location exhibit temporal consistency across the entire 2000-2014 period at the 99% confidence level.A change point was detected in the North-Western province of Zambia February 2001. The key is to determine whether this break point is artificial (e.g., resulting from a change in the satellite observing record) or results from natural climate variability. From careful inspection of the monthly time series and other climatic indices, it appears that this change point is likely the result of a naturally occurring transition from a period of relatively high wind speed, to one with lower overall wind speed. Figure A-6 shows the MERRA monthly mean wind speeds at 50 m AGL for a site in the southwestern portion of the Eastern Province. Overall, there is a very weak positive linear trend of about 0.01 m/s/annum from 2000-2014 for this example. Similar order of magnitude trends, both positive and negative, are seen at the other 10 locations, giving some confidence of no significant systematic bias being present. 8.0 7.0 Wind speed (m/s) 6.0 5.0 4.0 3.0 2.0 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Figure A-6 Monthly mean MERRA analyzed wind speeds at 50 m AGL for the southern-most of the 11 points in the Zambia Eastern Province Another common approach is to compare the long-term temporal consistency with other available global reanalyses. In the present case, MERRA has been directly compared to the ERA-Interim reanalysis, which serves as a somewhat independent check of consistency. ERA-Interim data is obtained for grid cells closest the 11 locations for the same 2000-2014 period. Visual comparisons of normalized monthly mean wind speeds and 12-month rolling averages reveal good general agreement between MERRA and ERA-Interim at all sites, as illustrated for one example site Figure A-7. The differences between monthly values for the two datasets (MERRA minus ERA-Interim) are also evaluated, and reveal that the long-term trends in both datasets are highly consistent. This culminates in higher overall confidence for the temporal consistency of MERRA, and its overall representation of the monthly, seasonal, annual, and interannual variability of the winds within Zambia over the 2000-2014 period. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 26 1.6 Normalized wind speed (m/s) 1.4 1.2 1.0 0.8 0.6 0.4 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 MERRA MERRA 12 month rolling average ERA-Interim ERA-Interim 12 month rolling average Figure A-7 Monthly mean wind speed anomalies as represented by MERRA and ERA-Interim for 2000-2014 DNV GL has performed all reasonable checks to establish that the period used has no significant inconsistencies; however, it is reiterated that robust ground measurements represent the most reliable source of data to use in this verification process. As the performance of reanalysis datasets in the region become better understood, increased confidence can be placed on these results. Step 2: Perform 10 km resolution simulations for most recent multi-year period, and 2 km resolution nested simulations for the most recent 365-day period The mesoscale downscaling approach used for this project is based on a computationally efficient and statistically robust method for generating long-term dynamically downscaled datasets. A coarse resolution (10 km) mesoscale model is run for the entire period to be downscaled, while for only a fraction of that period a nested version of that same model is run at high (2 km) resolution. The period over which the coarse and high-resolution runs overlap is called the training period, while the remaining portion is termed downscaling period. For each time of the latter, the best matching coarse estimates (termed “analogs”) over the training period are found. The downscaled solution is then constructed from the set of high-resolution values that correspond to the best matching coarse analogs. This method has been rigorously validated and published in open peer-reviewed scholarly journals (Delle Monache et al. 2011, 2013). It has been demonstrated to yield downscaled time series that closely approximate the standard “brute force” approach to dynamical downscaling, in terms of bias, root-mean-squared difference, and goodness-of-fit (Rife et al. 2014). Other properties of the time series are replicated very well by the analog method, including the autocorrelation and the hour-to-hour variability A complete hourly time series with the 10 km resolution grid is performed over the entire 15 April 2004 - 15 April 2014 period is first generated. Next, a nested 2 km grid is used to provide an hourly time series over the last year within the 10 km resolution simulation (15 April 2013 to 15 April 2014). These outputs provide the basis for the high resolution downscaling performed with the analog ensemble, which is further DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 27 described in the following step, to yield a complete 10-year times series at 2 km resolution over the entire country. Step 3: Perform analog ensemble (AnEn) downscaling to generate a complete 10-year time series at 2 km resolution As described above, a published and rigorously validated method is employed to downscale the 10-km gridded analyses to the 2-km grid. The new technique, termed the analog ensemble (AnEn), was originally developed by Delle Monache et al (2011) and further advanced in collaboration with project Co-Lead Rife (Delle Monache et al. 2013). The procedure is shown Figure A- 8. In this example, we seek to downscale a 10 km WRF simulation at Sample Time 1 (filled black dot) to 2 km resolution. We first find all dates in the 2013-2014 training period where the 10 km WRF hourly simulations are a close match to the 10 km solution at Sample Time 1 (open circles), and rank them according to their similarity to Sample Time 1. Next, we obtain the 2 km WRF simulations for these ranked dates (open squares), and then a weighted mean of those 2 km simulations is used to form the analog ensemble mean prediction at Sample Time 1. This process is repeated for every hourly output time in the 2004-2014 period. For wind mapping applications, we have found that the optimal number of analogs is about 5, which are used to form a weighted average of the 2 km WRF simulations on those dates. As shown in Rife et al. (2014) this yields a result that closely matches the “brute force” mesoscale calculation in terms of R2, bias, and root-mean-squared error for hourly wind speed and wind direction. Figure A- 8 Analog ensemble procedure. See text for explanation. Adapted from Delle Monache (2012). It is important to note that the analogs are sought independently for each analysis time and grid point, using the wind speed and wind direction at the five heights required in the TOR 1, and the Obukhov length (a measure of vertical thermal stratification). Thus, when we “localize” the search procedure in both space and time, this greatly reduces the degrees of freedom and significantly increases the probability of finding good 1 The TOR (Selection # 1126863) specifies that wind speed and wind direction are to be provided at the following heights AGL: 10 m, 50 m, 80 m, 100 m, and 200 m. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 28 matches. Additionally, it is not necessary to match all aspects of the gridded reanalysis, such as upper-level winds or temperatures. For example, Delle Monache et al. (2011a, 2013) found that many good analogs are found by matching only the wind speed, wind direction, temperature and pressure at a specific location and time. Rife et al. (2014) found that very good analogs are found using wind speed and wind direction only. In the present application, including the Obukhov length in the analog matching procedure ensures that we are choosing analogs from the same thermal stability regime. Other authors have also demonstrated that analog-based approaches are highly effective if they include a localization strategy in space and time (Van den Dool 1989, 1994; Hamill and Whitaker 2006). It should also be noted that a set of sensitivity experiments revealed that the performance of AnEn is largely insensitive to the choice of year used for the truth data, regardless of whether the weather patterns during that year are impacted by large scale modes of variability such as the Monsoon, and the El Niño Southern Oscillation (Rife et al. 2014). The complete multi-year integration performed for the simulations yields a continuous hourly time series that well represents the full range of wind and thermal stratification conditions over Zambia. This is a dramatic improvement over previous wind mapping approaches used for World Bank funded projects. Previous methods used a random sampling technique, which has uncertainties as high as 45% for both wind speed and direction (Rife et al. 2013a). These uncertainties are entirely eliminated in the time-series approach used for this project, and the time series method has the added benefit of allowing the interannual variation in the wind resource to be quantified. Also, because the proposed modeling approach creates a continuous hourly time series, updates to the wind atlas only require performing simulations over the most recent period. For example, the Phase 1 calculations are performed over the 15 Apr 2004 to 15 Apr 2014 period. If updates to the atlas are desired in the summer of 2015, say 01 Jul 2015, then we only need simulate the 16 Apr 2014 to 01 Jul 2015 period to bring the atlas database fully up-to-date. All the relevant wind statistics can then be easily re-calculated for the WAsP-lib files, and the new statistics will fully incorporate the information from the most recent year of data. By contrast, the random sampling method would have to be fully re-run “from scratch”, and there is no guarantee that the most recent year’s information will be adequately represented in the new results, since for each calendar day, a year is randomly chosen, typically from among the most recent 15 years. For example, for 1 January the year 2005 might be chosen, and for 2 January the year 2009 may be chosen, and so on through 31 December (Rife et al. 2013a) to form a single “representative year”. The result of these continuous hourly 10-year simulations is used to create the various outputs for the interim numerical wind atlas. Step 4: Generalized wind climate modeling for the interim wind atlas The hourly outputs of WMS at each grid point of the 2 km mesoscale domain are used to calculate the generalized wind climate wind statistics according to the industry-standard WAsP-format “wind atlas” (“WAsP-lib”) file, consistent with the European Wind Atlas method (Troen and Petersen, 1989). This was performed using the Danish Technical University (DTU) Generalized Wind Climate software package. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 29 When creating the WAsP-lib files there is an option to account for vertical thermal stability, as it is defined by the WRF model (Hahmann et al. 2014) 2. DTU conducted a number of sensitivity and validation tests as part of the Wind Atlas for South Africa, and concluded that accounting for thermal stability in the generalization process resulted in a lower overall wind speed error relative to hub height measurements. DNV GL has performed similar tests at other locations around the world and found overall improvements where nighttime stable conditions are prevalent. Although Zambia lacks measurements to estimate the occurrence and frequency of stable conditions, its high terrain elevations (average elevation is 1000 m AMSL), and two distinct dry seasons very likely lead to strong nocturnal radiative cooling, which is conducive to forming vertical temperature inversions. Examination of the WRF-based time series of the Obukhov stability parameter indicates that there is a greater frequency of stable conditions during the dry seasons in Zambia. Thus, we have chosen to account for thermal stability to create the generalized wind climate WAsP- lib files. Step 5: Preliminary mesoscale wind speed uncertainty index map It is well known that mesoscale model simulations are subject to a number of unavoidable errors and uncertainties, including the input datasets, the lateral boundary conditions, the numerical approximations used in the dynamical core of the model, and the imperfect representation of the complex physical processes that strongly drive the winds within the boundary layer. Monte Carlo methods have been developed to mitigate these errors, and provide statistically defined estimates of the uncertainty in the mesoscale results. Within the numerical weather prediction community such techniques are generically termed “ensemble methods”. The ensemble approach used for this project has been proven extensively for over 20 years (Strensrud et al. 1999; Eckel and Mass 2005, Hacker et al. 2011; Lee et al. 2012, among others). It amounts to constructing several unique versions of the model by selecting different suites of physical process parameterizations, and each model version is then used to perform a separate parallel downscaling simulation (e.g., Stensrud et al. 1999). Each simulation starts from the same initial condition (MERRA in this case), and this procedure is commonly referred to as a “multiphysics ensemble”. This type of ensemble specifically quantifies the uncertainty related to the necessarily imperfect representation of the physical processes such as land surface interactions, clouds, rainfall, and atmospheric radiative transfer. The “ensemble spread” or standard deviation among the ensemble of simulations provides an objective estimate of the meteorological uncertainty, allowing us to place a statistically defined error bar on the resource estimate for any point within the numerical wind map/atlas. The WRF model has a large list of options for each type of physical parameterization, and thus it is straightforward to create a significant number of ensemble members by simply varying these parameterizations. Because parameterizations employ a wide variety of formulations and assumptions about their representation of the physical processes, the individual simulations each will yield an equally plausible (and unique) representation of the atmospheric state at any instant, within the likely range of meteorological and numerical model uncertainty. Hacker et al. (2011) and Lee et al. (2012) provide excellent guidance for the unique sets of parameterizations that yield the best overall results within a multi- physics ensemble for boundary layer applications, and for this project we use the approximate intersection of their configurations to select 9 total model versions (see Figure A-9). It is important to note that full range of uncertainty information due to errors in the physics is obtainable with as few as 8-10 ensemble 2 Hahmann, A. N., J. Badger, P. Volker, J. R. Nielsen, C. Lennard, J. C. Hansen, and N. G. Mortensen,2014: Validation and comparison of numerical wind atlas methods: the South African example. European Wind Energy Conference & Exhibition 2014, Barcelona, Spain. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 30 members (e.g., Du et al. 1997; Clark et al. 2009, 2011; Lee et al. 2012; Wandashin et al. 2001). Thus, for this project we chose 9 multiphysics ensemble members. WRF physics 1 WRF Ensemble physics 2 Ensemble Raw variance MERRA WRF Ensemble physics 3 WRF physics 9 Simulation 1 Simulation 2 Simulation 3 Simulation 4 Simulation 5 Preliminary wind speed uncert. index Simulation 6 Simulation 7 Simulation 8 Simulation 9 Figure A-9 The multiphysics ensemble approach Since the aim of this process is to understand the overall uncertainties in the interim wind atlas at the regional level, we perform the 9-member multiphysics ensemble simulations on the 10 km outer grid over Zambia and the surrounding region for three representative years: a year where the overall wind speeds are most similar to the long-term mean (termed “normal”); a second year where the overall wind speeds are higher than the long-term mean (termed “high”); and third year whose overall winds speeds are lower than the mean (termed “low”). The years were chosen based on the decade long 10 km WRF simulation used to create the interim wind atlas. This has the advantage of being very cost effective in terms of computational time and expense, and the results can be directly related to those on the 2 km grid, since the relationships between the 10 km and 2 km grids are well established during the AnEn downscaling stage. An annual mean hourly wind speed map is computed for each of the 9 ensemble member simulations and the standard deviation is computed among the 9 maps (termed the ensemble standard deviation). Because the geographic patterns of ensemble standard deviation are very similar for each of the three simulated years, the final mesoscale standard deviation is computed to be the average of the 3 ensemble standard deviation maps. Given the lack of ground measurements to validation the wind atlas and at absence of microscale modeling undertaken at this stage of the project, the preliminary wind speed uncertainty index is initially set to be equal to the standard deviation of the nine multi-physics mesoscale ensemble members. Areas with a high index value, and therefore high standard deviation indicate an area where there is a lack DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 31 of consensus between the nine multi-physics ensemble members, and shows the apparent increased difficulty in modeling the flows in these areas. This dataset will provide tremendous value to stakeholders and commercial developers, as it provides the first available statistically defined uncertainty index for the estimated wind resource at each region within the map, thereby helping target the best placement of meteorological masts and potential wind farm sites. Indeed, the uncertainty map is a key input to the Candidate Site Identification Report as part of the Phase 1 deliverables. It will also help to significantly advance the current state-of-the-art in wind mapping, and helping the industry to understand the true value and utility of modern ensemble prediction methods, which have become the mainstay of many other fields, including weather forecasting and climate prediction. For example, all of the world’s national weather prediction centers have well-established ensemble modeling systems that are used in daily operations to support a large variety of commercial and public sectors, including forecasting for operational wind and solar power plants. GENERATING A POWER TIME SERIES Many of the derived outputs required as part of this project require the generation of a power time series. Firstly, a “generic wind turbine” must be defined. The TOR stipulates that generic pitch controlled wind turbine with 100 m rotor, 3 MW rated power and at 100 m hub height should be used. DNV GL has taken a range of power curves with similar dimensions from different manufacturers and averaged them to produce a single “generic” power curve that fulfil these criteria. Power curves at the standard 1.225 kg/m3 air density have been used for this purpose. Through this process we have also ensured that the power curve exhibits a reasonable peak power coefficient (maximum Cp of 0.45). The derived generic wind turbine power curve is shown below in Table A- 1. Table A- 1 power curve for generic wind turbine model Wind Power speed (m/s) (kW) 2 0 3 29 4 117 5 255 6 458 7 738 8 1102 9 1571 10 2117 11 2677 12 2931 13 2990 14 - 25 3000 DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 32 The details of this generic turbine are just one of the topics DNV GL is keen to solicit feedback for, from both Client and key Stakeholders alike as part of the Phase 1 workshop. Initial review of the preliminary wind speed map indicates that for much of Zambia a larger rotor diameter turbine may be more appropriate. However, the characteristics of the turbine proposed here are considered to capture the salient features relating wind speed frequency distributions through to an energy value for initial planning purposes. The next step involves computing a time series of density for every 2 km grid cell. The approach taken here was to first compute a 12 x 24 matrix of air density at each 2 km grid cell using the single year simulation from 15 April 2013 to 15 April 2014, rather than from the full 10-year simulation. The rationale for this approach is that the interannual variability of air density is low, as air density is a measure of total columnar mass of the atmosphere at any point. Although hourly and daily average values of air density may vary from one year to the next due to changes associated with transient weather systems, the seasonal and annual averages (e.g., the climatologic values), of air density will not, as the total atmospheric mass content neither increases or decreases (mass conservation). Indeed, the 10-year simulations from the DNV GL WMS show negligible interannual variability of monthly or seasonal values of air density. Air density is computed according to the IEC 61400-12-1 standards 3. 1 1 1 = � − � − �� Where: B is the barometric pressure [Pa] T is the absolute temperature [K] φ is the relative humidity (range 0 to 1) R o is the gas constant of dry air [287.05 J kg-1 K-1] R w is the gas constant of water vapour [461.5 J kg-1 K-1] P w is the vapour pressure [Pa]. The vapour pressure is calculated according to: = 0.0000205exp(0.0631846 ∗ ) The above parameters are derived directly from the DNV GL WMS model outputs with a resolution of 2 km and a height of 100 m AGL. Once the 12 x 24 matrix of air density values is created at each mesoscale grid point, a full 10 year time series of density at a height of 100 m AGL can be generated by using the 12 x 24 density table as a look up table for each individual hourly record. Next, for each hourly record, the density value derived for the hour is used to create a power curve for the specific hour, for each 2 km grid cell, according to the recommendations of the IEC 61400-12-1 standard, and then the simulated wind speed at 100 m is used to calculate the power output from the density specific power curve. This is repeated for each hourly record in the 10-year time series across the entire country yielding a continuous 10 year time series of hourly power output in GW. REFERENCES 3 IEC 61400-12-1 “Wind turbines – Part 12-1: Power Performance Measurement of Electricity Producing Wind Turbines”, First edition, 2005-12. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 33 Clark, A. J., et al., 2011. Mon. Wea. Rev., 139, 1410–1418. Dee et al, 2011. Q. J. Roy. Meteor. Soc., 137, 553-597 Delle Monache, L., et al., 2011. Mon. Wea. Rev., 139, 3554–3570. Delle Monache, L., 2012. AWEA Wind Resource & Project Energy Assessment Seminar, Pittsburgh, Pennsylvania. Delle Monache, L., et al., 2013. Mon. Wea. Rev., 141, 3498-3516. Draxl, C., et al., G., 2012. Wind Energy.. doi: 10.1002/we.1555. Du, J., S. L. Mullen, and F. Sanders, 1997. Mon. Wea. Rev., 125, 2427–2459. Eckel, F. A., et al., 2005. Wea. Forecasting, 20, 328–350. Hacker, J. P., et al., 2011. Tellus A, 63, 625–641. Hamill, T., et al. 2006a. Bull. Amer. Meteor. Soc., 87, 33–46. Hamill, T., et al., 2006b. Mon. Wea. Rev., 134, 3209–3229. Lee, J. A., et al., 2012. Mon. Wea. Rev., 140, 2270–2286. Leung, L. R., and Y. Qian, 2009. Geophys. Res. Lett., 36, L03820, doi:10.1029/2008GL036445. Liang, Xin-Zhong, et al., 2012. Bull. Amer. Meteor. Soc., 93, 1363–1387. Mass, C., et al., 2002. Bull. Amer. Meteor. Soc., 83, 407–430. Mass, C. et al. 2008. Wea. Forecasting, 23, 438–459. Mearns, L. O., et al., 2012. Bull. Amer. Meteor. Soc., 93, 1337–1362.Mullen, S., R. Buizza, 2002. Wea. Forecasting, 17, 173–191. Parks, K., Y et al., 2011. NREL/SR-5500-52233, Golden, CO: National Renewable Energy Laboratory. Potter, A. W., et al., 2008. Wind Engineering, 32, 325–338. Reynolds, et al., 2002. J. Climate, 15, 1609–1625. Rienecker, M.M., et al., 2011. J. Climate, 24, 3624–3648. Rife, D. , et al., 2013a. J. Appl. Meteor. Clim., 42, 47-63. Rife, D., et al., 2013b, WindPower 2013 Conference, Chicago, IL. Rife, D., et al., 2014, WindPower 2014 Conference Rodell, M., et al., 2004. Bull. Amer. Meteor. Soc., 85, 381–394. Skamarock, W. C. 2004: Mon. Wea. Rev., 132, 3019–3032 Skamarock, W. C., et al., 2008. NCAR/TN-475+STR, 113 pp. [http://www.mmm.ucar.edu/wrf/users/docs/arw_v3.pdf]. Stensrud, et al., 1999. Mon. Wea. Rev., 127, 433-446. Troen, I., and E. L. Petersen, 1989. Risø National Laboratory, 656 pp. Wandishin, et al., 2001. Mon. Wea. Rev., 129, 729–747. Wang, X. L., 2008:. J. Appl. Meteor. Climatol., 47, 2423-2444. Wang, X. L., 2008b. J. Atmos. Oceanic Tech., 25, 368-384. DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 34 www.dnvgl.com APPENDIX B – PRELIMINARY AND UNVALIDATED MESOSCALE OUTPUTS DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 35 www.dnvgl.com Figure B-1 Preliminary and unvalidated mesoscale wind sped map, 10 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 36 www.dnvgl.com Figure B-2 Preliminary and unvalidated mesoscale wind sped map, 20 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 37 www.dnvgl.com Figure B-3 Preliminary and unvalidated mesoscale wind sped map, 50 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 38 www.dnvgl.com Figure B-4 Preliminary and unvalidated mesoscale wind sped map, 80 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 39 www.dnvgl.com Figure B-5 Preliminary and unvalidated mesoscale wind sped map, 100 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 40 www.dnvgl.com Figure B-6 Preliminary and unvalidated mesoscale wind sped map, 200 m AGL, as simulated by the DNV GL Wind Mapping System DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 41 www.dnvgl.com DNV GL – Document No.: 702833-USSD-R03-D, Issue: D, Status: FINAL Page 42 www.dnvgl.com ABOUT DNV GL Driven by our purpose of safeguarding life, property and the environment, DNV GL enables organizations to advance the safety and sustainability of their business. We provide classification and technical assurance along with software and independent expert advisory services to the maritime, oil and gas, and energy industries. We also provide certification services to customers across a wide range of industries. Operating in more than 100 countries, our 16,000 professionals are dedicated to helping our customers make the world safer, smarter, and greener.