WPS8188 Policy Research Working Paper 8188 Natural Disaster Damage Indices Based on Remotely Sensed Data An Application to Indonesia Emmanuel Skoufias Eric Strobl Thomas Tveit Poverty and Equity Global Practice Group September 2017 Policy Research Working Paper 8188 Abstract Combining nightlight data as a proxy for economic activity the size of the annual fiscal transfers from the central gov- with remote sensing data typically used for natural hazard ernment to the subnational governments. Ex post, or after modeling, this paper constructs novel damage indices at the the incidence of a natural disaster, damage indices are useful district level for Indonesia, for different disaster events such for quickly assessing and estimating the damages caused as floods, earthquakes, volcanic eruptions and the 2004 and are especially useful for central and local governments, Christmas Tsunami. Ex ante, prior to the incidence of a disas- emergency services, and aid workers so that they can respond ter, district-level damage indices could be used to determine efficiently and deploy resources where they are most needed. This paper is a product of the Poverty and Equity Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at eskoufias@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Natural Disaster Damage Indices Based on Remotely Sensed Data: An Application to Indonesia∗ Emmanuel Skoufias (World Bank) Eric Strobl (University of Bern) Thomas Tveit (University of Cergy-Pontoise) JEL Classification: Q54, C63, R11, R5,O18 Keywords: Remotely Sensed Data, Natural Disasters, Natural Hazard model, Damage Index, Floods, Earthquakes, Volcanic Eruptions ∗ This paper was financed in part by the Disaster Risk Finance Impact Analytics project of the Disaster Risk Financing and Insurance Program of The World Bank Group 1 Introduction Quickly assessing and estimating the damage caused after the incidence of a natural disaster is important for both central and local governments, emergency services and aid workers, so that they can respond efficiently and deploy resources where they are most needed. Recently, remote sensing technologies have been used to analyze the impact of disasters, such as hurricanes (Myint et al., 2008; Klemas, 2009), floods (Haq et al., 2012; Wu et al., 2012, 2014; Chung et al., 2015), landslides (Nichol et al., 2006), earthquakes (Fu et al., 2005; Yamazaki & Matsuoka, 2007), wildfires (Holden et al., 2005; Roy et al., 2006), volcanoes (Carn et al., 2009; Ferguson et al., 2010) and tsunamis (R¨ omer et al., 2012). These remote sensing techniques are useful for providing quick damage estimates shortly after the disasters giving emergency services a chance to respond quickly and local governments an overview of estimated costs and necessary repairs. In addition to their usefulness in the aftermath of a disaster, estimates of the potential damage as- sociated with a natural disaster are also useful for policy making prior to the realization of the natural hazard event. In many cases the incidence of a natural hazard event can turn into a natural disaster simply because of inadequate preparation ex-ante. Indonesia, for example, is highly exposed to natural disasters by being situated in one of the worlds most active disaster hot spots, where several types of disasters such as earthquakes, tsunamis, volcanic eruptions, floods, landslides, droughts and forest fires frequently occur. The average annual cost of natural disasters, over the last 10 years, is estimated at 0.3 percent of Indonesian GDP, although the economic impact of such disasters is generally much higher at local or subnational levels (The Global Facility for Disaster Reduction and Recovery, 2011). The high frequency of disasters experienced has important impacts on expenditures by local governments that could be anticipated, at least in part, through upward adjustments in the annual fiscal transfers from the central government to the subnational governments.1 Such ex-ante adjustments in the level of fiscal transfers would be more useful if they could be based on estimates of the potential damages associated with the incidence of a natural disaster as opposed to estimates of the intensity of the potential natural hazard that might occur. However, although in recent years there has been much progress towards the modeling of the main natural hazards, there continues to be a scarcity of estimates of the damages associated with the incidence of these disasters. The value of damage caused by a natural disaster is typ- ically a complicated function of the size of population living in that area, the level and type of economic activity carried out, the value of the physical infrastructure in place, and the resilience of infrastructure and people’s livelihoods to the natural hazards. This paper fills some of the gaps in the literature by using different remote sensing sources and data on the physical characteristics of the events to construct four damage indices for natural disasters in Indonesia. The indices cover floods, earthquakes, volcanic eruptions and a tsunami, and are all weighted by local economic activity in an area, and then aggregated up to a district level.2 All data used in the construction of the indices are free and publicly available, making the methods used a potentially very useful alternative for both central and local governments to quickly get a rough estimate of the damages caused by a disaster (either ex-ante or ex-post).3 Importantly, all of the indices constructed take into account local exposure. Given limited access to highly disaggregated local economic activity data, nightlight intensity derived from satellite imagery has proved to be a good proxy; see, for instance, Henderson et al. (2012), Hodler & Raschky (2014) and Michalopoulos & Papaioannou (2014). By utilizing the grid cells of approximately 1 square kilometer we can break down areas in cities and districts into where they are busiest, and thus take into account not only the local physical characteristics of a natural disaster but also the local economic activity exposed to it. The paper is structured as follows. Section 2 of the paper discusses in more detail the incidence and types of natural disasters. Section 3 discusses the nightlights data. Sections 4-7 discuss in detail the construction of the four damage indices, while section 8 concludes. 1 For example, Indonesia experienced 4,000 disasters between 2001 and 2007 alone, including floods (37%), droughts (24%), landslides (11%) and windstorms (9%) (The Global Facility for Disaster Reduction and Recovery, 2011). 2 A tropical cyclone index was also constructed, but no hurricanes had strong enough winds to cause any damage on land. 3 In a separate paper, Skoufias et al. (2017), we correlate the damage indices of these disasters at the district level with the ex-post allocation of district expenditures in different sectors and by economic classification. 2 2 Natural Disasters in Indonesia Natural disasters are prevalent events across most parts of Indonesia. According to the Indonesian Na- tional Disaster Management Authority (BNPB) there were more than 19,000 natural disasters in the period 2001 - 2015 (National Disaster Management Agency, BNPB, 2016), making Indonesia a useful country for any natural disaster analysis. The most frequent disasters are floods and landslides (52 per- cent), strong winds (21 percent) and fires (15 percent), while the most damaging ones are earthquakes, tsunamis and volcanic eruptions, which all cause major damage to buildings and infrastructure in ad- dition to the human casualties. The deadliest year according to the BNPB data was 2004, where there were more than 167,000 deaths due to natural disasters and 166,671 of them stemming from the tsunami in December 2004. 2.1 Floods The tropical climate of Indonesia often leads to annual floods. The BNPB data registered more than 10,000 incidents of floods or landslides leading to more than 3,500 fatalities from 2001 through 2015. During the period from 1985 to 2016, The Darthmouth Flood Observatory (DFO) registered 3,808 floods of magnitude 4 or more and 1,175 floods of magnitude 6 and up.4 Of these floods, there were 126 large scale floods with a centroid within Indonesia in the period from 2001 to 2016 as can be seen in Figure 1. Of the 34 provinces, 27 experienced having a centroid of a large scale flood event during these years.5 Figure 1: DFO Large Scale Floods in Indonesia 2001-2016 Source: G.R.Brakenridge (2016) 4 Magnitude is defined as: M = log(D ∗ S ∗ AA), where D is the duration of the flood; S is the severity on a scale consisting of 1 (large event), 1.5 (very large event) and 2 (extreme event); and AA is the size of the affected area. Flood events registered by DFO have mainly been derived from news and governmental sources. 5 The provinces where no large scale centroid was present were Bangka Belitung, Riau Islands, Kalimantan Barat, Yogyakarta, Sulawesi Barat, Kalimantan Utara and Maluku. Note that some of these, like Kalimantan Utara, Kalimantan Barat, Sulawesi Barat and Yogakarta, did most likely experience large scale flood during these years, but that the centroid was in another province. The remaining three provinces consist mainly of smaller islands, so the flooded area will most likely not constitute a large scale flood event. 3 2.2 Earthquakes Due to Indonesia’s location inside the Pacific Ring of Fire, one of the most seismically active areas in the world, it is often struck by earthquakes. BNPB counted almost 400 earthquakes from 2001 to 2015, with the largest number of casualties coming from the tsunami created by a 9.0 earthquake located off the coast of Aceh, otherwise known as the earthquake that caused the 2004 tsunami. Apart from that, there were more than 8,000 registered fatalities due to earthquakes over the same period. Overall, this makes earthquakes the deadliest of the natural disasters that strike Indonesia. Figure 2 shows how common earthquakes are in Indonesia by displaying contour maps6 of all earth- quakes of magnitude 5.0 and above that struck Indonesia from 2004 through 2014. In total, the United States Geological Survey (USGS) registered 261 earthquakes.7 Figure 2: Earthquakes in Indonesia 2004-2014 Source: USGS 2.3 Volcanic activity Indonesia has the highest number of active volcanoes in the world, numbering almost 150. Of these, many have had eruptions in both more historical times and after the year 2000. The most famous eruption is probably the explosion of Krakatau in August 1883, when two-thirds of the Krakatau Island erupted and disappeared, killing more than 35,000 people and causing a global mini ice age and weather disruptions for years. BNPB have registered 92 eruptions over our 15-year time period and more than 60 major volcanoes that have had eruptions since 1900. The most recent one is the 2010 Mount Merapi eruption 6 These maps are also known as ShakeMaps, which are produced by USGS. 7 There are 1,002 earthquakes registered by USGS that were of magnitude 5.0 or more that had a point with a PGA of at least 0.05 within Indonesia. Many of these points create little to no damage. The 261 earthquakes mentioned above are quakes that are mostly contained within Indonesia. 4 that killed 324 people and dislocated more than 320,000. In addition to the BNPB data, during the years 2004 through 2015 the Darwin Volcanic Ash Advisory Centre (DVAAC) had 587 days where they issued a red warning, implying an ongoing or imminent volcanic eruption. The most active volcanoes - measured by number of days with red warnings - are shown in Table 1, with the top 5 volcanoes constituting almost 75 percent of the red warnings. These are also volcanoes that have been in the media, with Merapi already mentioned and Sinabung, which had several eruptions in 2010, 2013 and 2014. These two volcanoes are located close to densely populated areas, with Sinabung located in North Sumatra and Merapi in central Java. Table 1: Most Active Volcanoes 2004-2015 Volcano Number of Days with Red Warning Sinabung 224 Merapi 92 Manam 74 Egon 36 Soputan 31 2.4 2004 Christmas Tsunami The Christmas Tsunami in 2004 is the worst singular natural disaster during the modeling period, and one of the worst natural disasters in world history. As seen in the photo in Figure 3 the destruction was absolute in parts of Indonesia. The total death toll across Indonesia and 13 other countries was more than 230,000 people and there were many more missing. In addition, the World Bank (2005) estimated a total economic impact of 4.5 billion US dollars. The official BNPB data for Indonesia estimates 166,671 deaths due to the tsunami. The cause of the tsunami was an earthquake of magnitude 9.0 150 miles south-south east of Banda Aceh on the morning of 26 December. This quake created a tsunami with waves more than 20 meters high at the highest. Due to the fault line of the earthquake being in a north-south direction, the greatest strength of the tsunami was in an east-west direction (Athukorala & Resosudarmo, 2005). This led to the largest damages being in the northern part of Sumatra, in the province of Aceh, where entire villages were wiped out as seen in the photo of Banda Aceh (Figure 3). Figure 3: Destruction in Banda Aceh Source: The Atlantic (2014) 5 3 Nightlight Data Natural disasters are inherent local phenomena in that they either affect only parts of areas and/or affect parts within areas differently. It is thus important to take the local population/asset exposure into account when constructing more aggregate proxies. Arguably one would like to have measures of exposure as spatially disaggregated as possible. For a country like Indonesia, data are usually sparse and at a very aggregated spatial level. An alternative approach is thus to use nightlights as a proxy for local economic activity. As a matter of fact, nightlights have found widespread use where no other measures are available; see, for instance, Henderson et al. (2012), Hodler & Raschky (2014) and Michalopoulos & Papaioannou (2014). In Hender- son et al. (2012), Indonesia is used as an example of using nightlights to capture an economic downturn following the Asian financial crisis in the late 1990s. Their results show that swings in GDP change can generally be captured. Nevertheless one has to account for factors such as cultural differences in light usage, latitude and gas flares. In our case this is unlikely to affect our results since we use nightlights to capture exposure within a country rather than across countries. The nightlight imagery we employ is provided by the Defense Meteorological Satellite Program (DMSP) satellites. In terms of coverage each DMSP satellite has a 101 minute near-polar orbit at an altitude of about 800km above the surface of the earth, providing global coverage twice per day, at the same local time each day, with a spatial resolution of about 1km near the equator. The resulting images provide the percentage of nightlight occurrences for each pixel per year normalized across satel- lites to a scale ranging from 0 (no light) to 63 (maximum light). Yearly values were then constructed as simple averages across daily values of grids, and are available from 1992.8 We use the stable, cloud-free series; see Elvidge et al. (1997). The data revealed 414,644 cells which had a nightlight value greater than 0 in them at least once during the period 2001-2013. Figure 4 - containing all cells with nightlights in 2012 - shows that the large cities and densely populated areas on Java, Sunda Islands, coastal Kalimantan and Sumatra are fully covered in lights. Inner parts of Kalimantan and most parts of New Guinea are more sparsely lit. Figure 4: Cells with Registered Nightlights in 2012 8 For the years where satellites were replaced, DMSP provides an average from both the new and old satellite. In this paper we use the imagery from the most recent satellite but as part of our sensitivity analysis we also re-estimated our results using an average of the two satellites and the older satellite only. The results of these latter two options were almost quantitatively and qualitatively identical. 6 4 Flood Damage Index The modeling of floods can be done by remote sensing (Brakenridge & Anderson, 2006; Wu et al., 2012; Haq et al., 2012) or through a combination of weather data and GIS systems as for example in Knebl et al. (2005); Asante et al. (2007); Dessu et al. (2016). We utilize the latter, as remote sensing is useful for assessing whether an area is flooded or not, but it is weaker on modeling the intensity of the flood. Moreover, cloud cover generally limits the accurate detection of floods from remote sensing sources. To model floods we have decided to use the Geospatial Stream Flow Model (GeoSFM) which is a software that is “designed to use remotely sensed meteorological data in data sparse parts of the world”(Artan et al., 2008). GeoSFM was developed by USGS and USAID and is a hydrological model- ing tool used to model stream flows across large areas, in particular areas where highly localized data are lacking. It has been used in regions such as the Great Horn of Africa (Asante et al., 2007; Mati et al., 2008; Dessu et al., 2016) and Nepal (Shrestha et al., 2011), with Dessu et al. (2016) finding that the model captures 76% of the monthly average variability, making it useful for flood simulation. The inputs needed to model stream flow for basins are soil- and terrain-based - such as digital el- evations models (DEM) and land cover and soil data - and weather-based, such as precipitation and potential evapotranspiration (PET) data. The HYDRO1K data set from USGS, which is a DEM made for hydrological modeling based on the USGS’ 30 arc-second DEM of the world, is used as elevation input. The land cover data are the Global Land Cover Characterization (GLCC) data set also from USGS, while the soil data are from the FAO Digital Soil Map of the World. The daily precipitation data are from the 3-hourly data set from the Tropical Rainfall Measurement Mission Project (TRMM) and the PET data are 6-hourly data from the Global Data Assimilation Sys- tem (GDAS), both data sets are aggregated up to daily data. The PET data are available from February 2001 and onwards, while the precipitation data are available for the period 1998-2014. Given that we only have nightlight data through 2013, we will focus on floods for the period 2001-2014. GeoSFM uses the inputs to construct basins based on the terrain and then uses a linear soil moisture accounting routine to model surface runoff and soil moisture based on precipitation and PET. It is worth noting that although a more complex and better non-linear routine is also supported, it does not work well for our more generalized macro-modeling with fairly low resolution data. Finally, GeoSFM models the stream flow for each basin for each day of our time period. Note that GeoSFM does not model coastal floods, nor does it model flash floods in areas where there are no rivers or streams of a certain length. Figure 5 shows that there are parts of Indonesia and even one province - Riau Islands - which have no basins. Another weakness is that it does not take into consideration the specific terrain within each basin. Floods are generally very localized events and the low resolution of our data makes it impossible to model the intensity of the stream flow within a basin and also causes some river outlets to be slightly inland instead of running all the way to the ocean. 4.1 Creation of Index and Results The first part of constructing the index involves defining when a flood event is happening. In Wu et al. (2012) they propose four runoff based methods to define a flood threshold, and in addition Wu et al. (2014) propose a slightly modified flood threshold definition with a point being flooded when: R > P95 + σ and Q > 10m3 /s (1) where R is the routed runoff in millimeters, P95 is the 95th percentile value and σ is the standard deviation of the routed runoff. Q is the discharge in cubic meters. We found that with the GeoSFM modeled data, runoff was not a good proxy for flooding, due to it only capturing a limited number of floods. Discharge, Q, was a better proxy, leading to a new - but very similar - equation: Q > P95 + σ and Q > 10m3 /s (2) 7 Figure 5: Basins by Province By manually checking against the DFO floods, we find that our data do hit several of the large scale events in Figure 1. 4.1.1 Damage Index Due to floods being very localized, the modeling of damage is difficult, and no standard exists in the literature. Penning-Rowsell et al. (2005) base destruction on value of housing stock and the Standards of Protection and then uses an estimate of number of properties affected by different return period floods. Scawthorn et al. (2006) use a combination of building stock and velocity of the stream flow, whereas Kreibich et al. (2009) look at different parameters such as velocity, depth, energy head, stream flow and intensity. They find velocity to be a poor parameter for assessing damage, while water depth and energy head show the best results. Stream flow and intensity are also weak as parameters. Finally, Merz et al. (2010) assess different damage influencing parameters and point to the fact that most “damage influencing factors are neglected in damage modeling, since they are very heterogeneous in space and time, difficult to predict, and there is limited information on their (quantitative) effects”. Overall, there is limited support in the literature for a strong correlation between these parameters and damages on anything but a very localized scale. As for assessing the damage itself, Merz et al. (2010) discuss damage functions and the two main approaches, which involve one empirical approach where damage data are collected after the flood and one synthetic approach where they construct potential what if-scenarios. Once again the assessments rest on very localized data, which we do not have for Indonesia. Overall, it means that we cannot expect anything more than rough estimates. A common denominator for the papers mentioned above is that there is some measurement of intensity. Given that stream flow is an intensity proxy, we have used that to construct a simple measurement for intensity. The equation is: 0 : F lood = 0 Ib,t = ¯b Qb,t −Q (3) σb : F lood = 1 8 where Ib,t is the intensity of the flood in basin b at date t, Qb,t is the stream flow in the same basin at the same time and Q ¯ b and σb are mean and standard deviation of stream flow in b. The intensity is set to zero if the flood threshold - 95th percentile plus 1 standard deviation above the average - has not been exceeded. By normalizing, we obtain a measure that is comparable across all regions and that is independent of the absolute river flows. The assumption is that people living close to rivers will be prepared for variations in water levels, and that people living close to rivers with highly variable stream flows are more prepared for these events than people living close to more stable rivers. To aggregate the flood impact each basin is weighted based on the nightlights in it. The weights per basin, Wb,t−1 , used are: I Lb,t−1 i Li,t−1 Wb,t−1 = = J (4) Lp,t−1 j Lj,t−1 where Lb,t−1 is the sum of lights in basin b one year, t − 1, before the flood year and Lp,t−1 is the same at a province level. Finally, the weights from (4) are multiplied with the intensity from (3) to get the overall flood impact, FI b,t in that basin on the province: FI b,t = Wb,t−1 ∗ Ib,t (5) One thing to note here is that for basins that span several provinces or districts, we have assumed the same intensity, but the weight is based on nightlights within each individual province. 4.1.2 Results The stream flow was simulated for 5,082 consecutive days, from 1 February 2001 to 31 December 2014.9 Table 2 shows that the top 10 basins with most flood days had close to 200 days of flooding over the 14-year period. As expected, these basins do overlap with some of the busiest flood areas according to the DFO, as shown in Figure 6. The lower part of Table 2 reveals that the driest basin had a mere 12 days of flooding. All 33 provinces with a basin had days that went above our flood threshold set in Equation (2). All months in our model have flood events, but there are big differences. The range goes from 527 events every March and down to 154 events every August, with the traditional rainy season (November- March) producing the highest number of flood events, whereas the dry season months (June-October) are the driest. Aggregating the numbers for the rainy season, there are 2,215 events every year across the basins, while there are only 995 events every year during the dry season. Total number of basins that are partly or fully inside Indonesia is 495, and these basins have a total of 55,605 flood events or slightly more than 112 per basin. In other words, the average basin has been flooded for a total of 8 days a year over the 14-year period in question. This is not entirely unexpected given the climate in Indonesia and the way our threshold is made. Also, if we compare with the DFO data where they have 3,808 floods of magnitude 4 or higher through their period from 1985-2016, which converts to almost 123 fairly large scale flood events per year, our model provides a reasonable proxy for events. Even though the results seem logical on a per basin basis, the time steps in the model are 1 day at a time, which is too slow for the unfolding of a flood event, implying that downstream basins that would normally fill up very quickly will now only be filled up the day after, and then the next basin will be filled two days later and so forth. This means that the amount of days with floods are inflated. We believe that this does not affect our results much, though, as the number of events per province will not affect the end results, since we weigh by affected nightlight and not by number of days of floods. Despite the numerous floods in Indonesia, they generally do not affect a large percentage of the pop- ulation, as per Table 3. The mean of nightlights when excluding areas with 0 nightlight is 3.39 percent. 9 For Bali we did it for 5,080 days due to problems with 30 and 31 December 2014. 9 Figure 6: Top 10 Most Flooded Basins and DFO Floods Table 2: Basins with Most and Least Flood Events Basin Number Affected Provinces Number of flood events 2 Bangka-Belitung 192 705 Sumatera Selatan 190 133 Aceh 189 632 Jambi, Sumatera Barat 189 282 Sumatera Utara 187 872 Jawa Barat 186 868 Jawa Barat, Banten 183 916 Jawa Tengah, Jawa Timur 183 558 Sulawesi Barat, Sulawesi Selatan 180 709 Papua 177 256 Kalimantan Barat, Kalimantan Tengah 12 444 Riau 17 197 Sulawesi Tengah 21 316 Kalimantan Barat 24 314 Kalimantan Barat 27 If we assume that 3.39 percent of the approximately 250 million people of Indonesia are affected, the floods would impact 8.5 million people. 10 Table 3: Descriptives of Weights and Intensity (excluding zero damage observations) Statistic N Mean St. Dev. Min Max Weights 45,005 0.034 0.060 0.00003 0.557 Intensity 45,005 4.516 2.608 0.989 50.944 Damage Index 45,005 0.155 0.351 0.0001 12.780 4.1.3 Comparison of Model versus DFO Floods The DFO flood database is mostly based on news sources, providing an overview of the big floods in Indonesia. To check the database against the GeoSFM model, the focus will be on the largest events of magnitude 6 and above. Given how the DFO data do not give any intensity estimates and focus primar- ily on displacement numbers and area, while our model is driven by intensity the comparison will only focus on whether GeoSFM results do overlap in time and/or province with the centroid of the DFO floods. Table 4 shows the DFO data on the left side, first column being the start month of the flood, followed by duration, magnitude, the province where the centroid of the flood is, dead and displaced. The right side shows the modeling results where the focus is on duration. The first column under model results shows the number of days for the centroid province, then overall number of days with floods anywhere in Indonesia during the period, then looking at the same island - using that as a proxy for neighboring regions - where one examines total days the island provinces were flooded during the flood and finally how many of the days of the flood duration that a province on the same island was flooded. Generally, the model performs well, in particular on Sumatra and Kalimantan (Borneo), with the example where the 2008 flood was captured for all 25 days in the centroid province. Overall, it shows at least one flooded basin on Kalimantan and Sumatra for 85% of the days the major floods happened. The results are somewhat worse on Java, where only 37% of the days have a flood. A primary reason for this might be that Java is very narrow and hence the streams are short and might not be cap- tured in our model. Java also has larger percentage of land not covered by a basin, ref Figure 5, also due to its narrowness which makes the low resolution landcover data underestimate the size of the basins. 4.1.4 Aggregated numbers Finally, to aggregate up to a district or province level, we have used a simple method for the total damage experienced per year: T B P Dj,T = FI j,b,t (6) t b where j is the province or district, T is the year, sum of t is all the days for year T , sum of b are all the basins in the province or district and FI b,t is the flood impact from Equation 5. Normally one might use an average flood impact across the year, but by doing this, we capture repeated flood events and areas that experience generally high flooding. Using the above method, Table 5 provides the aggregated data for all provinces across all years. The impact is fairly even for the most impacted ones, with the impact numbers for the top 10 ranging from 44 to 55. Furthermore, Sumatera Selatan, Lampung and Yogyakarta make up 8 of the top 10 impacted provinces. The overall picture fits with the DFO floods in Figure 1, with the populous provinces in Java, Sumatra and Sulawesi being impacted, whereas the smaller island provinces and parts of Kalimantan are not affected much. For some of the island provinces the numbers are probably underestimated, no basins will have been constructed and modeled there due to the many small islands. Finally, Table 6 shows the most impacted districts over the years 2001 through 2014. The impact is much larger than for the provinces as one would expect due to the more localized data and impact. The districts are also more geographically spread out than the provinces. 11 Table 4: DFO Floods Compared with GeoSFM Results DFO Data Model Results Flooded days Total Flood Days Total Flood Days Days Flooded on Island Date Duration Magnitude DFO Centroid Province Dead Displaced in Province Sum All Provinces Sum on Island During Period Jan 2002 17 6.1 Jawa Timur 147 380,000 5 17 8 7 12 Dec 2003 45 6.9 Jambi 148 350,000 18 45 180 41 Jan 2005 31 6.4 Sumatera Selatan 9 0 16 30 105 30 Jan 2006 20 6.2 Jawa Barat 19 10,000 6 20 13 9 May 2007 25 6.0 Kalimantan Tengah 0 3,000 23 25 79 25 Mar 2008 25 6.3 Riau 0 60,000 25 25 149 25 Apr 2010 17 6.2 Kalimantan Tengah 0 0 14 17 38 17 Feb 2012 8 6.2 Sumatera Selatan 0 1,200 6 8 27 8 Jan 2014 31 6.2 Jawa Barat 23 20,000 9 31 31 13 Table 5: Aggregated Flood Intensity Data by Province Province 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Aceh 12.674 7.400 14.712 11.213 18.127 15.643 23.155 20.799 17.918 20.076 17.665 10.116 14.864 10.571 Bali 3.614 11.780 13.083 11.676 10.359 10.783 8.080 5.786 9.042 7.851 7.882 7.609 8.580 Bangka-Belitung 1.380 1.756 1.194 1.525 1.973 0.803 0.794 0.306 0.456 0.745 1.165 0.606 0.623 0.691 Banten 7.118 7.632 13.076 14.360 11.899 9.668 27.980 19.316 14.921 21.456 12.783 15.288 7.630 4.890 Bengkulu 6.394 6.756 25.205 13.606 14.273 10.266 11.683 16.207 13.145 17.983 17.417 6.351 21.045 7.279 Gorontalo 54.770 16.760 37.260 25.281 35.026 37.234 27.620 21.607 31.719 5.607 10.702 3.680 20.092 15.921 Irian Jaya Barat 1.368 0.647 1.242 0.565 1.167 0.921 0.985 0.508 1.107 0.356 0.343 1.837 1.710 1.307 Jakarta Raya 15.099 7.370 12.681 3.931 18.294 9.950 22.154 40.751 23.394 22.670 24.347 25.294 21.573 9.362 Jambi 44.338 11.141 10.589 11.677 19.131 13.241 16.989 34.894 31.627 32.430 15.023 21.463 32.401 30.362 Jawa Barat 16.616 17.697 29.439 18.629 30.536 23.842 36.484 28.958 27.030 35.907 25.600 27.714 29.462 11.156 Jawa Tengah 29.271 41.395 42.206 35.024 26.358 24.512 28.825 27.113 31.001 22.795 20.560 20.117 25.344 14.672 Jawa Timur 13.567 17.287 25.731 21.652 22.093 19.205 28.864 19.975 24.997 28.016 12.384 16.611 28.322 11.372 Kalimantan Barat 3.833 6.570 7.145 6.320 13.792 5.801 8.921 9.740 8.574 6.321 4.663 6.273 13.122 17.092 Kalimantan Selatan 26.550 10.917 19.599 21.410 15.397 12.302 21.088 29.202 21.391 16.995 10.598 29.563 32.648 21.549 Kalimantan Tengah 11.208 16.090 38.512 37.649 27.149 20.731 32.001 29.097 26.581 8.915 17.965 18.792 31.707 23.723 13 Kalimantan Timur 7.603 3.046 11.437 7.066 9.124 9.208 15.114 14.168 24.098 15.236 15.408 11.198 16.020 10.496 Kalimantan Utara 1.098 3.949 8.894 1.245 5.471 6.455 12.065 24.721 16.620 6.085 4.615 5.597 7.374 10.416 Lampung 25.228 20.499 30.012 32.566 31.452 21.910 23.642 37.172 39.414 49.285 25.333 22.493 43.979 47.995 Maluku 0.000 0.000 0.216 0.000 0.000 0.014 0.000 0.000 0.000 0.000 0.358 0.000 0.278 0.705 Maluku Utara 0.904 0.959 0.241 0.510 1.006 2.406 1.574 1.873 0.353 1.230 1.749 2.629 0.913 0.796 Nusa Tenggara Barat 0.014 0.083 0.035 0.021 0.080 0.034 0.059 0.093 0.036 0.038 0.064 0.068 0.048 0.009 Nusa Tenggara Timur 6.168 9.770 15.109 15.209 15.937 7.311 10.590 10.489 6.032 9.166 9.506 8.517 12.688 5.676 Papua 17.567 5.837 18.616 15.338 18.461 11.763 30.377 31.351 24.332 31.378 25.779 24.328 35.261 28.788 Riau 7.063 20.727 17.103 11.778 20.699 17.433 35.765 26.883 23.667 29.925 25.391 15.855 20.114 41.465 Sulawesi Barat 6.880 9.094 2.650 3.918 4.512 4.888 5.244 14.403 4.637 5.044 5.727 3.625 3.596 14.663 Sulawesi Selatan 14.729 9.733 14.866 12.397 8.112 15.667 11.431 14.619 12.996 9.001 10.157 7.805 11.202 12.707 Sulawesi Tengah 2.798 2.603 10.433 24.693 10.258 14.202 2.255 12.771 8.280 0.427 9.320 4.415 7.999 5.688 Sulawesi Tenggara 2.861 2.092 6.197 2.396 13.335 5.188 14.381 11.368 3.927 4.990 4.428 2.474 4.780 1.015 Sulawesi Utara 4.253 2.709 8.590 6.575 5.250 6.360 6.551 7.062 3.813 8.950 4.976 5.710 4.719 6.018 Sumatera Barat 8.944 16.785 26.946 21.787 28.464 18.737 35.011 24.624 24.595 18.317 9.770 10.559 17.472 18.169 Sumatera Selatan 27.938 17.191 47.852 33.865 34.345 23.193 39.282 55.345 34.210 34.032 24.159 23.462 51.858 27.107 Sumatera Utara 13.788 11.627 17.135 11.155 19.981 14.664 15.477 19.144 16.001 18.513 15.272 15.037 21.185 20.517 Yogyakarta 37.598 27.613 42.810 24.707 25.250 37.895 20.736 49.592 44.909 29.360 35.296 41.184 41.429 30.332 Table 6: 10 Most Impacted Districts District Province Year Flood Impact Seruyan Kalimantan Tengah 2010 175.080 Aceh Tengah Aceh 2010 165.381 Bener Meriah Aceh 2010 139.742 Pasaman Sumatera Barat 2010 125.854 Sarolangun Jambi 2010 118.244 Lubuk Linggau Sumatera Selatan 2003 114.790 Keerom Papua 2009 114.289 Tana Toraja Sulawesi Selatan 2013 106.886 Klaten Jawa Tengah 2002 106.488 Sukoharjo Jawa Tengah 2002 106.488 14 5 Earthquake Damage Index The measurement of earthquake detection and intensity has improved with remote sensing techniques. There are different methods to assess intensity and damage, ranging from satellite images (Dell’Acqua & Gamba, 2012; Tralli et al., 2005; Gillespie et al., 2007) to contour maps generated by seismological ground stations (De Groeve et al., 2008; GeoHazards International and United Nations Centre for Re- gional Development, 2001; Federal Emergency Management Agency, 2006). This paper uses the latter method, by utilizing ShakeMaps from USGS, which are automatically generated maps providing several key parameters following an earthquake, such as peak ground accel- eration (PGA), peak ground velocity (PGV) and modified Mercalli intensity (MMI). More specifically, the ShakeMaps use data from seismic stations that is interpolated using an algorithm which is similar to kriging. To model the intensity in a given coordinate, the model also takes into account ground con- ditions and the depth of earthquake. Wald et al. (2005) point to the magnitude and epicenter location - which are parameters common for the entire earthquake - that have historically been used to determine how severe earthquakes were, but that the damage pattern is not just dependent on those two parame- ters, but also on other, more localized parameters that the ShakeMaps use to generate intensity measures. This is exemplified by several earthquakes such as magnitude 6.7 and 6.9 earthquakes in California in 1994 and 1989, respectively, where some areas further away from the epicenters got more damaged than closer areas. The reason why the more localized ShakeMaps with their ground shaking parameters are a better gauge than magnitude and epicenter distance is explained on page 13 of Wald et al. (2005) which states that: “..., although an earthquake has one magnitude and one epicenter, it produces a range of ground shaking levels at sites throughout the region depending on distance from the earthquake, the rock and soil conditions at sites, and variations in the propagation of seismic waves from the earthquake due to complexities in the structure of the Earth’s crust.” The ShakeMaps are interpolated grids with point coordinates spaced approximately 1.5 kilometers apart (0.0167 degrees). Figure 2 shows contoured maps of these points. The PGA is a measure of the maximum horizontal ground acceleration as a percentage of gravity, PGV is the maximum horizontal ground speed in centimeters per second and MMI is the perceived intensity of the earthquake, a subjective measure. Figure 7 - which is originally found in Wald et al. (1999) - explains the relationship between the different parameters and the potential damage from dif- ferent values. The assumption is that damage starts at an MMI level of V and a PGA of 3.9 percent of g . These levels are found for California in Wald et al. (1999), but the relationship has been found for other areas in the US in Atkinson & Kaka (2006) and Atkinson & Kaka (2007) and for places such as Costa Rica (Linkimer, 2007) and Japan, Southern Europe and Western US (Murphy & O’Brien, 1977). It should be noted that the numerical relationship differs from region to region. There are no known papers estimating these values specifically for Indonesia. Figure 7: ShakeMap Instrumental Intensity Scale Legend Source: Wald et al. (1999) The different measures are largely interchangeable, and in GeoHazards International and United Nations Centre for Regional Development (2001) report, they use PGA to measure damage, pointing to the fact that PGA, unlike MMI is an objective measure, implying that MMI is not easy to obtain reliably across the globe. Also, for large scale modeling, where it is unfeasible for one to model local conditions precisely, PGA serves as a good proxy for intensity of earthquakes. 15 5.1 Creation of Damage Index and Results 5.1.1 Damage Index To construct the damage index, two types of data will be used; the intensity data - expressed as PGA - and building inventory data, to assess what damage one could expect for different intensities. To take into account the building types in Indonesia, we use information from the USGS building inventory for earthquake assessment, which provides estimates of the proportions of building types ob- served by country; see Jaiswal & Wald (2008). The data provide the share of 99 different building types within a country separately for urban and rural areas. For Indonesia the building type information was compiled from a World Housing Encyclopedia survey, while the split between urban and rural is from the urban extent map of Center for International Earth Science Information Network - CIESIN - Columbia University et al. (2011). Without any other information available, we use this as an indication of the distribution of building types in Indonesia, but, necessarily, assume that the distribution is homogenous within urban and rural areas. Fragility curves by building type are derived from the curves constructed by Global Earthquake Safety Initiative project; see GeoHazards International and United Nations Centre for Regional Devel- opment (2001). More specifically, buildings are first divided into 9 different types.10 Each building type itself is then rated according to the quality of the design, the quality of construction, and the quality of materials. Total quality is measured on a scale of zero to seven, depending on the total accumu- lated points from all three categories. According to the type of building and the total points acquired through the quality classification, each building is then assigned one of nine vulnerability curves which provides estimates of the percentage of building damage for a set of 28 peak ground acceleration intervals. In order to use these vulnerability curves for Indonesia we first allocated each of the 99 building types given in the USGS building inventory to one of the 9 more aggregate categories of the GESI building classification. However, to assign a building type its quality-specific vulnerability curve we would fur- ther need to determine its quality in terms of design, construction, and materials, an aspect for which we unfortunately have no further information. We instead assume that building quality is homogenous across building type in Indonesia and experiment with seven different sets of vulnerability curves, each set under a different quality ratings scenarios (ranging from 0 to 7). Figure 8 depicts the building share weighted vulnerability curves of Indonesia for urban and rural areas. Figure 8: Vulnerability Curves Rural Urban 10 Wood, steel, reinforced concrete, reinforced concrete or steel with unreinforced masonry infill walls, reinforced masonry, unreinforced masonry, adobe and adobe brick, stone rubble, and lightweight shack or lightweight traditional. 16 To model estimated damage due to a particular earthquake event the data from the ShakeMaps and GESI are used. Then, one identifies the value of peak ground acceleration that each nightlight cell in Indonesia experiences by matching each earthquake point with its nearest nightlight cell. If the cell is further away than 1.5 kilometers or if it experiences shaking (PGA) of less than 0.05 the value is set to 0. In order to derive at a region j specific earthquake damage index, ED , the following equation is applied: I K pgak,q ED q,j,t = wi,t−1 DR i,j,k,t q = 0, . . . , 7 (7) i=1 k=1 where DR is the damage ratio according to the peak ground acceleration, pga , and the urban-rural qualification of cell i, defined for a set of 8 different building quality q categories. The weight wi is the same as before; the sum of the nightlights of the affected cells over the sum of the total provincial nightlights. 5.1.2 Results With the above method, we find that 27 of the 34 provinces were damaged by earthquakes at some point in time.11 Table 7 shows that the big islands Java and Sumatra have the most affected nightlight cells, given how densely populated they are and how much seismic activity is experienced there this is expected. Table 7: Times a Lit Nightlight Cell is Damaged by Earthquake by Province Province Times Nightlight Cell Damaged Percentage of Total Aceh 1,170 22.28 Sumatera Utara 722 13.75 Sumatera Barat 511 9.73 Sulawesi Tengah 354 6.74 Jawa Barat 353 6.72 Sumatera Selatan 283 5.39 Jawa Tengah 259 4.93 Jawa Timur 242 4.61 Bengkulu 165 3.14 Finally, there were 5,251 cases where the instance hit a nightlight cell that was lit. Table 8 shows that the individual nightlight cell weights are small, as expected, but the impact of having buildings of quality 4 is that within a cell that is hit, on average a bit more than 6 percent of the buildings are destroyed.12 Table 8: Descriptives of Weights and Intensity for Building Quality 4 (excluding weights of zero) Statistic N Mean St. Dev. Min Max Weights 5,251 0.226 0.407 0.009 9.328 Damage 5,251 0.062 0.043 0.046 0.547 Intensity 5,251 0.016 0.035 0.0004 0.859 Weights and Intensity multiplied by 1,000 11 The seven not affected were Bangka Belitung, Kalimantan Barat, Kalimantan Selatan, Kalimantan Tengah, Kalimantan Timur, Kalimantan Utara and Kepualauan Riau. 12 We did the same for buildings of quality 0 (the best) and 7 (the worst), something which led to maximum damage values of 33% for the best buildings and 84% for the worst versus 55% for our base case, showing how the overall damage is highly dependent on building quality information. 17 5.1.3 Aggregated data When aggregating, a similar method as in section 4 is used, but now the aggregation is done directly by nightlight cells instead of by basin. The equation is: L EDj,t = ED l,t (8) l where t is year, l is all nightlight cells in the province or district j and ED is the damage from equation 7. Table 9 provides the full overview of damage by province, showing the large differences between the provinces and how they vary from year to year, as one can expect with highly randomized events such as earthquakes. Using Yogyakarta - which was only impacted by earthquakes in 2006 - as an example, there was a loss of 0.4 percent of the total building mass, causing damages estimated to be approxi- mately 3.1 billion US dollars in addition to more than 5,000 deaths and tens of thousands of injured and displaced people. Apart from that, provinces on Sumatra make up 6 of the top 10 most damaging years. Even though Indonesia is often hit by severe earthquakes, even in the worst of years they only destroy about 1 percent of the buildings in a province. That being said, 1 percent of total building mass and infrastructure being damaged does constitute a significant portion of local budgets. As another example, the September 2009 earthquake in West Sumatra inflicted damages for an estimated 2.3 billion US dollars, with repair costs and losses of 64 million US dollars on government buildings (Raschky, 2013). The numbers per district are shown in Table 10, and they suffer much more damage than the provinces, with the most impacted district losing 5 percent of building mass. 18 Table 9: Aggregated Earthquake Damage Data by Province Province 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Aceh 0.040 0.728 0.160 0.152 0.044 0.010 0.269 0.345 4.597 3.869 0.000 Bali 0.557 0.014 0.184 Banten 0.003 0.268 0.032 0.142 0.022 Bengkulu 0.447 0.623 8.905 0.380 0.270 0.378 0.154 1.224 0.176 Gorontalo 0.021 0.929 0.000 0.404 0.000 Irian Jaya Barat 0.030 0.178 0.410 0.000 0.000 1.217 Jakarta Raya 0.060 Jambi 1.071 0.033 0.070 0.000 Jawa Barat 0.018 0.047 0.171 0.007 0.047 0.000 Jawa Tengah 0.404 Jawa Timur 0.012 0.190 Lampung 0.019 0.040 0.000 19 Maluku 0.000 0.253 0.378 0.176 0.000 0.000 0.049 5.202 0.105 Maluku Utara 0.095 0.769 0.644 0.000 0.000 0.147 0.238 4.216 0.918 Nusa Tenggara Barat 0.006 0.248 0.771 0.038 0.187 0.031 Nusa Tenggara Timur 0.215 0.000 0.048 0.070 0.274 0.026 0.804 1.056 Papua 1.070 0.000 0.000 0.056 0.021 0.235 0.272 0.687 0.392 0.048 Riau 0.003 0.056 0.360 Sulawesi Barat 0.000 0.123 1.431 Sulawesi Selatan 0.261 0.000 Sulawesi Tengah 0.000 0.654 0.185 0.180 0.659 0.050 1.513 5.346 0.000 Sulawesi Tenggara 0.110 0.052 0.060 0.464 Sulawesi Utara 0.069 0.050 0.651 0.011 0.263 0.111 Sumatera Barat 0.032 0.175 0.000 9.094 0.173 3.072 0.008 0.000 0.000 0.316 Sumatera Selatan 0.000 0.816 0.000 Sumatera Utara 0.002 0.208 0.072 0.010 0.041 0.022 0.175 2.175 0.048 0.000 Yogyakarta 4.318 Note: Multiplied by 1,000 Table 10: 10 Most Impacted Districts District Province Year Intensity Alor Nusa Tenggara Timur 2014 53.184 Waropen Papua 2010 45.070 Mukomuko Bengkulu 2007 44.215 Aceh Tengah Aceh 2013 41.736 Nabire Papua 2004 39.864 Bener Meriah Aceh 2013 37.750 Sangihe Talaud Sulawesi Utara 2009 32.947 Lembata Nusa Tenggara Timur 2012 31.692 Halmahera Selatan Maluku Utara 2013 31.511 Aceh Barat Aceh 2012 22.141 Note: Multiplied by 1,000 20 6 Volcano Damage Index A volcanic eruption consists of ash clouds, pyroclastic flows and lava flows, the latter two which are very difficult to model without extensive local data. Unfortunately, there is little to no academic research that has looked into large scale volcanic modeling for all aspects of eruptions. For modeling ash clouds, Joyce et al. (2009) points to remote sensing through satellite images that detect SO2 emissions as a potential method. To construct a damage index for eruptions, we use a two-fold process. First, volcanic ash advisory data are used from Volcanic Ash Advisory Centers (VAAC) to detect eruptions; second, satellite images containing sulphur dioxide data from the OMI/AURA satellite are used to model the intensity of the eruptions. The OMI/AURA images have been utilized by Carn et al. (2009) and Ferguson et al. (2010) to model eruption intensity. 6.1 Volcano Modeling There are at least two types of software that are used to model eruptions. Voris (Felpeto et al., 2007), which models ash clouds, lava flows and energy cones, and HYSPLIT from Air Resources Laboratory, which models air pollution dispersion. Voris relies on highly localized data due to the lava flow and energy cone modeling, which one will not have in most cases and that does not fit well for large scale modeling across time and space. HYSPLIT does have batch inputs, but still requires several inputs per eruption, some which are not easily obtainable. The third solution, which is related to the ash clouds mentioned by Joyce et al. (2009), is based off ash advisory data to determine whether an eruption happened and OMI/AURA satellite data to determine the scale of the eruption. This will not help with modeling lava flows and energy cones, but due to the very localized nature of the flows, there are no good sources or methods to model it for several eruptions from different volcanoes, leaving the ash clouds as a good proxy for all damages. 6.1.1 Ash advisory data Ash advisories from the Darwin VAAC (DVAAC), which are ash cloud warnings for airplanes, are used to determine whether an eruption is happening. The warnings show relevant data such as volcano name, position, summit height, height of clouds and a color code that reflects the condition of the air/volcano. There are 4 different codes ranking from the normal state, green, to imminent danger of or ongoing volcanic eruption, red. Over the period from 2004 until 2015, the DVAAC issued 12,962 warnings and of these more than 90 percent were either of code red or orange. Data on advisories from code orange or below were not used, due to eruptions of this scale not being large enough to be properly captured by the SO2 -data. By limiting the data to code red events, there are 1,785 events spread across 587 dates.13 6.1.2 OMI/AURA Satellite images To measure the intensity of an eruption, data from the Sulphur Dioxide images of the OMI/Aura project (Krotkov & Li, 2006) are used. These consist of satellite images from October 2004 and on- wards. The data have been used to model ash cloud intensity and movement in several articles such as Carn et al. (2009) and Ferguson et al. (2010). The satellite images have a spatial resolution of 13 ∗ 24km and are taken from 80km above ground. The spectral imaging shows the SO2 vertical column density in Dobson Units and there are 14 or 15 orbits per day, where one orbit covers an area approximately 2,600km wide. A dobson unit is a measure of density, and at sea level the typical concentration in clean air is less than 0.2DU. The images contain 4 values for column density based on the center of mass altitude (CMA), which is a measure of the altitude one assumes the center of the distribution is at. There are 4 different estimates for the vertical column density, ranging from 0.9km above ground to 18km above ground. 13 There were 571 different dates, but some of these dates issued red warnings to 2 or more volcanoes. 21 For volcanic activities one normally uses a CMA of either 8km or 18km (OMI team, 2012), where the former is a middle tropospheric column (TRM) and is for use in medium eruptions, while the latter is an upper tropospheric and stratospheric column (STL) and is for explosive eruptions. Despite this difference, the data are interchangeable in the sense that one can interpolate from one CMA to the others. OMI is more sensitive above clouds, which both measures mentioned will normally be. The standard deviation for both measures is as low as 0.1DU over Indonesia. The data for both STL and TRM are very similar and this paper uses the STL-data as that are most useful for the biggest events. 6.2 Creation of Damage Index and Results 6.2.1 Damage Index When constructing a damage index based on SO2 values from ash clouds, one has to set thresholds for distance from the event and from the centroid of the nightlight cell and also a lower sulphur dioxide-value. There are no papers or literature that have estimated any parameter values and there are no usable local data, so the thresholds have been set somewhat arbitrarily. First off, one wants to set a distance threshold estimating how far away the eruption could cause damage. Note that one wants ground results and not for the aviation industry, since planes can be affected very far away as evidenced by the total stand still of planes across Europe during the 2010 okull eruption. We decided to set a very relaxed condition with any point closer than 10 Eyjafjallaj¨ degrees of latitude and longitude included. Figure 9 portrays the plume approximately 7 hours after one of the biggest Merapi eruptions on 4 November 2010, where the plume moved relatively slowly and after 1000km it dissipated at the lower altitudes, which shows that our 10 degree threshold works well. Secondly, to match the nightlight data with the OMI/Aura data, a maximum distance between a nightlight point and the nearest SO2 point is set at 50km. The SO2 points are fairly scattered due to cloud covers, hence to get a more consistent grid of nightlight and SO2 values we have chosen a distance that is approximately two times the longest side of an OMI cell. Finally, a minimum SO2 value in Dobson Units is chosen. According to the Belgian Institute for Space Aeronomy, a typical normal level in air is 0.1DU and a strong eruption is above 10, which is the threshold value chosen. Once the thresholds have been set, the same nightlight weighting method as for our other indices is applied and then the weights are multiplied with the SO2 value to get an intensity value. The equation is: 0 : V SO2 < 10 V Di,t = (9) wi,T −1 ∗ VSO2 i : V SO2 ≥ 10 where i is the nightlight cell on date t, and wi,T −1 is the previously used weight where i is the nightlight cell, T − 1 is the nightlight strength from the prior year and it is divided by the sum of total nightlights in the province or district. 22 Figure 9: Merapi Ash-Cloud 4 November 2010 at 05.33UTC (7h post-eruption) 6.2.2 Results Applying our constraints, the 587 dates with a red warning have been reduced to 16 days. Of these, 7 are from the 2010 eruption on Merapi, the biggest event during the time period. Table 11 provides the affected nightlight cells by year and volcano and the results are closely correlated with the events of the period. The main one is the Merapi eruption in 2010, Soputan with volcanic explosivity index events of level 2 and 3 in 2004, 2005, 2007 and 2008 (Global Volcanism Program, 2013) and Sinabung with several eruptions in 2014, which all fit the model well. Table 11: Nightlight Cells by Year and Volcano Volcano 2004 2005 2007 2008 2010 2014 Total Kelut 2,311 2,311 Manam 62 62 Merapi 129,352 129,352 Sangeang Api 1,156 1,156 Sinabung 3,566 3,566 Soputan 6,164 586 4,672 3,704 15,126 Total 6,164 648 4,672 3,704 129,352 7,033 151,573 Table 12 refer to province impacts, and the eruptions affected numerous provinces on Java and Suma- tra, with Jawa Barat and Jawa Tengah being the most affected with more than 120,000 cells with an SO2 -value above 10 at some point. This is further underlined by nine of the ten most affected districts in Table 13 being in these two provinces, which are linked to the Merapi eruption in 2010. Apart from that, Sulawesi Utara were affected all the years from 2004 through 2008 by the eruptions on Soputan. The final table in this section, Table 14, provide descriptives of the SO2 variable and the nightlight weights, as well as the product of the two. Overall, the mean SO2 value during these events is almost 20, with a max close to 60, which is 600 times the normal level of 0.1DU SO2 . 23 Table 12: Affected Nightlight Cells by Province and Year Province 2004 2005 2007 2008 2010 2014 Total Aceh 2,078 2,078 Banten 940 940 Jawa Barat 52,034 52,034 Jawa Tengah 71,356 71,356 Jawa Timur 2,320 2,311 4,631 Nusa Tenggara Timur 1,156 1,156 Papua 62 62 Sulawesi Utara 6,164 586 4,672 3,704 15,126 Sumatera Utara 1,488 1,488 Yogyakarta 2,702 2,702 Total 6,164 648 4,672 3,704 129,352 7,033 151,573 Table 13: Top 10 Districts with Most Affected Nightlight Cells District Province Affected Cells Cilacap Jawa Tengah 16,114 Sukabumi Jawa Barat 10,724 Kebumen Jawa Tengah 10,080 Ciamis Jawa Barat 9,612 Banyumas Jawa Tengah 9,384 Cianjur Jawa Barat 8,260 Brebes Jawa Tengah 6,112 Minahasa Selatan Sulawesi Utara 5,960 Garut Jawa Barat 5,622 Bandung Jawa Barat 5,342 Table 14: Descriptives of Weights and Intensity (Excluding weights of zero) Statistic N Mean St. Dev. Min Max Weights 114,587 0.046 0.084 0.007 2.354 SO2 level 114,587 19.748 10.556 10.083 57.231 Intensity 114,587 0.855 1.707 0.075 42.857 Weights and Intensity multiplied by 1,000 6.2.3 Aggregated data To aggregate the data, the same method as before is applied, with: T I V Dj,T = VD i,t (10) t i where all days t of year T and all nightlight cells i in province or district j are aggregated. The province overview, shown in Table 16, is caused by the Merapi eruption, with 3 of the top 4 being due to that eruption. Jawa Tengah and Yogyakarta are the two most affected provinces, given their immediate proximity to Merapi. One thing to note is that Jawa Timur, which is east of Merapi, was hardly affected at all. For districts, the Merapi results are even more pronounced, with all districts in the top 10 being from the 2010 eruption. All the Jawa Tengah districts are west of the volcano. It is somewhat surprising to find that some of the districts in the immediate vicinity of Merapi are not on the list, but this can be due to the timing and quality of the satellite images. Given the time interval between images, the SO2 24 clouds could have traveled past the closest districts by the time an image was taken. Regardless, the results are uniform in that all affected districts are neighbors. Overall, the model seems to give a fair picture of when and where the eruptions were at their most intense, although the ground level intensity can be hard to specify. Table 15: 10 Most Impacted Provinces Province Year Intensity Jawa Tengah 2010 34.641 Yogyakarta 2010 18.607 Sulawesi Utara 2004 16.224 Jawa Barat 2010 10.271 Sulawesi Utara 2007 6.387 Sulawesi Utara 2008 4.861 Nusa Tenggara Timur 2014 2.356 Aceh 2014 2.134 Sulawesi Utara 2005 0.684 Jawa Timur 2010 0.660 Note: Multiplied by 1,000 Table 16: Aggregated Volcano Intensity Data by Province and Year Province 2004 2005 2007 2008 2010 2014 Aceh 2.134 Banten 0.163 Jawa Barat 10.271 Jawa Tengah 34.641 Jawa Timur 0.660 0.432 Nusa Tenggara Timur 2.356 Papua 0.061 Sulawesi Utara 16.224 0.684 6.387 4.861 Sumatera Utara 0.487 Yogyakarta 18.607 25 Table 17: 10 Most Impacted Districts District Province Year Intensity Purwokerto Jawa Tengah 2010 184.673 Banyumas Jawa Tengah 2010 168.916 Cilacap Jawa Tengah 2010 149.878 Kebumen Jawa Tengah 2010 141.635 Banjarnegara Jawa Tengah 2010 106.472 Purbalingga Jawa Tengah 2010 98.697 Purworejo Jawa Tengah 2010 92.593 Kulon Progo Yogyakarta 2010 72.177 Wonosobo Jawa Tengah 2010 69.568 Banjar Jawa Barat 2010 60.208 Note: Multiplied by 1,000 26 7 Tsunami Damage Index The final disaster damage index constructed is for the 2004 Christmas tsunami. There is little local dis- trict level damage data available, so it was decided to use the methodology from Heger (2016), whereby inundation maps are used to construct a district level damage index assuming a uniform damage across all flooded areas. 7.1 Creation of Damage Index and Results 7.1.1 Damage Index Despite all the media coverage and attention the 2004 tsunami had, there is not much detailed spatial information readily available for research. Heger (2016) has done some research on the causal effects of the tsunami in his PhD thesis, and we will follow his method closely to model flood impact. To construct an inundation map of the affected areas, a map based on MODIS satellite pictures from Anderson et al. (2004) is used. The map itself is fairly low resolution, but it provides a good overview of the inundated areas. In terms of the intensity of the flood, there are no existing data on that, but a uniform flood intensity across all flooded areas is assumed, just as in Heger (2016). The resulting map is shown in Figure 10, which shows that a large proportion of the Aceh coastline was struck by the tsunami. To make this map, the inundation map from DFO was used as a base. Spatial algorithms were then applied to detect the difference in color between inundated and non-inundated areas. This process started with overlaying the base map on a regular shapefile of Indonesia, then detecting the specific color of inundated areas, before constructing a new shapefile where all inundated areas (areas with the same color) have value 1 and all other areas have value 0. Figure 10: Inundation Map of 2004 Tsunami 7.1.2 Results Figure 11 shows the inundated area nightlight cells, combined with the nightlight cells in all of Aceh. The tsunami did not strike the most densely populated areas, 460 nightlight cells were hit, out of a total of 13,456 cells in all of Aceh. Given that the tsunami happened 26 December 2004, it is more appropriate to link the incident with the 2004 nighlights, instead of 2003, as is done for the other disaster types. Using the 2004 numbers, there were 364 lit inundated cells out of 7,607 total lit cells in Aceh. Interestingly, in 2005, the year after the disaster, there is a strong decline, with 306 and 6,352 lit cells for the inundated 27 areas and Aceh, respectively. The average light intensity has gone down, from 7.39 per cell to 6.33 in the inundated areas and from 6.37 to 5.18 in the province as a whole. Figure 11: Aceh Nightlights and Tsunami Affected Nightlights Finally, Table 18 shows the weights which are - again - defined as nightlight in the cell over total nightlight in the province. Although the numbers look very small, by multiplying the means by number of cells, one gets approximately 5.5 percent. Knowing that the census numbers for Aceh in 2000 gave a population of just under 4 million and in 2010 of just under 4.5 million and if one multiplies the population numbers with the affected cells number of 5.5 percent of the total, one gets 221,894 and 249,631, respectively. Given the official numbers of 166,671 dead due to the tsunami, an assumption of total destruction in all inundated areas seem a bit high, given that 166,671 of 230,000 is 72.47 percent. A damage of 75 percent in the inundated cells is chosen, giving a final damage index formula: TD i = Wi ∗ D (11) li where TD i is the province weighted damage from nightlight cell i, Wi = I l i.e. the light in cell i i i over the sum of all nightlight in the province and D is the flat damage number of 0.75. Table 18: Descriptives of Weights by Year All Cells Only Lit Cells a a Year Cells Mean (st.dev ) Total Cells Meana (st.deva ) 2003 460 0.0965 (0.1356) 295 0.1505 (0.1433) 2004 460 0.1206 (0.1369) 364 0.1524 (0.1373) 2005 460 0.1281 (0.1656) 306 0.1925 (0.1698) a. Multiplied by 1000 28 7.1.3 Aggregated data Aggregating the data is done using the same method as in all previous sections, where the nightlight cells across the province or district is summed up: I T Dj = TD i (12) i where all nightlight cells i in province or district j are aggregated. Since the tsunami only affected one province, it is easy to see the total damage done by it on Aceh. With our assumptions, the tsunami destroyed about 4 percent of the buildings in Aceh. This is clearer once broken down into district level damage. There were 6 districts affected by the tsunami, with Aceh Jaya, Banda Aceh and Pidie all experiencing damage of more than 20 percent. The other 3 affected districts - Aceh Barat, Aceh Besar and Bireuen - experienced damage between 5 and 10 percent due to the tsunami. Table 19: Aggregated Tsunami Damage by District and Province District Province Intensity Aceh Barat Aceh 0.071 Aceh Besar Aceh 0.078 Aceh Jaya Aceh 0.221 Banda Aceh Aceh 0.229 Bireuen Aceh 0.055 Pidie Aceh 0.210 Aggregated Aceh 0.042 29 8 Conclusion With the continuous increase in remote sourcing data, it has gotten much easier and cheaper to monitor and assess the damages from natural disasters. Joyce et al. (2009) and Gillespie et al. (2007) have done an extensive review of how satellite images can be used to map natural disasters, while this paper has contributed with providing new techniques that utilize other remote sourced data such as ShakeMaps and ash advisory data. Throughout, techniques based on freely available data have been used to construct damage indices for different disaster types. Generally the indices can be used in any area of the world, and if calibrated with local data, they could provide an excellent tool for local governments or stakeholders in early disaster assessments. The indices can be used to get quick damage estimates and inform where to provide relief, as well as in research such as what the authors have done in Skoufias et al. (2017), where the indices are used to analyze district budget redistributions following natural disasters. The main caveat is the indices have not been validated against local level damage data. If one had access to high resolution monetary or intensity data, the estimates would be much more precise. Table 20 gives an overview of the different data sources and software used. All disasters have used the DMSP global nightlights data to weight the indices based on economic activity. Recently, the VIIRS nightlight data provide an alternative for assessing economic activity or events, as showcased for GDP in China (Li et al., 2013; Shi et al., 2014) and Africa (Chen & Nordhaus, 2015) and for storms and floods in the US (Cao et al., 2013; Sun et al., 2015). If one is interested in events after 2012, the VIIRS data provide higher spatial resolution and track changes by month instead of yearly. Alternatives do exist for choice of software if one wants everything to be open source and free. Instead of ArcGIS, packages such as QGIS and RGDAL for R are potential alternatives. R and Python are used more and more for spatial data and can often provide the necessary tools to do the modeling. Instead of statistical software such as Matlab and Stata, R and Python once more provide excellent alternatives. Table 20: Disasters and the Data Sources and Softwares used Disaster Type Data Sources Software Used Hydro1K GeoSFM (ArcView 3.3) GLCC Python Flood FAO Digital Soil Map Stata TRMM ArcGIS GDAS USGS ShakeMaps USGS Building Inventory Python Earthquake World Housing Encyclopedia Stata Urban Extent Map (CIESIN) ArcGIS GESI Python Ash Advisory Data Matlab Volcanic Eruption OMI/AURA Satellite Images Stata ArcGIS Python Tsunami DFO Inundation Map ArcGIS Stata 30 References ebastien. 2004. DFO Event number 2004 - 193 - Anderson, Elaine, Brakenridge, G.R., & Caquard, S´ Aceh Province Inundation Map 1. http://www.dartmouth.edu/ floods/2004193.html. Online; accessed 1 June 2016. Artan, G.A., Asante, K., Smith, J., Pervez, S., Entenmann, D., Verdin, J., & Rowland, J. 2008. Users Manual for the Geospatial Stream Flow Model (GeoSFM): U.S. Geological Survey Open-File Report 20071440. User Manual. USGS. Asante, Kwabena O., Macuacua, Rodrigues D., Artan, Guleid A., Lietzow, Ronald W., & Verdin, James P. 2007. Developing a Flood Monitoring System From Remotely Sensed Data for the Limpopo Basin. IEEE Transactions on Geoscience and Remote Sensing, 45(6), 1709–1714. Athukorala, Prema-chandra, & Resosudarmo, Budy P. 2005. The Indian Ocean Tsunami: Economic Impact, Disaster Management, and Lessons. MIT Press, 4(1), 1–39. Atkinson, Gail M., & Kaka, SanLinn I. 2007. Relationships between Felt Intensity and Instrumental Ground Motion in the Central United States and California. Bulletin of the Seismological Society of America, 97(2), 497–510. Atkinson, Gail Marie, & Kaka, SanLinn I. 2006. Relationships between felt intensity and instrumental ground motion for New Madrid ShakeMaps. Department of Earth Sciences, Carleton University. Balk, D.L., Deichmann, U., Yetman, G., Pozzi, F., Hay, S.I., & Nelson, A. 2006. Determining Global Pop- ulation Distribution: Methods, Applications and Data. Pages 119–156 of: Advances in Parasitology. Elsevier. Brakenridge, R., & Anderson, E. 2006. MODIS-based flood detection, mapping and measurement: The potential for operational hydrological applications. Pages 1–12 of: Marsalek, Jiri, Stancalie, Gheo- rghe, & Balint, Gabor (eds), Nato Science Series: IV: Earth and Environmental Sciences. Springer Netherlands. Cao, Changyong, Shao, Xi, & Uprety, Sirish. 2013. Detecting Light Outages After Severe Storms Using the S-NPP/VIIRS Day/Night Band Radiances. IEEE Geoscience and Remote Sensing Letters, 10(6), 1582–1586. Carn, Simon A., Krueger, Arlin J., Krotkov, Nickolay A., Yang, Kai, & Evans, Keith. 2009. Tracking volcanic sulfur dioxide clouds for aviation hazard mitigation. Natural Hazards, 51(2), 325–343. Center for International Earth Science Information Network - CIESIN - Columbia University, Interna- tional Food Policy Research Institute - IFPRI, The World Bank, & Centro Internacional de Agri- cultura Tropical - CIAT. 2011. Global Rural-Urban Mapping Project, Version 1 (GRUMPv1): Ur- ban Extents Grid. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC). http://dx.doi.org/10.7927/H4GH9FVG. Accessed 02 01 2017. Chen, Xi, & Nordhaus, William. 2015. A Test of the New VIIRS Lights Data Set: Population and Economic Output in Africa. Remote Sensing, 7(4), 4937–4947. Chung, Hsiao-Wei, Liu, Cheng-Chien, Cheng, I-Fan, Lee, Yun-Ruei, & Shieh, Ming-Chang. 2015. Rapid Response to a Typhoon-Induced Flood with an SAR-Derived Map of Inundated Areas: Case Study and Validation. Remote Sensing, 7(9), 11954–11973. De Groeve, Tom, Annunziato, A, Gadenz, S, Vernaccini, L, Erberik, A, & Yilmaz, T. 2008. Real-time impact estimation of large earthquakes using USGS Shakemaps. Proceedings of IDRC2008, Davos, Switzerland. Dell’Acqua, Fabio, & Gamba, Paolo. 2012. Remote Sensing and Earthquake Damage Assessment: Ex- periences, Limits, and Perspectives. Proceedings of the IEEE, 100(10), 2876–2890. Dessu, Shimelis Behailu, Seid, Abdulkarim Hussein, Abiy, Anteneh Z., & Melesse, Assefa M. 2016. Flood Forecasting and Stream Flow Simulation of the Upper Awash River Basin, Ethiopia Using Geospatial Stream Flow Model (GeoSFM). Cham: Springer International Publishing. Pages 367–384. 31 Elvidge, Christopher, Baugh, Kimberly, Hobson, Vinita, Kihn, Eric, Kroehl, Herbert, Davis, Ethan, & Cocero, David. 1997. Satellite inventory of human settlements using nocturnal radiation emissions: A contribution for the global toolchest. Global Change Biology, 3(5), 387–395. Federal Emergency Management Agency. 2006. HAZUS-MH MR2 Technical Manual. Federal Emergency Management Agency, Washington, D.C. Felpeto, Alicia, Mart, Joan, & Ortiz, Ramon. 2007. Automatic GIS-based system for volcanic hazard assessment. Journal of Volcanology and Geothermal Research, 166(2), 106 – 116. Ferguson, David J., Barnie, Talfan D., Pyle, David M., Oppenheimer, Clive, Yirgu, Gezahegn, Lewi, Elias, Kidane, Tesfaye, Carn, Simon, & Hamling, Ian. 2010. Recent rift-related volcanism in Afar, Ethiopia. Earth and Planetary Science Letters, 292(34), 409 – 418. Fu, Bihong, Awata, Yasuo, Du, Jianguo, Ninomiya, Yoshiki, & He, Wengui. 2005. Complex geometry and segmentation of the surface rupture associated with the 14 November 2001 great Kunlun earthquake, northern Tibet, China. Tectonophysics, 407(1-2), 43–63. GeoHazards International and United Nations Centre for Regional Development. 2001. Final report: Global Earthquake Safety Initiative (GESI) pilot project. Report. GHI. Gillespie, T. W., Chu, J., Frankenberg, E., & Thomas, D. 2007. Assessment and prediction of natural hazards from satellite imagery. Progress in Physical Geography, 31(5), 459–470. Global Volcanism Program. 2013. Volcanoes of the World, v. 4.5.3. Venzke, E (ed.). Smithsonian Insti- tution. Downloaded 25 Jan 2017. G.R.Brakenridge. 2016. Global Active Archive of Large Flood Events, Dartmouth Flood Observatory, University of Colorado. http://floodobservatory.colorado.edu/Archives/index.html. Online; accessed 29 May 2016. Haq, Mateeul, Akhtar, Memon, Muhammad, Sher, Paras, Siddiqi, & Rahmatullah, Jillani. 2012. Tech- niques of Remote Sensing and GIS for flood monitoring and damage assessment: A case study of Sindh province, Pakistan. The Egyptian Journal of Remote Sensing and Space Science, 15(2), 135–141. Heger, Martin Philipp. 2016. The Causal Effects of the Indian Ocean Tsunami and Armed Conflict on Acehs Economic Development. Ph.D. thesis, London School of Economics. Henderson, J. Vernon, Storeygard, Adam, & Weil, David N. 2012. Measuring Economic Growth from Outer Space. American Economic Review, 102(2), 994–1028. Hodler, Roland, & Raschky, Paul A. 2014. Regional Favoritism. The Quarterly Journal of Economics, 129(2), 995–1033. Holden, Z. A., Smith, A. M. S., Morgan, P., Rollins, M. G., & Gessler, P. E. 2005. Evaluation of novel thermally enhanced spectral indices for mapping fire perimeters and comparisons with fire atlas data. International Journal of Remote Sensing, 26(21), 4801–4808. Jaiswal, K.S., & Wald, D.J. 2008. Creating a global building inventory for earthquake loss assessment and risk management: U.S. Geological Survey Open-File Report 2008-1160. Tech. rept. USGS. Joyce, K. E., Belliss, S. E., Samsonov, S. V., McNeill, S. J., & Glassey, P. J. 2009. A review of the status of satellite remote sensing and image processing techniques for mapping natural hazards and disasters. Progress in Physical Geography, 33(2), 183–207. Klemas, Victor V. 2009. The Role of Remote Sensing in Predicting and Determining Coastal Storm Impacts. Journal of Coastal Research, 256(nov), 1264–1275. Knebl, M.R., Yang, Z.-L., Hutchison, K., & Maidment, D.R. 2005. Regional scale flood modeling using NEXRAD rainfall, GIS, and HEC-HMS/RAS: a case study for the San Antonio River Basin Summer 2002 storm event. Journal of Environmental Management, 75(4), 325–336. Kreibich, H., Piroth, K., Seifert, I., Maiwald, H., Kunert, U., Schwarz, J., Merz, B., & Thieken, A. H. 2009. Is flow velocity a significant parameter in flood damage modelling? Natural Hazards and Earth System Sciences, 9(5), 1679–1692. 32 Krotkov, Nickolay A., & Li, Can. 2006. OMI/Aura Sulphur Dioxide (SO2) Total Column 1-orbit L2 Swath 13x24 km V003, Greenbelt, MD, USA, Goddard Earth Sciences Data and Information Services Center (GES DISC), , Accessed 30 August 2016. Li, Xi, Xu, Huimin, Chen, Xiaoling, & Li, Chang. 2013. Potential of NPP-VIIRS Nighttime Light Imagery for Modeling the Regional Economy of China. Remote Sensing, 5(6), 3057–3081. Linkimer, Lepolt. 2007. Relationship between peak ground acceleration and Modified Mercalli intensity ogica de Am´ in Costa Rica. Revista Geol´ erica Central. Mati, Bancy M., Mutie, Simon, Gadain, Hussein, Home, Patrick, & Mtalo, Felix. 2008. Impacts of land-use/cover changes on the hydrology of the transboundary Mara River, Kenya/Tanzania. Lakes & Reservoirs: Research & Management, 13(2), 169–177. Merz, B., Kreibich, H., Schwarze, R., & Thieken, A. 2010. Review article ”Assessment of economic flood damage”. Natural Hazards and Earth System Sciences, 10(8), 1697–1724. Michalopoulos, Stelios, & Papaioannou, Elias. 2014. National Institutions and Subnational Development in Africa. The Quarterly Journal of Economics, 129(1), 151–213. Murphy, J. R., & O’Brien, L. J. 1977. The correlation of peak ground acceleration amplitude with seismic intensity and other physical parameters. Bulletin of the Seismological Society of America, 67(3), 877–915. Myint, S. W., Yuan, M., Cerveny, R. S., & Giri, C. 2008. Categorizing natural disaster damage assessment using satellite-based geospatial techniques. Natural Hazards and Earth System Science, 8(4), 707–719. National Disaster Management Agency, BNPB. 2016. DiBi database (Data and Information on Disaster in Indonesia). Nichol, Janet E., Shaker, Ahmed, & Wong, Man-Sing. 2006. Application of high-resolution stereo satellite images to detailed landslide hazard assessment. Geomorphology, 76(1-2), 68–75. OMI team. 2012. Ozone Monitoring Instrument (OMI) Data Users Guide. User Manual. OMI. Penning-Rowsell, Edmund, Johnson, Clare, Tunstall, Sylvia, Tapsell, Sue, Morris, Joe, Chatterton, John, & Green, Colin. 2005. The benefits of flood and coastal risk management: a handbook of assessment techniques. Tech. rept. Middlesex University Press. Raschky, Paul A. 2013. Discussion Paper: Estimating the Effects of West Sumatra Public Asset Insurance Program on Short-Term Recovery after the September 2009 Earthquake. Tech. rept. 2013-35. ERIA. omer, H., Willroth, P., Kaiser, G., Vafeidis, A. T., Ludwig, R., Sterr, H., & Diez, J. Revilla. 2012. R¨ Potential of remote sensing techniques for tsunami hazard and vulnerability analysis – a case study from Phang-Nga province, Thailand. Natural Hazards and Earth System Science, 12(6), 2103–2126. Roy, D.P., Boschetti, L., & Trigg, S.N. 2006. Remote Sensing of Fire Severity: Assessing the Performance of the Normalized Burn Ratio. IEEE Geoscience and Remote Sensing Letters, 3(1), 112–116. Scawthorn, Charles, Flores, Paul, Blais, Neil, Seligson, Hope, Tate, Eric, Chang, Stephanie, Mifflin, Edward, Thomas, Will, Murphy, James, Jones, Christopher, et al. 2006. HAZUS-MH flood loss estimation methodology. II. Damage and loss assessment. Natural Hazards Review, 7(2), 72–81. Shi, Kaifang, Yu, Bailang, Huang, Yixiu, Hu, Yingjie, Yin, Bing, Chen, Zuoqi, Chen, Liujia, & Wu, Jianping. 2014. Evaluating the Ability of NPP-VIIRS Nighttime Light Data to Estimate the Gross Domestic Product and the Electric Power Consumption of China at Multiple Scales: A Comparison with DMSP-OLS Data. Remote Sensing, 6(2), 1705–1724. Shrestha, M.S., Artan, G.A., Bajracharya, S.R., Gautam, D.K., & Tokar, S.A. 2011. Bias-adjusted satellite-based rainfall estimates for predicting floods: Narayani Basin. Journal of Flood Risk Man- agement, 4(4), 360–373. Skoufias, Emmanuel, Strobl, Eric, & Tveit, Thomas. 2017. The Reallocation of District-Level Spend- ing and Natural Disasters: Evidence from Indonesia. World Bank Policy Research Working Paper (Forthcoming). 33 Sun, Donglian, Li, Sanmei, Zheng, Wei, Croitoru, Arie, Stefanidis, Anthony, & Goldberg, Mitchell. 2015. Mapping floods due to Hurricane Sandy using NPP VIIRS and ATMS data and geotagged Flickr imagery. International Journal of Digital Earth, 9(5), 427–441. The Atlantic. 2014 (December). Ten Years Since the 2004 Indian Ocean Tsunami. The Global Facility for Disaster Reduction and Recovery. 2011. Indonesia: Advancing a National Dis- aster Risk Financing Strategy Options for Consideration. Report. World Bank. Tralli, David M., Blom, Ronald G., Zlotnicki, Victor, Donnellan, Andrea, & Evans, Diane L. 2005. Satellite remote sensing of earthquake, volcano, flood, landslide and coastal inundation hazards. ISPRS Journal of Photogrammetry and Remote Sensing, 59(4), 185–198. Wald, David J., Quitoriano, Vincent, Heaton, Thomas H., Kanamori, Hiroo, Scrivner, Craig W., & Worden, C. Bruce. 1999. TriNet ShakeMaps: Rapid Generation of Peak Ground Motion and Intensity Maps for Earthquakes in Southern California. Earthquake Spectra, 15(3), 537–555. Wald, D.J., Worden, B.C., Quitoriano, V., & Pankow, K.L. 2005. ShakeMap Manual - Technical Manual, User’s Guide, and Software Guide. Technical Manual. USGS. World Bank. 2005 (January). Indonesia: Preliminary Damage and Loss Assessment: The December 26, 2004 Natural Disaster. Wu, Huan, Adler, Robert F., Hong, Yang, Tian, Yudong, & Policelli, Fritz. 2012. Evaluation of Global Flood Detection Using Satellite-Based Rainfall and a Hydrologic Model. Journal of Hydrometeorology, 13(4), 1268–1284. Wu, Huan, Adler, Robert F., Tian, Yudong, Huffman, George J., Li, Hongyi, & Wang, JianJian. 2014. Real-time global flood estimation using satellite-based precipitation and a coupled land surface and routing model. Water Resources Research, 50(3), 2693–2717. Yamazaki, Fumio, & Matsuoka, Masashi. 2007. Remote Sensing Technologies in Post Disaster Damage Assessment. Journal of Earthquake and Tsunami, 01(03), 193–210. 34