Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Using VIIRS nightlights to estimate the impact of the 2015 Nepal earthquakes

Using VIIRS nightlights to estimate the impact of the 2015 Nepal earthquakes We use Visible Infrared Imaging Radiometer Suite ( VIIRS) nightlight data to model the impact of the 2015 Nepal earth- quakes. More specifically, the data—showing nightlight emissions—are used to examine the extent to which there is a difference in nightlight intensity between cells damaged in the earthquake versus undamaged cells based on (1) mean comparisons; and (2) fixed effect regression models akin to the double difference method. The analysis is car - ried out for the entire country as well as smaller regions in and around the Central area and Kathmandu, which were the hardest hit areas. Overall, the regressions find a significant and negative effect from the initial shock, followed by a positive net effect from aid and relief efforts, which is consistent with what one would expect to find. However, the mean analysis results are inconclusive and there is substantial noise in the nightlight measurements due to how the values are produced and persistent cloud cover over Nepal. Keywords: Earthquakes, Night-time lights (NTL), VIIRS, Remote sensing, Econometrics by natural hazards, see for example Elliott et  al. (2015), Introduction Klomp (2016), Skoufias et  al. (2017), Nguyen and Noy The Gorkha earthquake that struck Nepal on 25 April (2019) or Gao et al. (2020). 2015 is one of the biggest natural hazards of the last dec- The aim of this paper is to build upon this literature by ade. The widespread damages totalled USD 10 billion in showcasing how high-resolution nightlight and popula- monetary damages, leaving 8857 dead, 22,304 injured tion data can be used to quantify highly localized changes and tens of thousands homeless. In addition to the main in economic activity as proxied by monthly nightlight earthquake, several significant aftershocks also occurred, data. By using population data as a secondary layer, we causing additional damages and deaths. Following any seek to avoid several known issues with the nightlight natural hazard of this scale, it is important to quickly and value measurements, such as blooming, airglow contami- accurately identify the areas which are most in need of nation and rural area values being more poorly correlated aid and where one can expect the highest damages, both with economic activity. In addition, while existing articles in terms of casualties and monetary losses. The two are have focused on identifying affected areas immediately often correlated as the level of human activity usually after impact, this paper uses mean analysis and fixed coincides with economic output. Often, the identification effects regressions to quantify the initial impact and tem - of damages are done by on the ground observations or by poral effect over the 12  months ex-post. This is similar manually examining images taken from air or from space, to Gao et  al. (2020), but whereas they examined purely but over the last two decades a growing body of litera- based on nightlight change, we use objectively measured ture has focused on using remotely sensed data and auto- earthquake data to identify nightlight cells that were mated algorithms to analyze and detect changes caused impacted by the earthquake and compared these with cells that experienced no damage. *Correspondence: Thomas.tveit@vwi.unibe.ch The methodology of this paper is the same as used in Department of Economics, University of Bern, Schanzeneckstrasse 1, 3001 Bern, Switzerland Skoufias et al. (2021), where the authors found that VIIRS Full list of author information is available at the end of the article © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 2 of 13 nightlight data provided limited results when analyzing performed on the combined data. First, a simple mean larger countries and numerous natural hazards. In this comparison is performed where we compare the night- paper, we use the same method, but across a smaller area light values between affected and non-affected cells for and on a more damaging hazard, making it arguably an the 12 months before and after the earthquake. Secondly, ideal setting for performing such an analysis. In short, we run a fixed effect regression model to identify and we use contour maps from the United States Geological quantifyimpact over the first 12  months ex-post. Both Survey (USGS) to identify damaged areas and model the analysis are performed on a national nightlight set as local damage at a cell level from vulnerability curves pro- well as sets on smaller regions that were more severely vided by International and Regional Development (2001). impacted by the earthquakes. The damage data is then geomapped alongside the VIIRS The mean analysis results are not significant, poten - and population data, which act as a proxy for the eco- tially due to the general volatility and measurement nomic activity and hence the asset exposure. Finally, errors in the data. However, it could also be caused by an the combined data is used to identify and quantify what influx of aid and recovery to all areas of Nepal and not impact the damage has on nightlight values in two dif- just the ones experiencing the most damage, i.e. if aid is ferent analysis. The first is a mean comparison between not well targeted one would experience the same pattern damaged and non-damaged nightlight cells and the other across all cells. When performing a pre-post analysis of is a fixed effect regression. the mean results, we do find that there is a significant dif - The nightlights data used in literature has primarily ference in the trajectories of the two groups. To further been from two sources: the Defense Meteorological Sat- support this, the fixed effects regressions did yield sig - ellite Program (DMSP) and the VIIRS. The former is dis - nificant results, with an initial negative and significant continued and consists of annual data from 1992 through shock in the month of the disaster, followed by a statisti- 2013 with the data released being grid cells at 30 arc-sec- cally significant increase in light for 4–6  months, before ond resolution (approximately 1 km by 1 km at equator) tapering off and becoming insignificant. The first nega - where each cell has a normalized digital number (DN) tive shock followed by a positive effect is consistent with value from 0 to 63. The VIIRS data is more recent (April what one could expect from a large-scale disaster, where 2012 and onwards), has improved spatial and temporal at first infrastructure is damaged and necessary aid is detail with monthly releases at a grid cell resolution of not present, followed by an influx of aid and relief efforts 15 arc-seconds and the DN values have no upper bound. over the coming months. Following this phase, one would However, despite the advantages of the VIIRS data, it expect a decrease in nighlight values as the initial relief does have some limitations. Firstly, it is much more vola- efforts are discontinued. Additionally, after 12  months tile than the DMSP data and due to stray light correction one can potentially expect the VIIRS values to reflect a it can contain negative light values. Furthermore, some “new” normal and any significant effects found after that months have no measurements because of cloud cover. are unlikely to be directly related to the earthquake.. Despite these shortcomings, the VIIRS data values does However, the regression results include an unexpected show some correlation with local GDP (Zhao et al. 2017) and sudden positive spike 12  months after the earth- and we will use it as a proxy for economic activity in this quakes. The cause for this is unknown and the most likely article. explanation is that it is caused by noise in the measure- To improve upon the VIIRS data and attempt to ments. The regressions were run with several different decrease the volatility, we will use a settlement layer as specifications as robustness checks, but the main results a secondary layer to better identify areas with human remained unchanged. activity. Using secondary data sources to increase the The remainder of the paper starts with an overview of likelihood of correctly identifying human settlements Nepal, the earthquakes and the base data, followed by a have previously been used for poverty (Jean et  al. 2016) general methodology overview, the analysis results and and urban extent (Baragwanath et  al. 2019). The data finally a conclusion. used is from Worldpop (2017), which utilizes satellite images and census data to construct a high-resolution layer of human settlement patterns. The inclusion of the Materials and methods additional data is used to exclude non-populated cells Nepal and the 2015 earthquakes and to better distinguish rural and urban areas. Nepal is a mountainous country bordering India to the Finally, to identify and model local damage, we con- south, west and east and Tibet (China) to the north. It struct a damage index based on ShakeMaps from the is a small country at roughly 147,000 square kilometers USGS. The damage estimates are then combined with and has a population of approximately 29 million people. the economic activity proxy and two types of analysis are Its nominal GDP in 2019 is USD30.6 billion. On a per T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 3 of 13 Fig. 1 Nepal and the Gorkha earthquake. The underlying blue colour shows Nepal, whereas the green to red outlines the extent and intensity of the earthquake, with red being the most impacted areas. Source: USGS Shakemap and ArcGIS Basemap capita purchasing power adjusted basis, Nepal ranks at 8857 mortalities and 22,304 injured. In addition, tens the 162nd position globally per the World Bank. of thousands were made homeless. Some damages were In addition, Nepal is prone to disasters due to a com- also due to landslides caused by the earthquake as seen bination of the vulnerability of the building environment in Xu et  al. (2017). In this paper, we will treat all dam- and the population, its geographical location, topogra- ages as earthquake damages as the effect on nightlights phy and climate. Nepal is frequently affected by different will be the same and the cause of the damage is the earth- types of natural hazards such as landslides, floods and quake. Furthermore, to disentangle the two at a regional avalanches, stemming from the topography and the cli- or national scale would require significant manual labor. mate. Due to lack of irrigation, drought and fire are other In the wake of the Gorkha earthquake there were sev- frequent natural hazards. Finally, Nepal is situated along- eral aftershocks, with the largest one being a magnitude side a major east–west tectonic boundary plate where 7.3  MW occurring on 12 May 2015 to the east of Kath- earthquakes are frequent. INFORM earthquake exposure mandu in the districts of Dolakha and Sindhupalchowk. rank them at an 9.9 out of 10 (INFORM 2019). There were 153 dead and 3275 injured following this In 2015, Nepal suffered two major earthquakes. The earthquake. first—and largest—earthquake struck on 25 April 2015 The data used to identify the damaged areas and model and is often referred to as the Gorkha Earthquake. Fig- the damage are from contour maps generated by seis- ure 1 shows the magnitude and size of the 7.8 MW earth- mological ground stations (International and Regional quake relative to Nepal and the epicenter in the Gorkha Development 2001; Agency 2006; De Groeve et al. 2008). District. According to the UN office for Disaster Risk These contour maps—ShakeMaps from USGS—are used Reduction the total damages added up to USD 10 bil- as a basis for localized impact and they provide several lion—approximately one third of the total GDP—with key parameters such as peak ground acceleration (PGA), Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 4 of 13 peak ground velocity (PGV) and modified Mercalli inten - the Nepal earthquake. Recently, Sahoo et al. (2020) have sity (MMI). The maps use data from seismic stations and utilized machine learning methods to intercalibrate combine this with information on ground conditions and between the two data sets and construct a longer annual earthquake depth to interpolate and construct a grid of data set. However, an increase in the length of the tempo- points spaced 0.0167 degrees apart. ral dimension at the cost of temporal detail is not benefi - cial in this study. Additionally, data that is prior to 2012 is Population layer unlikely to impact on the 2016 estimated effects. To properly capture economic activity through night- The use of VIIRS in economic analysis was done early lights it is important to correctly identify areas with peo- in its lifetime by Chen and Nordhaus (2015), where they ple, hence avoiding stray lights or other sources of light found promising results for VIIRS as an economic and that are not connected to human activity. One method is population indicator, also when compared to the DMSP to use a human settlement layer as a secondary data set, product. However, the same authors found that VIIRS such as in Corbane et  al. (2016). One assumes that the was not useful in growth predictions and that the asso- settlement layer provides a better static map of the spa- ciation between economic activity and light values were tial distribution of people and then the nightlight data dependent upon the chosen area size, with state level provide a temporal and spatial overview of how the eco- analysis performing worse than metropolitan area level nomic activity is distributed within the populace. analysis (Chen and Nordhaus 2019). They hypothesized In this paper, we use the Worldpop datasets to identify that this was in part due to the economic activity type settlement areas in Nepal (Worldpop 2017). These data - in metropolitan areas and that they are more likely to be sets have a spatial gridcell resolution of approximately related to electric light. In a more recent working paper, 100  m at the equator and estimate the number of per- Gibson et al. (2019) also find that VIIRS performs poorly sons per square for the years 2010, 2011 (based on cen- in rural, low-density areas in Indonesia, while it performs sus data), 2015 and 2020. This article uses only the 2015 well and much better than DMSP data in urban areas. estimates as these are from the same year as the earth- The VIIRS products have also seen use in the natural dis - quakes. Worldpop constructed the population files by aster literature with Zhao et  al. (2018) using the under- using national totals and then adjust these to match UN lying NPP-VIIRS DNB Daily Data to analyze selected population estimates. The adjustment is done by a Ran - natural disasters, including the Gorkha earthquake. They dom Forest model, which generates a weighted popula- found that the images could be used in damage detection tion density layer and then this layer is used as a basis to and for identifying power outages, but that the analysis place the population according to the assumed real geo- was limited by cloud coverage. This article will not use graphical distribution. the daily data because of computational limitations when In addition to identifying economic activity, the settle- focusing on much longer temporal and spatial effects. ment layers will be used to designate a cell as rural and The VIIRS data provide two variables: the average urban based upon population density. monthly light radiance from DNB and the number of cloud free days. Some months will have no cloud free VIIRS days, meaning that no radiance value is provided. This is To find and identify areas with economic activity and accounted for by calculating an interpolated value from their exposure to natural hazards, satellite images of the values of the month prior and after. In addition, a nighttime lights will be used as a proxy. Since natural known problem for low light value cells is that one often hazards are highly localized, one would prefer high-res- finds negative light values due to airglow contamination olution spatially disaggregated economic data, although (Uprety et  al. 2019). Optimally, one would want to cor- nightlights have been used successfully to model local rect for this and identify the real underlying value and economic activity (Henderson et  al. 2012; Gillespie decrease the volatility exhibited. Recently, two published et  al. 2014; Hodler and Raschky 2014; Michalopoulos articles (one about the Gorkha earthquake) have explored and Papaioannou 2014). These articles use the annual methodologies for correcting the measurements through DMSP data with 30-arc second grids, whereas more trend decompositions (Gao et al. 2020; Zhao et al. 2020). recent papers have also utilized the VIIRS Day/Night However, lacking any established methodology, we Band (DNB) provided by The Earth Observations Group experimented by using simple threshold testing to iden- (EOG) at NOAA/NCEI, see for example Chen and Nord- tify how it would affect the analysis. haus (2015), Zhao et al. (2017, 2018) or (2020). The latter data set provides monthly data from April 2012 till pre- Using the value of the month of the previous year was also explored, but the sent, whereas DMSP contains nightlight data for the years overall value mean change was less than 0.01 and there were only 1 cell that 1992 through 2013, making DMSP unusable for studying was affected by the earthquake that had no radiance value. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 5 of 13 Fig. 2 VIIRS value distribution for VIIRS values below 2 Fig. 3 Days with no cloud cover in a month across all months. The X-axis shows how many days per month that were cloud free, i.e. that had a light value measurement. The Y-axis shows the distribution of number of cloud free days per month. Adding up all columns equal 1 Figure 2 show the distribution of VIIRS values below 2 for all populated areas and the number of cloud free days per month. It identifies significant clustering between the values 0 and 0.5, containing more than 85 percent the same time, almost 10% of the monthly observations of the total observations. Furthermore, more than 5% of had 0 or 1 cloud free days, with 4% of the sample hav- the total have a negative value. Analyzing the range from ing no observations. The months with no cloud free days negative to 0.3, one finds that 75% of all points fall within were included either through interpolation or by using this range, implying that a large proportion of our obser- the prior year’s value. The median and mean are 10, and vations have negative or very small values. These are only more than 12% of months had one or less cloud free days. points that have been identified as being populated, Hence, the monthly nightlight values are calculated based meaning that they should be relatively free of disturbance on a rather small subsample of days per month. How- from non-human sources. ever, for a natural hazard which is not correlated with the Even when limiting the observations to cells with pop- weather, such as earthquakes, this is unlikely to bias the ulation in excess of 50 (approximately 30% of populated results. In the data, April 2015 had a maximum number cells), the distribution stays similar with 98 percent of of cloud free days at just 16, and a mean and median at 9, light values being below 2, 95% below 1, 58% below 0.3 whereas May had a maximum number of cloud free days and 3% being negative. This distribution pattern is con - at 19 with a mean of 11 and median of 12. Overall, the sistent independently of which population threshold is year of 2015 was quite similar to the total sample, with chosen. When looking at population numbers above a max of 20, mean of 9 and median of 10. However, the 1000 (less than 0.5% of the total) 9% of the points still relatively low number of cloud free days for April could have nightlight values below 0.3 albeit less than 0.01% of potentially impact the analysis. Another factor that could points have values below 0, implying that VIIRS is poten- bias the results would be the monsoon season months tially only useful for urbanized areas with a high popula- from June through August, this will be commented upon tion density. in the discussion section. Figure  3 provides an overview of the distribution of cloud free days, with each pillar representing the per- General empirical strategy centage of observations for all months depending on the The approach chosen is the same as in Skoufias et  al. number of cloud free observations. From this, we find (2021), where the authors constructed damage indices that no months have more than 23 cloud free days and that were combined with population and nightlight data only 1% of months have 20 or more observations. At to estimate the impact different natural hazard types had on local nightlight emissions across 5 different South East Asian countries. The damage index is based on the USGS Shakemaps Values below 2 constitute approximately 99.3 percent of the total. This was and the damage modeling is done using two data sets, done due to the maximum VIIRS value being 154 and hence a density graph would be meaningless going from − 0.71 to 154. Also, any bin above 2 is very the objective earthquake intensity data expressed by PGA small and would not contribute to the graph. and building inventory data. The latter data is the USGS See next section for details on population layers. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 6 of 13 building inventory for earthquake assessment, which pro- were explored, without it affecting the results signifi - vides estimates of the proportions (based on total num- cantly in either direction. ber of buildings) of building types observed by country; Following the construction of the data sets, two ana- see Jaiswal and Wald (2008). For Nepal the building type lytical methods were utilized to potentially identify and information was compiled from a World Housing Ency- quantify the effect the earthquakes had on nightlight clopedia (WHE) survey. The WHE survey uses fraction values. The first was a simple mean analysis. It consisted of population who lives or works in buildings of differ - of two graphs; one with the mean of the nightlight val- ent types as their definition of how the building mass is ues of cells that were struck by an earthquake and one split up. The distribution of building types and mass are with the cells that were not affected. Furthermore, it was assumed to be homogenous within areas due to the lack broken down into two categories, one comparing light of more granular spatial data. This is a strong assump - value means across all cells in Nepal and one comparing tion, but without detailed local data it is not possible to means only across cells in regions that were affected by correct for. Preferably, one would want an extensive data- the earthquakes. base with building by building damage, such as in Wang The second set of analysis consists of fixed effects et al. (2016). regressions with additional controls for time and spa- From the building and intensity data, we derive dam- tial effects. To correct for potential heteroscedasticity, age curves based on the curves constructed by the Global Driscoll-Kraay standard errors are used (Driscoll and Earthquake Safety Initiative (GESI) project; see Interna- Kraay 1998). The regression equation is as follows: tional and Regional Development (2001). GESI divide buildings into 9 different categories, which are then rated L = β + β ED + θ + e i,t 0 n+1 i,q,t−n i i,t on different quality measures such as design, construc - n=0 tion and materials used. Depending on the rating and where L is the light level in cell i in month t and ED whether the building is in a rural or urban area, one out i,t i,q,t represents the damage curve value in the same cell and at of 16 damage curves (8 for urban and 8 for rural) are cho- the same time. Lags are allowed from month t to t − 12. sen as most likely to correlate to a building of the specific β is the intercept, θ are the cell fixed effects and e is the category and quality. In this article, it is assumed that all 0 i i,t error term. buildings in a category is of the same quality given that The regressions are lagged for the 12  months following we have no further distributional or locational data. the hazards to allow for any short- and mid-term effects to More formally, to derive a cell i specific earthquake materialize with the coefficients giving the effect that the damage index, ED, the following equation is applied: earthquakes have on the nightlights for the month when the pga k,q ED = DR q = 0, . . . ,7 i,q,t earthquake happens and the 12 subsequent months. The i,k,t regression is run both nationally and at a province level. where DR is the damage ratio according to the peak A flow chart providing an overview of the method - ground acceleration, pga , and the urban–rural qualifi - ology and data can be found in Fig.  4. Furthermore, an cation k of cell i , defined for a set of 8 different building overview of the different data sources with resolutions quality q categories. and temporal and spatial scope can be found in Table  1. The next step is to match the localized damage data Finally, Fig. 5 provides a graphical comparison of the res- with any intersecting VIIRS cell and assign the nightlight olution of the different data sets. value to the corresponding month. For light cells that intersected several damage cells an average value was Results used. As earthquake damage estimates were modeled Figure  6 show the results from the mean analysis com- based on centroids, only the centroid intersection was paring the means of nightlight cells in the entire country used. (top panel) and the two most affected regions—Central Finally, population data was aggregated up to the same and Western—in the bottom panel. The comparison is cell size as the VIIRS data and then matched to sum- between cells that were damaged by the earthquake and marize the total population in each nightlight cell. To cells that were not. In both cases, the nightlight values include a cell in the data set, the aggregated population follow the same pattern implying that there is no differ - had to be a minimum of 5. This is based on the average ence between the two subgroups of cells. household size in Nepal in 2010 being 5 (Libois and Som- Figure  7 shows the results for the second set of analy- ville 2018). Additionally, a cell was designated as either sis, the fixed effect regressions for the entire country urban or rural depending on the number of people liv- and for Central and Western. The coefficient values and ing in it. Our base case was 20, but numerous thresholds T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 7 of 13 Fig. 4 Flow chart for methodology Table 1 Overview of the different data sources Data source Type of data Spatial resolution Temporal res Temporal coverage Spatial coverage VIIRS Nightlights 15 arc-seconds Monthly 2012-present 75 N 65S Worldpop Population 3 or 30 arc-seconds Annually or less 2000-present Global b c ShakeMaps Earthquake 0.0167° One per EQ 1996-present Global Heavily dependent on time of year, i.e. during summer the hemispheres will have less nighttime coverage EMost are 0.0167°, but for Nepal it is 0.0333 Earlier earthquakes are also covered, but the development did not start until 1996 significance are similar across both sets, with an initial 4–6 months, before it tapers off. Months 6, 7, 10 and 11 negative shock during the month of the Gorkha earth- are insignificant in the province analysis. In month 9, quake followed by a positive effect for months 1 through the country results show a negative and significant coef - 8. The positive effect is most pronounced for the first ficient value, whereas Central and Western stay positive. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 8 of 13 Fig. 5 Comparison of the spatial resolution of the ShakeMap, the VIIRS nightlight data and the WorldPop data Afterwards, months 10 and 11 show a negative trend for are populated and whether any populated area is consid- both sets, while month 12 exhibits a strongly significant ered to be rural or urban. The former threshold was set and positive jump. at 5, which is in line with an average household size in Nepal. Whereas for the latter numerous thresholds were Discussion explored, ranging from 5 (all urban) to 100. However, The aim of the paper was to get highly localized impact the choice had little impact on the regression or mean estimates by using VIIRS data combined with high-reso- results. When running the regressions for urban areas lution population layers and spatially disaggregated dam- only, i.e. only for cells that were deemed to be urban, the age data. The nightlight data has shown in prior studies, coefficient values change, but the significance and direc - some of which use the Gorkha earthquake as their case tion of the signs are the same as for the primary analysis. study, that one can identify local effects from natural dis - In addition, we visually inspected the data to check for asters (Zhao et al. 2018; Gao et al. 2020). A key assump- any differences depending on which year of population tion is that damage can be linked to economic output. was chosen. We found no discernible difference in terms This is not necessarily true for cases where the damages of populated areas and when comparing the population are structural or human in nature. However, it is assumed per pixel between years, the differences were very small that our modelled structural damages will affect the and indicative of a simple application of growth trend nightlight values, which has shown a close link with eco- that has been interpolated across the cells. nomic output, see for example Henderson et al. (2012) or Secondly, when modeling the damage, the building Michalopoulos and Papaioannou (2014). quality assumption will have an impact on the sustained Our analysis rests on several assumptions, which can differ based on the area and time span being analyzed. Firstly, the population data is used to define which areas We also considered which data set to use as our population basis. We did use 2015, and when comparing the data with 2011, the differences are very small and implies that a simple growth trend has been used to adjust the pop- ulation numbers. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 9 of 13 Fig. 6 Mean of nightlight cell values for Nepal and the Central and Western districts split between cells that were affected by earthquakes and those that were not damage as well as the final impact estimates on night - Fourthly, the nightlight values depend on two lights. When the analysis was performed, we did find a assumptions. The first one regards missing nightlight non-negligible effect on the estimates but lacking any values, usually missing due to lack of cloud free obser- meaningful information a mean quality assumption of vations. When no value was present, it was linearly 4 was used. interpolated from the last light value before the missing Related to the first two points, assuming that the observation and the first light value after the observa - building types and quality are homogenously dis- tion. Another option that was explored, was to use the tributed across Nepal is not likely to be true given prior year’s value, but this did not affect the final esti - the diverse topography and climate of Nepal. How- mates. The second assumption was related to the treat - ever, lacking more detailed data on the distribution it ment of VIIRS values close to or below 0. When the is neigh on impossible to correct for. The rural/urban VIIRS images are processed a dark offset is deducted distinction might help alleviate some of the issues and from the raw day/night value. This offset is sometimes given how the results did not change with differing severely impacted by airglow leading to negative light thresholds it gives some confidence in the results. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 10 of 13 Fig. 7 Regression results for Nepal and the Central and Western districts values. To account for this in our analysis, several should—detrend and clean the data to get more con- options were explored. Firstly, the regressions were run sistent and less volatile nightlight values. with any light value below 0.1 set to 0. Then absolute Finally, the nightlight data shows that during monsoon values were used with any absolute value below thresh- season, the number of cloud free days is on average very olds of 0.1, 0.3 and 0.5 set to 0. Overall, the threshold low. This is partially corrected for by interpolating the choice only had a minor quantitative impact on the nightlight values for the month before and after a month coefficients. Recently, the papers from Gao et al. (2020) with no cloud free days and also tested for by using the and Zhao et al. (2020) show that one can—and probably previous year’s nightlight value. This is not optimal and T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 11 of 13 could impact the estimates depending on how the lack sudden and significant coefficient increase in month 12 of observations is distributed, but we are unaware of any differs from the expected pattern, which is a return to other or better way to correct for the lack of observations. non-significance and coefficient values close to 0, i.e. the As for the results, the mean comparisons of the night- earthquake should not impact nightlight values for a very light values in Fig.  5 yielded no difference between long period of time. When exploring the underlying data affected cells and cells that were not damaged by the and general information about Nepal, we found no spa- earthquake. As can be seen, the trends are the same tial or seasonal pattern in the data or information about for both sets of cells and there are several peaks and new power plants or other events that could have caused throughs throughout the period, some of which may be the spike, leading us to believe that it might be due to the seasonal. The nightlight values exhibit a low during the general volatility of the data. Increasing the lag period is winter months and a high in fall and late fall. Some of this also problematic as it will decrease the sample size and effect could be due to measurement differences and cloud is likely to capture other events that can potentially bias cover, but it might also be caused by differences in use of the results. Once the VIIRS data is available for longer energy and covering up lights during the winter months. periods both pre and post events it will be more feasible The lack of difference between the two means could be to do longer lags and potentially identify more long-term due to aid and relief efforts targeting all areas but given effects, although it can always be argued that any long- the inherent volatility and noise in the VIIRS values it term impact—in particular when one does month by is more likely that any quantitative effects on the local month analysis—are unlikely to be correctly quantified economy has not been strong enough to be captured by due to the many factors that impact light and the meas- the VIIRS data when performing such a simple analy- urements. However, this can be an interesting avenue for sis. To further test this we performed a pre-post analy- future research. sis where we compared the change in nightlight value for Finally, Fig.  8 depicts a comparison between our mod- the month before the earthquake (March) with the value elled damage, gridded out via a kriging algorithm, and for May and June. In this instance, the cells affected by the Copernicus Emergency Management Service (EMS) the earthquake experienced a slightly smaller drop in grading map for Kathmandu. The Copernicus map uses absolute values (− 0.049 vs − 0.057). However, due to satellite images from DigitalGlobe and Pleiades CNES at the higher mean absolute value of the affected cells, the 0.5 m resolution to visually compare ex-post and ex-ante percentage change was − 16% versus − 22%. When per- images and identify changes and potential damages. The forming a t-test between the two sample means, we find focus in the Copernicus maps is at a building level, while the difference to be significant at a 5% level, indicating our modelling focuses on a much more aggregate level that the affected cells experienced a smaller drop than making the two data sets difficult to compare. In addition, the non-affected ones. One possible explanation for the Copernicus maps are only produced for 8 areas that this could be the influx of aid or an inherent difference cover approximately 35 square kilometres each. With this between the two groups. To explore this further, we per- in mind, Fig. 7 is provided as a visual comparison of how formed a mean comparison between the nightlight val- our model performed versus the Copernicus map. Look- ues for both affected and non-affected cells for May and ing at the main area of damage as identified in the Coper - June in 2012–2014 versus the mean in 2015. Once again, nicus map, we find that the outlined rectangular cell we find that the growth for the affected cells is bigger, i.e. with the most damaged buildings map is seemingly more the nightlight value for affected cells is relatively higher damaged than the neighbouring cells. However, looking than for the non-affected cells. This could indicate that at the overall map, the marked cell does not stand out. the two groups of cells are on different trajectories, that This is most likely due to a combination of much lower potentially the affected cells are in more urban areas. This resolution data as well as the kriging algorithm, which could also explain why the fixed effects regressions iden - leads to a smoothing of the damages. tify a significant difference. The regression results did find that a statistical and sig - nificant effect occurred in cells that were damaged by the Conclusion earthquakes. One potential explanation for the nightlight Having used the VIIRS data as an economic proxy one value impact could be that in the month of the quake, finds that the earthquakes in Nepal in 2015 did impact there is a nightlight loss due to the damages. Then, in the local economic activity if one uses fixed effects the following months, there is an increased activity from regressions. Overall, there is an initial decrease when aid, repairs and rebuilding. Finally, one sees a decline fol- the shock occurs, followed by an increase in activity lowed by a negative effect that could signify the end of once aid and relief efforts have started. This is consist - the rebuilding efforts and the aid influx. However, the ent with what one would expect, however, with the Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 12 of 13 Fig. 8 Comparison for Kathmandu between Copernicus grading map and modelled damage map amount of noise and cloud cover in the underlying Declarations VIIRS data some caution is in order. Competing interests Potential future research within the area could focus The authors declare no competing interests. on other large-scale earthquakes or natural hazard Author details types as the methodology is easily scaled and depends Department of Economics, University of Bern, Schanzeneckstrasse 1, on few inputs. Another option is to explore other algo- 2 3001 Bern, Switzerland. The World Bank Group, Poverty and Equity Global rithms or secondary data to potentially achieve more Practice, Washington, DC, USA. consistent and comparable nightlight measurements. Received: 9 September 2021 Accepted: 26 December 2021 Acknowledgements This paper was funded by the global knowledge program of the World Bank’s Poverty and Equity Global Practice. References Authors’ contributions Agency, Federal Emergency Management (2006) HAZUS-MH MR2 Technical TT: analysis, data gathering, idea, writing, editing. ES: editing, writing, idea, Manual. Washington, DC funding. ES: editing, writing, idea. All authors read and approved the final Baragwanath K, Goldblatt R, Hanson G, Khandelwal AK (2019) Detecting urban manuscript. markets with satellite imagery: an application to India. J Urban Econ. https:// doi. org/ 10. 1016/j. jue. 2019. 05. 004 Availability of data materials Chen Xi, Nordhaus W (2015) A test of the new VIIRS lights data set: population The data used in this study were derived from the following resources and economic output in Africa. Remote Sens 7:4937–4947. https:// doi. available in the public domain: These data were derived from the following org/ 10. 3390/ rs704 04937 resources available in the public domain: Population data Worldpop: https:// Chen Xi, Nordhaus WD (2019) VIIRS nighttime lights in the estimation of cross- www. world pop. org/ proje ct/ categ ories? id=3. Nightlights data VIIRS: https:// sectional and time-series GDP. Remote Sens. https:// doi. org/ 10. 3390/ eogda ta. mines. edu/ downl oad_ dnb_ compo sites. html. Earthquake data USGS rs110 91057 Shakemaps: https:// earth quake. usgs. gov/ earth quakes/ search/. Building inven- tory data: https:// pubs. er. usgs. gov/ publi cation/ ofr20 081160. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 13 of 13 Corbane C, Kemper T, Freire S, Louvrier C, Pesaresi M (2016) Monitoring the Worldpop (2017) Nepal 100m population, version 2. University of Southamp- Syrian humanitarian crisis with the JRC’s global human settlement layer ton. https:// doi. org/ 10. 5258/ SOTON/ WP005 31 and night-time satellite data. Publications Office of the European Union, Xu C, Tian Y, Zhou B, Ran H, Lyu G (2017) Landslide damage along Araniko Luxembourg, pp 1–14 highway and Pasang Lhamu highway and regional assessment of land- De Groeve T, Annunziato A, Gadenz S, Vernaccini L, Erberik A, Yilmaz T (2008) slide hazard related to the Gorkha, Nepal earthquake of 25 April 2015. Real-time impact estimation of large earthquakes using USGS Shake- Geoenviron Disasters 4:14. https:// doi. org/ 10. 1186/ s40677- 017- 0078-9 maps. In: Proceedings of IDRC2008, Davos, Switzerland Zhao N, Hsu F-C, Cao G, Samson EL (2017) Improving accuracy of economic Driscoll J, Kraay A (1998) Consistent covariance matrix estimation with spatially estimations with VIIRS DNB image products. Int J Remote Sens 38:5899– dependent panel data. Rev Econ Stat 80:549–560 5918. https:// doi. org/ 10. 1080/ 01431 161. 2017. 13310 60 Elliott RJR, Strobl E, Sun P (2015) The local impact of typhoons on economic Zhao X, Bailang Yu, Liu Y, Yao S, Lian T, Chen L, Yang C, Chen Z, Jianping Wu activity in China: a view from outer space. J Urban Econ 88:50–66. https:// (2018) NPP-VIIRS DNB daily data in natural disaster assessment: evidence doi. org/ 10. 1016/j. jue. 2015. 05. 001 from selected case studies. Remote Sens 10:1526. https:// doi. org/ 10. Gao S, Chen Y, Liang L, Gong A (2020) Post-earthquake night-time light piece-3390/ rs101 01526 wise (PNLP) pattern based on NPP/VIIRS night-time light data: a case Zhao N, Liu Y, Hsu F-C, Samson EL, Letu H, Liang D, Cao G (2020) Time series study of the 2015 Nepal earthquake. Remote Sens 12:2009. https:// doi. analysis of VIIRS-DNB nighttime lights imagery for change detection in org/ 10. 3390/ rs121 22009 urban areas: a case study of devastation in Puerto Rico from hurricanes Gibson J, Olivia S, Boe-Gibson G (2019) A test of DMPS and VIIRS night lights Irma and Maria. Appl Geogr 120:102222. https:// doi. org/ 10. 1016/j. data for estimating GDP and spatial inequality for rural and urban areas. apgeog. 2020. 102222 Working papers in economics, University of Waikato. https:// ideas. repec. org/p/ wai/ econwp/ 19- 11. html Publisher’s Note Gillespie TW, Frankenberg E, Chum KF, Thomas D (2014) Night-time lights time Springer Nature remains neutral with regard to jurisdictional claims in pub- series of tsunami damage, recovery, and economic metrics in Sumatra, lished maps and institutional affiliations. Indonesia. Remote Sens Lett 5:286–294. https:// doi. org/ 10. 1080/ 21507 04X. 2014. 900205 Henderson JV, Storeygard A, Weil DN (2012) Measuring economic growth from outer space. Am Econ Rev 102:994–1028. https:// doi. org/ 10. 1257/ aer. 102.2. 994 Hodler R, Raschky PA (2014) Regional favoritism. Q J Econ 129:995–1033. https:// doi. org/ 10. 1093/ qje/ qju004 INFORM (2019) 07. http:// www. inform- index. org/ Porta ls/0/ Inform/ 2019/ count ry_ profi les/ NPL. pdf . Accessed July 2019 International, GeoHazards, and United Nations Centre Regional Development (2001) Final report: global earthquake safety initiative (GESI) pilot project. Report, GHI, 86 p Jaiswal KS, Wald DJ (2008) Creating a global building inventory for earthquake loss assessment and risk management: US Geological Survey Open-File Report 2008-1160. Technical report, USGS, 103 p Jean N, Burke M, Xie M, Davis WM, Lobell DB, Ermon S (2016) Combining satellite imagery and machine learning to predict poverty. Science 353:790–794. https:// doi. org/ 10. 1126/ scien ce. aaf78 94 Klomp J (2016) Economic development and natural disasters: a satellite data analysis. Glob Environ Chang 36:67–88. https:// doi. org/ 10. 1016/j. gloen vcha. 2015. 11. 001 Libois F, Somville V (2018) Fertility, household size and poverty in Nepal. World Dev 103:311–322. https:// doi. org/ 10. 1016/j. world dev. 2017. 11. 005 Michalopoulos S, Papaioannou E (2014) National institutions and subnational development in Africa. Q J Econ 129:151–213. https:// doi. org/ 10. 1093/ qje/ qjt029 Nguyen CN, Noy I (2019) Measuring the impact of insurance on urban earth- quake recovery using nightlights. J Econ Geogr 20:857–877. https:// doi. org/ 10. 1093/ jeg/ lbz033 Sahoo S, Gupta PK, Srivastav SK (2020) Inter-calibration of DMSP-OLS and SNPP-VIIRS-DNB annual nighttime light composites using machine learn- ing. Gisci Remote Sens 57:1144–1165. https:// doi. org/ 10. 1080/ 15481 603. 2020. 18483 23 Skoufias E, Strobl E, Tveit T (2021) Can we rely on VIIRS nightlights to estimate the short-term impacts of natural hazards? Evidence from five South East Asian countries. Geomat Nat Haz Risk 12:381–404. https:// doi. org/ 10. 1080/ 19475 705. 2021. 18799 43 Skoufias E, Strobl E, Tveit T (2017) Natural disaster damage indices based on remotely sensed data: an application to Indonesia (English). World Bank policy research working paper no. 8188 Uprety S, Cao C, Gu Y, Shao X, Blonski S, Zhang B (2019) Calibration improve- ments in S-NPP VIIRS DNB sensor data record using version 2 reprocess- ing. IEEE Trans Geosci Remote Sens 57:9602–9611. https:// doi. org/ 10. 1109/ TGRS. 2019. 29279 42 Wang F, Miyajima M, Dahal R, Timilsina M, Li T, Fujiu M, Kuwada Y, Zhao Q (2016) Eec ff ts of topographic and geological features on building damage caused by 2015.4.25 Mw7.8 Gorkha earthquake in Nepal: a pre- liminary investigation report. Geoenviron Disasters 3:7. https:// doi. org/ 10. 1186/ s40677- 016- 0040-2 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Geoenvironmental Disasters Springer Journals

Using VIIRS nightlights to estimate the impact of the 2015 Nepal earthquakes

Loading next page...
 
/lp/springer-journals/using-viirs-nightlights-to-estimate-the-impact-of-the-2015-nepal-Ft9fwqTHWu
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2022
eISSN
2197-8670
DOI
10.1186/s40677-021-00204-z
Publisher site
See Article on Publisher Site

Abstract

We use Visible Infrared Imaging Radiometer Suite ( VIIRS) nightlight data to model the impact of the 2015 Nepal earth- quakes. More specifically, the data—showing nightlight emissions—are used to examine the extent to which there is a difference in nightlight intensity between cells damaged in the earthquake versus undamaged cells based on (1) mean comparisons; and (2) fixed effect regression models akin to the double difference method. The analysis is car - ried out for the entire country as well as smaller regions in and around the Central area and Kathmandu, which were the hardest hit areas. Overall, the regressions find a significant and negative effect from the initial shock, followed by a positive net effect from aid and relief efforts, which is consistent with what one would expect to find. However, the mean analysis results are inconclusive and there is substantial noise in the nightlight measurements due to how the values are produced and persistent cloud cover over Nepal. Keywords: Earthquakes, Night-time lights (NTL), VIIRS, Remote sensing, Econometrics by natural hazards, see for example Elliott et  al. (2015), Introduction Klomp (2016), Skoufias et  al. (2017), Nguyen and Noy The Gorkha earthquake that struck Nepal on 25 April (2019) or Gao et al. (2020). 2015 is one of the biggest natural hazards of the last dec- The aim of this paper is to build upon this literature by ade. The widespread damages totalled USD 10 billion in showcasing how high-resolution nightlight and popula- monetary damages, leaving 8857 dead, 22,304 injured tion data can be used to quantify highly localized changes and tens of thousands homeless. In addition to the main in economic activity as proxied by monthly nightlight earthquake, several significant aftershocks also occurred, data. By using population data as a secondary layer, we causing additional damages and deaths. Following any seek to avoid several known issues with the nightlight natural hazard of this scale, it is important to quickly and value measurements, such as blooming, airglow contami- accurately identify the areas which are most in need of nation and rural area values being more poorly correlated aid and where one can expect the highest damages, both with economic activity. In addition, while existing articles in terms of casualties and monetary losses. The two are have focused on identifying affected areas immediately often correlated as the level of human activity usually after impact, this paper uses mean analysis and fixed coincides with economic output. Often, the identification effects regressions to quantify the initial impact and tem - of damages are done by on the ground observations or by poral effect over the 12  months ex-post. This is similar manually examining images taken from air or from space, to Gao et  al. (2020), but whereas they examined purely but over the last two decades a growing body of litera- based on nightlight change, we use objectively measured ture has focused on using remotely sensed data and auto- earthquake data to identify nightlight cells that were mated algorithms to analyze and detect changes caused impacted by the earthquake and compared these with cells that experienced no damage. *Correspondence: Thomas.tveit@vwi.unibe.ch The methodology of this paper is the same as used in Department of Economics, University of Bern, Schanzeneckstrasse 1, 3001 Bern, Switzerland Skoufias et al. (2021), where the authors found that VIIRS Full list of author information is available at the end of the article © The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 2 of 13 nightlight data provided limited results when analyzing performed on the combined data. First, a simple mean larger countries and numerous natural hazards. In this comparison is performed where we compare the night- paper, we use the same method, but across a smaller area light values between affected and non-affected cells for and on a more damaging hazard, making it arguably an the 12 months before and after the earthquake. Secondly, ideal setting for performing such an analysis. In short, we run a fixed effect regression model to identify and we use contour maps from the United States Geological quantifyimpact over the first 12  months ex-post. Both Survey (USGS) to identify damaged areas and model the analysis are performed on a national nightlight set as local damage at a cell level from vulnerability curves pro- well as sets on smaller regions that were more severely vided by International and Regional Development (2001). impacted by the earthquakes. The damage data is then geomapped alongside the VIIRS The mean analysis results are not significant, poten - and population data, which act as a proxy for the eco- tially due to the general volatility and measurement nomic activity and hence the asset exposure. Finally, errors in the data. However, it could also be caused by an the combined data is used to identify and quantify what influx of aid and recovery to all areas of Nepal and not impact the damage has on nightlight values in two dif- just the ones experiencing the most damage, i.e. if aid is ferent analysis. The first is a mean comparison between not well targeted one would experience the same pattern damaged and non-damaged nightlight cells and the other across all cells. When performing a pre-post analysis of is a fixed effect regression. the mean results, we do find that there is a significant dif - The nightlights data used in literature has primarily ference in the trajectories of the two groups. To further been from two sources: the Defense Meteorological Sat- support this, the fixed effects regressions did yield sig - ellite Program (DMSP) and the VIIRS. The former is dis - nificant results, with an initial negative and significant continued and consists of annual data from 1992 through shock in the month of the disaster, followed by a statisti- 2013 with the data released being grid cells at 30 arc-sec- cally significant increase in light for 4–6  months, before ond resolution (approximately 1 km by 1 km at equator) tapering off and becoming insignificant. The first nega - where each cell has a normalized digital number (DN) tive shock followed by a positive effect is consistent with value from 0 to 63. The VIIRS data is more recent (April what one could expect from a large-scale disaster, where 2012 and onwards), has improved spatial and temporal at first infrastructure is damaged and necessary aid is detail with monthly releases at a grid cell resolution of not present, followed by an influx of aid and relief efforts 15 arc-seconds and the DN values have no upper bound. over the coming months. Following this phase, one would However, despite the advantages of the VIIRS data, it expect a decrease in nighlight values as the initial relief does have some limitations. Firstly, it is much more vola- efforts are discontinued. Additionally, after 12  months tile than the DMSP data and due to stray light correction one can potentially expect the VIIRS values to reflect a it can contain negative light values. Furthermore, some “new” normal and any significant effects found after that months have no measurements because of cloud cover. are unlikely to be directly related to the earthquake.. Despite these shortcomings, the VIIRS data values does However, the regression results include an unexpected show some correlation with local GDP (Zhao et al. 2017) and sudden positive spike 12  months after the earth- and we will use it as a proxy for economic activity in this quakes. The cause for this is unknown and the most likely article. explanation is that it is caused by noise in the measure- To improve upon the VIIRS data and attempt to ments. The regressions were run with several different decrease the volatility, we will use a settlement layer as specifications as robustness checks, but the main results a secondary layer to better identify areas with human remained unchanged. activity. Using secondary data sources to increase the The remainder of the paper starts with an overview of likelihood of correctly identifying human settlements Nepal, the earthquakes and the base data, followed by a have previously been used for poverty (Jean et  al. 2016) general methodology overview, the analysis results and and urban extent (Baragwanath et  al. 2019). The data finally a conclusion. used is from Worldpop (2017), which utilizes satellite images and census data to construct a high-resolution layer of human settlement patterns. The inclusion of the Materials and methods additional data is used to exclude non-populated cells Nepal and the 2015 earthquakes and to better distinguish rural and urban areas. Nepal is a mountainous country bordering India to the Finally, to identify and model local damage, we con- south, west and east and Tibet (China) to the north. It struct a damage index based on ShakeMaps from the is a small country at roughly 147,000 square kilometers USGS. The damage estimates are then combined with and has a population of approximately 29 million people. the economic activity proxy and two types of analysis are Its nominal GDP in 2019 is USD30.6 billion. On a per T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 3 of 13 Fig. 1 Nepal and the Gorkha earthquake. The underlying blue colour shows Nepal, whereas the green to red outlines the extent and intensity of the earthquake, with red being the most impacted areas. Source: USGS Shakemap and ArcGIS Basemap capita purchasing power adjusted basis, Nepal ranks at 8857 mortalities and 22,304 injured. In addition, tens the 162nd position globally per the World Bank. of thousands were made homeless. Some damages were In addition, Nepal is prone to disasters due to a com- also due to landslides caused by the earthquake as seen bination of the vulnerability of the building environment in Xu et  al. (2017). In this paper, we will treat all dam- and the population, its geographical location, topogra- ages as earthquake damages as the effect on nightlights phy and climate. Nepal is frequently affected by different will be the same and the cause of the damage is the earth- types of natural hazards such as landslides, floods and quake. Furthermore, to disentangle the two at a regional avalanches, stemming from the topography and the cli- or national scale would require significant manual labor. mate. Due to lack of irrigation, drought and fire are other In the wake of the Gorkha earthquake there were sev- frequent natural hazards. Finally, Nepal is situated along- eral aftershocks, with the largest one being a magnitude side a major east–west tectonic boundary plate where 7.3  MW occurring on 12 May 2015 to the east of Kath- earthquakes are frequent. INFORM earthquake exposure mandu in the districts of Dolakha and Sindhupalchowk. rank them at an 9.9 out of 10 (INFORM 2019). There were 153 dead and 3275 injured following this In 2015, Nepal suffered two major earthquakes. The earthquake. first—and largest—earthquake struck on 25 April 2015 The data used to identify the damaged areas and model and is often referred to as the Gorkha Earthquake. Fig- the damage are from contour maps generated by seis- ure 1 shows the magnitude and size of the 7.8 MW earth- mological ground stations (International and Regional quake relative to Nepal and the epicenter in the Gorkha Development 2001; Agency 2006; De Groeve et al. 2008). District. According to the UN office for Disaster Risk These contour maps—ShakeMaps from USGS—are used Reduction the total damages added up to USD 10 bil- as a basis for localized impact and they provide several lion—approximately one third of the total GDP—with key parameters such as peak ground acceleration (PGA), Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 4 of 13 peak ground velocity (PGV) and modified Mercalli inten - the Nepal earthquake. Recently, Sahoo et al. (2020) have sity (MMI). The maps use data from seismic stations and utilized machine learning methods to intercalibrate combine this with information on ground conditions and between the two data sets and construct a longer annual earthquake depth to interpolate and construct a grid of data set. However, an increase in the length of the tempo- points spaced 0.0167 degrees apart. ral dimension at the cost of temporal detail is not benefi - cial in this study. Additionally, data that is prior to 2012 is Population layer unlikely to impact on the 2016 estimated effects. To properly capture economic activity through night- The use of VIIRS in economic analysis was done early lights it is important to correctly identify areas with peo- in its lifetime by Chen and Nordhaus (2015), where they ple, hence avoiding stray lights or other sources of light found promising results for VIIRS as an economic and that are not connected to human activity. One method is population indicator, also when compared to the DMSP to use a human settlement layer as a secondary data set, product. However, the same authors found that VIIRS such as in Corbane et  al. (2016). One assumes that the was not useful in growth predictions and that the asso- settlement layer provides a better static map of the spa- ciation between economic activity and light values were tial distribution of people and then the nightlight data dependent upon the chosen area size, with state level provide a temporal and spatial overview of how the eco- analysis performing worse than metropolitan area level nomic activity is distributed within the populace. analysis (Chen and Nordhaus 2019). They hypothesized In this paper, we use the Worldpop datasets to identify that this was in part due to the economic activity type settlement areas in Nepal (Worldpop 2017). These data - in metropolitan areas and that they are more likely to be sets have a spatial gridcell resolution of approximately related to electric light. In a more recent working paper, 100  m at the equator and estimate the number of per- Gibson et al. (2019) also find that VIIRS performs poorly sons per square for the years 2010, 2011 (based on cen- in rural, low-density areas in Indonesia, while it performs sus data), 2015 and 2020. This article uses only the 2015 well and much better than DMSP data in urban areas. estimates as these are from the same year as the earth- The VIIRS products have also seen use in the natural dis - quakes. Worldpop constructed the population files by aster literature with Zhao et  al. (2018) using the under- using national totals and then adjust these to match UN lying NPP-VIIRS DNB Daily Data to analyze selected population estimates. The adjustment is done by a Ran - natural disasters, including the Gorkha earthquake. They dom Forest model, which generates a weighted popula- found that the images could be used in damage detection tion density layer and then this layer is used as a basis to and for identifying power outages, but that the analysis place the population according to the assumed real geo- was limited by cloud coverage. This article will not use graphical distribution. the daily data because of computational limitations when In addition to identifying economic activity, the settle- focusing on much longer temporal and spatial effects. ment layers will be used to designate a cell as rural and The VIIRS data provide two variables: the average urban based upon population density. monthly light radiance from DNB and the number of cloud free days. Some months will have no cloud free VIIRS days, meaning that no radiance value is provided. This is To find and identify areas with economic activity and accounted for by calculating an interpolated value from their exposure to natural hazards, satellite images of the values of the month prior and after. In addition, a nighttime lights will be used as a proxy. Since natural known problem for low light value cells is that one often hazards are highly localized, one would prefer high-res- finds negative light values due to airglow contamination olution spatially disaggregated economic data, although (Uprety et  al. 2019). Optimally, one would want to cor- nightlights have been used successfully to model local rect for this and identify the real underlying value and economic activity (Henderson et  al. 2012; Gillespie decrease the volatility exhibited. Recently, two published et  al. 2014; Hodler and Raschky 2014; Michalopoulos articles (one about the Gorkha earthquake) have explored and Papaioannou 2014). These articles use the annual methodologies for correcting the measurements through DMSP data with 30-arc second grids, whereas more trend decompositions (Gao et al. 2020; Zhao et al. 2020). recent papers have also utilized the VIIRS Day/Night However, lacking any established methodology, we Band (DNB) provided by The Earth Observations Group experimented by using simple threshold testing to iden- (EOG) at NOAA/NCEI, see for example Chen and Nord- tify how it would affect the analysis. haus (2015), Zhao et al. (2017, 2018) or (2020). The latter data set provides monthly data from April 2012 till pre- Using the value of the month of the previous year was also explored, but the sent, whereas DMSP contains nightlight data for the years overall value mean change was less than 0.01 and there were only 1 cell that 1992 through 2013, making DMSP unusable for studying was affected by the earthquake that had no radiance value. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 5 of 13 Fig. 2 VIIRS value distribution for VIIRS values below 2 Fig. 3 Days with no cloud cover in a month across all months. The X-axis shows how many days per month that were cloud free, i.e. that had a light value measurement. The Y-axis shows the distribution of number of cloud free days per month. Adding up all columns equal 1 Figure 2 show the distribution of VIIRS values below 2 for all populated areas and the number of cloud free days per month. It identifies significant clustering between the values 0 and 0.5, containing more than 85 percent the same time, almost 10% of the monthly observations of the total observations. Furthermore, more than 5% of had 0 or 1 cloud free days, with 4% of the sample hav- the total have a negative value. Analyzing the range from ing no observations. The months with no cloud free days negative to 0.3, one finds that 75% of all points fall within were included either through interpolation or by using this range, implying that a large proportion of our obser- the prior year’s value. The median and mean are 10, and vations have negative or very small values. These are only more than 12% of months had one or less cloud free days. points that have been identified as being populated, Hence, the monthly nightlight values are calculated based meaning that they should be relatively free of disturbance on a rather small subsample of days per month. How- from non-human sources. ever, for a natural hazard which is not correlated with the Even when limiting the observations to cells with pop- weather, such as earthquakes, this is unlikely to bias the ulation in excess of 50 (approximately 30% of populated results. In the data, April 2015 had a maximum number cells), the distribution stays similar with 98 percent of of cloud free days at just 16, and a mean and median at 9, light values being below 2, 95% below 1, 58% below 0.3 whereas May had a maximum number of cloud free days and 3% being negative. This distribution pattern is con - at 19 with a mean of 11 and median of 12. Overall, the sistent independently of which population threshold is year of 2015 was quite similar to the total sample, with chosen. When looking at population numbers above a max of 20, mean of 9 and median of 10. However, the 1000 (less than 0.5% of the total) 9% of the points still relatively low number of cloud free days for April could have nightlight values below 0.3 albeit less than 0.01% of potentially impact the analysis. Another factor that could points have values below 0, implying that VIIRS is poten- bias the results would be the monsoon season months tially only useful for urbanized areas with a high popula- from June through August, this will be commented upon tion density. in the discussion section. Figure  3 provides an overview of the distribution of cloud free days, with each pillar representing the per- General empirical strategy centage of observations for all months depending on the The approach chosen is the same as in Skoufias et  al. number of cloud free observations. From this, we find (2021), where the authors constructed damage indices that no months have more than 23 cloud free days and that were combined with population and nightlight data only 1% of months have 20 or more observations. At to estimate the impact different natural hazard types had on local nightlight emissions across 5 different South East Asian countries. The damage index is based on the USGS Shakemaps Values below 2 constitute approximately 99.3 percent of the total. This was and the damage modeling is done using two data sets, done due to the maximum VIIRS value being 154 and hence a density graph would be meaningless going from − 0.71 to 154. Also, any bin above 2 is very the objective earthquake intensity data expressed by PGA small and would not contribute to the graph. and building inventory data. The latter data is the USGS See next section for details on population layers. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 6 of 13 building inventory for earthquake assessment, which pro- were explored, without it affecting the results signifi - vides estimates of the proportions (based on total num- cantly in either direction. ber of buildings) of building types observed by country; Following the construction of the data sets, two ana- see Jaiswal and Wald (2008). For Nepal the building type lytical methods were utilized to potentially identify and information was compiled from a World Housing Ency- quantify the effect the earthquakes had on nightlight clopedia (WHE) survey. The WHE survey uses fraction values. The first was a simple mean analysis. It consisted of population who lives or works in buildings of differ - of two graphs; one with the mean of the nightlight val- ent types as their definition of how the building mass is ues of cells that were struck by an earthquake and one split up. The distribution of building types and mass are with the cells that were not affected. Furthermore, it was assumed to be homogenous within areas due to the lack broken down into two categories, one comparing light of more granular spatial data. This is a strong assump - value means across all cells in Nepal and one comparing tion, but without detailed local data it is not possible to means only across cells in regions that were affected by correct for. Preferably, one would want an extensive data- the earthquakes. base with building by building damage, such as in Wang The second set of analysis consists of fixed effects et al. (2016). regressions with additional controls for time and spa- From the building and intensity data, we derive dam- tial effects. To correct for potential heteroscedasticity, age curves based on the curves constructed by the Global Driscoll-Kraay standard errors are used (Driscoll and Earthquake Safety Initiative (GESI) project; see Interna- Kraay 1998). The regression equation is as follows: tional and Regional Development (2001). GESI divide buildings into 9 different categories, which are then rated L = β + β ED + θ + e i,t 0 n+1 i,q,t−n i i,t on different quality measures such as design, construc - n=0 tion and materials used. Depending on the rating and where L is the light level in cell i in month t and ED whether the building is in a rural or urban area, one out i,t i,q,t represents the damage curve value in the same cell and at of 16 damage curves (8 for urban and 8 for rural) are cho- the same time. Lags are allowed from month t to t − 12. sen as most likely to correlate to a building of the specific β is the intercept, θ are the cell fixed effects and e is the category and quality. In this article, it is assumed that all 0 i i,t error term. buildings in a category is of the same quality given that The regressions are lagged for the 12  months following we have no further distributional or locational data. the hazards to allow for any short- and mid-term effects to More formally, to derive a cell i specific earthquake materialize with the coefficients giving the effect that the damage index, ED, the following equation is applied: earthquakes have on the nightlights for the month when the pga k,q ED = DR q = 0, . . . ,7 i,q,t earthquake happens and the 12 subsequent months. The i,k,t regression is run both nationally and at a province level. where DR is the damage ratio according to the peak A flow chart providing an overview of the method - ground acceleration, pga , and the urban–rural qualifi - ology and data can be found in Fig.  4. Furthermore, an cation k of cell i , defined for a set of 8 different building overview of the different data sources with resolutions quality q categories. and temporal and spatial scope can be found in Table  1. The next step is to match the localized damage data Finally, Fig. 5 provides a graphical comparison of the res- with any intersecting VIIRS cell and assign the nightlight olution of the different data sets. value to the corresponding month. For light cells that intersected several damage cells an average value was Results used. As earthquake damage estimates were modeled Figure  6 show the results from the mean analysis com- based on centroids, only the centroid intersection was paring the means of nightlight cells in the entire country used. (top panel) and the two most affected regions—Central Finally, population data was aggregated up to the same and Western—in the bottom panel. The comparison is cell size as the VIIRS data and then matched to sum- between cells that were damaged by the earthquake and marize the total population in each nightlight cell. To cells that were not. In both cases, the nightlight values include a cell in the data set, the aggregated population follow the same pattern implying that there is no differ - had to be a minimum of 5. This is based on the average ence between the two subgroups of cells. household size in Nepal in 2010 being 5 (Libois and Som- Figure  7 shows the results for the second set of analy- ville 2018). Additionally, a cell was designated as either sis, the fixed effect regressions for the entire country urban or rural depending on the number of people liv- and for Central and Western. The coefficient values and ing in it. Our base case was 20, but numerous thresholds T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 7 of 13 Fig. 4 Flow chart for methodology Table 1 Overview of the different data sources Data source Type of data Spatial resolution Temporal res Temporal coverage Spatial coverage VIIRS Nightlights 15 arc-seconds Monthly 2012-present 75 N 65S Worldpop Population 3 or 30 arc-seconds Annually or less 2000-present Global b c ShakeMaps Earthquake 0.0167° One per EQ 1996-present Global Heavily dependent on time of year, i.e. during summer the hemispheres will have less nighttime coverage EMost are 0.0167°, but for Nepal it is 0.0333 Earlier earthquakes are also covered, but the development did not start until 1996 significance are similar across both sets, with an initial 4–6 months, before it tapers off. Months 6, 7, 10 and 11 negative shock during the month of the Gorkha earth- are insignificant in the province analysis. In month 9, quake followed by a positive effect for months 1 through the country results show a negative and significant coef - 8. The positive effect is most pronounced for the first ficient value, whereas Central and Western stay positive. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 8 of 13 Fig. 5 Comparison of the spatial resolution of the ShakeMap, the VIIRS nightlight data and the WorldPop data Afterwards, months 10 and 11 show a negative trend for are populated and whether any populated area is consid- both sets, while month 12 exhibits a strongly significant ered to be rural or urban. The former threshold was set and positive jump. at 5, which is in line with an average household size in Nepal. Whereas for the latter numerous thresholds were Discussion explored, ranging from 5 (all urban) to 100. However, The aim of the paper was to get highly localized impact the choice had little impact on the regression or mean estimates by using VIIRS data combined with high-reso- results. When running the regressions for urban areas lution population layers and spatially disaggregated dam- only, i.e. only for cells that were deemed to be urban, the age data. The nightlight data has shown in prior studies, coefficient values change, but the significance and direc - some of which use the Gorkha earthquake as their case tion of the signs are the same as for the primary analysis. study, that one can identify local effects from natural dis - In addition, we visually inspected the data to check for asters (Zhao et al. 2018; Gao et al. 2020). A key assump- any differences depending on which year of population tion is that damage can be linked to economic output. was chosen. We found no discernible difference in terms This is not necessarily true for cases where the damages of populated areas and when comparing the population are structural or human in nature. However, it is assumed per pixel between years, the differences were very small that our modelled structural damages will affect the and indicative of a simple application of growth trend nightlight values, which has shown a close link with eco- that has been interpolated across the cells. nomic output, see for example Henderson et al. (2012) or Secondly, when modeling the damage, the building Michalopoulos and Papaioannou (2014). quality assumption will have an impact on the sustained Our analysis rests on several assumptions, which can differ based on the area and time span being analyzed. Firstly, the population data is used to define which areas We also considered which data set to use as our population basis. We did use 2015, and when comparing the data with 2011, the differences are very small and implies that a simple growth trend has been used to adjust the pop- ulation numbers. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 9 of 13 Fig. 6 Mean of nightlight cell values for Nepal and the Central and Western districts split between cells that were affected by earthquakes and those that were not damage as well as the final impact estimates on night - Fourthly, the nightlight values depend on two lights. When the analysis was performed, we did find a assumptions. The first one regards missing nightlight non-negligible effect on the estimates but lacking any values, usually missing due to lack of cloud free obser- meaningful information a mean quality assumption of vations. When no value was present, it was linearly 4 was used. interpolated from the last light value before the missing Related to the first two points, assuming that the observation and the first light value after the observa - building types and quality are homogenously dis- tion. Another option that was explored, was to use the tributed across Nepal is not likely to be true given prior year’s value, but this did not affect the final esti - the diverse topography and climate of Nepal. How- mates. The second assumption was related to the treat - ever, lacking more detailed data on the distribution it ment of VIIRS values close to or below 0. When the is neigh on impossible to correct for. The rural/urban VIIRS images are processed a dark offset is deducted distinction might help alleviate some of the issues and from the raw day/night value. This offset is sometimes given how the results did not change with differing severely impacted by airglow leading to negative light thresholds it gives some confidence in the results. Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 10 of 13 Fig. 7 Regression results for Nepal and the Central and Western districts values. To account for this in our analysis, several should—detrend and clean the data to get more con- options were explored. Firstly, the regressions were run sistent and less volatile nightlight values. with any light value below 0.1 set to 0. Then absolute Finally, the nightlight data shows that during monsoon values were used with any absolute value below thresh- season, the number of cloud free days is on average very olds of 0.1, 0.3 and 0.5 set to 0. Overall, the threshold low. This is partially corrected for by interpolating the choice only had a minor quantitative impact on the nightlight values for the month before and after a month coefficients. Recently, the papers from Gao et al. (2020) with no cloud free days and also tested for by using the and Zhao et al. (2020) show that one can—and probably previous year’s nightlight value. This is not optimal and T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 11 of 13 could impact the estimates depending on how the lack sudden and significant coefficient increase in month 12 of observations is distributed, but we are unaware of any differs from the expected pattern, which is a return to other or better way to correct for the lack of observations. non-significance and coefficient values close to 0, i.e. the As for the results, the mean comparisons of the night- earthquake should not impact nightlight values for a very light values in Fig.  5 yielded no difference between long period of time. When exploring the underlying data affected cells and cells that were not damaged by the and general information about Nepal, we found no spa- earthquake. As can be seen, the trends are the same tial or seasonal pattern in the data or information about for both sets of cells and there are several peaks and new power plants or other events that could have caused throughs throughout the period, some of which may be the spike, leading us to believe that it might be due to the seasonal. The nightlight values exhibit a low during the general volatility of the data. Increasing the lag period is winter months and a high in fall and late fall. Some of this also problematic as it will decrease the sample size and effect could be due to measurement differences and cloud is likely to capture other events that can potentially bias cover, but it might also be caused by differences in use of the results. Once the VIIRS data is available for longer energy and covering up lights during the winter months. periods both pre and post events it will be more feasible The lack of difference between the two means could be to do longer lags and potentially identify more long-term due to aid and relief efforts targeting all areas but given effects, although it can always be argued that any long- the inherent volatility and noise in the VIIRS values it term impact—in particular when one does month by is more likely that any quantitative effects on the local month analysis—are unlikely to be correctly quantified economy has not been strong enough to be captured by due to the many factors that impact light and the meas- the VIIRS data when performing such a simple analy- urements. However, this can be an interesting avenue for sis. To further test this we performed a pre-post analy- future research. sis where we compared the change in nightlight value for Finally, Fig.  8 depicts a comparison between our mod- the month before the earthquake (March) with the value elled damage, gridded out via a kriging algorithm, and for May and June. In this instance, the cells affected by the Copernicus Emergency Management Service (EMS) the earthquake experienced a slightly smaller drop in grading map for Kathmandu. The Copernicus map uses absolute values (− 0.049 vs − 0.057). However, due to satellite images from DigitalGlobe and Pleiades CNES at the higher mean absolute value of the affected cells, the 0.5 m resolution to visually compare ex-post and ex-ante percentage change was − 16% versus − 22%. When per- images and identify changes and potential damages. The forming a t-test between the two sample means, we find focus in the Copernicus maps is at a building level, while the difference to be significant at a 5% level, indicating our modelling focuses on a much more aggregate level that the affected cells experienced a smaller drop than making the two data sets difficult to compare. In addition, the non-affected ones. One possible explanation for the Copernicus maps are only produced for 8 areas that this could be the influx of aid or an inherent difference cover approximately 35 square kilometres each. With this between the two groups. To explore this further, we per- in mind, Fig. 7 is provided as a visual comparison of how formed a mean comparison between the nightlight val- our model performed versus the Copernicus map. Look- ues for both affected and non-affected cells for May and ing at the main area of damage as identified in the Coper - June in 2012–2014 versus the mean in 2015. Once again, nicus map, we find that the outlined rectangular cell we find that the growth for the affected cells is bigger, i.e. with the most damaged buildings map is seemingly more the nightlight value for affected cells is relatively higher damaged than the neighbouring cells. However, looking than for the non-affected cells. This could indicate that at the overall map, the marked cell does not stand out. the two groups of cells are on different trajectories, that This is most likely due to a combination of much lower potentially the affected cells are in more urban areas. This resolution data as well as the kriging algorithm, which could also explain why the fixed effects regressions iden - leads to a smoothing of the damages. tify a significant difference. The regression results did find that a statistical and sig - nificant effect occurred in cells that were damaged by the Conclusion earthquakes. One potential explanation for the nightlight Having used the VIIRS data as an economic proxy one value impact could be that in the month of the quake, finds that the earthquakes in Nepal in 2015 did impact there is a nightlight loss due to the damages. Then, in the local economic activity if one uses fixed effects the following months, there is an increased activity from regressions. Overall, there is an initial decrease when aid, repairs and rebuilding. Finally, one sees a decline fol- the shock occurs, followed by an increase in activity lowed by a negative effect that could signify the end of once aid and relief efforts have started. This is consist - the rebuilding efforts and the aid influx. However, the ent with what one would expect, however, with the Tveit et al. Geoenvironmental Disasters (2022) 9:2 Page 12 of 13 Fig. 8 Comparison for Kathmandu between Copernicus grading map and modelled damage map amount of noise and cloud cover in the underlying Declarations VIIRS data some caution is in order. Competing interests Potential future research within the area could focus The authors declare no competing interests. on other large-scale earthquakes or natural hazard Author details types as the methodology is easily scaled and depends Department of Economics, University of Bern, Schanzeneckstrasse 1, on few inputs. Another option is to explore other algo- 2 3001 Bern, Switzerland. The World Bank Group, Poverty and Equity Global rithms or secondary data to potentially achieve more Practice, Washington, DC, USA. consistent and comparable nightlight measurements. Received: 9 September 2021 Accepted: 26 December 2021 Acknowledgements This paper was funded by the global knowledge program of the World Bank’s Poverty and Equity Global Practice. References Authors’ contributions Agency, Federal Emergency Management (2006) HAZUS-MH MR2 Technical TT: analysis, data gathering, idea, writing, editing. ES: editing, writing, idea, Manual. Washington, DC funding. ES: editing, writing, idea. All authors read and approved the final Baragwanath K, Goldblatt R, Hanson G, Khandelwal AK (2019) Detecting urban manuscript. markets with satellite imagery: an application to India. J Urban Econ. https:// doi. org/ 10. 1016/j. jue. 2019. 05. 004 Availability of data materials Chen Xi, Nordhaus W (2015) A test of the new VIIRS lights data set: population The data used in this study were derived from the following resources and economic output in Africa. Remote Sens 7:4937–4947. https:// doi. available in the public domain: These data were derived from the following org/ 10. 3390/ rs704 04937 resources available in the public domain: Population data Worldpop: https:// Chen Xi, Nordhaus WD (2019) VIIRS nighttime lights in the estimation of cross- www. world pop. org/ proje ct/ categ ories? id=3. Nightlights data VIIRS: https:// sectional and time-series GDP. Remote Sens. https:// doi. org/ 10. 3390/ eogda ta. mines. edu/ downl oad_ dnb_ compo sites. html. Earthquake data USGS rs110 91057 Shakemaps: https:// earth quake. usgs. gov/ earth quakes/ search/. Building inven- tory data: https:// pubs. er. usgs. gov/ publi cation/ ofr20 081160. T veit et al. Geoenvironmental Disasters (2022) 9:2 Page 13 of 13 Corbane C, Kemper T, Freire S, Louvrier C, Pesaresi M (2016) Monitoring the Worldpop (2017) Nepal 100m population, version 2. University of Southamp- Syrian humanitarian crisis with the JRC’s global human settlement layer ton. https:// doi. org/ 10. 5258/ SOTON/ WP005 31 and night-time satellite data. Publications Office of the European Union, Xu C, Tian Y, Zhou B, Ran H, Lyu G (2017) Landslide damage along Araniko Luxembourg, pp 1–14 highway and Pasang Lhamu highway and regional assessment of land- De Groeve T, Annunziato A, Gadenz S, Vernaccini L, Erberik A, Yilmaz T (2008) slide hazard related to the Gorkha, Nepal earthquake of 25 April 2015. Real-time impact estimation of large earthquakes using USGS Shake- Geoenviron Disasters 4:14. https:// doi. org/ 10. 1186/ s40677- 017- 0078-9 maps. In: Proceedings of IDRC2008, Davos, Switzerland Zhao N, Hsu F-C, Cao G, Samson EL (2017) Improving accuracy of economic Driscoll J, Kraay A (1998) Consistent covariance matrix estimation with spatially estimations with VIIRS DNB image products. Int J Remote Sens 38:5899– dependent panel data. Rev Econ Stat 80:549–560 5918. https:// doi. org/ 10. 1080/ 01431 161. 2017. 13310 60 Elliott RJR, Strobl E, Sun P (2015) The local impact of typhoons on economic Zhao X, Bailang Yu, Liu Y, Yao S, Lian T, Chen L, Yang C, Chen Z, Jianping Wu activity in China: a view from outer space. J Urban Econ 88:50–66. https:// (2018) NPP-VIIRS DNB daily data in natural disaster assessment: evidence doi. org/ 10. 1016/j. jue. 2015. 05. 001 from selected case studies. Remote Sens 10:1526. https:// doi. org/ 10. Gao S, Chen Y, Liang L, Gong A (2020) Post-earthquake night-time light piece-3390/ rs101 01526 wise (PNLP) pattern based on NPP/VIIRS night-time light data: a case Zhao N, Liu Y, Hsu F-C, Samson EL, Letu H, Liang D, Cao G (2020) Time series study of the 2015 Nepal earthquake. Remote Sens 12:2009. https:// doi. analysis of VIIRS-DNB nighttime lights imagery for change detection in org/ 10. 3390/ rs121 22009 urban areas: a case study of devastation in Puerto Rico from hurricanes Gibson J, Olivia S, Boe-Gibson G (2019) A test of DMPS and VIIRS night lights Irma and Maria. Appl Geogr 120:102222. https:// doi. org/ 10. 1016/j. data for estimating GDP and spatial inequality for rural and urban areas. apgeog. 2020. 102222 Working papers in economics, University of Waikato. https:// ideas. repec. org/p/ wai/ econwp/ 19- 11. html Publisher’s Note Gillespie TW, Frankenberg E, Chum KF, Thomas D (2014) Night-time lights time Springer Nature remains neutral with regard to jurisdictional claims in pub- series of tsunami damage, recovery, and economic metrics in Sumatra, lished maps and institutional affiliations. Indonesia. Remote Sens Lett 5:286–294. https:// doi. org/ 10. 1080/ 21507 04X. 2014. 900205 Henderson JV, Storeygard A, Weil DN (2012) Measuring economic growth from outer space. Am Econ Rev 102:994–1028. https:// doi. org/ 10. 1257/ aer. 102.2. 994 Hodler R, Raschky PA (2014) Regional favoritism. Q J Econ 129:995–1033. https:// doi. org/ 10. 1093/ qje/ qju004 INFORM (2019) 07. http:// www. inform- index. org/ Porta ls/0/ Inform/ 2019/ count ry_ profi les/ NPL. pdf . Accessed July 2019 International, GeoHazards, and United Nations Centre Regional Development (2001) Final report: global earthquake safety initiative (GESI) pilot project. Report, GHI, 86 p Jaiswal KS, Wald DJ (2008) Creating a global building inventory for earthquake loss assessment and risk management: US Geological Survey Open-File Report 2008-1160. Technical report, USGS, 103 p Jean N, Burke M, Xie M, Davis WM, Lobell DB, Ermon S (2016) Combining satellite imagery and machine learning to predict poverty. Science 353:790–794. https:// doi. org/ 10. 1126/ scien ce. aaf78 94 Klomp J (2016) Economic development and natural disasters: a satellite data analysis. Glob Environ Chang 36:67–88. https:// doi. org/ 10. 1016/j. gloen vcha. 2015. 11. 001 Libois F, Somville V (2018) Fertility, household size and poverty in Nepal. World Dev 103:311–322. https:// doi. org/ 10. 1016/j. world dev. 2017. 11. 005 Michalopoulos S, Papaioannou E (2014) National institutions and subnational development in Africa. Q J Econ 129:151–213. https:// doi. org/ 10. 1093/ qje/ qjt029 Nguyen CN, Noy I (2019) Measuring the impact of insurance on urban earth- quake recovery using nightlights. J Econ Geogr 20:857–877. https:// doi. org/ 10. 1093/ jeg/ lbz033 Sahoo S, Gupta PK, Srivastav SK (2020) Inter-calibration of DMSP-OLS and SNPP-VIIRS-DNB annual nighttime light composites using machine learn- ing. Gisci Remote Sens 57:1144–1165. https:// doi. org/ 10. 1080/ 15481 603. 2020. 18483 23 Skoufias E, Strobl E, Tveit T (2021) Can we rely on VIIRS nightlights to estimate the short-term impacts of natural hazards? Evidence from five South East Asian countries. Geomat Nat Haz Risk 12:381–404. https:// doi. org/ 10. 1080/ 19475 705. 2021. 18799 43 Skoufias E, Strobl E, Tveit T (2017) Natural disaster damage indices based on remotely sensed data: an application to Indonesia (English). World Bank policy research working paper no. 8188 Uprety S, Cao C, Gu Y, Shao X, Blonski S, Zhang B (2019) Calibration improve- ments in S-NPP VIIRS DNB sensor data record using version 2 reprocess- ing. IEEE Trans Geosci Remote Sens 57:9602–9611. https:// doi. org/ 10. 1109/ TGRS. 2019. 29279 42 Wang F, Miyajima M, Dahal R, Timilsina M, Li T, Fujiu M, Kuwada Y, Zhao Q (2016) Eec ff ts of topographic and geological features on building damage caused by 2015.4.25 Mw7.8 Gorkha earthquake in Nepal: a pre- liminary investigation report. Geoenviron Disasters 3:7. https:// doi. org/ 10. 1186/ s40677- 016- 0040-2

Journal

Geoenvironmental DisastersSpringer Journals

Published: Jan 3, 2022

Keywords: Earthquakes; Night-time lights (NTL); VIIRS; Remote sensing; Econometrics

References