Satellite nighttime lights reveal increasing human exposure to floods worldwide
Abstract
River floods claim thousands of lives every year, but effective and high-resolution methods to map human exposure to floods at the global scale are still lacking. We use satellite nightlight data to prove that nocturnal lights close to rivers are consistently related to flood damages. We correlate global data of economic losses caused by flooding events with nighttime lights and find that increasing nightlights are associated to flood damage intensification. Then, we analyze the temporal evolution of nightlights along the river network all over the world from 1992 to 2012 and obtain a global map of nightlight trends, which we associate with increasing human exposure to floods, at 1 km2 resolution. An enhancement of exposure to floods worldwide, particularly in Africa and Asia, is revealed, which may exacerbate the projected effects of climate change on flood-related losses and therefore argues for the development of valuable flood preparedness and mitigation strategies.
Key Points
- Nightlights identify flood-prone areas
- 1992–2012 nightlights are a proxy of human settlement evolution
- Increasing nightlights along rivers are associated to higher exposure to floods
1 Introduction
The human activity on the Earth, which affects natural processes and hydrological processes in particular, is recognized to have triggered a new geological era, the Anthropocene [Crutzen, 2002]. Anthropogenic induced modifications of the climate and the global water cycle [Oki and Kanae, 2006], which include changes in pattern seasonality and increasing occurrence of extreme events [Min et al., 2011; Pall et al., 2011], and a possible enhancement of human pressure on riverine areas are expected to produce an increase in river flood risk with consequently devastating effects on human society and the environment. The possible enhanced flood risk induced by demographic expansion, concentration of population in perifluvial areas, and climate change is indeed a source of major concern for riverine people, policy makers, and institutions [Becker and Grunewald, 2003; Di Baldassarre et al., 2010; Bouwer, 2011; de Moel et al., 2011; Michel-Kerjan and Kunreuther, 2011; Hirabayashi et al., 2013; Jongman et al., 2014; Kundzewicz et al., 2014; World Bank, 2014]. In the last decades flood-related damages and fatalities have dramatically increased [Emergency Events Database (EM-DAT): The OFDA/CRED International Disaster Database, 2013; hereafter, EM-DAT]: in 2011 and 2012 nearly 200 million people were affected by floods with a total damage of almost 95 billion U.S. dollars (USD). The development of efficient flood preparedness strategies is largely constrained by rather sparse information on flood hazard, exposure, and vulnerability to flooding events. Previous efforts to assess flood risk mainly focused on a local perspective and did not objectively account for the temporal evolution of exposure due to environmental change [see, e.g., Barnett et al., 2008; te Linde et al., 2011; Bloeschl et al., 2013; Winsemius et al., 2013]. First attempts toward a worldwide flood risk assessment have recently appeared in the scientific literature, though based on complex and uncertainty-prone modeling tools for the identification of areas and assets exposed to floods [Dilley et al., 2005; Peduzzi et al., 2009; Merz et al., 2010; Jongman et al., 2012; Hirabayashi et al., 2013; Meyer et al., 2013; Ward et al., 2013; Arnell and Gosling, 2014; United Nations Environment Programme, 2014].
We propose here an innovative methodology to worldwide assess human exposure to floods (i.e., assets that may be affected by floods) at a fine spatial resolution by using nightlight data [NOAA, 2013]. Nightlights have been widely employed as a proxy for population and settlement density [Elvidge et al., 1997], economic activity [Chen and Nordhaus, 2011], electric power consumption and distribution [Chand et al., 2009], and poverty and development status [Elvidge et al., 2009], as well as for addressing other environmental issues, such as light pollution [Bennie et al., 2014]. Herein, we associate nightlight data with digital information about the geographical coordinates of the river network to locate the areas subjected to flood exposure, i.e., a key determinant of flood risk. In detail, we map global nightlights in the immediate proximity of the river network from 1992 to 2012, with a spatial resolution of 1 km2. To test for a correlation between increasing nightlights and flood damage intensification, we analyze global data of economic losses caused by flooding events [EM-DAT, 2013]. Finally, we examine the increasing trends of nightlights in the proximity of the river network in the last 20 years to identify areas where the human pressure to rivers is raising and, therefore, the flood risk may enhance.
2 River Network Nightlights and Flood-Related Damages
To assess exposure to floods, we associate nightlight data to the global river network. Annual time series of nighttime lights as satellite images from the Defense Meteorological Satellite Program, Operational Linescan System are freely accessible from the NOAA National Geophysical Data Center [NOAA, 2013]. Data cover a period from 1992 to 2012, where six different satellites were used. In total 33 composites are available for the 21 year period, where some years present a coexistence of two different satellites. In this case, an average light value is determined from the two simultaneous satellites. Nightlight data are expressed as yearly averaged digital number (DN) values in the range 0–63, where 0 represents complete darkness and 63 substantially bright areas. DN values are proportional to radiance and mainly identify lights from cities and towns. Data, available in a raster format (GeoTiff), present a spatial resolution of 30 arc sec (0.00833°), which corresponds to nearly 1 km at the equator. The spatial extension of nightlight images is between 75°N and 65°S latitude and 180°W and 180°E longitude. Since original data are not onboard calibrated, nightlight values cannot be strictly compared among years. An intercalibration procedure, extensively applied in recent papers [Elvidge et al., 2009; Chen and Nordhaus, 2011], is therefore employed in this context and adjusted DN values are used.
Regarding the location of the river network, we use the U.S. Geological Survey (USGS) Hydrological data and maps based on SHuttle Elevation Derivatives at multiple Scales (HydroSHEDS) worldwide river network [Lehner et al., 2008]. River network data, available in a vector format, cover an area between 62°N and 55°S latitude and 138°W and 180°E longitude and present a 30 arc sec (0.00833°) spatial resolution. Even though HydroSHEDS data do not cover the entire globe (Canada, Northern Europe, and Russia are appreciably excluded), their spatial resolution is similar to that of nightlight data, and therefore, their use is preferable with respect to other databases.
Nocturnal light values gathered in correspondence of the river network from 1992 to 2012 are first analyzed: for each pixel laying on the river network, we obtain a DN value for all considered years. River network nightlights (see, as an example, Figure 1a) are then examined through a multistep approach focusing on increasingly finer spatial scales: we consider the entire globe, then we move to single continents and finally to countries. Canada, Russia, Norway, Island, Finland, and Sweden, which lay outside the HydroSHEDS edge, as well as all countries with less than 3 · 105 inhabitants are excluded from the analysis. Overall, we consider 178 regions (Table S1 in the supporting information). For each region (world, or continent, or country) and for each year we evaluate the empirical frequency distribution of the DN values, in order to visualize tendencies in time. In addition, we estimate a characteristic DN value, 
, by averaging values measured in pixels located along rivers. Thus, regional average values are analyzed from a temporal perspective to detect the possible presence of trends in time. Next, for each region we estimate a spatiotemporal average value of nightlights 〈DN〉 by taking the average of 
over the whole time range 1992–2012. To test if intense nightlights are associated to severe flood damages, we analyze country-based yearly data on flood-related economic and human losses, gathered from the publicly available international emergency events database [EM-DAT, 2013]. Damage data, expressed in terms of economic losses in USD, and based on market exchange rates in the year of occurrence, cannot be compared among different years and different countries. Raw damage data therefore need to be reported in terms of purchasing power parity (PPP) and normalized for changes in inflation (see also supporting information) [Barredo, 2009]. Data on surface area for each study region, freely available from the World Bank database [World Bank, 2013], are also employed. For each country we evaluate the 1992–2012 temporal average of flood-related variables (i.e., economic damage and people affected) and normalize them to the region surface area in order to compare economic and human damages among countries with different areal extensions. It is well known that flood damage data may be affected by relevant uncertainty due to different monitoring and assessment techniques. However, there is no information available on EM-DAT data reliability. In order to reduce uncertainty, we thus decide to focus on the long-term average of flood damages. Indeed, data aggregated on long time scales are more reliable than damages registered for single-flood events. We thus link flood-related variables to spatiotemporal average nightlights 〈DN〉. Aiming at the identification of a possible relation between nightlights in correspondence of the river network and economic and human damages associated to floods, we apply a log-log regression model to fit the observed data. The Student's t test is then applied to test for the significance of the relationship between the response (i.e., flood-related damage) and the predictor (i.e., nightlight), and the correlation coefficients R and p values are determined.
, which lend themselves to a direct comparison of nightlight trends among different regions. To this aim, we finally apply a linear regression model to fit 
values versus time:
(1)3 Results and Discussion
A statistically significant relationship between temporal averages of flood-related economic damages per unit area (USD/km2) within the period 1992–2012 and spatiotemporal average nightlight values along the river network at a national scale is found (Figure 2). Our analysis confirms that larger economic losses correspond to regions with high nightlight values (e.g., Hong Kong and UK). Such outcome clearly reflects the condition for which higher population densities in the proximity of the river network, and a significant availability of public economic resources correspond to high artificial luminosity and elevated exposure to inundations. Exceptions include Bangladesh and Thailand, characterized by comparatively high economic losses but low luminosity data, mainly related to the lack of electrical infrastructures [Central Intelligence Agency (CIA), 2013], whereas low values in both nightlights and damages are typical of low-income countries (e.g., Congo). Overall, this worldwide picture evidently confirms that increasing nightlights close to watercourses are associated to increasing exposure to flood risk.
Similarly, for each single country we consider the number of people affected by floods per unit area (Figure S1). In this case, we do not find any significant trend between affected people and nightlights, since flood risk awareness and the development of efficient flood prevention strategies in developed countries (i.e., showing elevated nighttime luminosity) typically reduce human losses due to flooding events [World Bank, 2014].
After the identification of a statistically significant relation between nightlight intensity along the river network and flood damages, we are now able to infer temporal trends in exposure to flood risk by analyzing the time evolution of nightlights. Note that direct inference of flood trends from the economic damage data would be hampered by the nonhomogeneous temporal coverage of the damage databases [EM-DAT, 2013], since damage reporting is still sparse in space and time. Nightlight data allow the identification, within each country, of the hot spots where exposure to flood risk is more rapidly changing. Our next step is thus to determine the temporal evolution of spatially averaged light intensities, 
, for each region of interest. On a global scale (Figure S2 and Table S1), we find an overall positive trend of river network nightlights, corresponding to a 1.2% increase per year, which undoubtedly points out a worldwide rise in exposure to flood risk. In order to seek for possible spatial patterns, we analyze the temporal tendencies both on the continental and national scale. For the sake of comparison, we employ 
and evaluate the percentage of per year variation in light intensity, s. Interestingly, an evident differentiation emerges: both considering country-based data (Figure 3) and continental-scale values (Table S1), Africa, Asia, and South America reveal appreciably positive trends; Oceania and Europe exhibit a moderately positive trend, whereas North America presents a slightly negative tendency. These outcomes likely resemble a distinction based on the economic development of single continents, where considerably positive nightlight trends coincide with developing and emerging regions, characterized by an overall lower-middle income, whereas slightly positive and slightly negative nightlight trends mainly correspond to developed continents, with an overall middle-high income [Chen and Nordhaus, 2011]. Decreasing nightlight trends may be also induced by light pollution reduction strategies, like policy-driven initiatives promoted in UK and in several U.S. countries. The temporal enhancement of nightlights close to rivers during the last 20 years suggests a potential increment of exposure to flooding events. River network nightlights constitute indeed an excellent performing proxy for human exposure to floods. However, alternative options could be used, like, for instance, nightlights within the whole extension of the considered study region (i.e., disregarding the position of the river network) or the study region gross domestic product (GDP) based on purchasing power parity (PPP) per unit area. These alternative proxies would give similar results when regressed against flood-related economic damages per unit area over the whole study period 1992–2012. This outcome reflects indeed a strong correlation among river network nightlights, country wide nightlights and GDP data, chiefly because the more illuminated areas are close to rivers (see, for instance, Figure 1a), and also because nightlights are a good indicator for public economic resources. However, a striking difference emerges when focusing on the progress in time of these proxies, therefore analyzing data observed at the annual time scale. In fact, the increasing trend that is clearly visible in the annual nightlights along rivers is much less evident for country wide nightlights. Nightlights show indeed a decreasing trend moving away from the river network, which confirms that people are getting closer to rivers. As a consequence, country wide nightlights cannot efficiently prove an enhancement of human exposure to floods, as the increasing trend of luminosity along rivers is dampened by different tendencies occurring when moving far from the river network. A similar result is obtained with GDP data, whose temporal trend is influenced by economic forcings that are not related to the increasing exposure to floods. These considerations prove that the analysis of the temporal trend of exposure to floods needs to be performed along rivers. By focusing on the area where floods occur, we can remove spurious tendencies that may arise at the country scale. The fine spatial and temporal resolutions of nightlights allow one to assess where exposure is more pronounced and where it is increasing at a faster rate. Regions in the African continent, where riverine nightlights are markedly increasing in time (Figure 3), provide interesting examples. The Nile River and Delta region represent, for instance, a peculiar hot spot characterized by a significant increment in nightlights and therefore in flood prone human settlements (Figure 1b). An additional example is epitomized by Sudan, which presents a distinct average per year increment of luminosity of nearly 6.4% from 1992 to 2012 (Figure S3 and Table S1). In the case of intense floods, this nightlight trend likely attributed to an enhanced concentration of people close to streams and rivers, may lead to severe flood damages, like those experienced during the 2007 flood, with nearly 150 fatalities, more than 560,000 people affected and 300 million USD of economic losses [EM-DAT, 2013]. Likewise, the River Niger flood in September 2012, which caused 65 fatalities and over 100,000 homeless people [EM-DAT, 2013], hit areas undergoing an uncontrolled urbanization in recent years, as confirmed by sharply increasing luminosity (Table S1). Similarly to the African continent, Southeastern Asiatic nations reveal a significant per year increment of nighttime lights, which in the case of harmful flood occurrences, is reflected in conspicuous economic and human losses. For example, the consequences of the 2010 flood in China (i.e., more than 1900 people dead, nearly 140 million people affected and 18 million USD of economic damages [EM-DAT, 2013]) can be reasonably linked to the 5.6% yearly increment of nightlights. In the same way, the catastrophic effects of flooding events in the Philippines in 2009, 2011, and 2012 and in North India in 2013 likely reflect a fast-evolving urbanization, as clearly shown by our data on nightlight variation in time (Table S1). Many European countries and the U.S. are instead characterized by weaker (or nonsignificant) trends in nightlight intensity, which in some cases may be influenced by policies for the reduction of light pollution.
4 Conclusions
The identification of flood-prone areas is a recognized urgent priority worldwide, but expeditious and spatially detailed methods for their characterization at the global scale are still lacking. In this paper we show that satellite-based nightlight maps may offer a strong propellant for the analysis of hydraulic exposure to floods. The unique and objectively measured data set of nightlights offers a spatially explicit—and time variable—assessment of human exposure to riverine inundations. Our results clearly reveal a consistent characterization of enhanced anthropogenic pressure close to streams and rivers, as derived from nightlight data, which translates into higher flood exposure and consequently increasing flood-related economic losses. We believe that our analysis, which correlates for the first time nightlights to human and economic damages due to floods, may provide a benchmark for the identification of regions where a priority attention should be dedicated.
Acknowledgments
The data for this article is available at NOAA Earth Observation Group (http://ngdc.noaa.gov/eog/dmsp/downloadV4composites.html), USGS HydroSHEDS (http://hydrosheds.cr.usgs.gov/index.php), and EM-DAT (http://www.emdat.be/database). Results derive from elaborations through a not-open-source code developed by the authors. More information can be gathered upon email request to serena.ceola@unibo.it. S.C. and A.M. acknowledge the financial support from the EU funded project SWITCH-ON-603587. F.L. acknowledges support from the MIUR funded project RBFR12BA3Y.
M. Bayani Cardenas thanks two anonymous reviewers for their assistance in evaluating this paper.
