Introduction

The recent rise in armed conflicts underscores the need to reduce suffering among affected populations. Timely detection of conflict developments is crucial for effectively responding to and mitigating emerging crises. This need has led humanitarian organizations to explore the use of satellite data to establish conflict monitoring systems that enable continuous screening of large conflict areas, complementing human efforts, which are inherently resource-constrained. While the use of optical and synthetic aperture radar (SAR) imagery for monitoring purposes has gained increasing attention (Sticher et al., 2023), the potential of nighttime light (NTL) data remains underexplored. NTL refers to artificial and natural light emitted at night from the earth’s surface and captured by satellite-mounted remote sensors.

In contrast to other satellite data, NTL can capture changes in human activity, making it a valuable source for detecting conflict dynamics even in the absence of structural changes (Wang et al., 2024; Zheng et al., 2023). Traditionally, NTL data has been used to monitor slow-changing human phenomena such as economic development, urban expansion, population distribution, and economic inequality (e.g., Chen & Nordhaus, 2011; Elvidge et al., 1997; Henderson et al., 2012; Weidmann & Theunissen, 2021; Zhang & Seto, 2013). However, the release of daily NTL data with improved spatial resolution (Li et al., 2022; Román et al., 2018) has expanded the potential to identify patterns or anomalies in NTL that reflect conflict dynamics—such as a sudden decline in NTL when infrastructure is destroyed, populations are displaced, or when communities intentionally reduce lighting to avoid being targeted.

Existing research provides evidence that various conflict dynamics are reflected in changes in NTL (e.g., Eun & Skakun, 2022; Li et al., 2022; Lin et al., 2024; Negash et al., 2023; Wang et al., 2024; Yu et al., 2023; Zhukov, 2023). However, for NTL to serve as an unbiased data source for conflict monitoring, it must be available in a form that ensures equal potential for detecting conflict-related changes across and within conflict contexts. A key requirement for identifying conflict-induced decreases in NTL is the presence of a minimal level of radiance that can be reliably observed from space under normal conditions—a concept we refer to as ”baseline NTL”.Footnote 1 To date, no empirical analysis has systematically examined the relationship between conflict areas and baseline NTL coverage (for general estimates of unlit or energy-poor areas, see McCallum et al., 2022; Min et al., 2024). To address this gap, our article asks the following research question: How is baseline NTL distributed across and within conflict-affected countries?

Based on theoretical and empirical work in conflict research and economic studies, there are multiple reasons to assume that large portions of conflict-affected populations may not be covered by baseline NTL. Studies show that armed conflicts tend to occur in less economically developed countries and, within those countries, in regions that are economically marginalized and where state authority is weak (Cederman et al., 2011; Collier, 2004; Fearon & Laitin, 2003; Gurr, 1970; Østby, 2008). Armed conflicts, in turn, exacerbate poverty and further undermine state capacity (Braithwaite et al., 2016; Collier et al., 2003; Hegre et al., 2017; Petrova et al., 2023). Given that NTL strongly correlates with economic development and a state’s ability to provide and maintain energy infrastructure, we expect conflict-affected populations to be less likely to be covered by baseline NTL compared to those in non-conflict areas. Within conflict zones, we expect that rural and peripheral populations are particularly unlikely to be covered by baseline NTL.

To test these expectations, we combine NTL measurements from the Visible Infrared Imaging Radiometer Suite Day-Night Band (VIIRS DNB) (Román et al., 2018) with high-resolution population and settlement data from the Global Human Settlement Layer and GRID3 (CIESIN, 2024; Freire et al., 2016; Schiavina et al., 2022), and with information on armed conflict from the Uppsala Conflict Data Program (UCDP) (Davies et al., 2023; Sundberg & Melander, 2013). We first estimate the global population living in areas without detected NTL. We then conduct statistical analyses to compare the proportion of people in areas without NTL between conflict-affected and more peaceful regions, as well as between settlements differing in type (rural versus urban) and location (proximity to national and regional capitals).

In addition to these tests of our expectations, we introduce a novel benchmark for assessing the feasibility of conflict monitoring in each country: the “NTL threshold”. The NTL threshold reports the average population density required in an NTL cell for an 80% likelihood of NTL detection. We demonstrate that this threshold provides a more informative measure of the practical limits of NTL-based monitoring than simple percentages of areas with or without NTL, and explore factors influencing this threshold.

Our analyses reveal that in 2020, nearly a quarter of the global population—and almost a third of the conflict-affected population—lived in areas without detectable NTL. Due to a generally lower GDP, conflict-affected countries have particularly high NTL detection thresholds. Focusing on conflict-affected countries in Sub-Saharan Africa, we further show that NTL baseline coverage exhibits a strong urban bias. Built-up areas have significantly higher NTL coverage than smaller settlements, and communities located far from regional capitals are much less likely to be covered by baseline NTL, even when accounting for population size.

These findings contribute to the growing literature on conflict monitoring, particularly regarding the use of satellite data in this context (e.g., Duursma, 2021; Grand-Clement, 2022; Hirblinger et al., 2024; Sticher & Verjee, 2023; Sticher et al., 2023; Willmot, 2017). Our analyses provide, for the first time, an empirical foundation for understanding the limitations of NTL in conflict monitoring, both across and within conflict contexts. By establishing concrete benchmarks, we help clarify the conditions under which NTL may serve as a valuable data source.

Overall, our study shows that it is precisely in the countries where conflict monitoring systems are most needed—namely, less economically developed and conflict-affected countries—that a substantive portion of the population risks being excluded from monitoring. Moreover, many people affected by conflict reside outside urban areas and far away from regional capitals. The dearth of detectable NTL in remote and rural areas suggests that NTL data, often regarded as objective, in fact shares an important bias with conflict event data derived from news sources (Dawkins, 2021; Miller et al., 2022).

Our findings have important implications for researchers and organizations aiming to use NTL for conflict monitoring. While monitoring based on NTL offers unique insights into human activity that other satellite data cannot capture, it should complement rather than replace other data sources and human monitoring. By highlighting both the potential and limitations of NTL in capturing conflict-induced changes, our study provides an empirical foundation for future research.

Conflict monitoring using satellite data

We define conflict monitoring as the continuous observation of countries or conflicts to detect relevant events or their consequences as they are unfolding. The focus on detection differentiates monitoring from the related concept of conflict prediction, which involves forecasting future events based on historical data (Rød et al., 2024). An essential feature of conflict monitoring systems is their ability to detect new developments beyond existing focus areas. This broader scope contrasts with intelligence gathering by human analysts, which is often selective and more spatially or temporally constrained due to limited resources (Duursma, 2021; Willmot, 2017). Techniques such as machine learning can aid in processing diverse data sources relevant to conflict monitoring, including mobile phone records, social media activity, crowd-sourced crisis maps, and satellite data (Arendt-Cassetta, 2021; Meier, 2015; Sticher et al., 2023). In this article, we focus specifically on conflict monitoring using satellite data.

Satellite data has been used for various purposes to collect information on conflict, including to detect damaged buildings (e.g., Fakhri & Gkanatsios, 2021; Holail et al., 2024; Mueller et al., 2021), conflict-related displacement (e.g., Pech & Lakes, 2017; Ren et al., 2020), changing patterns of economic activity in war zones (e.g., Guo et al., 2024) or environmental damage from conflict processes (e.g., Aung, 2021). Many of these studies rely on high-resolution imagery. However, for conflict monitoring, the need to continuously screen vast regions quickly makes the cost of acquiring and processing high-resolution images prohibitive (Sticher et al., 2023). Scholars are therefore exploring approaches that rely on open-access moderate-resolution imagery (Dietrich et al., 2025; Hou et al., 2024; Racek et al., 2025).

In the trade-off between costs, usefulness, and feasibility, one type of satellite data that remains underexplored for the purpose of conflict monitoring is nighttime light (NTL) data. Nighttime light refers to light reflected from the earth at night, emanating from artificial sources (such as streetlights and building lamps) and natural sources (such as fires). By capturing both access to and use of energy, NTL provides a unique data source for understanding changes in human behavior that do not necessarily lead to physical changes on the earth’s surface (Q. Zheng et al., 2023).

Until the release of daily NTL data (Román et al., 2018), NTL research primarily focused on slow-changing phenomena, as the data was only available on an annual basis (Elvidge et al., 2014). In particular, NTL was widely used as a proxy for economic development, including at the local level, where detailed development indicators are often unavailable (e.g., Buhaug et al., 2021; Cederman et al., 2015; Weidmann & Schutte, 2017; Zhou & Shaver, 2021). The daily data unlocked the potential to detect short-term events with rapid light intensity variations, driving a surge of research across various fields (Combs & Miller, 2023; Q. Zheng et al., 2023).

This research shows a correlation between conflict dynamics and NTL. NTL fluctuations align with shifting frontlines and theaters of operation (Eun & Skakun, 2022; Guo et al., 2024; Li et al., 2022; Lin et al., 2024; Negash et al., 2023; Wang et al., 2024; Yu et al., 2023). Many conflict events result in (at least temporary) decreases in NTL: energy availability may decline due to infrastructure destruction (Wang et al., 2024; Zhukov, 2023); governments cut energy supplies in rebel-controlled areas (Li et al., 2018); curfews reduce public activity (e.g., Li et al., 2022; Zheng et al., 2022); and locals turn off lights to avoid becoming targets (Zhukov, 2023). However, some events may also increase NTL, such as fires from attacks on buildings and infrastructure (Cerra et al., 2024; Naghizadeh, 2024) or light from search and rescue efforts. Population movements can drive NTL variations in different ways (Bennett & Faxon, 2021; Huang et al., 2023; Wang et al., 2024): abandoned areas may see a decrease, while makeshift camps may contribute to an increase.

The observed impact of conflict dynamics on NTL, combined with the availability of daily NTL data, creates an opportunity to integrate NTL into conflict monitoring (Salisbury et al., 2024). Unlike conflict data derived from news reports—subject to biases from media coverage, cell phone access, internet availability, and political repression (e.g., Eck, 2012; Miller et al., 2022; Weidmann, 2016)—NTL data collection is unaffected by such factors. This suggests that NTL-based systems could expand coverage to previously inaccessible conflict zones and offer a more objective alternative to news or social media sources.

However, for NTL to support unbiased conflict monitoring, it must be available in a form that ensures equal potential for detecting conflict-related changes across different contexts. In principle, NTL data is globally available on a daily basis (Román et al., 2018). However, detecting changes in NTL requires not just data access but also variation in NTL itself. In particular, for conflict events that reduce NTL, a baseline level prior to the event is necessary to detect such declines. How is this baseline distributed across and within conflict contexts?

The next section develops theoretical expectations about NTL distribution in conflict settings, laying the groundwork for empirically assessing the potential and limitations of NTL for conflict monitoring.

Theoretical background

Previous research using NTL indicates that approximately 1.18 billion people in the developing world live without electricity (Min et al., 2024), and that in 2015, 19% of the world’s built-up areas were without detectable NTL (McCallum et al., 2022). Drawing on theoretical and empirical work in conflict research and economic studies, we argue that populations affected by conflict are particularly likely to reside in areas without baseline NTL.

First, there is a well-established relationship between economic development and the onset of armed conflicts. Conflicts tend to occur in lower-income countries and, within them, in economically underdeveloped regions. Grievances related to economic and other disadvantages of people in poorer areas are one explanation (Cederman et al., 2011; Gurr, 1970; Østby, 2008). Lower opportunity costs for potential insurgents in impoverished regions to participate in rebellion are another explanation (Collier, 2004). In addition, in economically underdeveloped regions, state capacity to police and monitor territory tends to be weaker, creating more opportunities for insurgents to operate (Fearon & Laitin, 2003). Together, these arguments suggest that armed conflicts are more likely to occur in poor areas with weak state presence.

Second, conflicts directly impact economic development by disrupting economic activities and trade, destroying infrastructure, and often causing economic stagnation or collapse (Braithwaite et al., 2016; Collier et al., 2003; Hegre et al., 2017; Petrova et al., 2023). The mutually reinforcing relationship between armed conflict, economic underdevelopment, and weak state capacity has been termed a “conflict trap” (Collier et al., 2003). These dynamics suggest that conflicts not only tend to occur in poorer areas but also exacerbate poverty over time.

Third, prior studies show that NTL is strongly correlated with economic development (e.g., Chen & Nordhaus, 2011; Elvidge et al., 1997; Henderson et al., 2012; Weidmann & Schutte, 2017; Weidmann & Theunissen, 2021). Economically developed regions tend to have higher levels of measurable NTL compared to poorer regions. Combined with the previous arguments—that conflicts tend to occur in and contribute to poor regions—this suggests that populations living in areas affected by armed conflicts are less likely to be covered by baseline NTL.

We add two further considerations that may reinforce these dynamics. First, beyond indirect effects through economic development, the onset of conflict can directly affect NTL coverage. As discussed above, there is a well-documented relationship between conflict and NTL. For example, Ukraine lost 68% of its NTL intensity within a month following the Russian invasion in 2022 (Wang et al., 2024). Many areas previously covered by baseline NTL went dark, making it impossible to observe further NTL reductions (Xiao et al., 2024). In addition to short-term fluctuations, conflict may contribute to long-term declines in baseline NTL by destroying energy infrastructure and weakening a state’s capacity to maintain and operate existing energy systems.

Second, energy infrastructure is unevenly distributed within countries, with rural peripheries often at a disadvantage (Bouzarovski & Simcock, 2017). Armed conflicts frequently emerge in regions where the state has a weak presence or is entirely absent (Buhaug, 2010; Risse & Stollenwerk, 2018). In such areas, state institutions struggle to fulfill basic governance functions, including the provision and maintenance of energy infrastructure across the national territory (Singh, 2023). As a result, rural areas—particularly those distant from administrative centers—are less likely to have detectable baseline NTL, further limiting the potential for conflict monitoring using NTL data.

If we combine these insights, we arrive at two main expectations:

Expectation one: Populations in conflict-affected areas are less likely to be covered by baseline NTL compared to those in non-conflict areas

This follows from the strong link between conflict and economic underdevelopment. Conflicts are more likely to occur in lower-GDP countries, and once they begin, they exacerbate economic decline and degrade energy infrastructure, further reducing electricity availability and NTL coverage.

Expectation two: Within conflict zones, rural and peripheral populations are particularly unlikely to be covered by baseline NTL

This expectation is grounded in the relationship between armed conflict, weak state capacity, and uneven infrastructure distribution. Conflicts often arise where state presence is limited, restricting the provision and maintenance of energy infrastructure.

In short, we have strong reasons to assume that NTL is not equally distributed across and within conflict contexts, and that coverage may be biased against populations that are directly affected by armed conflict.

Methods

To test our expectations, we systematically examine the distribution of baseline NTL across and within conflict contexts, accounting for factors that are already known to affect NTL distribution.

For the first expectation, we begin by estimating the global proportion of individuals without remotely sensed NTL and then statistically test whether this proportion differs significantly between conflict-affected and non-conflict-affected areas. After testing this general expectation, we conduct regression analyses to explore how poverty, urbanization, and conflict affect the NTL threshold. The NTL threshold, introduced in this study, represents the population density at which baseline NTL is likely to be detected in each country. These additional analyses do not aim to establish definitive causal relationships but rather to introduce a practical benchmark for assessing the feasibility of conflict monitoring and to explore how poverty, urbanization, and conflict affect this feasibility.

To test the second expectation, we study the within-country distribution of baseline NTL across settlements in conflict-affected countries in Sub-Saharan Africa. First, we determine whether each settlement is “monitorable”, defined as having at least 50% of its area covered by NTL. We then statistically test whether the proportion of monitorable settlements differs significantly between built-up (urban) areas and small (rural) settlements. To assess the influence of peripheral location, we conduct logistic regression analysis to evaluate how distance from the regional capital influences the probability of a settlement being monitorable, while controlling for population size and whether the settlement is located in the state capital district. As before, our aim is not to establish definitive causal relationships but to identify meaningful associations between settlement type, location, and baseline NTL that can inform further research.

Before describing the methods to test our expectations in more detail, we first define baseline NTL and armed conflict.

Baseline nighttime light

Our main phenomenon of interest is baseline NTL, which simply indicates that the sensor detected a minimum level of nighttime light in a given area—i.e., that a pixel (approximately 450 × 450 m) is not entirely dark. To give a sense of the light required for baseline NTL to be present, Coesfeld et al. (2020) found that a set of 12 streetlights per pixel in an agricultural field, isolated from other light sources, was detectable. We use the annual composite from NASA’s Black Marble nighttime lights product suite, which is based on data from the VIIRS Day-Night Band (Román et al., 2018). The Black Marble data is pre-processed, having been corrected for atmospheric distortions, BRDF effects related to different surfaces and viewing angles, and seasonal variations.

We use annual rather than daily or monthly data, as the focus of our analysis is on capturing baseline NTL—the presence of radiance under normal conditions—rather than examining changes over time. We selected the year 2020 for our analysis to correspond with the latest available data on population density. Our cross-country analysis covers the entire globe for 2020,Footnote 2 while our within-country analysis focuses on conflict-affected countries in Sub-Saharan Africa during the same year.

Armed Conflict

An armed conflict in this study refers to a context where two armed groups, at least one of which is a state government, use force to pursue political change, resulting in casualties on both sides.Footnote 3 This encompasses conflicts more commonly known as civil wars and inter-state wars. We classify a country as conflict-affected if the UCDP/PRIO Armed Conflict Dataset (Davies et al., 2023; Gleditsch et al., 2002) identifies a state-based conflict in the country between 2016 and 2020. Setting the timeframe to the five years leading up to and including our study year, 2020, limits the focus on countries that were relatively recently affected by conflict.

To test whether baseline NTL coverage differs between conflict-affected and non-conflict-affected areas in 2020, we use subnational information on where exactly armed violence took place. This information comes from the UCDP Georeferenced Event Dataset (GED) (Sundberg & Melander, 2013). The recorded violence includes military battles between the main armed groups in a conflict, as well as other forms of organized violence, such as the intentional targeting of civilians and clashes between rebel groups. The data show that nearly 10 percent of the global population lived in violence-affected districts in 2020.

Global analysis

For our comparisons of baseline NTL across and within countries, we create a dataset that records the population living in pixels (NTL cells) with and without detected radiance. To create this data, we combine global NTL data with population estimates from the Global Human Settlement Population Grid (GHS-POP) from the GHS Layer for 2020 (Freire et al., 2016; Schiavina et al., 2022). The population counts in the GHS-POP raster data are estimates derived by disaggregating the best-available census data into grid cells using Landsat satellite data to identify built-up structures. We aggregate and resample the 3 arc-second GHS-POP cells to match the 15 arc-second resolution of the NTL data. This process yields, for each NTL pixel, information on the number of people living in the respective area and whether baseline NTL is present. Using this data, we calculate the share of each country’s total population residing in pixels without NTL and then aggregate this information for a global baseline.

To assess whether the proportion of people without baseline NTL differs between conflict-affected and non-conflict-affected areas globally, we use the disaggregated UCDP conflict event data (Sundberg & Melander, 2013) described above. For each country, we calculate the total population in conflict-affected and unaffected second-level administrative units in 2020, along with the share without NTL in each. We aggregate these numbers to the global level, report the difference in proportions across the two samples, and use a two-proportion z-test to ascertain that the difference is statistically significant.

For further analyses into the feasibility of conflict monitoring, we use the global NTL-population data to calculate the NTL threshold. The NTL threshold records, for each country, the average population required within a pixel to have a high (80%) probability of detectable nighttime radiance. It is a practical summary measure for the potential to detect variation in NTL even in more sparsely populated regions. To estimate the NTL threshold we employ a General Additive Model (GAM) for each country, regressing a binary measure of whether there is any remotely sensed radiance in the cell in 2020 on the cell population.Footnote 4 We report the adjusted population size that accounts for the different absolute size of 15 arc-second cells at different latitudes, so the reported value assumes a cell of roughly 450 × 450 m.

Finally, we conduct descriptive and regression analyses of how the NTL threshold is influenced by economic development, urbanization, and armed conflict. For economic development, we use Gross Domestic Product (GDP) data from the World Bank (World Development Indicators Database (2025)), divided by the country population to get a per capita estimate, and logged to address skew. Urbanization data is from the 2020 Urbanization Index (UN Population Division Database (2025)) and quantifies the proportion of the population residing in clearly defined urban areas. We include information on whether a country was recently conflict-affected, as defined above. For the statistical analyses of the effect of these variables on the NTL threshold we estimate simple Ordinary Least Square (OLS) models with a log-transformed dependent variable and with the country as unit of analysis.

Within-country analysis

For the within-country analysis, we focus on 21 conflict-affected countries in Sub-Saharan Africa. This regional focus is motivated by two factors. First, countries in Sub-Saharan Africa are disproportionately impacted by armed conflicts. Almost half of the armed conflicts in our sample took place in that area. Second, standardized settlement data is available for Sub-Saharan Africa in the form of GRID3 data (CIESIN, 2024), enabling us to assess NTL distributions across settlements in the entire region. Using settlement data is particularly relevant in conflict contexts, as conflicts do not stop at arbitrary data grid cells; instead, settlement boundaries offer more meaningful measures of how populations interact and are affected by conflict dynamics.

We conduct two key analyses to test the expectation that rural and peripheral settlements in conflict zones are less likely to be covered by baseline NTL. First, we distinguish between built-up areas and small settlements as defined in the GRID3 data. Built-up areas are urban areas where buildings are relatively close to each other and where there is a visible grid of streets and blocks, whereas small settlements are not as densely built and can be considered rural or semi-urban (CIESIN, 2024). There is a third category, hamlets, that we do not report because we find hardly any NTL baseline coverage for them.Footnote 5 To compare NTL coverage across the two settlement types, we code for each settlement whether it is “monitorable”. We define a settlement as monitorable if at least 50% of the settlement area has baseline NTL, using precise zonal statistics that take into account raster cells partially covered by a settlement area.Footnote 6 The monitorability metric is relevant because, even if not the entire settlement area is covered by baseline NTL, algorithms are more likely to detect patterns if substantial portions of a settlement have baseline NTL coverage.

For each country, we report how the proportion of settlements that are monitorable differs between urban (built-up) and more rural small settlements. We also use a two-proportion z-test to ascertain that the difference is statistically significant. Because the 50% threshold is at least somewhat arbitrary, we also report two additional metrics to compare NTL coverage across built-up areas and small settlements.Footnote 7 One is the portion of settlements that have any NTL coverage. The third metric is the median NTL coverage across all settlements in a country, which indicates how much of the area of settlements is typically covered by baseline NTL. This metric is calculated for settlements with at least some NTL coverage.

Our second test assesses whether settlements in the periphery are less likely to be covered by baseline NTL, as predicted as part of expectation two. To do so, we examine the effect of a settlement’s location on baseline NTL. Specifically, we use a logistic regression model to evaluate how distance from the regional capital influences the probability of a settlement being monitorable.Footnote 8 Distances are calculated using the great circle distance (haversine formula) between the settlement centroid and the coordinates of the state capital.

We control for population size and whether the settlement is located in the state capital district. Ideally, we would also control for indicators of socio-economic development at the settlement level; however, such granular data is not readily available. While we acknowledge that socio-economic underdevelopment is likely a central driver behind reduced NTL coverage in peripheral settlements—indeed, this relationship is explicitly part of our theoretical expectation—we do not aim to isolate its specific causal effect here. Instead, the purpose of this analysis is to empirically assess, for the first time, the overall extent to which peripheral areas systematically lack baseline NTL coverage in conflict-affected contexts. In our models, we use log transformations of the distance and population variables given their right-skewed distribution.Footnote 9

Results

Almost a quarter of the world’s population has no detectable NTL

In 2020, 77% of the global population lived in places that had satellite-detected nighttime light. For pattern and anomaly detection dependent on baseline NTL coverage, this number means that almost a quarter of the world’s population (23%) remain literally in the dark. These results add nuance in the form of population estimates to the study by McCallum et al. (2022), which found that 19% of populated places lacked detectable nighttime radiance in 2015. The findings are also consistent with Min et al. (2024), who estimate that 1.18 billion people in the developing world experience energy poverty. Globally and with a slightly different approach, we find that this number stands at 1.77 billion people.

Figure 1 shows large differences between countries. Many countries, especially smaller ones like Singapore, Bahrain or Kuwait have remotely sensed nighttime light in almost all populated areas, whereas in South Sudan—a country ravaged by a decade of war and massive displacement—only 6% of the population live in areas with NTL. The African continent has by far the highest share of people living in cells without NTL (51%). That over half of Africa’s population lives in cells not covered by NTL poses an important challenge to conflict monitoring, given that a majority of armed conflicts occur on the African continent (Davies et al., 2023).

Fig. 1: Percentage of population without nighttime light.
figure 1

a World map of percent population with no NTL. Green shades indicate countries with below average percentages of the population without NTL; red shades indicate countries with above-average percentages of the population without NTL. b Ten countries with lowest percentage of population without NTL. c Ten countries with highest percentage of population without NTL. d Percentage of population without NTL by UN region.

Full size image

In conflict zones, more people are left out from NTL monitoring

In areas affected by armed conflict, even more people have no baseline NTL. Specifically, 31% of the population living in districts affected by armed violence live in cells not covered by NTL, compared to 22% of the population in areas without this type of violence in 2020. The disparity in baseline NTL coverage, statistically significant at a level far below p < 0.001, persists even when only contrasting violence-affected and less violent regions within countries that experienced conflict at all.Footnote 10

For the analysis of conflict-affected populations, we have moved from the country to the district (second-level administrative unit) level. This disaggregation is necessary because some armed conflicts affect only particular regions of a country. Large swaths of Russia, for instance, remained unaffected by the wars in Chechnya in the 1990s and 2000s, and the same is true for India’s Kashmir conflict, which is concentrated in the Northwest of the country. For conflict monitoring, the areas of a country that are affected by violence are arguably of key interest. These affected areas can change from year to year, as can the set of countries that have armed conflicts at all.

To ensure that our results are not idiosyncratic to our snapshot in time (2020), we also conducted the above tests for districts affected by violence in 2019 and 2021.Footnote 11 The baseline NTL and population data are still from 2020 as these are expected to change only slowly from year to year, whereas armed conflicts may start or end more abruptly. The results are very similar and statistically significant (p < 0.001) for 2019 and 2021. In 2019, the share of the global conflict-affected population that had no NTL was 32%, compared to 22% of people living in areas not directly subjected to this violence. In 2021, the conflict-affected population without NTL was a bit higher, namely 34%, compared to 22% of people living in more peaceful regions.

Among the 55 countries with armed conflicts that are part of this analysis, the 15 countries with the highest proportion of the violence-affected population living in areas without baseline NTL are all located in Africa. Madagascar, Guinea, and Chad are extreme cases: More than 90% of the population affected by armed violence in these countries have no baseline NTL.Footnote 12

A practical benchmark: The NTL threshold and its determinants

So far, our findings suggest that the potential for conflict monitoring using NTL varies greatly between and even within countries, depending on the share of the population lacking NTL. However, a simple percentage of the population in a particular country that lacks baseline NTL offers limited practical guidance for those responsible for detecting humanitarian crises. For these practitioners, it would be beneficial to know the minimum requirements, for instance in terms of population density in an area, before setting up a monitoring system. It is for this purpose that we have developed the NTL threshold for each country: the population density required for there to be a high (80%) likelihood of baseline NTL.

We find that this NTL threshold ranges from just 1 person needed in an area of roughly 450 × 450 m (a NTL pixel) for those countries with full or almost full NTL coverage (e.g. Singapore, Bahrain), and up to around 9000 for the most extreme countries (e.g. Yemen, Comoros). Figure 2 divides countries into threshold categories. The individual categories contain roughly a fifth of the countries each, with the exception of the last category, which includes the roughly five percent countries with the most extreme thresholds. The photos for Uganda and Singapore, chosen for their location close to the equator, are meant to illustrate what a cell may look like at the different population densities—in two countries with very different levels of economic development.

Fig. 2: Countries and their ntl thresholds, with illustration.
figure 2

a People needed in a cell to have a 80% likelihood of nighttime light: b Illustrative example from Uganda; c Illustrative example from Singapore; d Countries in that NTL threshold category. Image source: Google Satellite.

Full size image

Across countries, the NTL threshold is correlated with the percentage of the population that has no nighttime light, with a correlation coefficient of 0.9.Footnote 13 And yet it is not the same. Take the Republic of the Congo and Ireland—two countries with similar population sizes and in both of which roughly a quarter of the population has no remotely sensed nighttime light. And yet they have vastly different NTL thresholds. With a threshold of 2556 people needed per cell, Congo is an outlier, i.e., the NTL threshold is much higher than we would expect given the share of population lacking NTL. Ireland, with a threshold of 22, is an outlier in the other direction, i.e., the NTL threshold is much lower than predicted.

There are several explanations for these differences. One is that population itself is very differently distributed: In Ireland, the population is more evenly distributed across the country. In Congo, on the other hand, the roughly same population is concentrated in ten times fewer cells than in Ireland. What also influences the NTL threshold is the presence of very densely populated areas with no NTL. The presence of such cells is uncommon in Ireland. In Congo, on the other hand, even densely populated cells do not reliably have NTL.Footnote 14 The NTL threshold, unlike the simple percentage of the population with NTL, therefore better takes into account these differences in population and light distribution, and gives a better indication of the practical limits of NTL monitoring systems in each country.

The NTL threshold of an individual country is well predicted by the country’s GDP per capita and the level of urbanization. In fact, as the results of an OLS regression in Fig. 3b show, 68% of the variance in the NTL threshold is explained by just these two variables. Analyses into the relative importance of the two predictors (through cross-validation, and also by simply leaving one variable out of the model) show that GDP is more influential in predicting the NTL threshold, but urbanization still adds predictive accuracy.Footnote 15 A 10% increase in GDP per capita is associated with more than a 10% decrease in the NTL threshold (holding urbanization constant). Overall, as the right-hand side column of the regression table shows, GDP and urbanization explain and predict the NTL threshold better than they predict the percentage of the population that has no remotely sensed NTL, again showing that the NTL threshold is a more useful measure for practical purposes.

Fig. 3: GDP, urbanization, and nighttime light.
figure 3

a GDP and urbanization of individual countries and their respective NTL threshold categories, indicated by the size of the country circles. b Results of OLS regressions of how GDP per capita and urbanization are related to NTL threshold and percentage of the population without NTL, respectively. *p < 0.1; **p < 0.05; ***p < 0.01.

Full size image

Figure 3a illustrates these relationships between GDP, urbanization, and NTL. Specifically, it shows that GDP and urbanization are strongly correlated, and that NTL thresholds (represented by bubble size) get higher as GDP and urbanization get lower. The red bubbles further illustrate that countries affected by armed conflict in 2016–2020 are disproportionately among those with high NTL thresholds in 2020. We also tested whether including a variable for armed conflict in the regression model helps us explain a larger share of the variance in NTL thresholds among countries, but this was not the case.Footnote 16 It appears that the high NTL threshold in conflict countries is primarily driven by their lower GDP levels—the “conflict trap”.

There are, of course, many factors that might influence the NTL threshold beyond GDP, urbanization, and conflict, such as the level of regional inequality, countries’ electrification strategy (Falchetta et al., 2020) and high levels of informal electrification, or even atmospheric conditions that affect remote-sensing measurements (Coetzee et al., 2023). The results presented here should be interpreted as descriptive evidence that point to further research avenues in explaining and predicting the NTL threshold and thus the feasibility of conflict monitoring systems using NTL.

NTL baseline coverage has a strong urban bias

Our global assessment reveals that violence-affected areas have a higher percentage of the population not covered by NTL, and crucially, that the population threshold for NTL coverage is very high in many countries with armed conflicts. These findings highlight difficulties in detecting NTL anomalies that could point to humanitarian crises in the very regions that are most affected by armed conflicts. To better understand how limitations in NTL coverage affect the feasibility of using NTL data to monitor and detect conflict-related changes, we conducted a more fine-grained analysis of NTL distribution within conflict-affected countries in Sub-Saharan Africa.

Monitorable settlements in Sub-Saharan African countries

Across all conflict-affected countries in Sub-Saharan Africa, 23% of built-up (urban) areas are monitorable by our requirement that a settlement should have NTL coverage for at least 50% of its total settlement area. For smaller rural and semi-urban settlements this proportion is substantially lower: only 1.5% of small settlements are monitorable by this standard. These results show a stark difference in the feasibility of conflict monitoring systems between urban and rural areas, as we expected. This difference in the share of monitorable settlements between urban and rural areas is statistically significant (p < 0.001). The same is true for the difference in proportions for each country.Footnote 17

Figure 4b illustrates the wide variation in the proportion of monitorable settlements across different conflict contexts. The share of settlements (both urban and rural) that are monitorable is highest in the Republic of the Congo (Brazzaville), followed by Sudan. Generally, there is less difference between countries in the extent to which their rural (or semi-urban) settlements are monitorable. Almost without exception, this number is below 5% and thus extremely low. The larger difference between these countries lies in the extent to which their urban areas are monitorable. However, even these often have low NTL coverage. In seven countries, less than one in ten built-up areas is monitorable.

Fig. 4: NTL baseline coverage in built-up and small settlements.
figure 4

a Examples of built-up areas with typical (approximately median) NTL baseline coverage in Congo (93%), Somalia (60%), and Chad (17%). Basemap source: Mapbox satellite imagery. b NTL baseline coverage in built-up and small settlements for war-affected countries in Sub-Saharan Africa. 1) Median values are reported for settlements with baseline NTL coverage. 2) Data for Ethiopia only available for built-up areas.

Full size image

The proportion increases when considering the presence of baseline NTL in settlements more generally. For example, in Burundi and Uganda, while less than 5% of built-up areas are monitorable according to our definition, almost 40% of these settlements have at least some NTL baseline coverage, indicating there is at least one NTL cell with detectable light within the settlement area. The portion of a settlement typically covered (median baseline NTL coverage) also varies considerably, ranging from high (92.4% in the Republic of the Congo) to very low coverage (6.5% in South Sudan). Figure 4a shows illustrative examples of how built-up areas with typical NTL baseline coverage appear from space, showcasing three scenarios: a built-up area with a high portion of the area covered (Republic of the Congo), medium coverage (Somalia), and low coverage (Chad).

Location is a strong predictor of monitorable settlements

Across settlement types, the location of a settlement within the conflict country strongly predicts NTL coverage. This is the result of our regression analysis of NTL coverage on distance from the regional capital. The results are presented in Fig. 5b. The odds ratios indicate that settlements located in the state (first-level administrative unit) that contains the national capital are significantly more likely to be monitorable through NTL. Within first-level administrative units, increasing distance from the regional (state) capital substantially reduces the odds of being monitorable, even after controlling for population size. These findings provide further evidence for an urban-rural bias in the coverage of settlements by baseline NTL data. They also suggest that settlements in the periphery, where the state is not or hardly present, are highly unlikely to be covered by NTL, suggesting an important limitation for the use of NTL for conflict monitoring purposes.

Fig. 5: Settlements close to regional capitals are better monitorable.
figure 5

a Monitorable settlements in three key states affected by the war in Sudan. Settlements with at least 50% NTL coverage are shown in red. Black squares indicate state capitals. b Key results of logistic regression analysis of how being located in the state containing the national capital and proximity to the state capital affect whether settlements are monitorable, controlling for population density. Full results in appendix. *p < 0.1; **p < 0.05; ***p < 0.01.

Full size image

To illustrate the tendency of NTL to be concentrated in urban centers, Fig. 5a shows nighttime light coverage on the example of Sudan, where a war between the Sudanese Armed Forces and the Rapid Support Forces broke out in 2023. The fighting led to high numbers of casualties, widespread destruction, and millions of civilians displaced. We use pre-conflict NTL data, illustrating baseline coverage before conflict-induced changes in activities. Figure 5 shows the distribution of monitorable settlements in three key states affected by conflict violence:Footnote 18 Khartoum state, West Darfur state, and South Darfur state.

Once more, it is evident that the distribution of NTL is heavily biased toward urban areas. Settlements near the national capital, Khartoum, are more likely to be monitorable using NTL data. In contrast, in the West and South Darfur regions, only settlements in close proximity to district capitals surpass the 50% NTL baseline coverage threshold.

Discussion

Recent studies suggest that shifts in NTL can reflect conflict dynamics, highlighting the potential to use NTL for conflict monitoring. While some conflict-related factors—such as the establishment of makeshift camps or the use of nighttime lighting as a security measure—can lead to an increase in NTL, most conflict events result in a decline. Such dynamics can only be observed in areas where baseline NTL is present. Even the most advanced algorithm cannot identify decreases in NTL without detectable light to begin with. For monitoring purposes, it makes no difference whether a settlement lacks detected nighttime light due to the absence of electricity or because its emitted light is too faint to be captured by the satellite at the time of overpass. Understanding the coverage and distribution of baseline NTL is therefore essential to assessing the potential of using NTL for detecting conflict-related dynamics equitably across and within conflict contexts.

Our research is the first to provide a global estimate of the population without remotely sensed NTL and to specifically examine NTL coverage in conflict-affected areas. Previous studies have either mapped NTL in settlements without linking them to population estimates (McCallum et al., 2022) or estimated populations without NTL only in developing countries (Min et al., 2024). To our knowledge, no previous research has focused on conflict-affected areas.

Our findings offer a sobering picture: We show that conflict-affected areas generally have lower NTL coverage than the global average. Almost a third of people affected by organized armed violence are without baseline nighttime light. Within these countries, NTL baseline coverage is lowest in rural and peripheral settlements that tend to be under-monitored already. Combined, these findings suggest that conflict monitoring through NTL is most challenging in the very areas where such detection would be most valuable: remote and inaccessible conflict areas.

While it might not be unexpected that NTL is more prevalent in urban centers, what is truly remarkable is the extent of this bias. In our analysis of conflict-affected countries in Sub-Saharan Africa, only 1.5% of smaller settlements—encompassing rural and semi-urban areas—meet our monitorability criterion of having NTL coverage across at least half of their area, and even smaller settlements (hamlets) rarely ever exhibit any measurable nighttime light. The implications of this urban bias are profound. Information captured by sensors is often regarded as “objective”, as it is not filtered through human analysis. However, the urban bias present in baseline NTL coverage resembles a similar bias in news-derived conflict event data, which is a key source for humanitarian organizations (Dawkins, 2021; Miller et al., 2022). There is a risk that, without accounting for this bias, researchers or organizations may use NTL-based analysis to validate news-derived data without realizing that they could be reinforcing the very biases embedded in those data.

In light of our findings, one clear avenue for research is to study how NTL fluctuations observed in urban areas may reflect conflict dynamics or humanitarian crises in rural regions that lack direct NTL coverage. Under what conditions and in what situations can NTL in urban areas serve as a proxy for broader regional dynamics? If signals captured in urban centers can offer indirect evidence of conflict dynamics in less illuminated regions, this would help mitigate the urban bias in NTL coverage, ultimately improving the reliability of conflict monitoring efforts.

Policy recommendations

Our findings have direct implications for humanitarian organizations developing conflict monitoring systems based on satellite data. Accounting for the absence of NTL in many conflict-affected regions, particularly in peripheral rural settlements, is crucial for interpreting any insights derived from NTL monitoring. As the astronomer Carl Sagan famously noted, “absence of evidence is not evidence of absence” (Sagan, 2011, p. 223). In the context of conflict monitoring, this means that no news is not necessarily good news. Because rural and peripheral areas often lack baseline NTL coverage, crises occurring in these areas may fall outside the scope of NTL-based conflict monitoring. Crises that cause a sudden decrease in NTL may only become detectable when the situation escalates severely, and the effects spread to larger towns or regional capitals.

Our findings also underscore the need for humanitarian organizations to treat NTL as a complementary rather than standalone data source for conflict monitoring. To obtain a more comprehensive picture, it is essential to combine NTL with other data sources—including human monitoring and openly available remote sensing data in other formats, such as fire data, moderate-resolution optical imagery, and synthetic aperture radar data. These additional sources offer different perspectives on conflict dynamics and complement the picture provided by NTL data.

Finally, our analyses carry policy implications not only for humanitarian organizations but also for institutions involved in the development and deployment of satellites, including space agencies and governmental bodies funding remote sensing initiatives. The late overpass time of sensors capturing NTL plays an important role in limiting its coverage. For example, NASA’s Black Marble, which is the most temporally and geographically fine-grained open-access NTL data to date, is captured by a satellite with a local overpass time of approximately 1:30 am local solar time (NASA, n.d.). While such a late overpass time is beneficial for reducing stray light and minimizing missing values due to seasonal variations, it comes at the expense of lower NTL baseline coverage, especially in remote conflict areas where NTL is generally very low.Footnote 19 Investing in the development and deployment of complementary sensors with overpass times aligned to periods of higher baseline NTL emissions—such as late evening or early morning (Stokes et al., 2021)—could help mitigate these limitations and enhance the capacity of humanitarian organizations to monitor conflict dynamics in the areas that need it most.