Satellite Night-Light Intensity as a Leading Indicator for Regional Consumer-Spending Dynamics: Methodological and Contextual Considerations¶

1. Introduction¶

The concept of using satellite-derived night-light (NTL) intensity as a leading indicator for consumer-spending shifts has gained increasing attention in academic and policy research, particularly in contexts where traditional economic data are either sparse or lack sufficient temporal resolution. NTL data, captured through satellite sensors such as the Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) and the Visible Infrared Imaging Radiometer Suite (VIIRS), provide a continuous and objective measure of artificial light emissions at night, which are often linked to urbanization, economic activity, and infrastructure development [21]. In regions where consumer behavior is closely associated with the presence and intensity of artificial lighting—such as in urban centers—NTL data can serve as a reliable proxy for shifts in spending patterns. However, the relationship between NTL and consumer behavior is not uniform across all environments, particularly in rural or informal economies where lighting may not directly correlate with economic transactions or consumer activity [22].

The central objective of this study is to critically evaluate the utility of satellite-derived NTL data as a leading indicator for regional consumer-spending dynamics. This involves a comprehensive review of the literature to identify prior applications, methodological constraints, and the role of contextual factors in shaping the relationship between NTL and economic behavior. The study further examines empirical case studies in both urban and rural settings, highlighting the differential effectiveness of NTL in capturing consumer activity. The integration of NTL with complementary high-frequency data sources—such as mobile phone usage, social media activity, and point-of-interest (POI) data—is also explored to assess how such integrations can enhance the accuracy and contextual relevance of consumer-spending models.

This analysis is particularly relevant in the context of modern economic forecasting, where the ability to detect early signals of economic change is crucial for policy formulation and resource allocation. NTL data offer a unique advantage in this regard due to their near-real-time availability and spatial coverage, making them especially valuable in data-scarce regions or during periods of rapid economic fluctuation. For example, during the early stages of the pandemic, NTL data combined with electricity consumption metrics were used to estimate GDP contractions in India, demonstrating the potential of such data to capture consumer behavior shifts in the absence of immediate, ground-level statistics [9]. Yet, the interpretive and technical challenges associated with NTL—such as sensor limitations, noise, and the influence of non-economic factors—necessitate careful calibration and contextual awareness in modeling efforts [10].

The structure of this paper is organized to provide a systematic evaluation of the topic. The literature review section outlines the historical applications of NTL data in economic and social analysis, with a focus on their strengths and limitations in modeling consumer behavior. The methodology section details the data sources and analytical techniques used to establish the relationship between NTL and consumer spending, while the case studies section presents empirical evidence from both urban and rural regions. The integration with complementary data sources is examined in a separate section, emphasizing the methodological nuances involved in combining NTL with high-frequency behavioral and infrastructural indicators. Finally, the discussion considers the contextual variations that affect NTL’s utility, including the influence of economic structure and political factors. The limitations and interpretive challenges of NTL as a proxy for consumer spending are also explored, culminating in a set of conclusions and recommendations for future research directions.

This paper aims to contribute to the growing body of literature by providing a structured, context-aware evaluation of NTL as a leading indicator for consumer-spending shifts. By addressing both the technical and interpretive challenges of NTL data and exploring their integration with complementary sources, the study seeks to refine the methodological framework for using satellite night-light data in economic forecasting. The findings are expected to inform both academic research and practical applications, particularly in regions where traditional economic data are limited, and where the visibility of consumer behavior through artificial lighting is a critical factor in understanding economic dynamics. The next section will review the existing literature on the applications of NTL data in economic and social analysis, setting the stage for a deeper examination of the methodological constraints inherent in their use.

2. Literature Review¶

The literature on satellite-derived night-light (NTL) data has increasingly explored their utility as proxies for socioeconomic activity, including urban expansion, GDP estimation, and poverty mapping. This section reviews key prior applications of NTL data in economic and social analysis, emphasizing their strengths and limitations in capturing consumer behavior at different spatial and temporal scales. Particular attention is given to the methodological challenges that affect the accuracy and interpretability of NTL as a leading indicator for economic shifts, including sensor resolution, noise, and the influence of non-economic factors. These discussions provide a foundation for understanding how NTL data can be refined and contextualized to better reflect consumer-spending dynamics, especially in diverse and data-scarce environments. The following subsection first outlines previous applications of night-light data in economic research, setting the stage for a deeper examination of the methodological constraints inherent in their use.

2.1. Previous Applications of Night-Light Data¶

Night-light data have been extensively applied in academic research as a proxy for socioeconomic indicators, particularly in the absence of reliable, high-frequency ground-level statistics. Early applications of such data, dating back to the 1970s, focused on measuring energy consumption and industrial activity, with foundational studies by Croft (1973, 1978) and Welch (1980) establishing the potential of satellite-derived night-light intensity to reflect economic development [9]. Over time, the use of night-light data has expanded to include GDP estimation, urbanization tracking, and poverty mapping, with studies such as Sutton et al. (1997) demonstrating the predictive power of DMSP NOAA OLS imagery in estimating population density with an R-squared value of 0.63 [21].

In the context of GDP estimation, night-light data have been used to infer economic activity at both national and sub-national levels. For example, Henderson et al. (2011, 2012) and Elvidge et al. (1997) have shown strong correlations between nighttime light intensity and GDP in various countries. However, the relationship is not uniform across all contexts. Chen and Nordhaus (2019) caution that the predictive power of night-light data for GDP is weaker in developed economies, where growth is driven more by technological and human capital development than by physical capital that emits visible light [25]. Similarly, Bickenbach et al. (2016) observed significant regional variations in the correlation between night-light growth and GDP growth in countries like Brazil and India, underscoring the importance of contextual factors in modeling economic dynamics [21].

Beyond GDP, night-light data have been employed in poverty mapping efforts, particularly in developing countries where census data are sparse. Jean et al. (2016) and others have demonstrated that night-light intensity can serve as a proxy for poverty levels and economic inequality, especially when integrated with machine learning techniques and daytime satellite imagery [23]. These studies highlight that while night-light data provide a broad-brush indicator of economic activity, their utility in capturing fine-grained consumer behavior is limited without additional contextual refinement. This aligns with the findings of the study on Swedish municipalities, which found that NTL data are a strong proxy for population and establishment density but a weaker proxy for wages, particularly in rural areas [18].

The application of night-light data in urbanization studies has also been significant. The paper "Night-Time Light Data: A Good Proxy Measure for Economic Activity?" emphasizes that NTL data are particularly effective in capturing the spatial and temporal expansion of urban areas, as the intensification of light correlates with infrastructure development and agglomeration effects [18]. Similarly, the study on Colombia by [21] shows that NTL data are useful in estimating Regional Domestic Product (RDP) at the municipal level, with VIIRS data outperforming DMSP in capturing economic activity in both urban and rural settings. These findings suggest that while NTL data can reflect the spatial extent of economic activity, their effectiveness as a proxy for consumer spending—particularly in rural or informal economic contexts—remains methodologically complex.

The limitations of night-light data in capturing consumer behavior are further elaborated in the literature. The study by Gibson et al. (2021) highlights that DMSP data are particularly flawed in low-density rural areas, where the relationship with GDP is either negative or imprecise [10]. In contrast, VIIRS data offer improved spatial accuracy and dynamic range, making them more suitable for sub-national and rural economic modeling. The authors also note that even with VIIRS data, the predictive power for consumer spending is constrained in informal economies where economic activity is not directly tied to visible light emissions [10].

Moreover, the use of night-light data as a leading indicator for consumer-spending shifts is still an emerging area of research. The study on India’s economic impact during the early stages of the pandemic shows how night-light data can be used to estimate GDP fluctuations at the quarterly level, with an elasticity of 1.55 observed for emerging markets and developing economies [1]. While this suggests that night-light intensity can provide early signals of economic changes, the paper also notes that atmospheric noise—such as cloud cover—can introduce measurement errors, particularly with VIIRS data [1].

Taken together, the literature indicates that night-light data have proven most effective in urban settings, where the spatial concentration of economic activity aligns more closely with visible light patterns. In contrast, their utility in rural and informal economic contexts is constrained by factors such as low population density, unrecorded economic activities, and sensor limitations. These findings highlight the need for preprocessing techniques that account for spatial heterogeneity and contextual economic structures, as demonstrated by the application of geographically weighted regression in the Swedish case [18]. As the field continues to evolve, the integration of night-light data with complementary high-frequency indicators—such as mobile phone usage or transportation patterns—may enhance their predictive power for consumer behavior, particularly in data-scarce regions [13].

The next section will explore the methodological challenges inherent in the use of night-light data for consumer-spending analysis, focusing on issues of sensor limitations, preprocessing, and spatial resolution.

2.2. Methodological Challenges in Night-Light Analysis¶

The use of satellite-derived nighttime light (NTL) data as a proxy for economic activity and consumer behavior presents several methodological challenges that must be carefully addressed to ensure the reliability and validity of such analyses. One of the most prominent issues is the spatial resolution of available datasets. The older Defense Meteorological Satellite Program Operational Linescan System (DMSP-OLS) data, with a spatial resolution of 2.7 km at nadir, are limited in their ability to capture fine-grained economic dynamics, particularly in heterogeneous or rural regions where economic activity is not uniformly associated with lighting infrastructure [11]. In contrast, the Visible Infrared Imaging Radiometer Suite (VIIRS-DNB) aboard the Suomi NPP and NOAA-20 satellites provides a higher spatial resolution of approximately 750 m, allowing for more nuanced and localized economic modeling [8]. However, even VIIRS data face limitations, particularly in distinguishing between low-intensity artificial lighting and natural reflectivity from surfaces such as snow or water, which can obscure the true signal of human economic activity [8].

A second critical limitation lies in the presence of noise and variability in NTL datasets, which can arise from multiple sources including cloud cover, lunar illumination, and sensor-specific artifacts. For instance, the DMSP-OLS data are known to suffer from top-coding, a process that saturates the brightest pixels, thereby limiting the ability to capture differences in economic intensity among highly developed urban areas [10]. The VIIRS-DNB dataset, while offering a dynamic range covering nearly seven orders of magnitude in radiance, is not immune to noise and requires preprocessing to mitigate effects such as stray light and auroral interference [11]. Techniques like the Patch Filtering Method (PFM) and thresholding have been proposed to address these issues, but their effectiveness varies depending on the geographic and economic context, particularly in data-scarce or informal economic environments [11].

The influence of non-economic factors further complicates the interpretation of NTL data as a proxy for consumer spending. For example, in rural areas of Sub-Saharan Africa and other regions with significant informal economies, the relationship between NTL and economic activity is often weaker or less direct due to the prevalence of unrecorded economic transactions and the limited use of electric lighting in everyday economic interactions [22]. In such contexts, the informal economy—characterized by cash-based transactions, limited infrastructure, and low electrification—reduces the visibility of economic activity through artificial lighting, making it difficult to derive accurate consumer behavior estimates from NTL data alone. This necessitates a context-aware analytical framework that incorporates complementary data sources, such as electricity consumption, transportation flows, and socio-economic indicators, to account for these confounding factors and improve model accuracy [9].

Moreover, the integration of NTL data with other high-frequency datasets introduces methodological nuances that must be carefully managed to ensure the validity of consumer-spending models. For instance, the study on Indian GDP during the early stages of the pandemic demonstrated that a 90-day averaging window was more effective than the commonly used 30-day window in capturing the true economic signal, particularly in the presence of transient shocks [9]. Similarly, the use of kernel density aggregation and logarithmic transformations has been shown to improve the consistency of NTL datasets when transitioning between different sensor platforms [11]. These findings underscore the importance of selecting and refining preprocessing methods to enhance the reliability of NTL data for consumer-spending modeling.

Technical limitations also extend to data consistency and calibration across time and sensor platforms. The transition from DMSP-OLS to VIIRS-DNB introduces a discontinuity in the data record, with differences in radiometric resolution, sensor specifications, and overpass times between the two systems [11]. For instance, the overpass time of DMSP is around 9:00 pm, while VIIRS data are collected around 1:30 am, potentially capturing different patterns of human activity [11]. This temporal mismatch can lead to biases when generating long-term trends or assessing short-term economic shocks, such as those caused by policy interventions or crises [9]. To mitigate these issues, researchers have proposed inter-sensor calibration methods, such as the use of reference areas with stable light levels and the application of Gaussian point-spread functions to align the spatial characteristics of the two datasets [11]. However, such methods are not universally applicable and often require region-specific tuning to account for local climatic and infrastructural conditions.

In summary, while satellite-derived NTL data offer a promising avenue for tracking economic activity and consumer behavior, their utility is contingent on addressing a range of methodological challenges, including spatial resolution constraints, noise and variability, the influence of non-economic factors, and the need for consistent calibration across different sensor platforms. These considerations are especially pertinent in regions with complex economic structures, such as informal economies in rural Sub-Saharan Africa or service-driven urban centers in emerging markets. The next section will explore the integration of NTL data with high-frequency complementary sources, such as mobile phone usage and social media activity, to enhance the predictive accuracy and contextual sensitivity of consumer-spending models.

3. Methodology¶

This section presents the methodological foundations for evaluating satellite-derived night-light intensity as a leading indicator of regional consumer-spending shifts. It begins by detailing the primary satellite and ground-based data sources used in the analysis, including the DMSP/OLS and VIIRS/DNB platforms, as well as complementary consumer-spending datasets such as retail sales, electricity consumption, and household wealth indices. The discussion extends to the analytical framework that models the relationship between these data and economic behavior, incorporating regression techniques, time-series methods, and machine learning algorithms. These approaches are further contextualized through validation, sensitivity analysis, and interpretability strategies, reflecting the complexity of modeling consumer behavior across urban and rural, formal and informal economic settings. The following subsection outlines the specific data sources employed in the study.

3.1. Data Sources¶

The "Data Sources" section outlines the primary satellite and consumer-spending datasets employed in the analysis of regional economic activity and consumer behavior. These datasets are selected based on their temporal and spatial resolution, availability, and relevance to the modeling of consumer-spending dynamics in both urban and rural environments.

Satellite-derived night-time light (NTL) data primarily include the Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS) and the Visible Infrared Imaging Radiometer Suite (VIIRS) aboard the Suomi National Polar-orbiting Partnership (NPP) satellite. The DMSP/OLS dataset, available from 1992 to 2013, has a spatial resolution of 30 arc-seconds (approximately 1 km) and provides digital number (DN) values ranging from 0 to 63. However, due to issues such as sensor degradation, lack of on-board calibration, and spatial blurring, the DMSP data are not directly comparable across years and are less suitable for sub-national or rural economic modeling [27]. To address these limitations, a stepwise calibration approach has been developed to generate a temporally consistent NTL dataset, enabling more accurate tracking of economic trends [11].

The VIIRS Day/Night Band (DNB) dataset, available from 2012 to 2023, offers a higher spatial resolution of 15 arc-seconds (approximately 0.5 km) and records NTL radiance instead of DN values, providing a more robust and calibrated measure of light intensity [16]. The VIIRS data are processed to remove sunlit, moonlit, and cloudy pixels, as well as outliers such as those from biomass burning, through techniques like the cloud-free average radiance grid and outlier filtering. Additionally, a convolutional neural network (NTLSRU-Net) has been used to upscale DMSP data to VIIRS resolution, allowing for the creation of a harmonized dataset spanning from 1992 to 2023 [16]. These preprocessing steps enhance the utility of VIIRS data in capturing the spatial and temporal dynamics of consumer spending, particularly in urban areas and data-scarce regions [2].

In terms of consumer-spending datasets, the literature has incorporated a range of sources, including retail sales, electricity consumption, and credit card transaction data. For instance, historical fast-moving consumer good sales data from retail shops have been used in conjunction with VIIRS NTL data to estimate urban consumption potentiality [2]. In the context of the Indian economy during the early stages of the pandemic, GDP fluctuations were estimated using NTL data integrated with electricity consumption and precipitation metrics, demonstrating the value of complementary data sources in improving predictive accuracy [9]. Additionally, the household wealth index derived from the Demographic and Health Surveys (DHS) has been used as a proxy for consumer behavior in rural sub-Saharan Africa, where traditional economic data are limited [22].

The temporal resolution of NTL data varies by platform and preprocessing methods. The DMSP/OLS data are typically available as annual composites, whereas VIIRS data are accessible at monthly intervals, offering greater flexibility for analyzing short-term economic shifts [11]. In some applications, a 90-day averaging window has been found more effective than the standard 30-day window in capturing transient economic shocks [9]. The higher temporal granularity of VIIRS data, combined with improved spatial resolution, makes it a preferred choice for modeling urban consumer behavior, particularly in service-driven economies [2].

For rural and informal economic contexts, the literature emphasizes the limitations of both DMSP and VIIRS data in capturing low-intensity lighting patterns. In such settings, the relationship between NTL and economic activity is often weaker, necessitating the integration of NTL with ground-level economic indicators to improve model accuracy [10]. For example, in Colombian municipalities, VIIRS data demonstrated stronger correlations with Regional Domestic Product (RDP) than DMSP data, particularly in urban areas, while the use of the Global Urban Footprint improved model fit for large cities but reduced accuracy in rural regions [21]. These findings suggest that the effectiveness of NTL as a proxy for consumer spending is contingent on the economic structure of the region being studied.

The next section will outline the analytical framework used to model the relationship between satellite-derived NTL data and consumer-spending patterns.

3.2. Analytical Framework¶

The analytical framework employed in this study to model the relationship between satellite-derived night-light intensity (NTL) and consumer-spending dynamics integrates both traditional statistical and advanced machine learning approaches. These methods are designed to address the non-linear and context-dependent nature of the relationship between NTL and economic behavior, particularly in regions with varying degrees of urbanization and formal economic activity. The models are supplemented with control variables and validated using a range of techniques to ensure robustness and generalizability.

Regression analysis is a foundational method in this framework. Ordinary Least Squares (OLS) regression is used to estimate the elasticity of consumer spending with respect to NTL intensity, as demonstrated in the Indian case study, where a 1.55 elasticity was observed for emerging markets and developing economies [1]. This approach allows for the quantification of the marginal impact of light intensity changes on economic activity. However, given the non-linear relationships and potential confounding effects in heterogeneous regions, more flexible regression techniques—such as generalized additive models (GAMs) and geographically weighted regression (GWR)—are also employed to capture local variations and interactions between NTL and socio-economic factors [5]. GAMs are particularly useful for decomposing the effect of individual variables on the model output, while GWR improves the model's ability to reflect spatial differences in how NTL correlates with consumer behavior.

Time-series modeling is another critical component of the framework. Techniques such as ARIMA (AutoRegressive Integrated Moving Average) and Holt-Winters exponential smoothing are used to analyze temporal trends in NTL and consumer spending, allowing for the identification of seasonal patterns, trend components, and anomalies [12]. These models are extended to Bayesian and probabilistic frameworks to incorporate uncertainty into the predictions, which is essential for capturing the variability of consumer behavior. Ensemble methods, including forecast combination and reconciliation, are also integrated to improve accuracy by leveraging information across multiple time series and hierarchical dependencies [12]. For instance, the use of a 90-day averaging window in the Indian case study demonstrated superior performance in capturing short-term economic shocks compared to a 30-day window [9].

Machine learning techniques are central to the modeling process, given their capacity to handle high-dimensional and non-linear data. Random Forest regression is widely utilized for its ability to estimate variable importance and capture complex interactions among features [3]. In the Madrid-based study, %IncMSE metrics from Random Forest models were used to assess the contribution of variables such as retail area, establishment popularity, and tourist accommodations to night-time spending [3]. Similarly, artificial neural networks (ANNs) are employed to model the dynamic interactions between NTL and consumer behavior, particularly in data-scarce environments. The paper highlights the use of partial derivatives to evaluate the relevance of input features in ANNs, although it notes that input effects can saturate quickly, necessitating the use of integrated gradients to better capture feature contributions [12].

A unique approach in the literature is the use of the dominance-based rough set approach (DRSA), which is particularly well-suited for modeling consumer behavior in data-scarce or informal economic contexts [12]. DRSA derives monotonic decision rules from ordinal data, enabling the modeling of consumer preferences even in the presence of inconsistent or sparse data. This method is especially relevant for rural Sub-Saharan Africa, where the informal economy dominates and where NTL data are often sparse or zero [17]. Furthermore, the integration of multiple rule classifiers into ensembles, such as through bagging and boosting, is discussed as a strategy to enhance model performance and reduce overfitting [12].

To ensure model interpretability and transparency—key considerations for academic and policy-relevant research—both local and global explanation methods are applied. Local explanation methods such as LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations) are used to explain individual prediction scores, particularly in Random Forest and neural network models [12]. These methods provide insights into the specific contribution of each feature to the prediction, allowing researchers to identify the most influential variables in consumer-spending models. Global explanation methods, such as partial dependency plots (PDP) and permutation-based feature importance, are also employed to assess the overall impact of NTL and other variables across the entire dataset [12].

Validation and sensitivity analysis are integral to the framework. The models are evaluated using out-of-sample, out-of-period, and out-of-universe validation techniques to assess their generalizability and robustness [12]. In the Indian case study, the model predicted a 24% Year-over-Year (YoY) GDP contraction during the early stages of the pandemic, closely matching the official government-reported decline of 23.9% [9]. This demonstrates the framework’s ability to generate accurate and timely predictions of consumer-spending shifts. Calibration curves and error metrics such as PCC (Pearson Correlation Coefficient), MSE (Mean Squared Error), and MAD (Mean Absolute Deviation) are used to evaluate the reliability of model outputs [12]. Additionally, sensitivity analyses are conducted to assess how different preprocessing techniques—such as data sharpening and thresholding—impact model accuracy, particularly in low-density and informal economic settings [5].

The framework also incorporates complementary data sources to enhance the predictive accuracy of NTL-based models. In the Indian case, electricity consumption and precipitation data were integrated with NTL to improve the model’s ability to capture economic shocks [9]. Similarly, in the context of rural Sub-Saharan Africa, the study integrated NTL with household wealth indices derived from principal component analysis, enabling a more nuanced understanding of how lighting patterns reflect economic behavior [17]. The inclusion of such variables helps mitigate the limitations of NTL data in capturing the full spectrum of economic activity, particularly in regions where traditional economic indicators are unreliable or unavailable.

Finally, the framework addresses the challenges of model interpretability and responsible analytics. In consumer-spending modeling, particularly in credit risk and churn prediction contexts, the use of white-box models such as logistic regression and decision trees is emphasized to ensure transparency and accountability in model outcomes [12]. Additionally, post-hoc interpretability techniques such as SHAP and LIME are applied to complex models to meet regulatory and ethical standards. The study on India also highlights the importance of incorporating uncertainty quantification methods such as Gaussian process regression and quantile regression, which provide confidence intervals for predictions and help assess the risk of misclassification [12].

In conclusion, the analytical framework is designed to be both flexible and context-aware, combining statistical and machine learning techniques to model the relationship between NTL and consumer-spending behavior. The inclusion of control variables, the application of preprocessing and calibration methods, and the integration of complementary data sources are all essential for enhancing the accuracy and reliability of these models. The next section will examine the integration of NTL data with high-frequency complementary sources such as mobile phone usage and social media activity to further refine consumer-spending predictions.

4. Case Studies¶

This section presents empirical case studies that investigate the utility of satellite-derived night-light intensity (NTL) as a leading indicator for regional consumer-spending shifts, emphasizing the contextual variability in model performance across urban and rural environments. The analysis is structured around two primary settings—urban and rural regions—each of which presents unique methodological considerations, data integration strategies, and limitations in capturing economic behavior through luminosity patterns. Urban areas benefit from higher economic density and infrastructure visibility, where sensor type, preprocessing, and multi-source integration significantly influence the predictive accuracy of NTL for consumer spending. In contrast, rural areas exhibit weaker correlations between NTL and economic activity due to factors such as informal economies, low electrification, and spatial heterogeneity, necessitating tailored modeling approaches. These case studies collectively highlight the nuanced relationship between NTL and consumer behavior across diverse geographic and economic contexts.

4.1. Urban Areas¶

Urban areas represent a focal setting for the analysis of satellite-derived night-light intensity (NTL) as a proxy for consumer-spending dynamics, given the high concentration of economic and social activity. The literature consistently shows that NTL data perform more reliably in urban contexts compared to rural ones, largely due to the density of infrastructure and the visibility of economic output through artificial lighting. For instance, in Indonesia, VIIRS data demonstrate a positive and significant elasticity with real GDP for second-level sub-national urban regions, whereas DMSP data yield negative and statistically insignificant results [10]. This discrepancy underscores the superior performance of VIIRS in capturing urban economic activity, attributed to its higher dynamic range, spatial accuracy, and calibration, which are essential for modeling consumer behavior in complex, heterogeneous urban environments.

The relationship between NTL and urban consumer-spending patterns is particularly strong when examining aggregated spatial units such as provinces or large cities, where economic activity is more homogeneously represented by light intensity. In China, for example, VIIRS data show a robust correlation with GDP in urban areas, especially for the industrial and services sectors, which are more directly tied to consumer behavior and infrastructure usage [10]. Conversely, the primary sector—dominated by agriculture—exhibits a weaker link to NTL, consistent with the fact that agricultural activity is not necessarily reflected in nighttime luminosity. These findings align with broader empirical evidence from North American and European studies, where NTL data are shown to capture variations in urban economic density and infrastructure development [18].

A critical factor influencing the accuracy of NTL data in urban modeling is the preprocessing and calibration of the datasets. The study on Jakarta, for example, finds that DMSP data fail to capture intra-urban heterogeneity in brightness, such as the lighting patterns around key economic facilities and transport hubs, even after applying Pareto adjustment methods [10]. In contrast, VIIRS data, with their improved spatial resolution and dynamic range, provide a more nuanced representation of urban economic activity, particularly in large and complex metropolitan areas. This highlights the necessity of preprocessing techniques that account for top-coding, sensor saturation, and spatial blurring effects to ensure that NTL data accurately reflect the distribution of consumer spending across different urban zones [18].

Furthermore, the integration of NTL data with other high-resolution urban indicators enhances predictive accuracy. In Dalian, China, the use of Luojia-1 satellite imagery, combined with Point of Interest (POI) data and OpenStreetMap (OSM) road networks, allows for the classification of urban functional zones—such as commercial, residential, and industrial areas—based on their light intensity and spatial distribution [28]. This multi-scale and multi-source approach demonstrates that urban areas exhibit distinct lighting patterns that correlate with consumer behavior, infrastructure development, and economic specialization. For example, commercial and transportation areas are characterized by high average brightness (\(L_{avr}\)), whereas industrial areas show a mix of over-bright and under-bright blocks depending on their location relative to major transport corridors and urban centers [28].

The predictive power of NTL data in urban settings is also reinforced by their ability to capture short-term economic fluctuations. In India, for instance, NTL data combined with electricity consumption metrics were used to estimate a 24% Year-over-Year GDP contraction during the early stages of the pandemic, closely matching the official 23.9% decline [9]. This suggests that NTL data, when properly calibrated and integrated with complementary indicators, can serve as a timely and accurate tool for monitoring consumer-spending shifts in urban centers, particularly in the context of sudden economic disruptions.

However, while NTL data are more effective in urban areas than in rural ones, their utility is not absolute. The study on India notes that NTL data may fail to capture certain sectors of urban economic activity that operate primarily during daylight hours or are less reliant on visible infrastructure [9]. Additionally, in urban areas with high levels of informal economic activity, such as those found in some developing economies, NTL data may still underrepresent certain aspects of consumer behavior [10]. These limitations emphasize the need for context-specific modeling approaches and the integration of NTL with other data sources—such as electricity consumption, transportation flows, and socio-economic indicators—to improve the accuracy of consumer-spending predictions.

In summary, urban areas are more conducive to the use of NTL data as a leading indicator of consumer-spending behavior due to the concentration of economic activity and infrastructure. The effectiveness of NTL data is contingent on the sensor type (VIIRS outperforming DMSP), preprocessing techniques, and integration with complementary data sources. While NTL data demonstrate strong predictive capabilities in urban environments, particularly for industrial and service sectors, they require careful calibration and contextual interpretation to account for variations in economic structure and activity patterns. The following section will examine the application of NTL data in rural areas, where the challenges of modeling consumer behavior are more pronounced due to lower population density and informal economic activity.

4.2. Rural Areas¶

Rural areas pose distinct challenges and opportunities in the application of satellite-derived night-light intensity (NTL) as a leading indicator for regional consumer-spending shifts. The sparse economic activity and uneven distribution of infrastructure in these regions often result in lower light intensity and reduced variability in NTL data compared to their urban counterparts, which can obscure the relationship between luminosity and economic behavior. For example, a study on Sub-Saharan Africa notes that the standard deviation of NTL values is smaller in rural areas, indicating less variability in light intensity and thus a weaker signal for economic activity [22]. This is further compounded by the prevalence of informal economies, where economic transactions are not reflected in official statistics or lighting infrastructure, reducing the explanatory power of NTL for consumer behavior modeling [22].

The effectiveness of NTL data in rural contexts is also influenced by the type of sensor and preprocessing techniques employed. Research in Colombia shows that VIIRS data, with their higher spatial resolution (approximately 750 m) and dynamic range, outperform DMSP data in capturing economic activity at the municipal level, even in rural areas [21]. However, the application of the Global Urban Footprint—a preprocessing tool that enhances the accuracy of NTL in urban settings—can paradoxically reduce the reliability of rural RDP estimates, highlighting the need for region-specific preprocessing strategies [21]. Additionally, the study on Indian GDP during the early pandemic demonstrates that a 90-day averaging window improves the ability of NTL to capture short-term economic shifts, particularly in rural or informal economic contexts where transient variations in economic activity are more pronounced [9].

Beyond sensor and preprocessing considerations, the integration of NTL with ground-level economic indicators is crucial for improving the accuracy of consumer-spending models in rural regions. In rural Sub-Saharan Africa, for instance, the household wealth index derived from the Demographic and Health Surveys (DHS) is often used as a complementary metric to validate NTL-based economic estimates [22]. This integration helps account for unobserved economic activity that is not captured by luminosity data, particularly in regions where electrification levels are low and informal economic transactions dominate [22]. Similarly, the study on Colombia notes that combining NTL data with population density and infrastructure availability enhances the model fit, particularly for rural municipalities where the relationship between light and economic activity is less direct [21].

The spatial and temporal resolution of NTL data plays a pivotal role in their applicability to rural economic modeling. The study on China’s construction land utilization introduces the Nighttime Light Development Index (NLDI), which measures the alignment of light intensity with population density. In rural regions, particularly in the central and western parts of China, the NLDI reflects significant development disparities, where low-intensity lighting coexists with high land development density [19]. This suggests that in rural areas, NTL may not capture the full extent of economic activity due to underdeveloped infrastructure and low electrification rates. The study also notes that while urbanization and infrastructure development in rural regions have led to a gradual decline in NLDI values, the relationship between NTL and economic development remains complex and context-dependent [19].

Moreover, the literature emphasizes the need for a context-aware analytical framework when applying NTL data in rural regions. The study on Indian GDP highlights that NTL data must be interpreted alongside other socio-economic and infrastructural variables, such as electricity consumption, population density, and proximity to power lines, to account for the indirect drivers of consumer behavior [9]. In rural Sub-Saharan Africa, where economic data are sparse and unreliable, the integration of NTL with mobile phone ownership and market access indicators has been proposed as a means of improving the predictive accuracy of consumer-spending models [6]. These findings underscore the importance of a multi-source, context-sensitive approach in rural settings, where the relationship between NTL and economic activity is less direct and more influenced by non-observable factors.

Despite these methodological considerations, the utility of NTL data in rural regions remains limited by several inherent challenges. The study on India notes that atmospheric noise—such as cloud cover—can introduce measurement errors in VIIRS data, particularly in regions with low light intensity [9]. Additionally, the distinction between low-intensity artificial lighting and natural reflectivity (e.g., from snow or water surfaces) is a persistent issue in rural NTL analysis, as it can obscure the true signal of human economic activity [11]. These limitations necessitate the use of advanced filtering and calibration techniques to isolate meaningful economic signals from environmental and technical noise [11].

In summary, while satellite-derived NTL data have the potential to serve as a leading indicator for consumer-spending shifts in rural regions, their effectiveness is contingent on the choice of sensor, preprocessing methods, and integration with complementary data sources. The weaker correlation between NTL and economic activity in rural areas, as compared to urban contexts, highlights the need for context-aware modeling strategies that account for informal economic structures, low electrification rates, and spatial heterogeneity. The following section will explore the integration of NTL data with high-frequency complementary data sources, such as mobile phone usage and social media activity, to address these limitations and enhance the accuracy of consumer-spending models in both urban and rural settings.

5. Integration with Complementary Data Sources¶

The integration of satellite-derived night-light intensity data with complementary information sources—such as mobile phone records, social media activity, and point-of-interest (POI) data—offers a robust framework for enhancing the accuracy and responsiveness of consumer-spending forecasts. These data types provide high-frequency, spatially rich insights into human movement, interaction, and economic engagement, which can refine the interpretation of luminosity patterns observed in NTL datasets. Particular attention is given to the methodological considerations of spatial and temporal alignment, preprocessing, and contextual calibration necessary to ensure meaningful integration across diverse environments. This section first examines the role of mobile phone data in augmenting NTL-based models, highlighting their potential and associated challenges.

5.1. Mobile Phone Data¶

Mobile phone data represent a high-frequency and geographically rich complementary source for tracking consumer behavior, particularly in data-scarce environments where traditional economic indicators are either unavailable or insufficiently granular. The potential of mobile phone data lies in their ability to capture real-time movement patterns, communication dynamics, and location-based interactions, all of which can serve as proxies for economic activity and consumer spending. When integrated with satellite-derived night-light (NTL) data, mobile phone data can provide additional context for interpreting luminosity patterns, especially in urban and semi-urban regions where consumer behavior is more spatially distributed and temporally variable.

One of the key advantages of mobile phone data is their temporal granularity, often available at daily or even hourly intervals. This high-frequency nature complements the typically monthly or quarterly resolution of NTL data, offering a more immediate and dynamic view of consumer activity. For example, in the context of urban economic modeling, location requests derived from social media platforms have been used as a proxy for human population dynamics and subsequently linked to NTL data to better understand the spatial and temporal variations in economic behavior [4]. Although the study in question did not use mobile phone data directly, the methodology is analogous, suggesting that mobile phone data—such as call detail records (CDRs) or location-based services—could similarly be leveraged to enrich NTL-based consumer-spending models.

Moreover, mobile phone data can help disentangle the influence of non-economic factors on NTL signals. In rural Sub-Saharan Africa, for instance, the relationship between NTL and consumer spending is often muted due to the informal nature of economic activity and low electrification rates [22]. Mobile phone data, which reflect human movement and communication, can serve as a more direct proxy for economic engagement in such settings. By combining NTL with mobile phone-derived metrics—such as the number of unique devices detected in a given area or the frequency of location-based interactions—researchers can capture a more comprehensive picture of economic activity that accounts for informal transactions and transient labor patterns.

A critical consideration in the integration of mobile phone data with NTL is the alignment of spatial and temporal scales. Mobile phone data are typically available at the granularity of individual locations or aggregated mobility flows, while NTL data are captured at the pixel or administrative level. To bridge this gap, spatial interpolation techniques and kernel density estimation can be employed to aggregate or distribute mobile phone activity in a manner compatible with NTL datasets [26]. Additionally, preprocessing steps such as anonymization, aggregation, and filtering for outliers are essential to ensure data quality and ethical compliance [26]. These methods are particularly relevant for urban environments, where the spatial heterogeneity of consumer behavior is more pronounced and where the integration of fine-grained mobility data can enhance the predictive accuracy of NTL-based models [10].

The methodological complexity of integrating mobile phone data with NTL is further underscored by the need for contextual calibration. In North American urban case studies, the predictive power of NTL data has been shown to depend on the choice of sensor (e.g., VIIRS) and preprocessing techniques [10]. Mobile phone data, when integrated into these models, can provide additional calibration points that account for variations in infrastructure access, population density, and consumer behavior. For instance, in the case of Alibaba’s field experiments on last-mile station optimization, the authors demonstrated how customer-level and station-level data could be used to refine targeting strategies and improve the effectiveness of offline interactions, which in turn influence online consumer behavior [26]. While the study focused on retail logistics, its methodological approach highlights the value of integrating location-based data with economic indicators to capture the spatial and behavioral nuances of consumer activity.

However, the integration of mobile phone data with NTL also presents challenges. One of the primary limitations is the potential for spatial misalignment, particularly when using mobile phone data at the individual level. The use of high-resolution NTL data from VIIRS, which have a spatial resolution of approximately 750 m, can mitigate this issue to some extent [11]. Nevertheless, the spatial resolution of mobile phone data varies depending on the level of aggregation, and mismatches between the two datasets can lead to information loss or biased estimates. To address this, researchers must employ techniques such as spatial overlay, buffer analysis, or geospatial clustering to align mobile phone activity with NTL signals [7].

Additionally, the temporal alignment between mobile phone data and NTL data requires careful consideration. For example, mobile phone activity may be more sensitive to short-term events—such as festivals, policy changes, or natural disasters—than NTL data, which tend to reflect longer-term trends in infrastructure and economic activity [9]. This temporal asymmetry can be leveraged to create more responsive models that capture both the immediate and sustained effects of economic shifts on consumer behavior. In the context of India’s early pandemic period, the study demonstrated how combining NTL with electricity consumption and precipitation data improved the ability to capture economic downturns [9]. A similar approach could incorporate mobile phone data to further refine the model’s sensitivity to real-time behavioral changes.

From a methodological perspective, the integration of mobile phone data with NTL can benefit from advanced statistical and machine learning techniques. For instance, generalized random forests and causal inference methods have been used in field experiments to estimate heterogeneous treatment effects based on geographic and contextual factors [26]. These techniques can be adapted to consumer-spending modeling by treating mobile phone data as a high-frequency variable that influences the relationship between NTL and economic outcomes. Furthermore, the dominance-based rough set approach (DRSA) has been proposed as a method for modeling consumer preferences in data-scarce environments, and its application could be extended to scenarios where mobile phone data are integrated with NTL to improve model fit and interpretability [12].

In conclusion, mobile phone data offer significant potential as a complementary source for tracking consumer behavior, particularly when integrated with satellite-derived NTL data. Their high-frequency nature, spatial richness, and ability to capture human movement and interaction patterns make them a valuable addition to economic modeling frameworks. However, the integration requires careful preprocessing, spatial and temporal alignment, and context-aware calibration to ensure that the combined data accurately reflect consumer-spending dynamics. The following section will explore the integration of NTL with another high-frequency data source—social media and point-of-interest data—to further expand the analytical toolkit for modeling consumer behavior.

5.2. Social Media and Point-of-Interest Data¶

The integration of satellite-derived nighttime light (NTL) data with high-frequency social media and point-of-interest (POI) data has emerged as a promising avenue for capturing real-time consumer behavior and spending trends. Social media platforms, particularly those with geotagged user activity, provide insights into human mobility, visitation patterns, and consumer preferences, while POI data delineate the spatial distribution of commercial and service-oriented locations. Together, these datasets offer a complementary perspective to NTL data, enhancing their ability to reflect the spatial and temporal dynamics of economic activity.

A notable example of this integration is the study conducted in China using Tencent’s social media-derived location requests (NLR) alongside VIIRS Day-Night Band (DNB) data [4]. At the provincial level, the linear relationship between NTL and NLR explained up to 68% of inter-regional variance in human activity. However, the relationship became less pronounced at finer spatial scales—such as city and county levels—where the influence of settlement patterns and population distribution altered the correlation between light intensity and consumer behavior. At the pixel level, the predictive power dropped to 33% of the nationwide variation in NLR, underscoring the impact of spatial heterogeneity on the effectiveness of such integrations. The study also found that similar NTL values could correspond to different magnitudes of human activity, as exemplified by the contrasting NLR figures in Guangdong and Jiangsu provinces. This variability highlights the necessity of contextual calibration when combining NTL with social media and POI data.

The methodological approach in this integration involved the use of quantile regression and Moran’s I statistics to account for spatial autocorrelation in both NTL and NLR data. Positive spatial autocorrelation was observed at city and county levels, with both NTL and NLR showing significant clustering (I = 0.20–0.26, P-value < 0.01). This finding reinforces the importance of spatial dependence in modeling consumer behavior and suggests that ignoring such autocorrelation can lead to biased or inefficient estimates. The proposed partition algorithm further leveraged the co-distribution of NTL and NLR to characterize the spatial structure of human settlements, particularly in relation to city size, form, and urbanization level. This joint perspective allows for a more nuanced understanding of how luminosity patterns reflect underlying consumer behavior, particularly in urbanized regions where social media activity is more concentrated and POI data are more granular.

In addition to geotagged social media data, POI datasets have been used to enrich NTL-based models by providing a spatial inventory of commercial and service activities. For instance, in Dalian, China, researchers combined Luojia-1 satellite imagery with POI data and OpenStreetMap (OSM) road networks to classify urban functional zones [28]. The analysis revealed that commercial and transportation hubs exhibited high average brightness (\(L_{avr}\)), while industrial areas showed a mix of lighting patterns depending on their proximity to urban centers. These findings suggest that POI data can help contextualize NTL signals by linking luminosity to the types of economic activities that drive consumer spending. Furthermore, the integration of POI data with NTL can improve the accuracy of models by capturing the spatial distribution of consumption points, such as retail outlets, restaurants, and entertainment venues, which are often concentrated in areas of high light intensity.

The use of social media and POI data is particularly relevant in urban consumer-spending modeling, where these datasets can provide insights into transient or event-driven shifts in economic activity. For example, during the early stages of the pandemic in India, the integration of NTL with electricity consumption and precipitation data was used to estimate GDP fluctuations [9]. A similar approach could incorporate social media-derived mobility data to capture real-time behavioral shifts, such as reduced foot traffic in retail districts or increased activity in home-based consumption. This would allow for a more responsive model that captures both the infrastructure-based and behavior-based dimensions of consumer spending.

However, the integration of NTL with social media and POI data is not without challenges. One of the primary concerns is the need for spatial and temporal alignment between the datasets. Social media activity and POI locations are often available at fine spatial granularities, whereas NTL data are typically captured at the pixel or administrative level. Techniques such as spatial overlay and kernel density estimation are required to harmonize these datasets, ensuring that the combined analysis reflects meaningful patterns of consumer behavior [7]. Additionally, the preprocessing of NTL data must account for sensor-specific artifacts, such as stray light and auroral interference, which can distort the relationship between luminosity and social media or POI activity [11].

The effectiveness of this integration also depends on the quality and representativeness of the social media and POI data. In rural or informal economic contexts, where social media usage may be limited or POI data may not capture small-scale economic activities, the predictive power of these datasets is reduced [22]. In such cases, the integration of NTL with social media and POI data must be supplemented with other complementary indicators—such as mobile phone ownership or transportation flows—to ensure a comprehensive understanding of consumer behavior. This is particularly relevant for regions in Sub-Saharan Africa, where the informal economy dominates and traditional economic data are sparse.

In summary, the integration of satellite-derived NTL data with social media activity and POI datasets provides a powerful framework for analyzing real-time consumer behavior and spending trends. By combining luminosity patterns with behavioral and location-based data, researchers can capture the spatial and temporal nuances of economic activity that are often missed by NTL alone. However, the success of this integration depends on careful preprocessing, spatial alignment, and contextual calibration to address the limitations of both NTL and the complementary data sources. The next section will examine the broader implications of these integrations, focusing on how they can be leveraged to refine consumer-spending models in both urban and rural environments.

6. Contextual Variations¶

The utility of satellite-derived night-light intensity as a leading indicator for consumer-spending shifts is not uniform across regions and is strongly influenced by the socioeconomic and political context in which the data are collected and interpreted. This section examines how the effectiveness of night-light data varies in relation to economic structure and political factors, both of which shape the visibility of economic activity and the reliability of consumer behavior inferences drawn from luminosity patterns. Specifically, the following subsections explore the differential performance of night-light indicators in service-based urban economies versus informal and agriculture-dependent rural contexts, and analyze how governance, policy interventions, and regulatory frameworks further modulate these relationships.

6.1. Economic Structure and Development¶

The economic structure of a region significantly influences the relationship between satellite-derived night-light intensity (NTL) and consumer spending. In urbanized and service-based economies, where economic activity is closely linked to infrastructure and energy use, NTL data tend to exhibit stronger correlations with consumer behavior due to the visible and spatially concentrated nature of such activity. Conversely, in manufacturing or agriculture-based economies—particularly in rural and informal economic contexts—NTL data often provide a weaker or less reliable signal for consumer-spending patterns. These variations are attributable to differences in the types of economic activity, the spatial distribution of lighting infrastructure, and the degree of formalization in the local economy.

Empirical evidence supports this distinction. In China, for example, night-light data demonstrate a more robust relationship with GDP in industrial and services sectors compared to the primary sector, implying that NTL is more indicative of consumer behavior in urban centers where services and retail activities are more prominent [10]. Similarly, in India, NTL data combined with electricity consumption proved effective in capturing GDP fluctuations during the early stages of the pandemic, particularly in urban regions where energy use is tightly coupled with economic and consumer activity [9]. This suggests that in service-driven urban economies, NTL data can act as a reasonably accurate proxy for consumer spending, especially when integrated with high-frequency indicators such as electricity demand.

However, in rural and informal economic contexts, the relationship between NTL and consumer behavior is often weaker or less linear. A study on rural Sub-Saharan Africa found that while NTL data are positively correlated with household wealth indicators—such as ownership of assets like bicycles and improved sanitation—these relationships are limited in their ability to capture the full complexity of consumer spending in low-electrification regions [17]. The authors note that in such settings, economic activity is frequently unrecorded and less reliant on electric lighting, leading to an underestimation of consumer behavior when using NTL as a standalone proxy. The same limitations were observed in the analysis of Colombian municipalities, where DMSP data showed a negative and statistically insignificant relationship with Regional Domestic Product (RDP) in rural areas, while VIIRS data provided more reliable estimates [21].

Sensor design and preprocessing techniques further modulate the effectiveness of NTL data in capturing consumer behavior across different economic structures. The paper by Gibson et al. (2021) highlights that VIIRS data, with their higher spatial resolution and dynamic range, are more suitable for capturing economic activity in rural and informal settings compared to DMSP data, which suffer from top-coding and spatial blurring [10]. While VIIRS data improve the ability to detect low-intensity lighting and differentiate between economic zones, the predictive power for consumer behavior remains limited in rural areas, particularly when the economy is informal or unrecorded. These findings underscore the importance of context-aware preprocessing and the integration of NTL with complementary data sources—such as electricity consumption or population density—to enhance the accuracy of consumer-spending models in rural and informal economies.

In agricultural or primary-sector economies, NTL data may capture indirect economic activity rather than direct consumer behavior. For instance, the study on the relationship between NTL and GDP in 169 countries found that agricultural output had a weaker association with light consumption than industrial or service-sector activity [15]. The authors suggest that this is due to the fact that agricultural activity does not generate significant nighttime lighting, and thus, NTL data may not fully reflect the underlying economic and consumer behavior in these contexts. This limitation is particularly relevant for developing countries where primary-sector employment is high and consumer behavior is not as closely tied to infrastructure or artificial illumination.

Moreover, the economic structure of a region interacts with the temporal resolution of NTL data to influence their predictive power for consumer behavior. In low-density rural areas, the study on India demonstrated that a 90-day averaging window improved the ability of NTL to capture economic shifts, suggesting that longer time horizons may be necessary to smooth out the irregularities introduced by informal economic activity [9]. This implies that in regions with a less stable or formalized economic structure, the integration of NTL with complementary data sources—such as mobile phone activity, transportation patterns, or electricity consumption—can provide more reliable insights into consumer spending dynamics.

In summary, the effectiveness of NTL data as a leading indicator for consumer-spending shifts is contingent upon the economic structure of the region. Urban and service-based economies tend to exhibit a stronger and more direct relationship between NTL and consumer behavior, while manufacturing and agriculture-based economies—especially in rural and informal contexts—require additional context-aware modeling and integration with complementary data sources to improve predictive accuracy. The next section will explore the influence of political and policy factors on the interpretation and utility of NTL data in consumer-spending modeling.

6.2. Political and Policy Factors¶

Political and policy factors significantly shape the utility and interpretability of satellite-derived night-light (NTL) data as a leading indicator for consumer-spending shifts. The effectiveness of NTL in capturing economic behavior is influenced by governance structures, regulatory environments, and policy interventions—particularly those that alter the spatial and temporal patterns of human activity. For example, large-scale policy shocks such as lockdowns, stimulus programs, and infrastructure investments can either amplify or obscure the relationship between NTL and consumer spending, depending on how they affect the visibility of economic activity through lighting patterns.

A notable case is the impact of lockdown policies during the early stages of the COVID-19 pandemic in India, where NTL data were integrated with electricity consumption to estimate GDP contractions. The study found that while NTL data captured a 24% Year-over-Year (YoY) GDP decline, this relationship was nuanced by the fact that many services continued to operate during non-lit hours, and manufacturing shifted to daytime activity [9]. This illustrates that policy-induced changes in operational hours and behavioral patterns can alter the traditional link between NTL and consumer behavior, necessitating context-specific modeling approaches.

Similarly, in rural Sub-Saharan Africa, where informal economic activity is prevalent, policy interventions such as electrification programs or rural development initiatives can directly influence the visibility of economic activity through increased light intensity. For instance, a study using NTL data in conjunction with household wealth indices from the Demographic and Health Surveys (DHS) found that the correlation between NTL and economic indicators in rural areas was weaker compared to urban settings [17]. This discrepancy may be attributed to the limited formal infrastructure in rural economies, where policy-driven electrification and improved access to basic services are key factors in enhancing the signal captured by NTL data. Therefore, understanding the political and institutional context—such as the presence of electrification policies or informal sector support—is essential for accurate consumer-spending modeling.

The regulatory environment also plays a role in shaping the availability and quality of complementary data sources that are often used in conjunction with NTL. For example, the integration of mobile phone data with NTL has been hampered in some countries by data privacy laws and restrictions on data sharing [7]. These regulatory barriers can limit the ability to refine NTL-based models with real-time behavioral indicators, particularly in rural or informal economic contexts where such data may be the most valuable. Additionally, the lack of standardized methodologies for preprocessing and interpreting NTL data in politically volatile or data-scarce regions can introduce further uncertainty. For instance, in conflict-affected countries, the rapid deterioration and subsequent recovery of lighting patterns may be more accurately captured by NTL data than official economic statistics, but such data must be interpreted with caution in the absence of stable governance and consistent policy frameworks [25].

Moreover, policy interventions that directly affect economic behavior, such as stimulus packages or targeted social assistance programs, can influence the spatial distribution of consumer activity and, by extension, the patterns observed in NTL data. A study on the uptake of food grain distribution in India used NTL data as a proxy for economic development to match subdistricts and estimate the impact of agent choice on program participation [24]. The results demonstrated that NTL data helped account for unobserved factors such as digital infrastructure and proximity to administrative centers, which are themselves shaped by policy decisions. This underscores the importance of considering the policy landscape when interpreting NTL data, as government-led initiatives can alter the relationship between light intensity and economic activity.

The influence of political stability on NTL data utility is also evident in regions experiencing governance disruptions. For example, in conflict-affected areas of Sub-Saharan Africa, the study on industrial growth found that NTL data were better at capturing recovery trends when compared to other countries at similar developmental stages [20]. However, such comparisons rely on the assumption that governance structures and policy environments are comparable, which is often not the case. This highlights the need for caution in using NTL data as a standalone indicator in politically unstable regions, where policy fragmentation or the informalization of economic activity may distort the relationship between luminosity and consumer behavior.

In sum, political and policy factors are critical in determining how effectively NTL data can serve as a leading indicator for consumer-spending shifts. Policy interventions, regulatory frameworks, and governance structures shape both the visibility of economic activity through artificial lighting and the availability of complementary data sources that can enhance NTL’s predictive power. These influences are particularly pronounced in informal and rural economies, where policy decisions can directly alter the economic landscape. The next section will examine the limitations and challenges inherent in the use of NTL data, focusing on technical and interpretive constraints that affect their application across diverse contexts.

7. Limitations and Challenges¶

The use of satellite-derived night-light intensity (NTL) as a leading indicator for regional consumer-spending shifts is promising yet fraught with limitations that must be carefully addressed to ensure the validity and reliability of economic models. This section examines the key constraints associated with NTL data, focusing on technical and interpretive challenges that affect their utility in capturing consumer behavior across diverse economic contexts. Technically, issues such as spatial and temporal resolution, sensor design, calibration inconsistencies, and noise variability significantly influence the accuracy of NTL as a proxy for economic activity. Interpretively, the absence of ground-truth data at sub-national levels, the impact of non-economic factors on luminosity, and the varying relevance of NTL across different economic structures further complicate its use. These limitations set the stage for a detailed exploration of the technical challenges that underpin the reliability of NTL data in subsequent subsections.

7.1. Technical Limitations¶

The technical limitations of satellite-derived night-light (NTL) data present significant challenges to their use as a proxy for regional consumer-spending shifts. These limitations primarily stem from issues related to spatial and temporal resolution, sensor design, noise, and calibration inconsistencies, which affect the accuracy and reliability of NTL-based economic models.

One of the most pressing concerns is the spatial resolution of NTL datasets. Older datasets, such as the Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS), have a spatial resolution of approximately 1 km, which is insufficient for capturing fine-grained economic variations in heterogeneous regions, particularly in rural or informal economic contexts where economic activity is less concentrated and more dispersed [10]. In contrast, the Visible Infrared Imaging Radiometer Suite (VIIRS) offers a higher spatial resolution of around 750 m, allowing for more accurate identification of localized economic activity and consumer behavior patterns [21]. However, even with improved resolution, VIIRS data may still struggle to distinguish between low-intensity artificial lighting and natural reflectivity, particularly in low-light environments or regions with significant natural light contamination, such as snow-covered areas [11]. This limitation underscores the need for preprocessing techniques that can isolate meaningful economic signals from environmental noise, particularly in rural areas where the relationship between lighting and consumer spending is weaker [9].

Another critical limitation lies in sensor design and data calibration. DMSP satellites were originally designed for weather monitoring, not for socioeconomic analysis, and their data suffer from issues such as top-coding, blurring, and lack of on-board calibration, all of which reduce their accuracy in capturing the full range of economic activity [10]. Top-coding, for instance, causes the brightest pixels to be saturated, limiting the ability to detect differences in economic intensity among developed urban areas [10]. While Pareto-adjustment methods have been proposed to address this issue, their effectiveness is limited, particularly in urban centers where spatial inequality is high. In Jakarta, for example, DMSP data failed to reflect intra-urban heterogeneity in brightness from key economic facilities even after applying such adjustments [10]. VIIRS data, on the other hand, were designed with consistent radiance measurement in mind and offer a dynamic range covering nearly seven orders of magnitude, making them more suitable for sub-national economic modeling [10]. Nevertheless, the transition from DMSP to VIIRS introduces a discontinuity in the data record, as the two platforms differ in overpass times and sensor specifications. This mismatch can bias trend analyses and complicate the interpretation of NTL data in contexts where temporal consistency is crucial [11].

Noise and variability further undermine the reliability of NTL data in modeling consumer behavior. Atmospheric factors such as cloud cover, lunar illumination, and stray light can introduce significant noise into NTL observations, particularly in regions with low light intensity where the signal-to-noise ratio is already compromised [9]. For example, in India during the early stages of the pandemic, NTL data required a 90-day averaging window to reduce noise and improve the accuracy of GDP estimates, compared to the more commonly used 30-day window [9]. Similarly, in rural Sub-Saharan Africa, the low standard deviation of NTL values in sparsely populated areas suggests that the data may not adequately capture the full spectrum of economic activity, especially in informal economies where lighting is not directly tied to consumer behavior [22]. This variability necessitates the development of advanced preprocessing methods, including cloud masking, lunar correction, and spatial smoothing techniques, to enhance the signal quality and ensure that NTL data can effectively reflect consumer-spending dynamics [14].

The temporal resolution of NTL data is also a key factor in their utility for consumer-spending modeling. While DMSP data are typically available as annual composites, VIIRS data offer monthly or even sub-monthly observations, enabling more frequent and timely tracking of economic behavior [11]. However, the usefulness of high temporal resolution depends on the stability and continuity of the data stream. For instance, the Indian case study demonstrated that a longer averaging window improved the detection of economic shocks, suggesting that in regions with high variability in economic behavior, the temporal structure of NTL data must be carefully considered to avoid underrepresentation of key dynamics [9]. Additionally, the limited temporal span of some newer satellite platforms, such as SDGSAT-1, remains a constraint for long-term modeling of consumer behavior [10].

Lastly, the integration of NTL data with complementary sources, such as electricity consumption or social media activity, introduces further technical complexities. While such integrations can enhance the predictive power of consumer-spending models, they require careful alignment of spatial and temporal scales between the datasets [7]. For example, the application of the Global Urban Footprint (GUF) mask, while improving model fit in large cities, can distort rural economic estimates, highlighting the need for region-specific preprocessing approaches [21]. Similarly, the combination of NTL data with high-frequency social media-derived location requests has shown strong predictive power at the provincial level but weakens at finer spatial scales due to the influence of settlement patterns and population distribution [4]. These findings suggest that while NTL data can serve as a valuable component of a multi-source modeling framework, their integration must be guided by rigorous preprocessing and calibration to avoid misalignment and bias in consumer-spending estimates.

In light of these technical limitations, the use of NTL data as a leading indicator for consumer-spending shifts must be approached with caution. The choice of sensor, the application of preprocessing techniques, and the integration with complementary data sources are all critical to improving the accuracy and reliability of NTL-based economic models. The next section will explore the interpretive challenges that arise from these technical constraints, particularly in the context of urban and rural economic structures.

7.2. Interpretive Challenges¶

Interpretive challenges in the use of satellite-derived night-light intensity (NTL) as a proxy for consumer-spending shifts arise primarily from the difficulty in interpreting the data in the absence of ground-truth economic data and the potential for misinterpretation due to non-economic factors that influence luminosity. These challenges are particularly pronounced in rural and informal economic contexts, where the relationship between NTL and economic activity is not as direct or reliable as in urbanized regions. The interpretation of NTL data is further complicated by the fact that lighting patterns can reflect a range of non-economic phenomena, including seasonal variations in electricity usage, changes in public infrastructure, and shifts in population distribution that are unrelated to actual consumer behavior [10].

A key interpretive challenge is the lack of ground-truth data at the sub-national level, particularly in low-income and rural regions. Without reliable, high-frequency, and spatially granular socioeconomic data—such as consumer expenditure surveys or retail transaction records—NTL data must be interpreted in isolation or in combination with limited alternative indicators. This can result in biased or incomplete inferences, especially when economic activity is not directly tied to visible light emissions. For instance, in rural Sub-Saharan Africa, where informal economic activity is the norm, the absence of comprehensive consumer behavior data limits the ability to validate NTL-based models of spending shifts [22]. This issue is compounded by the fact that NTL data are typically aggregated at the provincial or municipal level, making it difficult to isolate the contribution of consumer behavior from other factors such as agricultural output or infrastructure development.

Another significant interpretive limitation is the influence of non-economic factors on NTL intensity, which can lead to misattribution of economic changes. For example, cloud cover and atmospheric interference can obscure the true signal of economic activity, particularly in low-light rural areas where the signal-to-noise ratio is already low [9]. Additionally, sensor-specific artifacts, such as top-coding and blurring, can distort the perceived intensity of light in urban and rural areas alike, leading to overestimation or underestimation of economic activity [10]. The study on Jakarta, for instance, highlights how DMSP data fail to capture intra-urban heterogeneity in brightness, even after applying Pareto adjustment methods [10]. This suggests that the interpretation of NTL data must be guided by an understanding of sensor limitations and their potential to misrepresent the actual distribution of consumer activity.

Furthermore, the economic structure of a region plays a critical role in determining how NTL data should be interpreted. In service-based urban economies, where consumer behavior is closely aligned with visible infrastructure and energy use, NTL data can serve as a relatively accurate leading indicator [10]. However, in agricultural or manufacturing-based economies—particularly in rural and informal settings—NTL data may reflect indirect economic activity rather than direct consumer spending. For example, in regions where primary-sector employment dominates, the absence of significant nighttime lighting makes it difficult to use NTL as a proxy for consumption patterns, necessitating the integration with other data sources to improve model accuracy [15]. This underscores the importance of context-aware modeling, where NTL data are not interpreted in isolation but are instead contextualized within the broader economic and infrastructural landscape.

The temporal dynamics of NTL data also introduce interpretive complexities. While the literature has demonstrated that NTL data can detect short-term economic shocks—such as those caused by lockdowns or policy changes—the timing and duration of such effects may not align with the temporal resolution of NTL datasets. For instance, the Indian case study found that a 90-day averaging window was more effective than a 30-day window in capturing the full economic impact of the pandemic [9]. This suggests that the interpretation of NTL data for consumer behavior must consider the time lag between economic shifts and their visibility in satellite imagery, particularly in rural and informal economies where economic activity is more transient and less infrastructure-dependent.

In conclusion, the interpretive challenges associated with NTL data highlight the need for a nuanced and context-specific approach to modeling consumer-spending shifts. The absence of ground-truth data, the influence of non-economic factors, and the variability in economic structure across regions all contribute to the risk of misinterpretation. These limitations reinforce the importance of integrating NTL data with complementary high-frequency economic and behavioral indicators to enhance the accuracy and reliability of consumer-spending models. The following section will provide a summary of the findings presented thus far, setting the stage for a discussion of future research directions.

8. Conclusion and Future Directions¶

This study has systematically evaluated the utility of satellite-derived night-light (NTL) intensity as a leading indicator for regional consumer-spending shifts, examining the methodological, technical, and contextual factors that influence its effectiveness. The findings confirm that NTL data are most reliable in urbanized and service-based economies, where the spatial concentration of economic activity aligns with visible lighting patterns. In these settings, high-resolution sensors such as VIIRS, combined with appropriate preprocessing techniques and integration with complementary data sources like electricity consumption and social media activity, significantly enhance the predictive accuracy of NTL-based models. For example, the case study on India during the early stages of the pandemic demonstrated that NTL data could closely approximate GDP contractions, with a 24% YoY decline predicted and validated against official statistics [9].

In contrast, the application of NTL data in rural and informal economic contexts remains methodologically complex. The lower population density, uneven infrastructure, and the prevalence of unrecorded economic activity in such regions result in weaker correlations between NTL and consumer spending. For instance, in rural Sub-Saharan Africa, the relationship between NTL and household wealth is indirect and influenced by factors such as electrification rates and informal economic structures [22]. The study in China’s central and western rural areas further illustrates that NTL may reflect development disparities but does not fully capture the economic reality, particularly when infrastructure and lighting are underdeveloped [19]. These findings highlight the necessity of context-aware modeling and the integration of NTL with alternative indicators—such as mobile phone data, transportation metrics, or household wealth indices—to overcome the limitations of luminosity as a standalone proxy for consumer behavior.

The integration of NTL data with high-frequency complementary sources has also been explored, with notable methodological insights emerging. Mobile phone data, for instance, provide real-time behavioral and mobility patterns that can enrich NTL-based models, particularly in rural and informal economies where traditional economic indicators are unreliable [26]. Similarly, the use of social media location data and POI information has shown promise in urban settings, where the spatial distribution of economic activity is more visible and structured [4]. However, such integrations require careful preprocessing, spatial alignment, and contextual calibration to ensure that the combined data accurately reflect consumer behavior. The application of kernel density estimation, geospatial clustering, and quantile regression has been identified as essential for harmonizing NTL with these high-frequency data sources [7].

Despite these advancements, the use of NTL data as a leading indicator for consumer-spending shifts is constrained by both technical and interpretive challenges. Sensor limitations, such as top-coding and spatial blurring in DMSP data, and calibration inconsistencies between different satellite platforms, introduce noise and bias into the data record [10]. Additionally, the influence of non-economic factors—such as lunar illumination, cloud cover, and seasonal variations in lighting—must be carefully accounted for in preprocessing and modeling stages [11]. The absence of granular, sub-national consumer-spending data further complicates the validation of NTL-based models, particularly in low-income and data-scarce regions [10]. These challenges underscore the importance of methodological refinement and the development of robust preprocessing pipelines that enhance the signal-to-noise ratio and align NTL data with the temporal and spatial characteristics of consumer behavior.

Looking forward, future research should focus on three primary directions to further refine the utility of NTL data in economic forecasting and consumer-spending modeling. First, the development of hybrid models that integrate NTL with high-frequency behavioral and infrastructural data—such as mobile phone usage, transportation flows, and social media engagement—offers a promising path to improving the accuracy and responsiveness of consumer behavior estimates. These models should be tested across a range of economic contexts to determine their generalizability and effectiveness in capturing both short-term shocks and long-term trends in consumer behavior.

Second, the expansion of NTL-based analysis to underrepresented and data-scarce regions is essential for broadening the applicability of the method. While the literature has primarily focused on urbanized and industrialized settings, rural and informal economies remain underserved in terms of reliable consumer-spending indicators. Future studies should explore how NTL data, when combined with alternative proxies such as household wealth indices or informal economic activity metrics, can be adapted to better reflect the subtleties of consumer behavior in such contexts. This includes the development of region-specific calibration techniques and the application of advanced statistical and machine learning methods to account for the heterogeneity of economic structures.

Third, the refinement of preprocessing and calibration methodologies is a critical area for further research. As demonstrated in this study, the effectiveness of NTL data depends heavily on how they are processed, particularly when transitioning between different satellite platforms or when integrating them with non-NTL data sources. Future work should explore the development of standardized preprocessing pipelines that are adaptable to diverse economic and geographic conditions, ensuring the long-term consistency and reliability of NTL data for consumer-spending modeling. This includes the evaluation of machine learning-based preprocessing techniques, such as convolutional neural networks, for improving the alignment and calibration of NTL datasets [16].

In sum, while satellite-derived NTL data offer a powerful tool for tracking consumer-spending shifts, their utility is contingent on addressing a range of methodological and contextual challenges. The findings of this study emphasize the importance of sensor selection, preprocessing, and data integration in enhancing the accuracy of NTL-based models. Future research should build on these insights by developing more sophisticated modeling frameworks and expanding the application of NTL data to diverse economic contexts. The next section will provide a more detailed discussion of the implications of these findings for academic and policy research.

8.1. Summary of Findings¶

Satellite-derived nighttime light (NTL) intensity has shown variable effectiveness as a leading indicator for regional consumer-spending shifts, with the most consistent results observed in urbanized and service-based economies. In such settings, NTL data—particularly those from high-resolution sensors like VIIRS—can reliably capture shifts in consumer behavior when paired with appropriate preprocessing and integration with complementary data such as electricity consumption and social media activity. For example, in the Indian case study, NTL data combined with electricity usage and precipitation metrics enabled a 24% YoY GDP decline estimate, closely aligning with official statistics [9]. This demonstrates the potential of NTL to serve as a near-real-time proxy for consumer behavior, especially in urban centers where economic activity is more visibly linked to artificial lighting patterns. VIIRS data, with their enhanced spatial resolution and dynamic range, have proven superior to DMSP/OLS data in capturing these patterns, particularly when advanced preprocessing methods such as kernel density aggregation and logarithmic transformations are applied [10].

In contrast, the use of NTL data in rural and informal economies remains methodologically complex due to the weak and often non-linear relationship between light intensity and consumer spending. In rural Sub-Saharan Africa, for instance, the standard deviation of NTL values is smaller, indicating limited variability that correlates with economic activity [22]. This is attributed to the prevalence of unrecorded economic transactions and the low reliance on electric lighting in daily interactions. Similarly, in rural regions of China, NTL data reflect development disparities but fail to fully capture the nuances of consumer behavior, especially in areas with underdeveloped infrastructure and low electrification rates [19]. These findings emphasize the necessity of integrating NTL with alternative data sources—such as household wealth indices and informal economic indicators—to improve the accuracy of consumer-spending models in such settings [17].

The integration of NTL with high-frequency complementary data sources has emerged as a key methodological advancement in modeling consumer behavior. Mobile phone data, for example, provide real-time behavioral and mobility insights that can enhance the predictive power of NTL in both urban and rural contexts [26]. In urban areas, the combination of NTL with social media-derived location requests and point-of-interest (POI) data has shown strong correlations with consumer behavior at the provincial level [4]. However, the relationship weakens at finer spatial scales due to the influence of settlement patterns and population distribution, underscoring the need for region-specific spatial alignment and preprocessing techniques [7]. In rural regions, where traditional economic data are often limited, these complementary data sources offer a more direct means of capturing economic engagement, particularly when integrated with NTL data using tailored calibration approaches [7].

The alignment of spatial and temporal resolutions between NTL and complementary data sources remains a critical factor in the success of these integrations. Techniques such as spatial overlay, buffer analysis, and geospatial clustering are essential for harmonizing datasets and ensuring that the combined models accurately reflect consumer-spending dynamics [7]. Additionally, the application of advanced statistical methods—such as generalized random forests, quantile regression, and geographically weighted regression—has been shown to improve the predictive power of NTL-based models by accounting for spatial heterogeneity and non-linear relationships [5]. These methodological refinements are particularly relevant in regions with complex economic structures, where the interplay between luminosity patterns and consumer behavior is not straightforward.

The utility of NTL data as a leading indicator for consumer spending is further influenced by the economic and political context. In service-based urban economies, where consumer activity is closely tied to infrastructure and energy use, NTL data can serve as a reasonably accurate early signal of economic change [10]. However, in agricultural or manufacturing-based economies—especially in rural and informal settings—NTL data may reflect indirect economic activity rather than direct consumer behavior. Policy interventions such as lockdowns, electrification programs, and stimulus initiatives can also alter the visibility of economic activity through lighting patterns, thereby affecting the relationship between NTL and consumer spending [9]. These findings highlight the necessity of context-aware modeling strategies that account for governance structures, policy environments, and economic specialization when interpreting NTL data.

In summary, satellite-derived NTL data offer a valuable, albeit context-dependent, leading indicator for consumer-spending shifts, particularly in urban and service-based economies where the spatial concentration of economic activity aligns with visible lighting patterns. In rural and informal settings, their utility is limited and requires integration with alternative data sources to improve model accuracy. The combination of NTL with high-frequency behavioral and infrastructural indicators—such as mobile phone data, social media activity, and household wealth metrics—provides a refined approach to modeling consumer behavior. However, the success of these integrations depends on careful preprocessing, spatial alignment, and contextual calibration. These findings underscore the importance of a multidimensional and context-sensitive approach in leveraging NTL data for economic forecasting, particularly in diverse and data-scarce environments. The next section will outline key future research directions to further refine and expand the application of NTL in consumer behavior modeling.

8.2. Future Research Directions¶

Future research should focus on three key areas to refine the use of satellite-derived night-light (NTL) data as a leading indicator for consumer-spending shifts. These directions aim to address the methodological and contextual limitations identified in the literature and empirical case studies, with particular emphasis on sensor selection, preprocessing, and integration with high-frequency data.

First, the development of context-aware hybrid modeling frameworks that combine NTL with high-frequency behavioral and infrastructural data—such as mobile phone usage, social media activity, and transportation flows—offers a promising path to enhancing the accuracy and responsiveness of consumer-spending models. These integrations are particularly valuable in data-scarce environments, where traditional economic indicators are either unreliable or unavailable. For example, mobile phone data can provide real-time insights into human movement and economic engagement, while social media data can capture shifts in consumer behavior through location-based activity. The success of these integrations depends on rigorous spatial and temporal alignment, as well as preprocessing strategies that harmonize the data to reflect meaningful economic signals [7]. The application of advanced statistical and machine learning techniques—such as generalized random forests, quantile regression, and geographically weighted regression—should be expanded to better account for spatial heterogeneity and non-linear relationships [5].

Second, the refinement of preprocessing and calibration methodologies is essential for improving the reliability of NTL data, especially in transitioning between sensor platforms such as DMSP and VIIRS. As demonstrated in the Indian case study, the use of a 90-day averaging window improved the detection of economic shocks compared to the standard 30-day window [9]. Similarly, kernel density aggregation, logarithmic transformations, and machine learning-based approaches—such as convolutional neural networks—have shown promise in aligning datasets and improving the signal-to-noise ratio [16]. Future research should prioritize the development of standardized yet adaptable preprocessing pipelines, particularly for rural and informal economies, where the signal is often weak and non-economic noise is more prevalent [11]. These pipelines must account for sensor-specific artifacts, such as top-coding and blurring, to ensure the consistency and accuracy of NTL data over time and across regions.

Third, sensor selection and its impact on predictive accuracy should be a central focus for future studies, particularly in urban environments where consumer behavior is more structured and visible. The literature has consistently shown that VIIRS data outperform DMSP data in capturing sub-national economic activity due to their higher spatial resolution and dynamic range [21]. However, the effectiveness of VIIRS data in rural and informal settings remains limited, necessitating the development of region-specific calibration techniques that can enhance the signal of low-intensity lighting [17]. The development of context-aware modeling approaches that incorporate governance, policy, and economic specialization as key variables will further improve the interpretability and robustness of NTL-based models. These methodological refinements are critical for advancing the use of NTL data in both academic research and policy applications, particularly in regions where traditional economic data are limited.

In conclusion, the future of NTL-based consumer-spending modeling lies in the development of context-sensitive hybrid models, the refinement of preprocessing techniques, and the strategic selection of sensors tailored to specific economic and geographic conditions. These directions align with the broader need for a multidimensional approach to economic forecasting, particularly in rural and informal economies. The next section will outline the broader implications of these findings for academic and policy research, emphasizing the importance of integrating technical and contextual insights to improve the utility of NTL data in consumer behavior analysis.

References¶

1. Measuring Quarterly Economic Growth from Outer Space in: IMF Working Papers Volume 2022 Issue 109 (2022). Available at: https://www.elibrary.imf.org/view/journals/001/2022/109/article-A001-en.xml (Accessed: August 29, 2025)
2. Estimation of consumption potentiality using VIIRS night-time light data. Available at: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0206230 (Accessed: August 29, 2025)
3. https://www.sciencedirect.com/science/article/pii/S0264275125000265. Available at: https://www.sciencedirect.com/science/article/pii/S0264275125000265 (Accessed: August 29, 2025)
4. https://www.mdpi.com/2072-4292/10/7/1128. Available at: https://www.mdpi.com/2072-4292/10/7/1128 (Accessed: August 29, 2025)
5. Léopold Simar, Valentin Zelenyuk, Shirong Zhao. (2024). Inference for aggregate efficiency: Theory and guidelines for practitioners. European Journal of Operational Research.
6. Energy planning in Sub-Saharan African countries needs to explicitly consider productive uses of electricity. Available at: https://www.nature.com/articles/s41598-023-40021-y (Accessed: August 29, 2025)
7. Anastasia Griva, Eleni Zampou, Vasilis Stavrou, Dimitris Papakiriakopoulos, George Doukidis. (2024). A two-stage business analytics approach to perform behavioural and geographic customer segmentation using e-commerce delivery data. Journal of Decision Systems.
8. Nighttime Lights. Available at: https://www.earthdata.nasa.gov/learn/backgrounders/nighttime-lights (Accessed: August 29, 2025)
9. Using satellite images of nighttime lights to predict the economic impact of COVID-19 in India. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC9128334/ (Accessed: August 29, 2025)
10. https://www.sciencedirect.com/science/article/abs/pii/S0304387820301772. Available at: https://www.sciencedirect.com/science/article/abs/pii/S0304387820301772 (Accessed: August 29, 2025)
11. A harmonized global nighttime light dataset 1992–2018. Available at: https://www.nature.com/articles/s41597-020-0510-y (Accessed: August 29, 2025)
12. Koen W. De Bock, Kristof Coussement, Arno De Caigny, Roman Słowiński, Bart Baesens, Robert N. Boute, Tsan-Ming Choi, Dursun Delen, Mathias Kraus, Stefan Lessmann, Sebastián Maldonado, David Martens, María Óskarsdóttir, Carla Vairetti, Wouter Verbeke, Richard Weber. (2024). Explainable AI for Operational Research: A defining framework, methods, applications, and a research agenda. European Journal of Operational Research.
13. https://www.sciencedirect.com/science/article/pii/S0924271623001521. Available at: https://www.sciencedirect.com/science/article/pii/S0924271623001521 (Accessed: August 29, 2025)
14. Understanding the association between urban noise and nighttime light in China. Available at: https://www.nature.com/articles/s41598-024-83125-9 (Accessed: August 29, 2025)
15. https://www.sciencedirect.com/science/article/abs/pii/S0034425713000734. Available at: https://www.sciencedirect.com/science/article/abs/pii/S0034425713000734 (Accessed: August 29, 2025)
16. A global annual simulated VIIRS nighttime light dataset from 1992 to 2023. Available at: https://www.nature.com/articles/s41597-024-04228-6 (Accessed: August 29, 2025)
17. Nighttime lights as a proxy for human development at the local level. Available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6124706/ (Accessed: August 29, 2025)
18. Night-Time Light Data: A Good Proxy Measure for Economic Activity?. Available at: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0139779 (Accessed: August 29, 2025)
19. Nighttime light development index: a new evaluation method for China’s construction land utilization level. Available at: https://www.nature.com/articles/s41599-025-04626-0 (Accessed: August 29, 2025)
20. https://www.researchgate.net/publication/335147374_Industrial_Growth_in_Sub-Saharan_Africa_Evidence_from_Machine_Learning_with_Insights_from_Nightlight_Satellite_Images. Available at: https://www.researchgate.net/publication/335147374_Industrial_Growth_in_Sub-Saharan_Africa_Evidence_from_Machine_Learning_with_Insights_from_Nightlight_Satellite_Images (Accessed: August 29, 2025)
21. https://www.sciencedirect.com/science/article/pii/S235293852100183X. Available at: https://www.sciencedirect.com/science/article/pii/S235293852100183X (Accessed: August 29, 2025)
22. Shedding light on development: Leveraging the new nightlights data to measure economic progress. Available at: https://pmc.ncbi.nlm.nih.gov/articles/PMC11790135/ (Accessed: August 29, 2025)
23. https://www.mdpi.com/2072-4292/13/11/2067. Available at: https://www.mdpi.com/2072-4292/13/11/2067 (Accessed: August 29, 2025)
24. Rakesh Allu, Maya Ganesh, Sarang Deo, Sripad K. Devalkar. (2024). Technology-Enabled Agent Choice and Uptake of Social Assistance Programs: Evidence from India’s Food Security Program. Manufacturing & Service Operations Management.
25. https://www.imf.org/en/Publications/fandd/issues/2019/09/satellite-images-at-night-and-economic-growth-yao. Available at: https://www.imf.org/en/Publications/fandd/issues/2019/09/satellite-images-at-night-and-economic-growth-yao (Accessed: August 29, 2025)
26. Brian Rongqing Han, Tianshu Sun, Leon Yang Chu, Lixia Wu. (2024). Connecting Customers and Merchants Offline: Experimental Evidence from the Commercialization of Last-Mile Stations at Alibaba. MIS Quarterly.
27. Night-time lights: A global, long term look at links to socio-economic trends. Available at: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174610 (Accessed: August 29, 2025)
28. Nighttime light intensity and brightness suitability in urban functional zones. Available at: https://www.nature.com/articles/s41598-025-10378-3 (Accessed: August 29, 2025)