Adaptive Modeling of Satellite-Derived Nighttime Lights Time-Series for Tracking Urban Change Processes Using Machine Learning

TL;DR

The paper tackles the challenge of tracking urban change across diverse cities using daily VIIRS nighttime lights by learning city-specific baseline dynamics with unsupervised neural-network forecasts. It develops a multi-model forecasting framework (FCNN, CNN, LSTM) whose predictions are ensembled to form a robust baseline; urban change is detected as deviations from this baseline, with rich metrics (severity, direction, start/end rates, and model-consensus confidence) used to characterize the changes. Demonstrated across ten study areas and three change drivers (disasters, conflicts, urbanization), the approach achieves high recall and precision, all without labeled change data, and supports continuous, near-real-time monitoring. This scalable methodology enables interpretable, data-driven tracking of urban infrastructure changes and can be extended to higher-resolution data and additional urban-change contexts to support decision-making in disaster response and urban development planning.

Abstract

Remotely sensed nighttime lights (NTL) uniquely capture urban change processes that are important to human and ecological well-being, such as urbanization, socio-political conflicts and displacement, impacts from disasters, holidays, and changes in daily human patterns of movement. Though several NTL products are global in extent, intrinsic city-specific factors that affect lighting, such as development levels, and social, economic, and cultural characteristics, are unique to each city, making the urban processes embedded in NTL signatures difficult to characterize, and limiting the scalability of urban change analyses. In this study, we propose a data-driven approach to detect urban changes from daily satellite-derived NTL data records that is adaptive across cities and effective at learning city-specific temporal patterns. The proposed method learns to forecast NTL signatures from past data records using neural networks and allows the use of large volumes of unlabeled data, eliminating annotation effort. Urban changes are detected based on deviations of observed NTL from model forecasts using an anomaly detection approach. Comparing model forecasts with observed NTL also allows identifying the direction of change (positive or negative) and monitoring change severity for tracking recovery. In operationalizing the model, we consider ten urban areas from diverse geographic regions with dynamic NTL time-series and demonstrate the generalizability of the approach for detecting the change processes with different drivers and rates occurring within these urban areas based on NTL deviation. This scalable approach for monitoring changes from daily remote sensing observations efficiently utilizes large data volumes to support continuous monitoring and decision making.

Figures (6)

• Figure 1: Workflow of the proposed methodology to train the baseline temporal variation in NTL time-series between successive windows. The trained models are then applied to unseen test phase NTL observations where urban change processes are detected based on its deviation from model prediction. All steps of the approach are applied on an area weighted 1-D NTL time-series.
• Figure 2: Interpreting NTL time-series and derived metrics using the proposed approach. Example of (a) pre- and (b) post-disaster (Hurricane Maria) images of Caguas, Puerto Rico due to infrastructure damage resulting in power outage. (c) Rich temporal information captured through NTL time-series. By comparing the NTL observation at each time-step with the ensemble prediction, change metrics on severity, deviation, change rate and recovery inferences are extracted. Prediction error also shows different temporal stages in this urban area, namely (i) pre-change baseline, (ii) change, (iii) continuing recovery, (iv) full recovery to pre-change levels.
• Figure 3: Power outages from natural hazards as seen through NTL time-series in the four study areas. This change type produces abrupt drop with high change severity (as shown in the shaded area) followed by recovery.
• Figure 4: Impact of socio-economic changes caused by conflict as seen through NTL. Conflict is characterized by sudden negative and sustained deviation (as shown in the shaded area). Negligible recovery trends are observed and the NTL time-series stabilizes at the post-conflict levels.
• Figure 5: Changes caused by urbanization as seen through NTL time-series. Urbanization is characterized by gradual, positive deviation (as shown in the shaded area). A continuing increase in NTL is seen in the post-change onset stage.