Detecting Flood-Induced Population Mobility Using Social Media and Satellite Data

Flooding is the world’s most pervasive natural hazard and is projected to intensify with ongoing socio-environmental change. Beyond the immediate damage they cause to infrastructure and livelihoods, floods can prompt disruptive short- and long-term population movements. This study quantifies and characterises population mobility in response to severe floods in Bihar, India. Bihar is a flood-prone and socio-economically vulnerable locale that experiences recurrent monsoon flooding affecting millions annually. We estimate the proportion of the population that responds to flooding events, examine the spatial and temporal characteristics of mobility (including distance travelled and timing relative to flood onset), and assess heterogeneity in responses across demographic groups (gender and age) and settlement types (urban, suburban, and rural).

We adopt a data-driven, multi-source geospatial approach centred on gridded user-count data from Meta’s Data for Good programme, which provides high-frequency proxies for population presence based on aggregated Facebook user activity.  This Facebook data offers a rich source for tracking migration and displacement in response to crises such as disease outbreaks, flooding, and tropical cyclones across the globe, particularly in low- and middle-income countries where alternative mobility data are sparse. These data are integrated with complementary datasets, including night-time lights as a proxy for electricity access and economic activity, daily river-discharge records to capture hydrological extremes, WorldPop population surfaces, Global Human Settlement Layer – Degree of urbanisation (GHSL-SMOD), and satellite-derived flood extent maps. The combined framework enables identification of both spatial and temporal mobility responses to flooding while accounting for variations in urbanisation and infrastructure.

Our results show that active Facebook user counts decline by approximately 35% during flood periods. This reduction likely reflects a combination of factors, including power and connectivity outages, evacuation and displacement, and reduced access to mobile devices. We find that the correspondence between Facebook user counts and underlying population increases monotonically with the degree of urbanisation, suggesting greater data reliability in more urban contexts. Analysis of movement flows indicates that mobility during flooding is dominated by urban-to-urban movements, followed by urban-to-suburban transitions, with comparatively limited rural outflows. Demographic analysis further reveals differential impacts across gender and age cohorts, indicating uneven exposure and adaptive capacity within affected populations. Overall, this study demonstrates the value of integrating social-media-derived mobility data with remote sensing and hydrological information to generate timely, granular insights into flood-induced population dynamics. Such evidence can support more targeted humanitarian response, infrastructure planning, and long-term resilience-building efforts in flood-prone, data-scarce regions.