National-scale mapping locates the mismatch between perceived and assessed flood risk in India

Pratyush Tripathy, Surabhi Upadhyay, Okikiola Michael Alegbeleye, Emine Senkardesler, Deepika Pingali, Emily Goddard, Jagadish Thaker, Anthony Leiserowitz, Jennifer Marlon 

Flood risk management assumes that people in hazardous places recognize the danger they face, but evidence on flood risk perception remains sparse in South Asia, home to much of the world's flood-exposed population. India has more than 600 million people exposed to inland flood hazards and the highest flood-related mortality in Asia, yet no national evidence base on flood risk perception. Here we combine a nationally representative survey of 10,704 Indian adults with a Multilevel Machine Learning with Poststratification (MMLP) approach to estimate flood risk perception for all 633 districts, and compare these against assessed hazard from the flood record. Perceived and assessed risk diverge by more than 30 percentage points in over half of districts, well beyond the uncertainty in our estimates. Districts that most sharply under-perceive risk are concentrated in Assam, Kerala, and Andhra Pradesh, while over-perception is widespread across northern, central, and northeastern states. These mismatches follow state boundaries, with state explaining nearly three times as much of the variance as district-level hydrology, climate, education, vulnerability, and news combined. This makes flood risk perception a measurable input for adaptation planning, and provides a method transferable to any country where hazard is well measured but perception is not.