Data for 'Evaluating the reliability of news-derived flood data in a data-sparse region: the case of West Africa'

This deposit contains the derived data and figures supporting the paper "Evaluating the reliability of news-derived flood data in a data-sparse region: the case of West Africa". The study evaluates how reliably the Groundsource news-derived flood dataset records verified floods across fifteen West African countries over 2018 to 2025. Using 171 reference floods from the Global Disaster Alert and Coordination System (GDACS), it tests what determines whether a real flood is recorded, comparing the location and urban setting of the flood, its own severity and impact, and the national conditions of the country in which it occurred. The central finding is that recording is governed by where a flood happens rather than how severe it is: every measure of how urban a location is strongly predicts whether a flood was captured, while the flood's fatalities, displaced population, and other characteristics do not, and no national characteristic explains the variation. News-derived flood data therefore under-represent remote and rural floods in a predictable, distance-dependent way, regardless of severity. The deposit includes the analysis-ready event table (one row per reference flood, with the recorded or not-recorded outcome and every covariate used), the per-country recall and national-covariate outputs, the sampled location features, and the figures. A README and a data dictionary document the files and their columns. The source datasets used in the analysis (Groundsource, GDACS, Africapolis, WorldPop, VIIRS night-time lights, and the travel-time surface) are not redistributed here and are cited by their own DOIs in the README. The analysis code is available from the  author on request.