Urban Flood Exposure Across Three Cities: A Reproducible Open-Satellite Comparison and a Pre-Registered Test of Its Blind Spot

This is a self-archived preprint. It has not been peer reviewed and has not been published in a journal. Cross-city comparison of urban flood exposure is obstructed by incompatible national hazard maps and census geographies. This study applies one identical open-satellite procedure to Dakar (Senegal), Managua (Nicaragua) and Ulsan (Republic of Korea) within 20 km radii. Flood-prone land was identified from a 30 m digital elevation model as terrain lying within 5 m of the local minimum elevation with slope below 5 degrees, excluding permanent surface water. Population was taken from a 100 m gridded product and summed at its native resolution. Exposed population fractions were 25.9% (Dakar), 6.7% (Ulsan) and 0.6% (Managua). Absolute values varied by up to a factor of seven with threshold choice, whereas the rank order was invariant across all five configurations tested (Spearman rho = 1.000). The Managua result is interpreted as a failure of the method rather than a property of the city. That interpretation was converted into three numerical predictions with explicit falsifying conditions, committed to a public repository before execution, and tested on eight recurrently flooding cities separated only by terrain: all four steep cities returned below 1.5% and all three flat cities above 19%, with rho = -0.857 (p = 0.007). A secondary warm-season surface urban heat island analysis reproduced the same lesson through an unrelated indicator: two of three cities failed an acceptance criterion fixed before the run because their vegetated reference lies 128 m and 176 m above their built-up area, so the metric was partly measuring relief. Identifying such blind spots under criteria specified in advance, rather than the exposure values themselves, is the principal contribution of this work.