Kayode Adeniyi
AI Research Engineer
Mapping floods where there is no ground truth
Most geospatial AI workflows quietly assume there's a gauge somewhere, a validated flood extent, or a land cover layer you can trust. In the catchments I work in, there usually isn't. This talk is about what you do instead: validating a flood model against community memory and damage surveys, why borrowed land cover schemas fall apart on reclaimed land and stilt settlements, and what changes when you're mapping for people who'll act on the map in hours rather than for a paper.
