Multiscale Suitability Modelling of Flood Susceptibility in Kenya: Weighted Overlay with Uncertainty Propagation in Data-Sparse Settings

Flood risk in Kenya is conventionally assessed through single-scale weighted overlay procedures that obscure how susceptibility rankings change with the spatial unit of analysis. The framework formalises criterion weighting as a probabilistic exercise, treating expert-derived weights as distributions rather than fixed scalars, and propagates these distributions through the overlay algebra using Monte Carlo simulation. A hierarchical sensitivity structure links catchment-level, county-level and sub-county analytical units, permitting the decomposition of classification variance into components attributable to scale, criterion weight and input-data error. The method is demonstrated conceptually for Kenya, where gauge density is low and land-cover records are intermittent, using the criteria of rainfall intensity, drainage density, elevation, distance to channels, land cover, soil permeability and slope. Results indicate that susceptibility classifications are most volatile at the sub-county scale and that weight uncertainty dominates input-data error in determining classification confidence. The framework provides a diagnostic protocol for identifying scales at which flood susceptibility mapping becomes decision-reliable despite sparse ground data.