Comparison of regional flood frequency analysis methods for ungauged catchments in West Africa

Abstract

The estimation of the return levels of floods is constrained by sparse and quality-limited hydrological observations in West Africa, even though floods remain among the most damaging natural hazards in the region. Regional Flood Frequency Analysis (RFFA) provides a pathway to estimate design floods at ungauged catchments, yet the diversity of available approaches calls for a systematic comparison. We assess whether flood quantiles can be reliably regionalized across West Africa using an unprecedented dataset of 211 near-natural catchments. This study compare a Direct Regression Approach (DRA) with three index-flood methods based on spatial proximity, Principal Component Analysis (PCA), and Canonical Correlation Analysis (CCA), all of which are implemented using both statistical and machine-learning models. Evaluation of model performance using relative bias (rBias) and mean absolute relative error (MARE) indicates that index-flood-based approaches consistently outperform DRA. Among all combinations, the CCA–SVR framework achieves the highest accuracy (rBias = -0.03; MARE = 0.21) for both 20- and 50-year flood quantiles. These findings provide robust guidance for flood design in data-scarce environments and support more resilient flood risk management across West Africa.