An integrated ArcGIS Python toolbox for SCS-CN event runoff estimation and flood screening in data-scarce Ethiopian basins

Abstract

Flood early warning in data-scarce regions is constrained by sparse gauging and limited modelling capacity. We present an integrated ArcGIS Python toolbox automating SCS-CN event runoff estimation, from curve number generation and per-event antecedent moisture classification through to threshold calibration, evaluated in a semi-arid Ethiopian basin. From four quality-controlled stations, 587 rainfall events (2010–2023) were extracted; 98.6% occurred under dry antecedent conditions, corroborated by ERA5-Land soil moisture ( p < 0.001). Evaluation used nineteen documented flood windows from disaster databases and official reports, since discharge records are unavailable. Basin-mean runoff discriminated documented floods with AUC = 0.91 (95% CI 0.83–0.97; Cohen’s d = 2.6), rising to 0.95 for the eleven events triggered within the gauged area. A calibrated 1.25 mm runoff threshold detected thirteen of sixteen floods with approximately 62% fewer false alarms than the regional 30 mm rainfall criterion at equal detection, confirmed out-of-sample by temporal hold-out. Discrimination proved robust to the initial abstraction ratio and ± 10% curve number bias, whereas absolute thresholds were not, making local calibration essential. Limited gauge coverage and a small documented-event sample are acknowledged limitations, quantified via satellite-precipitation comparison and bootstrap intervals. The toolbox provides a transparent, reproducible flood-screening framework for data-scarce semi-arid basins.