Optimal timescale of antecedent precipitation index for flood peak prediction in data-scarce mountain river basins

Study region Two data-scarce mountain river basins in the Hanjiang River Basin, China: the Qinjiang River Basin (1578 km 2 ) and the Hengshan Basin (12,624 km 2 ). Study focus This study develops a data-driven dual-criteria framework (integrating elasticity and T statistic of correlation) to objectively optimize the antecedent timescale (AT) of the Antecedent Precipitation Index (API) for flood peak discharge prediction, addressing subjectivity in traditional API parameterization. The framework is applied to extensive historical flood events, and a Random Forest model incorporating AT-derived API and identified peak-producing rainfall is used to predict flood peak discharge. New hydrological insights for the region Quantitative analysis reveals that the hydrological impact of antecedent moisture conditions on flood peaks is maximized at specific, catchment-dependent timescales (AT opt = 16 days for Qinjiang River Basin, 13 days for Hengshan Basin), rather than the conventionally assumed 30-day period. This multivaluedness reflects fundamental differences in how basins regulate moisture storage and release; the optimal integration period is dictated by the interplay between basin physiography and the underlying runoff response mechanisms. By replacing arbitrary temporal assumptions with a dual-criteria framework that balances hydrological sensitivity and statistical robustness, this study advances the parameterization of antecedent precipitation. The resulting process-informed approach offers a transferable strategy to improve the accuracy and reliability of flood peak prediction across data-scarce mountain environments.