Abstract: Studies on applicability of the probability distributed model (PDM) to the river basins in China and investigation of its parameter change across different flood stages are limited. This study compares the performance of PDM-based runoff forecasting for two typical basins in Zhejiang Province, and examines model parameter changes through dynamic identifiability analysis (DYNIA) to infer dominant controlling factors in rainfall-runoff process under different time windows. The results show that forecasting performance for the two basins is generally satisfactory with the Nash-Sutcliffe efficiency coefficient both over 0.7, and the model performs better in the Longquan basin dominated by low flows than the Jinhua basin dominated by high flows. Three parameters (empirical coefficientss α, b and maximum storage capacity Smax) of various flood events are negatively correlated to antecedent soil moisture. The recession slope-related b is positively correlated to average evaporation in the Jinhua basin but negatively to mean rainfall in the Longquan basin. Identifiability of α, a coefficient controlled by soil moisture at a depth of 28 cm, is generally higher for flood peak and recession periods.

Key words: probability distributed model, dynamic identifiability analysis, runoff forecasting, identifiability, dominant controlling factors