Prediction of Monthly Flow Regimes Using the Distance-Based Method Nested with Model Swapping

Abstract

The distance-based method is effective for hydrological prediction in ungauged basins, but challenges emerge when predicting flow regimes, as flow magnitude varies with time non-monotonically. Besides, substantial prediction errors often arise in basins with varying local characteristics and dynamics compared to their neighbour basins. This paper proposed a novel two-stage prediction approach that integrates the distance-based method also addressing high-flow timing with model swapping to improve predictions of monthly flow regimes in ungauged basins. In the distance-based method, dissimilarity is assessed based on lateral (timing) and vertical (magnitude) differences of high flows besides point-to-point differences between flow regimes. Model swapping improves regional model predictions by incorporating localized information at selected sites through the classical nearest neighbour principle in the descriptors’ space. This approach was applied in a case study in northwestern Italy, where model swapping was conducted at 77 out of 124 stations. The results indicate that incorporating model swapping enhances prediction accuracy by reducing the dissimilarity between observed and predicted flow regimes by 15% compared to the distance-based method. Moreover, this approach demonstrated improved predictions of high flow (i.e., ≥ 80% of the maximum monthly flow) and its timing. These results support the potential of this integrated approach for predicting complex flow regimes and their like.