A Data-Driven Analysis of Streamflow Pattern Recognition and Seasonal Transition Changes

Abstract

Seasonal streamflow transitions play a critical role in water resource management, particularly in supporting flood prevention and drought mitigation. However, understanding how these transitions shift under climate variability remains limited, especially when conventional methods rely on fixed-calendar metrics or station-based trends. This study introduces a time series clustering framework that integrates dynamic time warping (DTW) and hierarchical clustering analysis (HCA) with change-point detection and trend decomposition to capture evolving intra-annual flow patterns and seasonal transitions. By analysing transition timing within each flow pattern group, the approach moves beyond static classification to uncover climate sensitivity that is often masked in basin-aggregated results. Applied to long-term inflow records from four major reservoirs in Taiwan, the analysis reveals both spatial and pattern-conditioned changes in wet-season onset. At Shihmen Reservoir, the station-based trend suggests a general advancement in transition timing. However, when hydrologic years are grouped by flow patterns influenced by climate drivers, some clusters indicate delays linked to late-season typhoons, while others show earlier transitions associated with frontal rainfall. This contrast illustrates how aggregated trends can obscure flow-type-specific responses to climate variability. The proposed framework offers a flexible and transferable means of diagnosing intra-annual hydrological variability. It provides practical tools for adaptive water management and planning in regions facing intensifying seasonal uncertainty and hydrometeorological extremes.