Temporal Lag and Response Characteristics of Runoff and Its Components in a Glacial Basin

Glaciers are indicators of climate change, and the study of climate change impacts on runoff and its components in glacial basins is a research hotspot worldwide. The formation mechanisms of runoff and its components are affected by multiple factors, and relevant studies remain scarce in glacial regions due to limited data resulting from natural and objective conditions. To identify the driving factors of runoff and its components in the Vakhsh River basin of Tajikistan, we selected the suitable baseflow separation method and obtained datasets of total runoff, baseflow and surface runoff. Furthermore, the impacts and temporal patterns of hydroclimatic variables on runoff and its components were quantitatively evaluated using satellite remote-sensing products, reanalysis datasets and in-situ observations through cross-wavelet transform and explainable machine learning approaches, specifically Categorical Boosting (CatBoost) coupled with SHapley Additive exPlanations (SHAP) analysis. Results show that, among nine evaluated baseflow separation methods, the Chapman–Maxwell method demonstrates superior performance. Whether in the dry season or the wet season, the effects of snow-related variables and the Normalized Difference Vegetation Index (NDVI) on runoff and its components are significant. Compared with the dry season, the rankings of total evaporation (ET) and surface solar radiation (SR) decrease in the wet season, while the ranking of total precipitation (Prec) and snow depth (SD) increases significantly. The periodic oscillation in the dry season is stronger than that in the wet season. The time lag of surface runoff is the shortest, that of total runoff follows closely, and that of baseflow is the longest. The time lags are within 18 d in the dry season and within 12 d in the wet season. Compared with the dry season, the time lags of snow-related variables (e.g., SD, snow depth water equivalent (SDWE)) in the wet season advance by approximately 2–4 d; those of Prec advance by roughly 3–4 d; those of PSM and ET advance by about 5–6 d; and that of SR advance by approximately 4–5 d. Underlying surface and climatic variations cause uncertainty and errors in time lags identified by cross-wavelet transform, which we clarify. The results of this study provide a scientific basis for addressing climate change in glacial basins.