Item

Abstract

Due to the long memory of snow processes, statistically based seasonal streamflow prediction models in snow-dominated catchments can successfully leverage, but also typically rely on, snowpack estimates. Using mountainous catchments in central Asia as a case study, we demonstrate how seasonal hydrological forecasts benefit from incorporating large-scale climate oscillations (COs). Firstly, we examine the teleconnections between the major COs and peak precipitation season in eight catchments across the Pamir Mountains and the Tian Shan from February to June. We then employ a machine learning (ML) framework that incorporates snow water equivalent (SWE) and dominant CO indices as predictors for mean discharge from April to September. Our workflow leverages an ensemble technique with multiple SWE estimates from near-time global data sources and diverse types of explainable machine learning models. We find that the winter states of the El Niño–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO) enhance SWE-based forecasts of seasonal discharge in the study catchments. We identify three instances in which the inclusion of COs as additional predictors could be instrumental for snowpack-based seasonal streamflow forecasting: (1) when forecasts are issued at extended lead times and accumulated SWE is not yet representative of seasonal terrestrial water storage, (2) when climate variability during the forecasted season plays a larger role in shaping seasonal discharge, and (3) when SWE estimates for a catchment are subject to larger uncertainty. Our approach provides a useful way to reduce uncertainties in seasonal discharge predictions in data-scarce, snowmelt-dominated catchments.