Climate-Driven Streamflow Responses in the Jinghe River Basin Under CMIP6 Scenarios Using a Regionalized EA-LSTM

Abstract

Climate change may alter both basin-outlet water availability and runoff responses within large semi-arid tributaries. We evaluated climate-driven runoff changes at four gauged catchments in the Jinghe River Basin using a 16-gauge, multi-gauge EA-LSTM. Thirty-one NEX-GDDP-CMIP6 model–grid configurations were evaluated against 1960–2014 observations from 66 stations, and four models with strong historical statistical agreement and complete scenario records formed the screened ensemble. Native NEX-GDDP-CMIP6 forcing was used directly. At the Zhangjiashan outlet, test-period NSE and KGE were 0.823 and 0.615, respectively, while PBIAS was −31.6%. Under late-century SSP585, annual runoff decreased at Jingheyuan and Zhanghe but increased at Jingcun and Zhangjiashan, ranging from −16.7% at Zhanghe to +16.8% at Jingcun; flood-season responses showed a similar upstream–downstream contrast. At Zhangjiashan, screened-ensemble means were 0.2–11.5 percentage points lower than the corresponding expanded-ensemble means across the scenario–period combinations. Within the present input set and learned relation, precipitation and runoff changes were associated (Pearson r = 0.918), although this does not establish physical attribution.