摘要
Accurately estimating future long-term runoff is important for basin-wide water resource planning and management.However,uncertainties exist in future runoff projections due to the limitations in observations and model structures.This study,which was inspired by the recent release of high-precision global runoff reanalysis datasets,introduces a spatially informed machine learning technique that combines empirical orthogonal function(EOF)analysis and a Gaussian process regression(GPR)model,named EOF-GPR,to correct basin-wide future runoff projections by distributed global hydrological models(GHMs).The Yarlung Zangbo-Brahmaputra River basin(YBRB)is chosen as a case study,and the proposed method is applied to better analyze the runoff response to future climate change.The results demonstrate the effectiveness of the EOFGPR method in increasing the accuracy of long-term runoff projections across the entire basin.Compared with that in the historical periods(1979-2014),the average runoff during the summer in the YBRB is projected to decrease in the future period(2022-2100),whereas the average runoff in autumn and winter is projected to increase.A demarcation point is observed between the Nuxia and Bodak gauging stations,with upstream and downstream regions showing opposite trends in maximum and minimum runoff dates and flood frequency.Downstream of this point,more frequent(up to a~14%increase)but less persistent extreme flood events(up to a~84%decrease)are predicted.Similar patterns are observed for basin-wide seasonal extreme drought events under future climate change.Compared with previous studies at stations,the EOF-GPR method offers a refined understanding of runoff and flood-drought change characteristics at the spatial scale of the whole basin,which is valuable for decision-making in water resource management and flood and drought mitigation within the YBRB and other basins worldwide.
基金
supported by the National Natural Science Foundation of China(Grant No.92047302)。
