基于长短记忆模型的鄱阳湖流域径流模拟及其演变的归因分析

Simulation and attribution analysis based on the long-short-term-memory network for detecting the dominant cause of runoff variation in the Lake Poyang Basin

气候变化和人类活动直接或间接的影响着全球和区域水文循环过程,是导致水文水资源时空分布的主要因素,同时也是流域-湖泊水文情势变化的根本原因.本文基于长短记忆模型构建了鄱阳湖气象-径流模型,同时引入了基准期的概念,定量区分了导致鄱阳湖流域径流变化的主要影响因素.研究结果表明:在同时考虑计算效率和模拟效果的前提下,采用10 d预测窗口大小来构建鄱阳湖气象-径流模型能够很好地捕捉径流的极值,并且对径流的短期波动也能有很好的体现.训练期模型在各个子流域的纳什效率系数均高于0.94,而在验证期,模型在各个子流域的纳什效率系数均高于0.90.基于径流模拟结果,定量区分了人类活动和气候变化对鄱阳湖径流的影响,研究结果显示:人类活动对径流的影响主要发生在春、秋季,其中,人类活动在春季主要会造成径流的增加,平均增加幅度约为139.47 m^(3)/s,而在秋、冬季,人类活动则会导致径流平均减少约34.37 m^(3)/s.对比二者的相对贡献率,可以发现,春季人类活动对径流造成的影响较大,平均相对贡献率为77.26%.而在其余季节,鄱阳湖流域径流过程的改变主要是由于气候变化,平均相对贡献率约75.84%.研究结果能够为鄱阳湖流域水资源管理提供科学依据和理论指导.

Climate change and human activities directly or indirectly affect the global and regional hydrologic cycle,which is the main factor leading to the temporal and spatial distribution of water resources.Based on the long-short-term-memory(LSTM)model,this paper attributes the contribution of climate change and human activities on runoff variation in the Lake Poyang Basin.Results show that the runoff process was simulated well by LSTM framework with a window size of 10 days for both accuracy and computational efficiency.The overall performance of the model was good with Nash-Sutcliffe Efficiency varied from 0.94 to 0.95 in the training period and 0.90 to 0.98 in the test period,respectively.Based on the simulation results,the contribution of human activities and climate change on runoff change in Lake Poyang Basin was then quantified.Results show that the runoff increased by 139.47 m^(3)/s due to human activities in spring,accounting for 77.26%of runoff change.Whereas climate change was the dominant factor in changing runoff in the other seasons,which led the runoff to decrease by 34.37 m^(3)/s,accounting for about 75.84%of runoff changes.The results can provide scientific basis and theoretical guidance for water resources management in Lake Poyang Basin.