基于多源卫星数据推求稀缺资料地区的径流过程

Estimation of runoff process in data sparse region based on multi-source satellite data

以金沙江上游流域为例,使用多源卫星数据,基于水量平衡方程估算产流量,采用CaMa-Flood水动力模型进行汇流计算,分别采用人工神经网络(artificial neural networks,ANN)、广义回归神经网络(generalized regression neural network,GRNN)和长短期记忆神经网络(long short term memory,LSTM)模型进行水文模拟。结果表明:(1)陆地水储量变化(terrestrial water storage anomaly,TWSA)主导了水量平衡径流估计值的总体不确定性;(2)CaMa-Flood模型具有较高的模拟精度,相关系数、纳什效率系数和水量相对误差分别为0.92、0.77和-0.63%;(3)LSTM模型在率定期和验证期间的径流模拟结果均好于ANN和GRNN模型。基于多源卫星数据估算径流过程,为稀缺资料地区的水资源开发利用和保护提供了一条新的途径。

By taking the upper Jinsha River basin as an example,this study used multi-source satellite data to estimate runoff yield based on water balance equation,and applied the CaMa-Flood hydrodynamic model to simulate confluence process.The artificial neural networks(ANN),generalized regression neural network(GRNN)and long short term memory(LSTM)models were calibrated for further hydrological simulations.The results show that:(1)terrestrial water storage anomaly(TWSA)dominates the overall uncertainty of water balance runoff estimates;(2)the CaMa-Flood model simulates runoff accurately,the correlation coefficient,Nash efficiency coefficient and relative error are 0.92,0.77 and-0.63%,respectively;(3)the LSTM model outper-forms the ANN and GRNN models during the calibration and validation periods.Estimation of runoff process based on multi-source satellite data provides a new apporoach for development,utilization and protection of water resources in poorly gauged catchments.