基于GRACE数据重建的黄河流域实际蒸散发及其时空演变特征分析OA北大核心CSTPCDEI

Multiple deep learning methods were used to interpolate gravity recovery and climate experiment(GRACE)data,and the random forest algorithm was used to spatially downscale GRACE data.The actual evapotranspiration in the Yellow River Basin was calculated based on the water balance equation.And the data were verified using four evapotranspiration products to analyze the spatio-temporal evolution of actual evapotranspiration in the Yellow River Basin.The results indicate that the overall interpolation accuracy of the long short-term memory neural network is superior to that of deep neural network and convolutional long short-term memory neural network.The average correlation coefficient between the actual evapotranspiration estimated based on GRACE data and four evapotranspiration products is 0.903,indicating that the applicability of the actual evapotranspiration results estimated based on GRACE data is good.The average annual actual evapotranspiration in the Yellow River Basin from 2003 to 2021 was 144.38 to 775.62 mm,with a spatial distribution pattern of more in the south and less in the north,and a seasonal variation pattern of more in summer and less in winter.From 2003 to 2016,it increased at a rate of 2.51 mm/a,and showed a downward trend after 2017.