基于深度神经网络联合AMSR2和MODIS数据估算全球蒸散发研究

Global Evapotranspiration Estimation Study based on AMSR2 and MODIS Data

地表蒸散发(ET)是水循环和能量循环的关键组成部分,具有极其重要的应用价值。研究旨在发展一种可靠且高效的深度神经网络(DNN)模型,基于MODIS可见光数据、微波AMSR^(2)亮度温度和数字高程DEM,实现全天候全球高分辨率每日ET的估算。利用FLUXNET和AmeriFlux通量网6种代表性土地覆盖类型的148个站点观测数据来训练和验证DNN模型,结果表明:DNN模型可以有效建立卫星数据(MODIS、AMSR2数据)与ET之间的关系;6种地类的ET估算结果验证的平均绝对误差(MAE)为0.16—0.63 mm/d,均方根误差(RMSE)为0.27—0.89 mm/d,除裸地的决定系数(R^(2))为0.37以外,其他地类的R^(2)均>0.7。通过对比模型估算的ET与MOD16A2和GLEAM的ET产品,结果表明3种产品的ET空间分布特征相似,ET值非常接近,估算得到的全球2020年日均ET为0—4 mm/d。

The surface Evapotranspiration(ET)is a key component of the water and energy cycles and has critical value for applications.This study is aimed to develop a reliable and efficient Deep Neural Network(DNN)model for all-weather global daily ET estimation with high spatial resolution,using remote sensing MODIS datasets,microwave AMSR2 brightness temperature products and digital elevation DEM data as input.The study used 148 site observations over six representative land cover types from FLUXNET and AmeriFlux to train and validate DNN models.The results showed that the DNN model can effectively established the relationship between satellite(MODIS,AMSR2)data and ET,and the Mean Absolute Error(MAE)of ET estimation results for the six land cover types ranged from 0.16 to 0.63 mm/day,and the Root Mean Square Error(RMSE)ranged from 0.27 to 0.89 mm/day,and the coefficient of determination(R^(2))of all land types were>0.7,except for bare land,where the R^(2)was 0.37.By comparing the ET estimation in this study with the ET products of MOD16A2 and GLEAM,the results demonstrated that the spatial distribution characteristics of the three ET products were similar and the ET values were very close with global average daily ET of 0~4 mm/day over 2020.