Dataset of ground truth of land surface evapotranspiration at regional scale in the Heihe River Basin (2012-2016) ETMap Version 1.0

Surface evapotranspiration (ET) is an important variable that connects the land energy balance, water cycle and carbon cycle. The accurate acquisition of ET is helpful to the research of global climate change, crop yield estimation, drought monitoring, and it is of great significance to regional and global water resource planning and management. The methods of obtaining evapotranspiration mainly include ground observation, remote sensing estimation, model simulation and assimilation. The high-precision surface evapotranspiration data can be obtained by ground observation, but the spatial representation of observation stations is very limited; remote sensing estimation, model simulation and assimilation methods can obtain the spatial continuous surface evapotranspiration, but there are problems in the verification of accuracy and the rationality of spatial-temporal distribution pattern. Therefore, this study makes full use of a large number of high-precision station observation data, combined with multi-source remote sensing information, to expand the observation scale of ground stations to the region, to obtain high-precision, spatiotemporal distribution of continuous surface evapotranspiration.Based on the "Heihe River Integrated Remote Sensing joint experiment" (water), "Heihe River Basin Ecological hydrological process integrated remote sensing observation joint experiment" (hiwater), the accumulated station observation data (automatic meteorological station, eddy correlator, large aperture scintillation instrument, etc.), 36 stations (65 station years, distribution map is shown in Figure 1) are selected in combination with multi-source remote sensing data (land cover) Five machine learning methods (regression tree, random forest, artificial neural network, support vector machine, depth belief network) were used to construct different scale expansion models of surface evapotranspiration, and the results showed that: compared with The other four methods, random forest method, are more suitable for the study of the scale expansion of surface evapotranspiration from station to region in Heihe River Basin. Based on the selected random forest scale expansion model, taking remote sensing and air driven data as input, the surface evapotranspiration time-space distribution map (etmap) of Heihe River Basin during the growth season (May to September) from 2012 to 2016 was produced. The results show that the overall accuracy of etmap is good. The RMSE (MAPE) of upstream (las1), midstream (las2-las5) and downstream (las6-las8) are 0.65 mm / day (18.86%), 0.99 mm / day (19.13%) and 0.91 mm / day (22.82%), respectively. In a word, etmap is a high-precision evapotranspiration product in Heihe River Basin, which is based on the observation data of stations and the scale expansion of random forest algorithm. Please refer to Xu et al. (2018) for all station information and scale expansion methods, and Liu et al. (2018) for observation data processing.

地表蒸散量（ET）是连接陆地能量平衡、水循环与碳循环的重要变量。精确获取ET对于全球气候变化研究、作物产量估算、干旱监测具有重要意义，对于区域及全球水资源规划与管理亦具有深远影响。获取蒸散量的方法主要包括地面观测、遥感估算、模型模拟与同化。地面观测可获取高精度地表蒸散量数据，然观测站点的空间代表性极为有限；遥感估算、模型模拟与同化方法可获取连续的地表蒸散量空间分布，但在精度验证及空间时间分布模式的合理性方面存在一定问题。因此，本研究充分利用大量高精度站点观测数据，结合多源遥感信息，将地面观测站点的观测范围扩展至区域，以获得高精度、连续的地表蒸散量时空分布。基于“黑河综合遥感联合实验”（水）、“黑河流域生态水文过程综合遥感观测联合实验”（hiwater），结合自动气象站、涡度协相关仪、大孔径闪烁仪等观测数据累计，选取了36个站点（65年站点数据，分布图见图1），并联合多源遥感数据（地表覆盖），运用五种机器学习方法（回归树、随机森林、人工神经网络、支持向量机、深度信念网络）构建不同尺度的地表蒸散量扩展模型。结果表明：与其他四种方法相比，随机森林方法更适用于黑河流域从站点到区域的地表蒸散量尺度扩展研究。基于所选的随机森林尺度扩展模型，以遥感及空气驱动数据为输入，生成了2012年至2016年生长季（5月至9月）黑河流域地表蒸散量时空分布图（etmap）。结果显示，etmap的整体精度良好。上游（las1）、中游（las2-las5）和下游（las6-las8）的均方根误差（RMSE）（平均绝对百分比误差MAPE）分别为0.65毫米/天（18.86%）、0.99毫米/天（19.13%）和0.91毫米/天（22.82%）。总之，etmap是基于站点观测数据与随机森林算法尺度扩展的高精度蒸散量产品。有关所有站点信息及尺度扩展方法的详细信息，请参阅Xu等（2018），有关观测数据处理的信息，请参阅Liu等（2018）。