地表蒸散发是整个生物圈、岩石圈、水圈、冰雪圈与大气圈中水分循环和能量传输的重要控制因素。通过无人机遥感获取的超高分辨率遥感数据，可以得到亚米级的地表蒸散发产品，有助于分析复杂下垫面中各种地表类型水热状况的精细差异，更好地解决生态水文过程中由于空间异质性带来的问题，有力支撑干旱应急响应、精准灌溉和生态环境保护与修复等。本文基于2019-2020年植被生长季（2019年5-9月，2020年6-10月）在黑河中游三个典型下垫面获取的无人机遥感数据，结合黑河流域地表过程综合观测网观测数据，反演与验证了地表反照率、叶面积指数和地表温度等地表参数；优选了地表蒸散发遥感估算模型，估算与验证了黑河中游典型地表类型的蒸散发，并开展黑河中游典型下垫面、绿洲﹣荒漠区域地表蒸散发的精细分析。主要结论如下：

（1）在地表参数的遥感反演与验证中，采用窄波段到宽波段反照率的转换方法来反演地表反照率，并引入了区分地表类型的转换公式。与传统不区分地表类型的方法相比，精度明显提升，其中地面观测值的决定系数（R²）从0.65提升至0.88，平均相对误差（MRE）从5.75%降低至3.23%，均方根误差（RMSE）从0.029降低至0.014。利用多光谱数据和激光雷达数据，分别采用了物理模型反演方法和基于间隙率的方法来反演叶面积指数。验证结果表明，在绿洲，物理模型方法反演精度较高，其中在大满与地面人工测量值的R²可达0.89，MRE为-4.9%，RMSE是0.34。而激光雷达方法在绿洲植被生长茂盛地区由于功率小等原因难以穿透整个植被冠层，反演精度较差。在荒漠区域，则把NDVI和激光雷达反演的叶面积指数建立回归关系式，然后推广到整个植被生长季。地表温度遥感产品与地面观测值的R²可达0.96，MRE为0.14%，RMSE是1.82 K，可满足地表蒸散发遥感估算的需要。

（2）在地表蒸散发的遥感估算与验证中，优选了SEBS模型和非参数化（NP）模型。其中基于SEBS模型估算潜热通量与涡动相关仪（EC）观测数据的R²达0.92，MRE为-8.64%，RMSE是69.9 W/m²；与大满站双波段闪烁仪（OMS）观测数据的R²达0.71，MRE为1.8%，RMSE是64.5 W/m²。基于NP模型估算潜热通量与EC观测数据的R²达0.88，MRE为-4.9%，RMSE是75.3 W/m²；与大满站OMS观测数据的R²达0.59，MRE为-18.6%，RMSE是70.8W/m²。由于参数化方案的精准选取，SEBS模型的精度较好，而NP模型结构简单，所需参数少，更适合无人机遥感实时监测及快速响应的需要。

（3）以三个试验区为例，通过分析黑河中游典型下垫面、绿洲﹣荒漠区域蒸散发的时空变化特征，发现：在黑河中游植被生长季，农田、湿地与荒漠的潜热通量都呈现出了先增加后减少的趋势，但绿洲的潜热通量受植被生长节律和太阳辐射的影响较大，荒漠的潜热通量虽与植被生长节律有关，但主要受降水影响。绿洲与荒漠的能量平衡各组分也存在明显差异。整体上，绿洲比荒漠的净辐射、潜热通量偏大，而感热通量、土壤热通量则偏小，这主要由于绿洲植被覆盖度高，土壤水分充足，可利用能量更多转化为蒸发潜热；而在荒漠区域，地表植被稀疏，土壤水分少，可利用能量更多转化为感热通量和土壤热通量。不同地块同种地物的蒸散发量也存在着差异。在绿洲区域，不同田块、同一田块不同位置由于作物生长状况、灌溉等因素的影响，其潜热通量差异在68.5~86.1 W/m²；在荒漠区域，不同地形可以有效地影响土壤水分差异，从而导致植被生长状况不同，其潜热通量差异在6.1~39.7 W/m²。

Evapotranspiration (ET) includes vegetation transpiration, soil and water evaporation (including snow and ice sublimation) and evaporation of precipitation trapped by vegetation canopy. ET is an important control factor of water circulation and energy transfer in the whole biosphere, lithosphere, hydrosphere, cryosphere and atmosphere. The ultra-high resolution remote sensing data obtained by UAV can be used as the input data of the ET estimation model to calculate the sub-meter level ET products, which can help us analyze the differences in hydrothermal conditions between surface types in complex substrates and better solve the problems caused by spatial heterogeneity in the ecohydrological process. Therefore, the ultra-high resolution remote sensing data obtained by UAV is important for drought emergency response, precise irrigation and ecological monitoring. This paper retrieved and validated surface parameters such as albedo, leaf area index (LAI) and land surface temperature (LST) based on the UAV remote sensing data, which acquired in three typical surfaces in the middle reaches of the Heihe River Basin (HRB) during the 2019-2020 vegetation growing season (May to September in 2019 and June to October in 2020), and the observation data from the comprehensive observation network of surface processes in HRB. Moreover, the current paper firstly optimized the ET models to estimate ET, and then validated ET. Finally, this paper precisely analyzed the spatio-temporal variation rules of ET in the middle reaches of HRB. The key conclusions are as follows.

(1) In the retrieve and validation of surface parameters: this paper adopted the conversion method of narrow to wide band albedo to retrieve albedo and introduced the conversion method of distinguishing surface types. Validation results showed that the decision coefficient (R²) was improved from 0.65 to 0.88, the mean relative error (MRE) was reduced from 5.75% to 3.23%, the root mean square error (RMSE) was reduced from 0.029 to 0.014, compared with the traditional conversion method of not distinguishing surface types. In the retrieve of LAI, this paer used the canopy radiation transfer model and the gap rate based retrieval method for multispectral data and LiDAR data, respectively. Consequently, the validation results showed that in the oasis area, the cost function based on the ProSAIL model and considering prior knowledge could accurately retrieve LAI with R² = 0.89, MRE = -4.9%, and RMSE = 0.34 in Daman. However, LiDAR acquired on UAV had difficulty penetrating the entire vegetation canopy in areas with lush vegetation growth in oases due to the its power and hence the accuracy was poor. Therefore, this paper built the regression relationship between NDVI and LIDAR to retrieve LAI in the desert region, and then extended to the whole vegetation growing season in the desert region. In the validation of LST, R² = 0.96, MRE = 0.14%, and RMSE = 1.82 K, which fully met the need of the ET surface energy balance model.

(2) In this paper, the SEBS model and the non-parametric (NP) model were optimized for the estimation of ET. Through the SEBS model, the validation results showed that the estimated ET with eddy covariance system (EC) observation data were R² = 0.92, MRE = -8.64%, and RMSE = 69.9 W/m². In addition, the validation results with Optical-Microwave Scintillometer (OMS) observation data were R² = 0.71, MRE = 1.8%, RMSE = 64.5 W/m². Through NP model, the validation results of estimated ET with EC were R² of 0.88, MRE of -4.9 W/m², and RMSE of 75.3 W/m² and with OMS were R² = 0.59, MRE = -18.6%, and RMSE = 70.8 W/m². Therefore, the SEBS model had better accuracy due to the accurate selection of parameterization scheme. Whereas the NP model was simpler with fewer parameters, which was more suitable for a real-time monitoring and rapid response for UAV remote sensing.

(3) The spatial and temporal variation of ET in the middle reaches of HRB and oasis-desert area was analyzed by taking three experimental areas as examples. Generally, the ET in farmland, wetland and desert showed a trend of increasing and then decreasing with seasonal changes. Specifically, ET in the oasis area was greatly influenced by the vegetation growth rhythm and solar radiation. Although ET in the desert area was also related to the vegetation growth rhythm, it was mainly influenced by precipitation. Futhermore, there were also significant differences in the components of energy balance between oasis and desert. Compared with desert, the net radiation and latent heat flux (LE) of oasis was higher, whereas sensible heat flux (H) and soil heat flux was less. This was mainly due to the high vegetation cover and sufficient soil moisture in oasis and thus the energy was more converted into LE. While in desert area, the surface vegetation was sparser and soil moisture was relatively low. The energy was more converted into soil heat flux and H. Moreover, ET of different plots still varied even with the same land surface type. In the oasis region, LE was differentiated in different plots and different positions of the same plot due to the influence of crop growth status and irrigation, which ranged from 68.5 to 86.1 W/m². In the desert region, the terrain could effectively affect the difference of soil moisture, which caused the diviation of vegetation growth conditions and the variance of LE ranged from 6.1 to 39.7 W/m².