进入20世纪以来，以升温为主要特征的全球气候变化愈发显著，以干旱和高温为主要代表的极端气候事件频发，面对严峻的干旱情势，研究干旱指标并及时对旱情进行评估和预判，对干旱预警和气候研究有重要意义。蒸散发作为干旱最直接的信号，值得更系统、更全面的研究。进一步描述潜在蒸散发（ETp）和实际蒸散发（ETa）的差异将有助于加深对蒸散发过程的了解，为农业灌溉计划的制定和旱情预警提供重要参考。

本研究以全球陆地为研究区，从机理层面进一步描述全球潜在蒸散发和实际蒸散发之间的关系；并利用多源遥感数据集，探究蒸散发特征值对气候变化、植被变化和水量波动的响应差异，聚焦水量变化，在不同情景下通过支持向量机模型预测未来的全球蒸散发特征值，为相关地区的干旱预警和水资源管理提供科学参考。论文主要成果如下：

（1）根据潜在蒸散发和实际蒸散发在概念上的差异，本研究发现了区域蒸散发削减量（Er）的概念，并用此来描述土壤水分和植被条件对蒸散发的限制作用。使用CRU和GLDAS的全球ETp和ETa数据，计算了全球像素尺度上的Er，并采用Mann-Kendall趋势研究方法探究了ETp、ETa和Er在全球尺度、纬向尺度和像素尺度上的时空变化规律，发现在全球尺度上ETp与ETa之间存在较大的数值差异，并且Er可以很好地表示ETa受限于土壤水分和植被条件而没能达到ETp的情况。

（2）对变化环境的响应差异方面，本研究利用多源遥感数据产品，聚焦气温、降水、归一化差异植被指数（Normalized Difference Vegetation Index，NDVI）和增强植被指数（Enhanced Vegetation Index，EVI）等因素，并利用基于Noah模型的全球径流数据以及GRACE全球水储量数据，通过水量平衡方程分别计算了由自然因素和人类活动因素导致的水储量变化量，来表示水量波动因素，通过Pearson相关分析和基于随机森林的Boruta算法在全球像素尺度上探究了ETp、ETa和Er对于气候变化、植被变化和水量波动的响应差异，结果表明，在全球61.84%的地区，气温是ETa的第一决定因素，稀疏的植被是Er的主要决定因素，即为实际蒸散发的主要限制因素。

（3）未来蒸发特征值的预测方面，本研究使用耦合模式相互比较项目（CMIP6）的未来气温和降水数据，以是否考虑水量因素为出发点，划分了三个不同的变量输入情景，利用支持向量机模型在全球像素尺度上预测了2030、2060和2090年的ETp、ETa及Er。结果表明，考虑水量变化的模型不仅可以极大地提高模型的模拟性能（具有较高的R2），而且可以模拟出欧洲的干旱以及非洲更强烈的蒸散发情况。因此，本研究提出的Er为蒸散发的估算提供了良好的参考，将水量因素引入机器学习模型得到的更强烈的未来蒸散量也有重要的参考价值。

In the 20th century, global warming has become more and more significant. Extreme climate events mainly represented by drought and high temperature have occurred frequently. Facing the severe situation, it is of great significance to study the drought indicators and execute drought warning and climate research. As the most direct signal of drought, evapotranspiration is worthy of systematic and comprehensive research. Further description of the difference between potential evapotranspiration (ETp) and actual evapotranspiration (ETa) will help to explore the limiting factors of evapotranspiration, deepen the understanding of the complex phenomenon of evapotranspiration, and provide an important reference for the formulation of agricultural irrigation plan and drought early warning.

Taking the global land area as the research area, this study further describes the relationship between global potential evapotranspiration and actual evapotranspiration from the mechanism level. The multi-source remote sensing data set is used to explore the response differences of evaporation eigenvalues to climate change, vegetation change, and water fluctuation. Focusing on changes in water volume, this study predicts future global evapotranspiration eigenvalues under different scenarios, providing a scientific reference for drought warning and water resources management in relevant regions. The main results are as follows:

(1) According to the conceptual difference between potential evapotranspiration and actual evapotranspiration, the concept of regional evapotranspiration reduction (Er) is defined to describe the limitation of soil moisture and vegetation condition on evapotranspiration. Utilizing the global evapotranspiration data products of CRU and GLDAS, the ER on the global pixel scale is calculated, and the Mann-Kendall test method is used to explore the temporal and spatial variation laws of ETp, ETa, and Er on the global scale, zonal scale, and pixel scale. It is found that there is a huge numerical difference between ETp and ETa on the global scale, and ER can well represent the situation in that ETa fails to reach ETp due to soil moisture and vegetation conditions.

(2) In terms of response differences to changing environment, this study uses multi-source remote sensing data products to focus on factors such as temperature, precipitation, NDVI, and EVI, and uses runoff data simulated by the Noah model and GRACE data set to calculate the changes of water storage change caused by natural factors and human factors through the water balance equation. Through Pearson correlation analysis and Boruta algorithm based on Random forest, the response differences of ETp, ETa, and Er to climate change, vegetation change, and water fluctuation are explored on the global pixel scale. The results show that in 61.84% of the world, the temperature is the first determinant of evapotranspiration, and sparse vegetation is the determinant of limiting evapotranspiration.

(3) For the prediction of future evaporation eigenvalues, this study uses the CMIP6 future climate data to divide three scenarios based on whether to consider the water factor. And the Support Vector Machine model is used to predict the ETp, ETa, and Er in 2030, 2060, and 2090 on the global pixel scale. The prediction results show that the model considering water change can not only greatly improve the simulation performance of the model (with high R2), but also simulate the drought in Europe and the more intense evaporation in Africa. Therefore, the Er proposed in this study provides a good reference for the estimation of evapotranspiration, and the stronger future evapotranspiration obtained by inputting the water factor into the machine learning model also has an important reference value.