Based on half-hourly data collected from eddy covariance systems, an urban evapotranspiration estimation model was built using random forest model, which introduced the source area information-vegetation coverage as input variables. Vegetation coverage were obtained by combining footprint modeling and land remote sensing data. The simulation results were used to fill the missing values and explore the major control factors of the evapotranspiration during both dry and wet seasons in Shenzhen, China. The results show that compared with the traditional RF and MDS model, urban evapotranspiration could be simulated with better accuracy considering vegetation coverage, R²=0.73, RMSE=20.5 W/m², MAE=13.3 W/m², pbias=0.8%. For the estimation of relatively high evapotranspiration in wet seasons, RF performs significantly better than MDS model. MDS model under-estimates by 12.4% while RF only by 4.7%. During wet seasons of Shenzhen, vegetation coverage is the major control factor of evapotranspiration. In dry seasons, temperature, net radiation, saturated water vapor pressure deficit are the major control factors.

基于城市涡动相关系统每30分钟蒸散发数据, 结合足迹源区分析及研究区域遥感数据, 获取源区植被盖度, 构建基于随机森林的蒸散发模拟方法。将此模拟方法应用于涡动系统空缺值插值, 并探究深圳干湿季蒸散发的主要控制因子, 得到如下结论。1) 与不考虑植被盖度的随机森林模型及边际分布抽样算法(MDS)相比, 考虑植被盖度的随机森林模型可以更高精度地模拟城市蒸散发。与实测数据相比, 模型的R²=0.73, RMSE= 20.5 W/m², MAE=13.3 W/m², pbias=0.8%。2) 对湿季日间较高蒸散发时期的空缺值进行插值, 考虑植被盖度的随机森林模型显著优于MDS模型。与实测数据相比, MDS插值模型低估 12.4%, 而随机森林模型低估 4.7%。3) 深圳湿季期间, 植被盖度是蒸散发的主要控制因素。在干季, 温度、净辐射以及饱和蒸气压差是蒸散发的主要控制因子。