| As an important part of atmospheric water cycle and energy balance,evapotranspiration has a serious impact on global climate change and terrestrial ecosystems.Therefore,continuous monitoring of land evapotranspiration at different temporal and spatial scales is essential to increase the availability of water resources and the allocation of turbulent energy on the land surface.Remote sensing technology can provide continuous key physical variables.Remote sensing is considered to be the most promising and feasible method for mapping the regional evapotranspiration model of the Earth’s surface.At present,there are a lot of remote sensing evapotranspiration inversion models.But some of them need to input ground data,which limits their application in areas where measured data are scarce.The eigenspace method relies less on the surface data and only inputs the surface temperature and vegetation index.However,it is easy to define the model boundary empirically,resulting in uncertainty in parameter calculation.There are some challenges in how to accurately select the dry and wet boundary of the model,to realize the spatio-temporal continuous simulation of evapotranspiration.Based on the theoretical mechanism of the surface temperature vegetation index(TS-VI)feature space and the previous studies,combining the spatial domain and the time domain,this study converts the retrieval of the theoretical trunk edge from the regional scale to the pixel scale.First,this study determines the boundary conditions pixel by pixel with the TS-VI feature space with spatio-temporal two-dimensional attributes.Then,the thesis obtaines modified temperature vegetation drought index.Besides the instantaneous evaporation ratio of any pixel is calculated by the improved bare soil surface temperature solution method.Paper combines with the daily net radiation realized by remote sensing.The final daily evapotranspiration is retrieved,it is time-space continuity,high accuracy and good universality.In order to evaluate the accuracy of the model,site measurement and model inversion results are selected for verification.The accuracy is evaluated mainly by correlation coefficient(r),mean absolute error(MAE),root mean square error(RMSE)and deviation(B).In the regional scale exploration of the southern Great Plains of the United States,the correlation between the predicted value of evapotranspiration and the measured value was more than 0.840,the average absolute error was less than 0.895 mm and the root mean square error was less than 1.100 mm.The application in the United States showed that the model was universal for large scale regions.Evapotranspiration had spatial heterogeneity,showing a high low high distribution pattern from west to east.Besides,the seasonal difference of evapotranspiration was significant,with peak value in summer.Different factors had different effects on evapotranspiration,the response of air temperature and precipitation was intense,and the estimation result of evapotranspiration in the subtropical warm climate was optimal.In the comparison experiments of various evapotranspiration models,thesis analyzed the change trend of the dry and wet edges of each model every day.And thesis obtained the accuracy of its simulation of instantaneous evaporation ratio,instantaneous evapotranspiration and daily evapotranspiration.The last,thesis proved that the accuracy of the pixel model was significantly higher than that of the regional model.Compared with the Tang model with the same pixel scale,although the indicators of this model were slightly worse overall,this thesis had certain advantages in terms of bias. |
