Global Terrestrial Evapotranspiration Estimation And Spatiotemporal Pattern Analysis

Terrestrial evapotranspiration(ET)plays an important role in connecting the global water cycle,carbon cycle and energy cycle.Climate change will intensify the hydrological cycle and change ET,thus affecting ecosystem services and feedback to regional and global climate.It is crucial to accurately estimate the spatial and temporal distribution of ET.Conventional evapotranspiration observation techniques are difficult to directly measure ET on a large spatial scale.Although existing models for estimating ET can meet some scientific research needs,due to the inconsistent input parameters,spatiotemporal resolution,coverage and accuracy of ET products,the estimated ET of different models also exhibits significant differences and uncertainties in spatiotemporal distribution and trend changes.In addition,natural changes in the Earth’s climate system can reduce the ability to identify climate trends.Building a global ET product benchmark with reliable estimates under variable climate conditions and accurately analyzing global ET spatiotemporal changes remains challenging.This thesis focuses on the issues of global ET assessment and spatiotemporal pattern and change analysis.The main research content is divided into the following two parts:(1)To evaluate 8 widely used ET remote sensing products at the eddy covariance(EC)flux sites and analyze the spatiotemporal pattern and trend changes of ET.According to the verification of 149 EC flux sites,the correlation coefficient between the eight ET remote sensing products and the observation of EC flux sites is 0.65-0.75,and the root mean square difference is 23.60mm/month-35.45mm/month.The robustness of each ET remote sensing product on different land cover type sites is poor.The spatial pattern of different ET remote sensing products shows good consistency,but the interannual estimation range is between 483 mm and 636 mm,with significant differences.Moreover,there is no consensus on the long-term change trend among different ET remote sensing products.From 1982 to 2015,GLEAM,MTE,and rea showed a significant upward trend,ERA and MERRA2 showed a significant downward trend,and Land Flux showed a nonsignificant upward trend.Although there are various ET products based on different models,each product has its advantages and limitations,and there is still great uncertainty in analyzing the spatiotemporal distribution and trend changes of these ET products.(2)To propose a spatiotemporal fusion model(STF)on the basis of convolutional encoding and decoding networks.STF introduces gated recurrent unit,attention mechanisms,dilated convolution,and skip connections,which can integrate EC flux site observations,ET remote sensing products,and land surface auxiliary information to establish a reliable and robust global ET dataset to improve ET estimation.The correlation coefficient between ET dataset(ST-rea)obtained from inversion and EC flux site observation is 0.85,and the root mean square difference is 18.46 mm/month.Compared with other eight ET remote sensing products,the STF model significantly improves the accuracy and reduces the uncertainty,and ST-rea shows robust performance on sites of different land cover types.According to the spatiotemporal change analysis conducted by ST-rea,we found that the global ET showed a significant increase of 0.4 millimeters per year from 1982 to 2015,with strong correlation between temperature,NDVI,and ET,and consistent trend changes.