Parameter Sensitivity Analysis And Optimization For Remote Sensing Based Evapotranspiration Model

Global warming has a significant impact on the terrestrial hydrological cycle.As the only way for water return to the atmosphere from the land surface,terrestrial evapotranspiration is an important link in the Earth's hydrological cycle and energy balance,which closely links the exchange of water carbon and energy between the geosphere and the atmosphere.A large number of studies had shown that more than half of the solar radiation received by the land surface and consumed by the evapotranspiration(ET)process.Therefore,it is of great significance to accurately estimate terrestrial ET for understanding of regional energy balance mechanisms,heat wave and drought monitoring,and the application in scientific irrigation and water resources management.Remote sensing based ET model as an important tool for estimating terrestrial ET pattern at global scale has been greatly developed in recent years.However,the uncertainties of model parameters,model structure,and forcing data have crucial influence on the applicability of model.Hence,how to reduce the uncertainty of the model and the improvement of simulation accuracy are the current research hotspots.Based on observation of multiple flux stations over different climate and environmental conditions,this paper validated and evaluated the applicability of two representative remote sensing based ET models(PT-JPL model and MOD16 model)under different biomes.The results showed that the original model could not achieve ideal results in most cases(the mean NSE=-5.81~0.37),and usually overestimated or underestimated the latent heat flux.Therefore,this paper employed a global sensitivity analysis method(Sobol')to quantitatively analyze the sensitivity of model parameters across diverse biomes.Then,we try to identify the key parameters that have significant impact on the simulated results with different landcovers.Moreover,the DE-MC method was used to optimize the key parameters for this two models.The evaluations for the optimized parameter sets were taken at different spatial scales,including the cross-validation of simulated results at site scale(original model:mean NSE=-0.87~-0.61;optimized model:mean NSE=0.13~0.25);comparison with the water balance data at 32 major basins at catchment scale(MOD16 model:R~2=0.81;PT-JPL model:R~2=0.82);mutiple comparisons with other major terrestrial ET datasets at the global scale.Result shown that the optimized parameters could provide an efficient way to improve the model performance.At last,the global terrestrial ET dataset(daily scale with 0.05~o spatial resolution)from 2001 to 2010 was generated based on the two optimized ET models,which providing the ET spatial distribution and the interannual dynamics.Futhermore,the trends,anomalies,inter-monthly variation patterns,and the proportion of different components of total evapotranspiration were systematically analyzed in chapter 6.The results indicated that the total amount of terrestrial evapotranspiration during this decade had shown a slowly rising trend,and there has been a significant negative correlation between abnormal changes in the year with high and low anomalies.Moreover,the proportions of soil evaporation,vegetation transpiration,and intercepted evaporation were counted as 49%,39%,and 12%respectively.Meanwhile,the net radiation is still an important factor that dominated the seasonal variation of global ET,whereas the vegetation coverage and soil moisture condition is the reason that determines the difference of ET spatial pattern at local scale.