Data Assimilation Of Land Surface Processes Based On Remote Sensing Evapotranspiration

Timely, accurate estimation of land surface fluxes has great importance in droughtestimation, irrigation management and water resources comprehensive utilization. Landsurface model is treated as a powerful tool in continuous estimation of surface fluxes,and has been widely used in hydrological researches. However, model error tends toaccumulate in the process of model simulation due to the uncertainty of the forcing data,besides; heterogeneity of land surface limits the applicability of model in regionalestimation. Remote sensing data can provide accurate spatial distribution withoutaccumulating errors over time. Coupling remote sensing data into land surface modelthrough data assimilation method can efficiently enhance the accuracy of surface fluxestimation, which provides a new approach in continuous regional estimation of surfacefluxes. The thesis focuses on the land surface data assimilation method based on remotesensing evapotranspiration model and its application in soil moisture and fluxesestimation.The Surface Energy Balance System (SEBS) model based on satellite remotesensing data was firstly used to estimate LE at satellite over-passing time over a typicalwheat/maize rotation cropland on The North China Plain. The estimation results werevalidated by the observation from eddy covariance (EC) system. The root-mean-squareerror (rmse) of the estimated LE for both wheat and maize seasons were under20%,which showed the accuracy and reliability of instantaneous surface fluxes estimation bySEBS model in the study area. Afterwards, four extrapolation methods calculating dailyET total from instantaneous LE at satellite over-passing time were tested. Comparisonresults at five different croplands specifically wheat, maize, sorghum, grass and baresoil indicated that all the four methods had significant underestimation oroverestimation at different satellite over-passing time in the morning and afternoon,respectively, which leaded to large errors in the estimation of daily ET. To address theproblem, a new method based on daily representative parameters to improve SEBSmodel was proposed in the thesis, which circumvented the traditional scaling up processin the daily ET estimation. The method was first validated by EC observations, and thenapplied to long term large scale regional ET estimation. In order to further improve the accuracy of long term simulation of land surfaceenergy fluxes, the thesis focused on the research of land surface data assimilationapproach. An Ensemble Square Root Filter (EnSRF) method based on EnKF techniquewas coupled to a widely used conceptual Hydrological model called HyMOD as a casestudy. The modeling and observation errors added to the HyMOD were discussed in thefirst place, followed by an ensemble size experiment. Once the appropriate model errorand ensemble size was chosen, a simulation study focus on the data assimilationperformance with the correlation between the streamflow observation and model stateswas thus processed. The EnSRF method was thus implemented to HyMOD and resultsfor flash flood forecasting over two small watersheds respectively from America andChina were analyzed. Results demonstrated the benefit and efficiency of implementingdata assimilation into a hydrological model to improve flood forecasting with potentialapplication to real-time alert systems. On that basis, the validated EnSRF method wascoupled to a Hydrologically-Enhanced Land Process (HELP) model. The remotelysensed LE estimation from the former chapters was used as the observation value in thedata assimilation system to update the model states such as soil water contents andsurface temperatures, etc., where the open-loop simulation without state updating wastreated as the benchmark run. Results showed that the rmse of soil water contentreduced by30%-50%compared to the benchmark run, while the surface fluxes also hadsignificant improvement to different extents, among which the rmse of LE estimationfrom wheat season and maize season reduced by33%and44%, respectively. Theseresults demonstrated that the effect of data assimilation in improving land surfaceenergy fluxes and soil water states is possitive, which showed that data assimilationsystem had great potential in hydrological simulation and water management.