Soil Moisture Estimation by Microwave Remote Sensing for Assimilation into WATClass

This thesis examines the feasibility of assimilating space borne remotely-sensed microwave data into WATClass using the ensemble Kalman filter. WATClass is a meso-scale gridded hydrological model used to track water and energy budgets of watersheds by way of real-time remotely sensed data. By incorporating remotely-sensed soil moisture estimates into the model, the model’s soil moisture estimates can be improved, thus increasing the accuracy of the entire model. Due to the differences in scale between the remotely sensed data and WATClass, and the need of ground calibration for accurate soil moisture estimation from current satellite-borne active microwave remote sensing platforms, the spatial variability of soil moisture must be determined in order to characterise the dependency between the remotely-sensed estimates and the model data and subsequently to assimilate the remotely-sensed data into the model. Two sets of data – 1996-1997 Grand River watershed data and 2002-2003 Roseau River watershed data – are used to determine the spatial variability. The results of this spatial analysis however are found to contain too much error due to the small sample size. It is therefore recommended that a larger set of data with more samples both spatially and temporally be taken. The proposed algorithm is tested with simulated data in a simulation of WATClass. Using nominal values for the estimated errors and other model parameters, the assimilation of remotely sensed data is found to reduce the absolute RMS error in soil moisture from 0.095 to approximately 0.071. The sensitivities of the improvement in soil moisture estimates by using the proposed algorithm to several different parameters are examined.