Application of Soil moisture in Regionalization framework for Predictions in Ungauged Basin and its Uncertainty quantification

Continuous streamflow prediction in ungauged basin and the quantification of its uncertainty has been challenging over the decades. Regionalization of parameters from gauged basin has been the most promising approach adopted by the researchers. However, the improvement of hydrological model prediction in regionalization with proper alternative observed data has been constantly explored. Various researchers have used remote sensing based products such as soil moisture and evapotranspiration as variables for calibration of hydrological models. While the regionalization based approach result in higher predictive uncertainty, performance of models calibrated using remote sensing products have been found to be sub-optimal. In this context, a combined approach of regionalization and remote sensing data may result in reduced predictive uncertainty and enhanced accuracy. In this study the predictive uncertainty quantification in ungauged basin is proposed using the regression-based regionalization framework between the catchment attributes and probability distribution function (PDF) of hydrological model parameters. The PDF of the hydrological model parameters is derived using the MCMC procedure in the DREAM algorithm. The proposed approach is evaluated using the data pertaining to 12 watersheds in the MOPEX database and assuming one of the catchments as pseudo-ungauged catchment. The uncertainty quantification in regionalization for streamflow prediction analysed by average of the prediction is better performing with NSE of 0.77 in pseudo ungauged basin (Sugar Creek EdinBurgh watershed). Further, the remote sensing soil moisture from GLDAS was compared with the model simulated soil moisture analysed using NSE to sub sample the regionalization parameter space in ungauged basin. The regionalization of the reduced parameter set to assess the change in uncertainty quantification is performed and found to have same performance NSE of 0.77 in ungauged basin for streamflow prediction with reduction in average width of 0.23 mm/day in ensembles of streamflow prediction. The ensemble of the simulations has similar performance compared to the model calibrated using streamflow (NSE 0.77). The outcome of the study indicates that the calibration of hydrological model using remote sensing soil moisture product as simulating variable have improved performance the model prediction in the parameter range obtained from the regionalization framework in the ungauged basin. Thus, the integration of regionalization approach with simulation of hydrological model using remote sensing products in the ungauged basin is recommended to apply in the real time applications of water resources management.