Estimation Of Spatially Distributed Model Parameters Using A Regularization Approach

Abstract

The high dimensionality of the parameter search space can be solved by the introduction of additional information about the parameters. In this research the information contained in the apriori parameter estimates, derived using the Koren approach (Koren et al. 2000), was used to identify regularization equations that constrain the parameter variability during the calibration process and reduce the dimension of the calibration problem. The study of spatial variability of apriori parameters with respect to the NRCS based curve numbers and the depth of soil showed some recognizable trends that could be exploited in the form of some simple regression equations. These equations, along with some inter parameter relations, were used as regularization equations. Calibration of the coefficients of the regularization equations instead of the SACSMA parameters (Burnash et al.1973) reduced the dimension of the problem from 858 to 33 unknowns and resulted in significant reduction in the objective function values.