Calibration of a groundwater model using pattern information from remote sensing data

Due to the chronic lack of verification data, hydrologic models are notoriously over-parameterized. If a large number of parameters are estimated, while few verification data are available, the calibrated model may have little predictive value. However, recent development in remote sensing (RS) techniques allows generation of spatially distributed data that can be used to construct and verify hydrological models. These additional data reduce the ambiguity of the calibration process and thus increase the predictive value of the model. An example for such remotely sensed data is the spatial distribution of phreatic evaporation. In this modeling approach, we use the spatial distribution of phreatic evaporation obtained by remote sensing images as verification data Compared to the usual limited amount of head data, the spatial distribution of evaporation data provides a complete areal coverage.
However, the absolute values of the evaporation data are uncertain and therefore three ways of using the spatial distribution pattern of evaporation were tested and compared. The first way is to directly use the evaporation pattern defined in a relative manner by dividing the evaporation rate in a pixel by the total evaporation of a selected rectangular area of interest. Alternatively, the discrete fourier transform (DFT) or the discrete wavelet transform (DWT) are applied to the relative evaporation pattern in the space domain defined before. Seven different combinations of using hydraulic head data and/or evaporation pattern data as conditioning information have been tested. The code PEST, based on the least-squares method, was used as an automatic calibration tool.
From the calibration results, we can conclude that the evaporation pattern can replace the head data in the model calibration process, independently of the way the evaporation pattern is introduced into the calibration procedure. (C) 2009 Elsevier B.V All rights reserved.