Calibration of hydrological models in the spectral domain: An opportunity for scarcely gauged basins?

Abstract

This study considers the use of the maximum likelihood estimator proposed by Whittle for calibrating the parameters of hydrological models. Whittle's likelihood provides asymptotically consistent estimates for Gaussian and non‐Gaussian data, even in the presence of long‐range dependence. This method may represent a valuable opportunity in the context of ungauged or scarcely gauged catchments. In fact, the only information required for model parameterization is essentially the spectral density function of the actual process simulated by the model. When long series of calibration data are not available, the spectral density can be inferred by using old and sparse records, regionalization methods, or information on the correlation properties of the process itself. The proposed procedure is applied to the case studies of two Italian river basins, for which a lumped rainfall‐runoff model has been calibrated by emulating scarcely gauged situations. It is shown that the Whittle estimator can be applied in such context with satisfactory results.