SPATIAL VARIABILITY OF SURFACE-PROPERTIES AND ESTIMATION OF SURFACE FLUXES OF A SAVANNA

During the HAPEX-Sahel experiment in 1992, a data set including atmosp heric forcing, soil temperature at several depths, surface fluxes and surface soil moisture from 0 to 15 cm was collected on a degraded fall ow savannah over a period of 18 days. The SiSPAT SVAT model was calibr ated using this data set to provide a realistic reference set of param eters. The sensitivity of surface fluxes to the specification of surfa ce properties based on this reference set of parameters was quantified focusing on runoff, evapotranspiration and soil moisture at the field scale. Runoff and latent heat flux, predominantly bare soil evaporati on, were found to be the most sensitive processes in relation to soil parameters. For transpiration, even with such a sparse vegetation, lea f area index was the most sensitive factor. A stochastic approach was used to analyze the sensitivity of surface fluxes to the spatial varia bility of surface parameters. For a variation of +/- 50% of the parame ters, no significant bias was obtained between the mean of the stochas tic simulations and the 1-D simulation performed with the median value s of the parameters. The analysis indicates that the variations of the components of the water budget are linearly related. For larger varia tions of the parameters, the bias is significant; therefore, simple ag gregation rules fail to capture the nonlinearities induced by water tr ansfer into the soil. When diurnal cycles are considered, the standard deviation of bare soil evaporation and surface soil moisture was foun d to be maximum for the intermediate wetting range. For this period, i t would be valuable to parameterize the spatial variability of surface properties into larger scale models. Finally, the SISPAT model, which solves equations derived from the Richards equation [Richards, L.A., 1931. Capillary conduction of liquids through porous mediums. J. Phys. 1, pp. 318-333.]. for soil water distribution is shown to be very sen sitive to the specification of soil parameters. This result would hold for similar models, showing that the Richards equation should be used with caution within large-scale models if a robust estimation of the long-term water budget is to be obtained. (C) 1998 Elsevier Science B. V.