Résumé
The geostatistics, which helps to understand more the spatial inference existing within the distribution of a variable, drives to the development of a hydro-geostatistics methodology, to estimating intrinsic parameters of hydrologic models. We study sixteen watersheds on the high Niger River and its tributaries (in West Africa). Using statistical considerations we de-correlate the difference of performance of two model versions in relation to the spatial variability of a physical property of soil--specifically, the Water Holding Capacity (WHC). The de-correlation is through applying a weighting factor to simulating runoff depths in semi-distributed modelling. The improvement, brought to semi-distributed hydrologic models, is alternatively evaluated through a process named inter-models protocol; during which, a correlation coefficient--between the series formed by relative differences of two model versions performances and the other formed by the relative dispersion coefficient of WHC on each watershed--passes from 0.80 to 0.21 in absolute value, for SimulHyd hydrologic model. This model--SimulHyd--is a variant from the hydrologic model GR2M; while replacing a constant A--denoting the maximal capacity of a conceptual "soil reservoir" inside GR2M- by the term 1/X1*WHC. The X1 parameter is inherent in GR2M and modulates modelling entries data, the other is X2 parameter that is apply to a conceptual "gravity reservoir" and modulates the output data. The weighting factor used within SimulHyd semi-distributed--and within GR2M in our studies--is built from the empirical variance of WHC values between grids of a considered watershed. We propose an "inter-basins protocol" whose vocation is to build a space named "NASH-WHC", which is different from the usual geographical space, where a semi-distributed model with the built weighting factor is better represented than without it. The result is a correlation coefficient of 0.48 for the spatial structure of basin-points in "NASH-WHC" space--instead of 0.36 when the built weighting factor is not used. The hydro-geostatistics methodology constitutes a geo-statistical extension of this inter-basins protocol, where tools and steps of classical geostatistics are applied to modelling and simulating hydrologic model parameters in the proposed "NASH-WHC" space.