Estimation des erreurs de prediction spatiale du deficit hydrique dues a la rasterisation des cartes des sols.

Average moisture deficits of maize crops over a 32 year period were simulated by the computer model GOA for the whole French territory. This study focuses on the choice of representative soils of each simulation area. GOA simulates the major physical and biological processes of the agricultural system including water stress, growth and yields. Simulations were based on soils data from the French Soil Map. Soil surveys were made at a scale of 1:1,000,000. The model is applied to a network of 57 weather stations by using decade weather values from the period 1958 to 1989. The combination of soil map and climatic data for simulation requires the extrapolation of climatic data from regular grid-cells. Each cell is composed of one or several mapping units (UCS). A mapping unit is an association of one or several soil types (UTS) from which we can obtain necessary data to run a GOA model. GAO is linked to a Geographical Information System (GIS) ARC/INFO which handles and derives input data from the database at the correct spatial scale and produces thematic maps from the model output. In this study, we investigate the accuracy of moisture deficit estimation associated with the choice of representative soil units for grid-cells. To choose representative units of grid-cells with corresponding physical data, three scenarios are distinguished: (1) choice of mapping unit which occupies the largest area within the cell,(2) calculation of weighting mean of all mapping units that occur in a cell,(3) gathering of soil units that have the same soil physical data needed to simulate moisture regime. The results of scenarios are compared with simulation runs made for each soil unit within the mapping unit (reference scenario). The prediction error is also estimated using different grid sizes (10 x 10 km, 20 x 20 km, 40 x 40 km). The mean absolute error and the mean square error were used as quality measure. Error in estimating moisture deficit by the first scenario is 21 per cent. This error decreases if the largest unit within the area cell occupies more than 65 per cent area of the cell and increases for lowest percentages. Spatial variability within a cell hardly influences error levels. Thematic gathering scenario (scenario 3) produces better estimation than the two other scenarios. The best prediction was obtained with cell size 10 x 10 km. Using scenario 3, cell size 40 x 40 km gave a smaller error than cell size 20 x 20 km with all scenarios. For the French Soil Map, cell size 40 x 40 km can be considered to be sufficient for general purposes when using Thematic Gathering scenario.