PROPAGATION OF THE ERROR OF SPATIAL PREDICTION OF SOIL PROPERTIES IN SIMULATING CROP EVAPOTRANSPIRATION

A major problem in using soil-water models for land evaluation is to d efine which combination of model complexity and sampling density provi des the most reliable predictions for a given investment. The overall error which affects the predictions arises from different sources. For instance, one is the error due to spatial estimation of soil data fro m soil classifications. The paper describes, for two models of differe nt complexity and two soil maps of different resolution, how this erro r is propagated to the predictions of crop evapotranspiration. Errors arose and were accumulated during the course of the simulations, but t hey were not amplified. The variance of the errors depended on the cli matic conditions of the simulations. Nevertheless, whatever the condit ions, uncertainties in soil properties were propagated to the predicti ons to a lesser extent by the simple model than by the complex one. Fo r example, combining the simple model and the 1/10 000 map required th e same experimental investment as combining the complex model and the 1/100 000 map, but the variance of propagated errors was 53% greater f or the complex model than for the simple one. Thus, if we consider onl y the simulation error derived from estimation errors in soil properti es and the sampling costs, it is justifiable to use simple models for predicting the soil water balance in space. However, decisions should be based on the overall precision of the simulations which is also aff ected by other sources of error, such as the error arising from the mo del itself.