Comparison between mechanistic and functional models for estimating soil water balance: deterministic and stochastic approaches

In many models used to simulate soil-water relationships, representations of the transport mechanisms in the soil-plant-atmosphere continuum, range from mechanistic to functional. The objective of this paper is to compare two functional models, FAO (Doorenbos and Pruitt, 1977) and Ritchie (1985) models, with a mechanistic model (Maraux and Lafolie, 1998) to simulate the soil water balance of maize and sorghum grown in sequence in Nicaragua. In the FAO model, the soil is described as a single reservoir which is characterized by its amount of water varying on a daily time scale, depending on the rain, drainage, and actual evapotranspiration. In the Ritchie model, the soil is regarded as a multilayered soil profile. The maximum evapotranspiration is divided between soil evaporation and plant transpiration, and drainage occurs if the amount of water arriving in the last layer corresponds to a water content greater than the field capacity. The mechanistic model is based on the Richards' equation. Comparison of the three models was first made according to a deterministic approach with parameters coming from the same database. We then considered a stochastic approach for which 800 hydraulic characteristics of the soil were generated, according to the spatial variability observed at the field scale and to the scaling theory applied to similar porous media. A distribution of the stochastic parameters used in the three models was thus derived. Results showed that the order of magnitude of the evapotranspiration was similar for the three models (902, 874, 842 mm cumulative evapotranspiration for a 203 day period for the MM, Ritchie, and FAO models, respectively). Adding a capillary rise mechanism in the functional models improved moderately the soil-water balance. Evapotranspiration and drainage showed moderate sensitivity to spatial variability in soil hydraulic properties coefficients of variation less than 1.6%), whereas final water storage (after 203 days) showed a greater sensitivity (coefficients of variation from 7.9-15.7%, depending on the model). (Résumé d'auteur