Small catchment agricultural management using decision variables defined at catchment scale and a fuzzy rule-based system: a mediterranean vineyard case study

[Departement_IRSTEA]Ecotechnologies [TR1_IRSTEA]INSPIRE<br/>Contact: [email protected]

إنجليزي. Physically based hydrological models are increasingly used to simulate the impact of land use changes on water and mass transfers. The problems associated with this type of parameter-rich model from a water management perspective are related to the need for (1) a large number of local parameters instead of only a few catchment-scale decision variables and (2) the technical skills and computational expertise necessary to perform these models. This study aimed to show that it is possible to define a reduced number of decision variables and rules to synthesise numerical simulations carried out through a physically based model. The MHYDAS model was run on a Mediterranean vineyard catchment located in southern France (Roujan, Herault) for an actual, common rainfall event to calculate the runoff coefficient. The simulation results concerned 3,000 samples of contrasted scenarios. The scenarios were characterised by four catchment-scale decision variables related to agricultural practices: the proportion of the area of non agricultural land, the proportion of the area subjected to full chemical weeding practices (with the complement being mechanical weeding), the spatial arrangement of the practices based on the distance to the outlet and the initial soil moisture content. The simulation results were used to generate fuzzy linguistic rules to predict the runoff coefficient, as computed by the physical model from the decision variables. For a common end of spring rainfall event, simulations showed that the runoff coefficient was most heavily influenced by the initial soil moisture and the proportion of the area of full chemical weeding practices and the proportion of the area of other land uses and their spatial arrangement also played a role. The fuzzy rule-based model was able to reproduce the hydrological output with good accuracy (R2 = 0.97). Sensitivity analysis to the rainfall magnitude showed that if the amount of rainfall was the key factor explaining the runoff coefficient absolute values, the structure of the rule base remained stable for rainfall events close to the one studied