Combining global sensitivity analysis and multiobjective optimisation to estimate soil hydraulic properties and representations of various sole and mixed crops for the agro-hydrological SWAP model

Abstract

Sensitivity analysis and multiobjective optimisation are established diagnostic instruments for the identification of uncertainty factors and structural deficits in environmental simulation models. Although the application of both techniques provides a comprehensive understanding of model behaviour, they are seldom practised in combination. In this study, the Sobol global sensitivity method and the multiobjective algorithm AMALGAM are combined to assess the agro-hydrological SWAP model for simulating the soil water balance of different sole and mixed crops based on hydrological and phenological field observations. Fifteen unknown model parameters are subjected to the sensitivity analysis (GSA) with the aim of finding their importance to model performance of matric potential (F 1) and soil water content (F 2). Subsequently, sensitive parameters are calibrated by optimising F 1 and F 2 simultaneously. The GSA showed that the description of the rooting density and potential evapotranspiration is of crucial important to F 1, and that soil properties are most relevant for F 2. Parameter interactions played a primary role in the response of matric potential, being irrelevant for F 2. Structural model deficiencies in reproducing both objectives simultaneously were found in the multiobjective analysis, meaning that deterioration in the fit to one of the objectives is in favour of the other. However, solutions exist that produce satisfying fits to both observational types, suggesting that the SWAP model has the capability of simulating the soil water balance of the crops considered. The results of the evaluation period revealed model deficiencies in simulating the process under an environmental regime significantly different from that of the calibration period, indicating the necessity of acquiring a broader spectrum of environmental regimes for parameter calibration. Overall, this study demonstrates how complex and variable the relationship between parameters and model outputs can be in environmental models and highlights the value of combining global sensitivity analysis and multiobjective optimisation in order to improve model performances.