Sensitivity and uncertainty analysis in agro-hydrological modelling of drip fertigated lettuce crops under Mediterranean conditions

Abstract

Most agro-hydrological modelling studies of fertigated crops published in the literature are based on classical model identification approaches, in which a single and optimal representation of the water/nitrogen balance and crop biomass production is sought, through the calibration of model parameters. Here, the concept of an optimal parameter set is rejected in favor of the concept of equifinality (multiple possibilities for producing simulations that are acceptable). Monte Carlo simulations of seven experimental trials were conducted with EU-Rotate_N, using random realizations of a total of 27 parameters, to identify and assess the dominant sources and magnitude of uncertainty in drip irrigated and fertigated lettuce crop modelling. Eight interacting parameters were identified as sensitive, with strong influence on model outputs, using two different approaches to sensitivity analysis. Sensitivity patterns are shown to vary depending on the conditions prevailing during individual trials. Model uncertainty was assessed using a non-formal Bayesian approach (the Generalized Likelihood Uncertainty Estimation GLUE). The amplitude of the uncertainty bands and position relative to the observations are shown to vary depending on the range of variables and conditions that the model is required to simulate. The statistical analysis of ensembles of simulations, conducted with random realizations of model parameters are shown to produce similar results to those resulting from the analysis of the field experiments, in the sense that similar factors were found significant in the determination of agronomical and environmental variables related to crop growth and soil water and nutrient balances. An example of the use of ensemble-analysis for decision making in the context of environmental risk management is given.