Exploratory Sensitivity Analysis of CropSyst, WARM and WOFOST: a Case-study with Rice Biomass Simulations (Analisi Esplorativa della Sensibilità di CropSyst, WARM and WOFOST: un Caso di Studio con Simulazioni di Biomassa del Riso)

Abstract

Knowledge about model uncertainty is essential for crop modelling and provides information crucial for a real understanding of models behavior and for parameterization purposes. This work addresses an exploratory sensitivity analysis on the parameters involved with biomass accumulation of three crop models decidedly differing in the way they interpret the concept of growth. The models used in the sensitivity analysis were CropSyst, WARM and WOFOST. We used the Morris screening method, supplied with software for sensitivity analysis SimLab, to determine, at a reasonable cost in terms of model evaluations, which parameters have a substantial influence on a static biomass output (rice aboveground biomass at physiological maturity). Assumptions about the settings of the analysis in terms of parameters estimated distribution and uncertainty are discussed. As case study, we performed the sensitivity analysis using meteorological and management data from an experiment carried out in Opera (Milan, Northern Italy) during 2006. The site can be considered representative of temperate rice in Europe. The screening revealed some important features of the models in terms of their input parameters. The variability of simulated rice biomass was generally high, and few physiological parameters emerged as mostly influential on the biomass output. The biomass-transpiration coefficient of CropSyst was the most important parameter determining final biomass. In WOFOST, CO2 assimilation rates and partitioning coefficients were found to be the most relevant parameters. The majority of the parameters in CropSyst and WOFOST (75-80%) resulted not very influential on the final biomass with the inputs used. On the other hand, WARM is more relevant than the others in that most of its input parameters actually cause variation in the model response.