Leveraging temporal variability in global sensitivity analysis of the Daisy soil-plant-atmosphere model

Abstract

Dynamic crop models, such as the Daisy soil-plant-atmosphere model, simulate many processes and encompass a large number of parameters. Global sensitivity analysis (GSA) aims to identify the most influential parameters and understand model structure and behaviour. However, little attention has been paid to the temporal dynamics of parameter sensitivity in crop models, even though it can provide greater insight into model structure. This study performs a comprehensive GSA on the Daisy model, including the soil-vegetation-atmosphere transfer (SVAT) module, focusing on crop yield as well as CO₂, N₂O and energy fluxes. The Sobol’ method was applied to two types of outputs: (i) outputs aggregated into a scalar with an objective function (RMSE or cumulative) and (ii) vector outputs analysed at each time step. The main objectives of this paper were to compare the temporal and aggregated applications of GSA and to identify influential parameters of Daisy under different environmental conditions. Both aggregated and temporal methods identified the same main parameters. Nevertheless, temporal analysis provided deeper insight into model behaviour and calibration guidelines, revealing dynamic changes in parameter sensitivity at weekly and hourly resolutions and identifying critical periods for calibration. Aggregated analysis was less time-consuming and focused on specific aspects due to the definition of the objective function. Finally, we discussed the risks and solutions for Daisy over-parameterisation as well as methods for parameter estimation based on information provided by the GSA.