Assessing the effect of using different APSIM model configurations on model outputs

Abstract

The three major sources of uncertainty in crop models are model inputs, structure and parameters. Model structure is one of the major contributors to this uncertainty, however, its quantification is difficult due to limitations in controlling confounding effects from parameter and input uncertainty. The objective of this study was to quantify the contribution of structural uncertainty to the variance in model outputs produced by the Agricultural Production Systems sIMulator (APSIM). Outputs investigated were yield, irrigation requirements, partial gross margin, drainage and nitrogen (N) leaching. Eight model structures differing in choice of soil water model, crop model and irrigation model were developed within a single APSIM version (v.7.10) and tested under three contrasting environments (climate × soil) across 120 years. We quantified: (i) the model structure uncertainty (from soil water, crop and irrigation models) using analysis of variance (ANOVA) and deviation analysis; and (ii) the variability of outputs due to model structure and climate using the coefficient of variation. Confounding effects from inputs, parameters and model users were controlled. Most structural uncertainty resulted from first order effects of the choice of model components (crop model: 12.2–98.9%, irrigation model: 0–78.4%, soil water model:1–33.7%) rather than second order interactions between components (0.1–18.9%). Furthermore, uncertainty from choice of sub-model/model used was not necessarily related to the structural complexity of these components. The effects of structural uncertainty on predictions commonly used to inform agronomic, ecological or policy decision making were strongly impacted by site and climate conditions i.e., high rainfall site (∼1330 mm year−1) had less uncertainty and variability as compared to low rainfall site (∼610 mm year−1), highlighting the need for any uncertainty assessment to cover the entire range of conditions for model application. Here we show the value of a component-based modelling framework for quantifying uncertainty in crop modelling studies.