Is a Crop Model Really Useful for Evaluating and Comparing Irrigation Strategies?.

A major use of crop models is to evaluate management strategies. An important question is how accurate models are for such evaluations. The purpose of this presentation is to illustrate how to quantify the uncertainty in the evaluation of irrigation strategies, when that evaluation is based on a crop model.

The procedure for quantifying uncertainty has three steps. First, the sources of uncertainty are identified. Here we consider uncertainty in the model parameters and model residual error. Secondly, the uncertainty in each source is quantified. Here we use a Bayesian approach. Finally, the uncertainties are propagated through to the quantities of interest.  This includes both the measured quantities, as a test of the reliability of the calculated uncertainties, and the quantities used to evaluate irrigation strategies. We consider several evaluation criteria including yield in a single year with known weather, yield averaged over years and the inter-annual variation in yield.

We apply this approach to model-based evaluations of irrigation strategies for maize in southwest France.

A major conclusion is that the uncertainty in predicting yield averaged over years (about 0.2t/ha)  is quite a bit smaller than uncertainty in predicting yield for a single year with given weather (about 1.3t/ha). This suggests that crop models may be quite useful for comparing management strategies, since a major criterion of comparison is yield averaged over years. We also emphasize that it is essential to verify the reliability of uncertainty estimates using data, and that the conclusions only apply to the population of situations from which the data are drawn.