CR 51:11-21 (2012)  -  DOI: https://doi.org/10.3354/cr01057

ABSTRACT: Statistical models are common tools for quantifying possible impacts of climate change on crops. Climate change involves shifts in both mean and variability of climate parameters, and experimental results and simulations have shown that both mean and variability can have the same-order effects on crop growth and yield. It is therefore important for impact models to be able to capture the effects of both these aspects. The aim of the present study was to test the ability of statistical crop models to predict the effects of changes in mean and variability of temperature and precipitation on grain yield of maize. Climate variables were aggregated over different time intervals to explore effects of temporal aggregation of predictor variables. To examine the predictive capabilities of statistical crop models beyond the ranges of observed data, we applied the ‘perfect model’ approach using the process-based crop model CropSyst. Statistical crop models were then fitted based on different sample sizes to explore minimum data requirements for predicting the effects of different synthetic climate scenarios. The analysis revealed that statistical crop models are generally able to quantify the effects of changes in mean and variability of temperature and precipitation with reasonable accuracy, provided that a minimum of 10 to 20 samples per predictor are available for fitting. Also, we can conclude that with total sample sizes of <300 observations, disaggregation of predictor variables increases the risk of model over-fitting. Disaggregation of predictor variables had small beneficial effects, particularly for ­predicting impacts of changes in climate variability, only with large sample sizes (>300 ob­servations).

KEY WORDS: Statistical crop model · Climate change · Impact assessment · CropSyst · LARS-WG