Comparison of temperatures simulated by GCMs, RCMs and statistical downscaling: potential application in studies of future crop development

We evaluated the performance of a general circulation model (HadCM3), a regional circulation model (HadRM3P) and an artificial neural network (ANN), in reproducing daily maximum and minimum temperature (T-max and T-min) at site scale (Florence, Italy) for the present climate. The T-max and T-min values that were observed and those reproduced by HadCM3, HadRM3P and ANN for both the present and future climate scenarios (IPCC scenarios A2 and B2) were then used as input data in a cropping systems simulation model (CropSyst). In particular, climatic impact on the phenological developmental stages of a summer crop (sunflower Helianthus annuus L.) and winter crop (durum wheat Triticum aestivum L.) were evaluated. In addition, the frequency of extreme climatic events during specific crop phenological stages (i.e. number of events with T-max and T-min above and below stressful thresholds) were evaluated. The comparison between observed T-max and T-min, values and those produced by HadCM3, HadRM3P and ANN for the present climate, provided evidence for a higher accuracy of the ANN model in simulating these variables. The crop phenological stages and the related extreme climate events were therefore also better reproduced using the ANN climate data. The use of HadCM3 and HadRM3P climate data in climate change impact assessments seemed to result in an overestimation of the impacts (i.e. greater reduction in the length of development phases and greater changes in the frequency of extreme climate events during the most sensitive development stages) compared with those obtained using ANN climate data.