Detection of major weather patterns reduces number of simulations in climate impact studies

With climate change posing a serious threat to food security, there has been an increased interest in simulating its impact on cropping systems. Crop models are useful tools to evaluate strategies for adaptation to future climate; however, the simulation process may be infeasible when dealing with a large number of G x E x M combinations. We proposed that the number of simulations could significantly be reduced by clustering weather data and detecting major weather patterns. Using 5, 10 and 15 clusters (i.e., years representative of each weather pattern), we simulated phenology, cumulative transpiration, heat-shock-induced yield loss (heat loss) and grain yield of four Australian cultivars across the Australian wheatbelt over a 30-year period under current and future climates. A strong correlation (r(2)approximate to 1) between the proposed method and the benchmark (i.e., simulation of all 30 years without clustering) for phenology suggested that average duration of crop growth phases could be predicted with substantially fewer simulations as accurately as when simulating all 30 seasons. With mean absolute error of 0.64 days for phenology when only five clusters were identified, this method had a deviation considerably lower than the reported deviations of calibrated crop models. Although the proposed method showed higher deviations for traits highly sensitive to temporal climatic variability such as cumulative transpiration, heat loss and grain yield when five clusters were used, significantly strong correlations were achieved when 10 or 15 clusters were identified. Furthermore, this method was highly accurate in reproducing site-level impact of climate change. Less than 7% of site x general circulation model (GCM) combinations (zero for phenology) showed incorrect predication of the direction (+/-) of climate change impact when only five clusters were identified while the accuracy further increased at the regional level and with more clusters. The proposed method proved promising in predicting selected traits of wheat crops and can reduce number of simulations required to predict crop responses to climate/management scenarios in model-aided ideotyping and climate impact studies.