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 × E × 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 (r2≈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 × 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.

中文翻译：

---

主要天气模式的检测减少了气候影响研究中的模拟次数

随着气候变化对粮食安全构成严重威胁，人们越来越关注模拟其对种植系统的影响。作物模型是评估适应未来气候战略的有用工具；然而，当处理大量 G × E × M 组合时，模拟过程可能不可行。我们建议通过对天气数据进行聚类和检测主要天气模式可以显着减少模拟次数。使用 5、10 和 15 个集群（即代表每种天气模式的年份），我们模拟了澳大利亚小麦带上 30 年以上四个澳大利亚品种的物候、累积蒸腾作用、热冲击引起的产量损失（热损失）和谷物产量。 - 当前和未来气候下的年份。所提出的方法与物候学的基准（即所有 30 年没有聚类的模拟）之间的强相关性 (r2≈1) 表明，可以像模拟所有 30 年一样准确地预测作物生长阶段的平均持续时间，而模拟次数要少得多季节。当仅识别出五个集群时，物候学的平均绝对误差为 0.64 天，该方法的偏差远低于校准作物模型报告的偏差。尽管当使用五个聚类时，所提出的方法对时间气候变化高度敏感的性状（例如累积蒸腾、热量损失和谷物产量）显示出更高的偏差，但当识别出 10 或 15 个聚类时，实现了显着的强相关性。此外，这种方法在再现气候变化的站点级影响方面非常准确。当仅识别出五个集群时，不到 7% 的站点 × 大气环流模型 (GCM) 组合（物候学为零）对气候变化影响的方向 (+/-) 的预测不正确，而准确度在区域级别和有更多的集群。所提出的方法被证明在预测小麦作物的选定性状方面很有前景，并且可以减少在模型辅助的意识形态和气候影响研究中预测作物对气候/管理情景的反应所需的模拟数量。当仅识别出五个集群时，不到 7% 的站点 × 大气环流模型 (GCM) 组合（物候学为零）对气候变化影响的方向 (+/-) 的预测不正确，而在区域层面上的准确性进一步提高，有更多的集群。所提出的方法被证明在预测小麦作物的选定性状方面很有前景，并且可以减少在模型辅助的意识形态和气候影响研究中预测作物对气候/管理情景的反应所需的模拟数量。当仅识别出五个集群时，不到 7% 的站点 × 大气环流模型 (GCM) 组合（物候学为零）对气候变化影响的方向 (+/-) 的预测不正确，而在区域层面上的准确性进一步提高，有更多的集群。所提出的方法被证明在预测小麦作物的选定性状方面很有前景，并且可以减少在模型辅助的意识形态和气候影响研究中预测作物对气候/管理情景的反应所需的模拟数量。