Abstract Seasonal climate forecasts can improve the accuracy of early-season estimates of crop yield and influence seasonal crop management decisions. Climate forecasting centres around the globe now routinely run global climate models (GCMs) to provide ensemble forecasts. However, raw GCM forecasts require post-processing to improve their reliability and to enable systematic integration with crop models. Post-processing to meet crop model input requirements is highly challenging and simple bias-correction methods can perform poorly in this regard. As a result of the difficulties, GCM forecasts are often sidelined in favour of other inputs such as climate analogues. In this study, we evaluate two variants of a recently-developed post-processing method designed to systematically and reliably calibrate and downscale GCM forecasts for use in crop models. In one variant, local GCM forecasts of rainfall, temperature and solar radiation are post-processed directly. The second variant is a novel adaption in which the predictive input is instead the GCM's forecast of a large-scale climate pattern, in this case related to the El Nino-Southern Oscillation. The post-processed climate forecasts, which are in the form of ensemble time series, are used to drive an APSIM-sugar model to generate long-lead forecasts of biomass in north-eastern Australia from 1982 to 2016. A rigorous probabilistic assessment of forecast attributes suggests that local GCM forecast calibration provides the most skilful forecasts overall although the ENSO-related forecasts give more skilful biomass forecasts at certain times, implying model combination could be worthwhile to maximise skill. The generated biomass forecasts are unbiased and reliable for short to long lead times, suggesting that the downscaling methods will be of value to trial in a range of crop forecast applications, and support the quantitative, meaningful use of GCM forecasts in agriculture.

中文翻译：

---

用于农业应用的 GCM 气候预测校准和降尺度的改进工作流程——澳大利亚甘蔗产量预测的案例研究

摘要 季节性气候预测可以提高作物产量早期估算的准确性，并影响季节性作物管理决策。全球气候预测中心现在定期运行全球气候模型 (GCM) 以提供集合预报。然而，原始的 GCM 预测需要进行后处理以提高其可靠性并实现与作物模型的系统集成。满足作物模型输入要求的后处理极具挑战性，简单的偏差校正方法在这方面可能表现不佳。由于这些困难，GCM 的预测常常被搁置一边，转而支持气候类似物等其他输入。在这项研究中，我们评估了最近开发的后处理方法的两种变体，旨在系统地、可靠地校准和缩小 GCM 预测以用于作物模型。在一种变体中，当地 GCM 对降雨量、温度和太阳辐射的预测直接进行后处理。第二个变体是一种新的适应，其中预测输入是 GCM 对大尺度气候模式的预测，在这种情况下与厄尔尼诺 - 南方涛动有关。采用集合时间序列形式的后处理气候预测用于驱动 APSIM-糖模型，以生成 1982 年至 2016 年澳大利亚东北部生物量的长期预测。对预测属性的严格概率评估表明，尽管与 ENSO 相关的预测在某些时候提供了更熟练的生物量预测，但局部 GCM 预测校准总体上提供了最熟练的预测，这意味着模型组合可能值得最大化技能。生成的生物量预测对于短到长的提前期都是无偏见和可靠的，这表明降尺度方法对于在一系列作物预测应用中进行试验具有价值，并支持 GCM 预测在农业中的定量、有意义的使用。