Climate model forecast biases assessed with a perturbed physics ensemble
Climate model forecast biases assessed with a perturbed physics ensemble
复制标题
DOI：
10.1007/s00382-016-3407-x
复制
发表时间：
2017-09
期刊：
影响因子：
4.6
通讯作者：
D. Mulholland;K. Haines;S. Sparrow;D. Wallom
中科院分区：
文献类型：
--
作者：
D. Mulholland;K. Haines;S. Sparrow;D. Wallom
扰动物理学系综常被用来分析长时间尺度的气候模式行为，但较少被用来研究较短时间尺度上的模式过程。我们将瞬变扰动物理系综和一组初始化预报结合起来，以推导出标准HadCM3模式中存在的区域过程误差，这些误差导致模型在预报的早期阶段漂移。首先，证明了微扰物理系综中的瞬时漂移可以用来定量恢复被微扰的参数。对漂移影响最大的参数在区域上有所不同，但在1个月的时间尺度上，上层海洋混合和大气对流过程尤为重要。然后，初始化预报中的漂移被用来恢复“等效参数扰动”，这使得能够识别在真实世界的HadCM3表示中可能出错的物理过程。大多数参数在不同地区显示正负调整，表明标准HadCM3值代表全局折衷。该方法是通过纠正风引起的海洋混合强度的异常广泛的正偏差来验证的，结果是在许多地区减少了预测漂移。因此，这种方法可以用来通过对物理过程的区域调整来减少模型偏差，从而提高初始化气候模式预报的技能，无论是通过调整还是通过有针对性的参数化精细化。此外，在较长期的气候研究中，这种地区性调整的模型也可能显著优于具有全球参数配置的标准气候模型。
Perturbed physics ensembles have often been used to analyse long-timescale climate model behaviour, but have been used less often to study model processes on shorter timescales. We combine a transient perturbed physics ensemble with a set of initialised forecasts to deduce regional process errors present in the standard HadCM3 model, which cause the model to drift in the early stages of the forecast. First, it is shown that the transient drifts in the perturbed physics ensembles can be used to recover quantitatively the parameters that were perturbed. The parameters which exert most influence on the drifts vary regionally, but upper ocean mixing and atmospheric convective processes are particularly important on the 1-month timescale. Drifts in the initialised forecasts are then used to recover the ‘equivalent parameter perturbations’, which allow identification of the physical processes that may be at fault in the HadCM3 representation of the real world. Most parameters show positive and negative adjustments in different regions, indicating that standard HadCM3 values represent a global compromise. The method is verified by correcting an unusually widespread positive bias in the strength of wind-driven ocean mixing, with forecast drifts reduced in a large number of areas as a result. This method could therefore be used to improve the skill of initialised climate model forecasts by reducing model biases through regional adjustments to physical processes, either by tuning or targeted parametrisation refinement. Further, such regionally tuned models might also significantly outperform standard climate models, with global parameter configurations, in longer-term climate studies.