[Stochastic representation of the influence of the subgrid‐scales on the resolved scales in weather and climate models has been shown to improve ensemble spread and resolved variability. We propose a statistical scale‐aware space‐time model to characterize the contribution of mesoscale wind variability to air‐sea exchanges. In an earlier study, we analyzed the difference between “true” fluxes computed from a high resolution simulation and “resolved” fluxes obtained by coarse graining. This discrepancy is modeled in space and time, conditioned on the coarse‐grained wind and precipitation fields, to parameterize the enhancement of fluxes by mesoscale velocity variations. Stochastic parameterization models have traditionally been developed for particular model resolutions without the explicit capability to adapt to model resolution. We present an approach to develop stochastic models that adapt to resolution in a scale‐aware fashion. The scale‐aware parameterization is developed from empirical results for systematically coarse‐grained high‐resolution numerical model output. The statistical model is fit from numerical model output at three different coarsening resolutions. From this scale‐aware parameterization, we derive a stochastic parameterization of flux enhancement by subgrid velocity variations for arbitrary resolutions and characterize the conditional distributions and space‐time structures of the flux enhancement across model resolutions. ]

中文翻译：

---

海面通量亚网格尺度速度增强的尺度感知时空随机参数化

[在天气和气候模型中，随机表示次网格规模对已分解规模的影响可改善集合传播和已分解变异性。我们提出了一个统计尺度感知时空模型，以表征中尺度风变率对海-气交换的贡献。在较早的研究中，我们分析了通过高分辨率模拟计算出的“真实”通量与通过粗粒度获得的“分辨”通量之间的差异。这种差异是在时空上建模的，其条件是粗糙的风场和降水场，以通过中尺度速度变化来参数化通量的增强。传统上已经针对特定的模型分辨率开发了随机参数化模型，而没有明确的能力来适应模型分辨率。我们提出了一种方法，可以以规模感知的方式开发适应分辨率的随机模型。基于经验的结果开发了可感知规模的参数化，用于系统粗粒度的高分辨率数值模型输出。统计模型是从数值模型输出中以三种不同的粗化分辨率进行拟合的。从这种尺度感知的参数化中，我们得出了子网格速度变化对任意分辨率的通量增强的随机参数化，并描述了模型分辨率下通量增强的条件分布和时空结构。] 统计模型是从数值模型输出中以三种不同的粗化分辨率进行拟合的。从这种尺度感知的参数化中，我们得出了子网格速度变化对任意分辨率的通量增强的随机参数化，并描述了模型分辨率下通量增强的条件分布和时空结构。] 统计模型是从数值模型输出中以三种不同的粗化分辨率进行拟合的。从这种尺度感知的参数化中，我们得出了子网格速度变化对任意分辨率的通量增强的随机参数化，并描述了模型分辨率下通量增强的条件分布和时空结构。]