[论文解读] Stable Machine-Learning Parameterization of Subgrid Processes in a Comprehensive Atmospheric Model Learned From Embedded Convection-Permitting Simulations

Zeyuan Hu, Akshay Subramaniam+7 more|arXiv (Cornell University)|Jun 28, 2024

Distributed and Parallel Computing Systems被引用 6

一句话总结

这项工作在 MMF 中训练一个稳定的 ML 参数化，仿真 cloud-resolving model 的完整物理，并实现带真实地理信息的5年在线混合气候模拟，使用具有微物理约束的 U-Net。

ABSTRACT

Modern climate projections often suffer from inadequate spatial and temporal resolution due to computational limitations, resulting in inaccurate representations of sub-grid processes. A promising technique to address this is the Multiscale Modeling Framework (MMF), which embeds a kilometer-resolution cloud-resolving model within each atmospheric column of a host climate model to replace traditional convection and cloud parameterizations. Machine learning (ML) offers a unique opportunity to make MMF more accessible by emulating the embedded cloud-resolving model and reducing its substantial computational cost. Although many studies have demonstrated proof-of-concept success of achieving stable hybrid simulations, it remains a challenge to achieve near operational-level success with real geography and comprehensive variable emulation that includes, for example, explicit cloud condensate coupling. In this study, we present a stable hybrid model capable of integrating for at least 5 years with near operational-level complexity, including coarse-grid geography, seasonality, explicit cloud condensate and wind predictions, and land coupling. Our model demonstrates skillful online performance, achieving a 5-year zonal mean tropospheric temperature bias within 2K, water vapor bias within 1 g/kg, and a precipitation RMSE of 0.96 mm/day. Key factors contributing to our online performance include an expressive U-Net architecture and physical thermodynamic constraints for microphysics. With microphysical constraints mitigating unrealistic cloud formation, our work is the first to demonstrate realistic multi-year cloud condensate climatology under the MMF framework. Despite these advances, online diagnostics reveal persistent biases in certain regions, highlighting the need for innovative strategies to further optimize online performance.

研究动机与目标

• 在现实地理条件下，激励改进气候模型中次网格对流与云的表示。
• 在 MMF 框架内开发一个 ML 代理，以全物理仿真方式仿真 cloud-resolving model (CRM)。
• 通过整合微物理约束实现长期在线稳定性和现实的云气候学。
• 比较 ML 架构（MLP 与 U-Net）及输入特征，以最大化离线和在线表现。
• 展示 ML 参数化与基于 Fortran 的气候模型耦合，并将性能与参考 MMF 模拟进行比较。

提出的方法

• 比较适用于垂直柱数据的基线 MLP 以及 1D/2D U-Net 架构。
• 扩展输入以包括大尺度强迫、对流记忆和纬度。
• 通过预测总云凝结并根据温度将其分配为液态/冰态来纳入微物理约束。
• 实现基于对流顶高度的云去除约束以提升在线稳定性。
• 在来自 10 年 E3SM-MMF 运行的 ClimSim 低分辨率真实地理数据上进行训练，并通过离线技能评估和在线 5 年模拟测试。
• 使用 Pytorch-Fortran 将经过训练的模型耦合到 E3SM 以进行基于 TorchScript 的推理。

Figure 1: Assessment of the microphysical constraints. (a) The fraction of liquid cloud over total cloud as a function of temperature. This temperature-based phase partitioning holds exactly on each grid in the cloud-resolving model. (b) Histogram of the actual liquid cloud fraction in the E3SM grid

实验结果

研究问题

• RQ1 在真实地理条件下，能够用机器学习参数化在 MMF 内部多年度模拟中仿真 CRM 吗？
• RQ2 扩展输入的 U-Net 能否在 MMF 模拟中优于基线 MLP，在离线和在线技能方面？
• RQ3 对云凝结的微物理约束是否改善在线稳定性和云气候学的现实性？
• RQ4 相对于参考 MMF 模拟，在线 ML-MMF 模拟的长期偏差和变异性是多少？
• RQ5 纳入大尺度强迫和对流记忆如何影响模型性能与稳定性？

主要发现

• 在受约束的 U-Net 下，利用真实地理信息和全物理仿真的稳定 5 年混合 MMF 模拟已实现。
• 受约束的 U-Net 显示出改进的在线稳定性并减少偏差，5 年尺度经向平均温度偏差 <2 K 且水汽偏差 <1 g/kg。
• 全球经向平均风偏差维持在 ~5 m/s 左右，云分布保持现实。
• 离线结果显示，扩展输入的 U-Net 的表现优于 MLP，尤其在云凝结物和风向趋势方面。
• 未约束的 U-Net 可能产生非物理的云相以及过量的高层大气液态/冰态云，强调了微物理约束的重要性。
• 五年气候态趋势和降水统计在关键模式上高度相似于参考 MMF，但某些区域/极端降水尾部仍需进一步改进。

Figure 2: (a) Offline R 2 scores across various variables for MLP, U-Net, and U-Net with physics constraints. Variables are the full target variables listed in Table 2 , including temperature tendency ( $\frac{dT}{dt}$ ), water vapor tendency ( $\frac{dQ_{\text{v}}}{dt}$ ), liquid cloud mixing ratio