[论文解读] Regional data-driven weather modeling with a global stretched-grid

Thomas N. Nipen, Håvard Homleid Haugen+12 more|arXiv (Cornell University)|Sep 4, 2024

Distributed and Parallel Computing Systems被引用 5

一句话总结

使用基于图神经网络的全局拉伸网格数据驱动天气模型，在北欧区域提供高分辨率预报（2.5 公里），以 ERA5 和 MEPS 数据为训练，并与 MEPS 与 IFS 进行评估。对温度和降水具有竞争力，并突显通过拉伸网格实现区域-全球无边界耦合。

ABSTRACT

A data-driven model (DDM) suitable for regional weather forecasting applications is presented. The model extends the Artificial Intelligence Forecasting System by introducing a stretched-grid architecture that dedicates higher resolution over a regional area of interest and maintains a lower resolution elsewhere on the globe. The model is based on graph neural networks, which naturally affords arbitrary multi-resolution grid configurations. The model is applied to short-range weather prediction for the Nordics, producing forecasts at 2.5 km spatial and 6 h temporal resolution. The model is pre-trained on 43 years of global ERA5 data at 31 km resolution and is further refined using 3.3 years of 2.5 km resolution operational analyses from the MetCoOp Ensemble Prediction System (MEPS). The performance of the model is evaluated using surface observations from measurement stations across Norway and is compared to short-range weather forecasts from MEPS. The DDM outperforms both the control run and the ensemble mean of MEPS for 2 m temperature. The model also produces competitive precipitation and wind speed forecasts, but is shown to underestimate extreme events.

研究动机与目标

• 激励区域天气预报需求，用区域数据驱动模型为像 Yr 这样的高分辨率公共服务提供支持。
• 开发全球拉伸网格架构，在保持全球覆盖的同时，将区域区域集中特更高分辨率。
• 基于历史 ERA5 和 MEPS 数据训练并验证一个基于图神经网络的模型，以预测短期天气参数。
• 评估该模型在挪威及北欧地区相对于运行中的数值天气预报系统和观测的准确性与可靠性。

提出的方法

• 使用具有编码器-处理器-解码器架构的图神经网络，将输入网格数据映射到具多分辨率细化的潜在处理网格。
• 构建具有更高区域分辨率（2.5 公里）和较低全球分辨率的拉伸网格处理网格，实现跨域的无缝系统移动。
• 四阶段训练：ERA5 100 公里（Stage A），ERA5 31 公里（Stage B），结合 IFS+MEPS 的 31 公里和 2.5 公里区域（Stage C），以及 24 小时自回归展开（Stage D）。
• 采用强调区域域点的损失（占损失的 33%），尽管区域面积仅占全球的 1.2%；使用平方误差和逐变量权重。
• 使用 43 年 ERA5 数据进行预训练，然后以 MEPS 数据进行微调，以利用区域高分辨率信息。

Figure 1: (a) Map with annotated grid points centered around the Nordics. Global grid points are green, regional grid points are gray. (b) Input grid on the boundary between global and regional domain. (c) The encoder processes information into a mesh node from the 12 nearest grid points. (d) The pr

实验结果

研究问题

• RQ1 Can a global stretched-grid data-driven model provide accurate, high-resolution regional forecasts for the Nordics while maintaining seamless interaction with the global domain?
• RQ2 How does a GNN-based stretched-grid model perform in temperature, wind, and precipitation forecasts compared with MEPS and IFS at comparable lead times?
• RQ3 What training strategy and data sources yield the best regional performance for a stretched-grid DDM?
• RQ4 What are the strengths and limitations of this approach in representing mountainous and coastal climate features?

主要发现

• The stretched-grid DDM outperforms MEPS control for 2 m temperature RMSE across lead times and for 6 h precipitation RMSE.
• For wind, the DDM is competitive with MEPS control and ensemble mean, but under more extreme thresholds it underperforms relative to ensemble smoothing.
• The DDM demonstrates the ability to move weather systems seamlessly between global and regional domains, with comparable large-scale features to IFS and MEPS in several cases.
• The model shows strong regional performance for temperature but tends to underestimate extreme wind and precipitation events, suggesting post-processing or ensemble approaches could improve extremes.
• Training on ERA5 global data followed by fine-tuning with MEPS regional data yields the best regional performance, using 2.5 km regional resolution within a global stretched grid.

Figure 2: Model training follows a four-stage procedure. First, the DDM is pre-trained on 43 years of ERA5 data with a global resolution of 100 km (stage A) and 31 km (stage B). In stage C, we combine the 31 km global IFS dataset with the 2.5 km regional MEPS dataset with a training period of 3.3 ye