Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

WeatherGNN: Exploiting Meteo- and Spatial-Dependencies for Local Numerical Weather Prediction Bias-Correction

Binqing Wu, Weiqi Chen, Wengwei Wang, Bingqing Peng, Liang Sun, Ling Chen

TL;DR

WeatherGNN tackles local NWP bias by modeling both area-specific meteorological dependencies and cross-area spatial influences. It introduces a factor GNN to learn region-adaptive meteorological interactions and a fast hierarchical GNN to capture dynamic, multi-scale spatial dependencies under Tobler's laws, with linear-time complexity in the number of grids. Empirically, it achieves state-of-the-art RMSE improvements (average $4.75\%$) on Ningbo and Ningxia, supported by thorough ablations showing the necessity of both branches and the static-dynamic hierarchy. The approach is scalable, geometry-informed, and capable of improving downstream applications such as wind power forecasting and regional weather decision-making.

Abstract

Due to insufficient local area information, numerical weather prediction (NWP) may yield biases for specific areas. Previous studies correct biases mainly by employing handcrafted features or applying data-driven methods intuitively, overlooking the complicated dependencies between weather factors and between areas. To address this issue, we propose WeatherGNN, a local NWP bias-correction method that utilizes Graph Neural Networks (GNNs) to exploit meteorological dependencies and spatial dependencies under the guidance of domain knowledge. Specifically, we introduce a factor GNN to capture area-specific meteorological dependencies adaptively based on spatial heterogeneity and a fast hierarchical GNN to capture dynamic spatial dependencies efficiently guided by Tobler's first and second laws of geography. Our experimental results on two real-world datasets demonstrate that WeatherGNN achieves the state-of-the-art performance, outperforming the best baseline with an average of 4.75 \% on RMSE.

WeatherGNN: Exploiting Meteo- and Spatial-Dependencies for Local Numerical Weather Prediction Bias-Correction

TL;DR

WeatherGNN tackles local NWP bias by modeling both area-specific meteorological dependencies and cross-area spatial influences. It introduces a factor GNN to learn region-adaptive meteorological interactions and a fast hierarchical GNN to capture dynamic, multi-scale spatial dependencies under Tobler's laws, with linear-time complexity in the number of grids. Empirically, it achieves state-of-the-art RMSE improvements (average ) on Ningbo and Ningxia, supported by thorough ablations showing the necessity of both branches and the static-dynamic hierarchy. The approach is scalable, geometry-informed, and capable of improving downstream applications such as wind power forecasting and regional weather decision-making.

Abstract

Due to insufficient local area information, numerical weather prediction (NWP) may yield biases for specific areas. Previous studies correct biases mainly by employing handcrafted features or applying data-driven methods intuitively, overlooking the complicated dependencies between weather factors and between areas. To address this issue, we propose WeatherGNN, a local NWP bias-correction method that utilizes Graph Neural Networks (GNNs) to exploit meteorological dependencies and spatial dependencies under the guidance of domain knowledge. Specifically, we introduce a factor GNN to capture area-specific meteorological dependencies adaptively based on spatial heterogeneity and a fast hierarchical GNN to capture dynamic spatial dependencies efficiently guided by Tobler's first and second laws of geography. Our experimental results on two real-world datasets demonstrate that WeatherGNN achieves the state-of-the-art performance, outperforming the best baseline with an average of 4.75 \% on RMSE.
Paper Structure (14 sections, 1 theorem, 10 equations, 8 figures, 2 tables)

This paper contains 14 sections, 1 theorem, 10 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. WeatherGNN
  4. Problem Definition
  5. Model Overview
  6. Learning Area-specific Meteorological Dependencies via Factor GNN
  7. Learning Dynamic Spatial Dependencies via Fast Hierarchical GNN
  8. Bias-Correction Output
  9. Experiments
  10. Datasets
  11. Baselines and Experimental Settings
  12. Performance Comparison
  13. Model Analysis
  14. Conclusion and Future Work

Key Result

Lemma 1

The complexity of factor GNN and fast hierarchical GNN is $O(N)$, where $N$ is the number of grids.

Figures (8)

  • Figure 1: (a) Pearson Correlation Matrices between Weather Factors at Different Grids. (b) Locations and Terrains of Two Grids. Grid A is on an uphill, while B is in a valley. The distance between Grid A and B is around 18 kilometers.
  • Figure 2: (a) Grid Distribution and Terrain of Ningbo Dataset. (b) DTW Similarity of Half-month 100m Wind Speed between Grid P and Other Grids. (c) DTW Similarity of 100m Wind Speed with a Six-hour Time Range between Grid P and Grid Q.
  • Figure 3: An Overview of WeatherGNN.
  • Figure 4: An Overview of Fast Hierarchical Message Passing.
  • Figure 5: Results of Hyperparameter Study
  • ...and 3 more figures

Theorems & Definitions (1)

  • Lemma 1