Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Probabilistic Weather Forecasting with Hierarchical Graph Neural Networks

Joel Oskarsson, Tomas Landelius, Marc Peter Deisenroth, Fredrik Lindsten

TL;DR

This work proposes a probabilistic weather forecasting model called Graph-EFM, combining a flexible latent-variable formulation with the successful graph-based forecasting framework, requiring only a single forward pass per time step and allowing for fast generation of arbitrarily large ensembles.

Abstract

In recent years, machine learning has established itself as a powerful tool for high-resolution weather forecasting. While most current machine learning models focus on deterministic forecasts, accurately capturing the uncertainty in the chaotic weather system calls for probabilistic modeling. We propose a probabilistic weather forecasting model called Graph-EFM, combining a flexible latent-variable formulation with the successful graph-based forecasting framework. The use of a hierarchical graph construction allows for efficient sampling of spatially coherent forecasts. Requiring only a single forward pass per time step, Graph-EFM allows for fast generation of arbitrarily large ensembles. We experiment with the model on both global and limited area forecasting. Ensemble forecasts from Graph-EFM achieve equivalent or lower errors than comparable deterministic models, with the added benefit of accurately capturing forecast uncertainty.

Probabilistic Weather Forecasting with Hierarchical Graph Neural Networks

TL;DR

This work proposes a probabilistic weather forecasting model called Graph-EFM, combining a flexible latent-variable formulation with the successful graph-based forecasting framework, requiring only a single forward pass per time step and allowing for fast generation of arbitrarily large ensembles.

Abstract

In recent years, machine learning has established itself as a powerful tool for high-resolution weather forecasting. While most current machine learning models focus on deterministic forecasts, accurately capturing the uncertainty in the chaotic weather system calls for probabilistic modeling. We propose a probabilistic weather forecasting model called Graph-EFM, combining a flexible latent-variable formulation with the successful graph-based forecasting framework. The use of a hierarchical graph construction allows for efficient sampling of spatially coherent forecasts. Requiring only a single forward pass per time step, Graph-EFM allows for fast generation of arbitrarily large ensembles. We experiment with the model on both global and limited area forecasting. Ensemble forecasts from Graph-EFM achieve equivalent or lower errors than comparable deterministic models, with the added benefit of accurately capturing forecast uncertainty.
Paper Structure (97 sections, 22 equations, 51 figures, 13 tables, 5 algorithms)

This paper contains 97 sections, 22 equations, 51 figures, 13 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deterministic
  4. Ensembles from perturbations
  5. Generative modeling
  6. Hierarchical
  7. Background
  8. Problem Definition
  9. Graph-based Weather Forecasting
  10. Interaction Networks
  11. Global mesh graphs
  12. Weather Forecasting with Hierarchical Graph Neural Networks
  13. Hierarchical Graph
  14. Graph-FM: Deterministic Forecasting
  15. Graph-EFM: Probabilistic Forecasting
  16. ...and 82 more sections

Figures (51)

  • Figure 1: Overview of our Graph-EFM model, with example data and graphs for a LAM. The corresponding overview for the global setting is given in \ref{['fig:model_overview_global']} in \ref{['sec:model_details']}.
  • Figure 2: Graphical model for \ref{['eq:latent_var']}.
  • Figure 3: Results for global forecasting of mean sea level pressure (msl) at all lead times.
  • Figure 4: Example Graph-EFM ensemble forecast for specific humidity at 700 (q700), for lead time 10 days. The bottom row shows 3 ensemble members, randomly chosen out of the 80.
  • Figure 5: Example forecasts for net solar longwave radiation (nlwrs) at lead time 57.
  • ...and 46 more figures