Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An ensemble of data-driven weather prediction models for operational sub-seasonal forecasting

Jonathan A. Weyn, Divya Kumar, Jeremy Berman, Najeeb Kazmi, Sylwester Klocek, Pete Luferenko, Kit Thambiratnam

TL;DR

This work presents an operational multi-model ensemble of data-driven weather prediction models coupled to an ocean model to deliver global forecasts at $1^{\circ}$ for up to $4$ weeks. By training five diverse data-driven architectures and using autoregressive forecasting with SST coupling, the authors demonstrate near-state-of-the-art subseasonal forecasts and show that probabilistic forecasts can be competitive with, and occasionally exceed, the ECMWF extended-range ensemble. They implement a principled hindcast-based bias correction and show that combining data-driven models with traditional NWP improves reliability and spread calibration. The approach highlights the practical potential of data-driven ensembles for operational S2S forecasting, while acknowledging limitations in extreme-event evaluation, precipitation skill, and the need for further methodological refinements.

Abstract

We present an operations-ready multi-model ensemble weather forecasting system which uses hybrid data-driven weather prediction models coupled with the European Centre for Medium-range Weather Forecasts (ECMWF) ocean model to predict global weather at 1-degree resolution for 4 weeks of lead time. For predictions of 2-meter temperature, our ensemble on average outperforms the raw ECMWF extended-range ensemble by 4-17%, depending on the lead time. However, after applying statistical bias corrections, the ECMWF ensemble is about 3% better at 4 weeks. For other surface parameters, our ensemble is also within a few percentage points of ECMWF's ensemble. We demonstrate that it is possible to achieve near-state-of-the-art subseasonal-to-seasonal forecasts using a multi-model ensembling approach with data-driven weather prediction models.

An ensemble of data-driven weather prediction models for operational sub-seasonal forecasting

TL;DR

This work presents an operational multi-model ensemble of data-driven weather prediction models coupled to an ocean model to deliver global forecasts at for up to weeks. By training five diverse data-driven architectures and using autoregressive forecasting with SST coupling, the authors demonstrate near-state-of-the-art subseasonal forecasts and show that probabilistic forecasts can be competitive with, and occasionally exceed, the ECMWF extended-range ensemble. They implement a principled hindcast-based bias correction and show that combining data-driven models with traditional NWP improves reliability and spread calibration. The approach highlights the practical potential of data-driven ensembles for operational S2S forecasting, while acknowledging limitations in extreme-event evaluation, precipitation skill, and the need for further methodological refinements.

Abstract

We present an operations-ready multi-model ensemble weather forecasting system which uses hybrid data-driven weather prediction models coupled with the European Centre for Medium-range Weather Forecasts (ECMWF) ocean model to predict global weather at 1-degree resolution for 4 weeks of lead time. For predictions of 2-meter temperature, our ensemble on average outperforms the raw ECMWF extended-range ensemble by 4-17%, depending on the lead time. However, after applying statistical bias corrections, the ECMWF ensemble is about 3% better at 4 weeks. For other surface parameters, our ensemble is also within a few percentage points of ECMWF's ensemble. We demonstrate that it is possible to achieve near-state-of-the-art subseasonal-to-seasonal forecasts using a multi-model ensembling approach with data-driven weather prediction models.
Paper Structure (8 sections, 5 figures)

This paper contains 8 sections, 5 figures.

Table of Contents

  1. Introduction
  2. Methodology
  3. Data
  4. Models
  5. Training strategy
  6. Statistical bias correction
  7. Results
  8. Discussion

Figures (5)

  • Figure 1: Spatial distribution of the bias in 2-meter temperature. The bias is computed as the difference between the ensemble mean of the operational forecast and the ERA5 reanalysis. Numbers in each subplot are the global average.
  • Figure 2: CRPS and ESSR for 2-meter temperature. (a) CRPS for 2-meter temperature, as a function of lead time (weeks). (b) Ratio of ensemble spread to skill.
  • Figure 3: Spatial distribution of the CRPS score for 2-meter temperature. The CRPS indicates the quality of the probabilistic forecasts (lower is better) and reduces to the mean absolute error for a deterministic forecast.
  • Figure 4: Rank histogram for 2-meter temperature for each ensemble. The rank histogram shows the distribution of the rank of the operational analysis verification within the ensemble.
  • Figure 5: CRPS for other surface parameters. (a) CRPS for 2-meter dewpoint temperature, as a function of lead time (weeks). (b) As in (a) but for total cloud cover.