Development of Predictive Models: Medium Range and Sub seasonal Forecasting

Argonne is using vision transformers to build highly accurate medium-range weather forecasting models. They run at a fraction of the cost of traditional Numerical Weather Prediction and will help the energy industry respond to weather extremes.

Argonne used observation-based reanalysis to train a custom, weather-specific vision transformer (HR-Stormer) to predict the global atmosphere 14 days into the future. Using novel training techniques and creating a model-based ensemble system, Stormer produces more accurate forecasts at unprecedented speed. Once trained, our model can produce a 14-day forecast composed of 32 ensemble members in under a minute, something not feasible using previous methods.

We compared Stormer to existing models and found it performs competitively at short to medium range forecasts and outperforms current models beyond 7 days. The accelerated speed of our model will allow for new possibilities in data assimilation, ensemble forecasting, and uncertainty quantification.

Models of weather data — Fig 1. Illustration of an example 5-day forecast of near-surface wind speed (color-fill) and mean sea level pressure (contours). On December 31, 2020, an extratropical cyclone impacted Alaska setting a new North Pacific low-pressure record. Here, we evaluate the ability of Stormer to predict this record-breaking event 5 days in advance. Using initial conditions from 0000 UTC, 26 December 2011, Stormer was able to successfully forecast both the location and strength of this extreme event with great accuracy.

This model system is now being extended to develop forecasting ability to sub-seasonal scales and beyond, which is critical in planning for energy system resilience and expected energy demand. These models will allow for the integration of weather into energy demand forecast models that are critical for planning natural gas needs in the near-term.

Additionally, as the forecasting at these time scales has lower certainty compared to the next few days forecast, it is critical that these models have probabilistic forecasting capability. We are building probabilistic models for the sub-seasonal to seasonal (S2S) forecasts.