Continuous Ensemble Weather Forecasting with Diffusion models

Martin Andrae1
Tomas Landelius2
Joel Oskarsson1
Fredrik Lindsten1

1Div of Statistics and Machine Learning, Dept of Comp and Info Sci, Linköping Univ, Sweden
2Swedish Meteorological and Hydrological Institute, Norrköping, Sweden

[Paper] -->
[Code]


Example forecasts for WeatherBench

Example 10-day forecasts with 6h resolution and 50 ensemble members.
Forecasts are done with the ARCI-24/6h model, iterating on 24h and interpolating on 6h.

Temperature at 850hPa (t850)


Geopotential at 500hPa (z500)


Temperature at 2m (t2m)


U-wind component (u10)


V-wind component (v10)


Windspeed at 10m (ws10)

Example forecasts for Quasi- Geostrophic model


Truth
Sample #1
Sample #2
Sample #3
Continuous Ensemble Forecasting (Our method)


Truth
Sample #1
Sample #2
Sample #3
Iterative Forecasting (Same network but sampled autoreressively)

This website was made using a template which can be found here.