Topic 5F - ML For More Accurate Weather Forecasts

Comparison between Met Norway in situ observations and Netatmo observations for precipitation during an extreme rain event. Netatmos observations show a good agreement with MetNorway and cover a far greater area

Met Norway

Oslo 4 August 2019 (right) 10-min snapshots of Netatmo rainfall (left) same as right but also with Netatmo MSLP anomalies (shaded) and radar precip intensity (contours)

Met Norway/Roel Stappers/Thomas Nipen

High-resolution weather phenomena

Met Norway

This image shows the raw NWP model image at 2.5km resolution (left), the statistical downscaling process (middle), bias correction - adding observations(right)

Met Norway

These images show the downscaling process along a coastline. The raw model output is on the right, and the image on the left shows coastal gradient correction. Temperatures on the coast can be up to 10 degrees wrong - so land/ocean fraction can be used when interpolating to finer resolution

Met Norway

Example of direct weather prediction for 5 days lead time. The input to the neural network are fields on a latitude-longitude grid. The fields can be several levels of the same variable and/or different variables. The goal is to predict the same fields some time ahead. (b) Iterative forecasts are created from data-driven models trained on a shorter lead time, for example, 6 hr, which are then iteratively called up to the required forecast lead time.

Stephan Rasp, Peter Dueben et al., 2020

This is an Extreme forecast index (EFI) of weekly precipitation in Europe

ECMWF

This is an ECMWF Mean sea level pressure: Probability distribution forecast

ECMWF

This charts show 500 hPa geopotential height (contours) and temperature at 850 hPa (shading) from the ECMWF HRES model.

ECMWF

This is an example of a 2m temperature (colour shading) and 30m winds (black arrows) forecast

ECMWF