Weather News

AI shows promising signs with Australian tropical cyclone forecast

Ben Domensino, Friday March 22, 2024 - 14:13 AEDT


A new weather model powered by artificial intelligence has outperformed traditional weather models in predicting the location and strength of a potential tropical cyclone to the northwest of Australia.


Meteorologists were tracking the development of a tropical low off the northwest coast of Australia last week because of its potential to develop into a tropical cyclone. This low, which eventually become Tropical Cyclone Neville, was initially very difficult to predict due to forecast models struggling to align on its future track and strength.


The image below shows where four different weather models were predicting the low to be located at 11pm AEDT on Sunday, March 17, based on model runs initiated on March 12. So, this was a forecast looking five days ahead at that time.





Image: Forecast comparison for Tropical Low 08U, showing predictions for 11pm AEDT on Sunday March 17, based on model runs initiated on March 12.


The models in the top panels of the image above show tropical cyclones close to Australia’s northwest coast, while the bottom left panel also shows a tropical cyclone but much further off the coast. These three panels (top left, top right and bottom right) all show predictions from three highly regarded global-scale numerical weather prediction (NWP) models.


The bottom right panel in the image above shows a weaker low pressure system hundreds of kilometres to the northwest of Australia. This prediction came from a new AI-driven model from the European Centre for Medium-Range Weather Forecasts (ECMWF).


So, which model performed best?


The images below show where the tropical low ended up at 11pm AEDT on Sunday, March 17. The first image is an enhanced infrared satellite image showing the cloud being produced by the low, while the second image is a mean sea level pressure analysis from the Bureau of Meteorology.





Image: Enhanced infrared satellite image captured at 11pm AEDT on March 17, 2024. Source: RAMMB/CIRA





Image: Mean sea level pressure analysis chart for 11pm AEDT on March 17, 2024. Source: Bureau of Meteorology


The satellite image and MSLP chart above both show that the tropical low ended up being located close to where the ACCESS-G and ECMWF-AIFS models were predicting five days out. That’s one point for the AI model and one point for an NWP model. However, both the GFS and ECMWF-HRES models failed to accurately predict the low’s location five days in advance.


The MSLP analysis chart above also shows that the low’s central pressure was 999 hPa at 11pm AEDT on March 17. Here are the central pressures that were predicted by each model five days out:




ECMWF model: 981 hPa


GFS model: 968 hPa


ACCESS-G model: 981 hPa


ECMWF-AIFS model (AI): 997 hPa




Based on central pressure, the AI-based ECMWF-AIFS was by far the most accurate model five days out, predicting it within 2 hPa of the actual pressure. The three NWP models were out by 18 to 31 hPa. Another point for the AI model.


This case study shows that AI-based weather models have potential to be a valuable tool for tropical cyclones. However, this is only one time step from one system, so more real-world tests are needed before we can assess how useful AI-based weather models will be for operational meteorology.


- Weatherzone

© Weatherzone 2024

Site search


Enter a postcode or town name for local weather, or text to search the site. » advanced search

Summery Sydney, wintry Melbourne

15:39 AEST It's been one of those days when Australia's two largest cities could hardly have been more different in terms of weather.

Help with Farmonline Weather