Predicting extreme weather using data science
Dark clouds and strong winds, usually indicators of severe weather to come. In the old days, people did not have other than this information to act upon to prevent damage as much as possible. Later, new means of communication came into use to warn people at an earlier stage for extreme weather, when at site no sign was visible. That was a big step forward.
Nowadays, atmospheric conditions are monitored at a massive scale and intensively in time. Think of data gathered from weather stations, satellite images, weather radar images, reports from ships, airports and amateur meteorologist. But the question remains, can these data streams be used to accurately predict extreme weather events?
Geurt Jongbloed, Professor of Statistics and chairman of the Department of Applied Mathematics (DIAM), wraps his brain around this question. Together with assistant professor Juan-Juan Cai, PhD student Jasper Velthoen and the Royal Netherlands Meteorological Institute (KNMI) post-processing department, he works on methods to improve the quality of extreme weather predictions using all sorts of data available. These data consist of actual measurements, but also on outputs generated by the weather prediction models in the past. Elegant mathematical results from a subfield of statistics called ´Extreme Value Analysis’ can be adapted and used to specifically improve the prediction of extremes.
KNMI & mathematics
At KNMI, sophisticated models from mathematical physics are used to forecast the weather using a combination of physical models and measured data on huge computer systems. “For normal weather, the quality of the forecast for a couple of days in advance is strikingly high. If it comes to predicting extreme weather events, there is definitely need and room for improvement”, says Geurt Jongbloed.
How did you get in contact with KNMI on this subject?
“I supervised a PhD student jointly with people from the climate department at KNMI in the past. The meteorology department of KNMI has contact with water authorities, ProRail and commercial forecasting services and the need for improvement of extreme weather warning systems was broadly felt. Our Delft track record in the field of extreme value theory combined with fruitful collaborations with other department within KNMI in the past, lead to the current collaboration.”
What has been achieved thus far in this project?
We developed a model that uses weather prediction model outputs to estimate probabilities of extreme events. These are the ingredients for the colour coded alarm system used at KNMI. First results show improvement on the currently used approach.
What’s next?
“A next step is to include more of the data available in our model. Furthermore, especially for the water authorities there is also a strong need to consider predictions for ‘joint extremes’. It is possible that strong wind and high rainfall alone do not ask for preventive measures, but that the combination of the two does. Then it is important to model these quantities jointly, rather than separately. A nice aspect about this, is that it’s also at the frontline of current theoretical research in extreme value theory. Working on mathematically challenging problems with societal relevance. That really fits within our Delft Institute of Applied Mathematics!”