Extreme weather events driven by climate change are intensifying, and have the potential to wreak havoc around the world, particularly in economically disadvantaged areas. In the Department of Computer Science, Dr Andrew Barnes is pioneering methods to predict storm paths and extreme weather events with greater accuracy and efficiency, while making these insights accessible to countries with limited resources.

Storms form from the movement of air in the atmosphere. Using AI, Andy can understand how the air moves, calculate where the storms originate, and track their path. He stores this information in databases, with other distinct storm features, such as whether they occur in the summer or winter, their windspeed, and the amount of rainfall. He then uses an AI technique called clustering to train his models to classify different types of storms and enable precise forecasts.

Andy's research could have a huge impact on preparedness for storms and other weather events. By harnessing unique AI technologies that require as little data as possible, he can empower communities with fewer resources to adapt and mitigate the impacts of extreme weather. As climate change accelerates, his research demonstrates the transformative potential of AI in safeguarding our planet against extreme weather events.

Mitigating extreme weather events

There are several ways Andy’s models can be harnessed for the benefit of others. For example, they can draw generalisations about the weather, such as: ‘this summer will be windier than last summer’; or ‘this next storm will be wetter than the previous one’. These observations can be used to inform the general public about current and future weather events and help people prepare.

Andy aims to harness AI-driven insights to inform strategic responses to climate change. His primary focus was in the UK, however, he has also carried out research in Spain and South Africa, looking at the specific types of storms and flooding they trigger. While these countries collect lots of data about extreme weather, Andy wants to emphasise efficiency by using as little data as possible. By doing so, this research can be applied to less economically developed countries with less data available.

Developing countries with fewer resources to predict and mitigate extreme weather events are most likely to suffer from the impacts of climate change. Risks include storms destroying architecture, triggering landslides, disrupting water sources, and overall increasing the risk to life. Andy hopes to create reliable models that generate a better understanding of extreme weather and help mitigate risks in these regions.

Techniques behind predicting storms

Andy spends lots of time pre-processing the data he receives. He finds, collects, and combines datasets from multiple sources, such as the MetOffice, the European Centre for Medium-Range Weather Forecasts, and NOAA. Most current research focuses on a few features (such as rainfall or flooding), and then training massive datasets.

A common rule of thumb in machine learning is 'the more data the better', so these models often work quite well, but are infeasible for regions with fewer resources. Instead, Andy is focussing on making models trained on less data, for example by optimising feature selection.

Another part of preprocessing is the creation of storm tracks. These use mathematical models to abstract data from very large datasets. Andy uses a variety of techniques to preprocess the data. He uses distance metrics - data to tell us, for example, how far a storm has travelled. These are typically measured using 'Euclidean distance', or the distance between two points, although Andy is looking into a combination of different metrics to use.

Storms often don’t travel in a straight line between two points, so calculating their paths is often an unintuitive process requiring careful measurement. Using different distance metrics means paths can be measured more accurately and therefore need less data.

He also classifies storms using 'clustering'. This is a subsection of unsupervised machine learning, where models look for patterns in data without knowing what it is. The model then groups or ‘clusters’ the data, sorting it into groups with similar features. He has tried a variety of clustering algorithms and continues to experiment to find the optimal ones.

Future forecasts

Andy's work in harnessing AI for weather prediction marks a significant step in helping to safeguard communities against the escalating impacts of climate change.

By preprocessing data and deploying cutting-edge clustering algorithms, Andy not only enhances the accuracy of storm forecasts but also ensures the accessibility of this vital information to regions with limited resources.

As extreme weather events become more frequent and severe, Barnes's research could empower vulnerable populations to adapt to and mitigate risks effectively.

This article was written by Caroline Morton, a 2nd-year undergraduate, pursuing an integrated master's in Computer Science with Artificial Intelligence in the Department of Computer Science. It was produced as part of the Science Communication Ambassador project in the Faculty of Science.