An Early Warning System for Storms

An Early Warning System for Storms: Using AI to Power 'Nowcasting' Across Africa

Across the UK, networks of land-based weather stations, radars and satellites make up Britain’s vast infrastructure for weather monitoring. These feed live data into Met Office models to deliver precise hour-by-hour weather forecasts several days in advance, helping people make day-to-day decisions and bringing benefits across a range of sectors. Yet, in parts of the developing world, like Africa, patchy radar networks paint a starkly different picture for weather forecasting. 

Africa is a continent of big extremes in weather, with storms that can pose a severe threat-to-life to communities and significantly drain local economies. Data from satellites and the continent’s small number of radar stations can help to predict overall weather patterns a few days in advance, but they fall short of forecasting the types of storms that develop rapidly over just a few hours or minutes. With infrastructure that is more vulnerable to extreme weather, there is a pressing need for national weather agencies to provide accurate, up-to-date storm warnings to help safeguard their communities.   

Through the University of Leeds-led FASTA programme, a new way to provide short-term predictions of extreme weather in Africa is emerging. Called nowcasting, FASTA uses satellite data and fast data processing to estimate the location and movement of African storms up to six hours in advance. This information is made freely available online and via a smartphone app, with near-live updates from the Leeds-based data centre every 30 minutes. 

Hosted in the UK’s National Centre for Atmospheric Science (NCAS) at the University, FASTA nowcasting is being developed using a strength of multidisciplinary expertise that is unique to Leeds - African meteorology, fluid dynamics, machine learning, algorithm development, data science and social science. Close partnerships with several national meteorological agencies in Africa mean local partners are building the capacity to adopt and adapt FASTA nowcasting for their own needs – putting this lifesaving information directly into the hands of people. 

FASTA’s nowcasting service was first piloted with the Kenya Meteorological Department and the University of Nairobi in 2022. Along with socioeconomic research to understand how different communities in Africa use smartphones to find information, feedback through workshops and social media helped to improve how storm information was presented so that it better suited local needs. The platform now has around 4000 users across the country. 

One of the big challenges facing the FASTA research team has been to apply rapidly developing advancements in Artificial Intelligence (AI) to their nowcasting model in a way that is both robust and trusted. Their ongoing research is focused on improving storm nowcasts, bringing together the fields of machine learning and fluid dynamics to make predictions more location-specific and to enhance their accuracy. With the private sector increasingly adopting AI to deliver advanced weather solutions, FASTA’s approach is to use rigorous academic methods to understand how AI can best be incorporated into weather forecasting, ensuring their own system is reliable and properly evaluated.  

Through FASTA and the work of its UK and African partners, nowcasting is providing short-term storm warnings for many communities across Africa. In Ghana, farmers use nowcasting to determine if the weather conditions are suitable to apply expensive fertilisers or insecticides to crops, so their investment is less likely to be washed away by heavy rain. Communities that use small canoes and wooden boats to fish close to the shore typically can suffer high mortality rates due to the impact of unpredictable storms. When a storm is approaching, nowcasting data is being used to raise flags on nearby beaches that highlight the increased risk to life. In urban areas, where even modest rainstorms can cause severe flooding that is a danger to life for many, nowcasting helps people respond rapidly to oncoming storms – putting out sandbags in good time and closing schools early so children can get home safely.  

AI-based nowcasting runs at a fraction of the cost of conventional physics-based forecasting models, using less computational power and requiring fewer people – some estimates put the computational cost as being around 600,000 times cheaper with AI. The research team is working to design a comprehensive ‘Leeds nowcasting algorithm’ that can be developed and shared with African partners to set up cost-effective, elite storm warning systems. This will support existing and new weather centres across the continent to generate storm forecasts that are bespoke for their own communities. 

Professor Douglas Parker, Professor of Meteorology at the University of Leeds, says: “In just two years, AI-based nowcasts and forecasts have become equally as good as any system we had before at predicting storms. Replacing the need for the vast computational effort that’s usually required to forecast the weather means that forecasts and nowcasts can be created in centres right across Africa, for their own regions and communities. With our multidisciplinary team of just 15 researchers, data scientists and students, we’ve co-built a system that millions of people across Africa can benefit from in the years to come.  

“For us, success will be that we’re no longer needed, and we’re already seeing nowcasting being adopted by our partners in Africa, changing it and modifying it so it works for their needs. It’s all about democratizing weather forecasting – and that’s what we’re enabling here at Leeds.”   

The FASTA app can be downloaded here: https://tinyurl.com/FASTAweather