Bespoke AI forecasting systems to support farming communities in West Africa
The Cumulus Project is an ambitious research initiative that aims to transform weather forecasting for agriculture in West Africa through the development of bespoke AI-driven subseasonal forecasting systems. Our goal is to demonstrate the benefits that AI subseasonal forecasting systems can bring to African farmers; to empower national meteorological agencies in Senegal and Ghana to run these systems independently; and lay the groundwork for scaling across Africa in a subsequent phase.
The project guiding principles include a commitment to co-creation, co-development and real-world impact. We are collaborating closely with our local partners, University Cheikh Anta Diop of Dakar (UCAD) in Senegal and Kwame Nkrumah University of Science and Technology (KNUST) in Ghana, where two postdoctoral research fellows will be based. These fellows will also spend time in the national meteorological services (NMS) in these countries (ANACIM and GMet), ensuring that the systems Cumulus develops, are not only scientifically robust, but also operationally relevant and locally owned.
The project will also prioritise developing real systems, rather than academic prototypes. An early milestone is to deploy basic AI forecasts up and running at UCAD and KNUST. These initial systems will inform model development, enhance the Cumulus team’s understanding of operational realities, develop AI skills among partners, and maintain a focus on agricultural user needs.
Technical Summary (pdf), File DownloadProject Cumulus is funded by the Gates Foundation and the Foreign, Commonwealth and Development Office.
The University of Leeds will work alongside the Alan Turing Institute and a consortium of partners to co-design a more accurate forecasting system which could help farmers improve crop yields and reduce economic losses.
By working with top universities and weather services in West Africa, this international partnership will drive innovation in AI weather prediction.
Weather forecasting in West Africa presents a unique set of challenges. As a region increasingly vulnerable to the effects of climate change, unpredictable weather patterns have a direct impact on food security and economic stability.
Farmers often lack access to adequate weather forecasts, forcing them to make critical decisions on planting or harvesting crops without the information they need.
Douglas Parker, Professor of Meteorology in the School of Earth and Environment at the University of Leeds, said: “African rainfall events can be among the most intense in the world, often developing rapidly over only a few hours.
“Using AI-based methods allow us to make more accurate predictions of the probability of rainfall, learning from observations in each country and bringing in information from weather and ocean patterns around the globe.
“By working with top universities and weather services in West Africa, this international partnership will drive innovation in AI weather prediction.”
The Earth’s rotation and solar heating drive the atmosphere, which means that weather in Africa behaves differently from weather in the mid-latitudes in Europe and the USA, where most forecasting methods were designed, highlighting the need for new models optimised for African conditions.
Traditional physics-based forecasting approaches used in the Global North are less effective in sub-Saharan Africa, so the Cumulus initiative will draw on the project team’s contributions to emerging AI technologies like Aardvark Weather and the Aurora Earth System Foundation Model.
These models will help to develop new AI-based forecasting methods tailored to African conditions which will deliver more accurate and locally relevant insights.
The Aardvark technology is fully driven by AI and will combine satellite imagery, ground observations and existing forecast data to create a clearer picture of the atmosphere.
The technology draws on both remote-sensing and local measurements, learning from data-rich regions to improve predictions where data is more scarce such as in sub-Saharan Africa.
The Aurora Earth System Foundation model shows how a single AI model could be adapted for a wide range of forecasting tasks.
The agility of such models will allow the Cumulus initiative to create systems tuned to local weather patterns as well as extending forecasts to sub-seasonal timescales (2-6 weeks) most useful for farmers and the fishing industry.
Crucially, these systems are affordable and adaptable, enabling West African partners to produce their own forecasts, build expertise and drive local innovation.
Dr Scott Hosking, Mission Director for Environmental Forecasting at the Alan Turing Institute, said: “Forecasting rainfall in the tropics is a unique challenge, further complicated by climate change and a historical lack of localised data. To protect lives and livelihoods in these regions, we cannot rely on off-the-shelf AI solutions.
“This partnership between the UK, Ghana, and Senegal focuses on developing AI weather models specifically designed for conditions in West Africa, to provide the precision needed to effectively manage food systems and strengthen community resilience."
The project is made possible due to funding from the Gates Foundation and UK International Development from the UK government, with the UK’s Met Office and the European Centre for Medium Range Weather Forecasting contributing as advisors.
Picture credit: Professor Douglas Parker.