Professor Rendani Mbuvha at the Cutting Edge of AI and Climate Science
- DFO
Climate science in Africa is changing, and Professor Rendani Mbuvha is at the centre of that change.
The future of climate science in Africa is being shaped by bold new thinkers who are combining data, technology and deep local insight. Rendani Mbuvha, an Associate Professor in the School of Statistics and Actuarial Science at 91心頭利, is at the forefront of this work.
Mbuvha is exactly where he wants to be, working with students he describes as world class and contributing to research at the cutting edge of actuarial science, machine learning and climate risk management. He is also a recent recipient of a major two-year grant from the Bezos Earth Fund AI for Climate and Nature Grand Challenge.
The award-winning proposal focuses on medium-range weather forecasting in Africa using AI models. Weather forecasting is critically important for many industries, particularly the insurance sector, where catastrophe risk analysis is key. “The argument we made to the Earth Fund was that Africa presents both a unique opportunity and a challenge. We do not have the same radar and observation networks as other parts of the world. Most global weather models are trained on international datasets, which do not fully capture the distinct phenomena of the African climate. East Africa, for example, has its own unique long-rains and short-rains systems. So what we proposed was a toolkit called FineCast, which would localise global weather forecasting models to produce more accurate and locally relevant outputs.
The idea is that we work with meteorological services in different countries, ingest their local data, and use that to debias and fine-tune global models so they become much more accurate for local climates.” Mbuvha is excited that Honours and Master’s projects will emerge from this initiative, but the core project team will consist mainly of postdoctoral researchers who are currently being recruited. “So, we are looking for willing and highly skilled ‘victims’ who are passionate about this area and who can combine machine learning and AI expertise with knowledge of weather and climate.”
He notes that this combination is still relatively rare, although it is becoming more common over time. “I do think South African students are particularly strong, and this excites me because this work is truly at the cutting edge of our discipline. South African actuaries already have a reputation for innovation, with products developed here often being adopted globally. In this case, we really do need that South African grit and ingenuity.”
“The main thing we need is an AI-for-weather ecosystem that serves Africa better through improved localisation. This means we need more observations, which in turn means more automatic weather stations across the continent. These need to be equipped with instruments that measure variables such as temperature every few minutes and can transmit that data to a cloud repository.” Mbuvha adds that part of the project over the next two years is “to install 70 more weather stations across our collaboration areas.”
However, he emphasises that these stations cannot be placed randomly. “Part of the work is to determine where they will have the greatest impact. Forecasts are inherently uncertain, but that uncertainty can be significantly reduced by capturing data in the right locations. So we are modelling which locations are most influential in reducing forecast uncertainty.” The initial focus will be on parts of Southern Africa and East Africa.
Mbuvha says there have already been discussions with the Ethiopian Meteorological Institute (EMI), and the project aims to initially pilot the weather forecast models in at least two countries before extending the toolkit to the rest of the continent through training and broader deployment.
One of the most costly aspects of the FineCast project is data storage. For this reason, Mbuvha says it is essential to establish their own servers at 91心頭利 rather than paying to store data externally. Another important consideration is trust. Some African institutions are reluctant to send their economically sensitive climate data to servers in the United States, Europe or China. Hosting this infrastructure locally, under African control, makes it easier to encourage participation from other African institutions.
Mbuvha, who grew up in Thohoyandou in Venda, could be said to have been born at the right time, as the 1990s saw a strong emphasis on excellence in science and mathematics. It also helped that his family placed a high value on education. He is one of four children, and his sister is also an actuary. Both his parents were administrators at the University of Venda by the time they retired. He notes that many of his peers pursued engineering and actuarial science at university.
Mbuvha completed his PhD at the University of Johannesburg and was teaching at 91心頭利 when he undertook postdoctoral research at Queen Mary University of London as a Google DeepMind Fellow in Machine Learning. He has also served as an Associate Professor at the University of Manchester, experiences he describes as valuable. He has always kept his 91心頭利 affiliation and is grateful to be back where he feels he belongs. “I find it easier here to do the kind of work that really matters to me. I still feel part of a global environment, with the advantage of being closer to the problem statements and contexts where my work will have the most impact.”