LIFD Early Career Researcher spotlight: Mendrika Rakotomanga
Thesis Title
Object-Based Deep Learning for Probabilistic Convective Storm Nowcasting from Satellite Data
School / Faculty
School of Mathematics
Supervisors
- Professor Douglas J. Parker
- Dr Nadhir Ben Rached
- Dr Cornelia Klein (UKCEH)
- Dr Seonaid R. Anderson (UKCEH)
Tell us a bit about yourself
Mendrika Rakotomanga is a PhD researcher in machine learning in the Department of Applied Mathematics at the University of Leeds. He develops and deploys machine learning systems based on satellite observations to improve convective storm prediction across Africa. He is a Mastercard Foundation Scholar alumnus from Stellenbosch University, where he obtained an MSc in Mathematical Sciences. He also holds a Master’s degree in Physics and Chemistry from École Normale Supérieure d’Antananarivo. Before starting his PhD at Leeds, he worked as a Data Science Research Fellow at the Max Planck Institute for Intelligent Systems.
What is your research about?
My research focuses on developing deep learning methods for convective storm nowcasting using satellite observations. I use object-based approaches that identify and track convective cores in satellite imagery and learn how storms evolve over time. The work also explores interpretability and scalability in machine learning models. The goal is to produce probabilistic predictions of storm occurrence several hours in advance, helping improve short-term forecasting and early warning systems, particularly in regions where radar observations are limited.
What did you wish you knew before starting a PhD?
Before starting a PhD, I wish I had known how much time is spent exploring ideas that do not work. Learning to navigate uncertainty and iterating on research questions are important parts of the process.
What are your plans for the future?
I would like to continue developing and deploying AI-based systems for weather prediction and Earth observation using remote sensing and physics-informed machine learning.