Mendrika
Rakotomanga

Machine Learning · Satellite Data · Nowcasting

I am a PhD researcher in Machine Learning and Applied Mathematics at the University of Leeds. I develop and deploy machine learning systems based on satellite observations to improve convective storm prediction.

Mendrika Rakotomanga

Research

Object-based deep learning for storm nowcasting

I develop models that convert satellite-derived convective-core objects into probabilistic forecasts of storm occurrence. This allows the model to learn from storm structure, location, intensity, timing, and interaction between neighbouring convective systems.

Convective storm nowcasting Satellite remote sensing Deep learning Interpretability Early warning systems
NCAST storm nowcasting animation
PANCAST mobile animation

Featured Project

PANCAST

PANCAST is an operational AI-based probabilistic nowcasting system for convective storm occurrence across Africa. It combines satellite observations, object-based storm representations, and spatio-temporal deep learning to support real-time forecasting.

The system is designed for operational forecasting and early warning applications across Africa.

Visit pancast.io

Publications

Selected academic work

Object-Based Deep Learning for Probabilistic Convective Core Nowcasting from Satellite Data

M. Rakotomanga, D. J. Parker, N. Ben Rached, S. R. Anderson, C. Klein, and S. C. Wells.

Under review in Quarterly Journal of the Royal Meteorological Society.

Experience

Research and professional experience

2023–present

PhD Researcher in Machine Learning and Applied Mathematics, University of Leeds.

2025

NC-International Visiting Scientist, UK Centre for Ecology and Hydrology.

2023

Data Science and AI Fellow, Max Planck Institute for Intelligent Systems and University of Tübingen.

Education

Academic background

2021–2022

MSc in Mathematical Sciences, AIMS South Africa and Stellenbosch University, Cum Laude.

2017–2020

Master’s in Physics and Chemistry, Ecole Normale Supérieure and University of Antananarivo, Magna Cum Laude.

Conferences

Selected talks and conferences

ENWFC 2026

Oral presentation on deep learning methods for convective storm nowcasting.

RMetS Annual Conference 2026

Oral presentation on object-based deep learning for convective storm nowcasting.

WMO AI Webinar

Shared work related to AI applications for weather forecasting and nowcasting.

IndabaX Madagascar

Keynote and outreach activities on AI-driven nowcasting for African weather challenges.

Technical Skills

Python PyTorch TensorFlow scikit-learn Distributed ML SLURM HTCondor ReactJS NodeJS PostgreSQL MongoDB LaTeX

Languages

Fluent in English, French, and Malagasy.

Limited proficiency in German.

Recognition

Awards and selected activities

EPS Leeds Partnership Awards

Highly Commended, Community Impact and Innovation, 2026.

AIMS–Tübingen Fellowship

Fellowship Programme in Data Science and Artificial Intelligence, 2023.

Mastercard Foundation Scholar

Full scholarship for MSc studies at AIMS South Africa and Stellenbosch University.

Contact