Research

Dust, Aerosol & Air Pollution Forecasting

Developing AI methods to understand and forecast dust, aerosols, and air pollution, from local to global scales. We investigate the transport and evolution of pollutants from natural and anthropogenic sources, with a focus on extreme events and their atmospheric drivers, and explore their impacts on air quality, human exposure, and health.

Weather and Precipitation Forecasting

Developing AI methods for high-resolution weather and precipitation forecasting by integrating weather radar, rain gauges, satellite observations, and meteorological measurements to correct observational biases and better predict the evolution of weather and rainfall across space and time.

Flash Flood Forecasting

Developing data-driven AI methods for flash flood forecasting, using graph neural networks to learn the complex relationships between rainfall, catchment response, and streamflow. Our goal is to improve prediction of rapid and extreme hydrological events. 

LLMs for Weather Reasoning

We explore large language models (LLMs) for understanding and generating weather systems, translating textual forecasts into spatial weather fields. Our broader goal is to develop LLM-based systems that can reason about the structure and evolution of weather through language.

Explainable AI for Earth Systems

Developing explainable AI methods to understand how transformer-based models learn atmospheric processes, linking AI predictions to physical drivers and underlying atmospheric processes. 

Seasonal Forecasting

Developing AI methods for seasonal weather forecasting using foundation models, adapting large-scale pretrained weather models to regional prediction and incorporating physics-informed learning. Our research aims to improve long-range precipitation forecasts, including regional and extreme rainfall patterns beyond the capabilities of conventional forecasting approaches.