Machine learning models to improve rainfall forecasts
I look at old weather forecasts and compare them to what really happened. I look for recurring differences and try to correct them. Our current area of focus is East Africa.
Rather than a single weather forecast, we are really interested in the distribution of possible forecasts. From this distribution we can calculate the chance of any particular weather event happening. But to do this, the distribution we use must be accurate.
Current dynamical weather models struggle to represent the true rainfall distribution, but they do output useful information. I apply various empirical methods, or machine-learning, to process weather model output and generate more accurate distributions of possible forecasts.
Research interests
Climate sensitivity
Rainfall predictability
High dimensional chaos
Empirical methods