The ability to provide accurate monthly rainfall forecasts is key to operational planning in many sectors such as the agricultural industry who are reliant on sub-seasonal rainfall for crop yields and harvest periods. Traditional rainfall forecasting approaches utilize numerous numerical methods and empirical models to produce a gridded estimate of rainfall over a large area, the grid cells of which often span multiple regions and struggle to capture extreme events.

In recent years, advancements in new machine learning methods capable of mining information from images has opened a host of potential applications for rainfall forecasting. This talk presents a novel method of forecasting regional monthly rainfall by training a neural network capable of interpreting forecasted meteorological patterns. Forecasted mean sea-level pressure and 2m air temperature patterns are used to train a convolutional neural network against a benchmark rainfall dataset (CEH-GEAR). The network is then used to make predictions using an unseen set of patterns the results of which are evaluated against predictions made by the ECMWF SEAS5 service.

About Andy Barnes

Andy is currently a final year PhD student in the Department of Architecture and Civil Engineering at the University of Bath and is part of the Water, informatics, and science engineering centre for doctoral training (WISE CDT). Andy's research focus is on novel applications of deep machine learning methods for the analysis and prediction of extreme rainfall events in Great Britain. Prior to joining the WISE CDT Andrew held several software engineering positions after achieving a first-class honours degree in Computer Science from the University of Plymouth.

Since joining the University of Bath Andy has been involved with and has led several public engagement projects including the augmented reality sandbox as showcased at the Swindon Science Festival in 2020. Further to this, he has given a multitude of public facing talks on topics ranging from atmospheric rivers to the water cycle and engaged with members of the public of all ages regarding various aspects of his research.