General Profile: 

Over the years I have moved back and forth between academia and industry. During this time my interests have drifted from pure mathematics to applications of computing on real-world problems. My career in industry has been as a software developer. I have worked for various start ups, from a system to assist with answering questions over SMS (remember the time when you didn't have Google in your pocket?) to a fashion recommender. I have moved to academia again and research as a way to apply the skills I have learned to work on new and important problems. I am interested in how AI can be used to help people combat climate change.

Research Project Summary:

Climate change is causing the intensification of rainfall extremes. Precipitation projections with high spatial resolution are important for society to adapt to and mitigate these changes. Physics-based simulations for creating such projections are computationally expensive. The natural variability of precipitation in both time and space means multiple projections are required by users of projections in order to cope with this uncertainty. Generative machine learning models offer approaches for generating further projections from lower resolution projections more cheaply, effectively emulating a high-resolution, physics-based model.

Climate datasets can be hard to work with as well as being expensive to create. Data consumers may find that the data they desire are not available or not understand their limitations. Potential users will provide user requirements and evaluation of this emulator through interviews, surveys and observations. An initial round of interviews has established the need for more high-resolution rainfall data that is trustworthy and the difficulties in defining extreme events to suit all use cases. From these user requirements a method is being developed for designing and evaluating a variational autoencoder trained to be a high-resolution rainfall model emulator.