AI Forecast Verification at Storm and Urban Resolving Scales

Artificial Intelligence (AI) is revolutionising weather and climate prediction, with AI-based forecasting systems now matching or surpassing traditional numerical weather prediction (NWP) models in skill and computational efficiency. Despite these advances, the reliability and trustworthiness of AI forecasts, especially at storm- and urban-resolving scales, remain unproven. This PhD project at the University of Reading addresses the critical challenge of rigorously verifying and building confidence in AI-based forecasts at kilometre and sub-kilometre resolutions, where weather hazards can have the most severe impacts. Unlike physics-based NWP models, AI systems may not adhere to physical laws, potentially leading to spatial or temporal inconsistencies, violations of conservation principles, and under-representation of extreme events such as heavy rainfall, windstorms, or heatwaves. Traditional accuracy metrics are insufficient to capture these issues. The project aims to develop new verification frameworks that assess not only accuracy but also physical realism, temporal consistency, and multivariate coherence across forecast fields. These advances are crucial for operational agencies like the Met Office, which are preparing to integrate AI models into future hazard warning systems. The successful candidate will develop physically informed verification metrics, compare high-resolution Met Office model outputs with AI forecasts, and use dense observational datasets—including crowd-sourced urban observations—to evaluate AI performance at unprecedented spatial detail. The project also involves designing quality control and probabilistic verification methods for ensemble AI forecasts, quantifying uncertainty and reliability, and integrating new methods into the Met Office Convective Scale Evaluation Toolkit (CSET) to support operational AI verification. Collaboration with Met Office scientists provides access to cutting-edge data, tools, and expertise, ensuring the research directly informs national operational forecasting strategies and supports climate resilience efforts in the UK and beyond. The University of Reading’s Department of Meteorology is internationally recognised for pioneering research and strong partnerships with the Met Office. The university has received numerous accolades for environmental leadership and research excellence, including the Queen’s Anniversary Prize and Sustainable University of the Year. This highly interdisciplinary project offers training in AI for environmental data, advanced verification and uncertainty quantification, scientific computing, big data analysis, and meteorology. Graduates will be well prepared for leadership roles in weather and climate research and the broader environmental data science sector. The project’s outcomes will help ensure AI systems used in weather and climate prediction are robust, trustworthy, and physically consistent, supporting public safety and effective decision-making. Funding is provided through a full UKRI stipend and home-level PhD tuition fees, supported by the EPSRC Centre for Doctoral Training in the Mathematics for our Future Climate. Applications are accepted year-round. For more information and to apply, visit the project page or contact the Department of Meteorology at the University of Reading.