Atmospheric Physics Building,Department of Physics, University of Oxford, Parks Road, Oxford, OX1 3PU
Dr Isla Finney, Lake Street Consulting Ltd
Andrea Simpson (andrea.simpson@physics.ox.ac.uk
Abstract
As an operational weather forecaster, an understanding of how numerical weather prediction models are structured and initialised enables better interpretation of their output and estimation of uncertainties in the forecast. Renewable energy (which in the UK means mainly wind and solar generation) is increasingly providing our electricity. As the installed renewable capacity increases, reliable estimates of uncertainty become increasingly important for dispatching both renewable and thermal plant. This in turn leads to lower costs and lower carbon emissions. We will look at how knowledge about the model structure improves use of ensemble forecasts, both from the same model and across models, in order to highlight potential electricity system issues.
AI weather models have made an entry into the operational forecasting domain, and are a rapidly evolving science! Competition within global AI forecast models for the lowest root mean square error (rmse) score for a deterministic forecast dominated through last year, but with increasing awareness 1) of how the limitation of rmse was detrimental to creating models with realistic output beyond the short-range and 2) that the lower computational power required to run the forecast step contrasted with significant expense of creating the ERA5 dataset on which most AI weather models have been based. We will look at the new wave of research looking to address these issues including direct data assimilation to forecast and why AI with NWP, or NWP with AI, could be more than the sum of the parts.
About the speaker: since completing her DPhil at AOPP and OCIAM, Isla enjoyed a Postdoctoral Fellowship at NCAR then moved to working in the energy sector (as an analyst, a meteorologist for energy trading desks, and an energy trader) before setting up her own company in 2014 with the aim to help clients work with the weather. Isla manages to find time for a little research, including collaboration with AOPP as demonstrated by 2 of the 3 papers referenced below.
References:
- J Stanger et al, Optimising the use of ensemble information in numerical weather forecasts of wind power generation, 2019 Environ. Res. Lett. 14 124086
https://dx.doi.org/10.1088/1748-9326/ab5e54
- J Dorrington et al, Beyond skill scores: exploring sub-seasonal forecast value through a case-study of French month-ahead energy prediction, 2020 Q.J.R. Meteorol. Soc. 146 733
http://dx.doi.org/10.1002/qj.3863
- L Hardy & I Finney, Leveraging state-of-the-art AI models to forecast wind power generation using deep learning, Meteorol. Appl. 32 2