Probabilistic approaches are essential for modelling future renewable electricity systems due to the inherent uncertainty in weather forecasting. Conventional methods employ historical weather data, but this restricts the range of scenarios to the last four decades, limiting the exploration of rare or extreme conditions.
In this research, I pioneer an innovative strategy leveraging expired ensemble weather forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF) to dramatically extend the dataset for scenarios, potentially presenting up to 10,000 years of synthetic weather data. These ensemble forecasts, initialized from slightly perturbed snapshots of atmospheric conditions, diverge over time due to weather's chaotic nature. Consequently, their late-stage predictions provide independent yet physically plausible projections of what the weather could have been, preserving correct spatial correlations.
This presentation will further discuss the methods employed for the parallel processing of this expansive dataset, made possible through high-performance computing. By parallelizing the computations across the dataset, we have been able to manage the vast amount of data effectively, enabling more efficient processing and analysis.
[1] European Centre for Medium-Range Weather Forecasts (ECMWF) Atmospheric Model Ensemble extended forecast https://www.ecmwf.int/en/forecasts/datasets/set-vi
Petr is a PhD student from the first cohort of the AI for Environmental Risk CDT, looking at innovative ways to model, understand and bring to existence a future carbon-neutral electricity grid.
