Speaker
Description
Accurate weather forecasting plays an increasingly important role in today's society, with implications in hydrological modelling, renewable energy production, and civil service operations. To improve the reliability of ensemble weather forecasts, post-processing of said forecasts is frequently employed. However, many variables of interest, such as precipitation or wind speed, exhibit highly skewed distributions, making them challenging to model.
We investigate the application of normalizing flows and flow matching to facilitate skewed distribution modelling, focusing on precipitation forecast post-processing. We show that both methods improve the reliability of the underlying forecast while post-processing the entire lead time at once, jointly.
