Achieving Skilled and Reliable Daily Probabilistic Forecasts of Wind Power at Subseasonal-to-Seasonal Timescales over France

TL;DR

The paper tackles the challenge of skilled and reliable daily probabilistic wind power forecasts at subseasonal-to-seasonal horizons for France. It introduces a pipeline that converts ECMWF S2S weather ensembles into daily wind-power forecasts up to 46 days ahead without temporal or spatial aggregation, followed by online post-processing to correct biases and under-dispersion. The study shows that S2S-based forecasts outperform a climatological baseline by about 45–50% in CRPS and ensemble-mean MSE, with post-processing boosting calibration and yielding an additional ~10% skill gain; EMOS and Quantile Regression emerge as the leading post-processing approaches. Practically, the results support improved grid stability, O&M scheduling, and market risk management, while the methodology remains adaptable to other weather-dependent energy sectors and longer horizons as S2S products evolve.

Abstract

Accurate and reliable wind power forecasts are crucial for grid stability, balancing supply and demand, and market risk management. Even though short-term weather forecasts have been thoroughly used to provide short-term renewable power predictions, forecasts involving longer prediction horizons still need investigations. Despite the recent progress in subseasonal-to-seasonal weather probabilistic forecasting, their use for wind power prediction usually involves both temporal and spatial aggregation achieve reasonable skill. In this study, we present a forecasting pipeline enabling to transform ECMWF subseasonal-to-seasonal weather forecasts into wind power forecasts for lead times ranging from 1 day to 46 days at daily resolution. This framework also include post-processing of the resulting power ensembles to account for the biases and lack of dispersion of the weather forecasts. We show that our method is able to outperform a climatological baseline by 50 % in terms of both Continuous Ranked Probability Skill Score and Ensemble Mean Squared Error while also providing near perfect calibration of the forecasts for lead times ranging from 15 to 46 days.

Figures (8)

• Figure 1: Forecasting pipeline of the study represented schematically. S2S weather ensembles are converted to S2S Wind Power forecasts before being post-processed to account for the bias and under dispersion. The details of the weather-to-power model and post-processing model are presented in the dashed boxes. The post-processing model follows an online procedure limiting the data available for training.
• Figure 2: $CRPSS$ evolution with lead time. The lead times between $1$ and $7$ days are hatched to remind that short-term weather forecasts would be more suitable to provide power forecasts at such horizons. For each lead time, the $CRPS$ is averaged over all the available samples.
• Figure 3: $MSES_{Ens}$ evolution with lead time. The lead times between $1$ and $7$ days are hatched to remind that short-term weather forecasts would be more suitable to provide power forecasts at such horizons. For each lead time, the $MSE$ is averaged over all the available samples.
• Figure 4: Reliability plots for different lead times. A perfectly calibrated forecast would align with the first bisector represented by the black dashed line. Lead times below $7$ days are not represented as short-term weather forecasts would be preferred to provide the corresponding short-term power forecasts. For each lead time, the observed frequency are computed over all the available samples. The quantile discretization step is 0,05.
• Figure 5: $CRPSS$ evolution with lead time. The lead times between $1$ and $7$ days are hatched to remind that short-term weather forecasts would be more suitable to provide power forecasts at such horizons. For each lead time, the $CRPS$ is averaged over all the available samples.