Determination of seasonal forecast skill in identifying extreme events of temperature, wind speed, and SPI

Renewable energy is the fastest-growing source of electricity globally, but climate variability and impacting events affecting the potential productivity of plants are obstacles to its integration and planning. Knowing a few months in advance the productivity of plants and the impact of extreme events on productivity and infrastructure can help operators and policymakers make the energy sector more resilient to climate variability, promoting the deployment of renewable energy while maintaining energy security.

The energy sector already uses weather forecasts up to 15 days for plant management; beyond this time horizon, climatologies are routinely used. This approach has inherent weaknesses, including the inability to predict extreme events, the prediction of which is extremely useful to decision-makers. Information on seasonal climate variability obtained through climate forecasts can be of considerable benefit in decision-making processes. The Climate Data Store of the Copernicus Climate Change Service (C3S) provides seasonal forecasts and a common period of retrospective simulations (hindcasts) with equal spatial temporal resolution for simulations from 5 European forecast centres (European Centre for Medium-Range Weather Forecasts (ECMWF), Deutscher Wetterdienst (DWD), Meteo France (MF), UK Met Office (UKMO) and Euro-Mediterranean Centre on Climate Change (CMCC)), one US forecasting centre (NCEP) plus the Japan Meteorological Agency (JMA) model.

In this work, we analyse the skill and the accuracy of a subset of the operational seasonal forecasts provided by Copernicus C3S, focusing on three relevant essential climate variables for the energy sector: temperature (t2m), wind speed (sfcWind, relevant to the wind energy production), and precipitation. The latter has been analysed by taking the Standard Precipitation Index (SPI) into account.

First, the methodologies for bias correction have been defined. Subsequently, the reliability of the forecasts has been assessed using appropriate reliability indicators based on comparison with ERA5 reanalysis dataset. The hindcasts cover the period 1993-2017. For each of the variables considered, we evaluated the seasonal averages based on monthly means for two seasons: winter (DJF) and summer (JJA). Data have been bias corrected following two methodologies, one based on the application of a variance inflation technique to ensure the correction of the bias and the correspondence of variance between forecast and observation; the other based on the correction of the bias, the overall forecast variance and the ensemble spread as described in Doblas-Reyes et al. (2005).

Predictive ability has been assessed by calculating binary (Brier Skill Score, BSS hereafter, and Ranked Probability Skill Score, RPSS hereafter) and continuous (Continuous Ranked Probability Skill Score, CRPSS hereafter) scores. Forecast performance has been assessed using ERA 5 reanalysis as pseudo-observations. 

In this work we discuss the results obtained with different bias correction techniques highlighting the outcomes obtained analyzing the BSS for the first and the last terciles and the first and the last percentiles (10th and 90th). This analysis has the goal to identify the regions in which the seasonal forecast can be used to identify potential extreme events.