Objective & Background
Renewable energy penetration is continuing to grow and accurate forecasts of wind and solar energy are therefore essential. All numerical weather prediction (NWP) models have errors, and if some parts of these errors are systematic, then they can be removed using a variety of post-processing techniques. Our aim is to improve the day-ahead forecast skill of renewable energy variables (REV): 10 metre wind speed, shortwave radiation and 2 metre temperature by relating systematic forecast errors to anomalous states of the atmosphere.
Historical forecast data for upper-level variables such as temperature, geopotential height and wind speed on different pressure levels, from 250hPa to 850hPa, were used to inform the forecast correction model. Statistical classification methods were used to identify spatial patterns over the North Atlantic region for the preceding 60 days. These can be used to highlight periods similar to the upcoming forecast. These periods form a training dataset which were used to correct each REV forecast. Historical REV forecast errors were correlated with the training dataset of each upper-level variable. The timestamps of the highest correlations were then used to train the forecast correction model for each day's REV forecast.
Three years of hourly ECMWF IFS forecast data for 24-48 hours ahead were used to drive the forecast models, and results were compared with data from more than 20 weather observation stations on the island of Ireland.
Preliminary investigations have been completed to find the most suitable variables and atmospheric levels to use in the post-processing method. Initial results suggest the spatial mean of upper level variables has a weak correlation with shortwave radiation error. On the other hand, 10 metre wind speed error and 2 metre temperature error exhibit strong correlations with all variables at certain atmospheric levels. The correlation values vary spatially for the different weather observation stations.