Objective & Background
The C3S Energy [C3SE] operational service aims to deliver key information on historical, seasonal forecast and projection periods for climate-related indicators relevant to the European energy sector. This presentation will be focused on a new electricity demand dataset built for all European countries.
Method
C3SE builds on pre-operational demonstrators developed by the two proof-of-concept CLIM4ENERGY and ECEM. The new ERA5 reanalysis is used in replacement of ERA-Interim for the historical period (1979-2018), and to adjust biases in seasonal forecasts and climate projections.
Electricity demand at country level is modeled using Generalized Additive Models (GAM). Individual models are set up for each country, using electricity demand data from ENTSO-E and climate variables from ERA5 (2m temperature, Global Horizontal Irradiance and wind speed at 10m). Additionally, calendar data are necessary to consider periods of low/high energy consumption. Models are built on a few recent years, validated on an independent period, and then used over the whole ERA5 period to reconstruct 1979-2018 demand time series. These allow to mimic what demand would have been considering the effects of climate variability on a long period, if other drivers of demand had been the same as in the training period.
Results
As demand is modeled using country averages of climate variables, the shift from ERA-Interim to ERA5 has only a marginal impact. The main changes come from the approach taken to model the demand time series. Long term trends are not modeled explicitly, and the reconstructed demand data aim at simulating what demand would have been over the past forty years, if the non-climatic drivers of electricity consumption had been the same all over the period. From this, the actual demand can however be reconstructed by adding the daily anomaly of the calculated time series to actual annual mean demand interpolated at daily timescale.
The accuracy of the reconstruction is evaluated against ENTSO-E data, on independent periods. For most countries, the quality of the models is very good, close to operational demand models, with mean absolute percentage errors between 1-3%. For some countries however, the results are not so good. The main reason lies in the quality of the demand data used to build the models. Some countries indeed have either missing data, or inconsistencies in the ENTSO-E time series.
Discussion
The new demand times series show a good agreement with ENTSO-E data. Issues in observed data are the main limitation in the methodology. We identify the main problems and suggest some solutions to overcome these.
Next steps will consist in using these demand models with C3S multi-model seasonal forecasts and climate projections from EURO-CORDEX simulations.
Further extension into the past of ERA5 (back to 1950) will also allow to build longer demand time series, which are in general requested by end-users to better represent the interdecadal variability, and capture strong anomalies in the past (e.g. winters of 1954 and 1963 extreme cold anomalies).