Benchmarks, datasets & evaluation
An archived dataset from the ECMWF Ensemble Prediction System for probabilistic solar power forecasting
In one sentence. Archived ECMWF ensemble dataset for 2017 to 2020 over Europe and North America supports probabilistic solar forecasting, with calibration and ML-versus-physics case studies.
Abstract
Ensemble numerical weather prediction (NWP) is the backbone of the state-of-the-art solar forecasting for horizons ranging between a few hours and a few days. Dynamical ensemble forecasts are generated by perturbing the initial condition, and thereby obtaining a set of equally likely trajectories of the future weather. Generating dynamical ensemble forecasts demands extensive knowledge of atmospheric science and significant computational resources. Hence, the task is often performed by international and national weather centers and space agencies. Solar forecasters, on the other hand, are primarily interested in post-processing those ensemble forecasts disseminated by weather service providers, as to arrive at forecasts of solar power output. To facilitate the uptake of ensemble NWP forecasts in solar power forecasting research, this paper offers an archived dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF) Ensemble Prediction System, over a four-year period (2017–2020) and over an extensive geographical region (e.g., most of Europe and North America), under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Two case studies are presented to demonstrate the usage of the dataset. One case study elaborates how ensemble forecasts can be summarized and calibrated, which constitute two common forms of probabilistic forecast post-processing. The other demonstrates how the dataset can be used in solar power forecasting applications, which compares machine learning with the physical model chain in terms of their irradiance-to-power conversion capability. The Python code used to produce the results shown in this paper is made available on GitHub.
Labels
- Topic
- Benchmarks, datasets & evaluation, Post-processing & calibration
- Type
- Benchmark or dataset
- Method
- Classical ML (trees, GP, linear, kernels)
- Time scale
- 6 h–3 days
- Domain
- Regional
- Data
- NWP analysis or forecasts (IFS, GFS, ...)
- Status
- Research
- Identifiers
- DOI:10.1016/j.solener.2022.10.062 OpenAlex:W4309030289
- Added
- 2026-10-22
Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.