A database of skill scores on solar forecasting

We screen 1,447 papers from Google Scholar, among which the full texts of 320 papers from 2006 to 2022 are thoroughly reviewed for data extraction. A database of 4,687 observations of most of the key factors that can influence the forecast accuracy was built. Analyzing the database using statistical methods (e.g., linear regression), the marginal impacts on skill score of many important factors can be quantified, including climate zone, forecast horizon, inputs, forecast models, resolution of forecasts, train and test data length, type of forecasts (i.e., solar resource or PV output forecasting), and publication date. As the database covers most of the literature on the field that provide skill score, the findings from the database can be applied globally.

In the database file, there is a sheet named "Database", which we used for our analysis. Our findings have been published in a working paper (https://arxiv.org/ftp/arxiv/papers/2208/2208.10536.pdf) and are prepared for a journal paper publication. 

Furthermore, there is a sheet named "LargerDatabase", which include additional variables that we have extracted from the literature but did not include in our analysis to avoid multicollinearity problem. For example, information of irradiance level of regions (GHI, DIF, DNI), the installed capacity of the power plant, country and region of the forecasts, and the use of different techniques (data processing techniques, data engineering and feature selection techniques...) can be found in the LargerDatabase sheet. These additional variables can be of interest to many scholars.