Methods for Site-Adaptation of Satellite-Based DNI Time Series: Application to Brazilian Northeast

The standard procedure today in solar industry is the acceptance of solar irradiance estimates via satellite images (normally time series longer than 15 years) and its adaptation to the site where the solar power plant will be. The adaptation was done by correlating the long time series via satellite with a short time series (1 year) of good quality devices (first class pyrheliometer and standard secondary pyranometer) for measurements on terrestrial surface. The purpose of this work was the development as well as the demonstration of four computational techniques (simple regression, multiple regression, ANN and SVR) which allowed the correction/mitigation of this bias. The estimators developed were applied for Patos, city in the Northeast of Brazil, and allowed a significant decrease of the bias through the root mean square deviation (RMSD) of the originally estimated values obtained from a satellite database. Bias correction for the employed metrics provided low bias values or very close to zero according to some metrics. Regarding the RMSD, the simple, multiple and ANN (MLP) linear regression algorithms produced slight improvements over the raw data from satellite, respectively of 2.5, 8.0 and 4.3%. When using the SVR, the resulting improvement was remarkable for the RMSD. In this case, the improvement was about 57.9%. The results obtained showed the capacity of SVR for a reliable adjustment with precision for the estimation of DNI via satellite, provided with high quality terrestrial data.