Accurate solar radiation site adaptation: Harnessing satellite data and <i>in situ</i> measurements

Accurate solar radiation data are essential to optimize solar energy systems and assess their feasibility. In this study, we propose a site-adaptation procedure based on a machine learning model trained to enhance the accuracy of solar radiation data using a combination of the National Solar Radiation Database (NSRDB) and in situ data collected in southern Colombia. The NSRDB provides high temporal and spatial resolution data, while in situ data offer accurate localized measurements specific to the study area. Our machine learning models were trained to learn the relationships between NSRDB data and in situ meteorological station data. The results demonstrate promising predictive capabilities, with the extreme grading boosting model effectively reducing mean absolute error, while a neural network model trained with the triplet loss function proved effective in minimizing mean bias error (MBE) and improving correlation between model-adjusted and in situ collected data. These findings make significant contributions to the field of solar radiation prediction, highlighting the effectiveness of amalgamating NSRDB and in situ data for precise solar radiation estimation, and promote the advancement of solar energy system design and decision-making processes.