Solar Irradiance Prediction in the Amazon Basin Using Machine Learning: A Sustainable Approach for Renewable Energy Expansion

Abstract

The urgent need for renewable energy sources has spurred global innovation in environmental protection and climate change mitigation. Among the viable options, solar energy stands out despite its intermittent nature. Brazil's predominantly green energy matrix is witnessing substantial solar energy expansion. Harnessing solar power in the Amazon basin offers a pathway to enhance living standards for local communities and cities without resorting to new hydroelectric plants or biomass burning, thereby avoiding significant environmental impacts. This study employs data science and machine learning tools to forecast solar irradiance (W/m²) in four cities within the Amazonas state, utilizing NASA satellite data from 2013 to 2022. We implemented decision-tree-based models and vector autoregressive (time-series) models with daily, weekly, and monthly aggregations. The prediction model achieved a mean absolute error of approximately 0.20 using adaptive boosting and light gradient boosting algorithms, aligning with the accuracy of similar studies. This research highlights the potential of satellite data for solar energy assessment in remote regions, offering a robust framework for sustainable energy planning in the Amazon basin.