Climate Forecast in the Northeast of Goiás: Big Data and Machine Learning Applications

Abstract

This work proposes the development of a predictive model that combines Big Data and Machine Learning techniques for climate analysis in the Northeast region of Goiás, Brazil. The main focus of the study is the forecasting of extreme weather events, such as heavy rainfall and severe droughts, aiming to support strategic decision-making in the agricultural and water resource management sectors. The adopted methodology involves the collection of multivariate data from various sources, including NASA and ESA satellites, INMET meteorological stations, and environmental sensor networks. The processing of this data will be conducted using robust frameworks such as Apache Hadoop and Spark, enabling real-time analysis of large volumes of information. The modeling phase will employ advanced machine learning algorithms, such as Random Forest, a method based on multiple decision trees that enhances accuracy by reducing the risk of overfitting, and XGBoost, a highly efficient gradient boosting technique that combines several weak models to build a strong and accurate predictive model. Neural networks will also be applied to identify complex patterns in critical climate variables such as precipitation, temperature, relative humidity, and greenhouse gas concentrations. The results obtained will have practical applications in regional agricultural planning, irrigation system optimization, and risk reduction related to adverse climate events, significantly contributing to the region’s sustainable development and food security, while generating economic and environmental benefits.