Rainfall forecast in Brazil using machine learning

Rainfall forecasting through machine learning can play a crucial role in several areas, such as agriculture, energy, infrastructure, and public safety. The machine learning models have the ability to anticipate climate patterns and extreme events, allowing plantation planning, water resource management, and forecasting energy demands, as well as adopting preventive measures against natural disasters. In this work, we explore three machine learning models (random forest, long short-term memory, and bidirectional long short-term memory) to predict the amount of precipitation in five Brazilian regions (South, Southeast, Central-West, Northeast, and North). We use three-variable reanalysis climate data: local temperature, Atlantic Ocean temperature, and total precipitation. The models are trained by means of the local and Atlantic Ocean temperatures as input features and the total precipitation as a label. Our results indicate that all models perform satisfactorily in their predictions. We verify that the random forest exhibits average absolute errors less than the errors related to the recurrent neural network models. Our results show the effectiveness of machine learning models in predicting rainfall patterns.