Using Xgboost models dor daily rainfall prediction

Machine learning models for predicting daily precipitation have gained traction in recent years. Understanding the benefits of using this technology in different regions is a relevant research topic. For this reason, this study aims to evaluate daily precipitation estimated forecasts from climate data between 1983 and 2019 in Itirapina, São Paulo, Brazil. We used a novel machine learning algorithm, XGBoost (eXtreme Gradient Boosting), to create several daily precipitation prediction models. Two tasks were modeled: the occurrence of daily precipitation (classification) and the amount of daily precipitation (regression). The results revealed that the occurrence of daily precipitation could be predicted with an accuracy of around 90%. Additionally, models were developed to predict the amount of daily precipitation with error rates of around 3mm. We observed that precipitation in the study area is directly associated with solar radiation, and estimated forecasts of precipitation and the corresponding months are characteristic of the tropical climate.