Multivariable modelling based on statistical and machine learning techniques for monthly precipitation forecasting in the eastern Amazon

Accurate precipitation forecasting is crucial for various sectors, such as agriculture, hydrology, and disaster management. In recent years, machine learning (ML) techniques have proven invaluable in improving the accuracy of rainfall prediction and identifying the complex relationships between precipitation and other meteorological variables.This paper presents acomprehensive analysis of the use of multivariable statistical and ML models to predict monthly rainfall at 13 locations in the eastern Amazon. Each model is trained separately for each month, allowing for a tailored representation of precipitation patterns and variations. Additionally, the performance of these models is evaluated via the time series cross-validation technique and an independent test.The results indicate that for the points Serra Sul, Açailândia, and Ponta da Madeira, the multivariable models yielded the best monthly performance in 72.23% of the cases, mainly during the rainy season.Our results highlighted several important aspects of precipitation prediction at different points across the selected study region, particularly concerning the influence of exogenous variables (mainly u10, t2m, TSA, and TNA) on precipitation in most months. Additionally, our findings indicate that the ARIMA, XGBoost, and CNN-1D models outperformed the other models in monthly rainfall forecasting for the Serra Sul, Açailândia, and Ponta da Madeira regions, respectively.