Weather and Marine Multi-output Prediction Using XGBoost on Automatic Weather Station Data

Climate change on a global scale has triggered an increase in sea levels and heightened the frequency of extreme weather events, especially in maritime countries such as Indonesia. These conditions necessitate the development of accurate and adaptive weather and marine prediction systems. This study proposes a multi-output prediction model using the eXtreme Gradient Boosting (XGBoost) algorithm based on BMKG's Automatic Weather Station (AWS) data from the BMKG. The data cover the period 2022-2025 with high temporal resolution and include five main parameters: wind speed, water level, water temperature, relative humidity, and wind direction. The hyperparameter tuning process led to the discovery of an optimal configuration capable of enhancing the model's accuracy. The evaluation results of the coefficient of determination (R²) and Root Mean Squared Error (RMSE) metrics show that the model can predict water temperature, water level, and relative humidity with very high accuracy, which is more than 85 percent. The model also performed well in predicting wind speed, although it still faced difficulties in handling wind direction due to its cyclical nature. Overall, the XGBoost approach proved effective in modeling weather and marine parameters simultaneously and has the potential to be integrated into environmental monitoring systems in Indonesia's coastal and archipelagic regions.