Fog Forecast Model based on Machine Learning

Abstract

This study introduces an innovative approach for fog forecasting based on Machine Learning (ML) algorithms. It involved utilizing eighteen years of surface and sounding meteorological data from Guarulhos International Airport and nearby Marte Airfield for training and testing ML models. Multiple categorical algorithms were trained and evaluated, with the top three models selected for further investigation. The results of the study highlight that the best-performing model, which is based on the Random Forest algorithm, can provide reasonably accurate predictions for the occurrence of fog. Specifically, it forecasts fog occurrence within a time window from 03 to 11 UTC with a reasonable degree of accuracy (Proportion Correct = 0.90 ± 0.03, Probability of detection = 0.96 ± 0.03, False Alarm Rate = 0.33 ± 0.01, Critical Success Index = 0.65 ± 0.02, and Bias = 1.43 ± 0.05). Additionally, the method indicates the most likely time for the onset and dissipation of fog events based on historical data. This research offers valuable insights into improving fog forecasting at the Guarulhos International Airport and demonstrates the potential of ML algorithms in enhancing predictive accuracy for weather-related events.