Forecasting Daily Maximum PM2.5 Using FBProphet to Support Unmanned Aerial Vehicle Operations: A Case Study of Aranyaprathet District, Sa Kaeo

Abstract

This study aims to develop a forecasting model for daily maximum PM2.5 concentrations in Aranyaprathet District, Sa Kaeo Province, Thailand, and to utilize the forecasting results to support mission planning of Unmanned Aerial Vehicles for border surveillance operations. The dataset consists of daily maximum PM2.5 values collected from air quality monitoring station 71T (King Rama IX Commemorative Park) over a six-year period from January 1, 2019 to December 31, 2024, totaling 2,192 observations. Data preprocessing included missing value handling using forward filling, anomaly detection via Z-score, and noise reduction using a 7-day rolling median to preserve temporal patterns. The FBProphet model was employed for time series forecasting, with the dataset split into Training and Testing sets at a ratio of 80:20. Hyperparameter tuning was conducted using Manual Grid Search to systematically identify optimal model configurations. Model performance evaluation shows that the Training set achieved an R² of 0.636 and an RMSE of 7.54 µg/m³, while the Testing set achieved an R² of 0.626 and an RMSE of 7.29 µg/m³, indicating reliable predictive capability without significant overfitting. The forecasting results reveal clear seasonal patterns, with elevated PM2.5 levels during the winter season and lower concentrations during the rainy season. The findings demonstrate that FBProphet is an effective and interpretable tool for forecasting daily maximum PM2.5 in data-limited environments. Furthermore, integrating PM2.5 forecasts into UAV mission planning provides a proactive decision-support approach to enhance flight safety, improve surveillance efficiency, and optimize resource allocation in border security operations.