Rancang Model Prediksi Potensi Hujan Berbasis Stabilitas Atmosfer di Kawasan Bandara

Abstract

This study aims to develop a rainfall potential prediction model for airport areas using an Artificial Neural Network (ANN) based on atmospheric stability indices. The case study was conducted at Soekarno–Hatta International Airport using radiosonde, AWOS, and METAR data for the period 2020-2024 at 00 and 12 UTC, with predictions of rainfall occurrence and intensity up to 12 hours after radiosonde observations. The research stages include data preprocessing consisting of data cleaning, outlier handling using the interquartile range (IQR) method, Min-Max normalization to a range of -1 to 1, and class balancing using the Synthetic Minority Over-Sampling Technique (SMOTE). Two ANN models were developed separately: a rainfall occurrence model (rain/no rain) and a rainfall intensity model (light, moderate, and heavy rainfall), with a time-based data split of 2020-2022 for training, 2023 for validation, and 2024 for testing. Three base modeling approaches; ANN based on feature selection, stepwise regression, and Principal Component Analysis (PCA), were combined into an ensemble model. The results show that for rainfall occurrence prediction, the feature selection model achieved the highest accuracy of 0.79, while the ensemble model improved the accuracy to 0.80. For rainfall intensity prediction, the ensemble model provided the best performance with an accuracy of 0.61. The selected atmospheric stability indices represent atmospheric instability, convective energy, and water vapor content, which are key components in the formation of convective rainfall in tropical regions. Overall, the ensemble-based ANN model shows potential to support short-term rainfall prediction in airport areas, although further performance improvements are still required, particularly for predicting extreme rainfall intensity. As an operational implementation, the model was integrated into a Graphical User Interface (GUI) that allows both manual and batch data input, displays rainfall probabilities and classes, and stores prediction results. This GUI facilitates forecasters and airport operational personnel in supporting short-term rainfall forecasting and decision-making.