Advancing Clear-Air Turbulence Detection with Hybrid Predictive Models for a Regional Aviation Corridor in Southeast Brazil

Severe Clear-Air Turbulence (severe CAT) remains a relevant hazard to aviation safety, often occurring without visible atmospheric indicators. This study presents a hybrid forecasting framework that integrates outputs from the Global Forecast System (GFS025) with advanced machine learning (ML) algorithms to predict severe CAT events over Southeast Brazil, within the region bounded by 43°W to 49°W and 19°S to 25°S, from January 2018 to December 2021. To enhance predictive performance and reduce model complexity, a statistically robust dimensionality reduction technique was applied using p-value filtering and False Discovery Rate (FDR) control, resulting in a refined set of 13 physically interpretable predictors. Key turbulence indices, such as Ellrod’s index (ELL2) and Brown’s index (BROWN), emerged as the most relevant features for classification. Nine ML algorithms were tested and evaluated through Receiver Operating Characteristic (ROC) analysis and Area Under the Curve (AUC) scores. The Multi-Layer Perceptron (MLP) model, with a single hidden layer of 10 neurons, achieved the highest AUC (0.95), followed closely by Random Forest (0.94), demonstrating the effectiveness of relatively simple architectures when coupled with feature selection. These findings underscore the value of combining physically consistent diagnostics with data-driven methods for regional severe CAT forecasting. The proposed approach offers a scalable and adaptable framework that supports enhanced aviation safety and provides a solid foundation for the continued development of operational turbulence prediction tools.