AI-Driven Prediction of Optical Turbulence and Communication Reliability

Abstract This work presents a comprehensive machine learning ML-based framework for predicting optical turbulence and assessing the reliability of Free Space Optical Communication (FSOC) systems under varying atmospheric conditions. Using a year-long dataset of meteorological measurements—including temperature, humidity, windspeed, and solar radiation (SR—five advanced ML models were evaluated: Random Forests (RF), k-Nearest Neighbors (KNN), Decision Trees (DT), Gradient Boosting Regressor (GBR), and Artificial Neural Networks (ANN). The Scintillation Index (SI), a key metric for optical turbulence, was modeled as the main indicator affecting FSOC performance. Among all models, the GBR exhibited superior predictive accuracy for both SI and the FSOC outage probability (P_out). Through rigorous validation and interpretability analysis using SHAP values, the model was shown to capture critical atmospheric dynamics that influence signal reliability. These results highlight the potential of AI-driven models to significantly enhance the robustness and performance of FSOC systems by providing precise and interpretable turbulence predictions.