Accurate characterization of atmospheric turbulence is essential for optical satellite communication, particularly ground-to-satellite downlinks in tropical regions such as Vietnam, where the refractive-index structure parameter $C_n^2$ governs turbulence strength and directly affects key link metrics - Rytov variance, Fried coherence length, and isoplanatic angle. However, no public dataset captures Vietnam's climatic and geographical diversity, limiting the development of turbulence-prediction models for the region. We introduce the Vietnam Atmospheric Dataset (VAD), the first comprehensive dataset for optical satellite link turbulence prediction in Vietnam, combining a 12-month campaign across six stations in three climatic regions - north (Sa Pa, Ha Noi), central (Nha Trang, Da Lat), and south (Ho Chi Minh City, Can Tho). VAD integrates co-located $C_n^2$ profiles from scintillometers and radiosondes with surface meteorological data and astronomical seeing estimates, totaling over 525,600 records - the largest public dataset of its kind in the region. To demonstrate its value, we develop an Adaptive Optimization Algorithm (AOA) combining gradient boosting and Bayesian optimization to estimate high-resolution $C_n^2$ profiles, achieving an RMSE of $1.42 imes 10^{-15} ext{ m}^{-2/3}$ and a MAPE of 8.3%, outperforming conventional and state-of-the-art models by up to 34%. VAD and its source code are publicly released to support future research on turbulence modeling and optical satellite communication in tropical environments.

Atmospheric turbulence,,Bayesian optimization,optical satellite communications,,refractive-index structure parameter Cn2,,Vietnam Atmospheric Dataset (VAD),,machine learning,,Bayesian optimization