Terahertz UAV–Vehicle Links in Non-Uniform Atmospheres: Spatially Resolved Channel Modeling, Closed-Form Power Control and Learning-Aided Placement

A terahertz link from an unmanned aerial vehicle must serve a cluster of road vehicles from one position and with one power level, while smoke plumes, rain cells and fog banks attenuate only some paths. Our spatially resolved channel model integrates gas, rain, fog and smoke attenuation per path. The minimum power level required to meet every vehicle’s signal-to-noise ratio threshold has a closed form, and with equal thresholds in a uniform atmosphere the midpoint of the outermost vehicles is optimal. With mixed latency classes or non-uniform atmospheres, random forest (RF), gradient boosting (GB) and multilayer perceptron (MLP) models predict only the position, and the closed form sets the power level, meeting every threshold for known link budgets. In six standard atmospheres, learned positions come within 0.01 dB of the optimum in milliseconds, GB being most accurate, whereas particle swarm optimization needs 0.8 s for similar accuracy and direct power prediction violates thresholds in 47–66% of cases. Localized media shift the optimum by up to 380 m, costing the midpoint rule up to 4.8 dB; RF (best here) and GB cut this loss four- to fivefold in smoke and rain. On a published network, all vehicles are served with, at most, 1.4 dB more power than reported.