Causal-Backdoor Evaluation for Computer Agriculture Drone-Spray Deposit Models Under Unobserved Wind Covariates: An Artificial Intelligence Sensitivity-Bound Audit

Drone-spray deposit models used in precision agriculture are typically evaluated by predictive accuracy on held-out flight data, yet such evaluations do not establish that the model's spray-deposit predictions are causally identified when wind covariates are unobserved. The audit formalises the deposit model as a structural causal model, derives a backdoor-adjustment estimand under a declared wind-adjustment set, and constructs a sensitivity bound that quantifies the maximum bias in the dose-response function attributable to unmeasured wind confounding. The bound is expressed as a function of the partial R² of the unobserved wind field with the deposit outcome and with the treatment, and is computed over a grid of plausible wind-confounding strengths. A design-based operating-characteristics analysis was conducted using a before-after-control-impact estimator with between-cluster SD = 6.0, residual SD = 4.5, and intraclass correlation = 0.64; power was estimated across 2000 replications per design cell for cluster counts of 8, 16, 24 and 32 and effect sizes of 3, 5 and 8 percentage points. The analysis indicates that 54 clusters are required for 80% power at a 5-percentage-point effect. The audit provides a formal criterion by which computer-agriculture deposit models can be declared causally robust, or not, prior to deployment in variable-wind field conditions.