Evaluation of Causal-Backdoor-Adjusted Yield Estimation for Computer Agriculture Satellite Cropland Masking Under Cloud-Cover Confounding: An Artificial Intelligence Sensitivity Bound for Smallholder Food-Security Targeting

Satellite-derived cropland masks used in smallholder food-security targeting are confounded by cloud cover that is correlated with both agricultural potential and yield outcomes, so that naive yield comparisons between masked and unmasked areas conflate the effect of masking with the effect of the atmosphere. The estimator is formalised within a directed acyclic graph that separates cloud-cover pathways from agronomic pathways, and a sensitivity bound is derived that expresses the bias of the unadjusted estimator as a function of the strength of the cloud-yield association and the prevalence of masking. A before-after-control-impact design is evaluated under a declared variance structure with between-cluster standard deviation 6.0, residual standard deviation 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 design required 54 clusters to detect a 5 percentage point effect at 80 per cent power. The intraclass correlation of 0.64 dominates the power envelope, that the minimum detectable effect falls steeply as clusters increase from 8 to 32, and that the backdoor adjustment removes a bias that would otherwise exceed the effect being sought. The contribution is a sensitivity bound that lets analysts state, before data collection, whether a proposed cropland-masking evaluation can identify a food-security effect of policy-relevant size.