Spatiotemporal Causal Discovery for Artificial Intelligence Agriculture Fertilizer Advisory Under Unobserved Plot-Level Confounding: A Computer Transportability Audit

Fertilizer advisory systems built on spatiotemporal causal discovery promise plot-specific nitrogen recommendations, yet their transportability from the trial network to the farm gate depends on assumptions that are rarely audited. the present analysis treats the before-after-control-impact design as the inferential backbone of such systems and characterises its operating characteristics under a declared variance structure in which the between-cluster standard deviation is 6.0, the residual standard deviation is 4.5 and the intraclass correlation is 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.56 clusters are required to detect a 5 percentage point effect at 80 per cent power, that the minimum detectable effect at 80 per cent power falls steeply as clusters are added, and that the interaction estimate is unbiased but imprecise at the smaller cluster counts. A transportability audit is proposed in which the design parameters, the confounding structure and the decision threshold are reported alongside every advisory model, so that the conditions under which a plot-level recommendation is identified are visible to the agronomist who acts on it. The contribution is a formal statement of the identification conditions for spatiotemporal fertilizer advisory and a demonstration that the before-after-control-impact design, correctly powered, is a defensible inferential engine for it.