Sensitivity-Analysis Evaluation for Computer Agriculture Soil-Moisture Sensor Networks Under Unobserved Probe-Drift Bias: An Artificial Intelligence Error-Propagation Bound

Unobserved drift in soil-moisture probes is a persistent threat to the reliability of computer-agriculture irrigation decisions, yet the sensitivity of regression-based evaluation estimators to such drift has not been bounded analytically. The bound is derived from a sensitivity-analysis argument that decomposes estimator bias into a drift-attenuation term and a drift-correlation term, and it is shown to be sharp under Gaussian drift. The replication study was conducted to characterise the behaviour of the ordinary least squares estimator under the stated process. The estimated bias for the intercept was -0.01651 and for the exposure coefficient 5e-05; the root mean squared error was 0.84929 for the intercept and 0.01533 for the exposure coefficient; and empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters across 600 replications. The bound predicts that drift-induced bias in the exposure coefficient is bounded by the product of the drift variance ratio and the exposure standard deviation, and the replication results are consistent with that prediction. The contribution is a formal sensitivity bound that converts an unobservable probe-drift magnitude into a computable worst-case bias on the exposure coefficient, together with an evaluation protocol that computer-agriculture sensor networks can use to audit regression-based irrigation recommendations.