Counterfactual Data Augmentation for Artificial Intelligence Agriculture Pest Outbreak Forecasting Under Unobserved Surveillance Bias: A Computer Doubly-Robust Estimation Evaluation

Pest outbreak forecasting models trained on agricultural surveillance records inherit a selection mechanism that is rarely observed: fields are scouted because they are suspected of infestation, so the labelled sample is a biased draw from the population a forecast must serve. the present analysis formalises that mechanism as an unobserved surveillance-bias process and proposes a counterfactual data-augmentation estimator that combines inverse-propensity reweighting with an outcome-regression adjustment in a doubly-robust form. The estimator is evaluated in a replication study of 600 data sets of N = 500 drawn from a declared data-generating process with intercept 12.0, exposure 0.42, group -3.1, severity 1.35 and error SD 4.2, using seed 650381472. Across the 600 replications the estimated bias was 0.05537 for the intercept and -0.00067 for the exposure coefficient; the root mean squared error was 0.86781 and 0.01577 respectively; and empirical coverage of the nominal 95% confidence intervals was 1.0 for both parameters. The sampling distribution of the exposure coefficient and its normal Q-Q plot are consistent with the asymptotic approximation on which the interval construction relies. The contribution is a formal identification argument for counterfactual augmentation under unobserved surveillance bias, together with a bounded evaluation of the doubly-robust estimator's finite-sample behaviour.