Benchmarking Evaluation of Counterfactual Data Augmentation for Computer Agriculture Crop-Yield Estimators Under Satellite-Cloud Gaps: An Artificial Intelligence Coverage-Error Bound for Famine Early-Warning Reliability

Satellite-cloud gaps corrupt the reflectance time series on which computer-agriculture crop-yield estimators depend, and the resulting coverage error propagates into famine early-warning reliability. The framework is evaluated through a replication-based benchmark in which 600 data sets of N = 500 were drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2) and the ordinary least squares estimator was characterised under the stated process. The benchmark produced an estimated bias of -0.07173 for the intercept and 0.00147 for the exposure coefficient, with RMSE values of 0.89562 and 0.0157 respectively, and empirical coverage of the nominal 95% confidence intervals of 1.0 for both parameters. The coverage-error bound formalises the conditions under which CDA-supported estimators preserve nominal coverage and therefore early-warning reliability, and it identifies the augmentation regimes in which coverage degrades. The contribution is a benchmark protocol and a bound that make the reliability consequences of cloud-gap augmentation auditable rather than assumed.