Conformal Prediction Intervals for Satellite-Based Crop-Yield Regression Under Distribution Shift: Coverage Guarantees and Insurance Payout Fairness in Smallholder Schemes

Satellite-based crop-yield regression is increasingly used to trigger index insurance payouts for smallholder farmers, yet the prediction intervals that accompany these regressions are typically constructed under an exchangeability assumption that fails when the deployment season, region or sensor configuration differs from the training data. The framework is evaluated through a replication-based replication study 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. Across the 600 replications, empirical coverage of the nominal 95% confidence intervals was 1.0 for both the intercept and the exposure coefficient, estimated bias was 0.04357 for the intercept and -0.00045 for exposure, and root-mean-squared error was 0.82824 for the intercept and 0.01517 for exposure. Marginal coverage is necessary but insufficient for payout fairness, and it derives a subgroup-conditional coverage condition that index insurance schemes must satisfy if they are to avoid systematically under-compensating the farmers who most need protection.