Differentially Private computed Generation of Artificial Intelligence Climate Station Records Under Heterogeneous Sensor Coverage: Utility Bounds and Downstream Forecast Skill Retention

Differential privacy offers a formal guarantee for releasing climate station records, but the utility of privatised data under heterogeneous sensor coverage remains poorly characterised. The framework is applied to a real reanalysis dataset for Accra, Ghana (2000–2023, 312 monthly observations), which exhibits a mean temperature of 26.521 °C and a warming trend of 0.1622 °C per decade. The utility of computed records is governed by an effective sample size that depends on the spatial coverage of sensors, and that forecast skill retention decays predictably as the privacy budget tightens. The derived bounds are validated through a replication study that mimics heterogeneous coverage, and the results indicate that even moderate privacy budgets can preserve forecast skill when sensor density exceeds a critical threshold. The paper concludes that privacy-preserving computed data can be a viable tool for climate analytics, provided that coverage heterogeneity is explicitly accounted for in the generation mechanism.