Causal forest estimation of irrigation-expansion feedbacks on convective initiation from paired satellite retrievals: a climate-analytics evaluation of downwind rainfall attribution

The attribution of downwind rainfall changes to irrigation expansion remains contested because irrigation modifies both surface moisture and boundary-layer thermodynamics, and because satellite retrievals of convective initiation are sparse and noisy. The framework combines a potential-outcomes formulation with a honest random-forest splitting rule that permits effect modification by land-surface and atmospheric covariates. A formal result establishes that the forest estimator is consistent for the conditional average treatment effect when the propensity score is known and the retrieval error is classical, and a sensitivity bound quantifies bias under differential misclassification. The framework is then applied in a methodological evaluation using NASA POWER monthly reanalysis for Lagos, Nigeria (2000–2023, 312 monthly observations), where the mean temperature was 26.414 °C and the warming trend was 0.1553 °C per decade. The evaluation clarifies the conditions under which downwind rainfall attribution is credible and identifies the retrieval-error regimes in which causal-forest estimates should be preferred to regression-adjustment alternatives.