Causal inference on mangrove-restoration carbon-credit additionality from Landsat time series under staggered enrollment: an artificial intelligence evaluation of permanence-risk overstatement

Carbon-credit methodologies for mangrove restoration typically assume that credited carbon storage persists for a fixed horizon, yet staggered enrollment and unobserved baseline dynamics make permanence risk difficult to identify from satellite time series alone. The framework combines a potential-outcomes model for staggered adoption with a hidden-Markov representation of unobserved baseline loss, and derives bounds on the bias of the conventional permanence deduction. Analysis of the identification conditions shows that the standard deduction is unbiased only when baseline loss is independent of enrollment timing, an assumption that fails whenever enrollment is targeted at high-threat sites. The derived bounds imply that permanence-risk overstatement is increasing in the correlation between enrollment timing and baseline threat, and that the sign of the bias can be recovered from observable enrollment-date and land-cover-covariate patterns. The contribution is a diagnostic that credit issuers can apply to existing registries without new field data, together with a formal statement of the conditions under which the conventional deduction is defensible.