Causal mediation analysis of heat-stress yield losses via pollination-window timing: a climate-analytics estimation of adaptation-window misallocation in hybrid-maize advisories

Hybrid-maize advisories in tropical highlands typically schedule planting and heat-mitigation interventions against seasonal mean temperature, yet the yield-relevant exposure is concentrated in a short pollination window whose timing shifts with warming. The framework is formalised as a set of structural equations with a window-timing mediator, identified under sequential ignorability, and estimated with a mediation formula that separates natural direct and natural indirect effects. The framework is then applied to a reanalysis record for Nairobi, Kenya: NASA POWER monthly MERRA-2 air temperature at 2 m for 2000–2023, comprising 312 monthly observations with a mean of 20.146 °C, a standard deviation of 1.094 °C, and a linear warming trend of 0.2477 °C per decade. The record is used to characterise the exposure distribution that advisories face, not to estimate maize yields, which were not observed. Under a plausible range of thermal sensitivity and window-displacement parameters, the mediation decomposition implies that a substantial share of warming-attributable loss can be indirect, and that advisories keyed to seasonal means systematically misallocate adaptation resources toward the direct pathway. The contribution is a transferable identification template and a diagnostic for advisory design, not a site-specific yield estimate.