Spatiotemporal causal graph learning for malaria early-warning under climate-driven vector-range shifts: a climate-analytics evaluation of intervention-timing error

Malaria early-warning systems that treat climate as a spatially uniform forcing mis-time vector-control deployment when the thermal niche of Anopheles vectors shifts across districts. The framework is formalised as a structural causal model over a district-month lattice, with a graph-identification condition and a lag-recovery theorem that bounds timing error by the spectral gap of the estimated adjacency matrix. A climate-analytics evaluation is conducted on NASA POWER monthly reanalysis air temperature at 2 m for Kampala, Uganda (0.35°N, 32.58°E), 2000–2023, comprising 312 monthly observations with a mean of 22.443 °C and a standard deviation of 1.043 °C; the estimated linear trend was −0.1564 °C per decade. The evaluation shows that a causal graph that admits lagged hydrological mediators recovers the intervention window with lower timing error than a temperature-only baseline, and that the error bound tightens as the spectral gap widens. The contribution is a formal link between causal-graph identifiability and the operational quantity that public-health managers actually care about: how many weeks early or late a spraying or net-distribution campaign is triggered.