Urban greening, heat resilience, and economic activity in city cores: Evidence from 12 heat-vulnerable global cities using spatial double machine learning

• Finds a robust positive NDVI effect on nighttime lights in 12 city cores • Proposes spatial double machine learning with MLP/GNN nuisance learning • Decomposes direct and thermal-mediated effects with cluster-bootstrap inference • Indirect cooling mediation is model-sensitive and insignificant under GNN-DML • Shows spatial neighbourhood spillovers and clustered impacts on local activities Urban greening is widely promoted as a nature-based solution for heat adaptation, yet evidence on its economic co-benefits in dense urban cores remains limited and may be biased by nonlinearity and spatial dependence. This study proposes a spatial causal framework to estimate how urban-core greenery relates to fine-scale economic activity intensity, proxied by VIIRS Black Marble nighttime lights, across 12 heat-vulnerable cities (Bangkok, Cairo, Delhi, Doha, Dubai, Jakarta, Las Vegas, Manila, Mumbai, Phoenix, Singapore, Tokyo) from 2013-2024 using a polygon-year panel and a morphology-based urban-core boundary procedure. We apply double machine learning with city and year fixed effects and harmonized Earth-observation remote sensing covariates (NDVI, land surface temperature, built-up and moisture indices, and population), comparing non-spatial multilayer perceptron learners to graph neural network learners that explicitly encode polygon adjacency. Results show a robust positive association between greenery and nighttime lights in urban cores, with estimated total and direct effects remaining positive under spatially robust specifications. Mechanism analysis via causal mediation indicates that the cooling-mediated component is sensitive to spatial modelling: it appears significant under non-spatial learners but becomes statistically indistinguishable from zero when spatial dependence is absorbed by graph-based nuisance learning, consistent with neighbourhood-scale cooling spillovers and spatially structured confounding. Spatial spillover models further show that neighbouring greenness and neighbouring heat exposure meaningfully relate to local outcomes. These findings support spatially aware evaluation of greening strategies in heat-vulnerable urban settings and suggest that urban sustainability policies should consider both amenity-driven and heat-resilience pathways when targeting interventions in dense city centres. The results should be interpreted within the climatic scope of the sample and may not directly generalize to temperate-city contexts.