Regional geographic gradients dominate the association between apparent temperature and fine-particulate air pollution across China’s Yangtze River Delta at 1 km resolution

Outdoor heat and fine-particulate air pollution are two of the most important environmental pressures in dense urban regions, yet they are usually mapped separately and at coarse spatial scales. We integrate three independent, openly available, model-derived products for the year 2020 onto a common 1 km grid across the Yangtze River Delta (YRD) urban agglomeration, namely outdoor apparent temperature from the HiTIC collection, fine particulate matter from the ChinaHighPM2.5 (CHAP) product, and land cover from the 30 m China Land Cover Dataset. We treat apparent temperature as a modelled proxy for outdoor thermal conditions rather than as a measure of human thermal comfort, since the data carry no information on population behaviour, exposure time or vulnerability. The harmonised analysis is based on 316,813 grid cells. Apparent temperature and PM2.5 exhibit opposing north-south gradients, and they are correlated at minus 0.37 when all cells are combined, but are mostly driven by geographic structures within the region and are not closely correlated in general. Both fields are almost perfectly spatially clustered (Global Moran’s I of 0.96 and 1.00), i.e. the effective number of samples is several orders of magnitude smaller than the nominal number of cells; the number of cells varies from about 11 to 222 cells in a sensitivity analysis for different sizes of bootstrap blocks, but the pooled correlation is negative in all cases, and the confidence interval exclude zero correlation in all cases. Stratifying by land cover attenuates the pooled slope and reverses it in the forest subgroup, a partial reversal consistent with Simpson’s paradox. The association also varies by season, from minus 0.77 in winter to plus 0.19 in summer, so annual means conceal much of the structure. A random forest using only apparent temperature and land-cover fractions reproduces the broad spatial pattern of the modelled CHAP surface with a cross-validated coefficient of determination of about 0.85, which should be read as reproduction of that product rather than as independent prediction of observed PM2.5. Forest fraction is the strongest correlate of lower PM2.5 and remains negative after full spatial adjustment, but because land cover is itself an input to both source products, part of this association may be inherited from their construction. The contribution of this work is an integrated, reproducible regional assessment and a demonstration that regional spatial gradients can systematically dominate observed correlations in high-resolution gridded environmental data; the findings are based on a single year and require multi-year validation.