Urban green space morphology (UGSM) yields more overall ecological benefits in high-density urban areas. However, the types of UGSMs and the scale effects of the spatial heterogeneity of UGSM indicators (seven MSPs) on PM2.5 pollution remain largely unknown. Multiscale geographically weighted regression (MGWR) was used to analyse the relationship between UGSM and PM2.5 at four geographical scales across five periods in Wuhan, generating a novel multiscale spatiotemporal evolution analytical framework (multi-SSTE). Morphological spatial pattern analysis and spatial autocorrelation analysis methods were used to identify UGSMs. The evolution of UGSM took the form of “polygon UGSM –> line–polygon UGSM –> point–line–polygon UGSM”. Over 60% of the changes in PM2.5 were explained by UGSMs, with the best performance in reducing PM2.5 obtained in the point–line–polygon UGSM. Compared with ordinary least squares (OLS) regression, MGWR significantly increased the R2 value (OLS: 0.15–0.51, MGWR: 0.55–0.95), but this advantage decreased with an increase in geographical scale. The results suggested that 2 km × 2 km was the best scale for analysis of the spatial heterogeneity between UGSM and PM2.5. In terms of seven MSPs, the cores had the most stable local spatial variable properties in the UGSMs in 2000, 2010 and 2015 (bandwidths: 43–50). Our research not only sheds light on the complex relationship between UGSM and PM2.5 but also contributes a common theoretical framework for other ecological environment issues.