Multiscale driving mechanisms and spatial non-stationarity of landscape ecological risk based on an OPGD–MGWR framework: evidence from township-level units in the Nanjing metropolitan area (NMA), China

Rapid urbanization and industrial restructuring reshape landscapes and may elevate landscape ecological risk (LER) with strong spatial heterogeneity and scale dependence. To support territorial spatial planning with governance-relevant units, we assessed LER for township-level units (towns and subdistricts) in the Nanjing Metropolitan Area (NMA), China, and explored its drivers using a coupled Optimal-Parameter Geodetector and multiscale geographically weighted regression (OPGD-MGWR) framework. Using GlobeLand30 land-use maps for 2000, 2010 and 2020, we constructed an LER index, examined spatial clustering (Global Moran's I and LISA), and quantified global determinants, interactions, and locally varying effects. From 2000 to 2020, cropland declined while built-up land expanded, and LER remained moderate to high, forming a persistent concentric pattern (low core–high periphery–low outer edge) with significant positive spatial autocorrelation (Moran's I > 0.43). OPGD identified highway accessibility and nighttime lights as the dominant factors, with most factor pairs showing bivariate or nonlinear enhancement. MGWR revealed clear scale heterogeneity: NDVI and distance to district/county centers had near-global negative effects, whereas nighttime lights and water area exhibited strong local non-stationarity. The proposed township-level framework links risk levels, dominant drivers, and effective scales, enabling differentiated risk governance across multi-jurisdictional metropolitan regions. • Landscape ecological risk was assessed at township level from 2000 to 2020. • Cropland loss and built-up growth reinforced persistent landscape ecological risk. • Risk showed a persistent concentric low–high–low spatial pattern. • Highway access and night lights were dominant risk drivers. • Multiscale regression revealed strong spatial heterogeneity in drivers.