The transition toward urban carbon neutrality has intensified attention on the coupling mechanisms between urban spatial morphology and carbon emissions. However, existing studies are often constrained by single-scale analyses and linear assumptions, limiting their ability to capture nonlinear interactions among multiple morphological indicators and spatial heterogeneity. Supported by multi-source remote sensing data, this study advances and integrates geographically weighted regression, ensemble machine learning, and game-theoretic feature attribution into an integrated spatiotemporal nonlinear regression (ISTNR) model, which is applied to systematically analyze how the evolution of urban morphology influences carbon emissions across different spatial and temporal scales in Japan. The results highlight significant regional heterogeneity and nonlinear interactions. In dense urban areas like Tokyo and Osaka, emissions are mainly driven by compactness, while in sparsely populated regions such as Hokkaido, shape complexity and fragmentation dominate. Spatial connectivity shows asymmetric effects across regions. Key thresholds are identified: emissions increase rapidly when urban area exceeds 60,000 ha or connectivity surpasses 0.85. Conversely, mitigation potential is highest when adjacent urban land comprises 55–70% and population density ranges between 0.004 and 0.007 persons/m2. Strong interaction effects are also observed—for instance, the amplification of emissions when total urban area and spatial connectivity rise simultaneously, and the reversal in the influence of landscape mesh complexity and land aggregation depending on spatial scale. This study establishes a flexible and interpretable nonlinear modeling framework, offering a novel methodological basis for identifying optimal morphological zones and guiding low-carbon spatial planning.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号