Aligning urban morphology with carbon emission intensity categories is essential for advancing sustainable urban development and achieving dual carbon objectives. This study utilizes data from 336 Chinese cities across 2010, 2015, and 2020 to construct multi-dimensional morphological indicators. Spectral clustering categorizes cities into four distinct classes: high-emission intensity, medium-emission ecological, medium-emission developmental, and low-emission. An integrated gradient boosting framework, combined with SHAP and PDP interpretability tools, identifies key morphological drivers and their nonlinear contributions to class assignments. Results demonstrate that morphological features exert nonlinear and threshold-dependent effects on carbon emission intensity category assignments, exhibiting substantial spatial heterogeneity across urban clusters. Core drivers, such as economic density and the landscape shape index, follow distinctly different decision pathways in each category. Furthermore, morphological factors produce non-additive interactive effects that generate region-specific shifts in classification probability. Through this classification-oriented approach, the study provides policymakers with a systematic and readily interpretable reference to inform the formulation of context-specific low-carbon spatial planning strategies.