As public service facilities, urban parks offer many benefits for daily life and social activities for residents. However, the accessibility of public parks to urban residents is often unevenly distributed in spaces that cannot be utilized fully. Here, we used the urban parks in Beijing, China as a case study and examined the relationship between urban park accessibility and population distribution at different administrative levels. Gini coefficient and Lorenz curve were used to evaluate the social equity of urban park accessibility, and the location quotient was used to identify the spatial difference between urban parks and resident population. The results of our study show that the urban park accessibility varies at district and subdistrict levels and that places with more urban parks usually have higher accessibility. Very importantly, the spatial equity is different from the social equity, a mismatch exists between the spatial distribution of urban parks and population, particularly for the elderly residents. These results generate valuable insights, as, in China and many developing countries, current urban public green space planning only uses the ratio of public green space to urban construction land and the per capita public green area.

Urban renewal decisions require joint consideration of parcel-level conditions and the surrounding urban environment, yet existing urban vacant land (UVL) typologies remain primarily anchored in intrinsic parcel attributes, with contextual information appearing mainly as descriptive backdrop. Whether the two attribute dimensions are typologically commensurable, in the sense of yielding classification structures of comparable reliability and stable correspondence, remains untested. This study examines the dimension-dependent classifiability of UVL using a probabilistic clustering approach. Gaussian Mixture Models (GMM) are applied independently to intrinsic and contextual features of 2,290 parcels in Shenzhen, China, and their clustering convergence, stability, and classification certainty are systematically compared. The results reveal a clear structural asymmetry. The contextual dimension converges toward stable clustering structures with high classification certainty (AvePP 92.0–93.6%) and reproducible assignments (Jaccard 0.58–0.67), whereas the intrinsic dimension exhibits persistent instability and low reproducibility (Jaccard 0.34–0.46). The two typologies exhibit negligible association (Cramér’s V = 0.08) and can be interpreted as largely independent. A certainty-based quadrant framework translates classification uncertainty into a parcel-level framework, differentiating parcels whose typological labels can be used with greater confidence from those requiring supplementary investigation or field verification. These findings demonstrate that the classifiability of UVL is dimension-dependent, reflecting the distinction between institutionally regulated and residually generated processes shaping urban space. Shenzhen’s strongly planned, high-density context provides a favorable setting for observing this mechanism. The framework contributes a certainty-aware approach to UVL typological assessment and a testable hypothesis on when dimensional asymmetry should be expected.