How consistent are landscape–cooling relationships across observations? Multidimensional ECOSTRESS evidence from the Beijing–Tianjin–Hebei region

Abstract Landscape–cooling relationships identified from individual thermal observations are often interpreted as planning evidence, yet their consistency across observations and cooling dimensions remains unclear. Rather than identifying influential factors within individual observations, this study evaluates how consistently landscape–cooling associations recur across heterogeneous observations and cooling dimensions. Using 126 urban green spaces (UGS) in the Beijing–Tianjin–Hebei region, we integrated four ECOSTRESS observations with landscape and urban variables. A Gaussian surface model quantified cooling intensity, distance, and efficiency, while random forest, SHAP, and partial dependence analyses evaluated their recurrence, nonlinearity, and conditionality. Cooling magnitude and heterogeneity varied substantially across observations, with the lowest values at 00:31 and the highest at 10:50 across all three cooling dimensions. Green-space area showed the most recurring and predominantly positive association, whereas vegetation condition, spatial cohesion, and morphological and contextual factors were more observation- or dimension dependent. Even recurring associations were nonlinear and conditional; area showed diminishing marginal gains around 20 ha in several models, while vegetation responses varied across structural and contextual combinations. These findings show that cross-observation consistency is a matter of degree and support distinguishing recurring from context-dependent evidence for differentiated UGS planning.