Quantifying causal effects of green–blue–gray landscape change on urban thermal outcomes: A spatiotemporal causal machine learning analysis across city scales

Peitao Shen , Xikai Chen, Junhan Wang, Yingchun Fu

Optimizing urban green–blue–gray (UGBG) landscapes has become increasingly important for mitigating surface urban heat island (SUHI) intensity under rapid urbanization. However, the causal effects of landscape pattern changes on SUHI remain poorly quantified, particularly regarding their heterogeneity among spatial configuration, urban scales, and temporal periods. This knowledge gap limits the development of adaptive and evidence-based landscape planning strategies for SUHI mitigation. To address this limitation, this study developed a causal machine learning framework to quantify the spatiotemporal causal effects of landscape pattern changes on SUHI across 836 urban core units in China. Four landscape pattern metrics representing landscape composition and configuration, including proportion of landscape (PLAND), effective mesh size (MESH), patch density (PD), and shape index (SHAPE), were evaluated for green landscapes, blue landscapes, and intergrated green–blue–gray (GBG) landscapes. A standardized counterfactual perturbation approach was applied by increasing each landscape metric by 1% relative to its original value to estimate the corresponding causal changes in SUHI intensity. The finding demonstrated that: The finding demonstrated that: (1) Increasing green landscape proportion consistently reduced SUHI across urban scales, with a 1% increase in green landscape PLAND producing approximately 1% cooling effects from small cities to megacities, and this cooling response remained stable over time. (2) Enhancing green landscape connectivity through increased MESH generated consistent cooling benefits across urban scales, with stronger cooling responses observed in small and medium cities, while these cooling benefits gradually weakened over time. (3) Patch density exhibited contrasting thermal responses between green and blue landscapes, as increased fragmentation weakened green landscape cooling but enhanced blue landscape cooling, demonstrating that similar spatial configuration changes may produce opposite thermal effect depending on landscape type. (4) Landscape shape complexity emerged as a major warming driver, with its adverse thermal effects intensifying with urban expansion and progressively amplifying over time. Increasing GBG landscape SHAPE caused stronger warming responses in larger cities, while green landscape SHAPE shifted from cooling effects in smaller cities to warming effects in large cities and megacities, indicating a scale-dependent transition of landscape complexity from beneficial to detrimental.
Overall, this study demonstrates that effective UGBG landscape optimization depends on understanding how different landscape patterns regulate SUHI across urban scale and over time. Increasing green landscape proportion provides a broadly applicable cooling pathway, improving green connectivity is particularly effective for small and medium cities, and controlling excessive landscape complexity is critical for large cities and megacities. These findings provide scale-dependent causal insights for optimizing urban green–blue–gray landscapes and advancing adaptive SUHI mitigation under future urbanization.