Analysing urban local cold air dynamics and climate functional zones using interpretable machine learning: A case study of Tianhe district, Guangzhou

Abstract

Deterioration of the thermal environment in built-up areas poses a serious threat to human health, comfort, and urban infrastructure, while also increasing energy consumption and carbon emissions. This underscores the need to optimize wind environments as a key mitigation strategy for urban areas. This paper analyzed the effects of human activities and natural factors on local cold air in Tianhe District, Guangzhou, from the perspective of local ventilation systems. The KLAM_21 (Kaltluft Abfluss Modell) was used to simulate local cold air flow and delineate climate functional zones. A random forest model, interpreted with the SHapley Additive exPlanation (SHAP) method, assessed the impact of various factors on local cold air dynamics. The study found that: (1) The northern mountainous area is a crucial cold source; (2) Some open spaces in the built environment fail to function as effective local cold air corridors; (3) High-intensity urban development hinders local cold air transmission; (4) Water bodies are more effective than green spaces in collecting and transmitting local cold air. This study provided technical methods for identifying climate functional zones and understanding local cold air dynamics, as well as theoretical support for the construction of local ventilation systems in urban areas.