Assessing and optimizing cooling intensity of UGS via improved metrics: A study based on machine learning simulation model

Abstract

The assessment of cooling intensity in Urban Green Spaces (UGS) has been widely studied to understand cooling intensity and its impacting factors, aiming to mitigate urban heat. However, current metrics face three key limitations: 1) they measure cooling intensity based on buffer distances, ignoring spatial variations within the same distance; 2) they assess cooling intensity at the UGS scale, failing to capture variations within a single UGS; and 3) they lack focus on how cooling intensity metrics can aid in optimizing UGS. This research proposes a new metric, Spatial Park Cooling Intensity (SPCI), along with a categorization method that accounts for spatial variations, enabling more accurate classification of cooling intensity. Applying SPCI to Singapore’s UGS, we classified the UGS into five categories and found that almost half of the UGS belong to the low cooling intensity categories. By employing an optimal machine learning model to simulate the correlation between cooling intensity and its influencing factors, and by applying Partial Dependence Analysis (PDA) to four representative sites, we found that increasing the proportion of trees—or a combination of trees and grasses—can effectively improve the SPCI of UGS. However, the increase in SPCI is not evenly spatially distributed. This research introduces a new metric that enhances spatial accuracy in cooling intensity assessments and, when combined with machine learning, provides a valuable tool for optimizing UGS with low cooling intensity.