Understanding Green Infrastructure-Temperature Relationships in Cold-Climate Cities: Evidence from 12 Canadian Urban Areas

Cities around the world, including those located in predominantly cold climates that were once thought to be relatively immune to warming, are experiencing rapid temperature increases and more frequent heatwave events, with substantial impacts on people’s well-being and critical urban infrastructure. Green infrastructure (GI) can help mitigate these impacts by cooling through shade and evapotranspiration, but expanding vegetation cover is increasingly difficult because of land competition driven by urbanization. This makes it critical to understand how to maximize the cooling effect of a given amount of vegetation. Key considerations include the fact that different types of vegetation can confer varying levels of cooling and that the spatial distribution of vegetation can influence its cooling impact.

In this presentation, we report preliminary findings from a large ongoing study comparing 12 cities in Southern Ontario, Canada. We mapped two types of vegetation (i.e., trees and shrubs/grass) using Sentinel satellite imagery, and examined how different aspects of their spatial patterns, quantified using landscape metrics, affect land surface temperature (LST) derived from Landsat imagery averaged over summer months. We evaluate and compare these relationships across cities of different sizes, from small cities with fewer than 500,000 residents to large metropolitan areas such as Toronto. We also investigate how the relationship between GI spatial patterns and LST varies across spatial scales, and we evaluate multiple modelling approaches, including spatial regression models, as well as advanced machine learning (ML) and deep learning (DL) models, including random forest and convolutional neural networks.

Our findings to date yield several insights:

• In all cities, the spatial pattern of GI exerts a significant influence on LST even after controlling for the total amount of vegetation. However, the relative importance of specific spatial pattern characteristics (e.g., connectivity, geometric complexity of patches) varies across cities, with distinct differences between larger and smaller urban areas.
• Consistent with existing literature, trees provide substantially greater cooling effects than shrubs/grass, although the magnitude of cooling varies meaningfully across cities.
• The influence of spatial pattern on LST is strongly scale-dependent, with relationships generally strengthening from finer to intermediate spatial scales, and also varying with the shape of analytical units.
• Spatial regression models prove essential for accurately characterizing vegetation–temperature relationships, as non-spatial models tend to overestimate effect sizes and increase the likelihood of falsely identifying significant relationships.
• While machine learning and deep learning models excel at prediction, spatial regression models continue to offer interpretative insights not captured by ML and DL models. We provide recommendations on the appropriate use of each model.

We believe these findings help fill an important knowledge gap on cold-climate cities, particularly in the Canadian context, where urban morphology may differ from that of other cold-region cities. Our results provide a more nuanced understanding of how vegetation type, spatial configuration, and scale interact to shape cooling, and they offer practical guidance for policymakers and practitioners on strategically deploying GI to maximize cooling benefits.