Mapping street tree shades for Australian cities using Google Street View photos and machine learning

Increasing a city’s tree canopy contributes to lowering urban temperatures by blocking shortwave radiation and increasing water evaporation. Creating more comfortable microclimates, trees also mitigate air pollution caused by everyday urban activities. Their absorptive root systems also help avoid floods during severe rains and storm surges. In this work, we evaluated and compared tree canopy covers in Melbourne and Wellington CBD areas using Google Street View (GSV) panoramas. This method considers the obstruction of tree canopies and classifies the images accordingly. By using GSV rather than satellite imagery, it represents human perception of the environment from the street level.

Taking the Google Street View (GSV) images, we developed a computer vision model to automate their classification into the categories: sky, vegetation, buildings, roads and misc. Be calculating the proportion of classes in these images, it is possible to calculate indices such as tree shade coverage at street-level for each location in the study areas. Computer vision algorithms have been chosen for this classification task as they have outperformed traditional algorithms across a variety of domains and can be adapted to new problems with relative ease.

We mapped and compared the distribution of these various indices using a GIS (Geographic information system) for Melbourne and Wellington CBD. This analysis allowed us to better understand the current state of urban vegetation, and use it as a benchmark for tree planting. In particular, tree shading in the both CBD areas were investigated to determine the areas that are most in need of additional vegetation coverage.