Expanding urban tree canopy cover is an important strategy for mitigating climate change, but estimating green space is an unresolved problem for spatial researchers.
We aimed to reimplement Cai et al.’s approach for estimating canopy cover using convolutional neural networks and compared it with a simpler architecture, DeepGreen. We also mapped and visualized the results for Providence’s canopy cover, and interpreted the model using the LIME explainer and Grad-CAM. We achieved a mean-absolute error for the tree canopy cover estimation of 0.0909 for Cai et al.’s approach and 0.0969 for the DeepGreen approach.