Characterizing the Urban Tree Canopy (UTC) to Elevate its Role in Mitigating Climate Change and Creating a Healthy and Vibrant Community in Ann Arbor, MI

Abstract

Ann Arbor, Michigan is affectionately called “Tree Town” due to its distinguishable and valued urban tree canopy (UTC). The city consists of roughly 18,605 acres with 6,015 acres identified as UTC in 2010 (Hanou, 2010). The City of Ann Arbor recognizes the importance and potential benefits the UTC can provide in its “Urban and Community Forest Management Plan” and also mentions them in the “A2Zero Climate Action Plan”. While the forest remains a priority, the management plan was implemented eight years ago, and the last canopy assessment occurred twelve years ago. Furthermore, this assessment delineates where the canopy is but does not provide details about canopy composition. Additionally, data is available regarding the City’s street trees, but certain locations, such as parks and private land, have little available canopy data. Like many city governments working to elevate positive UTC contributions to the community at large, Ann Arbor is in need of descriptive analyses to enhance current data. Our goals were to identify turfgrass, delineate native forest fragments, and classify trees by genus. By emphasizing these three areas, we sought to elevate the role of UTC as an ecosystem service, mitigator of climate change, and guiding factor in stewardship actions. Turfgrass was identified to inform residents and decision-makers about its spatial extent. This would highlight locations that could receive greater ecosystem service benefits by reducing turfgrass to expand the urban tree canopy. To complete this analysis, aerial imagery was clipped using a Canopy Height Model, Normalized Difference Vegetation Index (NDVI) thresholds indicating live green vegetation, and an Unsupervised Classification was performed. From our assessment, turfgrass accounts for 50% of the total canopy. Native forest fragments were delineated both manually and with an unsupervised classification. Manual delineation was based on historic imagery dating back to the 1940s and refined using a Canopy Height Model. The unsupervised classification aided in identifying smaller forest fragments. Our manual delineation accounts for 14% of the total canopy, with two-thirds located outside of City-owned property. The classification results indicated native forest fragments comprising 28% of the total canopy, with three-fourths located outside of City-owned property. Genus classification was attempted to allow one to easily characterize the types of forest communities at a large scale and to aid in identifying invasive species. Tree data was collected to utilize as training/testing data, various predictor layers were associated with tree points, and machine learning models were assessed. Overall, LiDAR segmentation and the Random Forest model performed the best with the highest accuracy genus classification at 54%. While we were able to provide descriptive analyses of the UTC to enhance the City’s current dataset, the accuracy and reproducibility of our methods should be improved with future work. All three areas of focus could benefit from a better spatio-temporal alignment of imagery and tree data along with collecting ground truth data to test accuracy.