Progress 07/15/18 to 03/14/19

Outputs
Target Audience:Azavea's researchdemonstrated the use of street-level panoramas to discern the presence of street trees using visual and computer-assisted methods and to derive the geographic coordinates and related tree data that can serve as a supplement or even a first-pass alternative to traditional inventory methods.The primary target audience includes municipal, regional, and state governments across the United States and potentially worldwide, as well as local utility firms and conservation and environmental nonprofits focused on the community forest. Changes/Problems:Currently, Treetective only uses a single source of street level imagery - the data provided by Google Street View images as accessed via the Google Maps API. While that was sufficient for testing the feasibility of a prototype, we have some concerns regarding usage costs and licensing restrictions if Google Street View is the only option for data collection. Additionally, we have some concerns regarding how interacting with the imagery may counter some of Google's terms of service, despite initial reassurances from our mapping contacts there that our work was allowed.Since the results of our Phase I testing have proven the accuracy of data gathering, we are proposing expanding the sources of street level imagery as part of a Phase II grant. What opportunities for training and professional development has the project provided?Dr. Berland is one of the foremost experts in the use of remote ground-level imagery for gathering tree inventory data. His collaboration with our project provided valuable insight when we selected data fields for collection, created a species list, and chose inventories to compare to the Treetective gathered data. His analysis was also vital to determining the accuracy of the gathered data and pointing out opportunities for expanding data collection and conducting further evaluation as part of a future grant. How have the results been disseminated to communities of interest?In addition to USDA/NIFA, the information in this report has been shared with Azavea staff, as well as with our consultant, Professor Adam Berland. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Azavea began work on the Phase I Treetective project by designing a prototype application for gathering tree location data from ground-level imagery. A key element of Treetective is the ease of entering tree data. We considered several visual layouts before selecting an option that integrated the data entry form and ground-level imagery into a single screen. After completion of the prototype application, we hired a student intern, Ethan Leatherbarrow, to perform sample inventories that would test Treetective's geolocation accuracy and completeness capabilities against field-verified data previously collected in the same geographic location.Mr. Leatherbarrow has community forestry experience through his work with UC Green, a network of volunteer greening enthusiasts in Philadelphia, where he has been responsible for tree planting and maintenance activities to enhance public green spaces. We believe his knowledge and experience is comparable to the summer intern/community volunteer that will likely use Treetective to support smaller communities. Mr. Leatherbarrow also provided feedback on the visual design and user experience of the softwarethroughout the data collection process, and we updated several aspects of the software to improve data entry efficiency based on his input. This short feedback loop between development, design, and user enabled us to create several iterations of the user interface, despite the short grant time period. Azavea staff and Professor Adam Berland at Ball State University reviewed several available field-verified inventories in preparation for selecting inventory locations for Mr. Leatherbarrow. Our list of available locations included Philadelphia, Cincinnati, Columbus, Denver, Minneapolis, Seattle, and New York City. We evaluated each location based on the following criteria: Date the inventory was completed Dates when Street View imagery was available Number of trees available in each inventory Licensing and usage restrictions for the data Mr. Leatherbarrow's knowledge of the tree species in that location Based on this criteria, we began with an inventory conducted in Philadelphia in May-August 2015 that included less than 500 total trees. We then selected an additional inventory completed in the Cincinnati area in 2013 that included around 2,000 trees. Dr. Berland, Dr. Lara Roman from the US Forest Service Philadelphia Field Station, and Azavea staff discussed data collection options and chose to gather information for diameter, genus, species, condition (alive/dead), and whether there was any visible injury, damage, or insect infestation. Mr. Leatherbarrow gathered data on 492 trees in Philadelphia and 2,873 trees in Cincinnati. Azavea staff met with him weekly to review his progress and discuss suggestions from improving the data gathering process and the user interface for the prototype application. Upon completing both cities, Mr. Leatherbarrow began data collection within the geographic area covered by UC Green, the community greening organization with which he is affiliated. Mr. Leatherbarrow was interested in creating a baseline tree inventory dataset to compare against several years of planting records that were stored in a variety of formats. While earlier attempts at inventorying UC Green's trees have been completed in the field by volunteers, the scope and the distance between planting sites has made this a difficult and time consuming task. By first gathering divergent planting records and then using Treetective to verify the physical presence, species, and diameter-at-breast-height of selected trees, an accurate inventory was developed that both aggregates the historic records and adds current data on the trees. With Treetective, only one volunteer or staff person was required for this effort, no travel was needed, and the work could be done from a single computer within just a couple of weeks. Though not as comprehensive as a visit to the site of the tree, our Phase I research indicates that Google Street View or similar imagery can provide important information about a tree's species or genus, diameter at breast height, and general condition without the labor- and cost-intensive fieldwork requirements that are beyond the means of many small communities. Perhaps our most crucial accomplishment to date was determining that the tree data gathered via Treetective resulted in an inventory with sufficient accuracy for use in planting and managing trees. Based on his academic paper on the subject, our consultant,Professor Berland, created an evaluation methodology for comparing field-verified data with data gathered using the Treetective prototype. He manually compared the locations, diameters, and genus for the trees in Philadelphia and Cincinatti and used geospatial analysis techniques to determine the relative accuracy of the data.The Philadelphia data showed a median location accuracy of 6.1 feet, a diameter accuracy with an R2 of 0.94, and an accuracy of 75% at the genus level. While this data is not sufficiently accurate for use in scientific studies related to tree growth or mortality, it is useful for creating a baseline inventory that includes trees by general size class and genus, identifying trees that need to be removed, determining locations for new plantings or tree replacements, and guiding maintenance activities such as pruning.

Publications