Partnering with Green City Watch to co-explore the application of an open-source tree-detecting algorithm
Editor’s Note: This piece was co-written by Nadina Galle (Green City Watch) and Jaclyn Youngblood (New Urban Mechanics), describing a grant-funded prototype from December 2019. After COVID-19 delays, both organizations are now co-publishing this piece, associated open datasets, and the open-source code library.
The foundation of any effective urban forest management program is a tree inventory. A powerful tree inventory must be as dynamic as the trees themselves. For the past year, our partners at Green City Watch have been building TreeTect, an AI-enabled enhanced digital tree inventory which combines ecological expertise, satellite imagery, and machine learning to draw actionable insights about urban trees in near real-time.
That sounds bold, we know. It was a challenge, but also essential.
Here in Boston, like in many cities, we lack a frequently updated, robust tree inventory, and it can be easy to lose sight of just how much urban green space we’ve lost because of it. However, with today’s technology, we now know that each year US cities have lost over 36 million trees and will continue to lose trees. Globally, that number is likely to be well into the hundreds of millions.
Jaclyn Youngblood from the Mayor’s Office of New Urban Mechanics (MONUM), the City of Boston’s civic innovation team, met Nadina Galle from Green City Watch the way many of MONUM’s introductions happen: serendipitously and through a connection from a seemingly unrelated project about food access.
In July 2019, we had our first of a handful of exploratory meetings, discussing our organizations and goals, and imagining the potential contours of a collaborative prototype between our teams.
Green City Watch’s proposal of a collaborative prototype to use satellite imagery and machine learning to detect trees — and information about their health — sat at the intersection of MONUM’s “sweet spots”: exploring novel applications of technology, prototyping more equitable infrastructural investments, and working on topics that complement or uplift work that City departments are actively interested in advancing in new or different ways.
The imagined prototype would allow both of our teams to start from a place of strength. Building on the City of Boston’s Streetcaster work to bring equity to street and sidewalk repairs, MONUM saw in the TreeTect idea an opportunity to explore equity in data access to build a more robust tree inventory.
Urban deforestation as a result of human activity not only decreases the health and quality of life in our cities, but it also worsens the effects of an unpredictable changing climate. Urban trees protect our cities from the effects of harsher storms and more intense heat waves, creating more resilient cities in the process. Not to mention that they actively sequester carbon dioxide and filter out pollution.
Though many think that deforestation happens only in the Amazonian jungle or deep into the heart of Borneo, our cities are facing tree loss at a massive and unprecedented rate. We need to ensure that arborists and urban foresters are equipped to act, not just gather information.
In cities, a total overview of a tree inventory takes around 3–7 years due to manual surveys; which is about half the life of a street tree. If you only have a handful of arborists at your service — which is true (or more than true) for many cities and towns around the U.S. — it’s imperative that you free up their time from collecting data to acting.
So often, those combating urban deforestation don’t have the tools they need to make a significant impact. Changing this starts with changing the way that cities collect information on their tree inventory. And that’s what we set out to do.
Prototyping in the City of Boston
Speak For the Trees, a local advocacy organization in Boston, had cataloged roughly a third of Boston’s public street trees.
However, there remained barriers to getting a more timely and robust picture of the full urban tree canopy. We were curious to explore whether the use of hi-resolution satellite imagery could help us do so. We wanted to learn by doing.
Often, urban tree inventories are incomplete even when they seem “done”. That’s because many of them focus only on publicly owned trees. However, in many cities around the world, over half of the trees and half of the canopy cover is on private lands. A city may have limited jurisdiction over these trees but still be interested in community benefits these healthy trees can offer.
In December 2019, in collaboration with our partners in the Department of Innovation and Technology’s Analytics Team, the Parks Department, the Environment Department, and the Boston Planning and Development Agency, Green City Watch applied its TreeTect technology to create a baseline inventory of over 2,000 trees that were detected in Nubian Square.
The internal City of Boston partners knew we wanted to explore this prototype in a place where tree information had already been collected by Speak for the Trees, so we could use the additional insight from the prototype to add to what we already knew. We also wanted to select an area that had disproportionately high heat island effects compared to the citywide average. Ultimately, Nubian Square was chosen as our prototype-testing location.
Across all internal City of Boston partners, we were curious whether TreeTect could help expand the geography and depth of our already existing tree inventory work. We were also interested in the potential to understand more about trees on private land, particularly as it relates to work we do with institutional partners who own land in the city, such as universities. We wondered about questions such as:
- How might we set species diversity goals?
- How can streetscape improvements accommodate tree planting and support tree health?
- What are the environmental and financial values of trees in the study area?
In an overarching way, we wanted to understand whether a technology like this could help us have more active and holistic management of our urban tree canopy, which we hoped would inform a more proactive strategy about individual tree care and maintenance.
There were a number of internal departments partnering on this work from the City of Boston side. As a true collaborative partner, Green City Watch facilitated a series of discussions to help us internally align on priorities, build hypotheses to test with the prototype, and engage in some creative imagining for the future. These conversations were critical to ensuring the broad knowledge of our internal stakeholder groups was brought to bear on the research and prototype.
During those conversations, our collective explored more topics than the prototype ended up taking on. That often happens with exploratory prototypes and it’s a sign that there are more good questions to be engaged. One of those, which is still very much of interest to MONUM and Green City Watch, is whether and how TreeTect can be coupled with a civic engagement initiative, such as Adopt-a-Tree, to better support residents in being active stewards of their neighborhood trees.
With the Parks Department, we also discussed the potential to remotely identify dead trees before residents call them into 311, which might allow us to more proactively remove potentially dangerous trees while doing regular tree care and maintenance in an area, before a large storm comes through and makes those dead or sick trees a potential hazard.
The TreeTect prototype, which was first-of-its-kind in the US, used satellite imagery and machine learning to pinpoint each individual tree location, size, and condition. By mapping the health of individual trees across the area, we detected thriving hotspots and identified how trees of various conditions are distributed throughout the Nubian Square area (see map above). As a result of the project, the data provided a baseline map which could lead to the revival of unhealthy trees, the removal of dead trees, and the filling of empty tree pits with new trees.
Using this approach, tree managers can actively build a healthier urban environment by acting on, rather than gathering, data. This is especially necessary when considering that they are facing increasing pressures due to diminishing budgets, urban deforestation, and the effects of climate change.
While it is commonly known that trees are beneficial for cities and their residents, urban tree canopies continue to disappear. Increased density of urban forests improves air quality, creates a natural cooling effect, and has a range of health benefits. Yet, the average lifespan of an urban tree is between 13 to 20 years old, and while the majority of social and ecological benefits come with age, most never reach adulthood (> 25 years old).
By accumulating data through hi-resolution satellite imagery and analyzing it through machine learning, this process frees up crucial man-hours for the local authorities to tend to trees in the field. The results of the prototype, which ran over the course of six weeks, saw data accumulated from trees on both public and private land — which was not possible when done in the traditional manner — and could help prioritize new plantings and removals to create healthier green canopies.
Moving forward, open-source
The Nubian Square prototype showed us that technology can and should be involved in creating greener, more livable cities. However, to build livable cities, we must first understand our starting point and “taking nature online” allows us to do just that. In the right context, and if we’re asking the right questions, technology can often help us get there faster and cheaper.
Green City Watch’s mission is to revolutionize the way we value nature, bring transparency to local government, and regenerate our cities. We also believe in open data and mobilizing science into the public realm.
To get there faster and cheaper than ever before, we are releasing our workflow to detect urban tree locations via Github. Using our repository you can install the workflow in your own AWS environment. The code is made to run in AWS lambda and will allow you to optimize the tree detection algorithm for any area.
Give it a try and compute the tree inventory for your city!
If you’re curious about the prototype process, feel free to reach out to MONUM at firstname.lastname@example.org. Much like our partners at Green City Watch, we see collaboration, sharing what we’ve learned with others, and paying it forward as integral to our work in civic innovation.
Green City Watch is thrilled to welcome contributions from developers and organizations worldwide to provide cities and citizens with a better overview of their tree inventory. Together, we can map more trees in more cities and develop the technology to identify a host of other features, like species identification, and empower tree managers to take nature online.