The Unseen Value: Quantifying Somerville’s Urban Forest Investment

The health of a city’s infrastructure often extends beyond roads and bridges; it encompasses the natural environment, particularly its urban forest. Somerville, Massachusetts, a vibrant community just north of Boston, recognizes this, boasting nearly 14,000 publicly-owned trees that contribute significantly to the city’s ecosystem and, surprisingly, resident well-being. Maintaining this vital resource isn’t free, requiring substantial ongoing investment.

Understanding the growth and survival rates of these trees is crucial for efficient resource allocation and maximizing public benefit. Somerville’s commitment to data-driven decision-making led them to seek a way to unify two separate tree inventory datasets—one from 2009 and another from 2017—to gain a clearer picture of their urban forest's evolution. This project, undertaken in partnership with Harvard University's Data Science Master’s program, aimed to create a "tree matching" process, linking trees across both inventories.

The challenge lies in the lack of a direct link between records in the two datasets. Without a means of connecting these records, analyzing trends in tree growth and survival becomes significantly more difficult. This limitation highlights the broader issue of integrating disparate data sources for improved urban planning and environmental management.

Reconciling Disparate Datasets: A Data Science Challenge

The core of this project revolves around a seemingly simple problem: matching records of individual trees across two datasets taken eight years apart. While intuitively straightforward, the process is fraught with complexities stemming from data inconsistencies, measurement errors, and the natural dynamism of an urban forest. Successfully linking these datasets isn’t merely about finding matches; it’s about accurately identifying trees that have been planted, removed, or experienced significant changes in characteristics.

The absence of ground-truth data—the ability to physically confirm matches between trees in 2009 and 2017—presents a significant hurdle. Evaluating the accuracy of any matching model becomes an exercise in statistical inference, requiring careful consideration of potential biases and errors. This underscores the importance of prioritizing interpretability over sheer algorithmic complexity; a simpler, more understandable model is preferable to a “black box” that yields accurate results but lacks transparency.

The Somerville Urban Forestry group requires a solution that not only produces accurate matches but also allows them to understand how those matches were made. This requirement necessitates a focus on explainability, ensuring that the process can be scrutinized and refined by subject-matter experts.

Leveraging Diverse Data Sources for Enhanced Accuracy

To tackle this matching problem, the project team expanded beyond the initial tree inventory d