Virtually Tracking Planted Urban Tree Survival with Street-Level Imagery

Urban tree planting initiatives have become a widespread form of urban greening in cities around the world (Eisenman et al. 2022; UNECE 2024), with planting campaigns motivated by the ecosystem services provided by urban canopy (Silvera Seamens 2013). Street trees generally comprise a small fraction of overall urban canopy cover—which can be dominated by trees in parks and private properties—but street tree planting and stewardship are often the front lines of engagement for urban tree programs (Roy 2017; Elton et al. 2023). Additionally, trees are an essential component of street design to enhance urban livability (Coleman et al. 2022). In cities and towns around the globe, streetscapes can be the primary opportunity for tree planting in neighborhoods with high building density where houses do not have yards (Hall et al. 2012), including low-income and low-canopy communities (Riedman et al. 2022), making street tree planting and care critical to advancing environmental justice (Grant et al. 2024). Street trees are particularly important for mitigating urban heat islands and providing pedestrian thermal comfort (Ettinger et al. 2024; Jayasingh et al. 2024), and they are also central to green stormwater infrastructure systems in aging neighborhoods facing flooding challenges (Berland et al. 2017). Furthermore, the planting, maintenance, and removal of trees along streets dominate municipal tree management budgets (Hauer and Peterson 2016). The survival of planted street trees is therefore vital for the sustained benefits of this public resource, with a pronounced relevance in low-canopy neighborhoods.

Urban tree monitoring for mortality and survival enables urban forestry professionals and researchers to understand the performance of planting programs (Wattenhofer and Johnson 2021). A literature review revealed that typical annual mortality during the first 5 years after planting was 6.6% to 7.0%, with post-establishment annual mortality dropping to 2.8% to 3.8% (Hilbert et al. 2019). Based on those annual mortality rates, the typical population half-life for a planted tree cohort is 13 to 18 years; in other words, half the trees are dead or removed by about 15 years after planting. Improved long-term survival would yield greater ecosystem services (Widney et al. 2016). It is therefore imperative to understand the factors associated with mortality to enable realistic performance expectations and targeted program interventions (Wattenhofer and Johnson 2021; Moore 2022). With recent massive investments in urban tree planting initiatives in countries around the world, coproduced monitoring efforts across research and practice present an opportunity for learning and adaptive management (Campbell et al. 2016; Eisenman et al. 2025). Knowledge coproduction is a collaborative process to bring together different types and sources of knowledge for problem solving (Armitage et al. 2011). In the context of evaluating tree planting projects, such coproduction typically involves collaboration between researchers in government or academia and urban forestry professionals working for municipalities, states, nonprofits, or private companies (Campbell et al. 2016; Roman et al. 2018a; Vogt and Abood 2020).

Factors associated with urban tree mortality include taxonomic groups, site characteristics, planting technique and season, nursery stock, maintenance levels, construction, and socioeconomic characteristics (Cregg and Ellison 2018; Breger et al. 2019; Hilbert et al. 2019; Roman et al. 2022). Tree age and size are often related to mortality, with younger or smaller trees having higher mortality rates, reflecting vulnerabilities during early establishment (Hilbert et al. 2019; Bigelow et al. 2024). With respect to socioeconomic characteristics, there can be higher mortality in neighborhoods with lower socioeconomic status, but results are sometimes contradictory (Hilbert et al. 2019). Tree growth, health, and mortality are impacted by characteristics of planting sites, including soil condition and the extent of nearby impervious surfaces (Sanders and Grabosky 2014; Scharenbroch et al. 2017; Bigelow et al. 2024). For recently planted trees, there is a strong association between mortality outcomes and poor planting technique or inadequate maintenance (Yang and McBride 2003; Breger et al. 2019), and high survival has been found where there are professional expertise and consistent maintenance, including irrigation (Koeser et al. 2014; Roman et al. 2014; Roman et al. 2015). When residents are responsible for irrigation, tree survival can be enhanced through collective watering strategies or signed watering agreements (Mincey and Vogt 2014). More broadly, failure to maintain trees (including irrigation, mulching, and pruning) results in lower tree health and survival, as well as higher costs for replacement plantings and structural challenges as trees mature (Vogt et al. 2015a). The many factors related to urban tree mortality also vary by local ecological, institutional, and social context, making investigations into local planting initiatives critical to understand mortality patterns at scales relevant to program management.

Past mortality studies of planted urban trees have been carried out through field work, often representing a sample of the trees planted through a given program (e.g., Koeser et al. 2014; Vogt et al. 2015b; Elmes et al. 2018; Salisbury et al. 2022). However, recent research has illustrated the feasibility of virtual data collection through street view (SV) imagery, which has the potential to expand the temporal and geographical scale of monitoring. SV imagery, which is generally captured by cameras mounted on vehicles, offers a ground-level panoramic view along streets. In comparison to field work, SV manual data collection can be faster and less expensive, especially since there are no transportation costs (Berland and Roman 2020). However, there are limitations: tree diameter, species identification, health evaluations, and risk assessment can be challenging or impractical for SV interpretation (Berland and Lange 2017; Berland et al. 2019; Meunpong et al. 2019). Therefore, depending on data quality needs, SV imagery is not appropriate for all urban forestry research and management applications. However, SV is especially well-suited to tracking the mortality of recently planted trees for which there are already baseline records with accurate species and location (Berland and Roman 2020). Although the aforementioned studies involved manual inspection of SV imagery, there have been advances in using machine learning to conduct automated inventories (Branson et al. 2018; Laumer et al. 2020; Beery et al. 2022; Choi et al. 2022) and quantify streetscape tree cover or green view index (Li et al. 2015; Seiferling et al. 2017). Given the proliferation of tree planting initiatives in cities around the world, monitoring through SV is a promising avenue to evaluate planting performance.

Another strategy for young tree monitoring that has gained prominence in recent years is volunteer-based data collection, sometimes called citizen science or civic science (Roman et al. 2018a). Across the United States of America (USA), volunteers have been utilized for tree inventories (Boukili et al. 2017; Bancks et al. 2018; Crown et al. 2018; Johnson et al. 2018) and monitoring recently planted trees (Roman et al. 2018b). Such civic science programs engage local communities and are part of larger trends of extensive volunteer engagement in urban forestry in the USA (Hauer et al. 2018). Additionally, for both field and virtual data collection, volunteer data quality is high for simple tasks like determining mortality status and tree counts (Roman et al. 2017; Berland et al. 2019).

In this study, we tracked recently planted street tree survival and mortality from a planting program in Philadelphia, Pennsylvania, USA. We generated data using manual interpretation of SV imagery, complemented by civic science monitoring records, to demonstrate the application of virtual monitoring to local program assessment. We asked the following questions: (1) how do survival rates in this local program compare to typical rates from a literature review (Hilbert et al. 2019); and (2) how is establishment survival of these street trees related to site, neighborhood, and species characteristics? We then situate these findings with a reflection on the potential for large-scale virtual monitoring of tree planting initiatives.

We used virtual monitoring of street tree survival following pilot testing by Berland et al. (2019) and a coproduced transdisciplinary research approach (Campbell et al. 2016; Vogt and Abood 2020) involving research scientists, planting program staff, and students codeveloping the methods and interpreting the findings. Research scientists conceptualized the study and analyzed the data, while an undergraduate student devised the data collection protocol, and a planting manager advised on conceptualization, data collection, and interpretation.

The mortality and survival rates for recently planted street trees found in this study align with typical establishment mortality from the Hilbert et al. (2019) literature review. In that review, across 16 studies reporting annual mortality rates in the first 5 years after planting, the 25th, 50th, and 75th percentiles were 2.81% to 3.76%, 4.40% to 6.48%, and 7.09% to 9.33%, respectively. Our mortality rates for planted street trees in Philadelphia were at the high end of this range for planting to first summer (7.5%). Notably, planting to first summer was less than a full year of time, as trees were planted in November and April, and SV imagery was observed May through October. This finding suggests that mortality rates may be markedly higher in the first few months after planting, compared to later in establishment. The annual mortality rates in our study dropped considerably as time progressed, lowering after the first-to-second summer (5.6%) to the second-to-third (3.7%) and third-to-fourth (3.5%) summer intervals. Indeed, these mortality rates several years after planting are similar to annual mortality observed across all ages and size classes of street trees in Philadelphia (3.7%)(Bigelow et al. 2024) and Cambridge, Massachusetts, USA (3.6%)(Boukili et al. 2017). Therefore, heightened mortality may be most pronounced in the first two years after planting. Accordingly, managers might consider interventions to boost street tree survival during the first two years, when trees are most vulnerable.

In our study, the variables that predicted survival in most models were: (1) level of impervious surface; (2) social vulnerability; and (3) planting season. The level of impervious surface was the only significant predictor of survival for the model concerning planting to first summer, and this variable was included in most other models. Previous research has similarly found that trees surrounded by higher levels of impervious surface were more likely to experience mortality, as well as slower growth, lower health, and heightened heat stress (Sanders and Grabosky 2014; Scharenbroch et al. 2017; Bigelow et al. 2024). Street tree stress in highly paved sites also relates to limited water availability (Schütt et al. 2022). Given the generally high levels of impervious surfaces surrounding the trees in our study (median 95.9%)(Table 2), strategic site alterations could be implemented to promote tree survival. Arborists and landscape designers have developed various strategies to enhance biophysical growing conditions, including structural soils, biochar or compost amendments, and permeable pavers (Morgenroth et al. 2013; Mullaney et al. 2015; Somerville et al. 2020; Scharenbroch et al. 2022). These strategies were not used for the street trees in our study due to lack of funds. Budgets for planting initiatives in the USA are used mostly for purchasing trees, and the number of trees planted is often touted as a goal (Eisenman et al. 2022; Eisenman et al. 2025). In the USA, spending scarce resources on expensive site alterations is a pervasive challenge in urban forest management. Future research that examines the return-on-investment of soil and site alterations, in terms of tree health, growth, survival, and ecosystem services outcomes, can inform decision making about such investments.

Social vulnerability appeared significant in most models (except planting to first summer). There has been extensive recent scholarship on distributional inequities in urban tree canopy cover, showing general trends for more canopy where there are more white and wealthy residents (Gerrish and Watkins 2018; Watkins and Gerrish 2018), albeit with substantial variation in relationships across local socioecological contexts and temporal dynamics (Riley and Gardiner 2020; Locke et al. 2023). Research about canopy cover inequities has contributed to practitioner discourse in urban forest management around environmental justice, such as the American Forests Tree Equity score (Leahy and Serkez 2021). A study from Portland, Oregon, USA, showed distributional inequities in planting patterns (Donovan and Mills 2014) which can reinforce canopy inequities. Our analysis in Philadelphia contributes to this growing body of literature on urban forest equity by showing the connection between social vulnerability and establishment mortality (even after controlling for levels of impervious surface), indicating inequitable early tree performance. Such differences may be related to perspectives and behaviors of residents, maintenance practices, vandalism, and injuries from vehicles (Richardson and Shackleton 2014; Geron et al. 2023). Future research should investigate the potential tree performance benefits (e.g., survival, growth, health) of paid and unpaid tree maintenance programs (particularly in low-income and under-resourced neighborhoods) and systematic evaluation of tree injuries from vandalism and vehicles.

Prior research has demonstrated that programmatic support for tree care, such as maintenance carried out by organized volunteers or paid seasonal staff, produces high establishment survival (Koeser et al. 2014; Mincey and Vogt 2014; Roman et al. 2015; Vogt et al. 2015b). For example, relatively low establishment annual mortality—0.6% annual mortality in the first 6 years after planting—was observed for street trees in East Palo Alto, California, USA, due in part to investments in maintenance through a paid youth internship program alongside arboricultural expertise in species selection and installation (Roman et al. 2015). In another example, young street trees in Holyoke, Massachusetts, USA, had higher survival in comparison to yard trees, largely due to seasonal local staff and foresters carrying out irrigation for the street trees (Breger et al. 2019). However, the role of maintenance was not well explored in the present study. We used available information about Tree Tenders groups to generate a variable for group activity level, and higher activity was only significantly related to survival for one model (planting to fourth summer)(Table 5). We conclude that Tree Tenders group activity level is not likely representing maintenance and may be a proxy for other unmeasured neighborhood effects. As with research on soil and site alterations, there is a need for return-on-investment studies of establishment maintenance, including quantifying the intensity, frequency, duration, and extent of maintenance (Vogt et al. 2015a). Recent urban forest plans and a literature review have stressed that maintenance is deeply tied to equity and community perceptions (Myers et al. 2023; PPR 2023; City of Oakland 2024), yet postplanting tree care is not generally well-funded in the USA (Eisenman et al. 2022), making it imperative to evaluate the direct impacts of maintenance on trees and residents to inform policy and management. Such evaluations would require field observations, resident surveys, and/or administrative records that were outside the scope of our virtual monitoring methods, demonstrating one limitation of virtual data collection.

We found that trees planted in the Fall were more likely to survive compared to trees planted in the Spring. Past research has also found a relationship between planting season and establishment survival, but results have not been consistent (Hilbert et al. 2019), which may relate to regional climate differences, variation in the exact timing of planting within a given season, or potentially species-specific and stock-specific tolerances for planting timing. For the bare root tree stock planted through the PHS Tree Tenders program in Philadelphia, it seems that the timing of Spring plantings is not ideal. Spring plantings in this program have historically occurred the weekend after Arbor Day, which is the last Friday in April. Urban forestry advisors from Pennsylvania State University and Cornell University assert that Spring bare root plantings should occur early in the spring before buds burst (Buckstrup and Bassuk 2003; Jackson 2023). Although the April timing may be acceptable for areas of New England in the USA with long, cold winters, local leading arborists in Philadelphia consider early March to be more suitable timing for bare root plantings in this area (J. Lubar, personal communication, 2025 January 15). However, such timing may be logistically impractical for PHS staff due to their organization running the Philadelphia Flower Show in early March, a major event which is “the nation’s largest and oldest indoor flower show” (Crimmins 2024). In light of these logistical challenges, spring planting could prioritize taxa like Amelanchier spp. and Gleditsia spp. that perform relatively well when planted in spring (Table 4). Notably, PHS plantings in Spring 2025 occurred in early April, and it remains to be seen if this timing change impacts tree survival. The timing of optimal planting may continue to shift due to climate change, making it imperative to consider local climate patterns as well as stock dormancy signs in deciding when to plant.

When drought tolerance was significant (in 3 models), it was opposite of our expected direction; trees with low to medium drought tolerance were more likely to survive. We are uncertain how to explain this finding, which contradicts other street tree research in the same city (Bigelow et al. 2024), and with broader understandings of street tree planting sites as being water limited (Vico et al. 2014), thus making drought tolerant taxa more amenable to street tree settings. It is possible that the particular species in the drought tolerance categories had other traits that were related to survival, or that the drought tolerance categories we used were too coarse. Other measures, such as leaf turgor loss (Sjöman et al. 2018), would enable more direct evaluation of tree physiology and drought tolerance. Additionally, while our analysis was focused on mortality status and derived from virtual monitoring of SV imagery, field data on shoot extension, crown vigor, and soil conditions would enable more finegrain assessment of tree health and growth during establishment (Levinsson et al. 2017; Scharenbroch et al. 2017; Breger et al. 2019).

Tree survival and mortality data is critical to evaluate planting programs and accurately estimate benefits based on empirical data (Ko et al. 2015). Our study demonstrates the value of SV image interpretations to provide mortality data on street trees planted through a local program over multiple years. Having (nearly) all trees from each planting year being tracked makes this study unique compared to other studies that tracked samples of planted trees. Consistent with past research, we found that tree survival was related to level of impervious surface, socioeconomic characteristics, and planting season. These findings have management ramifications pertaining to planting site modifications, equity in planted tree performance, and the timing of planting events. There is potential for scaling up our approach using machine learning approaches to track street tree survival across multiple cities and to determine whether the patterns we observed hold in other biophysical, socioeconomic, and institutional contexts.

The authors reported no conflicts of interest.