Tree Type and Urban Growing Conditions Associated with Street Tree Stress: Lessons from Two US Cities

Site Descriptions

Choosing Sampling Locations

In both cities, we sampled trees within the public rights-of-way based on a stratified random sample of streets. Each city was split into a grid that matches sampling schemes devised for ongoing research in each city (1.5 km × 1.5 km in Chicago, 1 km × 1 km in Durham). Grid cells were selected based on stratification (performed separately for each city) using percent renter occupied housing, median home value, and percent tree canopy cover (USDA Forest Service 2021; US Census Bureau 2022). Tree canopy cover is known to influence and be influenced by other elements of urban conditions (e.g., temperature, inequities)(Locke et al. 2021; Wang et al. 2023), rentership rates can be related to patterns of tree survival (Vogt et al. 2015), and home values are influenced by various elements of the social and physical context, including tree canopy cover in some cases (e.g., Sachs et al. 2023). The values for each cell were the area-weighted mean of the census variables or the mean for canopy cover. These values were then categorized based on their quartile compared to all grid cells within the city (Durham) or region (West Side of Chicago). Using a random number generator, grid cells were then chosen to include at least one grid cell from the upper and lower quartiles of each of the stratification variables and to ensure unique combinations of quartiles. In Durham, where a city tree inventory was available and street trees were less common, selected grids had to include at least 10 city-managed street trees. This selection criteria was to make sure that randomly drawn street segments had a chance to include at least one city-managed street tree that we had permission to sample. However, it means that the sample population is biased towards areas near the downtown (with low overall canopy cover), areas with historical city tree planting (generally wealthy, white), and current priority tree planting areas (City of Durham 2018).

A random 0.8-km street segment, representing a sample of the streets in the grid cell, was created for each cell based on a protocol developed by the US Forest Service (e.g., Bigelow et al. 2024). A fine-scale grid (100 m × 100 m) was overlaid on geographic information system (GIS) layers of public roads to create 100-m road segments. Highways, onramps, and alleys were excluded. One 100-m road segment was randomly selected from within each coarse grid cell (1.5 km² or 1 km²) as a starting point. This 100-m road segment was extended to 0.8 km in Google MyMaps (Google, Mountain View, CA, USA), using a random number generator to decide which turns to make (right, left, or straight ahead). To obtain samples of more than 500 trees in more than 10 neighborhoods, we sampled 13 street segments in Chicago’s West Side and 24 in Durham (Figure S1). For each 0.8-km street segment, we sampled all street trees within city inventories (Durham) or within 3 m of the street (the public right-of-way in Chicago).

Data Collection

Sampling occurred from 2021 to 2023 in Chicago and 2022 to 2023 in Durham between May and August. To account for interannual variation in stress, some trees were resampled in each city. The total number of trees and tree-year observations is: 998 unique trees with 1,424 observations in Chicago and 541 unique trees with 754 observations in Durham (see Table S1 for details on number of trees resampled each year). Trees were sampled by trained research staff, including several co-authors. Training was performed to ensure all members of the team consistently described tree characteristics, site conditions, cues to care, and tree stress and sampling. In line with Healthy Trees, Healthy Cities (HTHC) protocols, sampling was performed in teams of two to augment consistency and allow discussion of borderline conditions. Additionally, for each year and in both locations, 10% to 20% of the trees were resampled as part of quality control protocols to ensure tree stress values were consistent across teams and through time. Errors estimated from this resampling are available in Table S2 and Table S3.

We collected site condition variables, cues to care, and tree characteristics for each tree to use as predictors for our models of tree stress (see below).

Data Cleaning and Model Preparation

To construct models of tree stress, predictor variables were selected from tree characteristics, site conditions, and cues to care. Potential predictors were: site condition (site type, land use, percent impervious surface, powerlines, sidewalks, crown light exposure); tree characteristics (species group, basal area, species urban tolerance); cues to care (mulch following BMPs, pruning following BMPs, landscaping); and baseline variables (street segment, tree ID, year)(Figure 1). Basal area values were log-transformed for analyses due to their otherwise skewed distribution. In addition, due to limited sample sizes, some categories were simplified. For land use, categories were grouped while ensuring that categories commonly used in land-use-based analyses (residential, commercial, and parks or natural areas) remained. For example, in Durham where multi-family residential was not common, multi-family and single-family residential was grouped together into “residential”. This left 6 classes in Chicago and 5 classes in Durham. Site type was grouped by the extent of the rooting space, with sidewalk cutouts/other hardscape (the most restrictive), then sidewalk planting strips (the most common, moderately restrictive), then other categories that had more rooting space (see Table S4 for original and simplified categories). For species group, commonly sampled tree species (having at least 5% of the total observations, 71 observations in Chicago, 41 observations in Durham) were analyzed at the species level, and for the remaining individuals genera with at least 5% of the total observations were analyzed at the genus-level. All other individuals were grouped based on mature tree size (small ≤ 10.5 m, medium between 10.5 m and 15.25 m, and large > 15.25 m). A list of species and their grouping for analysis and genera is available in Table S5 (Chicago) and Table S6 (Durham).

• Download figure
• Open in new tab
• Download PowerPoint

Figure 1.

Predictors used to model tree stress in Chicago, IL, and Durham, NC. Site condition variables describe the underlying stressors present in a location, cues to care describe potential mediating variables that can reduce the impact of the stressors, and tree characteristics describe a trees’ tolerance for the stressors present. The baseline variables are included to account for the sampling design. Defoliation, discoloration, and dieback were all modeled together in one multivariate model while overall crown stress was modeled separately.

To allow for cross comparisons between Chicago and Durham, we included only predictors that varied in both cities and were not highly correlated (full list of excluded variables in Table S9). For example, whether or not a tree was next to a sidewalk was not included as a predictor because more than 90% of trees in Chicago were next to a sidewalk, leaving too few not near sidewalks for parameter estimation. We set up the models to initially include variables that are known to have impacts on tree survival and stress susceptibility, namely tree size (basal area), tree type (species group), land use, and site type (as a proxy for rooting space)(Hilbert et al. 2019; Roman et al. 2020; Bigelow et al. 2024). We then added other measured variables, excluding those that were highly correlated (variable inflation factor > 5)(see Table S9)(Shrestha 2020). Year and segment were used to account for differences in sampling across years and potential spatial covariance. Because some trees were sampled multiple times and those samples were not independent, tree ID number was included as a random effect. Generalized joint attribute modeling (GJAM) handles random effects such that trees that are not resampled will be grouped into a “rare groups” category because otherwise the lack of multiple observations makes parameter estimation impossible. The sampling frequencies of included variables can be found in Table S7.

The models for tree stress included some interactions between predictor variables (tree level, site condition, and cues to care variables), with multiplicative parameterization. The first interaction (basal area and species group) accounts for species and genus-level differences in the relationship between tree basal area and senescence and likelihood of conflicting with built infrastructure (Randrup et al. 2001; Qiu et al. 2021). The second interaction (urban tolerance and site type) accounts for an individual species’ ability to survive conditions (pollution, salt, drought, small space) that are most relevant for site types with limited rooting space (Dirr 2016; Bigelow et al. 2024).

Data Analysis and Modeling

Due to various inter-city differences, we modeled each city separately. Although the same variables were recorded in both cities, predictors do not have the same implications for trees in the two locations. For example, though both cities have commercial zones, Chicago has commercial areas embedded within residential areas, while in Durham they are primarily downtown or in strip malls. Because of these and other differences (climate, population density, species composition), we expected best-fitting models to differ between cities.

For each city there were two models: one with defoliation, discoloration, and dieback as joint response variables, and one with overall crown stress as a sole response variable. Defoliation, discoloration, and dieback were modeled using a multivariate ordinal model: a Generalized Joint Attribute Model (GJAM) (Clark et al. 2017). GJAM models response variables jointly, recognizing their nonindependence, but still provides an estimate of model fit and predictor parameter estimates for each individual response variable (Clark et al. 2017). The responses (defoliation, discoloration, and dieback) were modeled as integer values corresponding to the ordered categories described above (see Table S8). In this case, GJAM follows Lawrence et al. (2008): a multivariate probit model with the addition of parameter expansion to sample the posterior distribution and allow for a Bayesian approach. The second model has only overall crown stress as a response in a univariate ordinal model, also modeled off of Lawrence et al. (2008). This approach allows us to compare the variation explained for a combined metric (overall crown stress) to the 3 individual but overlapping ones (discoloration, defoliation, dieback). For all models, priors for predictor variables were set as gaussian centered around zero. To test model configurations, variables and/or interactions were iteratively removed based on sensitivity, though year, tree ID, and (log-transformed) basal area were always included in potential models due to their relevance for sampling design and basic tree biology (see above). The best model was then selected based on Deviance Information Criterion (DIC), commonly used for Bayesian model comparison (Spiegelhalter et al. 2002). For all evaluated models, trace plots were checked to ensure that models converged (e.g., S2, S3). All data cleaning and modeling were performed using R (R Core Team 2023) with the dplyr (Wickham et al. 2026) and GJAM packages (Clark et al. 2017). Other functions were written in C++. A full list of evaluated models and their DIC values can be found in Table S10 to Table S13, with the predictors tested summarized in Table S14.

Discoloration, Defoliation, and Dieback

Site Condition Variables

Both site type and land use were in the best-fitting models for defoliation, discoloration, and dieback in both cities, but the effects diverge (Table 3, Figure 2). In Chicago, trees in sidewalk planting strips were associated with low dieback compared to trees in sidewalk cutouts. In Durham, planting strips were associated with high defoliation and dieback compared to trees in the most restrictive rooting spaces (sidewalk cutouts, other hardscape). For land use, commercial zones were associated with higher discoloration and dieback than single-family residential zones in Chicago, but that was not the case in Durham (Figure 2).

• Download figure
• Open in new tab
• Download PowerPoint

Figure 2.

Relationships between site conditions (site type, powerlines, land use) and tree stress responses (defoliation, discoloration, and dieback) in the multivariate model for Chicago and Durham. Box colors signify the type of association: negative (light blue), positive (red), or with confidenceintervals (CIs) that cros zer (hite). The abbreviations for land use are: commercial and mixed-use (COMM), managed park (MP), instituional (INST, multi-family residential (MFR), and natural area/vacant lot (NATV). The abbreviations or site types are sidewalk planting strip (SP), median(M), and other maintained area (OM). The baseline category for site type is sidewalk cutout in Chicago and sidewalk cutout and other hardscape in Durhm. The baseline category for land use is single-uit residential in Chicago and residential for Durham. Full parameter estimates with standard errors and CIs can be found in Tables S15 and Table S17.

Tree Characteristics

The associations between tree characteristics and tree stress diverge between the two cities, but in both cases tree stress is associated with species group (Figure 3). Urban tolerance was also in the best models in both cities along with the interaction between urban tolerance and site type, though the confidence intervals for the interaction crossed zero for all three metrics in both cities (Figure 3). In Chicago, basal area was associated with lower discoloration and higher dieback, and in both cities the interaction between basal area and species group had different associations depending on the stress metric and species group of interest (Figure 3). In both cities, the baseline species group is the most commonly sampled species (Gleditsia triacanthos in Chicago and Lagerstroemia cultivars in Durham), both of which are urban tolerant species. Compared to this baseline species, in Chicago, Fraxinus spp. (struggling with emerald ash borer) were associated with high dieback, while in Durham, Quercus phellos was associated with high dieback, and in both cases the interaction with basal area indicates that the association with high stress increases as the trees get larger (Figure 3). However, many of the other species or genera included were not associated with higher discoloration or dieback compared to the baseline urban tolerant species (Figure 3).

• Download figure
• Open in new tab
• Download PowerPoint

Figure 3.

Relationships between tree characteristics and stress respons (difoliation, discoloration, and dieback) in the multivariate model for Chicago and Durham. Box colors signify the types of associatio: negative (light blue), positive (red), or with cofidence intervals (CIs) that cross zero (white). Numbers correspond to species grups, labeled on the left (Chicago and right (Durham) sides of the figure. The diamonds wih the letters “int” in them represet the interction between basal area and the species group above the diamond. The abbreviation “spp.” stands for “species” and the abbreviations “tol.” here stand s for “tolerance”. The baseline categories for species group are the most common, urban tolerant species in each city: G. triacanthos (honeylocust) in Chicago and Lagerstroemia spp. (crape myrtle) in Durham. The list of species in each species group with sample sizes are available in Table S4 and Table S5. Full parameter estimates with standard errors and CIs can be found in Table S15 and Table S17.

Overall Crown Stress

Tree Characteristics

Overall crown stress highlighted the importance of basal area and species group, especially in Durham. In both cities, higher basal area was associated with lower overall crown stress, and in Chicago urban tolerance was also associated with lower overall crown stress (Figure 4). In Durham, the parameter estimates for species group indicated that all groups had higher overall crown stress than the baseline urban tolerant group (Lagerstroemia spp.)(Table S16). The interaction between basal area and species group was not in the best-fitting model for Durham, but in Chicago, the interaction indicated that larger basal area may be associated with higher stress for medium and large stature trees, among others (Figure 4).

• Download figure
• Open in new tab
• Download PowerPoint

Figure 4.

Relationships between tree characteristics (genus, basal area, and urban tolerant) and overall crown stress in the univariate model for Chicago and Durham. Box colors signify the direction of the relationship: negative (light blue), positive (red), or with confidence intervals (CIs) that cross zero (white). Numbers correspond to species groups, labeled on the left (Chicago) and right (Durham) sides of the figure. The diamonds with the letters “int” in them represent the interaction between basal area and the species group above the diamond. The abreviation “spp.” stands for species. The baseline caegories for species group are the most common, urban tolerant speies in each city: G. triacanthos (honeylocust) in Chicago and Lagerstroemia spp. (crap mytle) in Durham. The list of specis in each species groupwith sample izesare available in Tables S5 and S6. Full parameter estimates withstandard errors and CIs can be found in Table S16 and Table S18.

Defoliation

Defoliation is primarily driven by herbivory or pathogens, and many pathogens and herbivores have restricted hosts and diet preferences (Prins and Verkaar 1992; Fraser 1997). Therefore, defoliation can reflect species or genus level differences in herbivory or pathogen vulnerability. Though we did not directly sample for herbivores or pathogens, in some cases their influence was evident. For example, in Chicago, Celtis occidentalis had high defoliation (Figure 3). This came almost exclusively from Psyllid galls, which were not found in the other species groups (Yang and Mitter 1994).

The relationship between site conditions and defoliation in Durham and Chicago don’t point to clear overlaps and, if anything, highlight how different the conditions in the two cities are for defoliation (Figure 2), though direct sampling of herbivore or pest densities would help clarify how/if they vary across land use and site types in both cities.

Discoloration

The association between tree characteristics and discoloration, specifically species group and basal area, were consistent with planting stress and stress from restricted rooting spaces. Larger trees were associated with lower discoloration either directly (Chicago) or indirectly (Durham)(Figure 3), consistent with smaller trees being more recently planted and suffering from transplant shock (Watson 2005; Czaja et al. 2020) and/or having more limited root systems that may be less able to access deep water. In Durham, though basal area was not individually associated with lower discoloration, its interaction with species group indicated that it might be for some tree types with deep roots (e.g., Q. phellos and other oaks, Prunus spp.). That said, root depth is also heavily driven by soil conditions and not just tree species, and other species groups can also grow deep roots in urban settings (e.g., U. americana)(Day et al. 2010; Czaja et al. 2020). Also, the sampled Prunus trees were all relatively small (average DBH 4.6 cm compared to 29 cm for the baseline Lagerstroemia and 55 cm for Q. phellos), so the effect may be a sampling artifact. Although basal area itself was associated with lower discoloration in Chicago, the interaction between basal area and some species groups (e.g., Tilia spp., other large trees) indicates that in some cases high basal area may have less of a beneficial effect, especially for trees with large mature height (Figure 3). This moderated benefit in Chicago could be due to the restricted rooting spaces (all the trees were either in sidewalk cutouts or planting strips, compared to 53% in Durham) (Table S7) that provide less space for extensive root systems (Day et al. 2010; Czaja et al. 2020).

The relationship between discoloration and site type and land use diverged in Chicago and Durham, but heavy car traffic might have played a role in both. In Chicago, commercial areas had high discoloration compared to single-family residential areas (Figure 2). Commercial districts in Chicago tended to have sidewalk cutouts and heavily trafficked roads, which are more likely to have restricted rooting spaces and elevated levels of PM2.5 and other pollutants from cars that can influence discoloration (Dos Santos-Juusela et al. 2013; Mu and Chen 2021; Talebzadeh and Valeo 2022). Although commercial areas were not similarly associated with discoloration in Durham, trees in street medians had high discoloration (Figure 2), which is consistent with stress from being near heavy car traffic (Dos Santos-Juusela et al. 2013; Czaja et al. 2020).

Dieback

Dieback can be a sign of longer-term or intense tree stress (Camarero 2021). Small trees experiencing significant dieback are less likely to have sufficient leaf area to compensate for the loss and so may die, be replaced, and thus be removed from the sample (as seen for five trees with DBH < 6 cm on one segment in Chicago). These removals could explain the association with larger basal area and higher dieback in Chicago (Figure 3). Unsurprisingly, some species were also associated with high dieback relative to the most common, urban tolerant species in each city (Figure 3). The high dieback in Fraxinus trees in Chicago is likely related to emerald ash borer, though the city has provided some treatments (Bureau of Forestry 2014) with some sampled trees having no signs of treatment and others with tags indicating treatments as recent as 2020. For Q. phellos in Durham, despite some urban tolerance, they tend to have higher dieback in street tree conditions compared to lawn conditions (Salisbury and Grabosky 2020). However, the result could also be a sampling artifact. Of the 10 largest trees (all with DBH > 94 cm), 7 of them were Q. phellos. These large Q. phellos may have high dieback because they are starting to senesce or display retrenchment.

The two cities had diverging relationships between dieback and sidewalk planting strips. In Chicago, the low dieback for trees in sidewalk planting strips (Figure 2) is consistent with the planting strips having more rooting space than the small sidewalk cutouts (Bigelow et al. 2024). However, in Durham, the opposite is true, and sidewalk planting strips are associated with more dieback relative to sidewalk cutouts. Though sidewalk cutouts were in a variety of areas in Chicago, in Durham 82% of the trees in the most restrictive conditions (sidewalk cutouts and other hardscape) were in just 2 segments, both of which were in the downtown area. These areas have relatively high temperatures and high car traffic, which can be stressful for trees, but are also areas where nearby businesses may provide extra care for the trees that our sampling did not account for (like irrigation) that might offset these stresses. There may also be differences in tree pit design in Durham compared to Chicago that might make the sidewalk cutouts less stressful (e.g., use of pavement suspension systems, porous pavement). That said, dieback was relatively poorly explained by the models (pseudo-R² = 0.225 in Chicago, pseudo-R² = 247 in Durham), so other unexamined drivers are likely important (e.g., pruning history, soil conditions).

Relationship Among Defoliation, Discoloration, and Dieback

Residual association between dieback and discoloration in Chicago (Table S19) was consistent with dieback from prolonged water stress. Dieback could indicate the impacts of drought on top of chronic stress (Camarero 2021), and Chicago experienced dry spells in both 2021 and 2023 (Ford 2021, 2024a). The same was not true for Durham, which did not experience droughts in our sampling period (Davis and Dello 2023, 2024). The association between defoliation and discoloration in Chicago (Table S19) could be related to pathogens that cause both (e.g., psyllid galls) or could be because existing stress (like from drought in Chicago) can make trees more vulnerable to new stressors (e.g., drought stressed trees being more susceptible to herbivory)(Gely et al. 2020). Future research tracing tree defoliation, discoloration, and dieback over time and through droughts could clarify the relationship between these stress metrics, though they likely vary by species.

Overall Crown Stress

Overall crown stress, though poorly explained in Chicago (pseudo-R² = 0.198), did highlight the value of species selection. The association between urban tolerance and lower overall crown stress in Chicago (Figure 4) is consistent with long-standing arborist practices, choosing especially urban tolerant trees for the harshest conditions in cities (Minckler 1941). In Durham, though urban tolerance was not individually associated with low overall crown stress, all species groups were associated with high overall crown stress relative to the baseline urban tolerant species group (Lagerstroemia spp. or crape myrtle). Lagerstroemia have been and continue to be bred for tolerance to a variety of urban stressors and site types (Orlóci et al. 2025), thus it is not surprising that the Lagerstroemia had relatively low overall crown stress.

The site condition variables in the best-fitting models highlighted a key divergence between site types in the two cities that was also present with dieback. Again, sidewalk planting strips were associated with lower stress than the restrictive sidewalk cutouts in Chicago (consistent with the lower stress of more rooting space)(Bigelow et al. 2024), but the opposite was true in Durham (Table S18). The reason why sidewalk planting strips were relatively stressful in Durham is not immediately clear from the data collected here, and, as discussed above, could be related to tree care, tree pit design, or other unmeasured variables.

Cues to Care

Despite the positive impacts of tree care on tree survival and health (Roman et al. 2015; Esperon-Rodriguez et al. 2025), we did not see strong associations between cues to care and reduced tree stress (though most of the parameter estimates are negative, their confidence intervals cross zero)(Table S15 to Table S18). The impacts of pruning are hard to detect over short timescales and pruning needs vary by species (Clark and Matheny 2010; Poulton Kamakura et al. 2025). The benefits of mulch can be context dependent (Esperon-Rodriguez et al. 2025; Poulton Kamakura et al. 2025), and without testing the composition of the mulches and the underlying soil conditions, we may not be able to detect when or if mulch has positive benefits for street trees. Urban soils are notoriously heterogeneous, and soil structure and nutrient loads can vary with current site types along with historical land uses and broader sociopolitical patterns not examined here (Pickett and Cadenasso 2009; Czaja et al. 2020). Additionally, the function of cues to care partly depends on how they relate to the cultural values and expectations of residents (Li and Nassauer 2020), and we did not interview residents to see how they viewed the cues to care we catalogued. There are also a variety of sociopolitical factors that can influence the effectiveness of stewardship practices, including whether there is a consistent group providing irrigation or a collective watering strategy (Breger et al. 2019) that cannot be assessed without robust surveys or interviews. Further work could investigate whether landscaping, mulch, or pruning impact how nearby residents, businesses, and others interact with trees as well as how tree care is provided.

Limitations

This study focused on a subset of potential variables influencing tree stress but was limited in the number of trees sampled, the areas covered, and the predictor variables characterized. Though a sample size of 998 individual trees for Chicago and 541 individual trees in Durham does allow us to understand some interactions between predictor variables, the sample sizes were too small to investigate other interactions (between, for example, forms of tree care). We also only had 3 years of sampling in Chicago and 2 years of sampling in Durham. On top of that, not all trees were resampled every year, which limited our ability to understand interannual variation. The spatial extent was also limited such that, in Durham, only areas that had some city managed street trees could be sampled, excluding some of the areas with high canopy cover not prioritized for street trees and some areas without sidewalks. In Chicago, we only focused on the West Side of the city, which includes neighborhoods that vary in sociodemographic characteristics but is not representative of the rest of the city and especially excludes the wealthiest parts of the city, the densest parts of the downtown, and some areas heavily impacted by industrial pollution. Thus, the results in Durham are more relevant to areas with city-managed street trees, and results in Chicago are more relevant to the West Side and might differ from those seen across the whole city.

There were other potentially influential predictor variables that we did not sample, including and especially soil conditions and microclimatic conditions. As discussed in Poulton Kamakura et al. (2026), we did initially include soil electrical conductivity and soil compaction. However, pilot soil electrical conductivity measurements varied heavily with rain conditions, and sampling could not easily be adjusted to avoid rainy days (Figure S5). For soil compaction, values similarly depended on whether technicians moved mulch to sample and whether it was raining (Figure S6). However, soil conditions are known to have substantial impacts on tree stress, especially as it relates to available nutrients, soil moisture, and available soil volume (Day et al. 2010; Czaja et al. 2020). Microclimatic conditions can also influence tree stress, especially as it relates to water needs and physical stresses from wind (Czaja et al. 2020). We did sample crown light exposure as one form of microclimatic condition that is important for tree growth and health; however, other sampling of tree microclimates proved challenging because we only sampled most trees once a season, and microclimates vary with synoptic weather patterns as well as diurnally (Oke et al. 2017). To facilitate comparison across a given city, sampling temperature or wind, for example, would have had to occur at the same time of day and, ideally, under similar synoptic weather conditions. We did not have the personnel for that kind of sampling. Irrigation is also important for tree survival especially soon after planting (Breger et al. 2019). We at one point included irrigation infrastructure as part of the sampling scheme. However, pilot sampling in Chicago made clear that often residents watered the street trees with hoses brought out from their houses that would be difficult to detect unless we actively observed the watering, interviewed residents, or arrived soon after watering. Some of the included predictor variables (e.g., percent impervious surface in the root zone, site type, cues to care) can be proxies for some of the variables mentioned above (e.g., impervious surface is related to temperature, site type to rooting space), but more direct sampling would clarify the importance of soil conditions and microclimate.