Introduction

Trees growing in metropolitan region landscapes provide many benefits to the environment and to human societies, including factors such as mitigation of urban heat island effects, stormwater interception, wildlife habitat, and carbon (C) sequestration (Nowak and Dwyer 2007). The intentional urban forest (human planted and managed trees along streets, in landscape plantings, and in urban parks) is only one component of a larger landscape that also includes remnant and spontaneous trees growing in natural areas and interstitial stands (Zipperer et al. 1997). The environment experienced by trees in metropolitan landscapes spans a wide range of land uses, site-level conditions, and regional-scale environmental gradients. The composition and structure of urban forests also vary considerably across these gradients (Iverson and Cook 2000; Zipperer 2002; Nowak et al. 2008).

The longevity, growth, and resiliency of the trees that make up the urban forest is a very important factor in the provisioning of benefits from these trees and the durability and resilience of those benefits (Nowak et al. 2004; Roman and Scatena 2011). However, the wide variation in growing conditions, forest structure, and management strategies across the urban forest landscape likely also has a strong impact on the survival and growth of urban trees and dynamics of urban forests (Iakovoglou et al. 2001; Nowak et al. 2004; Vogt et al. 2015). For example, at a site scale, the volume, structure, and composition of soil in the rooting zone can have a considerable impact on tree growth (Whitlow and Bassuk 1987; Lindsey and Bassuk 1992). Environmental factors that are evident at a broader scale can also be important; for instance the urban heat island effect can influence both tree growth and resilience to environmental perturbations such as drought (Cregg and Dix 2001). Variables that are linked to specific land-uses, such as pollution associated with transportation corridors or fertilization in agricultural areas, can also affect growth (Benoit et al. 1982; Muir and McCune 1988).

Variation in the growth of urban trees is generally either assessed in relation to specific characteristics of forest stands (e.g., species composition, stand structure) and the sites on which they occur (e.g., soils, landscape position), or based on broader land-use categories and urban-rural gradients, which seek to synthesize these factors (Kjelgren and Clark 1992; Iakovoglou et al. 2001). Studies that assess generalized urban to rural gradients can allow for a regional perspective, but often overlook much of the variation that occurs within the more urbanized portion of the gradient (McDonnell et al. 1997; Iakovoglou et al. 2001; Iakovoglou et al. 2002). A focus purely on land-use categories can synthesize environmental and socio-economic factors, but can also miss variation that occurs within classes, especially when arrayed across large urban regions (Cadenasso et al. 2007). Reductive, site-level approaches focused on measured environmental conditions could be superior in some respects, offering a more mechanistic understanding of the factors that impact growth variability (Clark and Kjelgren 1990; Kjelgren and Clark 1992; Pataki et al. 2011; Litvak et al. 2012). For example, a number of previous studies have focused on the effect of characteristics of specific planting sites on urban tree growth (Pan and Bassuk 1985; Whitlow and Bassuk 1987; Kjelgren and Clark 1992; Pretzsch et al. 2015). However, such an approach requires identification and accurate measurement of the primary factors associated with variation in growth, which could vary greatly among locations across a large urban region. In this study we have attempted to compare and combine these approaches in evaluating factors that affect urban tree and forest growth.

Healthy and resilient urban forests could play a very important role in the adaptation of urban landscapes to global climate change as well as aiding in mitigation of climate change effects (Gill et al. 2007; Kirshen et al. 2008; Ordóñez and Duinker 2014; Brandt et al. 2016). The effects that global climate change could have on urban forests are uncertain and could be highly variable within and among urban regions (McPherson et al. 1997; Ordóñez and Duinker 2015). Specific local- and regional-scale shifts in temperature and precipitation regimes associated with global climate change are difficult to predict accurately, especially in highly urbanized regions. However, the frequency and intensity of extreme climatic events (i.e., droughts, floods, storms, and heat waves) is projected to increase, and such events could be especially influential in urban ecosystems (Meehl et al. 2007). Urban forests could help alleviate the effects of these events, through functions such as shading of buildings and reduction of stormwater flows (McPherson et al. 1997), but the capacity of urban forests to diminish the deleterious effects of climate change will depend on maintaining healthy tree populations and canopy cover (Ko et al. 2015).

Therefore, urban tree growth and survival as well as response to extreme climatic events will all be very important for adaptation of urban regions to climate change. However, the response of the urban forest to extreme climatic events may differ greatly with species composition and structure of the forest, across land-use types, and in relation to specific environmental factors and regional gradients (Cregg and Dix 2001; Pataki et al. 2011). In order to understand how urban forests will respond to future climatic changes, planners and managers need knowledge of how the growth and resilience of the trees that make up these forests will be affected by extreme climatic conditions. These actors also need to better understand how such responses vary among species and across the urban landscape to inform future adaptation-focused planting strategies (Brandt et al. 2016). To address this goal we pursued the following specific research objectives: 1) Characterize variation in individual tree growth and forest stand productivity among species and land-use types across a large urban metropolitan region; 2) Evaluate the response of urban tree growth and forest productivity to extreme drought across species and land-use types; and 3) Assess potential environmental drivers of urban tree growth, urban forest productivity, and the response of these ecosystem functions to extreme drought.

Methods

Study area

The present study capitalized on a network of plots established by the Morton Arboretum, USDA Forest Service, and City of Chicago for the Chicago Regional Tree Census (hereafter “Urban Tree Census”; Nowak et al. 2013). The overall Urban Tree Census data set consists of 2076 0.04 ha circular plots that span the seven-county Chicago metropolitan region as defined by the Chicago Metropolitan Agency for Planning (http://www.cmap.illinois.gov/). An initial network of 745 plots was established in the City of Chicago in 2007, while the remaining 1331 plots were established in 2010 in a stratified-random fashion across suburban Cook County and the other 6 counties (DuPage, Kane, Kendall, Lake, McHenry, and Will). To examine urban forest tree growth and response to climate variation across the Chicago region landscape, we resampled a subset of the original Urban Tree Census plots. We limited our subsample to plots that had trees present for increment core collection (at least 3 stems ≥10 cm in diameter at breast height or “DBH”). We then stratified the plots that met this criteria by common urban land-use types (as defined in original Urban Tree Census field surveys; Nowak et al. 2013) as follows: (1) Agriculture (“AG”; current farm land, riparian and ditch buffer zones in farm fields), (2) Commercial-industrial-transportation (“CIT”; commercial land, parking lots, current and vacant industrial sites, major transportation corridors, utility right-of-ways), (3) Forest (“FOR”; natural and restored forest, wetland, and savanna sites) (4) Park (“PARK”; parks, golf courses, schools and institutions with grass or landscaped understory), and (5) Residential (“RES”; single and multi-family residences and neighborhoods). We randomly selected 20 plots from each of these land-use categories for resampling, but only 10 agricultural plots in the overall data set met the criteria of having 3 trees ≥10 cm DBH, thus our final data set included 90 total plots (Table 1).

Table 1 Characteristics of sample plots and increment core data set
Full size table

Field data collection and processing

In 2013, we remeasured all trees on each plot (DBH and maximum canopy height) and extracted two increment cores (at 1.37 m height and perpendicular to each other on the tree bole) from all trees with DBH ≥10 cm. A total of 609 trees were sampled and analyzed from the 90 resample plots (Table 1). Increment cores were dried, mounted, and sanded with progressively finer sand paper. Following visual cross-dating (Fritts 1976; Stokes and Smiley 1996), annual ring widths were measured to the nearest 0.001 mm on a stage micrometer (Velmex Inc., Bloomfield, NY, USA) with Measure J2X software (Voortech 2005). Ring width series were then statistically crossdated and validated with the dplR package in R (Bunn et al. 2016). Chronologies were standardized and detrended using standard dendrochronological methods (Cook and Kairiukstis 1990; Bunn et al. 2016). We used a rigid modified negative exponential curve for standardization due to the short time series and to limit the errors produced by low-frequency functions that may appear near the end of a chronology since the study area experienced a widespread severe drought in 2012 (García-Suárez et al. 2009).

For each tree, growth was converted to basal area increment (BAI) based on annual ring width and DBH at time of sampling (Bunn et al. 2016). Sample depth for the increment core record decreased rapidly prior to 1980, with approximately 50% of series persisting into the early 1980s (Supplementary Material, Fig. A1). Chronologies were truncated for the most recent 30-year period (1983–2012) to ensure sufficient sample depth and to include the most recent droughts in response analysis. Individual tree BAI series were then averaged to total and annual mean land-use BAI for the 1983–2012 period.

In addition to tree characteristics, we also collected data in the field on ground cover (% cover of classes at the plot-level) and surface soils (10 soil cores to 20 cm depth composited at the plot level and collected using a 2.5 cm diameter Oakfield soil probe; Oakfield Apparatus, Inc., Fond du Lac, WI, USA). Surface soil characteristics included as predictors included bulk density, pH, electrical conductivity, dissolved organic carbon, percent sand, calcium, magnesium, phosphorous, sodium, and cation exchange capacity (full details of soil analysis included in Supplementary Material, Appendix B). Ground cover categories included percentage cover of bare soil, litter/duff, herbaceous vegetation, rock, water, maintained grass, concrete, gravel, asphalt, and building structure. Continuous ground cover and soil variables were combined into orthogonal variables illustrating dominant gradients using a principal components analysis (PCA - using veganR package; Oksanen et al. 2016). For the groundcover PCA the first two components explained 80% of the total variance (PC1 = 51.6%, PC2 = 28.8%) and the first component was strongly related to maintained grass cover vs. natural litter cover (Supplementary Material, Fig. A3). For the soil PCA the first two components explained 61% of the total variance (PC1 = 45.2%, PC2 = 16.0%). The first component was most strongly related to cation exchange capacity, pH, and Ca concentration while the second component was strongly related to soil texture (i.e., Percent Sand; Supplementary Material, Fig. A3).

DBH remeasurements were used to calculate total plot-level basal area (BA) and relative basal area by species (Table 1). From these data we quantified plot-level tree species diversity using Simpson’s Diversity Index (McCune and Grace 2002). Compositional patterns were extracted using nonmetric multidimensional scaling (NMS) ordination on a matrix of species BA by plot. NMS was performed using PC-ORD v.5.31 (McCune and Mefford 2006) with the “slow-and-thorough” auto-pilot setting, using 250 runs of real data and 250 Monte Carlo randomizations to assess the robustness of the solution. The NMS ordination resulted in a three-dimensional solution which was significant based on Monte Carle test (stress = 28.6, p = 0.039). The 3D solution explained 41.4% of the variation in the original data matrix. Each of the three axes explained a similar component of the variation: Axis 1 = 10.9%, Axis 2 = 13.2%, and Axis 3 = 17.4%. Axes from the resulting ordination solution were used as predictors representing the major compositional gradients in the overall data set (Supplementary Material, Fig. A2).

Data analysis

To address Obj. 1, variation in individual tree growth (defined by annual wood production as BAI) was compared among land-use classes and among common genera with analysis of variance (ANOVA; all tests considered significant based on p < 0.05 criteria) with Tukey-Kramer post-hoc test for comparison of means using Sigmaplot v. 13 (SYSTAT 2014). In addition, plot-level forest productivity was compared among land-use classes also using ANOVA.

To address Obj. 2, we assessed the response of individual tree growth and plot-level forest productivity to specific known drought years. Regional Palmer Drought Severity Index (PDSI) data (NOAA 2016) were used to verify widely recognized regional droughts that occurred in 1988, 2005, and 2012 all of which were characterized as “extreme” droughts at a regional scale (USDA 2016). Growth response to drought was evaluated using drought resistance and resilience indices (D'Amato et al. 2011; Fahey et al. 2013). Resistance was defined as the ability of a tree to sustain growth under drought conditions, and was calculated as follows:

$$ \mathrm{Resistance}\;\mathrm{Index}={BAI}_{\mathrm{droughtyear}}/\mathrm{mean}\ {BAI}_{5-\mathrm{yr}\ pre-\mathrm{drought}} $$

Resilience was defined as the ability of the tree or forest to return to pre-drought growth levels in the period following drought and was defined as:

$$ \mathrm{Resilience}\;\mathrm{Index}={\mathrm{meanBAI}}_{5\mathrm{yr}\ \mathrm{post}-\mathrm{drought}}/{\mathrm{meanBAI}}_{5\mathrm{yr}\ pre-\mathrm{drought}} $$

In both indices a value of 1 indicates no growth response, values <1 indicate a negative growth response, and values >1 indicate a positive response. Drought resistance and resilience of individual trees were compared among land-use classes and common tree genera (Table 1) using ANOVA. Drought resistance and resilience of plot-level forest productivity were compared among land-use types using ANOVA.

To identify factors driving variability in individual tree growth, stand productivity, and drought sensitivity across the urban forest landscape (Obj. 3), productivity and drought response data were related to community characteristics, physical site conditions, and position along urban-rural and socio-economic gradients. Multiple linear regression in an information-theoretic framework (Burnham and Anderson 2002) was used to test a set of predictors on BAI, BAI per unit BA (“Rel BA”), coefficient of variation of BAI over time (“CV BAI”), resistance to the 2012 drought, resistance to the 2005 drought, and resilience following the 2005 drought. Socio-ecological characteristics of urban landscapes (e.g., canopy height, impervious surface area, median income) can be highly variable over time; to minimize the effects of this variability on multiple regression analysis we utilized an abridged 15-year time period (1998–2012), which removed the 1988 drought from consideration, but retained the other two droughts.

Potential drivers of growth, drought resistance, and drought resilience included: 1) characteristics of the forest community – plot BA, canopy height, species diversity (Simpson’s Diversity Index), species composition (NMS axes), and percent BA represented by native species; 2) environmental factors – ground cover (PCA axes), soil characteristics (PCA axes); 3) position on the geographical urban-rural gradient as distance to city center (Navy Pier), plot-level impervious surface cover (field based assessment), and impervious surface cover within a 100-m buffer (from National Land Cover Database; Xian et al. 2011); and 4) socio-economic status – median income in the census block in which the plot occurs (US-Census-Bureau 2010).

For each response variable, a suite of models was evaluated using the corrected Akaike Information Criterion (AICc). Model merit was ranked by ΔAICc, the difference between the AICc value for the selected model and the lowest AICc value of all models, including the null. This assessment compares the strength of each model. As ΔAICc decreases, the probability that the selected model is the best-fit increases. Thus we accept and interpret only models with ΔAICc < 2 (Burnham and Anderson 2002). We also refrained from interpreting models that contained uninformative parameters – where one of the other models with ΔAICc < 2 included all but one of the same predictors (Arnold 2010).

Results

The increment core data set included 609 trees across a large number of species (45 total recognized), approximately half of which came from the five most common genera: Prunus, Fraxinus, Quercus, Acer, and Ulmus (Table 1). The trees sampled represented a large range in diameters, ranging from the minimum of 10.0 cm to 118.9 cm and the mean diameter did not differ significantly (ANOVA; F4,81 = 2.075, p = 0.92) among land-use categories (Table 1). The distribution of species and genera across land uses varied (Table 1), but there was not a strong separation of species composition across land-use categories based on ordination (Supplementary Material, Fig. A2).

Objective 1- growth and productivity

Annual mean basal area increment (BAI) varied over time (Fig. 1) and the pattern of annual BAI variation was highly correlated among land-use categories (all between 0.80 and 0.96; Supplementary Material, Table A1). Mean tree BAI over the 30-year period differed significantly among land-use classes based on ANOVA (F4, 604 = 9.81, p < 0.001; Fig. 2a). Individual trees had higher average BAI in PARK, RES, and CIT land-use categories, relative to FOR and AG land-uses (Fig. 2a). Similar trends can be seen in mean annual growth across the 30-year study period (Fig 1). Mean tree-level BAI also varied greatly among major genera, with Acer, Quercus, and Juglans exhibiting the highest average BAI (Fig. 2b).

Fig. 1
figure 1

Mean (a) and total (b) basal area increment (BAI) over time by land-use

Full size image
Fig. 2
figure 2

Results of one way analysis of variance assessing differences among a) land-use categories and b) major genera in mean basal area increment (BAI) across the study period (1982–2012). Results of individual comparisons indicated by letters, with land-use categories and genera that do not differ sharing letters

Full size image

Plot-level total BAI did not vary significantly among land-use categories (F4,80 = 1.046, p = 0.389), but BAI on a per stem basis did differ among land uses (F4,80 = 3.288, p = 0.015). However, when BAI was scaled by total BA present on the plot this effect was non-significant (F4,80 = 0.389, p = 0.816). The variability of BAI across years (expressed as the coefficient of variation in BAI) also did not differ significantly among land-use categories (F4,80 = 0.674, p = 0.612).

Objective 2 – drought response

There was a strong relationship between growing season PDSI and residual (detrended) BAI across the full data set (r = 0.65; Fig. 3), and this relationship was consistent across all land-use categories (AG. – r = 0.40, CIT – r = 0.52, FOR – r = 0.44, PARK – r = 0.72, RES – r = 0.50). There were noticeable dips in BAI in drought years as indicated in the PDSI record (Fig. 3). BAI resistance to the major droughts was generally low with ~20% reduction in detrended BAI during the 1988 and 2005 droughts and >30% reduction in the 2012 drought (Fig. 3). These represented by far the three lowest resistance values (only values <0.9) for any years in the record going back 50+ years (Supplementary Material, Fig. A6). However, resilience was high to the 1988 and 2005 droughts (3% and 7% increase in detrended BAI respectively).

Fig. 3
figure 3

Trends in basal area increment (BAI) and Palmer Drought Severity Index (PDSI) over the study period

Full size image

Tree drought resistance in the three major droughts (1988, 2005, and 2012) did not vary significantly among land-use classes (1988 – F4,287 = 0.503, p = 0.734; 2005 – F4,287 = 0.701, p = 0.592; 2012 – F4,287 = 1.57, p = 0.182; Fig. 4a). However, drought resistance did vary among major genera for each of the drought years (1988 – F4,287 = 3.51, p = 0.009; 2005 – F4,287 = 2.75, p = 0.029; 2012 – F4,287 = 2.82, p = 0.026, Fig. 4b) and there was not a significant interaction between genus and land use. Fraxinus and Quercus spp. were more resistant to drought in 2012 and 2005 while species in Acer were very resistant to the 1988 drought compared to other species (Fig. 4b). Drought resistance tended to be lower for the 2012 drought than the 2005 and 1988 droughts across most land uses and genera (Fig. 4). Drought resistance varied among common species for the 2005 drought (F18,451 = 1.63, p = 0.051), but not the 2012 drought (F18,451 = 1.47, p = 0.098). The highest resistance to the 2005 drought was found for bur oak (Quercus macrocarpa) and silver maple (Acer saccharinum), but differences among species were not significant following adjustment for multiple comparisons.

Fig. 4
figure 4

Results of one way analysis of variance assessing differences among a) land-use categories and b) major genera in resistance to the 1988, 2005 and 2012 drought events. Significance of ANOVA tests indicated in legend with * and ** indicating significance at p = 0.1 and p = 0.05 respectively

Full size image

Resilience to the 2005 and 1988 drought did not vary significantly among land-use classes (1988 – F4,287 = 0.779, p = 0.541; 2005 – F4,287 = 0.503, p = 0.734; Fig. 5a). Trees across almost all land-use categories (exception was RES in 2005) appeared to be resilient to drought, with growth in the 5-years following the drought event returning to or exceeding pre-drought growth (Fig 5). Drought resilience did vary significantly among the most common genera following the 2005 (F4,287 = 3.27, p = 0.012) but not the 1988 drought (F4,287 = 0.715, p = 0.583; Fig. 5b). Resilience of Prunus species was high following both droughts, but especially (and significantly greater than other genera) after the 2005 drought. The highest resilience to the 2005 drought was found for tree-of-heaven (Ailanthus altissima) and Siberian elm (Ulmus pumila), but differences among species were not significant following adjustment for multiple comparisons.

Fig. 5
figure 5

Results of one way analysis of variance assessing differences among a) land-use categories and b) major genera in resilience to the 2005 and 2012 drought events. Significance of ANOVA tests indicated in legend with * and ** indicating significance at p = 0.1 and p = 0.05 respectively

Full size image

Objective 3 – drivers of variability

The model with the highest predictive power of annual BAI included as predictors plot level BA, urban distance, species composition, soil, ground cover, and canopy height (Table 2). The model containing these predictors had a 65% weighting relative to the full model set. This model had an adjusted R2 of 0.53. Annual BAI was higher in locations with more total BA, farther from the urban core, shorter canopy height, with lower soil pH and organic matter content, and with oak-hickory versus pine dominance. The model with the highest predictive power of Rel BAI included as predictors stem count, urban distance, species composition, soil, and canopy height (Table 2). The model containing these predictors had a 40% weighting relative to the full model set. This model had an adjusted R2 of 0.29. Rel BAI had equivalent relationships with predictors as above, but growth per unit BA was greater in lower density stands. The model with the highest predictive power of CV BAI included as predictors plot BA, Simpson’s diversity index, and ground cover (Table 2). The model containing these predictors had a 22% weighting relative to the full model set. This model had an adjusted R2 of 0.28. CV BAI was higher with higher BA and greater species diversity.

Table 2 List of models for predicting basal area increment (BAI), BAI per unit of basal area (Rel BAI), and variation in basal area increment (CV BAI)
Full size table

The model with the highest predictive power of Resilience to the 2005 drought included species composition and soil factors as predictors (Table 3). The model containing these predictors had a 20% weighting relative to the full model set. This model had an adjusted R2 of 0.14. Resilience to the 2005 drought was higher in areas with higher soil pH and organic matter content and with greater maple dominance. The model with the highest predictive power of resistance to the 2012 drought included canopy height and urban distance as predictors (Table 3). Resistance to the 2012 drought was higher where canopy was shorter and farther from the urban core. The model containing these predictors had a 13% weighting relative to the full model set. However, this model had an adjusted R2 of only 0.05. The model with the highest predictive power of resistance to 2005 drought included urban distance, species composition, and soil as predictors (Table 3). Resistance to the 2005 drought was higher in areas with higher soil pH and organic matter content, with greater maple dominance, and closer to the urban core. The model containing these predictors had a 12% weighting relative to the full model set. However, this model had an adjusted R2 of only 0.05.

Table 3 List of models for predicting resilience and resistance to 2005 and 2012 drought events
Full size table

Discussion

Urban forest growth

Generalized land-use categories were not a strong or consistent predictor of variation in urban tree and forest growth rates across the Chicago region. Individual tree growth varied somewhat among land-use categories, with lower growth rates evident in forest and agricultural settings (Fig. 1). However, this effect was not evident in overall forest growth or relative basal area increment (i.e., per unit BA) and thus was probably mostly associated with either greater competition (high density) or smaller trees (lower mean diameter) evident in forest and agriculture plots (Table 1). Prior work has shown some relationship between broad land-use categorizations and urban tree growth, but for a much narrower range of conditions and smaller study region (Fahey et al. 2013). In the present study, in order to derive a useable sample size from the existing Urban Tree Census data, similar land-use categories were combined into more general categories than those used by urban planners (e.g., Commercial, Industrial, and Transportation were combined into a single category). This generalization may have affected the potential for detecting differences among land-use classes, but there is significant variation within even narrowly-defined land-use classes across large urban regions (Cadenasso et al. 2007). For example, the characteristics of areas classified as single-family residential differs strongly in suburban versus urban areas (Grove et al. 2014).

Neither position along a broadly-defined urban-rural gradient nor impervious surface cover (plot or area) were strong predictors of variation in urban forest growth. Prior studies have illustrated variation in forest growth across urban-rural gradients, but have generally focused on comparisons of urban natural areas to wildland forests on the urban fringe (Iakovoglou et al. 2001; Gillner et al. 2013; Pretzsch et al. 2015). In this study, impervious surface cover (at two scales – plot and 100 m buffer around the plot) and distance from the urban core were utilized as general metrics of the position of a plot along the regional urban-rural gradient (Medley et al. 1995). For ease of interpretation we focused on these simple metrics of position along the urban-rural gradient, but recognize that other metrics could be used to define these gradients, including synthetic variables combining geographic position and landscape structure (Theobald 2004; Grove et al. 2006; Pickett et al. 2008). Although position along the geographic urban-rural gradient was not in itself strongly related to urban forest growth, it was included as a predictor in models of growth variation. This result suggests that the effect of the biotic and abiotic factors that influence urban tree and forest growth (see below) vary somewhat across the urban landscape.

Variation in urban forest growth across the Chicago region was predicted most strongly by a combination of site factors, both “natural” and anthropogenic, and stand/community characteristics. Structural characteristics (such as total BA and canopy height), species composition, soil characteristics, and ground cover type were all included as predictors of basal area increment (BAI) and relative BAI. While BAI was greater with greater plot-level BA, there was actually greater BAI in plots with lower average canopy height, which could reflect fast-growing younger trees in successional stands. Plot-level BA was the strongest individual predictor of BAI (R2 = 0.35), but the inclusion of the other predictors greatly improved predictive power (R2 = 0.53 for the top model). Temporal variability in urban forest growth (CV BAI) was most strongly predicted by a combination of total plot BA, species diversity, and ground cover type. However, variability was positively related with species diversity, the opposite of the expectation if diversity was buffering variability in growth (McCann 2000). Prior work has illustrated links between soil properties and urban tree growth (Whitlow and Bassuk 1987; Kjelgren and Clark 1992; Jim 1998). However, although this study was paired with detailed analysis of soil properties, there was not a strong relationship between soil characteristics and urban forest growth. Even with detailed soil characterizations our understanding of soil properties may not have been sufficient to quantify the role that soil has in affecting urban forest growth (Jim 1998). For example, we were not able to directly measure some characteristics of soils (e.g., water potential) or variation in any factors over time at the scale at which this study was applied. These issues illustrate the difficulty in understanding linkages between fine-scale abiotic conditions and tree growth across large, extremely heterogeneous urban landscapes (Pataki et al. 2011).

Urban forest drought response

Urban forest growth across the Chicago region landscape was highly responsive to meteorological drought across the ~50 year study period. There has been relatively little prior work investigating links between drought and tree growth in urban regions (Chen et al. 2011; Bartens et al. 2012; Gillner et al. 2014; Moser et al. 2016). Although dendrochronological studies have illustrated that Chicago region forests are responsive to drought conditions (Bowles et al. 2005), the extension of this linkage to urban tree growth is not a foregone conclusion. There are many factors associated with urban forests and urban sites that could disrupt the linkage between drought and tree growth (Gillner et al. 2014). For example, some urban sites and land uses are often subject to significant supplemental water inputs, both intentional and unintentional (e.g., leaking pipes), which could limit the response of these trees to drought (Bijoor et al. 2012). However, there was relatively strong correlation between meteorological drought and growth across all land-use categories, and the categories that are most likely to be affected by supplemental watering (residential and park) illustrated the strongest drought effect on growth.

Urban forest growth was strongly influenced by intense regional droughts in 1988, 2005 and 2012 as growth was severely depressed during those years (i.e., not resistant). However, growth was highly resilient to the 1988 and 2005 droughts, with equivalent or greater growth in the years following the drought. Prior work has indicated significant drought resilience in urban trees across a range of species, but not in relation to a range of land-uses or across urban landscapes (Gillner et al. 2013; Gillner et al. 2014). Given the timing of the data collection we were not able to quantify resilience to the 2012 drought, which was the most extreme and showed the greatest growth decline relative to the previous years (Fig. 1). Resilience to this drought may not have been as robust given the much greater growth declines evident in the data. However, the lower resistance to the most recent drought relative to the previous droughts could reflect missing data for trees that eventually died or were removed. Such trees may not have yet succumbed to the effects of the 2012 drought or may not have been removed yet, but the potential effect of this issue is not clear.

Although there were strong short-term effects of drought on urban forest growth, these effects did not appear to differ greatly among land-use categories or major genera or species across the urban-rural gradient. There was not significant difference in resistance or resilience across land-use categories for any of the drought years that were evaluated (Figs. 4 & 5). The highest resistance levels tended to occur in the residential land-use category, which could reflect the influence of factors such as supplemental water input (Bijoor et al. 2012), but again these differences were not statistically significant and potential drivers of such patterns are unclear. There were significant differences in resistance and resilience among major genera (Figs. 4 & 5), but these were not consistent across drought years. The lack of any consistent differences likely reflect the issue of lumping at the genus-level, which combines species with somewhat different traits and life-histories as well as native and non-native species. As has been previously illustrated there were significant differences among urban tree species in drought resistance and resilience (Gillner et al. 2014), but the low sample size for most species in this study precluded strong statistical inference about these patterns. There were also few strong predictors of variation in drought resistance or resilience, probably in part because resistance and resilience did not vary greatly across plots (Supplementary Material, Fig. A5). The urban forest as a whole appeared to be highly resilient to drought effects across a broad range of stand and site conditions.

Conclusions and implications for management and planning

Patterns of variation in urban forest productivity do not appear to be easily predictable based on either generalized land-use categories and urban-rural gradients or specific site and stand conditions. The absence of strong patterns is not especially surprising given previous efforts to characterize variation in urban forest growth and the high variability in urban tree growth even within a given site/stand (Kjelgren and Clark 1992; Iakovoglou et al. 2001; Fahey et al. 2013; Ko et al. 2015). The ecological conditions of urban sites are very difficult to predict or fully quantify, in part because of the long-lasting impact of previous anthropogenic impacts on factors such as hydrology and soil characteristics (Boone et al. 2010). Urban forests are dynamic socio-ecological systems (both temporally and spatially) and the factors that affect tree growth may not be evident in a site-based analysis or be well characterized by broad land-use categorizations (Vogt et al. 2015). The unpredictability of urban forest growth could preclude efforts to understand variation in C dynamics across urban regions and complicate management efforts focused on maximizing urban forest productivity and C sequestration (Nowak and Crane 2002; Nowak et al. 2004). However, some of the factors that were most predictive of urban forest growth (BA, canopy height, species composition) could potentially be assessed through remotely sensed data (Gu et al. 2015; Alonzo et al. 2016; Gu and Townsend 2017). Future work should explore these relationships as well as the relationships between remotely sensed productivity surrogates and site-level data (Raciti et al. 2014).

The high resilience of urban trees and forests to “extreme” drought conditions (USDA 2016) across a wide variety of sites is encouraging, although not entirely unexpected. Many species and cultivars are selected for use in the urban environment based on their drought tolerance and the urban environment has already likely indirectly selected for heat and drought tolerant species in remnant and spontaneous forest stands (Whitlow and Bassuk 1987; Santamour 2004). However, the low resistance of urban forests and trees to extreme drought, especially in 2012, indicates that they are susceptible to the effects of such disturbances. Future droughts could more often reach the “exceptional” drought level and may also more commonly last for multiple years, neither of which occurred in the time period included in this study (Cook and Krusic 2016; USDA 2016). Such high intensity, multiple year “climate change-type droughts” could have greater impact on tree growth and survival than droughts in even the recent historical record (Breshears et al. 2005; Breshears et al. 2008). Urban forests may not be resilient to such events, especially when combined with the effects of pests and diseases, urban development, and urban heat islands (Cregg and Dix 2001; Chen et al. 2011; Pataki et al. 2011).