, Volume 8, Issue 5, pp 568–582

Forest Remnants Along Urban-Rural Gradients: Examining Their Potential for Global Change Research


    • Department of BiologyUniversity of Louisville
  • Christopher E. Tripler
    • Department of BiologyUniversity of Louisville
Mini Review

DOI: 10.1007/s10021-003-0172-6

Cite this article as:
Carreiro, M.M. & Tripler, C.E. Ecosystems (2005) 8: 568. doi:10.1007/s10021-003-0172-6


Over the next century, ecosystems throughout the world will be responding to rapid changes in climate and rising levels of carbon dioxide, inorganic N and ozone. Because people depend on biological systems for water, food and other ecosystem services, predicting the range of responses to global change for various ecosystem types in different geographic locations is a high priority. Modeling exercises and manipulative experimentation have been the principle approaches used to place upper and lower bounds on community and ecosystem responses. However, each of these approaches has recognized limitations. Manipulative experiments cannot vary all the relevant factors and are often performed at small spatio-temporal scales. Modeling is limited by data availability and by our knowledge of how current observations translate into future conditions. These weaknesses would improve if we could observe ecosystems that have already responded to global change factors and thus presage shifts in ecosystem structure and function. Here we consider whether urban forest remnants might offer this ability. As urban forests have been exposed to elevated temperature, carbon dioxide, nitrogen deposition and ozone for many decades, they may be ahead of the global change “response curve” for forests in their region. Therefore, not only might forests along urbanization gradients provide us with natural experiments for studying current responses to global change factors, but their legacy of response to past urbanization may also constitute space-for-time substitution experiments for predicting likely regional forest responses to continued environmental change. For this approach to be successful, appropriate criteria must be developed for selecting forest remnants and plots that would optimize our ability to detect incipient forest responses to spatial variation in global change factors along urbanization gradients, while minimizing artifacts associated with remnant size and factors other than those that simulate global change. Studying forests that meet such criteria along urban-to-rural gradients could become an informative part of a mixed strategy of approaches for improving forecasts of forest ecosystem change at the regional scale.


global changeclimate change, urban-rural gradientsurban forestsnitrogen depositionozoneelevated CO2space-for-time substitution


Over the last decade a few scientists (for example, Groffman and others 1995; Idso and others 2002; Ziska and others 2003) working in urban areas have suggested that because cities are surrounded by a heat island and produce the very pollutants (carbon dioxide (CO2), various forms of inorganic reactive nitrogen, ozone (O3)) that are causing environmental change at regional and planetary levels, cities may provide analogs for studying ecological responses to these global change factors. The notion that cities can be used as laboratories for this purpose is provocative and intuitively appealing, yet has not been methodically developed or tested, perhaps because intense, multi-disciplinary scientific research in urban areas is in its early stages and because there are many conceptual and logistic challenges imposed by landscapes characterized by such a high degree of spatial heterogeneity and contrast. The purpose of this paper is to explore further the idea that cities may provide us with equivalents for studying global change impacts on natural ecosystems. We will do this by evaluating the potential of using forests located along urban-to-rural land-use gradients in the U.S.A. as “natural experiments” for studying ongoing global change impacts and also as “space-for-time substitution experiments” that would allow us to use these forests as sentinels for verifying predictions of forest responses to global change in the coming century. We start by summarizing some predictions of U.S. forest responses to global change derived from other approaches– modeling and manipulative studies– and contrast their advantages and limitations with those inherent to the “natural experiment” approach in general and to studies along urban-to-rural gradients in particular. Next, the characteristics of cities that may make them useful global environmental change analogs are discussed. With this background, we then suggest a set of criteria that might optimize our ability to detect whether forests along urbanization gradients are responding to varying levels of abiotic “treatments” originating in the surrounding land-use/ land-cover types. We also discuss criteria that might aid in selecting study sites that minimize interference from urban effects that do not simulate global change factors. Finally, we propose tools and approaches that would allow us to determine if these forests can also be used as space-for-time substitution experiments for global change research. With this review, we hope to stimulate discussion of this potential model for global change studies as well as to generate greater interest in conducting a coordinated, methodical search for cities that contain appropriate forest remnants for this purpose. If successful, this approach would add value to integrated manipulative studies and modeling efforts. An integrated approach employing urban-rural gradients may have potential to provide a much clearer picture of ecosystem change at a regional scale than traditional approaches alone.


Due to ongoing fossil fuel combustion and land-cover change, current projections hold that greenhouse gases (CO2, CH4, N2O) will continue accumulating in the atmosphere over the next century, causing mean global temperature to rise by 1.7 to 4.9°C (Wigley and Raper 2001). During that interval, projections show global mean CO2 concentrations increasing from its present 373 ppmV levels to 540–970 ppmV (IPCC Working Group I 2001). Global precipitation patterns are expected to change and become more variable as well (IPCC Working Group I 2001). In addition to altered climate and higher CO2 concentrations, regional levels of air pollutants, like O3 and various forms of biologically reactive inorganic N like NOx gases, nitrate (NO3) and ammonium (NH4+) produced by agro-industrial complexes and cities, are also expected to rise (Galloway and others 1995). Together, these proximate abiotic drivers of global environmental change will increasingly impinge on many terrestrial systems and are expected to affect the productivity and species composition of biotic communities over large areas of the world (IPCC Working Group II 2001; National Assessment Synthesis Team 2001).

Predicting the effects of future environmental change on forests in particular, has become a research and management priority in the USA, because along with water resources, agriculture, human health and coastal zones, forests are both vital to the nation’s economic and environmental security, and sensitive to global change factors (McNulty and Aber 2001). Forests cover approximately 32% of the land surface in all 50 of the United States (302 million hectares (ha) (USDA Forest Service 2001). They store vast amounts of carbon in soil and wood, filter pollutants from air and water, harbor 80% of our nation’s wildlife species, and supply two-thirds of our run-off water (Miller 2002). Forests also provide nearly 400 million cubic meters of timber for lumber, pulp and fiber annually (USDA Forest Service 2001). But because U.S. forests consist of many different tree communities, are located on different soil types, and span a continent with large east-west, north-south and elevational climatic gradients, predicting their future responses to global environmental change is a complex challenge.

The scientific community has tackled this challenge in two major ways: 1) by modeling forest species and ecosystem responses to particular subsets of global change drivers, and 2) by performing seedling-to forest-plot-level experiments that manipulate one or two of these factors to assess how a system will respond to foreseeable changes. For example, Global Circulation Models (GCMs) linked to biogeochemical (BGC) and biodiversity models predict shifts in forest ecosystem processes and tree community distributions in the conterminous United States over the next century (the Vegetation/Ecosystem Modeling and Analysis Project (VEMAP) (VEMAP members 1995; Aber and others 2001; Dale and others 2001; Hansen and others 2001). Combinations of all BGC models driven by the Hadley Centre’s HADCM2SUL GCM forecast increases in mean carbon storage in forest vegetation that range from 3.3 to 13.2 Pg (Aber and others 2001). According to VEMAP forest community projections for seven GCMs linked to four biodiversity models, potential habitat in the eastern United States for spruce-fir and aspen-birch forests will decline by 97% and 92%, respectively (Hansen and others 2001). Seven important tree species may lose 90% or more of their potential habitat. These include sugar maple, quaking aspen, and paper birch whose optimal habitats eventually may recede almost entirely to Canada. In contrast, the potential habitat range for some community types may expand greatly (for example, 34% and 230% for oak-hickory and southern oak-pine, respectively; Hansen and others 2001). According to predictions, many other tree species will remain in their present ranges but become more or less abundant (Iverson and Prasad 1998), thus altering community structure and possibly ecosystem function. In addition, interaction effects are expected among the variables driving change. For example, using both the mechanistic models PnET-II and TEM 4.0, Jenkins and others (2000) showed that elevated CO2 would increase forest productivity in the northeastern U.S. by 30 to 38%. However, when PnET-II was adapted to evaluate O3 effects on forest productivity, Laurance and others (2000) found that realistic O3 levels could depress net productivity of mixed deciduous forests in the northeastern U.S. by 2 to 17% depending on soil water availability–thus constraining the ability of trees to respond fully to the potential stimulatory effects of increased CO2 and warmer temperatures. Although this study did not explicitly incorporate limiting nutrient interactions with CO2, N deposition could complicate matters even further in this regard. Hence, modeling efforts have provided a glimpse of the future, but are constrained by our poor understanding of the complexity of ecosystem responses to global change drivers. In this sense, modeling results and their inherent uncertainty illustrate the need for more widespread empirical studies that can invalidate particular scenarios and thereby improve the certainty and precision of the predictive models.

Models base their structure partly on controlled experimentation designed to reveal the potential impacts that global change variables will have on tree species and forest ecosystem processes. These include experiments that manipulate CO2, N, O3 and temperature from greenhouse to ecosystem scales (Bunce 1992; Bazzaz and Miao 1993; Norby 1996; Aber and others 1998; DeLucia and others 1999; Carreiro and others 2000; Shaver and others 2000; Zak and others 2000; Rustad and others 2000; Finzi and others 2002). Although in the short term elevated CO2 stimulates plant canopy growth and Leaf Area Index (Norby 1996; Amthor 1999; DeLucia and others 1999), over longer periods it may decrease above-ground plant growth (Körner 2000). On the other hand, elevated CO2 appears to consistently stimulate belowground NPP in woody plants (Rogers and others 1999). Elevated CO2 also increases leaf-level water-use and nutrient-use efficiencies for many plants, but not necessarily for some widespread species of oaks with normally low stomatal conductance rates (Bunce 1992; Dixon and others 1995). Greater N availability interacts synergistically with elevated atmospheric CO2 to increase primary production further (Johnson and others 1998; Zak and others 2000). Increased N deposition to the forest floor might also affect leaf litter decay rates, but the magnitude and direction of the response might vary with litter species and composition of the decomposer community (Magill and Aber 1998; Carreiro and others 2000). At concentrations that can affect plant growth performance, O3 can penetrate below forest canopies to affect plants closer to the ground (Fuentes and others 1992; Volin and others 1998). However, vulnerability to injury and decreased performance depends on species-specific variation in stomatal conductance (Volin and others 1998). Therefore, the effects of O3 and its potential interactions with elevated CO2 and moisture availability on tree performance and primary production are likely to differ among tree species and interact with conditions in their specific geographic locations.

Superimposed on the biogeochemical drivers is elevated temperature, which probably will directly affect many ecosystem processes, including primary production, decomposition, net ecosystem production, and evapotranspiration, by increasing the metabolic rates of plants, microbes and ectothermic animals (Shaver and others 2000). Rustad and others (2001) showed that increasing soil temperature by 1° to 3.5°C consistently increased soil respiration and N-mineralization rates across several forest sites. However, temperature responses of other soil processes like litter decay and trace gas flux also depended on soil moisture and N availability. The indirect effects of warmer temperature on forest processes via increased evapotranspiration and soil moisture loss will also be expected to vary with topography, soils, dominant tree species and precipitation patterns. By exerting direct effects on individual organisms as well as indirect effects on key biogeochemical processes, temperature acts as an overarching driver that further complicates our ability to predict the long-term responses of ecosystems to global change.

Predicting the interactive net effects of these four abiotic variables on intact ecosystems is difficult even with the aid of models and manipulative experiments, because of variable geographic factors and short-and long-term feedbacks within a system that may either dampen or amplify the effects of any single factor. Therefore, although powerful, both mechanistic manipulative studies and modeling efforts have some important limitations for predicting the future ecological functioning and species composition of forests as global environmental change factors intensify and interact. Experiments that manipulate temperature, precipitation, CO2, N and O3 individually or in two-by-two factorial fashion do increase our precision in knowing how certain species and ecosystem compartments may respond to specific global change phenomena. However, extrapolations from manipulated experiments to the ecosystem level can be limited by their restricted spatial and temporal scales, which often do not include feedbacks and synergisms from indirect effects that these driving variables have on the system. Unexpected ecological responses may arise when above-and below-ground organisms and processes are fully coupled and interact over time. This degree of complexity can only be incorporated into ecological experiments by studying intact forest systems. In addition, forest responses to differing levels and combinations of these abiotic factors are contingent on local to regional differences in dominant species, soils, and topography. Therefore, model-based predictions become increasingly more uncertain as the focal geographic location of interest becomes smaller.


Cities are sources of the very chemical pollutants that are driving environmental change globally. Cities also exert a heat-island effect on their surroundings. Therefore, forest remnants surrounded by an urban environment have been exposed for many decades to the major global change drivers that affect forests worldwide. Hence, when compared with similar reference forests farther from the city, urban forest remnants might serve as whole ecosystem “natural” experiments for confirming or invalidating plant and ecosystem responses found in manipulated experiments and predicted by GCM-linked biodiversity and biogeochemical models. But there are potential caveats to this approach as well. Many factors co-vary in space and time along urban-rural gradients and their individual effects may be impossible to separate. Moreover, because the approach is based on remnants, the effects of patch size and proportionately greater edge influence may confound the interpretation of results. Some urban and suburban forests may be heavily trafficked by people or extensively invaded by exotic species, whereas rural counterparts are not. Hence there are advantages and disadvantages to this approach and careful consideration of both is needed to determine the extent to which urban-rural gradients may become an important tool in global change research.

A major advantage of comparative “natural” experiments involving forest remnants along urban-to-rural gradients is that they might provide realistic information about integrated community-and ecosystem-level responses to multiple environmental factors. For instance, urban-to-rural land-use gradients would incorporate local to regional variation in terrain, soils, species and air pollutant influences that are not captured by GCMs, which are based on idealized landscapes and information from specific plot-level experiments. Therefore, information gathered from urban forest remnants and their reference sites would have the advantage of reducing uncertainties regarding forest responses to global environmental change at smaller geographic scales than those dealt with by models like GCMs. Also, unlike many manipulated experiments, the externally derived “treatments” are being “applied” to these forests via natural pathways such as precipitation and advection. Moreover, the treatment dose levels and intervals are realistic, being deposited more naturally than the rapid step-wise applications that characterize manipulative experiments. Forests within urbanized landscapes have been acclimating holistically and gradually to the environmental change around them, thus incorporating time lags and the net effects of potential feedback loops that develop within the system. In addition, because urban forests may have been exposed for decades to higher levels of these abiotic global change factors than corresponding rural forests, they may already be ahead of the environmental change “response curve” for their respective regions. Therefore urban forests (Figure 1A) along with less affected reference forests further away from the city (Figure 1B) might be useful as space-for-time substitutes (Pickett 1990) for global change research. The study of urban forest responses to factors that simulate global change could therefore help us anticipate and prioritize adaptive management strategies that are tailored to a particular region.
Figure 1A

A view from the southern end of Iroquois Park, a 304 ha urban forest remnant in Louisville, Kentucky. This forest is surrounded by tree-lined residential areas to the south and west, a busy commercial avenue to the east, and a golf course to the north. Downtown Louisville (population 700,000) can be seen near the center horizon about 5 km away. B. The forested Knob hills about 40 km south of downtown Louisville, Kentucky. These forests are surrounded by farms and small residential clusters, and are at the rural end of a series of forest research plots established along an urban-to-rural land-use gradient. (Photos: Margaret Carreiro).

There are inherent limitations to this kind of natural experiment approach as well. The net responses of such “uncontrolled” experiments involving multiple driving variables do not permit the effects of individual factors on the system to be determined directly. However, if improved mechanistic understanding for the responses is sought, then controlled manipulative experiments can be conducted within the systems being compared (for example, reciprocal transplants, bioassays, addition experiments, removal experiments). Likewise, for forests along urban-rural land-use gradients, once the net integrated responses of these systems are characterized and hypotheses generated about causation, subsequent manipulative experiments (for example, irrigation, nutrient additions or removals, soil warming) could be performed within the forests or in greenhouse settings to ascertain the relative importance of the driving variables to particular processes. In addition, recently developed statistical tools such as structural equation modeling (Maruyama 1998) and regression tree analysis (De’ath and Fabricus 2000) may be useful for estimating the relative importance of individual driving variables to a particular net forest response. Hence, by linking an urban-to-rural gradient approach, with its unique view of whole ecosystem responses to environmental change, to controlled manipulations and/or multivariate deconstruction analyses, we may be able to gain a much better understanding of ecosystem responses to global change. Such knowledge would in turn benefit modeling efforts by improving both the data and the structure applied to them.

Other factors unrelated to variation in global change factors may limit the potential of this approach. For example, appropriately sized forest remnants may not exist in some cities within forested biomes. Urban forest remnants are likely to be smaller than suburban and rural forests nearby (Medley and others 1995) and therefore edge-related phenomena become relatively more difficult to avoid in urban forests. Whether an urban forest remnant may be too large for such studies presents an interesting consideration about the degree to which the abiotic conditions in a highly altered, urban matrix may penetrate from the sides and from above to change the average condition for the forest canopy and the forest interior at ground level (Figure 2). However, the existence of urban remnants so large that they entirely determine their own climate and are unaffected by inputs from the external matrix environment is less likely to impede this research approach than there being a preponderance of urban remnants that are too small. Alternatively, large urban forests may exist but not meet other plot-level criteria (described in a later section) needed to make valid response comparisons with suburban and rural reference plots. High exotic species abundance and human trampling are also concerns, but our personal experience in urban and suburban forests suggests that these two latter notions about urban forest condition are not encountered as often as is generally assumed, particularly for large forest remnants.
Figure 2

Measurements of global change factors (F) (for example, temperature, CO2, atmospheric N and O3) outside forest remnants are depicted as being higher on average in urban than in rural environments. Whether these factors are maintained at higher levels inside urban forests as compared to rural reference forests is not well known. In the first possibility, a global change factor, for example temperature, decreases through the forest boundary and is on average higher in the urban than in the rural forest interior. Forest plots established away from changing conditions near the boundary could then be used for global change studies. In the second possibility, factor levels outside the urban forest are also higher than in its rural counterpart, but the levels achieved in the two forest interiors are similar. If this latter possibility is the case for all four global change factors, then forests along urban-to-rural gradients may not provide analogs for global change studies. Simultaneous paired measurements of temperature, CO2, atmospheric N deposition and O3 are needed in more urban and rural forests to determine whether the first or second possibility is the norm.

Several researchers (Körner 2000; Shaver and others 2000; Aber and others 2001) have called for the establishment of national forest monitoring programs to provide early detection of species change and the alterations in C, N and water cycling that would drive shifts in plant species composition via physiological limitations and altered biotic interactions. Forests along urbanization gradients could play an important role in such a network of monitored sites, if it can be demonstrated that urban forests have been and are now being subjected to higher levels of these abiotic factors than other similar forests nearby. This issue is addressed in the following section.


The first criterion that must be satisfied for urban forests to be useful as global change analogs is that they must have been in the past and currently be influenced by environmental factors that simulate global change. Intuitively it seems likely that they have been, but few hard data exist to prove it. The paucity of data available stems from the fact that urban ecology is only beginning to emerge as a distinct field, and cities have largely been ignored in general ecological studies. The data that are available offer hope that urban forests do experience global change-like influences, but these are only initial indications and much more data on a broader array of cities are needed to determine how closely urban factors mimic anticipated global change conditions.

Urban Heat Islands: Surrogates for Global Warming?

Thermal absorption properties of impervious cover, urban vertical roughness, generation of heat from combustion activities, and low soil moisture content create localized meso-scale (104 to 105 m) warming phenomena known as “urban heat islands” (Oke 1995). This temperature contrast between urban and nearby rural environments intensifies with urban expansion and can be as great as 12°C (Oke 1995). In the last half-century, many cities have already experienced temperature increases that match or exceed those projected in the IPCC 2001 report (a mean global rate of increase of 0.17 to 0.5°C per decade over the next 100 years). For example, Cardelino and Chameides (1990) used meteorological data and an empirical relationship between population and urban heat islands to infer that Atlanta, Georgia experienced a 2°C rise in temperature between 1974 and 1988 (1.3°C per decade) due to urban expansion.

Forest remnants surrounded by an urban built environment may well be in a dynamic thermal equilibrium with the city and thus on average warmer than a similar forest remnant surrounded by vegetation in nearby countryside (Figure 2). However, forests do create their own cooler microclimate within the urban matrix (Bonan 2002) and so the degree to which warmer temperatures in the built environment penetrate any single urban forest remnant requires measurement and documentation. Studies examining the influence of different urban land-cover types on the climate of embedded forest remnants (and vice versa) have apparently not been undertaken in a broad or concerted manner (but see Upmanis and others 1998). Few simultaneous measurements of air and/or soil temperatures in paired urban and rural forest remnants have been published. Pouyat (1992) found that from May to October 1990 soils at 2 cm depth in urban oak forest remnants in the Bronx, New York City were consistently 2–3°C warmer (and sometimes up to 5°C warmer) than rural counterparts 130 km distant. Given that the heat island effect is nearly ubiquitous, it seems likely that embedded urban forests will be warmer than neighboring rural forests, but more studies are needed to ascertain whether temperatures in urban forests generally are warmer than rural counterparts nearby and to measure the magnitude of such a potential difference.

Despite the lack of direct temperature observations in forests along urban-to-rural gradients, there is evidence that urban climate affects thermal conditions in embedded habitats differently from rural counterparts. Using satellite data to create a “greenness index”, White and others (2002) found that plants in urban areas in the eastern U.S.A. experience a growing season that was on average 7.6 days longer than that of paired rural areas. They attributed this phenomenon to the urban heat island effect and suggested that it served as a proxy for future global warming to predict the degree to which the growing season in the eastern deciduous forest biome in the United States may be lengthened in coming decades. Similarly, long-term phenological data (1951–1995) from ten cities in central Europe indicate that urban climate has accelerated the timing of spring flowering for different plant species by 3 to 16 days relative to stations in paired rural areas (Rötzer and others 2000). These studies suggest that the urban heat island can affect vegetation over large geographic areas and over decadal time scales.

If air and soil temperatures surrounding and within urban forests are indeed warmer than similar rural counterparts, then one of several criteria would be met for using urban forests to predict forest responses to global environmental change in their respective regions. The current information gap could be dealt with relatively easily and inexpensively by deploying temperature data loggers in forest remnants surrounded by urban, suburban and rural land use. This would be one of the first steps in determining whether urban-to-rural gradients would be useful proxies for global change.

CO2 Concentration in Cities: How Elevated is It?

Because cities are sites of intense fossil fuel combustion and in some cases cement production, it is reasonable to expect that CO2 concentrations in urban areas may be higher than the average tropospheric concentration (373 ppmV; Keeling and Whorf 2003). Several studies have documented that CO2 concentrations are greater in urban environments than global background (Grimmond and others 2002; Takahashi and others 2002; Pataki and others 2003b). Despite the potential for elevated CO2 to stimulate primary production rates of urban vegetation, we have little information about how the supply of CO2 to urban forests compares to that of surrounding suburban or rural reference sites. Idso and others (2002) found that a negative relationship existed between atmospheric CO2 concentrations and distance from the urban center, with a maximum CO2 concentration in the city center that was 67.4% greater than rural background (370 ppmV) and a maximum suburban concentration that was 32.6% greater than rural background. In Palermo, Italy mean CO2 concentrations of 390 ppmV and maximum daytime values of 470 ppmV were measured compared to a mean of 372 ppmV in the nearby countryside (Dongarra and Varrica 2002).

Determining whether urban forest remnants receive more CO2 than adjacent rural sites requires more measurements of both CO2 concentration and flux between the forest and its surroundings. This was done by Grimmond and others (2002) who reported that a single-family, vegetated residential area in suburban Chicago was on average a net carbon source during the growing season when vegetation provides a local sink. Therefore, it appears likely that urban and suburban forest canopies are surrounded by higher than global ambient CO2 levels. To determine the degree to which urban forests may be influenced by increased CO2 emissions in cities, more local-scale (102 to 104 m) studies are needed to establish relationships between different land-cover/land-use types that surround urban and their reference forests, the CO2 concentrations these forests are exposed to, and the net CO2 flux between the atmosphere and the forest canopy integrated over the growing season. Such measurements require more equipment and are more expensive than temperature measurements, but appropriate technologies are well established and readily available.

Atmospheric Nitrogen (N) Deposition to Urban Areas: Do Urban Forests Receive More Fertilizer?

Most of the world’s emissions of fossil-fuel-derived NOx gases are released in urban-industrial areas (Galloway and others 2002). These emissions are expected to rise by 60% in the next twenty years, increasing the impact of N enrichment on many ecosystems (Galloway and others 1995). Despite availability of N concentration data for city air (US EPA), information on actual deposition rates of N to vegetated areas in metropolitan areas has only recently been obtained. Lovett and others (2000) measured atmospheric deposition and throughfall fluxes of NO3 and NH4+ to oak forests located along an urban-rural gradient in the New York City area and found that over the growing season, particulate N inputs from the atmosphere were 17 times greater in New York City forests than in nearby suburban and rural forests. This trend for increased particulate deposition of N to urban forests during the growing season was corroborated in a study of urban oak forests along an urban-rural gradient in Louisville, Kentucky (Carreiro and Tripler unpublished data). Nitrogen in net and total throughfall over the growing season was 33% greater in an oak forest in Louisville than in a nearby rural forest, with a suburban forest being intermediate. In addition, the amount of inorganic N entering the urban forests in these two cities is likely to be underestimated because uptake of gaseous N forms like NOx, which are higher in cities than in rural areas (Baumbach and others 1989), was not measured. If the trends observed for N inputs in New York City and Louisville occur in other cities, then relative to forests in outlying areas, urban forests receive a large N subsidy in dry deposition during the growing season. This steep urban-to-rural N deposition gradient, due largely to localized particulate fall-out, would increase the usefulness of these forests as global change analogs, because variation in photoperiod, climate, and soils that occur over larger, regional deposition gradients would not exist to confound interpretation of ecosystem responses.

Ozone: Can It Cap Forest Responses To Increased Temperature, CO2, and N?

As forest plant species in the United States respond to expected climate change, increased CO2 and N deposition, they will also have to contend with elevated tropospheric O3 in many urban and rural locations. Ozone, an important component of urban photochemical smog, is the product of light-catalyzed chemical interactions between NOx and reactive volatile organic compounds (VOCs) of anthropogenic and biogenic origin. Although cities produce O3, the damaging effects of this strong oxidant on plant tissues are not just local because O3 can continue to be produced and transported far from urban-industrial centers that generate chemical precursors (Smith 1990). Ozone concentrations in urban atmospheres increase with emissions production, decreasing latitude and local weather patterns. In the southern and north central regions of the United States, ozone levels have risen in the past ten years due mostly to a 3% increase in NOx emissions during that interval (EPA National Air Quality Trends 2000). Therefore, O3 will likely grow in importance as a factor that will affect natural vegetation and ecosystem dynamics for many more years.

Ozone’s effects on ecosystems lie in its ability to damage tissue. Ozone can penetrate below the forest canopy (Fuentes and others 1992) and levels of 60–170 ppb sustained for 4 h are known to cause plant injury (Smith 1990). But levels as low as 50 ppb for several hours daily over a 16-day period, although not visibly damaging to tissues, can reduce plant growth and performance (MacKenzie and El-Ashry 1989) by increasing respiration rates, and reducing leaf retention and longevity (Volin and others 1998). Ozone may also affect decomposition processes by altering leaf litter quality (Findlay and others 1996). Ozone’s potential in reducing tree growth along an urban-to-rural gradient was recently shown by Gregg and others (2003). By placing potted Eastern cottonwood (Populus deltoides) saplings in different locations along an urban-rural gradient in the New York City area, they showed that saplings at the rural end of this gradient produced less biomass than saplings in the city. Using a combination of approaches, including lab incubations, experimental manipulations, and statistical examination of air quality data, they were able to infer that higher chronic O3 exposures in the rural end of the gradient were probably responsible (although acute levels of O3 exposure were higher in the city). The extent to which O3 can affect growth of trees in intact urban and paired rural forests, however, still needs to be determined. Although O3 concentration data from monitoring sites are readily available, more measurement of O3 flux to forests along urban-to-rural gradients is needed to establish relationships with forest tree growth responses in situ.

Interaction Effects among Factors

Interaction effects among O3, CO2, N, and temperature would be expected. Warmer temperatures would increase the rates of chemical reactions that produce tropospheric O3. By reducing stomatal conductance, elevated CO2 may protect trees from the adverse effects of O3 and moisture stress (Volin and others 1998) and this could favor trees in urban relative to rural forests. Due to differential species responses to varying combinations of CO2 and O3, the long-term integrated responses of forests to these and other factors like N deposition and warmer temperatures are uncertain. In fact, as opposed to the results obtained by Gregg and others (2003), preliminary evidence from paired urban and rural forests in the Louisville, Kentucky area indicates that average sapling growth rates of chestnut oak, white ash, shagbark hickory, American beech and sugar maple have been adversely affected by conditions in the city over the past five years (Tripler, Carreiro and Canham, unpublished data). Although the reasons for these responses are as yet unclear, this pattern indicates that some factor or factors, perhaps O3 and/or lower soil moisture, have been constraining the growth of young trees in this urban forest relative to conspecifics in a nearby rural forest under the same light conditions. This pattern also illustrates the importance of basing predictions of tree responses to urban vs. rural environments on their growth in intact habitats.

Due to the interaction uncertainties mentioned above, determining the magnitude of variation in the abiotic forcing factors along an urban-rural land-use gradient is essential if forest remnants are to be used as whole ecosystem laboratories for examining global change questions. In addition, for some cities, direct measurement of these factors may reveal instances where steep gradients for all four global change drivers do not occur along a particular urban-to-rural gradient. In such cases, the potential to use these forests for understanding interaction effects among the remaining factors may still exist.


In addition to measuring the levels of global change drivers along urban-rural gradients, forest remnants along these land-use gradients must be found and criteria for selecting the best candidate remnants developed. Criteria for choosing study plots within remnants must also be established to optimize our ability to detect differential responses of these forests to potential variation in the global change drivers and to minimize effects of factors not related to global change. There are new tools as well as established methods that would make the search for candidate remnants more efficient than in the past. Aerial photographs, maps, remote sensing and Geographic Information Systems (GIS) can be used to locate forest fragments in relation to different land-use/ land-cover types. These tools can also be used to determine population density, road density, percent built land and vegetative cover at various radial distances from the remnant so that forests can be classified into more precise and spatially detailed land-use/ land-cover categories. These finer-grained variables would facilitate comparative multivariate analyses between land use/cover attributes and global change drivers within a gradient and potentially among different cities.

Although urban forest remnants are likely to be smaller than most suburban and rural forest fragments (Medley and others 1995), choosing forests in similar size class categories in different land-use types would be preferred for such comparisons. Fragment size is an important consideration because a remnant should be large enough to accommodate study plots far enough away from boundary-edge gradients of rapidly changing light, microclimate and atmospheric inputs (Figure 2; Matlack 1993a; Chen and others 1995; Gehlhausen and others 2000; Weathers and others 2001) and localized biotic and sociological edge exchanges with the land-use matrix (Matlack 1993b) that could confuse interpretation of forest responses to the main global change drivers.

Once candidate forest remnants are located, additional criteria have to be met depending on the scale of the questions being asked. Forest ecosystem responses known to be tightly linked to global change drivers would likely be among the first measurements of interest. These would include measurement of physiological responses and ecosystem-level processes known to be sensitive to climate, air quality and nutrient availability such as tree growth rates, evapotranspiration, soil respiration, foliar nutrient status, decomposition, and soil N-mineralization. To make such comparisons the experimental design requires that land use (the independent variable and surrogate for the treatments) vary, in which case the forest “specimens” (the dependent response variables) should be as similar as possible. As ecosystem-level responses, like those described above, are strongly influenced by tree species and soil type (Finzi and others 1998), forest plots across the gradient should be located on the same soil type and dominated by the same tree species (Figure 3, plots 2 and 5). In addition, plots for these kinds of studies should also be chosen with respect to similarity in total basal area, stand age, slope and aspect, and possess no evidence of recent disturbance (fire, logging, major insect infestation, canopy gaps). Plots should be established as far from paths as possible to minimize the likelihood of trampling. Soil bulk density measurements should be made to ascertain that soil in urban and suburban forests is not more compacted than those in rural settings, as was done by Pouyat and others (1995) for forest plots in the New York City gradient (McDonnell and others 1997). Obtaining records to establish the land-use history of plots or remnants would also aid in interpretation of current forest responses. As rainfall may be as much as 10% to 25% greater downwind of large cities (Bonan 2002), precipitation inputs into forests should be obtained either directly or from meteorological stations nearby to make sure that it does not vary in a non-random manner along a particular urban-rural gradient. These criteria are stringent but would be necessary to increase confidence that species physiology and soil metabolism are responding to variation in abiotic global change drivers and not unrelated factors.
Figure 3

Simplified depiction of how forest plots within remnants along an urban-to-rural gradient may be selected and used for global change research. Only urban and rural end members of the gradient are shown (conifer and deciduous tree icons are used to denote any two contrasting species, and not specifically coniferous vs. deciduous tree responses). Urban and rural plots (Plots 2 and 5) are selected using criteria listed in the text including similar mature stand structure and age. Present mature stand structure indicates similarity in past structure (Plots 1 and 4) as the city’s impact and levels of global change drivers presumably increased around the urban forest. If elevated temperature, CO2, atmospheric N deposition and O3 are documented, forests may be used in the present as natural experiments to investigate current effects of global change factors on ecosystem processes. Seedling and sapling demography and growth performance may also be measured and modeled to estimate forest community structure in the future under presumed greater impact from thermal and pollutant drivers at the urban end of the gradient (Plots 3 and 6). Projected urban stand structure may forecast a potential state for regional rural forests in the future as O3 and N deposition increase at regional scales, and as temperature and atmospheric CO2 concentrations increase at the global scale.

Although it is likely that the urban and rural forest species communities have been responding for some time to the altered land use around them, constraining plot choices to those with similar mature tree composition would also provide starting point benchmarks against which differential changes in plant demography and other factors can be measured into the future (Figure 3, plots 3 and 6). Such selection criteria would also reduce concerns arising from the fact that many cities, and hence their urban forests, are not randomly located and therefore may possess different soils and plant community types than neighboring rural forests. Lastly, it should be recognized that forests along urban-rural gradients may not be as useful for understanding responses to global change that occur at large spatial scales (for example, fire and pest event frequency). However, they may still provide indicators of the potential for such events, such as forest floor desiccation or foliar nutrient changes that may predispose leaves to greater herbivory or pathogen attack.

Finally, the issue of exotic species abundance in forests along the gradient raises interesting considerations, particularly if their distribution along the urban-rural gradient is asymmetrical (usually assumed to be greater at the urban end). On the one hand, high coverage or biomass of exotic canopy trees, understory plants or invertebrates in plots should be avoided because they may strongly influence ecosystem processes in ways that prevent forecasting for regional forests that do not currently contain these exotics. However, urban conditions, including those caused by the abiotic global change drivers themselves, can “incubate” and select the very exotics that are perhaps most likely to invade regional forests in the future. In this case, invaded urban forest remnants would allow us to examine community and process-level outcomes of the interactions between native and exotic species that may be important to a particular region in coming decades. This issue highlights the importance of choosing forest remnants and plots along a gradient based on the questions posed by the investigators.


Current variation in abiotic variables (temperature, CO2 and O3 concentration, and N deposition) should be measured along the gradient of sites so that their contemporaneous relationships with species and ecosystem responses can be evaluated. If these forests can be studied over a three to ten year period, then year-to-year variation in precipitation, temperature extremes and means, and pollutant concentrations and fluxes can place boundaries on the magnitude of the ecosystem’s shorter term responses to such variation. However, if it can be established that urban remnants have in the past been receiving greater inputs of heat, CO2, N and O3 than suburban and rural counterparts, it may also be possible to use the gradient of forest sites as space-for-time substitutes for global change research because urban forests may already be exhibiting response signals that will take rural forests decades yet to develop (Figure 3, the different understory seedling-sapling composition in plots 2 and 5). But, there are caveats to this approach as well. As Pickett pointed out in his analysis of the use of space-for-time substitution in ecology (1990), the study of a spatial chronosequence of habitats does not fully substitute for the passage of time because time itself is not really the variable of interest. Time is a proxy for past conditions and other variables that influenced the current structural and functional status of an ecosystem. Knowledge of past conditions gives us greater explanatory power for interpreting present-day patterns.

How might we obtain such a record of past abiotic influences on forests along urban-rural gradients? Archived national and local data sets and new biogeochemical tools may allow us to open the door to a forest’s past. Daily temperature and precipitation records from meteorological stations in many cities and outlying rural areas can provide an index of the rate of change in urban climate that go back at least several decades and sometimes further (Brazel and others 2000). A reconstruction of the pollution input history of forests along urban-to-rural gradients would also enhance our mechanistic understanding of current system condition. Soils and tree rings offer the possibility of reconstructing such a pollution- and resource-input history. As heavy metals of atmospheric origin accumulate in soils, their concentrations along the gradient of forest sites can be used as a relative indicator of past cumulative atmospheric inputs of less stable nutrients such as N derived from fossil fuel combustion (Friedland and others 1984). It may also be possible to estimate past variation in local climatic, and the duration and relative magnitude of anthropogenic CO2 and N inputs into forests along urban-rural gradients using radioactive 14C (Suess 1970; Takahashi and others 2002) and stable isotopes (13C, 15N, and 18O) in tree rings (Lajtha and Marshall 1994; Nadelhoffer and Fry 1994; Wahlen 1994; Saurer and others 1998; Dongarra and Varrica 2002; Pataki and others 2003a). Takahashi and others (2002) showed that the anthropogenic (fossil fuel combustion) and biogenic (soil respiration) contributions to total atmospheric CO2 at various heights above and within an urban oak forest in Japan could be quantified using the variation in Δ14C and δ13C signatures of these sources. Pataki and others (2003b) used δ13C and δ18O in the urban atmosphere of Salt Lake City to estimate the relative contributions of anthropogenic and local biogenic CO2 to the atmospheric total. These anthropogenic signals may be recorded in tree rings. For example, Dongarra and Varrica (2002) were able to show that an anthropogenic CO2 signal was detectable in the tree ring record of Platanus hybrida. From 1880 to 1998 the 13C/12C ratio in the tree rings of urban trees in Palermo, Italy declined more than that in rural conspecifics. They attributed the difference to there being higher urban CO2 concentrations (with low 13C/12C ratios) due to fossil fuel combustion, particularly over the last 50 years. Therefore, there is promise that radioactive C and stable isotope studies of tree rings may allow us to determine the length of time trees have been exposed to anthropogenic CO2 and perhaps help us estimate past atmospheric CO2 concentrations along urban-rural gradients as well.

If higher levels of past exposure to global change factors can be established for urban relative to rural forests, then urban forests may already have had time for metabolic responses to these factors to cascade and accumulate throughout the system. For example, differences along the gradient in within-species foliar and litter quality may be apparent, suggesting altered patterns of foliar damage or nutrient-use efficiencies. Different soil organic matter pools (labile to recalcitrant) may have accumulated or decreased. Although community-level change in the adult tree population may not yet be apparent in forests along urban-rural gradients, seedling and sapling growth performance for various tree species over recent years may be measured to determine if there are detectable differences in urban vs. rural forests. Modeled projections based on these seedling and sapling performances along urban-rural gradients could reveal likely species compositions of future regional forests under global change conditions (Figure. 3, plots 3 and 6).


Rastetter (1996) examined and promoted several strategies, including use of space-for-time substitution, for validating models of ecosystem response to global change factors. He makes a strong case that this approach would be powerful if a gradient of systems could be found that were reacting differentially to global change factors at temporal scales of interest for global change models (a 20 to 200 year time frame). It may be that urban forest responses are ahead of regional forest responses within that very time frame. Shaver and others (2000) suggested that large-scale monitoring of ecosystems can supplement manipulative approaches to provide early detection of ecosystem responses to global change. In addition, the U.S. Global Change Research Program has called for “new approaches, new knowledge and new capabilities” for evaluating the vulnerability of natural systems to global change, including a suite of multiple stressors driven by land-use transformation (National Assessment Synthesis Team 2001). Studies of forests along urban-rural gradients could provide such a new approach. By incorporating regional-scale variation in topography, soils and dominant species, studies of forests along urban-rural gradients may improve model scenarios for particular geographic locations. Urban forests have likely had several decades or more to adjust and respond to anthropogenic inputs of matter and energy in an integrated fashion that includes intermediate but non-equilibrium ecological responses. Their species and ecosystem responses are less likely to be artifacts of a sudden, large change in condition such as those often imposed by shorter-term (<10 years) experimental treatments. If criteria for remnant and plot selection can be developed that minimize effects of factors not directly related to global change drivers, forests along urban-rural land-use gradients may be positioned to provide us with a) intact experimental units for examining current forest responses to elevated temperature, CO2, O3 and N deposition; and b) space-for-time substitution experiments for refining global change model predictions for forests. New tools and conceptual approaches, such as GIS, radioactive and stable isotopes and structural equation modeling, may allow us now to reexamine the possibility of using forests in cities as analogs for global environmental change. If acceptable candidate remnants exist in even a few metropolitan areas in forest biomes across North America, then studies of forests along urban-rural gradients could be a powerful new approach that could complement traditional manipulative experimentation and modeling efforts, and improve forecasts of forest responses to regional and global environmental change.


We thank Mary Arthur, Jay Gulledge, Tara Trammell, Wei-Xing Zhu, Wayne Zipperer, Ann Kinzig and two anonymous reviewers for comments and suggestions on earlier drafts of this manuscript. This paper also benefited from stimulating discussions with Jay Gulledge as the manuscript evolved. We also thank Keith Mountain for expert piloting during aerial photography sessions. Funding from the University of Louisville Research Foundation is gratefully acknowledged. Tripler was supported on a National Science Foundation Minority Post-doctoral Fellowship (DBI-0208392).

Copyright information

© Springer Science+Business Media, Inc. 2005