Threshold responses and catastrophic state changes have been recognized as critical features of biotic communities and ecosystems (Scheffer and Carpenter 2003, Folke et al. 2004). Feedback mechanisms and emergent behavior on the scale of ecosystems (e.g., Fauchald 2010), and nonlinearity in physiological mechanisms underlying stress responses imply that biological communities may often exhibit threshold changes to environmental factors. Characterization of threshold responses is important because community shifts, such as reduced biodiversity or changes in the dominance of an economically important species, are often undesirable per se and because they may signal irreversible changes in underlying ecological processes (Scheffer 2009) and ecosystem function (e.g., Casini et al. 2009).

In general, biodiversity of different taxonomic groups is weakly correlated (Heino 2010). However, greater congruence of biodiversity could be expected at higher levels of stress as many species approach limits of environmental tolerance. Several researchers have examined the relative importance of environmental stressors in explaining variability of community composition of different taxonomie groups, primarily in lotic ecosystems (but see Brazner et al. 2007a, Niemi et al. 2009, 2011 for lentic examples). Most studies showed discordance in the responses of different taxonomic groups (Hirst et al. 2002, Johnson et al. 2006, Brazner et al. 2007a, Lewis et al. 2007, Johnson and Hering 2009). Understanding the nature of these dissimilar responses is essential to predicting the impacts of anthropogenic stressors on local ecosystems.

An important question is whether large-scale anthropogenic stress results in synchronous thresholds across taxonomic assemblages or whether taxa respond differentially, leading to complex or chaotic community dynamics that defy prediction. Synchronous responses would provide critical evidence in support of regulatory and management frameworks, and lack of such data has been cited as one of the major issues in application of the ecological thresholds concept in biological conservation (Huggett 2005). Considering different mechanisms underlying assemblage-specific responses to anthropogenic stress, synchronous threshold responses across several assemblages would indicate sweeping changes in ecological communities as many species approach limits of environmental tolerance and an increasing potential for functional changes in the ecosystems. Despite rapid development in this area in the last few years (Dodds et al. 2010), few, if any, investigators have examined the concordance of community thresholds across different taxonomie groups and a common framework for addressing these changes, such as a well-defined stress gradient (e.g., Allan et al. 2013), is lacking. Predicting nonlinear responses of different asseblages is essential to forecast the impacts of anthropogenic stress, understand the trade-offs between biodiversity and development (Huggett 2005), inform management- and policyrelevant issues, and identify priority sites for management intervention.

The Laurentian Great Lakes coastal wetlands are a suitable and relevant ecological system for testing the concordance of community thresholds across taxonomie groups because they are critical habitats for many biological assemblages, they are functionally significant as zones of nutrient and contaminant processing at the aquatic-terrestrial interface, and most important, they are distributed across a wide gradient of anthropogenic impact ranging from severely degraded to nearly pristine (Danz et al. 2007). The responses of Great Lakes coastal wetland communities to environmental stress are both scientifically interesting and of crucial importance because of the enormous socioeconomic value of the health of the Great Lakes ecosystem, which contains 20% of the world's fresh water and provides countless ecosystem services ranging from commercial fisheries to recreation (Niemi and McDonald 2004, Niemi et al. 2007, Vaccaro and Reid 2011, Allan et al. 2013).

We used extensive multitaxon data sets from coastal wetlands representing the 762 watersheds bordering the 6044-km mainland shoreline of the US Laurentian Great Lakes (Danz et al. 2005, 2007, Reavie et al. 2006, Bhagat et al. 2007, Howe et al. 2007). These data were collected as part of a multidisciplinary effort to identify community responses to anthropogenic stress in the Great Lakes coastal zone (the Great Lakes Environmental Indicators project; Niemi et al. 2007, Morrice et al. 2008). Here, we relate ecological stress to community changes observed in 5 different taxonomie groups using Threshold Indicator Taxa Analysis (TITAN; Baker and King 2010), a change-point and indicator-species analysis-based approach that integrates responses of many individual taxa to determine cumulative community response along an environmental stress gradient. Our study is the first attempt to examine the nonlinearity in responses of lentic assemblages and to integrate simultaneous responses of several assemblage types to anthropogenic stress. Our goal was to assess the degree of cross-taxon congruence in response to 2 watershed-scale anthropogenic stressors (agriculture and development). More specifically, our objective was to assess whether community changes in response to large-scale anthropogenic stressors were similar among the fish, benthic macroinvertebrate, bird, wetland plant, and diatom assemblages of coastal wetlands.

METHODS

Stressor variables

The Great Lakes Basin has a long history of anthropogenic activities associated with ecosystem degradation (Evans 2005). Widespread clear-cutting across the landscape in the 1800s affected the entire basin, but later landscape disturbances were associated mainly with centers of population, which radiated from river mouths and estuaries and led to widespread habitat loss in tributaries and nutrient and industrial pollutant discharges in river mouths. The strong gradient of climate and landform (i.e., soil fertility) differences across the basin largely explain the land-use patterns that became established in the 20^th century, continue today, and led to dominance of landscapes in the eastern portion of the basin by intensive agriculture.

Figure 1.

Distribution of 337 sampling sites in US Great Lakes coastal ecosystems in the Northern (Laurentian Mixed Forest) and Southern (Eastern Broadleaf Forest) ecological provinces.

We focused on 2 major watershed-scale stressors: % agriculture and % developed land in the watershed. These stressors serve as surrogates, or indicators, of the degree of potential human impact in a watershed, and their impact is mediated through a variety of mechanisms. The advantages of our approach include ability to relate biological responses to readily available, composite measures of stress, whereas disadvantages include lack of connection of these variables to the precise mechanisms responsible for potential effects. Previous investigators clearly demonstrated that biotic assemblages in Great Lakes coastal ecosystems are affected by these stressors (Brazner et al. 2007a, b, Niemi et al. 2009, 2011), but threshold relationships were not examined. A detailed description of the procedures used to characterize watershed-scale anthropogenic stress is available elsewhere (Hollenhorst et al. 2011). Briefly, Arc-Hydro^® with 10-m Digital Elevation Models was used to delineate the boundaries of downstream catchments for each representative watershed and proportions of agriculture and development in those watersheds were derived from the US Geological Survey National Land Cover Dataset (2001;  http://landcover.usgs.gov/usgslandcover.php).

Biotic community variables

We sampled 5 biological assemblages (birds, fishes, benthic macroinvertebrates, diatom algae, vegetation; Fig. 1) in 2002 and 2003. We obtained bird data for each sampling unit from 10-min point-count listening surveys at 1 to 5 points within the wetland, each sampled twice during the June/early-July breeding season (see Howe et al. 1997, 2007 for details). We sampled fishes at each site with 2 sets of large and small fyke nets set overnight just offshore of the 2 dominant shoreline land uses (Brady et al. 2007). Upon net retrieval, we identified fishes to species and released them. We used D-frame nets to sample benthic macroinvertebrates and later identified them in the laboratory to the highest possible resolution (genus for most insects). We estimated vegetation cover for each taxon in randomly selected 1-m² quadrats (Johnston et al. 2010). We collected diatoms from natural substrates at 0.5- to 3-m depths with a push corer, Ponar sampler, or by rock scrubbing (see Reavie et al. 2006 for details). For each assemblage, we pooled within-site data to yield 1 data point per site. Sites sampled for fish and benthic macroinvertebrates overlapped almost completely, and sites sampled for other assemblages overlapped somewhat (Fig. 1). Sites sampled for each assemblage covered similar gradients of % agriculture but somewhat different gradients for % developed land (Table 1) related to issues of site accessibility and the need to satisfy conditions for site selection (e.g., minimum depth for fishing).

Table 1.

Range of stressor values for the sites sampled for each of the 5 assemblages. Agriculture and development are expressed as a percentage of a watershed. Min = minimum, max = maximum.

RESULTS

Levels of anthropogenic stressors for the sampled coastal wetlands spanned a gradient from 0.0 to 89.2% agriculture and 0.0 to 92.9% developed land (Table 1). Synchronous thresholds were found for % developed land but not for % agriculture (Fig. 2A, B). Concordant thresholds in community composition in stress-sensitive (z-) taxa for the 5 taxonomie groups occurred at 4 to 6% developed land in the watershed in the Northern ecoprovince and 7 to 10% in the Southern ecoprovince (Fig. 2A). Wider empirical confidence intervals (i.e., QIs) in community thresholds among the 5 assemblages were observed in response to % agriculture than to % developed land. Average thresholds for % agriculture differed among the 5 assemblages, but the QIs overlapped (Fig. 2B). Thresholds for diatom and fish assemblages occurred at substantially lower % agriculture than those for bird and benthic macro invertebrate assemblages (Fig. 2B). The community thresholds identified differed between ecoprovinces. For all assemblages, thresholds occurred at higher levels of % developed land and % agriculture in the Southern than in the Northern ecoprovince (Fig. 2A, B).

Figure 2.

Thresholds (0.05–0.95 bootstrap quantile intervals [QIs]) for coastal fish, invertebrate, bird, diatom and wetland plant community thresholds in response to % developed land (A) and % agriculture (B) in the watershed for the Northern and Southern ecoprovinces of the Great Lakes. QIs can be asymmetrical because of skewed distribution of bootstrapped replicates. Veg = vegetation, Inverts = macroinvertebrates.

Threshold identification

Confidence in the existence of community breakpoints was greatest for the % developed land stressor. This confidence was supported by the steepness of cumulative threshold frequency curves ( Fig. S1A, B (2013133_FigureS1.pdf)); the number of significant ( Fig. S2A (2013133_FigureS2.pdf)), pure, and reliable (i.e., consistently assigned to a particular group and with consistently significant IndVals across bootstrap replicates) indicators; QIs associated with individual taxa (Fig. 3A–E); and individual taxa-stressor relationships ( Fig. S3 (2013133_FigureS3.pdf)). However, birds in the Southern ecoprovince (Fig. 3I) and fishes in the Northern ecoprovince (Fig. 3A) were exceptions to this trend because the location of the change point was based on a very small number of reliable indicator taxa. In contrast to the sensitive-community responses, degraded communities exhibited more-gradual changes in response to stress ( Fig. S1A, B (2013133_FigureS1.pdf)).

Along the % agriculture stressor gradient in the Northern ecoprovince, 4 of the 5 assemblages exhibited threshold-consistent responses characterized by steep cumulative threshold frequency curves ( Fig. S1C (2013133_FigureS1.pdf)) and many reliable indicator taxa with narrow (0.05–0.95) QIs (Fig. 4A, C– E). The macroinvertebrate assemblage did not have a sufficient number of reliable indicator taxa and did not yield steep cumulative frequency curves characteristic of a typical threshold response (Fig. 4B,  Fig. S1C (2013133_FigureS1.pdf)). Along the % agriculture stressor gradient in the Southern ecoprovince, responses of all assemblages were inconsistent with the threshold model based on the criteria described above (Fig. 4F–J,  Fig. S2B (2013133_FigureS2.pdf)), possibly because the 2 stressors were confounded and because of the scarcity of relatively pristine watersheds (see Discussion).

No matches were found between stressor medians and community thresholds ( Table S2 (2013133_TableS2.pdf)), indicating that this method is relatively robust to the distribution of sites across the stress gradients. This situation is most obvious for % agriculture, where median values of the stressor were relatively consistent across assemblages, whereas thresholds were very different. This difference seemed a bit smaller for % developed land because of congruent thresholds, but large (e.g., 13 sites for diatoms) and inconsistent differences existed between sites with median % development and sites with % development at community threshold. Community thresholds tended to be lower than the median % development.

Figure 3.

Change-points (0.05–0.95 bootstrap quantile intervals [QIs]) for significant sensitive indicator fish (A, F), invertebrate (B, G), diatom (C, H), bird (D, I), and vegetation (E, J) taxa along the stressor gradient % developed land in the Northern (A–E) and Southern (F–J) province that contribute to threshold determination. Responses consistent with the threshold model are characterized by a large number of individual indicators and narrow QIs around their change points. Indicator value p-value cut-off = 0.05; purity and reliability cut-offs = 0.8. Identity of indicator taxa is not discussed in this paper. Note that representation of z+ (tolerant) indicator responses (grey), although not discussed in the context of this paper, is important for comparison of results across methods and systems. The scale reflects the entire overall range of the stressor gradients.

Figure 4.

Change-points (0.05–0.95 bootstrap quantile intervals [QIs]) for significant sensitive indicator fish (A, F), invertebrate (B, G), diatom (C, H), bird (D, I), and vegetation (E, J) taxa along the stressor gradient % agriculture in the Northern (A–E) and Southern (F–J) province. See Fig. 3 for explanation of figure.

DISCUSSION

The % developed land in a watershed at which community thresholds occurred was consistent across stress-sensitive taxa from 5 assemblage types: birds, fish, benthic macroinvertebrates, wetland vegetation, and diatoms. The approximate location of these thresholds along the stressor gradient agrees well with previous findings that changes in benthic macroinvertebrate and fish communities occur at low to very low (1–5%) levels of urbanization or impervious land cover associated with development (Wang et al. 2000, Stepenuck et al. 2002, Baker and King 2010, Hilderbrand et al. 2010, King et al. 2011, Utz and Hillderbrand 2011, Kail et al. 2012). Our threshold estimates for these 2 assemblages were slightly higher than 1 to 5% developed land, but still similar to published values given the length of the development gradient (0–92%).

Various approaches have been used to characterize community thresholds (e.g., Booth and Jackson 1997, Wang et al. 2000, Qian et al. 2004, Trebitz 2012). Utz et al. (2009) used cumulative frequency distributions to detect similar responses to development in stream macroinvertebrates. Gradient-boosted generalized linear models detected decreased macroinvertebrate richness in small streams draining watersheds with >3% impervious cover (Maloney et al. 2012). Threshold detection approaches have been a subject of recent controversy (Cuffney et al. 2011, Daily et al. 2012, but see Baker and King 2013), but the consistency of findings across methods, systems, and taxa is very encouraging and increases our confidence in the existence of a threshold response and its location along the stressor gradient. Documentation of such biological sensitivity to development has been used to justify the importance of creating protected areas in streams (Hilderbrand et al. 2010), a scenario that may need to be considered for coastal and lacustrine systems as well. This sensitivity of assemblages also emphasizes the need for additional research on local habitat factors that could mitigate watershed-scale anthropogenic stress and on interactions between local- and watershed-scale factors in light of climate-change scenarios (e.g., Wiley et al. 2010).

Our findings support the conjecture that anthropogenic stress, particularly in the form of development, in water-sheds exerts strong effects on ecological communities of coastal wetlands in the Laurentian Great Lakes (Brazner et al. 2007a, Niemi et al. 2011). In the Southern ecoprovince, strong responses to development were observed despite the negative correlation between development and agriculture, although absence of pristine habitats and their associated assemblages could have led to higher thresholds in the south than in the north. The mechanisms behind the effects of development on the resident communities may include pollutant and sediment discharge, increasing amount of impervious surfaces, and direct habitat destruction (reviewed in Allan 2004). A suite of factors acting on multiple scales (Brazner et al. 2007a) could have led to the observed congruence of thresholds across communities. Threshold congruence also could be caused by integration of the effects of watershed-scale stressors at the local scale. For example, watershed-scale degradation in water quality and disturbance of vegetation associated with increasing agriculture and development lead to consistent local-scale stressors on associated diatom (via water quality) and bird, fish, and macroinvertebrate communities (via habitat degradation). Aquatic and nearshore wetland vegetation is particularly important for persistence of wetland communities, and its destruction has cascading effects on wetland communities (Uzarski et al. 2009) by removing the structural matrix on which many assemblages rely (Diehl and Kornijów 1997, Kovalenko et al. 2012). King et al. (2011) argued that nonlinear responses are characteristic of assemblages facing novel stressors, such as those associated with changes in water quality or hydrology subsequent to development. Regardless of the mechanism, such congruence could indicate the presence of a true ecosystem threshold when changes in several biological assemblages occur around the same stressor value.

Communities could be less sensitive to % agriculture in the watershed than to the more-direct habitat disturbance associated with increasing development, especially if some mediating local factors, such as riparian buffers or other best-management practices, are in place. Many studies have demonstrated sensitivity of terrestrial birds and other taxa to habitat loss and fragmentation in developed landscapes. In some cases (Palomino and Carrascal 2007, Garaffa et al. 2009), species exhibited threshold responses to landscape degradation. However, these studies illustrated changes associated with displacement of habitat by agriculture or urbanization. The effects of agricultural activities on coastal wetlands may be more indirect and, therefore, less likely to be manifested as threshold responses. For example, DeLuca et al. (2004) documented threshold changes in marsh bird assemblages of Chesapeake Bay in response to human land use, but the effects were seen at local scales, not at the watershed scale. Agricultural stressors act through a variety of mechanisms, including sediment and fertilizer run-off and nonpoint source pollution (reviewed in Allan 2004). Previous studies also showed a wide range of responses, absence of clear community thresholds, or changes that were noticeable only when most of the land was converted to agriculture (e.g., Wang et al. 2000, Allan 2004, Riseng et al. 2011). In the case of our data, the absence of truly pristine locations in the Southern ecoprovince, a consequence of a long history of agricultural land use, and covariation of the developmental and agricultural stress gradients may have contributed to absence of threshold responses to stress in associated communities.

The interprovince differences in stress thresholds in our study correspond with biogeographical differences in associated communities and their sensitivity to the stress gradient. For instance, diatom assemblages in the Southern ecoprovince are tolerant to the contemporary environmental stress in the region, such as higher nutrient concentrations, so a threshold was observed at a higher level of stress than for the Northern ecoprovince. Hanowski et al. (2007) noted that the composition of coastal wetland bird communities differs substantially between Northern and Southern ecoprovinces of the Great Lakes. Other investigators have found large differences in macroinvertebrate communities in rivers on and off the Precambrian shield (Neff and Jackson 2011) and in catchments across mountain, piedmont, and coastal physiogeographies in the Mid-Atlantic region (King et al. 2011). Human land use is often correlated with natural latitudinal gradients (reviewed in Allan 2004), but for the within-ecoprovince analyses, we demonstrated that land use had an overriding effect on community thresholds when agricultural land use and latitude were significantly correlated. In an analysis of communities along the US coast of Lake Superior, where latitudinal variation is minimal, Niemi et al. (2011) showed that at least some of the observed variation was associated with gradients of agricultural and urban development.

We are unaware of other analyses of congruence in community threshold responses, so we are unable to compare our results to patterns in other systems. However, the threshold congruence patterns observed in our study run contrary to previously reported diversity trends among different taxa. Most investigators have shown discordance in the responses of different taxonomic groups (Hirst et al. 2002, Johnson et al. 2006, Brazner 2007a, Lewis et al. 2007, Johnson and Hering 2009). Fahrig (2001) simulated threshold responses of birds and other organisms to habitat loss, but she expected these thresholds to be asynchronous among species and species groups. Stralberg et al. (2009) predicted that varying responses of bird species to climate change would lead to novel community assemblages, and this lack of congruence would be expected to generate even greater differences in multitaxon assemblages. In a recent review, Heino (2010) reported low correspondence in diversity among various taxonomic groups across all freshwater systems and taxa, a result that limits application of cross-taxon congruence in conservation (i.e., use of certain assemblages as a proxy for others). Similarly low concordance in biodiversity was observed in a meta-analysis of biodiversity trends among terrestrial taxa (Flather et al. 1997). The difference between community thresholds and biodiversity trends might be explained by the fact that aggregate metrics, such as biodiversity, are confounded by synchronous changes in sensitive and tolerant taxa (King et al. 2011, Baker and King 2013).

Application of thresholds in management is complicated by uncertainties associated with their predicted location along the stressor gradient and, in most cases, limited understanding of the mechanisms underlying a system's response (Groffman et al. 2006, Dodds et al. 2010). The most promising warning signals of ecosystem thresholds are increasing autocorrelation (Scheffer et al. 2009) and variance (Carpenter and Brock 2006) in the system's pattern of fluctuations over time. Elucidating such patterns requires long-term data to distinguish proximity to a threshold from natural variability in system's response (Andersen et al. 2009). Thresholds associated with community shifts can be especially difficult to detect because of inherent variability in community composition and problems with detectability of less-common species. In our case, threshold similarity across different assemblages increases our confidence in its location. On the other hand, despite a relatively strong concordance in responses of different taxa, uncertainty associated with the response of each taxon (5–95% QIs) leads us to recommend a multitaxon approach rather than use of a particular assemblage as an indicator for all other groups to increase confidence in identifying environmental thresholds. This recommendation is consistent with those of many assessment programs (e.g., Barbour et al. 1999).

Changes in assemblage composition in response to development stress were more consistent with the threshold model than were changes in response to stress associated with agriculture. Nevertheless, with its limitations explicitly acknowledged, our approach can be used to infer the extent of environmental disturbance that is likely to cause significant changes in community composition. Therefore, this approach can be used to inform watershed-restoration and landuse-planning decisions. Such information is especially valuable for identifying sites at greatest risk of significant change. Sites at risk that have not yet approached the threshold could be designated as priority sites for management intervention. Concordant changes in several communities emphasize the risk of functional changes in ecosystems and the need for management action. Furthermore, concerns regarding concurrent changes in several assemblages, some more charismatic than others, are easier to convey to stakeholders than concerns about a precipitous decline in a particular group for which public awareness may be nearly nonexistent.

Our analyses of Great Lakes coastal wetlands indicate that a relatively large number of sites are subject to levels of anthropogenic stress above that necessary to elicit a threshold response. This conclusion is not surprising, especially in the Southern ecoprovince, because of high population density, a long history of settlement and agriculture, and anthropogenic pressure on the Great Lakes coastal ecosystem. A combination of our results and the entire set of Great Lakes watershed-level stress scores ( http://beaver.nrri.umn.edu/gleidb/gleidb/gdb_main) can be used to identify areas that must be prioritized for management intervention to sustain biological communities associated with minimally degraded conditions. These results could be used to make long-term projections of watershed-scale changes on resident biological communities.

In summary, we detected congruent, marked changes in bird, fish, benthic macroinvertebrate, wetland plant, and diatom assemblage composition in response to watershed development across the Great Lakes coastal margin. Cross-taxon responses to agricultural stress were less congruent but still overlapped. Our results indicate that watershed-scale anthropogenic stress leads to pronounced simultaneous changes in many biotic assemblages, with potential for nonlinear functional changes in aquatic ecosystems.