Coastal wetlands are important contributors to large-lake productivity and biodiversity and mediators of lake—watershed interactions. This study explores whether the size of the watershed in which coastal wetlands are embedded (a measure of strength of connection to the terrestrial landscape) influences their background condition and response to anthropogenic landuse. Water quality, substrate, vegetation structure, and composition of zoobenthos, turtles, crayfish, and fish were characterized in 32 Lake Superior coastal wetlands in the summers of 2000–2001, and related to watershed size categories via ANOVA and to watershed development (percent agricultural and urban landuse) via linear regression. Lake Superior coastal wetlands had relatively low levels of watershed development – apparently not enough to significantly alter fish composition. However, watershed development was associated with significant changes in substrate, turbidity, plant structure, and zoobenthos, and in most cases these effects were stronger in wetlands having big rather than small watersheds. An alternate classification contrasting exposure to versus protection from river influences was not effective at resolving responses to watershed development. Watershed size had little effect on background conditions in minimally disturbed wetlands, although turtles were more abundant in large-watershed wetlands. The role of watershed size in mediating responses to landuse merits further study, but our findings suggest that receiving inflows from bigger watersheds affects coastal wetlands primarily by amplifying transmission of disturbance rather than via direct impacts of flow.
Hydrological connections to the adjacent lake are a defining feature of coastal wetlands in the Laurentian Great Lakes (Wetzel, 1990; Brazner et al., 2000; Bouvier et al., 2009). In contrast, hydrological connections to the terrestrial watershed are highly variable among coastal wetlands, with some completely lacking tributaries or even a definable catchment (e.g. fringing wetlands), while others have large watersheds and substantial river inflows. The magnitude of tributary inputs relative to lake inputs is known to organize patterns in coastal wetland morphology, water chemistry, and biota (Keough et al., 1999; Trebitz et al., 2005; Morrice et al., 2009). Furthermore, variability among wetlands in tributary inflows is much larger than variability in seiche-driven lakewater inflows (Trebitz et al., 2009). Watershed size, which substantially dictates tributary size (Richards, 1990; Johnston and Shmagin, 2008), could therefore be an important factor in mediating the extent to which anthropogenic alterations of watersheds affect downstream wetlands (DeCatanzaro and Chow-Fraser, 2011). Prior research has demonstrated consistent associations between coastal wetland conditions and anthropogenic landuse when characterized simply on a percent of watershed basis (Crosbie and Chow-Fraser, 1999; Lougheed et al., 2001; Uzarski et al., 2004; Trebitz et al., 2007b, 2009), and points to a series of linked effects first on water quality, then on primary producers and habitat structure, and finally on composition of wetland fauna. Accounting for watershed size might further resolve understanding of how watershed landuse changes affect coastal wetlands.
Here, we take advantage of data spanning a suite of coastal wetlands on Lake Superior to examine the role of watershed size in influencing wetland habitat and biota. We ask whether watershed size influences wetland background conditions (i.e. in the absence of disturbance) and whether the response to watershed disturbance differs in strength among watershed size categories. Disturbance is measured by the percent of the watershed in agricultural or urban land uses (percent development). Having a large number of sites in a single lake minimizes the biogeographic effects that have confounded investigations of disturbance in cross-basin studies of the Great Lakes (Lougheed and Chow-Fraser, 2002; Brazner et al., 2007a). The habitat and biota endpoints examined represent various wetland attributes expected to be directly or indirectly responsive to watershed and river size (e.g. wetland morphology) or to landuse alteration in the watershed (e.g. water quality, plant structure, fish and zoobenthos composition), as well as a few for which this relationship is still untested (substrate type, crayfish, some zoobenthos and turtle endpoints).
This study included 32 coastal wetlands along Lake Superior's southern shore (Figure 1). Sites were identified by scanning maps and aerial imagery for wetlands capable of providing habitat for Great Lakes fishes (significant inundated area, surface connection to Lake Superior) that were safely accessible with a small boat. Along the eastern part of the basin, almost every suitable wetland was sampled, while along the more wetland-rich western shoreline, a subset of wetlands was chosen to cover a range of adjacent land uses. In a few locations, more than one wetland within a complex was sampled (lower St. Louis River, Bark Bay area, Sturgeon River area). We actually sampled 33 coastal wetlands but excluded one from analyses because its watershed was much more developed than the others, giving it undue weight in analyses. However, the relatively low levels of watershed development across the other 32 wetlands reflect conditions in Lake Superior rather than sampling design (i.e. we did not avoid sampling in more developed watersheds).
Wetland morphology and watershed characteristics were established via geospatial analyses. Wetland shorelines were digitized from orthophotographs and used to compute inundated area. Watersheds were digitized with reference to hydrographic and elevation data layers, then intersected with year 2001 National Land Cover Data (NLCD; Homer et al., 2004) to obtain the distribution among undeveloped (water, forest, shrub, wetland, grassland, bare soil), agriculture (pasture, cultivated), and urban (low, medium, or high-intensity development) NLCD categories. Landcover was summarized as percent developed land within watersheds (agriculture + urban combined). Each wetland had its own watershed, although in the lower St. Louis River, the watersheds of the more downstream wetlands include that of the more upstream ones. Wetlands were divided into a small and a big watershed size group using a small gap in the size distribution at 9.0 × 107 m2 as the threshold (Figure 2A). Wetlands were also assigned to two hydromorphic classes, intended to capture the degree of exposure to or shelter from river influences. Wetlands whose inundated area was mainly along a central river channel were classified as high river exposure, whereas wetlands having only small tributaries (≤ 2d order) or having substantial off-channel areas were classified as low river exposure (based on orthophotographs).
Data collection and summarization
Sampling took place in July and August of either 2000 or 2001. Three locations were sampled per wetland–in vegetated, off-channel areas when available–choosing two suitable for large-frame fyke nets and one suitable for small-frame nets. The exception was Galloway wetland, which only had enough room for two net sets. Habitat characteristics recorded at each location included water temperature, conductivity, and turbidity (Hydrolab® multiprobe), slope of the bottom (inverse of distance offshore of the 1 m depth contour), submerged plant cover (%) and species richness, and the number (count) of substrate types present (possibilities were sand, silt, muck, gravel, cobble, wood, detritus, or boulder). Habitat measures were averaged across sampling locations within wetlands.
Zoobenthos were collected with funnel traps (2 arrays per sampling location) deployed open-side down on the sediment surface for 24 h. Each array consisted of three 15 cm diameter glass funnels inserted into collection jars held in a plexiglas frame. Jar contents were filtered through 50 μm mesh, preserved with buffered formalin, and counted under a dissecting microscope. Crustaceans (cladocerans and copepods) were identified to species where possible, while other taxa were identified to genus. Zoobenthos were pooled across samples within wetlands and summarized as total catch, total richness, Shannon-Wiener diversity, rotifer richness, crustacean richness, percent of catch as crustaceans or as rotifers, ratio of crustaceans to rotifers, and ratio of cladocerans to copepods. Zoobenthos endpoints focused on rotifers and crustaceans because these made up most of the catch.
Fish, turtles, and crayfish were collected with fyke nets deployed for 48 h and tended daily. Nets were set in pairs in lead-to-lead configuration parallel to shore, with large-frame nets (0.9 × 1.2 m frame, 5 and 13 mm mesh) at ∼1.0 m water depth and small-frame nets (0.45 × 0.9 m, 5 mm mesh) at ∼0.4 m depth. Crayfish and turtles were enumerated to species; life-stage was also recorded for fish (young-of-year, juvenile, or adult; based on size ranges in Becker, 1983). Catches were pooled across nets within wetlands and summarized as total crayfish, turtle numbers by species, total fish numbers, and fish richness and Shannon-Wiener diversity. Fish composition was also expressed as the percentage of the catch consisting of individuals from various life-history categories that had shown responsiveness to watershed development in an earlier study (Trebitz et al., 2009), namely nest-guarding species, species that are large-bodied (as adults), and species that are turbidity tolerant or turbidity intolerant. Life-history classifications were derived using Becker (1983), Barbour et al. (1999) and Trebitz et al. (2007a).
Our central question is whether watershed size influences the response of coastal wetlands to watershed landuse. We used linear regression to relate wetland habitat and biota (response variables) to watershed development (the predictor variable), with regressions run separately for each watershed size group. Non-overlapping confidence intervals for significant (p ≤0.05) slope coefficients or slopes that were significant for one group but not the other were taken as evidence for differing response rates between groups. Both watershed size groups had similar N and comparable range of watershed development (0–30%), ensuring the validity of comparing regressions between them. Since exposure to river influences might also affect sensitivity to watershed conditions, significant regressions were repeated using exposed vs. protected as the groups evaluated. To provide context for understanding responses to watershed development, we also asked whether watershed size or river exposure influence wetland conditions independent of disturbance. To do so, we used ANOVA to compare habitat and biota endpoints between watershed size or river exposure groups for the subset of wetlands having minimal (<10%) watershed development. All these analyses were run using SYSTAT 9 software.
To help evaluate any watershed responses observed, we wanted to compare the present fish data to data from a basin-wide study (Trebitz et al., 2009) that evaluated many of the same endpoints. For this comparison to be valid, we needed to establish similarity between fish composition based on fyke-net data (present study) and electrofishing data (previous study). To do so, we visually assessed the match achieved when overlaying the present Lake Superior wetlands on a non-metric multi-dimensional scaling ordination of wetlands from Trebitz et al. (2009). This overlay was done using the NMS-scores procedure in PC-ORD (McCune and Mefford, 2006), which is analogous to fitting a regression model with one data set, then using it to examine predictions for an independent data set. Relative abundance matrices of by fish species by wetlands were used as input to these analyses. We also computed the correlation between percent watershed development and the ordination scores and plotted these as vectors on the ordination as an indicator of gradient direction and strength.
Before proceeding with statistical analyses, we screened for collinearity among candidate response variables with Pearson correlation analysis. Variables dropped due to excessive collinearity (correlation >0.75) were turtle total number (retaining numbers of the two primary species), zoobenthos total number and richness (retaining number and richness of crustaceans and rotifers), and % rotifers (retaining % crustaceans). There was no strong collinearity among habitat variables (largest correlation was 0.56–plant richness vs. cover). Since linear regression and ANOVA assume normality and homoscedasticity of the residuals (Kéry and Hatfield, 2003), plots of residuals versus response variable values were examined for indications of non-normality and heteroscedasticity (e.g. fan-shaped or curved rather than level residual plots, Draper and Smith, 1981). Percent intolerant fish was dropped from analysis because most wetlands had very few, leading to heteroscedastic residuals. Watershed size was right-skewed over several orders of magnitude and was log10-transformed to attain normality; the other variables were not excessively skewed and were left untransformed.
Wetland habitat, biota, and watershed characteristics
The 32 coastal wetlands sampled spanned the full east-west extent of Lake Superior's southern shore (Figure 1) and a substantial gradient in watershed size (Figure 2A). Eleven wetlands were classified as exposed to river influence (area mostly along a single central channel), whereas 21 were classified as protected from river influences (substantial side-channel or back-bay areas). The ratio of exposed to protected wetlands was similar across watershed size groups (Figure 2A) and the number of each type was similar among the minimally-disturbed wetlands (9 exposed, 11 protected). Two-thirds of the study wetlands had <10% watershed development, all had <30% watershed development (Figure 2B), and development that was present tended to be low intensity (e.g. hay/pasture rather than row-crop cultivation, lower-intensity NLCD categories of urbanization). This is in distinct contrast to coastal wetlands elsewhere in the Great Lakes, where watershed development levels range much higher (Figure 2B) and includes more intense agricultural and urban landuse. The study wetlands generally had clear water (mostly <20 NTU turbidity) and supported structurally diverse aquatic vegetation (mostly >10 species and >40% cover; Table 1).
Nine phyla of zoobenthos were recovered with funnel traps, including annelids, arthropods, chordates, coelenterates, flatworms, gastrotricha, rotifers, mollusks, and nematodes. Out of the ∼150 taxa identified (typically to species or genus), rotifers and crustaceans were by far the most abundant, making up ∼94% of all zoobenthos organisms counted. The crustaceans tended to be vegetation-associated taxa such as chydorids, whereas the rotifers were a mix of vegetation-associated and planktonic taxa. The most common rotifer genera (found at >50% of wetlands) were Asplanchna, Cephalodella, Conochilus, Euchlanis, Keratella, Lecane, Lepadella, Lophocharis, Monostyla, Notholca, Ploesoma, Polyarthra, Synchaeta, Testudinella, Trichocerca, and Trichotria. The most common crustaceans (>50% occurrence) were harpacticoids, ostracods, and the genera Alona, Bosmina, Camptocercus, Ceriodaphnia, Chydorus, Diaphanosoma, Eucyclops, Eurycercus, Latona, Macrocyclops, Microcylops, and Simoce- phalus.
Fyke nets were effective for capturing crayfish and turtles as well as fish. Crayfish were caught in 73% of wetlands, in numbers ranging from 1 to 95 (Table 1), and were essentially monospecific Orconectes virilis (virile crayfish). Painted Turtles (Chrysemys picta) and Snapping Turtles (Chelydra serpentina) were found in 84% and 51% of wetlands, respectively, while a single wood turtle (Glyptemys insculpta) was caught in the entire study. The proportion of catch represented by Painted rather than Snapping Turtles (85%) was similar to that found by DeCatanazaro and Chow-Fraser (2010). We caught fifty-four species of fish, with richness ranging from 5 to 29 per wetland (Table 1). The most common species were Yellow Perch, Rock Bass, White Sucker, Pumpkinseed Sunfish, Northern Pike, Johnny Darter, Golden Shiner, Spottail Shiner, And Brown Bullhead (species names in Table 2). Other species that sometimes dominated the catch (>25% at any one wetland) were Common Shiner, Logperch, Black Crappie, Common Carp, Bluntnose Minnow, Emerald Shiner, Trout-Perch, Threespine Stickleback, and young-of-year Bullheads or Sunfish (Table 2).
Fish species composition was broadly consistent with that found in Lake Superior wetlands in a different, Great Lakes wide study (Trebitz et al., 2009) that included some of the same sites. We assessed this by overlaying fish composition from the present study on an ordination computed previously, analogous to fitting a regression with one data set, then using it to make predictions for an independent set of sites. Lake Superior wetlands from both data sets occupied the same general region of the ordination space (Figure 3), indicating that fish composition is broadly comparable despite differences in gear (fyke-netting vs. electrofishing). Gear performance differences (Reutz et al., 2007) no doubt contribute to species-level composition differences between the studies. However, we take the broad concurrence to mean that is it valid to compare responses of aggregated composition endpoints (e.g.% turbidity tolerant fish) between the studies.
Relationships to watershed size, river exposure, and watershed development
Twenty-one wetlands with <10% watershed development were included in analyses asking whether background conditions differed among watershed size groups. We found no significant ANOVA effects of watershed size on any wetland morphology or habitat endpoints (Table 1). Wetland inundated area did increase linearly with watershed size (R2 = 0.35, p<0.01, Figure 4D), but the ANOVA was not quite significant (R2 = 0.15, p = 0.08). There were significantly more Painted Turtles and Snapping Turtles in big-watershed wetlands (Table 1, Figure 4A-B). We did not find significant differences between watershed size groups for any fish endpoints, and only 1 of 7 zoobenthos endpoints differed among size groups (crustacean to rotifer ratio, Table 1, Figure 4C).
All 32 coastal wetlands were included in analyses examining responses to watershed development and how they differed between watershed size groups. We had anticipated watershed development responses for 14 endpoints based on Great Lakes literature, but found a significant response for only 8 of them (failing to do so for 2 water quality and 4 fish endpoints, Table 1). We also tested some endpoints (substrate variety, crayfish number, several zoobenthos endpoints) for which relevant literature was absent or equivocal, but found no significant responses to watershed development for any of these (Table 1). Among endpoints having a significant response to watershed development, the relationship was stronger in big-watershed wetlands in 5 cases (Table 1). Turbidity increased more steeply with development in big-watershed wetlands, and substrate variety, plant cover, plant richness, and crustacean richness responded significantly only in the big-watershed group (Table 1, Figure 5A-E). However, the zoobenthos diversity response was similar in both size groups (Figure 5F), and percent crustaceans responded significant only in the small-watershed group (Figure 5G). Percent turbidity tolerant fish also responded significantly only in small-watershed wetlands (Table 1), a trend which appeared to be heavily influenced by one wetland (Figure 5H) in which the catch was 82% carp (carp were caught in ∼1/3 of wetlands, but nowhere else in such numbers, Table 2). The endpoints for which the response to watershed development depended on watershed size were different from the endpoints for which watershed size influenced background conditions (compare linear regressions to ANOVAs; Table 1).
Exposure to river influences was not useful in resolving wetland patterns. Among minimally disturbed wetlands, we found no significant ANOVA differences between the exposed and protected groups for any habitat or biota endpoints. Of the 8 endpoints that had significant relationships to watershed condition when grouped by watershed size, only three continued to be significant when grouped by river exposure, and counterintuitively, it was protected rather than exposed wetlands where the response was significant (turbidity, plant richness, zoobenthos diversity). We attribute this finding to having more power to detect development effects in the protected (N = 21) than the exposed group (N = 11).
Effect of watershed size on minimally disturbed wetlands
We had expected physical impacts of watershed size on wetlands even in the absence of disturbance, since bigger watersheds have larger tributaries, with more power to carve out basins, move sediments, and physically disturb vegetation. Not finding such effects in ANOVAs comparing watershed size groups was therefore somewhat surprising, although our protocol of sampling backwater rather than channel areas probably helped minimized river flow influences (e.g. on vegetation – Lougheed et al., 2001). Lack of effects on water quality endpoints (during summer baseflow) is consistent with DeCatanzaro et al. (2009), who reported no watershed size effect on conductivity or water clarity in relatively undisturbed coastal wetlands in Georgian Bay, Lake Huron. We had not expected an effect of watershed size on faunal endpoints. Larger systems cumulatively support more habitat types and more organisms (Tonn and Magnuson, 1982; Findlay and Houlahan, 1997) and wetland size did increase with watershed size, but we sampled all wetlands with a constant level of effort which should minimize any such patterns. The reason for the higher catch of Painted and Snapping Turtles in wetlands with big watersheds is unclear.
Effect of watershed condition on wetlands
Conversion of watersheds from largely natural landcover to agricultural and urban (developed) landuse increases the concentration of nutrients and sediments in runoff (Hopkinson and Vallino, 1995; Chen and Driscoll, 2009; Han et al., 2010). Responses to landuse changes in Great Lakes coastal wetlands generally parallel those observed in other types of aquatic systems (e.g. Smith, 1998; Allan, 2004) – increasing nutrient and sediment inputs stimulate planktonic algae growth, decrease water clarity, degrade aquatic vegetation structure, and ultimately alter the composition of macroinvertebrates and fish (Crosbie and Chow-Fraser, 1999; Lougheed et al., 2001; Albert and Minc, 2004; Uzarski et al., 2004). We had, therefore, expected that habitat and biota would significantly respond to watershed development in our study. We did indeed find such relationships for habitat and zoobenthos, but were surprised by the scarcity of significant effects on fish. Out of 5 fish endpoints that had been significantly related to watershed landcover in a Great Lakes-wide study (Trebitz et al., 2009), only one showed a response to watershed condition in these Lake Superior wetlands.
We think there are several reasons why watershed condition effects on fish were difficult to resolve. The stressor gradient in Lake Superior is small relative to that across the Great Lakes (Figure 2B), and consequently the power of watershed development to alter fish composition may be limited (compare vector magnitudes, Figure 3). Significant natural differences across lakes or latitude are superimposed on the Great Lakes-wide stressor gradient (e.g. Brazner et al., 2007a) and strengthen apparent stressor effects across lakes relative to within lakes. For example, fish composition depends in part on thermal regime, which is perhaps incidentally correlated to watershed development across the Great Lakes because both vary with latitude. Also, the occasionally large numbers of young-of-year caught in fyke nets increased background variability in fish composition (compared to electrofishing data), making a landuse signal harder to detect. Some way of down-weighting young-of-year fish in calculating composition endpoints would be desirable, but is beyond the scope of this study.
The literature is more limited regarding landuse effects on other biotic groups. Responses for several zoobenthos endpoints from our funnel-traps matched those reported from sweep-net data (Uzarski et al., 2004; Table 1). We tested additional zoobenthos endpoints that Uzarski et al. could not report on (because sweep-nets do not effectively retain rotifers and enumerating a fixed number of organisms as they did yields no abundance information) but these did not appear sensitive to watershed development. Lougheed and Chow-Fraser (2002) found zooplankton (not zoobenthos) numbers positively related to water quality changes that typically accompanying watershed disturbance (e.g. decreased water clarity), but did not examine landuse directly. Herpatofauna richness declined with loss of forest cover around inland wetlands (Findlay and Houlahan, 1997) but our turtle catches were not sufficiently speciose to examine richness as an endpoint. Roads can represent migration barriers and mortality sources for turtles, but have little direct effect on the more aquatic species (Gibbs and Shriver, 2002) such as Painted Turtles. DeCatanzaro and Chow-Fraser (2010) attributed positive relationships between coastal wetland turtle numbers and road density to eutrophication-related increases in food, so perhaps we failed to find a trend in turtle numbers because levels of development were not sufficient to make any of the wetlands eutrophic. We could find no literature on crayfish in coastal wetlands beyond documentation of their occurrence (e.g. Simon and Thoma, 2006), so their response to landuse is unknown. Our results do not suggest that crayfish or turtle abundances are sensitive to watershed development, but since they are readily sampled with fyke nets, examining their response across a larger stressor gradient before ruling them out as indicators is warranted.
The role of watershed size in the response to watershed landuse
Most studies of landuse effects on coastal wetlands have simply related wetland conditions to watershed landuse percentages without examining watershed size at all. One study that did examine watershed size (via regression tree analysis, Brazner et al., 2007b) found it a significant co-factor for anthropogenic disturbance effects on various biotic groups. Our finding that rates of response to watershed development can differ among watershed size classes extends this work beyond biota themselves to habitat endpoints (e.g. water quality, vegetation structure, substrate) that would mediate such effects. In fact, interacting watershed size – watershed landuse responses may be most readily apparent for habitat endpoints (as we found), whereas effects on biota may be harder to resolve because they are further removed in the causal pathway and have more complex dynamics.
Why might we expect more prominent responses to landuse in coastal wetlands having bigger watersheds? One reason is that the potential for lake-water dilution of runoff is diminished with larger tributary flows (Trebitz et al., 2002; Morrice et al., 2009). Another is that bigger watersheds accumulate more runoff and thus deliver greater loads of sediments and nutrients even if concentrations are similar (Richards, 1990; Kurtz et al., 2006). While this sounds compelling, we know of no study establishing loading as a predictor of biotic composition in coastal wetlands, whereas studies have consistently found associations with stressor concentration (Lougheed et al., 2001; McNair and Chow-Fraser, 2003; Cooper et al., 2006; Reavie, 2007; Seilheimer et al., 2009). Landuse effects may be harder to detect in small watersheds because spatial arrangement becomes increasingly important in dictating the ability for runoff to reach downstream locations (Strayer et al., 2003; Fraterrigo and Downing, 2008). Nevertheless, assessment of landuse over whole watersheds rather than over more local scales (e.g. some fixed buffer distance from the waterbody) has generally had the best predictive capability in past studies (Mensing et al., 1998, Strayer et al., 2003, Brazner et al., 2007b). Bigger watersheds may have physical impacts (scouring, etc.) under disturbance that are not evident in the natural state, as conversion to agriculture and urban landuse tends to reduce assimilative capacity and increase flashiness of stream flows (Verry, 1986; Richards, 1990; Hopkinson and Vallino, 1995).
Factors besides size that produce flow differences among watersheds include topographic relief, stream density, soil type, and water storage capacity (Fraterrigo and Downing, 2008). The potential for such differences to affect coastal wetlands is illustrated by Detenbeck et al. (2006), who found that minimally disturbed wetlands fed by streams with flashy hydrology had higher turbidity and nutrient levels than those fed by streams with stable hydrology. Such landscape characteristics contribute substantially to variation in hydrologic regime across the Great Lakes basin but their variability across the Lake Superior south shore is only minor (Richards, 1990; Johnston and Shmagin, 2008), and thus should be of little importance for our data set.
Conclusions and implications for wetland assessment and management
Watershed development levels were relatively low for the 32 Lake Superior coastal wetlands we studied, yet nevertheless associated with significant changes in habitat and biota. The development effect tended to be more prominent in wetlands having big rather than small watersheds. Our findings have several implications for wetland assessment and management. One is simply that in Lake Superior, an emphasis on protecting relatively unimpacted wetlands and watersheds as well as on restoring degraded ones is warranted. Preventing adverse environmental impacts can be much less expensive than remediating them, but the regulatory, policy, and funding tools available are often stronger for remediation than for protection. We recognize that “unimpacted” is a relative term; there are legacy impacts (e.g. historic logging; Fizpatrick et al., 1999) that still affect present-day conditions in Lake Superior wetlands, and protection efforts will not restore coastal wetlands that have already been lost.
Our work reinforces previous observations that systematic land-use differences across latitude and lake in the Great Lakes basin (Niemi et al., 2007) impact the spatial scale at which management actions and ecological understanding can be developed. Working with data for wetlands from a single Great Lake eliminated concerns about biogeographic confounding of stressor-driven patterns that have been an issue in cross-Great Lake studies (e.g. Lougheed and Chow-Fraser, 2002; Brazner et al., 2007a). However, this also meant that the anthropogenic stressor gradient was reduced relative to that present over multiple lakes (Niemi et al., 2007), which probably contributed to our finding relatively few significant responses for biotic endpoints. Lake-specific indicators may be desired by managers or seem appropriate when endpoints exhibit important geographic patterns, but short stressor gradients make developing and calibrating lake-specific indicators challenging regardless of whether the lack is heavily disturbed sites (in Lake Superior) or minimally disturbed sites (e.g. in Lake Erie).
Watershed size may be an important classification variable in resolving responses to land-use changes, as such relationships were often stronger in wetlands in our study having big watersheds rather than small watersheds. Watershed size, however, had little influence on background expectations for habitat structure and aquatic fauna, which simplifies indicator development. Although the mechanisms are not yet clear, finding that watershed size appears to interact with landuse development rather than having independent effects suggests that bigger watersheds and tributaries affect coastal wetlands primarily by amplifying stressor delivery rather than through direct impacts of flow. Regulating landuse intensity without considering watershed size may be insufficiently protective of some coastal wetlands, but additional work on how watershed size mediates sensitivity to landuse stressors is warranted.
We thank Matthew Starry for additional GIS processing, and Michael Sierszen, John Morrice, and Jack Kelly for discussions and feedback on various aspects of this work. Comments from two anonymous reviewers helped to improve the manuscript. Although this work was fully funded by the U.S. Environmental Protection Agency, the views expressed are those of the authors and do not necessarily reflect the views or policies of the Agency.
This article not subject to United States copyright law.