Extent of Sedge-Grass Meadow in a Lake Michigan Drowned River Mouth Wetland Dictated by Topography and Lake Level

Water-level fluctuations are critical in maintaining diversity of plant communities in Great Lakes wetlands. Sedge-grass meadows are especially sensitive to such fluctuations. We conducted vegetation sampling in a sedge-grass dominated Lake Michigan drowned river mouth wetland in 1995, 2002, and 2010 following high lake levels in 1986 and 1997. We also conducted photointerpretation studies in 16 years dating back to 1965 to include responses to high lake levels in 1952 and 1974. Topographic data were collected to assess their influence on areal extent of sedge-grass meadow. Dominant species in short emergent and submersed/floating plant communities changed with water availability from 1995 to extreme low lake levels in 2002 and 2010. Sedge-grass meadow was dominated by Calamagrostis canadensis and Carex stricta in all years sampled, but Importance Values differed among years partly due to sampling in newly exposed areas. Photointerpretation studies showed a significant relation between percent of wetland in sedge-grass meadow and summer lake level, as well as the number of years since an extreme high lake level. From the topographic map created, we calculated the cumulative area above each 0.2-m contour to determine the percent of wetland dewatered in select years following extreme high lake levels. When compared with percent sedge-grass meadow in those years, relative changes in both predicted land surface and sedge-grass meadow demonstrated that accuracy of lake level as a predictor of area of sedge-grass meadow is dependent on topography. Our results regarding relations of plant-community response to hydrology are applicable to other Great Lakes wetlands.


Introduction
The Laurentian Great Lakes system in North America consists of Lakes Superior, Huron, Michigan, Erie, and Ontario, as well as their connecting channels and the St. Lawrence River, which connects the lakes to the Atlantic Ocean. The lake surface of the basin covers about 244,000 km 2 , The relations between water-level fluctuations and plant communities of Great Lakes wetlands have been extolled for decades (e.g., Keddy and Reznicek 1986;Hudon 1997;Wilcox 2004;Wilcox and Nichols 2008;Keddy and Campbell 2020;Smith et al. 2021). High lake levels cause die-back of upland and canopy-dominating wetland species, and succeeding low lake-level periods expose sediments and allow regrowth of smaller-statured wetland plants from seeds or propagules. Over the past nearly 5000 years, water levels on Lake Michigan-Huron (one lake hydrologically) have fluctuated on a quasi-periodic cycle (Baedke and Thompson 2000;Argyilan et al. 2018). Large, longer-term fluctuations from high to low lake levels recur about every 160 years, with fluctuations of about 30-33 years occurring over this same time period. The latter fluctuation pattern is responsible for much of the vegetation change (Wilcox 2004).
Models have been created that demonstrate the relation of such hydrological fluctuations on vegetation. Wilcox and Nichols (2008) modeled the response of a bulrush marsh in Saginaw Bay of Lake Huron to the reduction of water levels following an extreme high in 1986. The model made use of plant community data collected along transects that followed elevation contours with specific water-level histories to make predictions of plant community change as a function of meters above water vs. number of years out of water. Keddy and Campbell (2020) made use of information reported in the literature to model relative marsh area vs. duration of dewatering in relation to invasion by and flooding-out of woody species.
The plant community most affected by water-level fluctuations is sedge-grass meadow (SGM). The SGM occurs along the higher elevation fringe of the wetlands and, in relation to flooding and dewatering, is in competition with emergent vegetation at the lower extent and woody vegetation at the upper extent (Keddy and Reznicek 1986;Keddy and Campbell 2020). Gathman et al. (2005) sampled wetland plant communities in northern Lake Huron in 1996 when water levels were above average, 1997 with high water levels, and 1998 when levels were back near those of 1996. Stem counts of dominant sedge and grass species in the wet meadow decreased by > 70% from 1996 to 1997 and increased in 1998, but only by 16%. Their study did not capture changes that would have occurred by 1999, when lake levels decreased an additional half meter, but we expect that SGM increases were much greater. They also did not collect data on areal extent of SGM nor on elevations. Werner and Zedler (2002) and Peach and Zedler (2006) addressed microtopographic surface area in SGM provided by Carex stricta tussocks but not wetland topography or surface area.
On Lake Ontario,  and Wilcox and Bateman (2018) assessed changes in SGM in relation to water levels using photointerpretation of aerial photographs. Wilcox and Xie (2007) added topographic relief data to model response of SGM to water-level fluctuations. However, Lake Ontario water levels have been regulated since about 1960 when the St. Lawrence Seaway began operation (Hudon et al. 2006;Wilcox and Xie 2007), so the results do not necessarily reflect natural Great Lakes processes. Although the Keddy and Campbell (2020) model imposed the concept of areal extent, it did not contain numerical area data, which they acknowledged would require site-specific topography.
As part of three studies at Arcadia Marsh on the eastern shore of Lake Michigan, we had the opportunity to sample changes in SGM and other plant communities in 1995SGM and other plant communities in , 2002SGM and other plant communities in , and 2010 related to water-level decreases from extreme lakelevel highs in 1986 and 1997 ( Fig. 1). Observations made prior to the last study in 2010 led us to recognize that area of SGM in various years was likely determined by the relation between topography (and bathymetry) and lake levels, which was not yet quantified. Thus, we added collection of elevation data to the study, as well as photointerpretation analyses, which allowed us to include decreases from extreme lake-level highs in 1952 and 1974 to the study (Fig. 1). Our overall study objectives were to assess changes in composition of plant communities across the three study years that had different waterlevel histories and to draw correlations between topography and bathymetry and lake level in determining areal extent of SGM, which had previously not been made. We believe that our results and interpretation add greatly to understanding vegetation dynamics in wetlands subject to large fluctuations in water levels, especially in relation to topography.

Fig. 1
Hydrograph showing historical water levels for Lake Michigan-Huron, 1940 to 2020 (NOAA 2021), with arrows labeling years for photointerpretation. Arrows for vegetation sampling years were also labeled with *. Vegetation sampling in 1995 with no photointerpretation was labeled with a dashed arrow

Study Area
Arcadia Marsh is a 170-ha drowned river mouth wetland  near the village of Arcadia in Manistee County, Michigan, USA (Fig. 2). The wetland follows the corridor of a stream formed by the confluence of Bowens, Tondu, and Lucker Creeks crossing a wide basin upstream from Arcadia Lake, which is connected to the eastern shore of Lake Michigan by a channel. The wetland is separated from Arcadia Lake by a 1.0-km-long roadcrossing (M-22) with two large culverts that restrict flow during high flow periods. Surface sediments in the aquatic zone are largely decomposed peat, with some sand and silt. Numerous ditches were constructed through much of the wetland in an attempt to drain it for agriculture, which was unsuccessful because water levels are controlled by Lake Michigan levels. The nearby adjacent land is used for agriculture. The most prominent vegetation type is sedgegrass meadow dominated by Calamagrostis canadensis and Carex stricta, although some areas have been invaded by species such as Phalaris arundinacea, Typha angustifolia, Typha × glauca, and more recently Phragmites australis. To focus on portions of the marsh most affected by lake level, an area of approximately 27.6 ha was selected for study (Fig. 2), defined by wetland edge to the south, road to the west, a ditch to the north, and parallel to an agricultural excavation on the east.

Vegetation Sampling
Initial vegetation mapping and plant sampling were conducted during 27-28 July 1995 (Wilcox et al. 2002). Major vegetation types clearly definable on aerial images, which were photographed in 1994 in anticipation of their need, were identified and ground-truthed in the field with Fig. 2 Map showing the Arcadia Marsh study site along the eastern shore of Lake Michigan in Manistee County, MI, USA photographs in hand. Vegetation types were categorized as interior sedge-grass meadow, Phalaris/Typha-invaded sedge-grass meadow, short emergent, cattail, and submersed/ floating. We then sampled ten 1-m × 1-m quadrats in each vegetation type according to a randomly dispersed, haphazard design (blind toss over shoulder). All taxa in each quadrat were identified to species, if possible, and estimates of percent cover were assigned to each taxon in the quadrats at one-percent increments to 10, then at five-percent increments. Data on T. angustifolia and T. × glauca were combined due to the tedious task of identifying each plant individually. Similar sampling was conducted during 13-16 August 2002 and 20-22 July 2010, with 20 quadrats sampled in interior sedge-grass meadow and submersed/floating.

Photointerpretation
New 1:5000 CIR photos in 1994 and 1:6000 CIR photos in 2002 and 2010 were contracted through private vendors. Following an extensive search, we selected the best available imagery from 1965 through 2010 for analysis (Table 1). Available orthophotos were loaded into ArcMap GIS (Esri) such that each site had a layer file projected to UTM zone 16 N. Heads-up digitizing was used to identify the boundaries of plant community types and delineate them with polygon features. Due to variation of photo resolution and quality, the resolution at which vegetation was delineated differed across years, although a scale of 1:1500 or better was used when drawing polygons. The border extent was as described above.
We identified and quantified nine vegetation types, although not all of them were present in all years: sedgegrass meadow, short emergent, submersed aquatic, cattail, common reed, shrub, tree, open water, and upland. Initial ground-truthing of the mapped 1994 photos was conducted in 1995, and follow-up ground-truthing was done in 2002 and 2010. Although the sedge-grass meadow vegetation type was broken into SGM and invaded SGM categories for field vegetation sampling, that subdivision could not be recognized in most photos, so it was not used in photointerpretation. Ground-truthing for other years was not possible, so the location of existing community boundaries in the 1994, 2002, and 2010 photos, as well as their spectral signatures, helped identify the extent to which borders had expanded or contracted in the other years.
We used auto-complete polygons within the editor toolbar in ArcMap to produce polygon features around vegetation stands identified in each year (e.g., 2010: Fig. 3). To ensure that the extent of wetland delineated was the same in each year, the clip feature in ArcMap was used so that only the delineated polygons within the wetland border were used in the analyses. Total area was determined and compared to ensure that the boundaries were the same for each year.

Topographic Data Collection
Topographic and bathymetric survey data were collected during July 2010 sampling using two methods. A Trimble RTK GPS was used for real-time, differentially-corrected elevations in areas above water, totaling 647 points with estimated elevation accuracy of 3 cm. A Garmin hand-held GPS in combination with water-depth measurements from a boat and a benchmark on the bridge culvert that was surveyed by RTK GPS was used for below water areas, totaling 381 points (Appendix Fig. 1).
All Digital Elevation Models (DEMs) used the International Great Lakes Datum 1985 (IGLD85) for reference. All spatial data sets were created based on the Universal Transverse Mercator Zone 16 North coordinate system, the North American Datum of 1983, and used internal units of the SI (metric) system.

Processing Methods
Extent Prior to processing survey data to generate a DEM, we needed to establish an extent polygon for subsequent processing steps. We began by constructing a polygon within a Geographic Information System (GIS) based on the site boundaries of wetland edge on the south, road on the west, ditch on the north, and parallel to an agricultural excavation on the east. LIDAR DEMs provided by the United States Army Corps of Engineers Joint Airborne  Portions of these data were clipped and included to supplement the survey data points described above.

Merging Hand-Held GPS Boat-Based and RTK GPS Data
Sets GPS data were merged by selecting point features from each set, copying those features with the select tool in Arcmap's editor tool bar, and pasting them into a new shape file with the same tool. Horizontal positional accuracy, vertical elevation values, and naming conventions were used to cross-validate the independent attribute fields with the input data that were used to produce the merged shape file.
Adjusting Elevations to the IGLD85 Datum Vdatum (version 2.3.0) produced by the National Oceanic and Atmospheric Administration was used to convert North American Vertical Datum of 1988 (NAVD88) elevation values to the IGLD85 system.
Interpolation An Inverse Distance Weighted (IDW) interpolation method was applied to the survey point data within Arcmap's Spatial Analyst extension using the default settings. These settings included a second power weighting function combined with a variable search radius that required twelve points to estimate grid cell values.
Clipping The resulting DEM was clipped based on the extent polygon we created. This eliminated unrealistic data from beyond the July 2010 survey extent that may have led to a poor depiction of actual ground surface and/or bathymetric elevation values.

Map Generation
The clipped, interpolated DEM was used to create a topographic map with 20-cm contour intervals. Using the contour tool within ArcPro, the contour interval was set to 0.2 m, and the base contour was set to an elevation of 175 m, with a Z factor of 1. The contour type was "contour polygon" to enable summarizing the area between the contours. The contour polygon layer was then clipped to the extent of the study area, and using the summarizewithin tool, the area between each contour was calculated in square meters.

Data Analyses
Plant community data from field sampling were sorted by vegetation type and year. For evaluation, Importance Values (IV) for all taxa were then calculated by vegetation type in each year as relative mean percent cover + relative frequency × 100. The area of wetland mapped in each vegetation type was determined for each photo year, and data were converted to percent of wetland. We took several approaches to assess the relations of the prominent sedge-grass meadow with variable Lake Michigan water levels.
We ran regressions of %SGM mapped in each photo year against mean three-month, growing season, recorded lake level, and against the number of years since the last extreme high summer lake level defined as > 177 m IGLD85 (1952:177.26 m;1974:177.27 m;1986;177.37 m;1997:177.16 m). We also made comparisons between %SGM from photointerpretation results and the topographic data from Arcadia Marsh, which were developed with 0.2-m contour intervals. We calculated the cumulative area above each contour and interpolated to 0.01 m levels to determine the percent of wetland that would be predicted to dewater when Lake Michigan water levels receded from extreme highs for comparison with %SGM in specific photo years. Exposure of the land surface would be expected to promote development or reestablishment of sedge-grass meadow (Wilcox 2004;Keddy and Campbell 2020).

Vegetation Sampling
The interior sedge-grass meadow was dominated by C. canadensis and C. stricta in all years sampled, with other prominent taxa including Impatiens capensis and Persicaria amphibia in 1995, Campanula aparinoides in 2002, and Carex aquatilis and Carex lacustris in 2010 ( Table 2). The decrease in summertime peak water levels in 2002 ( Fig. 1) was accompanied by an increase in C. canadensis but decreased again in 2010 sampling. Carex stricta decreased in 2002 and decreased further in 2010.
The invaded sedge-grass meadow was dominated by C. stricta, C. canadensis, and Typha sp. in 1995 (Table 2). However, sampling of this vegetation type was largely in a different area in 2002 and 2010 because lower water levels exposed areas of previously flooded sedge-grass meadow. Carex stricta and Typha then decreased in sampling, with P. arundinacea, C. lacustris, and C. canadensis becoming the dominant species.
Areas sampled for short emergents also changed with changes in water level. During higher water levels in 1995 ( Fig. 1), dominant species were Sparganium eurycarpum and Lemna minor, with Ceratophyllum demersum and Sagittaria latifolia prominent (   (Table 2). Other prominent species across sampling years included L. minor, C. demersum, and Potamogeton crispus. The most dominant species in 2002 and 2010 was Stuckenia pectinata.
The cattail vegetation type was dominated by Typha sp. in all years sampled (Table 2). Phalaris arundinacea and C. canadensis were also prevalent in 2002 and 2010.

Photointerpretation
Three-month summer water levels (Table 3)   During extreme high three-month summer water levels in 1986 (177.37 m), 94.0% of the sampled wetland was in S/F, with little SGM remaining (4.8%) (Table 3). However, as water levels receded in 1987 (177.04 m) and further by 1988 (176.63 m), S/F was reduced to 77.9% and SGM increased to 20.6%. Continuation of low water levels in 1992 (176.54 m) resulted in an increase in SGM (27.3%), a decrease in S/F vegetation, and a return of SE and small areas of C. Water levels increased more than 35 cm in 1993 (176.86 m), resulting in increased S/F and a slight increase in SGM (28.8%) at the expense of SE. Water levels decreased slightly in 1994, with a reduction in S/F and SE, while SGM increased to 35.2%.
High three-month summer water levels returned in 1997 (177.16 m), and photointerpretation of 1998 photos (176.87 m) showed another increase in S/F to 66.3% and a reduction in SGM to 20.6% and SE to 4.9% (Table 3). A long period of low water levels began in 1999 (Fig. 1), with photos from 2002 (176.31 m) showing SGM at 61.0%, SE at 20.6%, and S/F reduced to 12.1%. Changes in SGM and S/F were minor during continued low water levels in 2004, 2006, 2009, and 2010 (range 176.15 -176.44 m), while SE decreased and C increased. Common reed (CR) was mapped in 2004 and increased from 5.9% to 11.3% by 2010.

Topography and Bathymetry
The topographic map showed little variation in elevation across much of the study area (Fig. 4). Elevations from 175.8 m to 176.8 m (IGLD85) contained 89.8% of the total study area, which ranged in elevation from 175.0 m   (Table 4). Elevations from 176.0 m to 176.4 m accounted for 47.3% of the total study area. For later analyses, the cumulative area of wetland above each contour interval was also calculated, which confirmed that much of the wetland area had little relief.

Discussion
Our objectives were to assess changes in wetland vegetation in relation to large fluctuations in water levels, which we addressed in three ways. 1) We tracked plant community composition changes related to recorded lake-level history, which showed that key SGM species responded to flooding and availability of exposed moist soil habitat but potentially also to competition, while composition in other vegetation types was more closely tied to water-depth preferences. 2) We tracked changes in area of SGM vs. recorded growing season lake levels and to number of years since the last extreme high lake level, both of which were significant. 3) We assessed area of SGM in relation to the combined influence of recorded lake level and topography, which highlighted the importance of land-surface elevation.

Sedge-Grass Meadow
The large increase in Importance Values of C. canadensis in both interior and invaded sedge-grass meadow in response to the reduction in growing season water levels from 176.62 m in 1995 to 176.31 m in 2002 (Table 2) follows its known habitat preference for moist soil vs. standing water (Costello 1936;Voss and Reznicek 2012). A return to lower 1995 IV levels in 2010, when water levels dropped further to 176.25 m, may have been related to competition from C. lacustris, which has a spreading growth form and can be productive (Bernard and MacDonald 1974;Yetka and Galatowitsch 1999). Calamagrostis can occur on C. stricta tussocks (Costello 1936;Peach and Zedler 2006) but is often found in slightly higher and drier elevations in wetlands (Costello 1936;Keddy and Reznicek 1982;Keddy1984;Kercher and Zedler 2004). Both species can grow vegetatively by tillering and can expand readily into open areas, especially when water levels are lower (Costello 1936;Budelsky and Galatowitsch 2004). Carex stricta was prominent in both interior and invaded sedge-grass meadow in 1995, likely because it forms tussocks and can tolerate deeper water (Costello 1936;Peach and Zedler 2006). In the invaded sedge-grass meadow, increased competition from C. lacustris may have caused reductions in C. stricta by 2010, but the large increase in P. arundinacea, which has been shown to out-compete C. stricta (Wetzel and van der Valk 1998;Budelsky and Galatowitsch 2004;Kercher and Zedler 2004), was more likely responsible for the decrease in IV from 53 in 1995 to 13 in 2002 and 2010. The large decrease in Typha from 1995 to later years with lower lake levels is likely explained by its requirement for water, as demonstrated in studies of cattail invasion in sedge-grass meadows elsewhere (Wilcox et al. 1984.

Other Vegetation Types
Changes in lake level result in changes of water depth, which can drive plant species distributions (e.g., van der Table 3 Percent of delineated Arcadia Marsh mapped by photointerpretation in study years as sedge/grass meadow (SGM), short emergent (SE), submersed/floating (S/F), cattail (C), and common reed (CR). Mean, three-month, growing season Lake Michigan water levels (June, July, August) in meters IGLD85 are also provided (WL)  (Table 2) were likely related to decreases in growing season water levels across years, resulting in changes in locations of areas that were sampled -the short emergents were in different places. However, species typical of more shallow  (Table 2). By 2002, lower lake levels reduced standing water to a relatively narrow channel that previously contained no Nuphar. The channel was rather turbid in 2002 and 2010, which explains dominance by turbidity-tolerant C. demersum, P. crispus, and S. pectinata (Adamus and Brandt 1990).

Area of SGM vs. Growing Season and Extreme High Lake Levels
One of our objectives in this publication was to assess the responses of sedge-grass meadow to fluctuations in Great Lakes water levels, as SGM is an important habitat for wetland fauna (e.g., Wilcox 1995;Farrell 2001;Riffell et al. 2001;Farrell et al. 2006;Cooper et al. 2008), and it has faced losses in many Great Lakes wetlands (e.g., Albert 2003;Stanley et al. 2005;Frieswyk and Zedler 2007;Wilcox and Bateman 2018). Percent SGM vs. recorded mean three-month lake level, which is biologically important because it represents much of the growing season, showed a significant relation (R 2 = 0.756, p = 0.000) (Fig. 5A). The relation of %SGM vs. years since high lake level was also significant (R 2 = 0.703, p = 0.000) (Fig. 5B) and has importance because sedge-grass meadow may be reduced or even eliminated by extreme flooding, as will be addressed in our next assessment.
The percent of Arcadia Marsh returning to SGM when water levels decreased from high lake-level years differed among years. Following a 1.24-m reduction in three-month summer lake level from 1952 to photo year 1965 (Fig. 1,  Table 3), all but 15% of the marsh was SGM, despite an intervening higher lake level of 176.73 m in 1960. Although there was a 0.60 m decrease in lake level from the 1974 high to 1978, summertime water levels in 1976 were still at 177.12 m, and the two-year lag time was not sufficient to allow regeneration of SGM (Wilcox and Xie 2007). The 0.83-m lake-level reduction from 1986 to 1992 resulted in SGM increasing by 22.5% in six years, while the nearly identical 0.85-m reduction from 1997 to 2002 resulted in an increase of 40.4% in five years from photo year 1998, in which lake level had already decreased by 0.29 m. In a period of five to 13 years following a reduction in growing season water levels by 0.83 m or more from an extreme high lake level, the increases in %SGM, as could be measured from available photos, ranged widely. Each of those time periods had lag times of five or more years, suggested as critical by Wilcox and Xie (2007). The disparity in response comes from the starting and ending elevations of lake level in these comparisons combined with the topography of the wetland. Growing season water levels were 176.02 m in 1965, 176.54 m in 1992, and 176.31 m in 2002. Simply, lower lake levels may expose more of the underlying soils and create more habitat for sedge-grass meadow (Wilcox 2004;Keddy and Campbell 2020).

Area of SGM vs. Lake Levels and Topography
While water depths often drive wetland plant distributions, those depths are dependent on both water level and topography (Poiani and Johnson 1993). Comparisons between %SGM from photointerpretation results and the topography and bathymetry showed that topographic data predicted 76.2% exposure of the wetland land surface in 1965, while %SGM was 84.4% (Fig. 6). The three-month summer lake level was very low (176.02 m), and despite an intervening moderate, single-year high in 1960 (Fig. 1), there was a lengthy time lag following the 1952 extreme high lake level that would promote growth of sedge-grass meadow (Wilcox and Xie 2007). Following the 1974 extreme high, predicted exposure of land surface was 11.4% in 1978, with lake levels at 176.67 m, and %SGM was 24.4%. Although percent land cover was less than %SGM in 1978, both values were much less than for 1965.
Following the 1986 extreme high lake level, predicted land surface exposed was 13.9% in 1988 (176.63 m) and Following all post-extreme years, percent land-surface exposure was always less than corresponding %SGM, likely due to accuracy of topographic mapping based on a finite number of RTK-GPS and water-depth data points and LIDAR DEM data from a relatively flat area. However, relative changes in both predicted land surface and %SGM shown in our results demonstrate that accuracy of lake level as a predictor of area of sedge-grass meadow is dependent on topography. Lake Michigan water levels increased again beginning in 2014 and extended to a threemonth extreme high of 177.44 m in 2020 (Fig. 1). When water levels decrease dramatically again, as expected based on historical data and paleo-lake-level studies (Baedke and Thompson 2000;Argyilan et al. 2018), the response of SGM in Arcadia Marsh and other Lake Michigan/Huron Marsh study area mapped as sedge-grass meadow (SGM) A) vs. mean three-month summer water levels of Lake Michigan (June-August) in photointerpretation study years (y = 177-0.0147x, R 2 = 0.756, p = 0.000), B) vs. number of years since extreme high lake level exceeding 177 m (IGLD85) (y = -149 + 0.16x, R 2 = 0.703, p = 0.000) Fig. 6 Percent of Arcadia Marsh study area ( Fig. 4) with (black) elevation greater than three-month summer water levels of Lake Michigan (June-August) and (gray) mapped as sedge-grass meadow (SGM) in years with low lake levels following extreme highs (> 177 m IGLD85) in 1952(> 177 m IGLD85) in , 1974(> 177 m IGLD85) in , 1986(> 177 m IGLD85) in , and 1997