predicting wetland plant community responses to proposed...

J. Great Lakes Res. 33:751–773Internat. Assoc. Great Lakes Res., 2007

INTRODUCTION

Human activities have modified natural water-level fluctuations in the Great Lakes through exca-vation of connecting channels and regulation at theoutlets of Lakes Superior and Ontario. Regulationof Lake Ontario began around 1960 with operationof the St. Lawrence Seaway. Lake levels are largelycontrolled at the Moses-Saunders hydroelectric dam

*Corresponding author. E-mail: [email protected]

between Cornwall, Ontario and Massena, NewYork. Under the current regulation plan (Plan1958D with deviations), high lake levels normallyexperienced during high water-supply periods havebeen lowered and low lake levels during low water-supply periods raised. The lake-level range hasbeen compressed from approximately 1.5 m to 0.7m, or half of what it was prior to regulation orwould have been without regulation (Wilcox et al.2005).

Water-level fluctuations are a natural phenome-

751

Predicting Wetland Plant Community Responses to ProposedWater-level-regulation Plans for Lake Ontario: GIS-based Modeling

Douglas A. Wilcox1,* and Yichun Xie2

1U.S. Geological SurveyGreat Lakes Science Center

1451 Green RoadAnn Arbor, Michigan 48105

2Department of Geography and GeologyEastern Michigan University

205 Strong HallYpsilanti, Michigan 48197

ABSTRACT. Integrated, GIS-based, wetland predictive models were constructed to assist in predictingthe responses of wetland plant communities to proposed new water-level regulation plans for LakeOntario. The modeling exercise consisted of four major components: 1) building individual site wetlandgeometric models; 2) constructing generalized wetland geometric models representing specific types ofwetlands (rectangle model for drowned river mouth wetlands, half ring model for open embayment wet-lands, half ellipse model for protected embayment wetlands, and ellipse model for barrier beach wet-lands); 3) assigning wetland plant profiles to the generalized wetland geometric models that identifyassociations between past flooding / dewatering events and the regulated water-level changes of a pro-posed water-level-regulation plan; and 4) predicting relevant proportions of wetland plant communitiesand the time durations during which they would be affected under proposed regulation plans. Based onthis conceptual foundation, the predictive models were constructed using bathymetric and topographicwetland models and technical procedures operating on the platform of ArcGIS. An example of the modelprocesses and outputs for the drowned river mouth wetland model using a test regulation plan illustratesthe four components and, when compared against other test regulation plans, provided results that metecological expectations. The model results were also compared to independent data collected by photoin-terpretation. Although data collections were not directly comparable, the predicted extent of meadowmarsh in years in which photographs were taken was significantly correlated with extent of mappedmeadow marsh in all but barrier beach wetlands. The predictive model for wetland plant communitiesprovided valuable input into International Joint Commission deliberations on new regulation plans andwas also incorporated into faunal predictive models used for that purpose.

INDEX WORDS: GIS modeling, generalized wetland geometric models, lake-level regulation plans,mathematical modeling, plant community profile.

752 Wilcox and Xie

wetland models, and technical procedures for creat-ing the models on the platform of ArcGIS. In thefollowing sections, we present the general approachbehind the modeling effort, the nature of input data,model design and the model mathematical founda-tion, the construction of geometric models for indi-vidual wetland sites, and the processes used forbuilding generalized wetland geometric models offour wetland geomorphic types. We then describehow the models and mathematical routines generatepredictions of plant communities that would resultunder new regulation plans and describe the modelaccuracy and possible error sources. We also pro-vide an illustration of the predictive model to evalu-ate one Lake Ontario lake-level regulation plan andtest the results against independent data.

GENERAL APPROACH

Many challenges were encountered in a previouseffort to study the responses of Lake Ontario wet-land ecosystems to water-level changes (Wilcox etal. 1992, Wilcox and Meeker 1995), including a)time-consuming manual construction of wetlandmodels and manual calculation of wetland areas be-tween different elevation intervals; b) inflexibilitywhen adjusting model parameters; c) non-scalable,two-dimensional data sets; d) lack of an interactivevisual means to display findings readily; and e)most profoundly, the inability to adopt an integratedand comprehensive approach due to the lack of dataexploration and modeling capability that GIS nowprovides (Batty and Xie 1994). Therefore, thesechallenges led us to develop integrated wetlandmathematical models based on GIS geo-processingtechniques to search for systematic and effectivesolutions. The integrated, GIS-based mathematicalmodel predicting the responses of wetland plantcommunity to proposed new water-level regulationplans is depicted in Figure 1. The scientific logicunderlying this mathematical modeling makes apresumption of continuity of natural transforma-tions (Berkhout and Hertin 2000).

Field measurements from representative wetlandstudy sites were needed for modeling work; 32 siteswere selected and segregated into four geomorphictypes (drowned river mouth, open embayment, pro-tected embayment, barrier beach) for developmentof four models, each specific to a geomorphic type(Wilcox et al. 2005, Hudon et al. 2006). The mod-els operate under the assumption that the reactionsof wetland plant communities to future water-levelchanges will be consistent with their reactions in

non in the Great Lakes due to natural climatic vari-ability. For example, Lake Michigan was less thanhalf its current size during the mid-Holocene warm-ing period about 8,500 years ago, and lake levelswere more than 4 m higher than today during theNipissing II phase 4,500 years ago. Since that time,the lake has experienced extreme high and low lakelevels approximately every 150–160 years andlesser events approximately each 30–33 years(Thompson and Baedke 1997, Baedke and Thomp-son 2000). Water-level changes such as these di-rectly affect the biological communities of theGreat Lakes (Working Committee 2 1993, Wilcox1995, Maynard and Wilcox 1997, EnvironmentCanada 2002). The effects are greatest in shallowwater, where even small changes in lake level canresult in conversion of a standing water environ-ment to an environment in which sediments are ex-posed to the air, or vice versa, resulting in death byflooding or in plant seed-bank germination (Keddyand Reznicek 1986, Wilcox 1995). Lake-level regu-lation disrupts this natural process.

There is a real need to understand the correlationbetween water-level patterns and biologicalprocesses that determine wetland plant communitydiversity, abundance, and distribution. This infor-mation can then be linked to habitat requirementsfor wetland fish and wildlife communities. To-gether, the information gained would enable devel-opment of water-regulation criteria important towetland communities and assessment models foruse in evaluating alternative water-regulation plans(Wilcox et al. 2005, Hudon et al. 2006).

In 2001, the International Joint Commission(IJC) undertook a bi-national study of the regula-tion plan for Lake Ontario, with the potential objec-tive of developing a new plan that better serves theinterests of hydropower, shipping, water supply,recreational boaters, riparian landowners, and theenvironment. Some of the wetland-related objec-tives of the environmental portion of the study wereto demonstrate qualitative and quantitative changesin wetland plant communities resulting from pastregulation (factoring out other environmental influ-ences), to determine water-level patterns that bestmaintain faunal habitat diversity (as determined byplant community diversity, abundance, and distribu-tion), and to develop predictive models and perfor-mance indicators to evaluate proposed newregulation plans for the lake.

This paper focuses on the conceptual foundationof the predictive models, based on generalized geo-metric (combined bathymetric and topographic)

Modeling Wetland Response to Regulation of Lake Ontario Water Levels 753

the past. Using the approach of Wilcox and Meeker(1991) for studies on regulated reservoirs, quantita-tive vegetation sampling was conducted along tran-sects that follow elevation contours with specificpast lake-level histories (number of years since lastflooded and since last dewatered during the grow-ing season). Plants respond to water depths, not tolake-level elevations, so associations between lakelevels and topographic/bathymetric informationwere also needed. Using field data and GISmethodologies, topographic/bathymetric models foreach individual study site were constructed andthen combined by geomorphic type to develop fourgeneralized, three-dimensional, geometric models.The geometric models convert any instantaneouslake level to area or percent of wetland with spe-cific water depths or water-depth histories. Pro-posed new regulation plans supply data in the formof predicted lake levels (average static level in eachof 48 quarter months of each year) over a 101-yearperiod, so computer codes were entered into themodel to identify the periodic high lake levels inthose data sets that would flood wetlands and thelowest growing-season lake levels that would dewa-ter them. Those specific lake levels and their peri-odicities were then tied to the plant communityresponses observed in the field, and areal portions

of the geometric models were assigned to specificplant community types. Repeated iterations of thisroutine in the models, for each of the 101 years in aregulation plan, resulted in time-weighted predic-tions of the area/percent of wetland of each geo-morphic type that would be occupied by the plantcommunities identified in field sampling.

STUDY SITES

The 32 study sites selected for this work weredistributed across the Lake Ontario-Upper St.Lawrence River area upstream from the dam andincluded eight wetlands of each of four geomorphictypes: open embayment, protected embayment, bar-rier beach, and drowned river mouth (Fig. 2). Halfof the sites for each geomorphic type were inCanada and half were in the United States. The sitesavailable for study within each geomorphic typewere restricted to specific reaches of the shorelinefor which topographic and bathymetric data wouldbe obtained as part of the overall IJC study. Privateland ownership added further constraints at somelocations. Among the available sites, wetlands wereselected that retained hydrologic connection to thelake even in low lake-level years and were least im-pacted by other human disturbances (thus affectedprimarily by lake levels and restricting effects fromhuman alterations in the watershed (Wei and Chow-Fraser 2005)). These sites are intended to representmost of the total of 879 geomorphically distinctwetlands, totaling 25,847 hectares, identified in awetland inventory developed in this study (Appen-dix A; view appendix at http://iaglr.org/jglr/appendices/). That inventory is a seamless, digital,vector-based coastal wetland database created forthe entire Lake Ontario basin and Upper St.Lawrence using a combination of existing Ontarioand New York wetland databases and aerial photointerpretation (Wilcox et al. 2005). Later analysesthat extended the lakeward edge of the wetlands tomatch the depth boundaries of the models increasedthe total area of wetland to 31,652 ha. We recog-nized that some Lake Ontario wetlands that aregreatly impacted by human activity (Wei andChow-Fraser 2005) may not respond to a new regu-lation plan in the same manner as those used forconstructing the model. However, there was nomeans to control for those differences, and selectionof any given regulation plan likely would not affectthe outcome in the impacted sites.

FIG. 1. Flow chart for the GIS-based mathemat-ical model predicting responses of wetland plantcommunities to regulated lake-level changes.

754 Wilcox and Xie

SITE GEOMETRIC WETLAND MODELS

The predictive model requires construction ofsite-specific geometric models for each sampledwetland, based on both topographic and bathymet-ric features. The morphometry of each site within agiven wetland type can vary greatly. Therefore, thegeometries of eight sites were used to generate themodels for each type. The models are intended torepresent average wetlands within a type and not befaithful to any specific site. Topographic and bathy-metric data for this effort were collected using air-borne LIDAR and boat-based depth soundings aspart of a larger IJC effort that included the need forelevation data for assessing shoreline erosion is-sues, as well as for specific wetlands. Unlike depthsoundings, LIDAR data do not have point-specificaccuracy, but it was not necessary for creation ofgeneralized models. Three steps are involved inconstructing a wetland site geometric model usingthese data: interpolating the topographic surface, in-terpolating the bathymetric surface, and mergingthe topographic and bathymetric surfaces. Severalconcepts and techniques are important in these

processes: how to design the sample points, whichinterpolation method is appropriate, and what spe-cial considerations must be made when merging thetopographic and bathymetric surfaces.

Topographic and bathymetric surfaces are com-monly constructed from data points, which are col-lected through field sampling and observation(Petrie 1991). In practice, there are two steps for lo-cating sample points, transect lines first and thensampling locations. In this study, transect lines wereplaced in areas where they provided the best repre-sentations of terrain and plant community charac-teristics of specific wetlands. In addition, the fullcoverage of the study area was taken into consider-ation. The spacing between the sampling transectlines was commonly 100 m. The sampled points onthe transect lines for both topography and bathyme-try were usually densely spaced (about 5–10 m),and the spacing was close enough to produce a gridof half-meter elevation increments for most interpo-lation methods. The topographic and bathymetricdata were reviewed carefully to detect obviousanomalies. The most accurate and best suitable data

FIG. 2. Map of wetland study sites along the shore of Lake Ontario and the Upper St. Lawrence River.


were selected for the analysis according to the prin-ciple that uneven but adequate density of measuredpoints should be matched with local roughness ofthe terrain surface (Makarovic 1977).

There are two major groups of interpolationmethods—global and local. The global methods fita particular mathematical function to the entire dataset in such a manner that all data points exist onthis chosen functional surface (Hardy 1971, Lam1983, McCullagh 1991, ESRI 2004). The local in-terpolation functions assume there exists an auto-correlation effect in the surface that decreases withdistance away from the point where the interpola-tion is made. One of the local interpolation meth-ods, IDW (inverse distance weighted), was chosenin this study because the IDW method is intuitiveand efficient and works best with relatively denselyand evenly distributed points. Moreover, in the con-text of topographic/bathymetric modeling, IDW ex-plains best the linear feature of slope changes. IDWinterpolation explicitly implements the assumptionthat things that are close to one another are morealike than those that are farther apart. To predict avalue for any unmeasured location, IDW uses themeasured values surrounding the prediction loca-tion. Those measured values closest to the predic-tion location have more influence on the predictedvalue than those farther away. Thus, IDW assumesthat each measured point has a local influence that

diminishes with distance. It weights the pointscloser to the prediction location greater than thosefarther away, hence the name inverse distanceweighted.

The diminishing local influence of IDW is cali-brated through two parameters, the neighborhoodsize (the radius) and the power. The radius is thedistance in map units specifying that all input sam-ple points within the specified radius will be used toperform interpolation. The power parameter con-trols the significance of the surrounding pointsupon the interpolated value. A higher power resultsin a steep curve, which defines less influence fromdistant points. A measure of Goodness of Fit is cal-culated to determine the best combination of thetwo parameters automatically for the IDW interpo-lations. For each interpolation, about 100 samplingpoints were randomly removed from the data set.Different permutations and combinations of thePower (1.6 to 4.4 at an increment of 0.4) and theRadius (6–16 with an increment of 2) values weretested on a randomly chosen wetland for each wet-land type (Fig. 3). Then, we computed RMSE (theroot mean square error, Equation 1) to see whichcombination of the parameters generated the small-est RMSE.

FIG. 3. Calibration results of the IDW interpolation parameters: radius and power. Different permuta-tions and combinations of the Power (1.6 to 4.4 at an increment of 0.4) and the Radius (6–16 with anincrement of 2) values were tested on a dataset randomly chosen from 32 wetlands. On the X-axis, Powershows eight values, 1.6, 2.0, 2.4, 2.8, 3.2, 3.6, 4.0, and 4.4. A power value of 2.4 and the radius value of 8mgave the smallest RMSE (the root mean square error).

(1)

756 Wilcox and Xie

N is the number of randomly chosen samplingpoints; Pi is the interpolated value at i samplingpoint; and Oi is the originally observed value at isampling point. A power value of 2.4 and the radiusvalue of 8 m were selected for final interpolationbased on RMSE because this pair gave the leastvariance. We then used these parameters to interpo-late the topography and bathymetry of the eightwetlands of each geomorphic type.

The separation line between bathymetry andtopography data follows a contour line, and the wa-tered portion of a wetland was usually small com-pared with the drier portion. Because of theseconsiderations and the spacing characteristics of thesampling transects and points, separate topographicand bathymetry interpolation processes were imple-mented instead of a seamless interpolation methodto avoid dominance of topographic data in interpo-lation at the edges of bathymetric and topographicdata. Bathymetric and topographic boundaries werecreated and used as mask polygons during the inter-polation process. Mask polygons help in restrictingthe area of interpolation.

The above procedures of creating wetland sitegeometric models were implemented in ArcGISSpatial Analyst following these steps: go to “Op-tions” in Spatial Analyst; select the boundary file asthe Analysis Mask; in the extent tab, select “sameas boundary file” (remember to change options forthe bathymetry and topographic interpolation withtheir respective boundary files before interpolat-ing); leave other fields with default values and click“OK”; use “Interpolate” to execute the interpola-tions with the output grid cell size of 0.5 m; andmerge the two interpolated topographic and ba-thymetry grids using Calculator in Spatial Analyst“sitegrid = merge([bathy_grid], [topo_grid]).”

GENERALIZING WETLANDGEOMETRIC MODELS

Generalized geometric models created from thesite-specific models were needed for each of thefour types of coastal wetland for use in predictingplant community distributions and calculating theareas occupied by each plant community. The gen-eralized models should accurately measure areasand approximately characterize topographic shapesof specific wetlands. The first requirement dictatesthat the generalized models observe the area ratiosbetween various elevation ranges to allow accurateprediction of wetland areas. The second require-ment is more for representation purposes so that the

generalized geometric models help visualize thewetland types in study. Several geo-processingtechniques were integrated to generate the geomet-ric wetland models: 1) averaging areas between ele-vation intervals for the four types of wetlands instudy and producing a summary table of the areasbetween various intervals; 2) averaging and graph-ing the slope (distance-depth) profile for these typesof wetlands; and 3) constructing a generalized geo-metric model for these types of wetlands based onthe results from Steps 1 and 2.

First, we needed to compute average areas be-tween various elevation intervals so that we couldhave quantitative measures for generalizing a geo-metric model, which support the area calculation re-quirement. A routine, Query-Area-By-Elevation-Intervals, was developed to perform spatial query ofsurface areas between elevation intervals and tocompute average values (Appendix B; view appen-dix at http://iaglr.org/jglr/appendices/). In brief, thiscode queries the wetland area at 0.25-m elevationintervals (the range of main interest and, thus,model development is between 73.00 m and 75.75m International Great Lakes Datum 1985 (IGLD1985). The routine then loops through eight samplewetlands of a specific type calculating the averageareas between elevation intervals. The final resultsare written as tables of average areas between theelevation intervals. The query results for thedrowned river mouth wetlands are given in Table 1as illustration.

Second, we needed to know the slope (depth)profile of a wetland type in study, which helps tovisualize the geometric shape of this wetland type.It is a straightforward task to create the slope(depth) profile because the area average values arecomputed against specific elevation intervals. Alldata that are needed to create a profile exist in thesummary tables. From the tables, the percentageand distance curves could be easily developed withany spreadsheet or GIS software. The slope-dis-tance profiles can be generated directly using Ar-cGIS GRID (Spatial Analyst) profile tool (Fig. 4).

Third, wetland types within the Great Lakes re-flect the influences of watershed hydrology andshoreline geomorphology (Keough et al. 1999, Al-bert et al. 2005). Within Lake Ontario—Upper St.Lawrence River, four distinct geomorphic types arecommon. They include wetlands protected fromwave attack by barrier beaches, thus retaining or-ganic sediments and developing a flatter topo-graphic profile; protected wetlands in river mouthsthat are back-flooded by the lake and also have or-


ganic sediments and a flatter topographic profile;wetlands exposed to wave attack in open embay-ments, thus having predominantly inorganic sedi-ments and a steeper topographic profile; andwetlands of intermediate wave exposure in pro-tected embayments that typically have flatter topo-graphic profiles and organic sediments. We wantedto visualize various wetland types using differentgeometric shapes, so that we would have intuitiveand realistic presentations of the wetland types instudy. However, due to varied shapes of the wet-lands, selecting a geometric form for a wetlandshape is a rough approximation. Four generalizedgeometric models were chosen to approximate fourwetland types from the perspectives of approxi-mated topographic shapes and wetland hydrologiccharacteristics (Fig. 5): drowned river mouth = rec-tangle; open embayment = half circle; protectedembayment = half ellipse; and barrier beach = el-lipse. The four geometric models were constructedby using the Circle, Ellipse, and Rectangle drawingtools in ArcGIS, respectively. In this paper, we usethe drowned river mouth wetland type as an exam-ple to introduce the basic steps of creating thebands between elevation contours. As describedpreviously (see Table 1), the area percentage be-

TABLE 1. Query results for drowned rivermouth wetlands presenting elevation range (Z, mInternational Great Lakes Datum 1985) and areaand percent of wetland. The area of wetland isbased on the average total area in square metersbetween the model boundaries (73.0–75.75 m) ofthe eight wetlands used to generate the model.

Z Range Average Area Percentage

73–73.25 17321.50 1.8773.25–73.5 27763.53 2.9973.5–73.75 27989.44 3.0273.75–74 46621.59 5.0274–74.25 98994.59 10.6674.25–74.5 58171.91 6.2774.5–74.75 32755.34 3.5374.75–75 103036.19 11.1075–75.25 143240.53 15.4375.25–75.5 136238.22 14.6875.5–75.75 111458.75 12.0175.75–76 92844.19 10.00

Total* 928333.47 96.56

* The total average area (928333.47) is the sum of theaverage areas between all elevation intervals from 70 to80 m at 0.25 m intervals. Therefore, the total percentagebetween the elevations 73–76 m is not equal to 100.

tween any two contours with a specific elevationrange is required to calculate the areas accurately.In other words, the area percentage values are usedto determine the widths (radii) of the elevationbands. The area percentages are standardized intothe band widths based on a chosen mapping scaleor size. The depth (slope) information is embeddedin the elevation bands. Therefore, we only neededto compute the radii of bands (Table 2).

Four steps in ArcGIS were needed to create thebands. First, two empty shape files (one point andone polygon) were created in ArcCatelog for futureuse. Second, a point was created in the point shape-file using ArcGIS Editing Tool. Third, bands werecreated in the polygon shapefile by using the BufferTool with reference to that point and by enteringradii values that are prompted by the Buffer Tool.Fourth, the attribute table of the polygon shapefilewas opened (the newly created bands) and two newfields added: Begin-Elevation and End-Elevation.Fifth, the elevation values of each band were en-tered into the attribute table. Because the area per-centage values (or the band radii) are averaged overthe eight sampled wetlands for each wetland type,the constructed hypothetical geometric rings aretopological approximations of a specific type ofwetland in study.

For the 3D visualization and mathematical mod-eling, the four shape files (rectangle for drownedriver mouth, half circle for open embayment, halfellipse for protected embayment, and ellipse forbarrier beach) were processed further to generateTINs (triangulated irregular networks), which canbe extruded to create 3D views in ArcGIS 3D Ana-lyst or other 3D authoring packages (Fig. 5). Usingthe drowned river mouth wetland model as an ex-ample, we opened 3D Analyst in ArcGIS and se-lected “Create TIN.” We then checked thegeneralized band dataset and selected End-Eleva-tion as “Height Source,” as well as “Tag ValueField” and “Soft Clip” for Triangulate Option. Wecreated a TIN by defining these parameters. Afterthe TIN dataset was created, we opened ArcGISArcScene to explore 3D view of the generalizedgeometric wetland model through defining (adjust-ing) “Scene Properties,” “View Settings,” and“Layer Properties.” Following these general proce-dures, we created the 3D views of the four general-ized wetland geometric models in Figure 5.

The TIN datasets needed to be rasterized (ex-ported to ArcGIS) into GRID files for calculationsof area in the wetland mathematical modeling. Thegeneralized GRID file of each wetland type is the

758 Wilcox and Xie

FIG. 5. Three-dimensional depictions oftopographic models of four wetland geo-morphic types showing cross-sectionsdepicted in Figure 4.

FIG. 4. Slope-distance profiles of the four wetlandgeomorphic types.


final data input to the respective GIS mathematicalmodels (see the next section). The rasterization wasdone in ArcGIS 3D Analyst by selecting “TIN toRaster” and entering 0.5 m as “Cell Size” and 1.0as “Z Factor.”

PLANT COMMUNITY INPUT DATA

Plant community data were collected by sam-pling in quadrats along transects that followed topo-graphic contours representing different flooding/dewatering histories associated with past lake-levelchanges (Wilcox et al. 2005). Selection of samplingelevations was dictated and constrained by actuallake-level history, as a flooding or dewatering eventthat is exceeded in succeeding years is eliminatedfrom consideration (see Figure 6 for visual interpre-tation). Since the existing wetland vegetation in thelake developed in response to the history of highlake levels and low lake levels, the selected eleva-tions reflect lake-level history. The elevations(IGLD 1985) used for sampling in 2003 are shownin Figure 6 and are as follows: A) 75.60 m, lastflooded 30 years ago; B) 75.45 m, last flooded 10years ago; C) 75.25 m, last flooded 5 years ago; D)75.00 m, last flooded 1 year ago and last dewateredduring growing season 2 years ago (variable flood-ing and dewatering over past 3 years; E) 74.85 m,last dewatered during growing season 4 years ago;F) 74.70 m last dewatered during growing season

38 years ago; G) 74.25 m, last dewatered duringgrowing season 68 years ago.

In each of 20 randomly placed 0.5 × 1.0 mquadrats/transect/study site, set with the longer sidefollowing the elevation contour, the plant speciespresent were identified and percent cover estima-tions were made by visual inspection. Vegetationsurvey data were analyzed using summary statisticsand ordination/classification procedures (Wilcox etal. 2005). Importance Values (IV) were calculatedfor each taxa on each transect as the sum of relativefrequency and relative mean cover ratings; thesevalues were used in the ordinations/classifications.Correlations between specific elevations and ac-companying plant communities were assessedacross all wetlands sampled to determine the rangeof elevations in which the most diverse plant com-munities occurred and to identify any specific habi-tat requirements of individual plant species,including invasive taxa. Any correlations with otherphysical habitat parameters, such as bulk densityand percent organic matter of soil, were identifiedalso.

When vegetation data were analyzed by speciesprominence and non-metric multidimensional scal-ing (NMDS), transects A, B, and C (see above)showed similarities across geomorphic types; tran-sects E and F were similar in all types except pro-tected embayments; and transect G was distinctfrom all other transects (Wilcox et al. 2005). Data

TABLE 2. Elevation ranges (m, International Great Lakes Datum 1985), areas (m2based on the average total area between the model boundaries), and widths (m) of the gen-eralized rings model for the drowned river mouth wetlands generated with data from eightwetland sites.

MinElev MaxElev Area Area_sum Ring Width Sum Ring Width

73.00 73.25 17321.50 17321.50 12.37 12.3773.25 73.50 13881.77 31203.27 9.92 22.2973.50 73.75 13994.72 45197.99 10.00 32.2873.75 74.00 23310.80 68508.79 16.65 48.9374.00 74.25 49497.29 118006.08 35.36 84.2974.25 74.50 29085.95 147092.03 20.78 105.0774.50 74.75 16377.68 163469.71 11.70 116.7674.75 75.00 51518.10 214987.81 36.80 153.5675.00 75.25 71620.26 286608.07 51.16 204.7275.25 75.50 68119.11 354727.18 48.66 253.3875.50 75.75 55729.38 410456.56 39.81 293.18

Ring Width = Area / LengthSum Ring Width = Area_sum / LengthThe long dimension of the drowned river mouth wetlands is 1,400 m on average from the samples (i.e., Length = 1,400 m).

760 Wilcox and Xie

were analyzed to ascertain similarities among wet-lands of the same geomorphic type for use in themodel. In general, the plant communities at eleva-tions that had not been flooded for 5 or more years(transects A, B, and C) were dominated by sedgesand grasses (meadow marsh), and those that hadnot been dewatered for 4–38 years (transects E andF) were dominated by cattails. The interveningtransect D that was intermittently flooded and de-watered over a 5-year span was also dominated bycattails but also contained a combination of sedges,grasses, and other emergent species. Plant commu-nities that had not been dewatered in the growingseason for many years (transect G) were dominatedby floating and submersed species. We recognizedthat seasonal and short-term (seiche) water-levelchanges may affect plant communities. However,they occurred across the entire past lake-level his-tory upon which the models were based and thusare inherently captured within the models.

PREDICTING WETLAND PLANTCOMMUNITY RESPONSES TO LAKE-LEVEL

REGULATION PLANS

With generalized geometric models created andwetland plant communities associated with actualflooding and dewatering histories defined, the nextstep is to predict the area of each plant communitythat will occur with future flooding and dewateringpatterns generated by proposed new regulationplans.

Proposed Future Lake Levels

For each new regulation plan for Lake Ontario,the IJC Hydrologic and Hydraulic Technical Work-ing Group provided 101 years of data (labeled1900–2000) representing modeled lake levels inwhich the regulation criteria in a new plan were ap-plied to the net basin supply data from 1900through 2000. The data were presented as quarter-

FIG. 6. Hydrograph for Lake Ontario (1918–2000) showing elevations selected for transects to samplewetland plant communities.


monthly lake-level elevations referenced to IGLD1985.

To evaluate the effects on upland and emergentvegetation from flooding during high lake-levelyears, our computer program was coded to identifythe highest quarter-monthly value in the entire planand assign all elevations above that value to vegeta-tion category U (never flooded, transition to up-land). The three adjacent (four total) highest valueswere then identified, and the elevation of the lowestof those quarter-monthly values was selected torepresent an elevation that had been flooded forfour quarter months during that peak lake-levelyear, which was likely long enough to affect uplandherbaceous vegetation and many upland woodyspecies, as well as some wetland species (Penfoundet al. 1945, Kadlec and Wentz 1974, Crawford1982, Jackson and Drew 1984, Kozlowski 1984). Asimilar process was used to evaluate other yearswith high lake levels; the specifics are described ina later section illustrating use of the models.

To evaluate the effects of low water years on sub-mersed and floating vegetation, the computer pro-gram identified the lowest peak quarter-monthlyvalue occurring during the growing season in theentire plan and assigned all elevations below thatvalue to vegetation category G (never dewateredover the past 68 years). Elevations above this valuewere dewatered during the growing season, thus af-fecting submersed aquatic and floating vegetation.Again, a similar process was used to evaluate otherlow lake levels in the plan.

The yearly water-level values described aboveare saved as a list called The Real Data List. Then,this list is examined with reference to the potentialwetland plant community responses to generate twoclass lists, “The Flood Class List” and “The Dewa-ter Class List.” These lists are generated by the sub-routine that can be viewed in Appendix C athttp://www.glsc.usgs.gov/_files/publications/ontari-ogis07.pdf and are used in the fourth sub-routine tomatch with a lake-level regulation plan, as will bediscussed in the following sections on plant com-munity response and illustration of use of the mod-els. Identification of the highest water-level yearsand the lowest growing season peak years in a plan(The Plan Data List) provides assessment criteriaon flooding and dewatering to which vegetation re-sponds. The method is analogous to that used to de-tect historical flooding and dewatering events,which were used to determine elevation contoursfor plant sampling.

Plant Community Response

Matching of high water levels in a regulationplan with plant community response is illustrated inthe pseudo codes (Appendix C; view appendix athttp://iaglr.org/jglr/appendices/). The match withthe flooding events first starts with finding thehighest planned water level in The Plan Data List(derived from the lake-level regulation plan instudy). When the highest water level is found, thehighest elevation that remains flooded for at leastone quarter-month becomes known, along with theyear in which it occurs. Starting with the most re-cent year in the plan data (2000), the number ofyears since the highest flood and its water-level ele-vation allow us to match the flood event with a wet-land plant community identified from the transectsampling. Within the generalized geometric model,this elevation determines the area of wetland af-fected by this water level as well. Therefore, theimpact of the planned (regulated) highest lake levelon the upland/wetland plant communities is pre-dicted. See the illustration in a later section forstep-by-step detailing of these and the remainingprocedures and outputs from running the model.

The next step is to determine the remainingplanned high water levels and match them withother flooding events. The algorithm is to truncate(or extract) a sub-dataset from The Plan Data List,which is the portion of the data from the previouslyidentified highest water level to the plan year beingevaluated (2000). In other words, this sub-datasetcontains data points between the most recent plan-ning year and the year immediately after the yearwhen the highest water level was previously identi-fied. Then, a routine similar to that described in thefirst step is applied to this truncated sub-dataset tofind the next highest water level, the year planned,and the water level (the height value); to matchwith a flooding event; to relate to the associatedwetland plant community; and to calculate the wet-lands area affected. This same procedure is loopeduntil all planned high water levels are matched withthe plant communities, which finishes a completerun.

The study beginning year is then moved back-ward one year to 1999 and another run is com-pleted. This process continues until it reaches thefirst plan year (1900). To provide data for earlyplan years, the entire lake-level sequence in ThePlan Data List is appended at the beginning of thelake-level data set. Overall, the procedures de-scribed above are applied to get model results for

762 Wilcox and Xie

each year in the 101-year plan sequence, as theprocess is executed until all plan years are ex-hausted and the number of the model outcome setsequal to the number of the plan years. In addition tomodel results for each individual year, the averagearea and percent wetland values for each plant com-munity type over the 101-year sequence are calcu-lated. The very same procedure (beginning with themost recent year in the plan and working sequen-tially backward to prior years) can be applied formatching with the dewatering events except forstarting with the lowest growing season high waterlevels, which denote dewatering. The pseudo codeand explanation for matching with the dewateringevents are similar to the match with the floodingevents.

MODEL SENSITIVITY ANDERROR SOURCES

All modeling activities involve some levels ofuncertainty (Burrough et al. 1996, Zhang andGoodchild 2002, Longley et al. 2005). This uncer-tainty can come from a lack of complete under-standing of the phenomena being studied, error inthe source (input) data, errors in geo-processingprocesses, and erroneous specifications of parame-ter values in the model. The input data in this re-search were produced by federal agencies of theUnited States and Canada and were processed withhighly professional quality control measures andwith excellent meta-data. We are not attempting byany means to guarantee that errors in the input dataare trivial. We explained the mathematical founda-tion of this integrated wetland modeling approachin a previous section. Here, we focus on the uncer-tainties that may arise from wrong selections of pa-rameters or from geometric generalizations.

The parameters (the power and the radius) for theIDW interpolation methods that were used to gener-ate wetland site geometric models in this studywere determined by running against the datasetsrandomly chosen across the 32 wetland samples.We also tested the IDW parameters randomlyamong the datasets of each wetland type (drownedriver mouth, Fig. 7a; open embayment, Fig. 7b; pro-tected embayment, Fig. 7c; and barrier beach, Fig.7d), respectively. The statistics used in the error de-tection were the root mean square errors (RMSE).

The different patterns of RMSE curves (Fig. 7)revealed three findings. First, the RMSE curveswere rising with the increase of the power values,which suggests that most sample wetlands in study

had gentle slope changes because small power val-ues had better performances.

Second, the RMSE displayed varied patternsalong with the increase of radius values. For in-stance, the RMSE decreased with the increase ofthe radius values in the drowned river mouth wet-land model, but increased with the increase of theradius values in the open embayment wetland. Theother two wetland types show more complexchange patterns. For the barrier beach wetland,Radii 6 and 8 generated lower RMSE ratios, Radii10 and 12 created higher RMSE ratios, and Radii14 and 16 had intermediate ratios. For the protectedembayment wetland, the RMSE first decreasedwhen the radius value increased but reversed thetrend when the power value passed 2.4. This find-ing suggests that the size of neighborhood had dif-ferent impacts on the IDW interpolation results. Alarger radius value worked better for the drownedriver mouth wetland. The RMSE curves convergedbetter for the protected embayment wetland, and thelow-middle radius values (8.0–10.0) generated thesmallest RMSE. However, the radius values for theother two wetland types showed noticeable fluctua-tions. Third, the open embayment and the barrierbeach wetlands displayed high RMSE values. Thedrowned river mouth and protected embaymentwetlands showed much lower RMSE values. Inbrief, the performances of the generalization for theprotected embayment wetlands and the drownedriver mouth were much better than the open embay-ment and the barrier beach wetlands. The modelprocedure worked best for the protected embaymentwetlands.

Averaging the area percentages between variouselevation intervals to construct the generalized geo-metric models for each wetland type is another con-cern of the model sensitivity. We used the statisticsof the standard deviation as a measure of the extentto which a distribution varies from its mean (Fig. 8).For each generalized geometric model (of a wetlandtype), the horizontal bar is the mean area percentagein an elevation interval, and the extended line is thestandard deviation (SD) over the eight samples. Thegeneralized model for drowned river mouth wetlandshas highly concentrated areas at upper elevationzones (74.75 m–76.00 m) and a single large area atthe elevation interval 74.00 m–74.25 m. The SD val-ues over four of the six zones are between 9.61 and16.40. The generalized model for open embaymentwetlands also has six zones over which the area per-centages are greater than 10.00. Four of those zonesare between 73.50 m and 74.50 m, and the other two


are between 75.00 m and 75.50 m. However, the SDvalues are much lower (around or slightly higherthan 5.00) than those of the drowned river mouthmodel. The generalized model for protected embay-ment wetlands only has three elevation zones thathave high area percentages. The three zones span73.75–74.00 m, 74.00–74.25 m, and 75.25–75.50 m.The corresponding SD values are also much higher,reaching 15.22. Moreover, there are three zones thathave lower area percentages but higher SD values(74.75–75.00 m, 75.00–75.25 m, 75.50–75.75 m).The generalized model for barrier beach wetlandshas three consecutively highly concentrated zones,which are located in the upper portions between75.00 and 75.75 m. The SD values are quite highalso. Moreover, the 74.50–74.75 m elevation zonehas a very high SD value.

When compared, the four wetland geometricmodels are distinct in terms of percent area distrib-utions along the model elevation profiles. Withinthe open and protected embayment geometric mod-els, lower elevation portions account for more ofthe total model area. Relative to the other wetlandtypes, the open embayment wetlands are more ex-posed to wave attack and ice scour, which reducesthe amount of organic sediment deposition. For thisreason, the open embayment wetlands are expectedto have a steeper topographic profile within theupper contour elevations and shallower slopes atlower elevation contours. The open embaymentwetlands are more regularly shaped, and thus, theopen embayment geometric model is more stableand predictable. However, the protected embaymentmodel seems to be very sensitive to the protection,which depends on coastal shorelines and geomor-phologic conditions. The study wetlands displayed

FIG. 7. Calibration of the IDW Interpolation Para-meters for four wetland geomorphic types. Differentpermutations and combinations of the Power (1.6 to4.0 at an increment of 0.4) and the Radius (6–16with an increment of 2) values were tested on adataset randomly chosen from each wetland type.The X-axis indicates seven power values; the Y-axisshows root mean square error (RMSE) as ratios. Thesix curves reflect the changes in RMSE at six radiusvalues. The ratios for each wetland type are lowerthan the total RMSE ratios for thirty-two wetlands(see Fig. 3) because they are computed for the eightwetlands of the same type.

764 Wilcox and Xie

FIG. 8. Mean percent areas and standard deviations within elevation intervals of thefour generalized geometric models. Mean percent areas are the averaged percentagesof area in a specific elevation interval vs. the total wetland area. Eight sample wet-lands were used in calculations for each wetland type.


results of data analyses from quantitative samplingof plant communities. Analysis of NMDS ordina-tions and mean percent cover data for the mostprominent species identified four distinct plantcommunities—ABC, D, EF, and G (Wilcox et al.2005). Transects A, B, and C were quite similar inspecies composition, with sedges, grasses, andsome upland species. In transect D, these sedges,grasses, and the cattails present in EF comprised theprominent species. Transects E and F were largelydominated by cattails, with some submersed speciesand a few sedges and grasses. In transect G, sub-mersed and floating-leaf vegetation (e.g., waterlilies) were much more prominent than emergentvegetation, and sedges and grasses were not ob-served. The demarcation between the EF and Gvegetation types at the study sites was typicallyvery distinct. Although we did not sample eleva-tions above 75.72 m that had not been flooded formore than 30 years due to the constraints imposedby actual past lake levels, we recognized that theywere still within the potential range of extreme highLake Ontario water levels. The plant communitiesat those unflooded elevations showed strong transi-tion to upland vegetation and were given a separatecategory (U). Similar actual lake-level constraints,as well as shallower basins, prevented sampling atlower elevations that had not been dewatered formore than 68 years; however, on-site observationssuggested that transect G represented those areaswell.

Based on past flooding/dewatering history thatresulted in these five communities, we set the fol-lowing rules and procedures for assigning vegeta-tion types to portions of the drowned river mouthgeometric model (not individual sites) in any givenyear in a regulation plan. Professional judgmentbased on discussions among prominent Great Lakeswetland scientists was used to determine breakpoints between some classes.

1. Elevations above the highest peak in the entireregulation plan: assign to U (transition to Up-land) and go up to elevation of 75.75 m (top ofmodel)

2. For other peak lake levels used to make “lastflooded” determinations, use lowest of four adja-cent quarter-month values (including peak).Starting with the next highest lake level follow-ing the highest peak and then moving sequen-tially to each more recent peak value, assignvegetation types as follows.

obvious deviations in the area percents from themean.

Drowned river mouth and barrier beach wetlandsare typically well-protected from wave attack. Theprotection features allow for thick sediment accu-mulation and result in a shallower topographic pro-file within the upper elevation contours of themodel range (Wilcox et al. 2005). In the context ofmodel sensitivity, the drowned river mouth modelperformed relatively stably. The barrier beachmodel showed large fluctuations.

AN ILLUSTRATION USING THEPREDICTIVE MODELS TO ASSESS

ALTERNATIVE WATERREGULATION PLANS

The predictive models for each of the four wet-land geomorphic types were tested using potentialregulation plans for Lake Ontario. The plans testedwere the current Plan 1958D with deviations(1958DD) and two plans (X and Y) developed byusing 1958DD as a base. The differences in lakelevel among any proposed regulation plans are dic-tated by how much water is released to the lowerSt. Lawrence River under any given net basin sup-ply (the amount of water entering Lake Ontariofrom its immediate watershed and from the upperGreat Lakes). Plans X and Y simulated releasingmore water to the lower river than did 1958DD in1900–1903, the 1920s, 1930s–early 1940s, 1960s,and late 1990s when basin supplies were low (IJCdata), thus resulting in more years with low lakelevels. More years with low levels were added inPlan Y than in Plan X (Fig. 9). The low lake levelsadded to Plan 1958DD were never lower than thosethat actually occurred during post-regulation; highlake levels never exceeded those of 1958DD. Basedon existing knowledge of moisture requirements ofthe plant species involved, as well as observationsand data from the Lake Ontario studies, testing ofthese plans should show an increase in area ofmeadow marsh (ABC) and a decrease in cattail(EF) vegetation as more low lake levels are added.

Predictions for Plan Y UsingDrowned River Mouth Wetland Model

As an illustration of the model procedures, wepresent the results for testing the drowned rivermouth wetland model using hydrologic input datafrom the test Plan Y, along with details regardingeach step in the process. First, we summarize the

766 Wilcox and Xie

FIG. 9. Hydrographs for Lake Ontario showing predicted lake levels if regulation plans1958DD, X, and Y were implemented using total basin supplies for the period 1900–2000.Plans X and Y were created from 1958DD but allowing more low lake levels to occur duringlow water-supply periods.


a. Last flooded > 30 years: assign to U (transi-tion to Upland)

b. Last flooded 5–30 years: assign to (ABC)c. Last flooded < 5 years or last dewatered in

growing season < 4 years: assign to (D)d. If the final most recent “last flooded” peak

year selected is < 5 years ago and its eleva-tion selected from the four highest quarter-month values is less than the most recentdewatered year elevation, use the single maxi-mum quarter-month value for making the“last flooded” determination.

3. Elevations below the smallest growing seasonpeak value in the entire plan: assign to G(never dewatered during growing season) and godown to elevation of 73.0 m (bottom of model).

4. For other lower growing season peak lake levelsused to make “last dewatered” determinations, use single maximum quarter-month value of the peak and assign vegetationtypes as follows.a. Last flooded < 5 years or last dewatered in

growing season < 4 years: assign to (D)

b. Last dewatered in growing season 4–39 years:assign to (EF)

c. Last dewatered in growing season 40 years ormore: assign to (G)

When the quarter-monthly lake levels for Plan Ywere entered into the mathematical routine of thedrowned river mouth model, the model started inyear 2000 to make assignments of wetland area/per-cent to portions of the drowned river mouth geo-metric model (see Figs. 5a and 9c, Table 3). Thesteps for making “last flooded” assignments in PlanY proceeded as follows (see Table 3): 1) identifiedthe highest lake level that occurs in the plan (75.69m, 1993; see Fig. 9c), determined the wetland area(and percent wetland) between that elevation andthe upper limit of the topographic/bathymetricmodel (75.75 m, 3.3%), and assigned the area/per-cent wetland to U (never flooded, transition to Up-land); 2) identified the next most recent peak lakelevel (1998), selected the lowest of the four quarter-monthly lake-level values surrounding and includ-ing that peak (75.33 m), determined the wetlandarea (and percent wetland) between that peak

TABLE 3. Model output for calculation of area and percent wetland in vegetation types U, ABC,D, EF, and G using 1900–2000 quarter-monthly lake level data from Plan Y and the drownedriver mouth model. Results are shown for year 2000, as well as the summary calculation thatappears at the end of 101 years of prediction output. Elevation is in meters International GreatLakes Datum 1985; area is based on the average total area in square meters between the modelboundaries of the eight wetlands used to generate the model. In the 101-year calculation, Sumand % refer to area.

Calculation For Year 2000:Year Elevation Years Class AREA Percentage

Flood 1993 75.69 Forever U 26750.63 3.31998 75.33 7 ABC 177353.55 22.12000 74.75 2 D 289877.61 36.1

Dewater 1964 74.59 Forever G 288686.69 35.91965 74.65 36 EF 7863.08 1.01999 74.70 35 EF 6552.57 0.82000 74.75 1 D 6552.57 0.8

CLASS U ABC D EF G TotalArea 26750.63 177353.55 296430.18 14415.65 288686.69 803636.69(%) 3.33 22.07 36.89 1.79 35.92 100.00

Calculation Results for 101 Planning Years:

Class U ABC D EF G TotalSum 10766821.34 13640033.98 20499700.06 6230592.84 30030157.83 81167306.06(%) 13.26 16.80 25.26 7.68 37.00 100.00

768 Wilcox and Xie

years or last dewatered in growing season < 4years).

The ‘last flooded” and “last dewatered” proce-dures were then repeated for each remaining year inPlan Y (1999–1900), with yearly assignments ofvegetation types to percentages of the drownedriver mouth geometric model. When evaluatingearly years in a regulation plan, there are limitednumbers of prior years from which to make calcula-tions. This problem was overcome by attaching acopy of the regulation plan at the beginning of theplan under evaluation because each regulation planis considered in the IJC study protocols to be de-rived from a repetitive sequence of 101-year netbasin supplies.

The model output is annual predictions of thearea and percent of vegetation/time classes that willoccupy the elevation range (73.0–75.75 m) given inthe model (Table 3). These predictions were thentime-weighted by summing the areas/percents foreach vegetation/time class and dividing by the num-ber of years analyzed. The final output is time-weighted percent of wetland expected to fall intoeach vegetation/time class during the period por-trayed by the regulation plan. In addition to the cal-culations shown in Table 3, summary tables aregenerated that display the four highest quarter-month values for each year and the elevation se-lected for analysis per Step 2 above (Table 4) andthe percent wetland assignments for each of the in-dividual 101 years evaluated (Table 5).

Comparison of Plan Results

As a requirement of the overall IJC study, a wet-land habitat performance indicator (Area ofMeadow Marsh Vegetation) was developed usingstudy results. This indicator was selected because itis sensitive to hydrologic change (Wilcox et al.1984, Keddy and Reznicek 1986, Keddy 2000), itrepresents a habitat that supports the greatest diver-sity of plant species, it can contain a diversity ofstructural habitats that support a wide range offauna, and it is the plant community shown to havebeen affected most by regulation of lake levels. Tomake the performance indicator more sensitive tothe hydrologic conditions that promote meadowmarsh expansion (low lake levels), calculations ofArea of Meadow Marsh Vegetation were made foronly those years in which low total basin suppliesprovided an opportunity for low lake levels tooccur. To meet these criteria, a time span beganfour years after the average quarter-monthly total

(75.33 m) and maximum peak value of 75.69 m(22.1%), determined the number of years from theyear being analyzed (2000) and the last peak usedto assign area/percent wetland (1993, 7 years), andassigned the area/percent wetland to ABC (lastflooded 5–30 years ago); 2i) identified the nextmost recent peak lake level (2000), recognized it asthe last peak < 5 years ago and with selected eleva-tion (74.75 m) less than the most recent dewateredyear elevation, invoked criterion 2d above and se-lected actual peak value for analysis (74.75 m), de-termined the wetland area (and percent wetland)between that peak (74.75 m) and last peak value as-sessed (75.33 m, 36.1%), determined the number ofyears from the year being analyzed (2000) and thelast peak used to assign area/percent wetland (1998,2 years), and assigned the area/percent wetland to D(last flooded < 5 years or last dewatered in growingseason < 4 years).

The steps for making “last dewatered” assign-ments proceeded as follows: 3) identified the lowestgrowing season peak lake level that occurs in theplan (74.59 m, 1964; see Fig. 9c), determined thearea (and percent wetland) between that elevationand the lower limit of the topographic/bathymetricmodel (73.0 m, 35.9%), and assigned the area/per-cent to G (never dewatered during growing season);4) identified the next most recent lower growingseason peak lake level (1965), determined the wet-land area (and percent wetland) between that peak(74.65 m) and the lowest growing season peakvalue of 74.59 m (1.0%), determined the number ofyears from the year being analyzed (2000) and thelast peak used to assign area/percent wetland (1964,36 years), and assigned the area/percent to EF (lastdewatered in growing season 4–39 years); 4i) iden-tified the next most recent lower growing seasonpeak lake level (1999), determined the wetland area(and percent wetland) between that peak (74.70 m)and the last growing season peak value assessed(1965, 0.8%), determined the number of years fromthe year being analyzed (2000) and the last peakused to assign area/percent wetland (1965, 35years), and assigned the area/percent to EF (last de-watered in growing season 4–39 years). 4ii) identi-fied the next most recent lower growing seasonpeak lake level (2000), determined the wetland area(and percent wetland) between that peak (74.75 m)and the last growing season peak value assessed(1999, 0.8%), determined the number of years fromthe year being analyzed (2000) and the last peakused to assign area/percent wetland (1999, 1 year),and assigned the area/percent to D (last flooded < 5


basin supply during the January–June period wasless than 7,100 m3/s and ended whenever the supplyexceeded 8,000 m3/s. This resulted in 22 individualyears being selected for use in calculating the per-formance indicator, which thus measures the com-parative ability of regulation plans to generate thelow growing-season lake levels required by themeadow marsh community during time periodswhen water supplies are low and low lake levels arepossible. Greater supplies in the remaining 79 years

would not have allowed an increase in the area ofmeadow marsh vegetation under any regulationplans that might be developed. When results fromthe entire 101-year regulation plan period are com-pared, there is an increase in the average percent ofmeadow marsh (ABC) vegetation from 1958DD(12.8%) to the plan with additional low lake-levelyears (Plan X: 14.2%) to the plan with even morelow lake-level years (Plan Y: 16.81%), which is theexpected result. However, when using the perfor-mance indicator that uses data only for the 22 yearswhen differences between regulation plans mightallow more meadow marsh to develop, the increasein average percent meadow marsh is more profound(1958DD = 18.4%; X = 24.2%; Y = 35.2%).

Testing the Model

We recognize the importance of also testingmodel results against independent data. However,no alternative regulation plans for Lake Ontariohave been implemented yet, and 101 years is a longtime to wait for data suitable for testing. Therefore,another approach was taken as the next best alterna-tive. In the initial phase of the IJC study, U.S. andCanadian researchers assessed vegetative change inthe 32 study wetlands, including eight drownedriver mouths, related to regulation using retrospec-tive photointerpretation of aerial photographs at ap-proximately decadal intervals from 2001 to themiddle/late 1950s (Ingram and Patterson 2003,Wilcox et al. 2003). The identifiable vegetationtypes that were ground-truthed in 2002 includedthose dominated by cattails and by sedges/grasses(meadow marsh), paralleling the EF and ABC cate-gorizations of vegetation types for transect sam-pling used to develop the model.

Unfortunately, there are inherent problems incomparing data sets of different origin. Retrospec-tive photointerpretation often loses resolution inolder photographs, as vegetation signatures may beless distinct in some historic photographs and back-tracking vegetation types by signature and locationis halted. In addition, sampling percent cover inquadrats along transects allows equal representationof taxa of short physical stature and taller, canopy-dominating taxa, while photointerpretation is gener-ally influenced primarily by those taller plants.Nevertheless, the photointerpretation data providedpotential for testing the model.

We ran the drowned river mouth wetland modeldescribed above using actual lake-level quarter-monthly data from 1900 to 2004. If the model oper-

TABLE 4. Model output for determination oflake-level in each year (1900–2000) of Plan Y thatis selected to represent the “last flooded” elevationused in the model. The annual peak and threeadjacent quarter-monthly levels are identified, andthe lowest of the four values is selected. Years1991–2000 are shown.

Water levels (m, InternationalGreat Lakes Datum 1985) Selected

for four quarter-months including lakeYear and surrounding the annual peak level

2000 74.75 74.75 74.75 74.73 74.731999 74.68 74.70 74.68 74.67 74.671998 75.36 75.38 75.37 75.33 75.331997 75.29 75.33 75.32 75.29 75.291996 75.18 75.19 75.19 75.18 75.181995 74.88 74.89 74.89 74.88 74.881994 75.09 75.11 75.12 75.11 75.091993 75.65 75.69 75.68 75.65 75.651992 75.15 75.14 75.13 75.12 75.121991 75.33 75.34 75.32 75.28 75.28

TABLE 5. Model output showing summary ofpercent wetland assigned to vegetation types U,ABC, D, EF, and G using 1900–2000 quarter-monthly lake level data from Plan Y and thedrowned river mouth model. Results are shown foryears 1991–2000 are shown.

Year U ABC D EF G

2000 3.33 22.07 36.89 1.79 35.921999 3.33 22.07 36.89 1.79 35.921998 3.33 0.00 50.96 9.79 35.921997 3.33 0.00 50.96 9.79 35.921996 3.33 0.00 50.96 9.79 35.921995 3.33 0.00 50.96 9.79 35.921994 3.33 0.00 36.76 23.99 35.921993 12.21 0.00 25.75 26.13 35.921992 12.76 16.03 9.16 26.13 35.921991 12.76 8.57 30.19 11.84 35.92

770 Wilcox and Xie

ates as expected, there could be a correlation be-tween the percentages of meadow marsh and cattailidentified by photointerpretation and the percent-ages predicted by the model for the same years de-spite the inherent problems with different data sets.To overcome the problems mentioned above, thedata were handled as follows. Photointerpretationdata from the Crooked Creek site were not used incomparisons of either meadow marsh or cattail be-cause this site was not used in model developmentdue to its status as an outlier in the ordinations(Wilcox et al. 2005). The Brush Creek and JordanStation sites were considered inappropriate forcomparisons for meadow marsh because theyshowed very little meadow marsh even in pre-regu-lation photographs, perhaps due to loss of resolu-tion during backtracking. Canopy influences onphotointerpretation were handled by two actions.We added the vegetation type mapped as shrubswamp to meadow marsh percentages because fieldobservations indicated that shrubs were generallyan invader of meadow marsh at higher elevations,with meadow marsh still present but masked byshrub cover in the photographs. We also recognizedthat the model distinguishes between D and EF veg-etation types because of understory differences, butthe canopy in both is dominated by cattails andwould have been mapped as cattails. Therefore, forcomparisons with photointerpreted cattail percent-ages, model predictions for D and EF were com-bined.

We conducted a simple linear regression to com-pare percent meadow marsh mapped by photointer-pretation in specific years with percent ABCpredictions from the model for those years. We alsoran a regression to compare mapped cattail withmodeled D + EF in specific years but confined theanalysis to post-regulation years because the modelwas constructed based on plant communities thatalready contained large areas of cattail associatedwith regulation (Wilcox et al. 2005). The predictivemodel is intended to be representative of the spec-trum of sites and is not expected to have site-specific accuracy; however, percent mappedmeadow marsh among specific sites did indeedshow a significant correlation with modeled percentABC (R2 = 0.266; p = 0.007) (Fig. 10). The lowvalue for R2 is likely a result of scatter caused bycomparing site-specific mapping data with modeldata based on averaged site geomorphometries thatare not intended to be site-specific. For example,three data points that would result in a reduced R2

had predicted (model ABC) values of zero; the ob-

served values of 6, 13, and 15% from mapping werefrom photographs taken in 1953 and 1954. Veryhigh lake levels in 1952 (Fig. 6) would cause themodel to predict no meadow marsh in subsequentyears, but if the individual morphometry of thosethree sites included some area that was not floodedtoo deeply in 1952, meadow marsh could have sur-vived and appeared in photographs from 1953 and1954. Averaged morphometry in the model is notexpected to duplicate actual morphometry at anyone site; existence of a significant correlation wasencouraging given those conditions. Percentmapped cattail was not significantly correlated withmodeled percent D + EF (R2 = 0.030; p = 0.415), asfew data points had small values for either mappedor modeled percent cattail, resulting in a clumpeddistribution of points. Floating mats of cattail oc-curred at some sites along the edge adjacent to thewater and likely do not respond to water-levelchanges. We also note that this use of photointer-preted changes in percent cattail as water levelsvary with time may not be a valid measure for com-parison with model results because most cattail in-vasion was landward rather than lakeward (Wilcoxet al. 2005). These regression results suggest thatthe performance indicator (Area of Meadow MarshVegetation) selected for use in evaluations of poten-tial new regulation plans for Lake Ontario was validfor drowned river mouth wetlands within the con-straints under which it was developed. Similarly,mapped percent meadow marsh for open embay-ment wetlands (R2 = 0.116; p = 0.037) and pro-

FIG. 10. Percent of wetland mapped as meadowmarsh in 26 aerial photographs from five drownedriver mouth wetlands (1953–2001) vs. percentABC vegetation type derived from running thedrowned river mouth model using actual quarter-monthly lake levels (1900-2004). Linear regres-sion: R2 = 0.266, p = 0.007, Y = 0.7969X + 6.928.


tected embayment wetlands (R2 = 0.107; p = 0.047)also showed significant correlations with modeledpercent ABC. The correlation with barrier beachwetlands (R2 = 0.065; p = 0.159) was not signifi-cant, which is not surprising because there is con-siderable variability in geomorphometry among thebarrier beach wetlands, and again, the model wasnot intended to apply to the geometry of any ofthem individually. Correlations of percent mappedcattail with modeled D + EF were not significantfor open embayment, protected embayment, or bar-rier beach wetlands, perhaps for reasons similar todrowned river mouths.

APPLICATION OF MODEL RESULTS

The regulation plans proposed for considerationby the International Joint Commission were evalu-ated by models for each wetland geomorphic typeas described above. The Integrated EnvironmentalResponse Model (IERM) developed by the IJC En-vironmental Technical Working Group also incor-porated these models (LimnoTech, Inc. 2005),along with performance indicators for many faunalgroups. However, the IERM converted the percent-ages of meadow marsh community across all geo-morphic types to area of meadow marsh for theentire Lake Ontario/Upper St. Lawrence Riverbasin by making use of the wetland inventory(Wilcox et al. 2005, Appendix A; view appendix at

http://iaglr.org/jglr/appendices/). That inventoryshows 9,157 ha of drowned river mouth wetland,7,002 ha of barrier beach wetland, 3,337 ha of openembayment wetland, and 6,352 ha of protected em-bayment wetland. We recognize, however, thatmodel results are not meant to apply to any specificwetland and that other factors may influence vege-tation changes at highly disturbed sites.

The abundance of other predicted wetland vege-tation types was also important in alternate planevaluation, as was the structural nature of the plantcommunities. Using quadrat data from samplingalong transects in 2003, the mean vegetative coverwas sorted by structural category to represent dif-ferent habitat types (e.g., Table 6; drowned rivermouth) (Wilcox et al. 2005) so that model resultsfor each regulation plan could be converted to per-cent of wetland in each habitat type. Those habitatpredictions were incorporated into several faunalmodels used in the overall IJC study and in theIERM. Faunal performance indicators such as theblack tern, Virginia rail, and least bittern reproduc-tive indices incorporated comparisons of the rela-tive supply of deep and shallow emergent marshhabitats among alternate water-level regulationplans. Yellow rail and king rail preferred breedinghabitat indices incorporated changes in plant com-munity structure. Models for spawning habitat sup-ply for various fish guilds and northern pike

TABLE 6. Mean percent cover by combined transects for unique structural groups of plants found indrowned river mouth wetlands (excludes Crooked Creek) derived from sampling quadrats along tran-sects A-G in 2003 (from Wilcox et al. 2005).

A, B, C D E, F, GMEAN COVER MEAN COVER MEAN COVER MEAN COVER

Structural Category ( 420 quads ) ( 140 quads ) ( 280 quads ) ( 140 quads )

Tree/shrub 18.99 3.78 0.07 0.00Vines 5.29 0.78 0.84 0.00Ferns 1.16 0.00 0.00 0.00Moss 0.07 0.00 0.00 0.01Forbs 28.74 4.77 0.57 0.00Grasses 13.71 8.19 8.19 0.09Sedges 7.96 1.96 0.26 0.00Broad Leaf Emergent 0.21 0.37 0.41 0.19Thin-Stem Emergent 2.97 0.25 0.39 0.02Thin-Stem Persistent Emergent 1.42 50.91 39.36 0.55Floating 0.00 8.68 25.73 20.99Submerged Broad-Leaf 0.00 0.00 0.00 0.55Submerged Narrow-Leaf 0.00 0.00 0.64 36.67Algae 0.00 0.00 0.00 21.18Miscellaneous 1.39 0.82 0.04 0.00Total Mean Cover 81.92 80.51 76.49 80.26

772 Wilcox and Xie

young-of-year net productivity also incorporatedthe relative availability of structural vegetationtypes (LimnoTech, Inc. 2005).

The development of quantitative relationships be-tween water levels and wetland plant communities,generalized geometric wetland elevation models,and estimates of wetland area within the study re-gion provides powerful predictive tools to evaluatepotential impacts of alternate water-level regulationplans on Lake Ontario-Upper St. Lawrence coastalwetland habitats. Manipulations of the current Plan1958DD water-level regulation criteria clearlydemonstrate that small changes in specific criteriacan have dramatic impacts on coastal wetland plantcommunities.

The regulation plans used for testing the predic-tive models were developed with recognition thatthe IJC Study Board and Commissioners must eval-uate the interests of all stakeholders and avoidundue impacts to any interest. Therefore, potentialregulation plans likely are viable options only ifthey do not exceed extreme lake levels that wouldotherwise be produced under the current regulationplan. From the wetland standpoint, viable plansshould place the frequency of high and low lakelevels in concert with total basin supplies. Such anapproach represents a realistic opportunity to ad-dress problems facing this important and complexecosystem.

ACKNOWLEDGMENTS

We thank Kurt Kowalski, Joel Ingram, andMartha Carlson for insights provided during devel-opment of these models. Reviews of an early draftof this manuscript were provided by Joel Ingram,Jan Keough, and Bill Werick. Jeff Schaeffer pro-vided statistical advice, and Martha Carlson pre-pared many data files used in our analyses, as wellas some of the figures in the manuscript. HeatherMorgan also prepared figures. Environment Canada(Canadian Wildlife Service) also provided figures.Field data used in developing and testing the modelwere collected jointly by USGS and EnvironmentCanada, with major input from Kurt Kowalski, JimMeeker, and Joel Ingram. Helpful suggestions wereprovided by three anonymous reviewers and Asso-ciate Editor Christiane Hudon. This manuscript iscontribution #1433 of the USGS-Great Lakes Sci-ence Center.

REFERENCESAlbert, D.A., Wilcox, D.A., Ingram, J.W., and Thomp-

son, T.A. 2005. Hydrogeomorohic classification forGreat Lakes coastal wetlands. J. Great Lakes Res. 31(Suppl. 1):129–146.

Baedke, S.J., and Thompson, T.A. 2000. A 4,700-yearrecord of lake level and isostasy for Lake Michigan. J.Great Lakes Res. 26:416–426.

Batty, M. and Xie, Y. 1994. Modeling inside GIS: part I:model structures, exploratory spatial data analysis,and aggregation. Int. J. Geogr. Inf. Sys. 8:291–307.

Berkhout, F., and Hertin, J. 2000. Socio-economic sce-narios for climate impact assessment. Global Environ-mental Change 10:165–168.

Burrough, P.A., van Rijn, R., and Rikken, M. 1996. Spa-tial Data Quality and Error Analysis Issues: GISFunctions and Environmental Modeling. In GIS andEnvironmental Modeling: Progress and ResearchIssues. eds. M.F. Goodchild, L.T. Steyaert, B.O.Parks, C. Johnston, D. Maidment, M. Crane, and S.Glendinning, pp. 29–34. Indianapolis, IN: GIS WorldBooks.

Crawford, R.M.M. 1982. Physiological response toflooding. In Physiological plant ecology II. encyclo-pedia of plant physiology, O.L. Large, P.S. Nobel, C.B. Osmond, and H. Ziegler, eds., pp. 453–477. Berlin,Germany: Springer-Verlag.

Environment Canada. 2002. Where land meets water:understanding wetlands of the Great Lakes, D.A.Wilcox, N. Patterson, T.A. Thompson, D. Albert, R.Weeber, J. McCracken, T. Whillans, and J. Gannon,contributors. Toronto, Ontario: Environment Canada.

ESRI. 2004. ArcGIS Spatial Analyst: Surface interpola-tion. http://www.esri.com/software/arcgis/exten-sions/spatialanalyst/about/features5.html.

Hardy, R.L. 1971. Multiquadric equations of topographyand other irregular surface. Journal of GeographicResearch 76:1905–1915.

Hudon, C., Wilcox, D.A., and Ingram, J.W. 2006. Mod-eling wetland plant community response to assesswater-level regulation scenarios in the Lake Ontario-St. Lawrence River basin. Environmental Monitoringand Assessment 113:303–328.

Ingram, J.W., and Patterson, N.J. 2003. Evaluation of theeffect of water level regulation on wetland plant com-munity abundance and distribution . IJC LakeOntario–St. Lawrence River Water RegulationReview Study Year 2 Report, International JointCommission, Ottawa, ON and Washington, DC.

Jackson, M.B., and Drew, M.C. 1984. Effects of flood-ing on growth and metabolism of herbaceous plants.In Flooding and plant growth, ed. T.T. Kozlowski,pp. 47–128. Orlando, FL: Academic Press.

Kadlec, J.A., and Wentz, W.A. 1974. State of the art sur-vey and evaluation of marsh plant establishment tech-niques: induced and natural. Volume 1: Report of


research. U.S. Army Corps of Engineers, Vicksburg,MS. D-74-9.

Keddy, P.A. 2000. Wetland ecology: principles and con-servation. Cambridge, UK: Cambridge UniversityPress.

———, and Reznicek, A.A. 1986. Great Lakes vegeta-tion dynamics: the role of fluctuating water levels andburied seeds. J. Great Lakes Res. 12:25–36.

Keough, J.R, Thompson, T.A., Guntenspergen, G.R., andWilcox, D.A. 1999. Hydrogeomorphic factors andecosystem responses in coastal wetlands of the GreatLakes. Wetlands 19:821–834.

Kozlowski, T.T. 1984. Responses of woody plants toflooding. In Flooding and plant growth , T.T.Kozlowski, ed., pp. 129–163. Orlando, FL: AcademicPress.

Lam, N.S. 1983. Spatial interpolation methods: a review.American Cartographer 10:129–149.

LimnoTech, Inc. 2005. Development of an IntegratedEcological Response Model (IERM) for the LakeOntario—St. Lawrence River Study. LimnoTech, Inc.,Ann Arbor, MI. http://www.limno.com/ierm/IERM_report_draft-042905.pdf

Longley, P.A., Goodchild, M.F., Maguire, D.J., andRhind, D.W. 2005. Geographic Information Systemsand Science. West Sussex, England: Wiley.

Makarovic, B. 1977. Composite sampling for DTMs.ITC Journal 1977-3:406–433.

Maynard, L., and Wilcox, D.A. 1997. Coastal wetlands:State of the Lakes Ecosystem Conference proceedings.Environment Canada, Toronto, ON and U.S. Environ-mental Protection Agency, Chicago, IL. EPA 905-R-97-015b.

McCullagh, M.J. 1991. Digital terrain modeling andvisualization. In Terrain modeling in surveying andcivil engineering, G. Petrie and T.J.M. Kennie, eds.,pp. 128–145. New York, NY: McGraw-Hill, Inc.

Penfound, W.T., Hall, T.F., and Hess, A.D. 1945. Thespring phenology of plants in and around the reser-voirs in north Alabama with particular reference tomalaria control. Ecology 26:332–352.

Petrie, G. 1991. Modeling, interpolation and contouringprocedures. In Terrain modeling in surveying andcivil engineering, G. Petrie and T.J.M. Kennie, eds.,pp. 112–127. New York, NY: McGraw-Hill, Inc.

Thompson, T.A., and Baedke, S.J. 1997. Strand-plainevidence for late Holocene lake-level variations inLake Michigan. GSA Bulletin 109:666–682.

Wei, P., and Chow-Fraser, P. 2005. Untangling the con-founding effects of urbanization and high water levelon the cover of emergent vegetation in Cootes Par-

adise Marsh, a degraded coastal wetland of LakeOntario. Hydrobiologia 544:1–9.

Wilcox, D.A.1995. The role of wetlands as nearshorehabitat in Lake Huron. In The Lake Huron ecosystem:ecology, fisheries, and management: Ecovision WorldMonograph series, M. Munawar, T. Edsall, and J.Leach, eds., pp. 223–245. Amsterdam, The Nether-lands: S.P.B. Academic Publishing.

———, and Meeker, J.E. 1991. Disturbance effects onaquatic vegetation in regulated and non-regulatedlakes in northern Minnesota. Can. J. Bot .69:1542–1551.

———, and Meeker, J.E. 1995. Wetlands in regulatedGreat Lakes. In Our living resources: a report to thenation on the distribution, abundance, and health ofU.S. plants, animals, and ecosystems, E.T. LaRoe,G.S. Farris, C.E. Puckett, P.D. Doran, and M.J. Mac,eds., pp. 247–249. Washington, DC: U.S. DOI,National Biological Service.

——— , Meeker, J.E., and Elias, J. 1992. Impacts ofwater-level regulation on wetlands of the GreatLakes. Phase 2 Report to Working Committee 2, IJCWater-Levels Reference Study, International JointCommission, Washington, DC.

———, Apfelbaum, S.I., and Hiebert, R.D. 1984. Cattailinvasion of sedge meadows following hydrologic dis-turbance in the Cowles Bog Wetland Complex, Indi-ana Dunes National Lakeshore. Wetlands 4:115–128.

———, Kowalski, K.P., Roten, H., and Carlson, M.L.2003. Relationships between Lake Ontario water lev-els and wetlands. IJC Lake Ontario–St. LawrenceRiver Water Regulation Review Study Year 2 Report,International Joint Commission, Ottawa, ON andWashington, DC.

——— , Ingram, J.W., Kowalski, K.P., Meeker, J.E.,Carlson, M.L., Xie, Y., Grabas, G.P., Holmes, K.L.,and Patterson, N.J. 2005. Evaluation of water-level-regulation influences on Lake Ontario and Upper St.Lawrence River coastal wetland plant communities:final project report. International Joint Commission,Ottawa, ON and Washington, DC.

Working Committee 2. 1993. Levels reference study,Great Lakes-St. Lawrence River Basin: annex 2 –land use and management. IJC Water-Levels Refer-ence Study, International Joint Commission, Wash-ington, DC.

Zhang, J., and Goodchild, M.F. 2002. Uncertainty inGeographic Information. Erewhan, North Carolina:Taylor and Francis.

Submitted: 4 December 2006Accepted: 28 June 2007Editorial handling: Christiane Hudon

predicting wetland plant community responses to proposed...

Documents