, Volume 32, Issue 6, pp 1175–1188

Forty Years of Change in the Bulrush Marshes of the St. Lawrence Estuary and The Impact of the Greater Snow Goose


    • Department of Applied GeomaticsUniversité de Sherbrooke
  • Richard A. Fournier
    • Department of Applied GeomaticsUniversité de Sherbrooke
  • Marcelle Grenier
    • Canadian Wildlife Service, Environment Canada
  • Josée Lefebvre
    • Canadian Wildlife Service, Environment Canada
  • Jean-François Giroux
    • Département des sciences biologiquesUniversité du Québec à Montréal

DOI: 10.1007/s13157-012-0347-z

Cite this article as:
Allard, M., Fournier, R.A., Grenier, M. et al. Wetlands (2012) 32: 1175. doi:10.1007/s13157-012-0347-z


During its spring and fall migrations, the Greater Snow Goose (Chen caerulescens atlanticus) stages in the marshes along the St. Lawrence Estuary in southern Quebec, where it feeds on three-square bulrush (Schoenoplectus americanus) rhizomes. The goose population has grown from 70 000 birds to around one million over the last 40 years, thus increasing pressure on these tidal marshes. To determine the impact of geese on the ecological integrity of the marshes over this period, we used IKONOS satellite imagery and aerial photographs to classify vegetation types. We estimated changes in bulrush cover using the eCognition image analysis software (Trimble). We examined the spectral, textural, and contextual characteristics of the identified classes. The proportion of bulrush cover has declined significantly in the lower marsh since around 1980, and bulrush has been gradually replaced by wild rice (Zizania aquatica var. brevis). We also documented the erosion between the lower and upper marshes along most of the shoreline.


Remote sensingBulrushMarshGreater Snow GooseObject-based classificationErosion


During its spring and fall migrations between the nesting grounds in the Canadian Arctic and the wintering areas on the East Coast of the United States, the Greater Snow Goose (Chen caerulescens atlanticus) population stages for a few weeks in southern Quebec. Its traditional migratory area is the St. Lawrence Estuary bulrush marshes. However, a population explosion over the last few decades is threatening the ecological equilibrium of this habitat. The snow goose population, estimated at 3,000 at the beginning of the last century, increased to 25,000 in the mid 1960s and between 800,000 to 1,000,000 individuals since 1998 (J. Lefebvre, unpubl. data). The snow goose eats the rhizomes of the Three-square (previously American) Bulrush (Schoenoplectus americanus), and several studies have indicated that the goose population had exceeded the support capacity of the marshes starting as far back as the early 1980s (Giroux and Bédard 1987; Lefebvre et al. 2001). Furthermore, the primary productivity of these marshes is estimated to have declined by more than 60 % (Giroux and Bédard 1987). This situation represents a source of stress for the other animal and plant species living in the marshes, as well as contributing to shoreline erosion.

Previous research on the spatial variation of wetlands shows adverse impacts on habitats and provides useful quantitative information for establishing regulations to protect habitat and guidelines for land-use management (Kent and Mast 2005). Beginning in 1999, a series of special conservation measures were implemented to stabilize the snow goose population size. One of the goal of those measures was to protect these unique marshes. Amendments to hunting regulations and habitat management have contributed to major changes in the species’ behaviour, as have its demographic explosion and degradation of the marshes (Bélanger and Lefebvre 2006). However, the traditional staging grounds are still being visited each year and the ecological integrity of the bulrush marshes is far from assured. Examining the spatial and temporal variations of bulrush populations may indicate areas that are at risk due to goose grazing pressure or other factors.

The impact of snow goose herbivory has often been evaluated using exclosures and plant sampling (Smith and Odum 1981; Giroux and Bédard 1987; Bélanger and Bédard 1994a). However, such studies have usually been conducted in a few localised marshes because of the intensive labour associated with above- and below-ground vegetation sampling (Giroux and Bédard 1988a). Monitoring the integrity of these coastal marshes subject to goose grazing may be facilitated by the use of remote imagery where field data can be used for training or validation (Miller et al. 1996).

A variety of remote sensing images are available in support of studies on spatio-temporal variability of wetland vegetation. However, developing a semi-automated image analysis method able to handle different images types and the spatial heterogeneity of wetland is a challenge (Fournier et al. 2007). A first limitation is the ability to handle digital images from various sensors and resolutions but also the different images types of film used for aerial photography. Other limitations include images acquired on different dates, the varying height of tides, and the scarcity of field data. The object-based classification method for image analysis stands out as a practical solution to handle multiple images and spatial heterogeneity (Kurtz et al. 2012). This method is frequently used to map wetland (Fournier et al. 2007; Grenier et al. 2007), however it needs to be adapted to each new context.

The goal of this study was to propose the object-based classification method to describe the spatial variation of Three-square Bulrush in the marshes of the St. Lawrence River over the last 40 years by mapping these habitats using satellite images and aerial photographs. In addition, we wanted to determine if it was possible to describe temporal variations in the distribution and abundance of two other important plant species in the marshes: wild rice (Zizania aquatica var. brevis) and arrowhead (Sagittaria spp.). In this paper, we developed guidelines for image processing when using different image types with various spatial resolutions over the span of 40 years.


Study Area

The bulrush marshes of the St. Lawrence Estuary are the staging grounds for the Greater Snow Goose and cover approximately 3750 ha (Giroux and Bédard 1987). Four traditional migratory staging grounds were studied: the marshes of Cap Tourmente National Wildlife Area (NWA), Montmagny, Isle-aux-Grues, and Cap Saint-Ignace, Quebec (Fig. 1). The Cap Tourmente NWA was created in 1978 mainly to protect the Three-square Bulrush marsh. Tens of thousands of geese can be found daily in these marshes for about a month in fall and for another in spring. Traditional fall hunting is allowed at certain well-defined locations within the marsh. On the south shore of the St. Lawrence River, the goose also represents an important source of economic activity. Since 1986, the Montmagny site has been designated as migratory bird sanctuary, a status that confers a particular protection to the birds and their staging grounds. In the present study, approximately 150 ha of marsh in Montmagny were examined, whereas a little less than 100 ha were studied at Cap Saint-Ignace. At Isle-aux-Grues, the marsh that rings the island is designated as a staging area for migratory birds. Approximately 300 ha of the lower marsh located on the north side of the island were included in this study. Marshes have two distinct zones: upper marsh and lower marsh. The lower marsh is covered daily by the tide while the upper marsh is covered only sporadically. A talus slope sometimes separates the two areas because of the impact of erosion by tides.
Fig. 1

Location of the study areas

Plant Community of the Bulrush Marshes

The lower marsh plant community is mainly composed of Three-square Bulrush, wild rice, and arrowhead (Giroux and Bédard 1987; Bélanger and Bédard 1994a). Three-square Bulrush, a member of the Cyperaceae family, is the traditional food source of the Greater Snow Goose during its migratory stopovers. It is a rhizomatous perennial grass that can reach nearly 2 m in height. Its stems are very resistant to wave action, whereas its rhizomes are resistant to erosion by ice. When the rhizomes are eaten by geese, the accumulated carbohydrate reserves of the bulrush decrease, which can be detrimental to shoot growth in the spring. This is a reason why spatial changes in this species of bulrush are potentially a good indicator of the impact of geese. Bulrush reproduces relatively easily, as rhizome fragments left by the geese in the fall are sufficient to produce a new plant the following spring. Geese eat the bulrush rhizomes by plunging their heads as deep as 20 cm into the mud (Bélanger and Bédard 1994b). This is when they have the biggest impact on the propagation of the plant (Bélanger and Bédard 1994a).

Wild rice, the second most dominant plant in the marshes, is an annual grass 1 to 2 m in height that grows in water at a depth of 0.5 to 1 m. Deeper water makes wild rice reproduction difficult; also, currents or wave action can sometimes uproot the plants. However, on the shores of the St. Lawrence River, the variety brevis has developed some resistance to harsh conditions, especially wave action. The third most dominant plant in the riparian marshes of the St. Lawrence is arrowhead, a perennial plant that usually does not exceed 1 m in height. It is typically found at the transition between water and terrestrial environments, being well-rooted and able to resist variations in water levels.

The marsh plant community composition is spatially and temporally variable. However, only the three main species were selected for classification because the other plants present, such as Torrey’s Bulrush (S. torreyi), Strong Bulrush (S. lacustris), Spike-rush (Eleocharis spp.), or Beggar-ticks (Bidens spp.), are not abundant enough to be identified using satellite imagery or aerial photographs. Although bulrush, wild rice, and arrowhead are difficult to differentiate by remote sensing before the end of June, their appearance can vary during the entire growing season. The flowering of wild rice in late July or early August is helpful in differentiating the species from bulrush and arrowhead. The flowering of bulrush and arrowhead offers no advantage for remote sensing because of their small flowers. The senescence of the plants near the end of the growing season (end of August, early September) can be a useful parameter for species differentiation. Wild rice plants become yellowish in colour earlier than bulrush, of which only the tips of stems show this colour change. However, from year to year, it is difficult to predict the exact timing of the changes observed during the growing season.

Available Data Sets

Satellite images and aerial photographs were selected for various years over the last four decades to measure the spatial variations in Three-square Bulrush over time. The classifications were done for at least three distinct time periods for each study area: 1960–1979, 1980–1999, and 2000–2010. For the last period, IKONOS images from 2002 (July 31 for Cap Tourmente NWA and August 23 for the other marshes) provided by Environment Canada were used. Although the 4-m resolution of the IKONOS images was not as high as that of the aerial photographs, they had the advantage of being multispectral. For selecting aerial photographs, the criteria were as follows: (1) the scale of the photos was preferably larger than 1:15 000 to be appropriate for the classification objectives, (2) the photos were of good quality (contrast between hues and values, sharpness, and absence of clouds), (3) the photos were taken during the summer when vegetation growth was sufficiently advanced, but before the geese had arrived for their fall staging, and (4) the photos were taken at low tide, so that the lower marsh would be visible. The period between mid-July and mid-September seemed the most appropriate for the identification of the vegetation groups in the marshes. It was deemed preferable to use the same time period for all the images in order to avoid confusion between seasonal variables and long-term decreases or increases (Jano et al. 1998).

The identification of single-species areas for the three plant species that were used as training areas in the classification was essential to obtain characteristics of their spectral values and texture, as well as the context of these areas in the analyzed images. Data collected in the field were used for validating the group classifications. Sketches of the plant cover in 30 m by 30 m plots with estimates of the percent cover of the three major plant species were made during August 2004 and 2005 at 50 locations distributed among the four marshes. Plots were randomly located after initial spatial stratification to ensure coverage of the entire study area. In addition, some single-species areas were sampled during the field work.

Data Analysis

We applied an accurate geometric correction to the aerial photographs which was important for a reliable comparison between the map information and the field data, as the plant cover of the lower marsh was very heterogeneous. A mosaic was not created with the aerial photographs, and the images were analyzed scene by scene. This was because radiometric differences, such as luminosity and hue, were too large to allow the adoption of general rules for entire flight lines. Radiometric corrections might have considerably reduced the dynamic amplitude. The maps were designed to take into consideration the spatial distribution of plants in the lower marsh, especially Three-square Bulrush. The lower marsh was delimited as a starting point for two strata: (1) the dominant plant species (bulrush, wild rice, or arrowhead) and (2) the percent cover of bulrush. The stratification by dominant species showed the most common species by measuring the concentration of the plant cover at each location in the marsh. In areas where there was a mixture of species, the dominant species determined the class attribution. Bulrush cover is a measurement of the concentration of the plant based on the cover observed in the image, thus yielding a percent cover of the area. Field sketches of plant cover played an important role in making the determinations based on images. Three classes of percent cover of bulrush were determined (low, medium, high) because the spectral resolution of the aerial photos and the rarity of field data did not allow for more narrowly defined classes.

The first step after acquisition of field data was to perform several directed classification tests to verify whether the spectral response of the different species could be discriminated. The results were promising for the satellite images. By contrast, the analysis by pixel was totally inefficient for the older aerial photos where two major problems were encountered: (1) the absence of field data, and (2) the low dynamic amplitude of the panchromatic images. A photo-interpretation identification of homogeneous areas (spatial objects) seemed the most efficient solution to adequately classify the different aerial photos. The object-based classification method is similar to this procedure by allowing segmentation of the images into several homogeneous areas and distinguishing among these areas by using over 100 spatial, textural, and contextual criteria (Blaschke 2003). The eCognition (Trimble) software was used for this analysis. When creating objects, the weight of the homogeneity criterion of color was higher than the shape criterion because, except for arrowhead, plant communities did not come together in a very specific way. As for the scale parameter, even if it varies from one image or one sector to the other, it was set to create small objects because of the heterogeneity of the plant communities. For example, about 100 pixels were grouped on average for each object resulting from the segmentation of the IKONOS images. The areas of the homogeneous polygons were compatible with the sketches of the plant cover drawn in the field.

During classifications, a top-down hierarchical approach was used meaning that general classes were first identified and then divided into sub-classes by adding several discrimination functions, either from attributes specific to each spatial object (mean, ratio, standard deviation), Haralick texture measurement, or relationship criteria with other classes at different levels. The principles of fuzzy logic were employed in this case because the limits between classes were not sharp enough for plant communities. At the lowest level of segmentation, the field data allowed for the creation of an identification key based on several characteristics unique to each class. Also, the Feature Space Optimization tool of the eCognition program identified membership functions that facilitated discrimination among classes. Several textural functions (Grey-Level Co-occurrene Matrix GLCM mean, GLCM standard deviation, GLCM correlation) applied to the near-infrared (NIR) band were particularly useful for discriminating the dominant species classes, whereas other functions (Ratio Red, GLCM correlation green and NIR, GLCM standard deviation red, minimum pixel value blue, compactness) allowed for the classification of the percent cover of bulrush. Next, the same identification key was used to classify other IKONOS images. However, it was impossible to create an identification key for the aerial photographs because they were produced using different types of film (i.e., colour, infrared, panchromatic) and their quality (i.e., contrast and sharpness) varied considerably. Thus, for the 43 aerial photos used for the classifications in all study areas, 43 separate identification keys had to be created. Also, the limited field data made the process long and difficult because the classifications could only be achieved by detailed photo interpretation.

The first classifications were made using the 2002 IKONOS images; the other classifications used aerial photographs, starting with the most recent images and progressing backwards in time to the oldest, because the process was based on the hypothesis that patches of bulrush found in 2002 should be found more or less at the same locations in preceding years since rhizomes remain at the same place from year to year. The rhizomes are thus responsible for the fact that the bulrush spatial distribution is stable through time, although patches can be larger or smaller depending on the effect of geese and other external factors. The goal was then to associate certain homogeneous areas created from the segmentation by eCognition to the classes required based on several criteria: mode of reproduction (annual or perennial), hue, value, texture, seasonal differences, spatial pattern, and context. These indices, as well as the observation of the behaviour of the membership functions used in the classification of the IKONOS images, were used to identify training areas on aerial photographs. The Feature Space Optimization tool of the eCognition program was then used to identify the membership functions that allowed the best separation among classes. Once the classifications were completed, the areas were calculated to provide the contribution of each to all the marshes in the study. Given changes in cloud cover, the area of a marsh, or other elements that could bias the comparison of areas among study years, a common area with data for all years was determined for each marsh in the study. The comparison of the relative areas of each class from the different time periods allowed for spatial variation within the marshes to be quantified. These results were then compared with the snow goose population size to evaluate the impact of herbivory on the marshes.


In order to validate the results of the classifications performed based on the IKONOS images, a new set of field data was collected in the summer of 2006. In all, 120 objects were identified using a stratified random sampling design. Confusion matrices were calculated using the field data (Jensen 2005). Normally, error matrices are created automatically by including a vectorial layer containing the validation data in an image processing program. This method is well adapted to very categorical classification results (e.g., water, forest, urban area), unlike for this study, where the two types of stratification that were evaluated have classes determined using percent cover. For the dominant species stratification, where a marsh area is composed of approximately 60 % wild rice and 40 % bulrush, the classification result would be influenced by the dominant species, which would be wild rice in this case. However, if the bulrush class is obtained following the classification process, it is possible to say that this result is not entirely false because this species certainly is present in the area. To remedy this situation, fuzzy logic analysis was used to eliminate categorical results (true or false) while considering results that were partially true or partially false (Lunetta and Lyon 2004).

Several authors have discussed the use of fuzzy logic to validate a classification project (Townsend 2000; Woodcock and Gopal 2000; Okeke and Karnieli 2006a, b; Grenier et al. 2008). This method allows to evaluate whether differences between a classification and field samples are acceptable or unacceptable. In the stratification by dominant species, three situations were possible. The category “exact” corresponded to situations where the class produced by the classification was the same as that observed in the field. The category “acceptable” corresponded to a situation where the dominant species determined by the classification was found in a proportion that was ≤20 % lower than the dominant species observed in the field. Any other situation resulted in the category “inexact.” A table of decision rules was produced where exact results were attributed a weight of 2 to accentuate their importance relative to results categorized as acceptable, which received a weight of 1; inexact results received a weight of zero. The results obtained were then multiplied by these weights to produce a fuzzy accuracy coefficient. The same principle was used for validating the results of the classification of the percent cover of bulrush.

The validation methods described above were applicable to classifications obtained from the IKONOS images because field data were collected within a relatively short time period (i.e., 4 years difference). This was not the case for classifications obtained from aerial photos dating from the 1970s and 1980s. For the most part, either no corresponding field data were available or the data were inadequate for the purposes of this study. To circumvent this lack of data, the two types of classifications from aerial photographs were compared against each other. For example, if the results of one classification at a specific location gave a dominance of wild rice, it was normally impossible to obtain the class “high” for the percent cover of bulrush with the other classification. A matrix was constructed to verify the probability of each situation by determining if they were “very likely,” “likely,” or “unlikely.”

Shoreline Erosion

For the marshes at Cap Saint-Ignace, Cap Tourmente NWA, and Montmagny, the limit between the upper and lower marsh was drawn by photo-interpretation for each year of the study. Shoreline recession was measured at 30 equidistant locations in each marsh for three periods and compared with repeated measures ANOVA. This allowed us to verify whether variations in the abundance of the plant species were caused by changes in the distribution of the lower marsh following erosion of the shore.


The classifications were used to create maps of the dominant species and the coverage of Three-square Bulrush within each marsh for three different periods of time (Figs. 2 and 3). The areas of each class were calculated to measure the range of variation over time.
Fig. 2

Classification of dominant species and bulrush percent cover (low, medium, high) for the Cap Saint-Ignace (top) and the Cap Tourmente NWA (bottom) study areas

Fig. 3

Classification of dominant species and bulrush percent cover (low, medium, high) for the Isle-aux-Grues (top) and the Montmagny (bottom) study areas

The comparison of the total areas of the marshes was important because it allowed for an assessment of the gains and losses in area during the period studied. Unfortunately, for certain years the sites were under high tide or cloud cover, thus limiting direct comparisons. Areas were therefore transformed into percentages of total area (Table 1).
Table 1

Area (ha) covered by dominant species (A) and bulrush (B) in four marshes along the St. Lawrence estuary. Percentages are shown in parentheses


Cap Saint-Ignace

Cap Tourmente NWA

















2.4 (3.5)

2.5 (5.7)

2.7 (4.3)

27.4 (10.6)

24.4 (10.5)

20.7 (9.0)

44.9 (19.1)

50.2 (20.9)

50.2 (21.3)

6.8 (9.8)

6.8 (7.3)

13.7 (13.2)


35.6 (79.5)

37.0 (75.7)

27.8 (58.0)

146.8 (73.5)

141.6 (67.0)

107.9 (53.8)

166.7 (71.1)

154.8 (67.0)

152.0 (65.9)

70.0 (70.0)

65.3 (71.2)

49.9 (60.8)

Wild rice

10.7 (17.0)

9.2 (18.6)

18.3 (37.6)

44.2 (15.9)

54.0 (22.5)

91.4 (37.2)

25.0 (9.8)

31.6 (12.1)

34.0 (12.7)

23.2 (20.2)

29.8 (21.5)

36.7 (26.0)



4.1 (5.9)

5.0 (6.4)

7.8 (19.4)

26.4 (12.5)

33.5 (17.0)

64.6 (29.6)

43.9 (21.8)

35.7 (18.5)

44.6 (19.0)

9.2 (12.0)

14.1 (17.3)

38.6 (31.2)


10.0 (20.8)

12.7 (31.4)

15.4 (26.4)

97.2 (43.0)

91.8 (40.2)

71.1 (30.0)

86.9 (35.9)

93.1 (38.9)

83.4 (33.5)

29.1 (25.1)

22.9 (23.5)

16.2 (14.5)


35.2 (73.3)

31.5 (62.2)

26.1 (54.2)

92.3 (44.4)

92.1 (42.9)

81.5 (40.4)

105.9 (42.3)

107.8 (42.5)

108.6 (47.4)

63.7 (62.9)

65.7 (60.4)

46.2 (54.2)

At all sites, except Isle-aux-Grues, the results showed reductions in bulrush and increased coverage of wild rice. However, this decrease appeared to have been relatively constant for Montmagny between the three periods. The decrease in bulrush at Isle-aux-Grues was very slight (loss of 6 % between 1978 and 2002), whereas the decrease in the “high” percent cover class of bulrush was 20 % at both Montmagny and Cap Saint-Ignace. At Cap Tourmente NWA, Cap Saint-Ignace and Montmagny, the percent cover of bulrush decreased in both ‘high’ and ‘medium’ density with an increase of the ‘low’ density. Also, no major variations were detected in the vegetation between goose hunting areas and staging areas. This may be explained by an increase in night-time feeding by the geese, during the fall and spring, within the hunting areas (Giroux and Bédard 1988b). Detailed maps of areas where marsh was present in each of the three study years were drawn for each of the test sites. This allowed for better observation of the spatial distribution and the intensity of changes in plant species and in the percent cover of bulrush in a fixed area of the marshes.

At Cap Saint-Ignace, the results of the classification by dominant species show that the replacement of bulrush by wild rice occurred through time from 1970 to 2002, mostly in the centre of the marsh, whereas the lower and upper limits of the marsh appear to have been less affected by this change (Fig. 2). The percent cover of bulrush changed from high to medium between 1989 and 2002. This increase in the abundance of wild rice was particularly high in the part of the marsh east of the cape, as well as in the area near the peninsula at the southern edge of the study area. The map of the percent cover of bulrush corroborates these results, showing that the percent cover also decreased near the centre of the marsh. We noted from the map that a narrow strip of bulrush, absent in 1970, recently appeared all along the limit between the lower and the upper marsh.

At Cap Tourmente NWA, the comparison of the classification by dominant species for the different years of the study shows a definite increase in wild rice and a decrease of Three-square Bulrush (Fig. 2). The transition from bulrush to wild rice is evident over almost the entire area, except near the lower limit of the marsh. The presence of a very narrow strip of bulrush also appeared along the shoreline between 1977 and 2002. The classification of the percent cover of bulrush shows a marked increase in areas of “low” coverage in 2002.

The comparison of the classification by dominant species at Isle-aux-Grues clearly shows the presence of a much higher amount of arrowhead than at the other three marsh sites (Fig. 3). For each year of the study, arrowhead was found throughout the marsh but particularly along tidal drainage channels. Also, although we generally found a narrow strip of dense bulrush at the lower and upper limits of the marsh, this phenomenon was far less pronounced at this site than at any of the other three study sites. Both the proportion of dominant species and the percent cover of bulrush remained relatively constant between the three periods. We also noted a slight increase in wild rice and arrowhead and a decrease of bulrush (i.e., decrease of 5 % for bulrush over 25 years).

At Montmagny, the classification by dominant species from 1965 to 2002 show an increase in areas with wild rice and a decrease of bulrush, mainly between the last two periods studied (between 1980 and 1999 and 2000-onward; Fig. 3). Similar to Cap Saint-Ignace, this transition appears most important in the areas located at the centre of the marsh and close to the shoreline, whereas the lower limit of the shore remained more stable. In 2002, over 30 % of the entire marsh had a low percent cover of bulrush, whereas only 12 % had a low bulrush cover in 1965.


An example of how to calculate overall fuzzy accuracy of the classification based on dominant species is presented for the Cap Tourmente NWA (Fig. 4). When calculated for each site, the validation results were 83.3 % for the Cap Saint-Ignace marsh, 80.0 % for the Cap Tourmente NWA marsh, 82.1 % for the Isle-aux-Grues marsh, and 60.0 % for the Montmagny marsh. For the classification of the percent cover of bulrush, the overall fuzzy accuracy was 60.0 % for Cap Saint-Ignace, 85.7 % for Cap Tourmente NWA, 42.3 % for Isle-aux-Grues, and 56.7 % for Montmagny. The results of both types of classifications (dominant species and percent cover of bulrush) were compared to introduce an additional element of validation to compensate for the absence of field data. The results of both classifications at Cap Saint-Ignace for the three study years showed that 60–71 % of the area of the marsh had mostly “likely” situations, whereas only 5 to 9 %, depending on the year, had “unlikely” situations. When comparing both types of classification for Cap Tourmente NWA, we found that 95 % of the area of the marsh had situations that were “likely” or “very likely” in 1977. This total exceeded 96 % for the classifications produced with 1984 aerial photos. The reliability of the classifications was slightly lower for 2002, where nearly 11 % of the area of the marsh consisted of “unlikely” situations. Comparing the results of both types of classification for Isle-aux-Grues resulted in 92 % of the classified marsh that had “likely” or “very likely” situations for 1978 and 1989. This total was slightly lower than 90 % for 2002. For the Montmagny marsh, the comparison showed that more than 60 % of the marsh had “very likely” situations in 1965 and 1978, whereas “unlikely” situations occurred for 8.2 % of the area in 1965 and for 11.3 % of the area in 1978. The area of “unlikely” situations was lower than 3 % in 2002, when 80 % of the area consisted of “likely” situations.
Fig. 4

Example of how to calculate fuzzy overall accuracy of the classification using dominant species. a An error matrix based on field data for classification by dominant species is first constructed to verify accuracy. For the numbers in bold, the classification was exact. For the numbers underscore, in a deterministic approach, they would be discarded. In a fuzzy logic, verification has to be done. The overall accuracy represents 67.8 % and the Kappa = 51.9 %, which represents a moderate agreement between the classification map and the ground reference information, b Underscore numbers are acceptable if the class defined by the classification is present in 20 % or less of the dominant class observed in the field. If not, they are classified has inexact. As a second step, the results are multiplied by the appropriate weight (Exact = 2, Acceptable = 1, Inexact = 0). c The results are summed up and divided by the maximum value that can be reached if all results were exact to yield the overall fuzzy accuracy (76.3 %)

Shoreline Erosion

Shoreline recession varied among marshes (F2,87 = 151.735, P < 0.001) and time (F2,86 = 122.857, P < 0.001). In general, erosion was more important at Cap Tourmente than at Montmagny and Cap St-Ignace and was much more important in recent years than during the first two periods (Table 2). At Cap Saint-Ignace, the delimitation line between the lower marsh and the upper marsh showed a recession of 18 m on average between 1970 and 2002, which represented 21 % of the original marsh surface. This recession was less important between 1970 and 1989 in comparison with the period 1989–2002, when soil erosion removed 0.8 m yr−1. At Montmagny, the strip of upper marsh that was lost accounted for 41 m or 18 % of the original surface. This reduction appeared to be particularly fast between 1970 and 1978, with a loss of 1.6 m yr−1. Finally at the Cap Tourmente NWA, the retreat of the limit was quite variable over the entire marsh but averaged 56 m between 1977 and 2002 or 17 % of the original surface (Fig. 5). The disappearance of the upper marsh appears to have accelerated during the last 20 years in comparison to the preceding 20 years (Table 2).
Table 2

Mean and standard deviation of the recession (m) of the limit between the upper and lower marsh measured at 30 equidistant locations at three sites along the St. Lawrence estuary during three periods. The recession rate per year for each time period is also presented


Cap Saint-Ignace

Cap Tourmente NWA











Mean (m) (σ)

3.6 (2.6)

3.4 (3.2)

10.8 (4.6)

11.3 (6.2)

11.2 (5.2)

33.0 (9.1)

3.5 (1.8)

12.8 (5.6)

25.2 (9.2)

Erosion rate/year (m)










Fig. 5

Changes in the upper limit of the lower marsh at the Cap Tourmente NWA showing shoreline recession between 1964 and 2002


The general objective of this study was to create maps for comparing spatial and temporal changes in the vegetation of bulrush marshes in relation to the grazing by Greater Snow Geese. Our results showed three major changes related to the ecological integrity of these marshes: (1) the decline of Three-square Bulrush, (2) the change in the spatial distribution of Bulrush, and (3) the extension of the upper limit of the lower marsh.

Evidence supporting the decline of Three-square Bulrush was found in the important reduction of its abundance observed at three of the four study sites: Cap Saint-Ignace, Cap Tourmente NWA, and Montmagny. At Cap Saint-Ignace, the decrease in bulrush occurred almost exclusively between 1989 and 2002. At Cap Tourmente NWA, the decline was equivalent to 0.7 ha per year between 1977 and 1984, and it has increased to 1.9 ha per year since 1984. At Montmagny, the decrease in bulrush seemed to have occurred more gradually because the decline rate remained relatively stable between 1965 and 2002. The demographic explosion of the Greater Snow Goose started in the 1980s and 1990s, which might explain the major bulrush decreases between 1984 and 2002 at both Cap Tourmente NWA and Cap Saint-Ignace, where pressure from goose grazing probably exceeded the renewal rate of bulrush. At the Montmagny site, the impact of the geese also seemed clear, but it is important to note that the decline in bulrush had already started in 1965, when the increase in the goose population was still relatively small. At that time, almost all of the geese fed within the marsh, as opposed to today, when a large part of their feeding also occurs on agricultural lands. This may explain why the geese seemed to have had such a negative effect on bulrush at Montmagny in the 1960s and 1970s. However, we cannot completely discard the potential role of maritime activity or climate changes. At Isle-aux-Grues, the lower decrease in bulrush area is particularly interesting when compared to other sites. This marsh does not have direct impact of the St. Lawrence seaway, because it is protected by islands on its northern side. This would have the effect of reducing the impact of waves from ship wakes, as well as reducing mechanical damage from ice floes and ice scouring in the spring. Therefore, these factors, as well as the goose demographic explosion, may be significant in the continued decline of Three-square Bulrush in other areas of the St. Lawrence Estuary.

Although it has been shown that bulrush have experienced major reductions over the years, it is still the most common plant species at all study sites. Large tracts of marshes still contain high densities of bulrush, in spite of the impact of geese grazing and the other factors that can affect the ecological integrity of these habitats. Our observations of the spatial and temporal variations in the dominant species showed that the bulrush decline was highest at the centre of the marshes. The results of the classification of bulrush percent cover showed that the class “high” has remained the most frequent at all study sites, whereas areas classified as having “medium” cover have generally declined to a “low” cover class over the years. However, in spite of all the recorded changes, we observed that bulrush was still practically the only species present at the lower limit of the marsh. This can be explained by the fact that the root systems of bulrush at this location are organized in a way that makes the plants more resistant to wave and tidal action. The rhizomes are intertwined, more deeply rooted, and the substrate is more difficult to penetrate (Bélanger and Bédard 1994b). It’s more difficult for geese to eat these rhizomes, which could improve survival of bulrush in these zones. Furthermore, bulrush seems to survive well at the limit between the lower marsh and upper marsh because it can tolerate the lack of water at low tide better than the other species like wild rice. In short, we generally saw a decline of bulrush in areas that were exposed to intense grazing by geese and prone to colonization by other plant species like wild rice and arrowhead.

Between 1970 and 2002 at Cap Saint-Ignace, the proportion of marsh dominated by wild rice increased from 22 to 38 %, whereas bulrush decreased by 16 %. At Cap Tourmente NWA, the increase in wild rice was 21 % between 1977 and 2002, whereas bulrush decreased by 18 %. The increased proportion of wild rice was 4 % at Isle-aux-Grues and bulrush declined by 6 %. At Montmagny, the decrease in areas dominated by bulrush reached 20 % between 1965 and 2002, which led to an increase of 13 % for wild rice and 7 % for arrowhead for the same period. With the exception of the Montmagny site, the areas dominated by arrowhead have remained relatively unchanged for the last four decades. At Montmagny, the proportion of arrowhead seemed to have grown considerably in 2002. However, it could be that a coarser segmentation for this site led to an overestimate of the area occupied by arrowhead, which was usually found in smaller patches. Given the stability of arrowhead populations, it is safe to assert that it was mainly wild rice that was gradually occupying the space formerly occupied by bulrush within the lower marsh. The regeneration of bulrush depends on the quantity of rhizomes left in the soil following grazing by geese. A minimal size of rhizome is necessary to ensure the growth of new bulrush shoots in the spring, and when it is not sufficient, the area can then be colonized by wild rice (Giroux and Bédard 1987). The areas located near the centre of the lower marsh seemed to be the most susceptible to this transition, particularly the areas dominated by bulrush, where it had a medium percent cover. The root system of bulrush favours its establishment at both the lower and upper limits of the lower marsh. In contrast, the ease with which tidal action can uproot wild rice plants and the difficulty for wild rice to tolerate the total absence of water, may explain in large part why the lower marsh showed some stability in bulrush density and dominance. If the goose population size continues to increase within the current climate conditions, it is most likely bulrush will be replaced by wild rice in the central areas of lower marshes along the St. Lawrence Estuary. Nevertheless, none of the main plant species present in the marshes seems in danger in the short or mid-term. However, it is fairly difficult to predict what effect the decrease in the abundance of bulrush could have on fauna or the plant’s capacity to multiply from a restricted number of individuals.

Shoreline Erosion

At Montmagny, approximately 9.3 ha of upper marsh along 3.5 km of shoreline disappeared between 1970 and 2002. At Cap Saint-Ignace, approximately 6.4 ha of the upper marsh disappeared during this same period. This represented a 0.5 % decrease per year. As for Cap Tourmente NWA, nearly 32 ha of upper marsh and riparian areas disappeared between 1964 and 2002, over a distance of approximately 9.8 km of shoreline. This phenomenon could have a devastating impact on the ecological integrity within the transition zone between the upper and lower marshes (Dauphin 2000). From the maps we produced, we observed that the upper marsh area that has disappeared is mainly being transformed to bare soil. Eventually, these areas may be colonised by Three-square bulrush through seed germination (Giroux and Bédard 1995). The total area of the lower marsh was therefore increasing to the detriment of the upper marsh at several locations within the study sites. The disappearance of the upper marsh could have devastating effects on the fauna and flora of this habitat, especially since the erosion of the shoreline talus is making this process practically irreversible in many areas. At Isle-aux-Grues, the erosion rate seems to be lower, based on the available information and on the fact that this marsh is the only not directly exposed to the St. Lawrence River seaway. Indeed, as is the case for almost all of the shores of the St. Lawrence River, it is the wave action from passing ships that is the main factor responsible for this erosion. The uprooting of vegetation during spring snowmelt (creating ice-scoured pools) and the early melting of ice caused by climate changes may also contribute to increased erosion. Ice blocks represent a very important source of erosion, as they sometimes remove part of the substrate (ice scouring) during ice break-up (Bélanger and Bédard 1994a). The modification of plant communities of the lower marsh, the variation in the water level, as well as the trampling of vegetation by hunters could also be factors contributing to erosion. Several authors believe that the activity of geese in the marshes represents an important local erosion factor, even though it does not represent the principal factor behind the erosion of the upper marsh (Lacombe 1982; Giroux and Bédard 1987; Bélanger and Bédard 1995; Dionne and Bouchard 2000). Bulrush, much more resistant than wild rice to tidal action, could have a beneficial effect by reducing the energy of waves before they reach the shoreline. The increase in the goose population could thus have an indirect negative effect by contributing to the recession of the lower limit of the upper marsh. Whatever the cause, the estimated rate of recession in the present study is comparable to the rates estimated by Dionne (1986), Dionne and Bouchard (2000), and Bernatchez and Dubois (2004).

The validation results of the classification by dominant species showed that three out of four study sites had satisfactory fuzzy precision coefficients that were above 80 %. However, at Montmagny, this coefficient was only 60 %. Field visits to the sites were made in 2005 and 2006, and the main trend that was observed was a slight decrease in wild rice in 2006 at all study sites, but particularly at Montmagny and Cap Tourmente NWA. This decrease in wild rice may have been sufficient to have had a certain impact on the dominance of spatial objects, where this dominance was previously shared between two or three species. Most of the fuzzy precision coefficients resulting from the classification of the percent cover of bulrush were relatively low, except at Cap Tourmente NWA, where the coefficient was 85.7 %. The other coefficients were low, ranging between 42 and 60 %, but certain phenomena may explain these results. First, as was the case for the classification by dominant species, annual variations in wild rice may have had a significant effect. Because the percent cover was evaluated based on aerial and satellite images, if tall wild rice plants were less numerous, bulrush plants would become more visible and the density of the bulrush cover would appear higher. In addition to differences between years, it is possible that phenological differences could have also caused identification errors in the analyzed images. Indeed, the flowering of wild rice as well as the height of the plants might obscure the bulrush underneath. Images taken at the end of June, after the appearance of bulrush but before arrowhead and wild rice, may reveal certain differences. In short, it is possible that the percent cover of bulrush near the end of the summer slightly underestimated its real abundance.


We used images with a very high spatial resolution and aerial photographs at a very large scale to classify images using the object-based approach for identifying plant species. In spite of some limitations, mostly related to the quality and the date of acquisition of the analyzed images, the object-based approach using the eCognition program was shown to be successful. It allowed us to identify spatial characteristics for separating several plant groups and distinguishing among species by their textural appearance. The statistical functions used in the classification were chosen because they can be easily applied to other areas for similar studies. This identification key could be a useful tool for map-based monitoring of the study area. The final results of this study showed a significant decrease in bulrush populations for three of the four study sites. The next step would be to develop rigorous methods to statistically compare changes over time.

The choice of image type to use is often dictated by financial resources. Here, four types of images were used; only the black and white aerial photographs should be avoided altogether in the future because of the limited spectral information that they offer. Satellite images have the advantage of offering more spectral information and reducing the analysis time by showing a large area within a single image. However, aerial photographs can provide more spatial detail (scales at 1:8 000 and 1:10 000, in this case) and are often a less costly option (when they already exist) than satellite images. At that scale, several aerial photographs are needed to cover the entire area, which greatly increases the analysis time and even a radiometric correction does not provide perfect uniformity among all photos. Ideally, the use of very high resolution satellite images, such as IKONOS or Quickbird images, should be the first choice. Several tests using RADARSAT-1 images (Fournier et al. 2007; Grenier et al. 2007) or polarimetric images in C-band (Touzi et al. 2007) could provide interesting additional information. Late August seems to be the most appropriate time for the classification of bulrush because the colour of wild rice at this time of year better discriminates it from bulrush. However, the precise time at which the different growth phases of plant species occur varies considerably from year to year, making it impossible to definitively state the best time of year for image acquisition. These recommendations should be considered to overcome the limitations encountered during this study, while attempting to make the best use of available resources.


This study is part of a larger project funded by the Arctic Goose Joint Venture. The IKONOS images used in this study were kindly provided by Guy Létourneau of the St. Lawrence Centre of Environment Canada in Montréal. Thanks also to Léo Provencher of the Department of Applied Geomatics at the Université de Sherbrooke for his most useful advice over the course of this project.

Copyright information

© Society of Wetland Scientists 2012