Differences in Breeding Bird Assemblages Related to Reed Canary Grass Cover and Forest Structure on the Upper Mississippi River

The floodplain forest of the Upper Mississippi River (UMR) hosts an abundant and diverse breeding bird community (Emlen et al. 1986; Grettenberger 1991) that contributed to the UMR's designation as a Globally Important Bird Area by the American Bird Conservancy (Chipley et al. 2003). Forest is the most prevalent terrestrial land cover type on the UMR, and, compared to adjacent upland forest, the composition of the breeding songbird community differs, and the relative abundance of birds is almost twice as great (Knutson et al. 1996, 1999). Notably, UMR floodplain forest supports more prothonotary warbler Protonotaria citrea, American redstart Setophaga ruticilla, brown creeper Certhia americana, and yellow-bellied sapsucker Sphyrapicus varius than upland forests during the breeding season (Knutson et al. 1996). The reasons for greater abundance of birds and different species composition in UMR floodplain versus upland forests have not been investigated, but they may be due to differences in primary productivity (e.g., Shure and Gottschalk 1985), food availability, habitat structure, and different rates of nest predation.

The UMR floodplain forest is composed of flood-tolerant tree species and has a fairly homogeneous structure compared to diverse species composition and structure in the upland forest due to the relatively flat topography (Kirsch et al. 2013). Most areas of UMR floodplain forest are mature, with closed canopies; a well-developed subcanopy; an open understory; tall and thick ground cover vegetation (primarily wood nettle Laportia canadensis); and an abundance of snags (Kirsch et al. 2013). This forest is also highly fragmented by aquatic areas and wetlands, but most landforms and islands within the UMR floodplain are thousands of years old (Knox 1996).

These forests were historically and are presently dominated by silver maple Acer saccharinum, with green ash Fraxinus pennsylvanica, elm Ulmus spp., river birch Betula nigra, and cottonwood Populus deltoides as frequent codominants or part of the subcanopy and understory (Knutson and Klaas 1998; Yin 1999). However, this floodplain forest has been influenced by a long history of extensive logging, conversion to agriculture and urban areas, and hydrological changes resulting from construction of channel training structures and the navigation system of 27 locks and dams between Minneapolis, Minnesota, and St. Louis, Missouri, in the 1940s. Importantly, completion and operation of the lock and dam system permanently flooded >50% of the remaining forest, and current flow regimes (Pinter et al. 2006, 2008) have caused the decline of swamp white oak Quercus bicolor, pin oak Quercus palustris, cottonwood Populus deltoides, and black willow Salix niger (Knutson and Klaas 1998; Urich et al. 2002).

Invasive organisms also have had a large effect on this forest. Notably, Dutch elm disease (caused by ascomycete microfungi and spread by elm bark beetle Hylurgopinus rufipes) killed large American elms Ulmus americana in the 1970s. Currently, elms only reach the subcanopy before dying (Romano 2010). Emerald ash borer Agrilus planipennis is now established in several areas on the UMR and threatens continued prevalence of green ash (and black ash Fraxinus nigra). Green ash is the second most dominant tree in the upper reaches (Kirsch et al. 2013); current flow regimes favor green ash regeneration, and it is the dominant sapling on the UMR (Yin et al. 2009). However, invasive reed canary grass Phalaris arundincacea is a threat to the entire forest because it prevents tree seedling establishment and growth. Reed canary grass is an aggressive invasive in many midwestern wetland systems (Galatowitsch et al. 1999; Lavoie et al. 2005) and widespread in the floodplain, occurring in wet meadows, along forest edges, and in canopy openings (Knutson and Klaas 1998; Urich et al. 2002; Romano 2010).

Reed canary grass is flood tolerant (Rice and Pinkerton 1993; Kercher and Zedler 2004a) and extremely competitive in nutrient-enhanced and hydrologically altered wetlands (Green and Galatowitsch 2002; Kercher and Zedler 2004b; Perry et al. 2004; Kercher et al. 2007) such as the UMR (Goolsby et al. 2001; Pinter et al. 2006, 2008). Reed canary grass grows rapidly, creating a large amount of biomass (Reinhardt-Adams and Galatowitsch 2005) that retards tree regeneration by effectively shading seedlings and occupying establishment sites (Knutson and Klaas 1998; Urich et al. 2002; Romano 2010; Thomsen et al. 2012). Reed canary grass invasion could have a widespread detrimental effect on floodplain forest sustainability as ash trees die, leaving canopy gaps. Furthermore, the dominant silver maples, a large proportion of which are around 75 y old, are expected to senesce and die in the next 50 y. Tree regeneration in UMR floodplain forest is currently weak, as there are few species other than green ash in the understory (Yin 1999; Urich et al. 2002; Romano 2010).

Altered hydrology, senescence of dominant silver maples, low rates of tree regeneration, and reed canary grass invasion are likely to drive a shift from largely closed canopy, mature floodplain forests to a savannah-like woodland with reed canary grass ground cover. There is great concern for the breeding bird community because this chain of events could occur over a large scale. As part of a study of bird responses to forest structure and differences related to landscape setting (Kirsch 2009), we assessed 1) reed canary grass distribution in relation to forest edge and interior areas and 2) whether breeding bird assemblages differed between UMR floodplain forest sites where reed canary grass dominated the ground cover vs. forest sites with little or no reed canary grass. We predicted that reed canary grass cover and frequency of occurrence would be less in forest interior areas than forest edges. We also predicted that birds would not respond to reed canary grass because this forest is naturally fragmented and the bird assemblage is largely composed of edge-tolerant and open-woodland species (Knutson 1995).

This study was conducted in the UMR floodplain between Hastings and Red Wing, Minnesota, and includes navigation Pool 3 and the upper part of Pool 4 (Figure 1). Typically, the upper portion of each UMR pool is a complex of floodplain forest and backwater sloughs, with relatively small ponds, lakes, and streams (Fremling and Claflin 1984). The lower portion is a large open expanse of water, with scattered small wooded islands (Fremling and Claflin 1984). The study area encompasses the mouths of the Vermillion and Cannon rivers (Figure 1). The Vermillion River runs parallel to the UMR within the UMR floodplain for ∼22km, from just southeast of Hastings, almost the entire length of Pool 3 with the mouth just below Lock and Dam 4. The Vermillion River is narrow and incised, with several backwater connections to the UMR along the upper half of Pool 3. The Cannon River flows almost perpendicular to the UMR from the west, but its delta is approximately 4 km wide and the mouth is below Lock and Dam 4, near Red Wing. The Cannon River is relatively shallow once it enters the UMR floodplain, and historically would dewater over its delta before reaching the UMR main channel.

We selected survey locations within interior forest patches, edges associated with interior patches, and remaining edge areas, by using geographic information system (GIS) coverages based on 2000 aerial photography (Dieck and Robinson 2004). We defined forest interiors as patches greater than 100 m from an edge with any other land cover type, and edge areas of forest were within 100 m of any edge. We used ArcGIS 8.1 (Esri, Redlands, CA) to first dissolve all forest polygons into one class of forest and then placed 100-m buffers inside forest edges. About 7% of the forest in the study area (108 patches) was more than 100 m from any edge; however, only 34 of these interior patches were 2 ha or greater (i.e., large enough for 150 × 80 m transects used to conduct spring songbird surveys; Kirsch 2009). Therefore, we restricted our survey site selection to these 34 patches.

We selected sample sites (points) to balance our ability to make inferences related to forest interior patches, edge associated with interior patches, and remaining forest areas that were not associated with an interior patch (remaining edge areas). We randomly selected 17 of the interior forest patches and paired an edge location with each (i.e., areas within the 100-m buffer around the interior forest patch). Sampling locations in interior patches were placed 50–150 m interior from the innermost border of the 100-m buffer. We selected the location to sample within the associated 100-m buffer by randomly choosing a number between 0 and 360 (in increments of 10) to indicate a radial direction from the center of the interior patch, and then placed the sampling location 50 m from the edge along the selected radius. To make inference to the rest of the forest area, we randomly selected 17 additional points in remaining forest. We refer to these sites as “random sites” for simplicity, and they were placed at least 50 m from any edge. Pairs of interior patch and associated edge sampling sites were a minimum of 250 m from any other interior, edge, or random sites.

We conducted point-center quarter sampling (Mueller-Dombois and Ellenberg 1974) during late June 2008 at plots located at 0, 50, 100, and 150 m along transects in each sampling location. Transects started at the sampling locations (points) selected above. Transects in interior patches ran parallel to the long axis of each patch; paired edge transects were parallel to the edge. Paired interior and edge transects were a minimum of 100 m from each other. Transects that originated at random points were arbitrarily oriented, but they remained at least 50 m away from any edge. Each plot was divided into four 90° radial subplots. Within each subplot, we recorded species, diameter-at-breast height (dbh), and distance from the center of the plot for the tree (>8 cm dbh) and sapling (≤8 cm dbh) closest to the plot center. To avoid potentially double counting trees or saplings in adjacent plots, we did not sample beyond 25 m from the plot center. We counted the number of standing snags (>8 cm dbh and >2 m tall) within 25 m of each plot center. We estimated basal area at each plot by using a size 10 angle gauge. We estimated height of a representative canopy tree and understory sapling or shrub within 25 m of the plot center by using a clinometer. We visually estimated percentage of cover of herbaceous vegetation, including reed canary grass, within a 10-m circle around the plot center (Mueller-Dombois and Ellenberg 1974). At originally selected points, we visually estimated cover of canopy and ground vegetation layers within a 10-m circle after-each-bird-survey (Mueller-Dombois and Ellenberg 1974).

We surveyed breeding birds from 4 June to 2 July 2008 at the originally selected points by using 50-m radius, 10-min point counts (Ralph et al. 1993). We ensured that point counts in interior patches were at least 125 m from point counts in the associated edge. We recorded all birds seen and heard and mapped their approximate locations of the first detection within 50 m. Three observers conducted bird surveys and alternated edge and interior surveys so that edge sites were not always surveyed first. We conducted surveys between 30 min and 4 h after civil sunrise, and we surveyed each point three times. We did not conduct surveys in rain, sustained wind greater than 12 kts, or dense fog.

We created two data sets for habitat and bird data that were then used in different steps of multivariate analyses. Habitat data included averaged estimates of tree, basal area, number of snags, and reed canary grass cover and estimated tree importance values from the four plots from each transect, and canopy cover and ground cover estimated at point count locations. We estimated importance values for each tree species by transect by using data from the four point-center quarter plots (distance and dbh) as the sum of relative abundance, relative frequency, and relative dominance (Curtis 1959; Mueller-Dombois and Ellenberg 1974). We combined importance values for green and black ash and dropped several tree species from the data set because they occurred on fewer than five transects. We used a square-root transformation to normalize importance values. We used PRIMER-E version 6 (Clark and Gorley 2006) to create the second habitat data set, a resemblance matrix, by first normalizing variables and then estimating Euclidian distances among sites. For the bird data, we averaged counts of each species over all three surveys. We square-root transformed average counts to down-weight abundant species, and we retained all rare species. The second bird data set was a resemblance matrix among sites created using Bray–Curtis similarities in PRIMER-E.

We used PRIMER-E for nonparametric multivariate and permutation analyses. We characterized habitat gradients at sites with principal components analysis (PCA). Variables that yielded the first three principal components (PCs) accounting for the most variation in the data were average basal area, tree height, number of snags, reed canary grass cover, and importance values for tree species. We determined degree to which site similarities based on bird data matched site similarities based on habitat data with the RELATE routine and 1,000 random permutations to test the null hypothesis that there was no relationship between habitat and bird similarity matrices (ρ = 0 when the null hypothesis is true). We used LINKTREE to find splits in the bird community data that best corresponded with breaks in values of habitat variables (Clark et al. 2008). The first split separates sites based on bird data that are most clearly related to a break-point for a habitat variable. Successive splits become less definitive than each prior split. We used the SIMPROF permutation test (similarity profile, which was run 1,000 times on the bird data after each split) to consider only those splits that differed from a null model of no difference in bird assemblages between the newly split group and the remainder of sites. We report analysis of similarity (ANOSIM) R, the maximized high-dimensional separation between groups, and B%, the percentage of difference between the split group and the remainder of sites in terms of the original similarity matrix. With each split, B% declines because the split group and the remaining group become more similar; hence, it becomes more difficult to discern which bird species may have accounted for the differences upon which later splits were made.

Strong separations of sites were based on ground cover less than 15% and reed canary grass cover greater than 53%. These two splits yielded three groups of sites that coincided with areas that were hydrologically influenced by each of the three rivers in the study area. We used ANOSIM to examine differences among sites associated with each river based on habitat variables. We used nonmetric multidimensional scaling and plotted sites labeled by river to visualize differences among bird assemblages. We used the SIMPER routine to examine which bird species contributed to assemblage differences related to reed canary grass cover, but excluded the four Vermillion River sites because LINKTREE results suggested these bird assemblages were different from those in the remainder of sites. SIMPER decomposes Bray–Curtis average dissimilarities between all possible pairs of sites between groups into percentage contributions by each species (Clark and Gorley 2006: 140) and is considered an exploratory analysis of which species contribute to differences between groups or are indicative of a group (Clark and Warwick 2001). Finally, we examined differences in average abundance and species richness, diversity, and evenness (the latter two factors estimated in PRIMER-E) among the three groups of sites based on 95% confidence limits of group estimates.

We used SAS version 9.1 (SAS Institute 2003) to model site occupancy and detection probability (MacKenzie et al. 2006) of the 26 most common bird species. We modeled the effect of more reed canary grass (>53% cover LINKTREE breakpoint), vs. less reed canary grass cover, again excluding the four Vermillion River sites. There were four models for each species. We varied occupancy, detection probability, or both with more or less reed canary grass cover for the first three models, and we held occupancy and detection constant for the fourth model. We compared these models by using Akaike's Information Criterion for small samples sizes (AIC_c; Burnham and Anderson 2002), where models with ΔAIC_c (i.e., the difference in AIC_c values between the current and the most appropriate model) < 2 had the most support for being the best among the set, models with 2 < ΔAIC_c < 10 had less support, and models with ΔAIC_c > 10 had almost no support.

Reed canary grass occurred on 63% of transects (32 transects: 11 interior, 12 edge, and 9 random). Average cover of reed canary grass on transects was similar among interior, edge, and random sites (interior mean = 19.3%, SD = 31.37; edge mean = 19.7%, SD = 33.48; random mean = 14.3%, SD = 28.10). Reed canary grass occurred on almost 33% of all breeding bird survey plots (16 plots: 8 interior, 5 edge, and 3 random).

The first three PCs accounted for 56.8% of the variation in the habitat data (Table 1; Figure 2). The first PC (28.8% of the variation) depicted a gradient from sites with high reed canary grass cover and black willow importance values to sites with high basal area, canopy cover, and tree height. The second PC (18.1% of the variation) depicted a gradient from sites with more tree species and greater elm and box elder Acer negundo importance values to sites with greater silver maple importance values. The third PC (9.9% of the variation) depicted a gradient from sites with greater hackberry Celtis occidentalis importance values to sites with greater cottonwood importance values.

Based on results from LINKTREE analysis (see below), habitat and bird similarity matrices were most similar within three groups of sites, with each hydrologically associated with one of the three rivers in the study area. We did not anticipate that sites would be grouped by River, but evidence for these groups based on habitat variables was strong (ANOSIM R = 0.52, P < 0.001; Figure 2). Habitat on Cannon and Vermillion River sites differed from each other the most (ANOSIM R = 0.75, P < 0.001), followed by UMR compared to Vermillion River sites (ANOSIM R = 0.67, P < 0.001), and then UMR compared to Cannon River sites (ANOSIM R = 0.43, P < 0.001).

We identified 2,999 bird detections to 56 species: 24 Neotropical migrant, 19 short distance migrant, and 13 resident species. American redstart, house wren Troglodytes aedon, yellow warbler Setophaga petechia, American robin Turdus migratorius, warbling vireo Vireo gilvus, and common grackle Quiscalus quiscula comprised just over 53% of all detections. Site resemblance matrices of breeding bird assemblages and habitat variables were significantly related (ρ = 0.50, P < 0.001), and LINKTREE analyses revealed clear splits in the bird data based on habitat variables. The first split, ground cover less than 15% (R = 0.73, B% = 88), separated the four Vermillion River sites from the remaining sites. The second split, reed canary grass cover greater than 53% (R = 0.39, B% = 61), separated nine sites that included only those most influenced hydrologically by the Cannon River. The third split, ground cover values between 60 and 15% (R = 0.39, B% = 56), separated five sites, all associated with the UMR.

Post hoc nonmetric multidimensional scaling analyses revealed the differences among sites grouped by River (Figure 3; minimal two-dimensional stress = 0.25, in 25 of 100 runs). Vermillion River sites are on the far left and do not overlap with other sites, whereas the Cannon River sites are on the right periphery of the distribution, but a few slightly overlap the central distribution of UMR sites. Bird assemblages in sites with more vs. less reed canary grass cover were 42.9% dissimilar (SIMPER analysis). Fourteen species accounted for 52% of the cumulative percentage of dissimilarity (Table 2), with abundance of common yellowthroat and American redstart differing the most between sites with more vs. less reed canary grass cover (Table 2). Although the number of detections (abundance) per site did not differ among the three river areas (Table 3), species richness, diversity, and evenness were lower in Vermillion River vs. Cannon River and UMR sites (Table 3).

Comparison of occupancy models revealed a likely effect of reed canary grass cover on occupancy (Ψ) or detection probability (p) for 7 of the 26 species examined (Table 4; Table S1, Supplemental Material). Detection probability in sites with more reed canary grass was greater for song sparrow Melospiza melodia, black-capped chickadee Poecile atricapillus, Baltimore oriole Icterus galbula, and rose-breasted grosbeak Pheucticus ludovicianus, and reduced for American redstart, great crested flycatcher Myiarchus crinitus, and yellow-bellied sapsucker (Table 5; Table S1, Supplemental Material). Occupancy probability in sites with more reed canary grass was greater for common yellowthroat Geothlypis trichas, black-capped chickadee, and rose-breasted grosbeak, and reduced for Baltimore oriole and great crested flycatcher (Table 5; Table S1, Supplemental Material). For the remaining 19 species the model that held both Ψ and p constant was competitive, with ΔAIC_c < 2 (Table 5; Table S1, Supplemental Material); and for 15 of these species, the model that held Ψ and p constant had the greatest support (ΔAIC_c = 0).

Counter to our expectations, reed canary grass occurrence and coverage in forest interior, edge, and random sites were similar. Multivariate analyses revealed that differences in bird assemblages were related to reed canary grass cover (53% cover breakpoint). Bird species associated with sites that had more reed canary grass were common yellowthroat, black-capped chickadee, and rose-breasted grosbeak; those associated with sites that had less reed canary grass were American redstart, Baltimore oriole, and great crested flycatcher. However, reed canary grass cover did not seem to affect overall bird abundance, species richness, diversity, or evenness.

The increased likelihood of common yellowthroat presence and greater abundance in areas with more reed canary grass cover makes sense because this species nests in thick grassy vegetation in a wide variety of habitats (Guzy and Ritchison 1999). However, explaining responses of bird species that nest in trees in woodland or forest habitat based on increased cover (and presence) of reed canary grass is not as straightforward. The LINKTREE analysis revealed that the bird community was associated with reed canary grass, and not with a forest structure variable such as basal area, canopy cover, or tree height. However, PCA revealed that reed canary grass cover was negatively associated with these variables. Considering these relationships, we compared our results to those from studies that compared bird assemblages in forests with differing canopy cover such as oak forest to savannah gradient (Davis et al. 2000; Brawn 2006; Grundel and Pavlovic 2007a, 2007b; Au et al. 2008), canopy gaps created by disturbance in mixed deciduous forest (Canterbury and Blockstein 1997), and timber management in eastern hardwood forests (DeGraaf et al. 1991; Germaine et al. 1997; Moorman and Gynn 2001).

Our observations for several species were similar to those reported in the literature. Common yellowthroat and rose-breasted grosbeak were more abundant in sites with more reed canary grass cover on the UMR. These species are more abundant in open canopy savannah, disturbed forest, and managed forest plots (thinned or group selection) vs. closed canopy woodland or control forest plots (DeGraaf et al. 1991; Canterbury and Blockstein 1997; Germaine et al. 1997; Davis et al. 2000; Moorman and Gynn 2001; Brawn 2006; Grundel and Pavlovic 2007a; Au et al. 2008). Black-capped chickadees were also more abundant in sites with more reed canary grass cover. Black-capped chickadees are generally not sensitive to canopy openness (Foote et al. 2010), although Au et al. (2008) found they are less common in savanna than burned and unburned woodland in Minnesota. American redstart and great crested flycatcher were less abundant in sites with more reed canary grass cover. American redstarts are more abundant in oak woodland (Au et al. 2008), undisturbed areas of forest (Canterbury and Blockstein 1997), and uncut forest (DeGraaf et al. 1991). Great crested flycatchers are less abundant in open canopy savanna than closed canopy woodland in Minnesota (Davis et al. 2000; Au et al. 2008). However, in Illinois, great crested flycatchers are more common in savanna than closed canopy forests (Brawn 2006). Baltimore orioles also were negatively associated with reed canary grass cover; however, they usually are associated with open canopy and savannah habitat (DeGraaf et al. 1991; Canterbury and Blockstein 1997; Davis et al. 2000; Brawn 2006; Au et al. 2008).

It seems notable that Baltimore oriole, a species of open woodlands, as well as American redstart and great crested flycatcher avoided UMR forest with more reed canary grass cover. Perhaps the structure of reed canary grass ground cover is in some way undesirable for these species. Thick reed canary grass cover would hinder one of the great crested flycatchers' modes of foraging, dropping to the ground from a perch to trap prey (Gabrielson 1915), although they usually do not spend much time on the ground. Reed canary grass cover would also be undesirable for keeping track of newly fledged great crested flycatcher, Baltimore oriole, and American redstart chicks, which often end up on the ground (Lanyon 1997; Sherry and Holmes 1997; Rising and Flood 1998). Thick thatch of reed canary grass cover would hinder parental care and perhaps increase chances of predation. As opposed to reed canary grass, wood nettle Laportea canadensis, the most common ground cover in UMR floodplain forest, has broad spreading leaves near the top of the plant, but it is open with virtually no litter cover underneath. Although abundance and presence of a few species were affected by reed canary grass cover, diversity and abundance of birds at sites were not. This suggests that current forest conditions associated with more reed canary grass cover, and the amount of forest in this condition, were not yet having a large effect on the breeding bird assemblage.

The influence of reed canary grass cover on the breeding bird assemblage was not as strong as lack of ground cover, which was caused by prolonged flooding in spring and early summer. All of the sites with virtually no ground cover (<15%) were on the Vermillion River. These sites hosted most of the prothonotary warblers, which favor nesting in cavities over open water (Petit and Petit 1996). But we detected fewer species at these sites perhaps because of mature forest structure (large basal area, tall closed canopy) and flooding (Knutson and Klaas 1997).

Projected future tree loss and canopy opening along with increases in reed canary grass occurrence and cover may at some point cause a greater shift in bird species composition. Controlling reed canary grass and promoting forest sustainability in this system will be difficult because altered hydrology and increased nutrient inputs from upland agriculture are large-scale problems; new propagules arrive with every high water event. In addition to the formidable invasiveness of reed canary grass in canopy gaps and open areas, it can also persist in shaded areas. We found dense stands of reed canary grass in areas with more than 90% tree canopy cover. Although it is thought that reed canary grass cannot gain a foothold or persist under a closed canopy (Maurer and Zedler 2002; Hovick and Reinartz 2007), these reed canary grass stands may have been composed of clones from plants in sunny areas that may subsidize the parts of the clone in shade (Maurer and Zedler 2004).

Numerous reed canary grass control and reforestation methods are available (e.g., Lavergne and Molofski 2006), but repeated chemical control will be necessary (Merriam 2014). Integrated, multimodal, and repeated management can be successful in restoring herbaceous and woody floodplain forest vegetation on a small scale (Lavergne and Molofski 2007; Thomsen et al. 2012). Other efforts on the UMR may have failed because they did not mechanically pretreat sites to remove thatch, break up rhizomes and expose bare ground, apply pre-emergent herbicides in fall, and follow up with grass-specific herbicide during the growing season (Thomsen et al. 2012). Managing hydrology is also important, but not yet specifically applied on the UMR (Lavergne and Molofski 2007). The U.S. Army Corps of Engineers and others have discussed using dredge spoil to bury reed canary grass, raising the elevation of sites, and then seeding with desired tree species (R. Urich, U.S. Army Corps of Engineers, St. Paul District, personal communication).

Our results are the first we are aware of to report bird assemblage shifts associated with reed canary grass cover in forests. Continued research and monitoring are necessary to verify the relations with reed canary grass that we observed because funding was available for only 1 y. To date, bird monitoring is not included in forest monitoring plans or in pre- and post-treatment monitoring, including treatments that seek to control reed canary grass and restore floodplain forest. Growing interest among resource managers on the UMR concerning sustainability of the floodplain forest, especially as an important habitat for birds, may improve opportunities to collect more breeding bird, forest structure, and reed canary grass cover data through research or as part of a forest monitoring program.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Table S1. Model results for occupancy and detection estimates of breeding birds with reed canary grass cover on Upper Mississippi River floodplain forest, 2008. Sites with more reed canary grass had more than 53% cover of reed canary grass (Cannon River sites, n = 10) and sites with less reed canary grass had less than 53% cover of reed canary grass (Upper Mississippi River sites, n = 37). Vermillion River sites (n = 4) were excluded.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S1 (81 KB XLS).

Reference S1. Dieck JJ, Robinson LR. 2004. Techniques and methods book 2, collection of environmental data, section A, biological science, chapter 1, general classification handbook for floodplain vegetation in large river systems. U.S. Geological Survey, Techniques and Methods 2 A–1.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S2 (27886 KB PDF).

Reference S2. Kirsch EM. 2009. Ecological survey of bird populations and floodplain forest in the Vermillion/Cannon River Bottoms Important Bird Area. Final report prepared for the Minnesota Department of Natural Resources, St. Paul, Minnesota.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S3 (1709 KB PDF).

Reference S3. Knutson MG, Hoover JP, Klaas EE. 1996. The importance of floodplain forests in the conservation and management of Neotropical birds in the Midwest. Pages 168–188 in Thompson FR, editor. Management of midwestern landscapes of the conservation of Neotropical migratory birds. U.S. Department of Agriculture North Central Forest Experiment Station General Technical Report NC-187.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S4 (2934 KB PDF).

Reference S4. Ralph CJ, Geupel GR, Pyle P, Martin TE, DeSante DF. 1993. Handbook of field methods for monitoring landbirds. Pacific Southwest Research Station, Forest Service, U.S. Department of Agriculture, Albany, California, General Technical Report PSW-GTR-144. Albany, California: Pacific Southwest Research Station, forest Service, U.S. Department of Agriculture.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S5 (1503 KB PDF).

Reference S5. Urich R, Swenson G, Nelson E. 2002. Upper Mississippi and Illinois River floodplain forests: desired future and recommended actions. Rock Island, Illinois: Upper Mississippi River Conservation Committee.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S6 (3354 KB PDF).

Reference S6. Yin Y. 1999. Floodplain forests. Pages 9.1–9.9 in U.S. Geological Survey, ecological status and trends of the Upper Mississippi River system 1998: a report of the long term resource monitoring program. La Crosse, Wisconsin: U.S. Geological Survey, Upper Midwest Environmental Sciences Center. LTRMP 99-T001.

Found at DOI: http://dx.doi.org/10.3996/012016-JFWM-002.S7 (1949 KB PDF).

Field technicians were Kathleen Carlyle and Steve Houdek. Pete Boma helped ready trailers and boats. Jason Rohweder and J.C. Nelson facilitated GIS manipulations used in selecting sample areas. Harry Roberts and his staff at Frontenac State Park, Minnesota, assisted with logistics. Volunteers Matt Groshek and Kevin Markwardt helped collect vegetation data. Kirk Lohman, Jack Waide, David Haukos, and Melinda Knutson provided valuable comments that greatly improved the manuscript.

Funding for this project was provided by the U. S. Fish and Wildlife Service State Wildlife Grant Program, the Minnesota Department of Natural Resources Non-game Wildlife Program, and the U.S. Geological Survey, Upper Midwest Environmental Sciences Center.

Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.