Northern Prairie Wildlife Research Center
We surveyed birds on 18-m fixed-radius plots in eight vegetation types based on hydrology and plant form and structure: scrub-shrub, wet meadow/scrub-shrub, wet meadow/scrub-shrub/loosestrife, wet meadow, wet meadow/loosestrife, inland cattail, coastal cattail, and coastal bulrush (Scirpus spp.). Our habitat classifications were based on Cowardin and coworkers (1979); dominant plants had greater than 30% cover (Fig. 1). We used a split class (e.g., broad-leafed deciduous scrub-shrub/persistent emergent; National Wetlands Inventory) to classify two vegetation types because scattered shrubs of at least 30% cover were present. We separated cattail sites into coastal and inland because hydrologies differed; coastal sites were intermittently exposed, whereas inland sites were semi-permanently flooded by groundwater and precipitation.
Sampling periods were divided into an early season during the second and third weeks of May, a mid-season during the first and second weeks of June, and a late season during the last week of June and first week of July. We conducted surveys between sunrise and 10:00 EST. Surveys were not conducted if sustained winds exceeded 24 km/h or during heavy rain.
We selected plots using the following protocol: first, an azimuth was determined that traversed the habitat. The center of the first plot was placed at least 18 m from the outer boundary of the vegetation on that azimuth. The center of the next plot was 70 m from the first plot on the same azimuth. This procedure was continued until observers surveyed three or more plots or reached a different vegetation type. If fewer than 3 plots were established on the first azimuth, we established a second azimuth, approximately perpendicular to the first azimuth, that traversed the vegetation type and permitted plot placement at least 70 m from other plots. Plots were set on this azimuth in the same manner as on the first azimuth. Plots were placed in different locations at the same site among time periods to avoid resampling the same plots and recounting the same nests. Coastal bulrush plots were not surveyed during the first periods of each year because they lacked structure; new vegetative growth was not yet established and the previous year's growth was eliminated by ice action. Neither did we survey three vegetation types (wet meadow/scrub-shrub/loosestrife, wet meadow, wet meadow/loosestrife) during the first period of 1994. We surveyed 258 plots in 8 wetland habitats.
Observers waited 5 min for normal bird activity to resume after arriving at a survey plot. We recorded all birds seen or heard on plots during a 7-min observation period. We recorded flying birds if their flight originated or terminated within the plot and we tallied individual birds only once. We played tape-recorded calls (Peterson 1990) of five secretive species [American Bittern (Botaurus lentiginosus), Least Bittern, King Rail (Rallus elegans), Virginia Rail (R. limicola), and Sora (Porzana carolina)] during the last 3 min using portable cassette recorders (Johnson et al. 1981, Marion et al. 1981, Johnson and Dinsmore 1986). We played calls for 25 - 30 sec followed by 10 sec of silence. We measured water depth and vertical cover 4 m from plot center at 0o, 120o, and 240o (Table 1). Observers measured vertical cover to the nearest 10 cm using a 2-m Robel pole placed at plot center and viewed while maintaining eye level 1 m above the water surface or ground level and looking back toward plot center (Higgins et al. 1994). Workers returned to plots later that day and searched the innermost 13-m radius (0.05 ha) portion for nests. A bird species was designated as breeding when nests or flightless young were observed in one or more periods or when adults were observed in two of three periods (Brown and Dinsmore 1986). A nest verified breeding status when eggs, young, or strong evidence of use such as egg shell fragments, down, or fecal sacs were present. We considered predated nests as breeding evidence when prey species could be determined. We also tallied species as breeding if they were observed within the sampled vegetation type but outside of plot boundaries on two of three visits.
We tallied breeding species richness (i.e., number of breeding species) for each vegetation type. We calculated avian diversities for each plot using the Shannon-Weiner diversity index. Density was the number of birds (both sexes) observed on a plot multiplied by 10 to obtain density per hectare.
We used ANOVA (PROC GLM; SAS 1990; SAS 6.12 for Windows) to assess fixed effects of vegetation, period, year, and their interactions on avian density and diversity. Residuals were normally distributed, but variances were not homogeneous because we never observed some species in one or more habitats (resulting in means and variances of zero). However, the overall F-statistic from ANOVA is robust to violations in assumptions of homogeneous variances (Sokal and Rohlf 1981). Early-period observations were eliminated from all analyses because of missing data. We considered plots as the experimental units because we decided a priori to restrict our inference to Saginaw Bay wetlands. We used α = 0.05 for all statistical comparisons. We initially analyzed fully specified models (all main effects and interactions included). We fitted each model using a backward, stepwise procedure by eliminating non-significant (P > 0.05) effects, beginning with highest-order interactions. Thus, our final models included only significant effects or interactions, and main effects or interactions contained in significant higher-order interactions. We used Fisher's protected least significant difference test to isolate differences among least-square means (LSMEANS, SAS 1990) for significant effects in the ANOVA (Milliken and Johnson 1984). We compared density and diversity of birds in loosestrife-dominated vegetation types (wet meadow/scrub-shrub/loosestrife and wet meadow/loosestrife) to those in other vegetation types using orthogonal contrasts (PROC GLM; SAS 1990), and estimated least-square means using estimate statements (PROC GLM; SAS 1990). We developed similar models for abundance of the six most commonly observed bird species: Sedge Wren (Cistothorus platensis), Marsh Wren, Yellow Warbler (Dendroica petechia), Common Yellowthroat (Geothlypis trichas), Swamp Sparrow (Melospiza georgiana), and Red-winged Blackbird.
Standard errors reported are for least-square means (SAS 1990). Because multiple comparison of means with heterogenous variances may be misleading (Sokal and Rohlf 1981), we further examined comparisons of non-zero means to means of zero using confidence intervals. For each mean of zero, we constructed a 90% upper confidence limit after assigning the highest standard deviation associated with any mean in the model. We then compared 90% lower confidence intervals for nonzero means to 90% upper confidence intervals for zero means; we considered failure of these intervals to overlap as statistically significant. Resulting confidence intervals for zero means are likely overestimated, yielding a conservative comparison. We note in tables instances where confidence interval comparisons did not corroborate multiple comparisons using Fisher's least significant difference test.