Northern Prairie Wildlife Research Center
Landscape-level indicators of condition used in the analysis followed those recommended by Cowardin and Sklebar (1997) and included: (1) cropland as a percentage of total upland habitat (PERCROP); (2) grassland as a percentage of total upland habitat (PERGRASS); (3) proportion of basins modified by drainage (PCTNMOD); (4) total linear drainage length (m), an indicator of drainage intensity (TOTDRAIN); (5) percent of cropland adjacent to all temporary, seasonal, and semipermanent basins (ALLADJC); and (6) percent of grassland adjacent to all temporary, seasonal, and semipermanent basins (ALLADJG). TOTDRAIN and PCTNMOD were derived from original NWI data and photointerpretation of 1996 aerial photography.
Other landscape variables used included total basin area (BASAREA (ha), as determined from GIS coverage), total number of basins (TOTBASIN), and indices in the change in wet area, indicating seasonal water loss for semipermanent basins (SEMICID), seasonal basins (SEASCID), and temporary basins (TEMPCID) between May and July. These indices were included because Cowardin and Sklebar (1997) found the indices were able to discriminate between extremes in landscape condition and were more sensitive than the raw data. We calculated the change index (CI) for area of water in each study area between May and July 1996 using the following formula (Cowardin and Sklebar, 1997):
- CI = | (May - July) / NWI |
- May = wet area of sampled basins in May;
- July = wet area of sampled basins in July;
- NWI = area of sampled basins from GIS coverage.
- CI was calculated for each basin class across each study area.
Roadside transects were located along all roads drivable in May in each study area. Wetland basins observable from the road right-of-way constituted the sample of wetland basins in each study area for pair counts. In 1995, the roadside transect and associated wetland basins available for sampling included everything 402 m (0.25 mi) on either side of the center of the road; in 1996, this width was reduced to 201 m (0.125 mi) to improve observability. A random subsample of 100 basins from the transect area was selected proportional to the area and frequency of each basin class. If there were less than 100 basins in the transect, all basins were sampled. All permanent basins within the transect were sampled. When we were unable to view a basin, we replaced it with a basin of the same wetland regime which was located within the transect area.
Three technicians conducted duck counts from vehicles along the right-of-way. For each basin, we recorded number of ducks by social groups (Cowardin et al., 1995), the proportion of the basin that could be observed, the proportion of the basin that was not obstructed by emergent vegetation, basin cover type (relative distribution of open water and emergent vegetation in the basin; Stewart and Kantrud, 1971), and areal percent of basin holding water. Ducks counted were later adjusted for visibility using the first 2 wetland measures and followed the assumption that ducks were dispersed equally across the basin.
Two counts were conducted each year: early counts were conducted during 1-15 May and late counts were conducted during 20 May-5 June. Data from early counts were used to estimate breeding pairs of mallards and pintails whereas data from late counts were used to estimate breeding pairs of blue-winged teal and gadwalls (Cowardin et al., 1988b). Breeding pairs of northern shovelers were estimated from the count occurring nearest 15 May. Numbers of breeding pairs were determined from social groups as described by Cowardin et al. (1995).
Numbers of breeding pairs were extrapolated from the adjusted duck counts to the entire study area using previously developed regression models that related estimated breeding pairs to pond area (area of the basin containing water; Cowardin et al., 1995). A correction, γ, for temporal and spatial variation was calculated as the ratio of observed to predicted pairs for the sampled basins. The estimate from the regression model for all basins on a study area was corrected by multiplying the regression estimate by γ. This method does not require the assumption that duck density on the sampled basins represents density on the entire study area. We assumed that the difference between observed pairs and pairs predicted by the regression for the sampled basins was representative of the unsampled basins within the study area.
Data Analyses. We used analysis of variance (ANOVA) techniques to assess differences in estimated numbers of breeding pairs, γ's, and landscape variables among three regions. Landscape variables changed very little between years, so only one value was used for both years. To achieve normality and constant variance, TOTDRAIN was transformed using a square root (Y + 0.5) transformation, and TEMPCID, SEASCID, and SEMICID were transformed using an inverse transformation of (Y + 1). The least squares means procedure (SAS Institute, Inc., 1997) was used to estimate parameters; LSMEANS reported here are back-transformed values.
We used ANOVAs to assess the potential influence or relationship of the explanatory variables (landscape variables) to the indicated pair counts and estimated γ. We dichotomized each explanatory variable into low and high categories using their respective medians as the dividing point. Following methods described in Milliken and Johnson (1992), we conducted a factorial ANOVA with main effects and two-way interactions to determine potential factors to use in a multiple regression analysis. We used 0.2 as a significance level for the screening process, but used 0.05 as a significance level for reporting mean comparisons. We then followed model selection methods described in Myers (1990) and Neter et al. (1990) to develop regression models and their explanatory capabilities towards the response variables. ANOVAs were conducted using the mixed model procedure (PROC MIXED) of SAS (SAS Institute, Inc., 1997), with mean separations for significant effects in the ANOVAs being done with Fisher's protected LSD criteria (Milliken and Johnson, 1992). Regression analyses were conducted using the regression procedure (PROC REG) of SAS (SAS Institute, Inc., 1997). ANOVAs were done separately for each duck species and their respective γ, and for each year because of differences in sampling design between years. Landscape parameters did not differ between years, but four study areas were censored from 1995 analyses because no duck counts were conducted.
We included γ as a response variable to assess the effect of landscape variables because γ adjusts for a variety of effects such as geographic within-year and/or year effects. In the calculation of γ, predicted numbers (the denominator) is based on water available in each basin. We examined various regression models using PROC GLM (SAS Institute, Inc., 1997), incorporating PERCROP, PERGRASS, TOTDRAIN, and PCTNMOD.