Northern Prairie Wildlife Research Center
We used a randomized block design to test for grazing and burning effects. Each treatment consisted of a multi-year regimen of burning, grazing and/or rest (Appendix 1). At the beginning of the experiment we assigned fields to 3 treatments and 1 control in each of 3 blocks. We chose treatments that were being used or that were under consideration for use by grassland managers who wanted to reduce brush and exotic grasses. Control fields were neither burned nor grazed. Our assignment of blocks was based on similarity of vegetation in 1980 and 1981, before treatments began. Assignment of fields within blocks to treatments was constrained by the ability to control a prescribed burn, and hence was not strictly random. This would only be a problem if our assignment led to biased measures of nest density or nest success of nesting waterfowl. We have no evidence that there was such a bias.
Three nest searches were conducted each year. We found nests by flushing hens with a 53-m cable-chain (Higgins et al. 1977) towed by vehicles. The stage of embryo development when the nest was found was ascertained by candling in the field (Weller 1956) and was used to estimate the nest initiation date. We checked nests every 3 weeks and after estimated hatching dates to determine fates. We estimated the hatching date by summing the clutch size (assuming 1 egg was laid per day) and the incubation period (Klett et al. 1986) and adding that number to the estimated nest initiation date. A nest was considered successful if at least 1 egg hatched. A nest was considered destroyed if there was no evidence of hatching and the nest bowl was burned, empty, or broken, or if partially consumed eggs were in the nest, or if the dead hen was found at the nest. A nest was considered abandoned because of human disturbance if, on the second visit, the hen was absent and there was no evidence the hen had returned to the nest (e.g., no increase in clutch size since the first visit). In all other cases of absence of the hen and her apparent abandonment of the nest, we considered the nest failed for unknown reasons. We calculated nest success with the modified Mayfield method (Mayfield 1961, Johnson 1979).
Because of changes in water levels at Lostwood, more dry land, suitable for nesting by ducks, was available in some years than in others. We standardized the area from which data on nests were used in the analyses with maps of water levels and nest locations made in the field each year. We determined the year in which water covered the most area (1984), then omitted from our analyses any nests found in other years that were within the high-water boundaries.
We used 2 measures of nest density. Annual nest density was the number of nests per field each year divided by the upland area of the field. Total nest density was the number of nests per field summed over the years 1982-88 and divided by the upland area of the field. Nest data from 1980 and 1981 (pre-treatment yrs) were not included in the analysis of total nest density.
We analyzed data separately for each species of waterfowl. We analyzed total nest density and annual nest density of all species except 2 uncommon species, green-winged teal (Anas crecca, 7 nests) and canvasback (Aythya valisineria, 1 nest). Statistical analysis of nest success was limited to the 3 most common species. For our analyses of total nest density, we used a completely randomized block design to test for treatment effects.
We analyzed nest success with an angular (inverse sine) transformation (Steel and Torrie 1980) of daily nest survival rate (Mayfield 1961), weighted by exposure days. Nests that were abandoned because of human disturbance were omitted from the success analyses. Because of lack of nests of some species in some fields, estimates of daily nest survival rates were not available from all fields in all years. To analyze success we used the procedures described by Milliken and Johnson (1984:378) for unbalanced data. Least-squares means of transformed daily nest survival rates were calculated (SAS Inst. Inc. 1987). Daily nest survival rates reported in this paper were obtained by reversing the transformation calculations with the least-squares means calculated by SAS.
We used repeated-measures analysis of variance (ANOVA) (Milliken and Johnson 1984) with a randomized block design for the analyses of annual nest density and nest success. We tested for the main effects of treatment and year, as well as treatment-year interaction. When the effect of year was significant, we used Fisher's Least Significant Difference (LSD) procedure to test for differences among years. When a treatment-year interaction was significant, we tested for differences among treatments within years and among years within treatments. In such tests for annual nest density, we controlled for the effect of nest density changes on the control fields by analyzing the difference in density between control fields and treatment fields within each block.
In our examination of differences in annual nest density among years within treatments, we hypothesized which effects were probable based on our knowledge of differences in treatment applications among years. We then constructed appropriate statistical contrasts using the CONTRAST statement in the GLM procedure of SAS (SAS Inst. Inc. 1987). For all treatments, we refer to 1980-81 as pre-treatment years and 1987-88 as post-treatment years (after all grazing and burning had ended). On spring graze fields we refer to 1982-84 as grazing years and 1985-88 as post-grazing years. We contrasted pre-treatment (1980- 81) versus post-treatment (1987-88) years for control, spring burn, spring graze, and summer burn/spring graze treatments; pre-treatment versus grazing years for spring graze treatment; grazing versus postgrazing years for spring graze treatment; linear trend throughout the study (1980-88) for control fields; linear trend for 3 years and 5 years after last grazed for spring graze treatment.
In our examination of differences in annual nest density among treatments within years, we performed the following contrasts: control versus all treatments in 1980,1981, and 1988 when no treatments were applied; control versus spring graze in 1982, 1983, and 1984 (grazing yrs); control versus spring burn in 1983, 1985, and 1987 (burn yrs); control versus summer burn/ spring graze in 1983 and 1986 (spring grazing yrs).
Statistical analyses were performed with SAS programs for microcomputers (SAS Inst. Inc. 1987). A probability level < or = to 0.05 indicated statistical significance. We evaluated statistical power in cases where no treatment effect was detected. We computed confidence intervals for the difference between the highest and lowest parameter (nest density or nest success) values and evaluated the width of the confidence interval (Goodman and Berlin 1994, The Journal of Wildlife Management 1995). For example, for mallard total nest density, we computed the difference and standard error (SE) of the difference for the control (highest density) and summer burn/spring graze (lowest density) treatments using the ESTIMATE statement in PROC GLM of SAS (SAS Inst. Inc. 1987). We then computed the 95% confidence interval as ± 1.96 X SE. We reported the estimated differences and 95% confidence intervals for cases in which no treatment effect was detected (Appendix 2).
We indexed vegetation structure with the 100% visual obstruction method described by Robel et al. (1970) and Kirsch et al. (1978). We rounded visual obstruction readings to the nearest 0.5 dm and grouped them into 6 classes: 0-0.49 dm, 0.50-0.99 dm, 1.00-1.49 dm, 1.50-1.99 dm, 2.00-2.49 dm and >/=2.50 dm. All 6 classes were used for comparisons of use and availability. To simplify visual presentation of treatment effects on visual obstruction readings, we used 3 classes: 0-0.49 dm, 0.50-2.49 dm, and >/=2.50 dm.
To assess availability of visual obstruction reading classes, we systematically sampled upland vegetation from transects that bisected each study field. We avoided wetlands and previously farmed areas. On each transect were 25 stations, spaced 25 paces apart. We sampled twice each year, once in late April (1980-88) and again in early June (1982-88). A visual obstruction reading also was taken at each nest site when the nest was found.
We compared use and availability of visual obstruction reading classes in each treatment with a Chi-square contingency test and Bonferroni simultaneous confidence intervals (Neu et al. 1974, Byers et 84, Thomas and Taylor 1990). For mallard and northern pintail, which nest early in the breeding season (Table 1.), we used the first readings, obtained between 15 and 29 April, as our index of available vegetation. To ensure that transect data reflect availability of vegetation searched by mallard and pintail for nesting, we restricted the comparisons between availability and use to nests found before 25 May. For the remaining species, we used the second readings, obtained between 31 May and 13 June, as our index of available vegetation. We restricted analyses of data of these species to nests found between 18 May and 24 June (lesser scaup [Aythya affinis]), 20 May and 24 June (blue-winged teal, American wigeon, and northern shoveler [Anas clypeata]), or 20 May and 29 June (gadwall).
Vegetation types were mapped on aerial photographs of each field during ground surveys in 1981 and 1987. We selected these years as representative of pre- and post-treatment years. We selected 6 vegetation types for analysis: brush, western snowberry without grass understory; brush/grass, western snowberry with >50% canopy cover and an understory of grass; grass/ brush, western snowberry with <50% canopy cover and an understory of grass; grass/forb, grass with >50% canopy cover and forbs with <50% canopy cover; forbs, 100% canopy cover of forbs; and shallow marsh emergent, vegetation typical of seasonal wetlands.
We ascertained availability of vegetation types by measuring areas of each type from the maps made during ground surveys. We tested for changes between 1981 and 1987 in proportions of available vegetation types with a multivariate analysis of variance (MANOVA; SAS Inst. Inc. 1987). We used univariate ANOVA to determine which vegetation types changed. We determined use of vegetation types by classifying the vegetation at each nest site. We compared use and availability of vegetation types with a Chi-square goodness-of-fit test and Bonferroni simultaneous confidence intervals (Neu et al. 1974, Byers et al. 1984, Thomas and Taylor 1990).