USGS - science for a changing world

Northern Prairie Wildlife Research Center

  Home About NPWRC Our Science Staff Employment Contacts Common Questions About the Site

Assessment of a Mallard Model in
Minnesota's Prairie Coteau

Methods


We established circular 51.8 km2 study sites representative of landscapes managed for waterfowl in the Prairie Coteau. Wildlife managers were provided the project description and asked to identify candidate study sites. Extremely dry wetland conditions were expected during the first year of study. To increase our chance of capturing a sample of hen mallards, we also asked managers to identify sites likely to retain water. Candidate sites were inspected in early April 1990 to select three for field work. We made no effort to pick sites randomly. Instead, we chose sites having as wide of a range in land use as possible while providing enough wetland area to support the breeding pairs needed in the study.

Field Work

A 2-person crew was assigned to each study site in 1990 through 1992. Crews were instructed on how to capture and radiomark hen mallards and in telemetry field techniques. Crews practiced finding transmitters in the field and handling and radiomarking captive hens before work began. Vehicle-mounted null-peak yagi-antenna systems (see Cochran 1980) were used to monitor radiomarked hens and hand-held yagis were used to find nests. We assigned crew members having the most experience with radiotelemetry to study sites having landscapes where we believed monitoring hens and detecting nests would be most difficult.

Hen mallards were decoy-trapped (Sharp and Lokemoen 1987) at each study site in a central 23.3 km2 core area (4.8 km x 4.8 km). We attempted to mark 25 hens at each site each year with radiotransmitters weighing 26-28 g (Dwyer 1972). Each hen was weighed (nearest 10 g) and the second greater secondary covert (GSC2) was pulled from one wing. Dry mass in grams (W) and black/white area in square millimeters (BW) of the GSC2 were used in a discriminant function (DF) to age hens as yearlings or adults (Krapu et al. 1979a). A corrected DF (DF = 23.46 - 0.48W - 0.135BW) accounted for a scaling error in the original publication (D. Johnson, Natl. Biol. Serv., pers. commun. in Reynolds et al. 1995). A DF score ≤-1.0 indicated the hen was an adult whereas hens with scores ≥1.0 were considered yearlings. Age could not be determined for intermediate DF scores.

Attempts were made to obtain 2 radiorelocations per day for each bird. Transmitters were equipped with a mortality switch, and we tried to determine the fate of hens when the switch was activated.

Mallard nest mortality in the prairies is known to be high (e.g., Higgins 1977, Cowardin et al. 1985, Klett et al. 1988); thus, crews were instructed to confirm the existence of a nest as soon as one was suspected. We looked for a nest whenever radiorelocations placed a hen at the same spot in potential nesting habitat for 2 consecutive days or when a nest was suspected for other reasons. Most nests were found with an incomplete clutch, and nest age in days was considered equal to the number of eggs present. Eggs in nests being incubated when found were candled (Weller 1956), and nest age was determined by adding the number of days the nest had been incubated to the number of days required to lay the clutch. Nests were visited again when the clutch was believed to be complete and hens were away. The last check was made as soon as telemetry relocations indicated the hen was no longer attending the nest. Nest locations were categorized into one of 14 habitat types (Table 1). We use the terms habitat and habitat type to be consistent with Mack (1991) although these categories might be more appropriately referred to as vegetation types (see Hall et al. 1997). Data were also recorded for mallard nests of nonradioed hens found incidentally.

Table 1. Habitat types in modeling mallard productivity in southwestern Minnesota, 1990-92.
Habitat type Description
Barren Barren areas including open water. Equivalent to mallard model habitat
TRUE.BARREN.
Burned Grass Grass cover burned since the previous summer. Modeled equivalent to mallard
model habitat CROPLANDFALLPLRO.
Conservation Reserve
Program
Conservation Reserve Program land. Equivalent to mallard model habitat CRP.
Hayland Hayland mowed at least once annually. Equivalent to mallard model habitat
HAYLAND.
New cover Areas plowed and planted to grasses since the previous summer. Modeled
equivalent to mallard model habitat CROPLANDFALLPLRO.
Other Patches of habitat (<5 acres) described by other habitat types, and linear and point
features (e.g., rockpiles in cropland or strips of grass cover between plowed fields).
Equivalent to mallard model habitat OTHER.
Planted trees Herbicided areas planted to trees. Modeled equivalent to mallard model habitat
OTHER.
Right-of-way Roadside right-of-ways (i.e., roadside ditches). Equivalent to mallard model habitat
R OF WAY.
Row crop Agricultural land used for row crops. Equivalent to mallard model habitat
CROPLANDFALLPLRO.
Small grain Agricultural land used for small grains. Equivalent to mallard model habitat
CROPLANDFALLPLGR.
Unmanaged grass Wild grassland not managed for wildlife (i.e., pasture). Equivalent to mallard model
habitat GRASLAND.
Wetland vegetation Wetland vegetation that could be used for nesting. Includes the 4 mallard model
habitats PERM.WETLAND, SEMI.WETLAND, SEAS.WETLAND, and
TEMP.WETLAND.
Wildlife planted
cover
Wildlife Management Area (WPA) or Waterfowl Production Area (WMA)
grasslands. Equivalent to mallard model habitat PLNTCOVR.
Woodland Forrested areas ≥5 acres in size. Equivalent to mallard model habitat
WOODLAND.
ª Mallard model habitats are described in Mack (1991).

Habitat Mapping

Habitats on each site were delineated and quantified by the U.S. Fish and Wildlife Service National Wetlands Inventory in a manner similar to that described by Cowardin et al. (1995). Initial data were from high-altitude (1:63,360) color-infrared photographs taken during the late 1970s and early 1980s. We updated these by extensive ground truthing. Unbroken parcels of the same habitat were mapped as closed polygons; however, roads, streams, and some oddly-shaped habitats were mapped as lines or points because of their small scale. Area of line and point features was determined, while maintaining their true map locations, by buffering (Appendix A). Resulting maps were digitized and converted to Map Overlay and Statistical System files (Pywell and Niedzwiadek 1980).

Summary of Field Data

A hen was considered resident of a study site in a given year if it: 1) nested on the study site, 2) was relocated >3 days after capture and died on the study site, or 3) was relocated at least as often as the hen with the minimum number of relocations that nested on the site and ≥60% of these relocations occurred on the study site.

Daily hen survival and habitat-specific nest survival rates were estimated using the Mayfield method (Mayfield 1961, Klett et al. 1986). Hens that died within 3 days of instrumentation were excluded from survival analyses. Exposure-days for a hen were the number of days between radiomarking and the date she died or that we lost her radiosignal. We counted all days in this interval unless we were unable to monitor a female for >7 consecutive days. In this case, hen exposure-days were reduced by the number of consecutive days when no radiocontact was made. We also did not count exposure-days if we lost contact with a bird within 7 days of marking. Nests abandoned as a result of our influence were excluded from nest survival analysis. Exposure-days for nests found incidentally were determined using the modified Mayfield method (Johnson 1979).

We partitioned our nesting data 4 ways (i.e., on or off a study site and radiomarked or unmarked hen). We used data from all nests to estimate habitat-specific nest survival and clutch size to maximize sample sizes. Proportional habitat-specific nest occurrence was determined from all nests initiated on the study sites by resident hens. Nest attempts per hen and number of hatched nests were determined from the sample of nests established by resident hens whether or not all of the nests were on the study site.

Effects of year, study site, habitat type, and their interactions on the age at which mallard nests were confirmed were examined using a 2-way ANOVA (SAS PROC GLM) (SAS Institute 1991). We restricted the analysis to data from 1991 and 1992 due to sample size considerations. Because few nests were discovered in several habitat types, we also collapsed these into 3 broader categories that we believed presented similar levels of difficulty in detecting mallard nest starts: 1) grasslands or tracts of planted herbaceous cover, 2) roadside right-of-ways (ROW), and 3) cultivated fields, wetland vegetation, and odd areas. Ages were log transformed to stabilize variances. Main effects models were used when there were no significant interactions. Pairwise comparisons were made using Tukey's test maintaining a comparisonwise significance level of 0.05.

We needed an estimated population mean for each parameter of concern to compare our field estimates with the model predictions. We considered each study site-year combination (n = 9) as a stratum. We used standard stratified sampling estimators (Cochran 1977) to compute population means and their 90% confidence intervals from our field data. The stratum-specific sampling weights for summer hen mortality and habitat-specific nest survival were the respective number of exposure-days for the strata. Sampling weights for nests per hen, hatch rate, hen success, average clutch size, number of hatched nests, and proportional habitat-specific nest occurrence were based on the respective stratum-specific sample sizes. We present summer hen mortality for a 182-day season and nest survival for a 35-day nesting period. These are more meaningful to most biologists than are daily survival rates, and they correspond to predictions output by the model. We also estimated the conditional probability that a hen would be killed if her nest was destroyed. We pooled data from all depredated nests of radiomarked hens to estimate conditional hen mortality because relatively few hens were killed when their nests were destroyed. Because conditional hen survival could not be estimated using the Mayfield method, Kaplan-Meier estimates of the probability that a hen was killed were computed using SAS PROC LIFETEST (SAS Institute 1990).

Modeling

The mallard model requires the user to specify input values describing characteristics of the nesting habitat and wetland conditions in the landscape being modeled as well as inputs for certain parameters describing the mallard population of concern (Table 2). The model can be executed using average input values suggested for various parts of the Prairie Pothole region or with inputs that the investigator customizes to be more appropriate (Mack 1991). We executed 2 models. The first (hereafter the default model) used western Minnesota default inputs (Mack 1991; T. Shaffer, U.S. Geol. Survey, pers. commun.). The second (hereafter the customized model) used inputs we customized to more closely describe habitat conditions during our study. We used the default model because wildlife managers may not have the resources to gather detailed input data specific to their areas of interest. However, customized inputs are preferred because field observations should be most similar to model predictions when model inputs describe study conditions as closely as possible. We present results from both models because the comparisons are informative.

Table 2. Inputs required for mallard model execution
(Mack 1991).
Descriptor Inputs
For each habitat Measure of availability (%
of landscape or area)
For each habitat 5 date-specific Robel
readings (cm)
For each habitat 5 dates associated with
the Robel readings
For each habitat 6 date-specific daily nest
mortality rates
For each habitat 6 dates associated with
the daily nest mortality
rates
For
semipermanent
wetlands
Daily proportion of basins
containing water during
season (161 days)
For each wetland
type
Measure of % of total
wetland area in landscape
that is suitable for nesting
For breeding
mallards
Date first pairs arrive
For breeding
mallards
Date last pairs arrive
For breeding
mallards
Number of resident pairs
For breeding
mallards
Proportion of adult vs.
yearling hens in resident
population
For breeding
mallards
Daily hen mortality rate
(away from nest)
For breeding
mallards
Conditional probability that
hen is killed given nest is
destroyed
For mallard
broods
Seasonal brood survival
rate
For mallard
ducklings
Seasonal brood survival
rate

We reasoned that the weather prior to and during the study influenced vegetative growth and wetland conditions in the Prairie Coteau. We used unpublished data from wildlife managers in western Minnesota to customize Robel readings (Robel et al. 1970) input for Conservation Reserve Program (CRP), wildlife planted cover, hayland, unmanaged grass, and ROW habitat types during the study years (Appendix B). We also changed 6 Robel input dates, which reflected mowing of hayland and ROW habitats, to ones suggested by wildlife managers. Because we changed the dates that hayland and ROW habitats were mowed, we also changed corresponding input daily nest mortality dates to describe the changes in nest survival that occurred with mowing. In addition, we used cumulative precipitation totals from the previous 1 July each year to customize input daily measures of the proportion of semipermanent wetlands containing water (Appendix C).

Basin characteristics and the amount of emergent cover varied widely among the wetlands on our study sites. We used digitized National Wetland Inventory data to customize input measures for the percent of the total area of each wetland type that was suitable for mallard nesting (Appendix D).

For both the default and customized models, we modeled productivity on each study site each year for the number of resident hens in our sample. Age ratios determined from the sample of resident hens also were used as inputs in both models. For the remaining inputs (Table 2), Western Minnesota default values (Mack 1991; T. Shaffer, U.S. Geol. Survey, pers. commun.) were used in both models.

The mallard model is executable in either a homing or nonhoming mode. In both, nests are simulated in habitats based on the habitat's attractiveness, which is a function of the Robel reading and areal coverage of each habitat in the modeled landscape. However, in the homing mode the placement of nests is also weighted in favor of habitats with greater nest survival, presumably because hens select habitats for nesting in which they have had previous success. We modeled mallard productivity for our study area using the nonhoming option because of the relative newness of CRP habitat (L. Cowardin, U.S. Geol. Survey, pers. commun.) and because homing is not well understood and thus cannot be realistically modeled (Cowardin et al. 1988:26).

Analysis

A model run represents a hypothetical population of 1,000 hens with the resulting output scaled to the number of resident hens specified as an input. However, the mallard model is stochastic. Thus in our analysis, each model was run 8 times for all 9 study site-year strata (72 runs for each model). We averaged model predictions for all 72 runs to account for stochastic variability and to compute a single set of predictions for the Prairie Coteau study sites. We compared these predictions with the corresponding stratified means and 90% confidence intervals estimated from our field data. The range of confidence limits of the stratified mean for each parameter estimate served as the "known" value against which we compared the average model prediction. If the average model prediction fell within the confidence limits of the estimated stratified mean, one might consider the model to be performing well with respect to that parameter. We assessed the departure of the average model prediction from the "known" value by expressing the deviation of the prediction from the upper or lower confidence limit as a percent of the estimated stratified mean. We also ranked habitat-specific predictions of proportional nest occurrence and nest success and compared model-based ranks with rankings from our field estimates.


Previous Section -- Study Area
Return to Contents
Next Section -- Results

Accessibility FOIA Privacy Policies and Notices

Take Pride in America logo USA.gov logo U.S. Department of the Interior | U.S. Geological Survey
URL: http://www.npwrc.usgs.gov/resource/birds/malmodel/method.htm
Page Contact Information: Webmaster
Page Last Modified: Friday, 01-Feb-2013 19:34:49 EST
Menlo Park, CA [caww54]