Habitat Selection by Mallard Broods on Navigation Pool 7 of the Upper MississippiRiver

Lynne T. DeHaan 1,2

1 Saint Mary's University, 700 Terrace Heights, Winona, Minnesota 559872 U.S. Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse,Wisconsin 54603

Keywords: mallard, habitat use, habitat selection, availability, radio-telemetry, homerange, compositional analysis, geo-spatial modeling

Abstract

Habitat use and selection was determined for radio-marked mallard broods on Pool 7 ofthe Upper Mississippi River for 1993 and 1994. Data were collected on a daily basisusing standard telemetry techniques. Habitat use was determined using methods thatconsider telemetry error in estimating brood locations. Compositional analysis was usedto determine habitat selection at two levels. The first level of analysis (second-orderselection) compared the area composition of habitat used in the home range to the area ofhabitats available throughout the study area. The second level of analysis (third-orderselection) evaluated the telemetry locations as habitat used and compared them to thearea of habitats available in the home range. At both levels of analysis for 1993 and1994, it was determined that use among habitats differed than what would be expected ifuse occurred at random. Emergent and rooted floating aquatic vegetation ranked high inthe third-order selection analysis among available habitat types for 1993 and 1994.Submersed aquatic vegetation and open water ranked high among available habitat typesin the second-order selection analysis for 1993 and 1994. There was a small detectabledifference in selection between 1993 and 1994 in the third-order selection analysis.

In addition to analyzing habitat selection by mallard broods, several importantissues concerning habitat use and selection studies are addressed. Included in this paperare discussions on the effects of triangulation error, misclassification error, definition ofavailability, home range estimators, and habitat use analysis methods. Results areaffected by each of these issues, so each needs to be considered when planning andconducting any habitat use and selection study.

Introduction

Many ecological and managementquestions regarding movement,behavior, habitat use, survival, and otherinterests of mobile species have existedfor decades. Through the developmentof radio-telemetry techniques,researchers have been able to gather andanalyze information on mobile species(Samuel and Fuller 1994). Use of radio-

telemetry data to evaluate issues such ashabitat use and selection is now commonpractice in wildlife research andmanagement (Samuel and Kenow 1992).Standard telemetry techniques can beused to locate an individual remotelywhile minimizing disturbances that canaffect daily activities of that individual(Samuel and Fuller 1994). Identifyinghabitats selected by mallard (Anasplatyrhynchos) broods can provide

2

managers with information about thehabitat requirements of mallard broods.This information allows researchers andmanagers to monitor which habitats arebeing used compared to habitats that areavailable to the animal.

Several habitat use and selectionstudies have been conducted using thehierarchy of selection proposed byJohnson (1980). The hierarchy containsfour orders of selection. First-order isthe selection of a physical orgeographical range by a species.Second-order determines the home rangeof an individual or group. Third-orderrefers to the usage of habitat componentswithin the home range. Fourth-orderidentifies specific resources (e.g. food)selected at a particular site. Availabilityof components differ between these fourselection orders. To evaluate the use andavailability of components on any orderof selection, a ranking system was alsodeveloped by Johnson (1980). Theranking system calculates the differencebetween the rank of usage and the rankof availability to measure preference.From this ranking system relativestatements can be made on usage andavailability of components. Using thisranking system allows the results to beless subjective and less sensitive toarbitrary definitions of availability andestimated use measurements (Johnson1980).

Many habitat use and selectionstudies have been conducted usingJohnson's (1980) methods. Mauser et al.(1994) found strong habitat selection bymallards occurring at the second-orderselection level, but not at the third-orderselection level in northeasternCalifornia. At the second-order ofselection, seasonal marshes with sometype of cover component were preferredcompared to open habitats or

permanently flooded marshes (Mauser etal. 1994). Gilmer et al. (1975)conducted a habitat-time interactionstudy in Minnesota and found thatmallards varied in their use of habitattypes, and habitat selection differedbetween day and night. Overall habitatpreference was for seasonal wetlands,sand bar-pond, and overhanging brushshorelines (Gilmer et al. 1975).Conversely, Rotella and Ratti (1992)concluded that habitat selection forcertain types of wetlands was notconsistent among mallard broods insouthwestern Manitoba. Broods that didshow selection for wetlands preferredseasonal, small semi-permanent, andlarge semi-permanent wetlands, whileother broods did not show any selection(Rotella and Ratti 1992). Since therewas a difference in selection or noselection, in that study, mallards may beconsidered as a generalist species.

The objective of this study was toanalyze habitat use and selection ofmallard broods on Pool 7 of the UpperMississippi River (UMR) using radio-telemetry data and land cover data.Habitat selection was determined at twolevels; within the home range of eachanimal and throughout the entire studyarea (Johnson 1980). Because thedefinition of what is available to ananimal is somewhat arbitrary, evaluatinghabitat use at two levels provides insightinto the hierarchical nature of selectionby mallard broods.

Study Area

The Upper Mississippi River (UMR) is avast system extending 1,300 miles fromLake Itasca in Minnesota to Cairo,Illinois. It supports commercial andrecreational traffic while providinghabitat to a diverse array of fish and

3

wildlife species. The UMR is controlledwith a lock and dam system to helpsupport commercial and recreationaltraffic. A result of the lock and damsystem has been the considerableamount of change in habitat compositionover several years.

An important annual activity thatoccurs on the UMR that also affectshabitat composition is flooding.Flooding can be important for manyspecies on the UMR, but can also bedetrimental to other species. In anycase, flooding affects vegetationdynamics and subsequently theavailability of habitats to individualspecies from year to year.

The Pool 7 study areaencompasses approximately 12,422hectares and extends from river mile 714near Trempealeau, Wisconsin to rivermile 697.5 near La Crosse, Wisconsin(Figure 1). The study area composition(Figure 2) mainly consisted of openwater (39%), woody terrestrial (19%),and agriculture-urban-other (19%; Table1). Much of the Pool 7 study area iscontained within the Upper MississippiRiver National Wildlife and FishRefuge, which is one of the five NationalWildlife Refuges that encompass300,000 acres of land and water alongthe river corridor.

Methods

Mallard brood location data (providedby K. P. Kenow, USGS Upper MidwestEnvironmental Sciences Center, LaCrosse, Wisconsin, unpublished data)were collected as part of a larger study todocument brood survival and movementon the UMR. The mallard nests werefound by searching islands on Pool 7.Eggs were collected from the nests andplaced in incubators until hatching.When a clutch began to hatch,individuals from the brood wererandomly selected for subcutaneousimplant of radio transmitters (Korschgenet al. 1996). The ducklings were thendeployed as broods back into existing

#

Minnesota

Wisconsin

Figure 1. Pool 7 study area.

Table 1. 1994 study area land cover classes used in the analysis.

ClassAbbreviation

Full Class Name Area(Hectares)

Area(Proportion)

OW Open Water 4874.72 0.39S Submersed Aquatic Vegetation 561.42 0.05RF Rooted Floating Aquatic Vegetation 738.92 0.06E Emergent Vegetation 815.18 0.07EG Emergents-Grasses-Forbs 151.68 0.009GF Grasses-Forbs 524.65 0.04WT Woody Terrestrial 2408.29 0.19AUD Agriculture-Urban-Developed-Other 2334.56 0.19SM Sand-Mud 12.55 0.001Total Area 12421.97 1.00

4

N

2000 0 2000 4000

Meters

1994 Land CoverOpen WaterSubmersed Aquatic VegetationRooted Floating Aquatic VegetationEmergent VegetationEmergents/Grasses/ForbsGrasses/ForbsWoody TerrestrialAgriculture/Urban/OtherSand/Mud

4

Figure 2. Study area habitat composition.

5

nests on the islands of Arrowhead,Broken Gun, Cormorant, or McIlvane inLake Onalaska of Pool 7 (Figure 3).After deployment, radio-markedducklings were tracked from Junethrough August in 1993 and Maythrough August in 1994 using standardtelemetry techniques. Data wererecorded and processed using a modifiedversion of the XYLOG programdeveloped by Dodge and Steiner (1986).The location of an individual wasdetermined using triangulation, which isthe process of estimating the location ofa transmitter by using two or moredirectional bearings obtained fromknown, remote locations (White andGarrott 1990). The standard deviation ofdirectional bearings for this study wascalculated to be 6.93 degrees based onmethods proposed by Lee et al. (1985).

Once the location data werecompiled, they were edited andsummarized. A total of seventeenbroods were monitored in 1993 andtwenty-two broods were monitored in1994. From each of the broods anindividual duck was selected to representthe entire movement of the brood. For1993, twelve of the seventeen broodswere represented, and for 1994,seventeen of the twenty-two broods wererepresented. The data for the broods thatwere not represented did not containcomplete information or did not containat least two readings to generate alocation solution, so they were not usedin the analysis. The data recordsassociated with each duckling includedthe duckling identification, date, time,receiver location, and azimuth to theduckling. In addition, the data wereevaluated and records were eliminatedthat contained incomplete readings ornon-intercepting azimuths. Also, thesize of the error ellipse associated with

400 0 400 800 Meters

N

Lake Onalaska

Broken Gun

Cormorant

Arrowhead

McIlvane

1994 Land / Water

LandWater

Figure 3. Islands used by nesting mallards inLake Onalaska.

each location solution was evaluated.Larger error ellipse values may signify adecrease in the accuracy of the azimuthsused in generating a particular locationsolution. Based on the distribution ofthe error ellipse sizes, location solutionsets that generated a 95% error ellipsegreater than 26 hectares were alsoeliminated from the analysis.

Land cover data were generatedfrom 1:15,000-scale color infra-redaerial photographs, which were taken inAugust of 1994. From the photography,the signatures were interpreted usingstandard operating procedures (Owensand Hop 1995) with a minimummapping unit of one acre. The photointerpreted transparencies were thentransferred onto a 1:24,000-scale USGStopographic map (Owens and Robinson1996) and digitized into a geographicinformation system (GIS) using standardoperating procedures (Arndt and Olsen1995). Using USGS topographic mapsfor transfer medium resulted in theaccuracy of the land cover data to bewithin fifteen meters. Over 160 speciesand species combinations of aquatic andterrestrial vegetation have been

6

identified on the UMR (L. Robinson,USGS, pers. comm.). For GIS andmapping purposes thirteen generalvegetation classes were generated for the1994 land cover data. After examiningthe characteristics of each class andreviewing the habitats most likely to beused by mallard broods, I combinedsome classes and ended up with nineclasses for the analysis (Table 1).

Newly developed software (K. P.Kenow, R. B. Wright, M. D. Samuel,and P. Rasmussen, Integration of SASand GIS software to improve habitat useestimates from radio-telemetry data,unpublished software) was used toconduct the habitat use analysis of thebrood location data. The softwareconsists of two components. AStatistical Analysis Systems (SAS)executable program, SUBSAMPL,generated 100 random points from theerror distribution of each estimated ducklocation (Figure 4) using maximumlikelihood procedures (Lenth 1981). Theoutput from SUBSAMPL generated a

N

100 0 100 200

Meters

#S#S

#S#S #S#S#S #S#S #S#S#S #S#S #S#S #S #S #S#S#S#S #S#S#S#S #S#S#S#S#S#S#S#S#S#S#S #S#S#S #S#S #S#S#S#S#S#S#S#S#S #S#S#S#S#S #S#S#S#S #S #S#S#S#S#S#S #S#S #S#S#S #S#S#S #S#S#S#S #S#S#S#S#S#S#S#S#S#S #S #S #S#S#S#S#S#S #S#S #S#S#S #S#S #S#S #S#S#S#S #S

#S #S#S#S#S#S #S#S #S #S#S#S#S #S#S #S#S #S#S#S#S #S #S#S#S#S#S#S#S #S#S#S#S#S #S#S #S#S #S #S#S #S#S #S#S#S #S#S #S#S #S#S #S#S #S#S#S#S #S#S #S#S#S #S#S #S#S #S#S #S#S #S #S#S#S#S#S #S#S#S #S#S #S#S#S #S#S

#S

#S

&\

&\

1994 Land CoverOpen WaterSubmersed Aquatic VegetationRooted Floating Aquatic VegetationEmergent Vegeta tionEmergents/Grasses/ForbsWoody Ter restria l

#S Random Sample Points&\ Location Solution

Figure 4. Example of random sampling pointswithin the 95% error ellipse of actual locationsolution.

file consisting of coordinate and attributeinformation that can be processed inARC/INFO. The geographicinformation systems (GIS) programHABUSE created a digital data set fromthe files containing coordinate andattribute information, determined thehabitat type for each random point in theerror ellipse, and summarized the habitattypes for all of the points. Theprobability of use for each habitat typewas calculated from the summarizedvalues.

Based on the resource selectionmethods of Johnson (1980), second- andthird- order selection were estimatedfrom information on habitat use relativeto availability. As part of these selectionmodels, the home range was determinedin ArcView (Environmental SystemsResearch Institute, Inc. 1998) using theminimum convex polygon (MCP)method. The MCP home range wasgenerated by connecting each bird'souter-most point locations. There areother techniques to obtain an animal'shome range, but the MCP method is thesimplest and most commonly used inhabitat selection studies (Samuel andFuller 1994).

Compositional analysis(Aebischer et al. 1993) was used to testfor habitat selection. It is one of thenewer techniques used to analyze habitatselection and addresses shortcomingsthat might exist in other habitat selectionanalysis methods. To deal with theseshortcomings, compositional analysisuses the animal rather than the radiolocation as a sample unit, performs alog-ratio transformation of the data toovercome non-independence ofproportional values, tests between-groupdifferences by referencing within-groupbetween-animal variation, and can

7

analyze data at different levels ofavailability (Aebischer et al. 1993).

The compositional analysisprogram (G. W. Pendleton, AlaskaDepartment of Fish and Game, Douglas,Alaska, unpublished software) tested thehypothesis that habitat use wasproportional to habitat availability. Thedifference between the log-ratiotransformation of habitat used and thelog-ratio transformation of habitatavailable was calculated for each habitattype indicating selection. Thesedifferences were then placed in a pair-wise comparison matrix to summarizethe selection analysis. Matrices of themeans and standard errors weregenerated, and the sign of each valuefrom the mean matrix was then extractedand placed in a simplified ranking matrix(see Results). The mean value matrixand the simplified ranking matrixprovided an indication of relativeselection among available habitat types.The significance of relative selectionwas related to how much the mean valuedeviated from zero, which also signifiedhow much use differed from randomuse. The rankings were determined fromthe number of positive signs generatedfor each habitat type compared to allother habitat types in the pair-wise

comparison matrix. For nine habitatclasses, the rankings ranged from 0 to 8with 0 being the least selected and 8being the most selected.

Results

Habitat use

Habitat use data were summarized toevaluate any use patterns that may haveoccurred by the mallard broods. Homerange estimates that were used in thehabitat use and selection analyses rangedfrom 2 hectares to 1,100 hectares for1993 and 55 hectares to 1,450 hectaresfor 1994. Location and home rangecomposition data for 1993 arerepresented in Tables 2a and 2b,respectively. Tables 3a and b representproportion of use for 1993. On average,habitats used consisted primarily of openwater (38%) and rooted floating aquaticvegetation (22%) based on the radiolocation data (Table 3a). Whenevaluating habitat composition of thehome range of the broods (Table 3b), thecomposition consisted primarily of openwater (61%) with each of the remainingvegetation classes representing less than10 percent of the average home rangecomposition.

Table 2a. Location data showing the proportion of habitat types used in 1993.

BroodNo.

No. ofLocations

OW S RF E EG GF WT AUD SM

4009 13 3.32 0.01 0.16 2.34 1.29 0.09 5.19 0.60 04025 23 2.92 0.25 17.53 0.94 0.14 0 1.22 0 04367 5 0.28 0 0 0.38 0 2.58 1.76 0 04422 24 7.25 0.55 6.52 4.35 1.16 0.48 3.69 0 04692 2 0.68 0.53 0.56 0.06 0.17 0 0 0 04782 9 1.08 1.22 1.25 0.51 1.08 0.82 2.26 0.78 04826 6 2.98 0.24 1.92 0.48 0.07 0 0.31 0 05759 4 2.57 0.05 1.36 0 0 0 0.02 0 05774 2 1.98 0.02 0 0 0 0 0 0 05855 12 5.37 0.68 4.21 0.89 0.46 0.04 0.35 0 05958 7 3.43 0 0.57 1.77 0.16 0.05 0.44 0.58 07243 12 4.13 1.07 0.81 0.88 0.47 2.51 2.13 0 0

8

Table 2b. Minimum convex polygon home range habitat composition in hectares for 1993.

Brood No. OW S RF E EG GF WT AUD SM4009 315.05 26.58 5.49 22.46 6.19 25.47 239.35 31.27 0.374025 208.80 24.70 44.59 6.08 1.49 0.08 12.19 0.14 04367 45.67 18.99 24.63 38.46 6.35 33.38 29.65 0.002 04422 108.02 25.07 28.39 19.90 4.31 1.87 16.80 0.02 04692 68.23 8.18 2.52 0.30 0.35 0.14 0 0.18 04782 512.77 166.57 124.11 107.93 38.80 10.11 80.46 62.05 04826 126.82 15.74 18.30 5.12 1.49 0.24 8.64 0.26 05759 31.22 2.13 0.99 0 0 0.48 0.34 0.19 05774 2.20 0.04 0 0 0 0.04 0 0.08 05855 218.59 30.05 63.12 16.22 3.95 2.22 11.47 0.19 05958 91.57 12.54 4.10 21.39 6.82 18.40 18.19 51.31 07243 457.74 123.53 93.07 155.49 33.15 48.72 83.42 0.18 0

Table 3a. Proportion of radio locations representing habitat use for 1993.

Brood No. OW S RF E EG GF WT AUD SM4009 0.255 0.001 0.012 0.180 0.099 0.007 0.399 0.046 0.0004025 0.127 0.011 0.762 0.041 0.006 0.000 0.053 0.000 0.0004367 0.056 0.000 0.000 0.076 0.000 0.516 0.352 0.000 0.0004422 0.302 0.023 0.272 0.181 0.048 0.020 0.154 0.000 0.0004692 0.340 0.265 0.280 0.030 0.085 0.000 0.000 0.000 0.0004782 0.120 0.136 0.139 0.057 0.120 0.091 0.251 0.087 0.0004826 0.497 0.040 0.320 0.080 0.012 0.000 0.052 0.000 0.0005759 0.643 0.013 0.340 0.000 0.000 0.000 0.005 0.000 0.0005774 0.990 0.010 0.000 0.000 0.000 0.000 0.000 0.000 0.0005855 0.448 0.057 0.351 0.074 0.038 0.003 0.029 0.000 0.0005958 0.490 0.000 0.081 0.253 0.023 0.007 0.063 0.083 0.0007243 0.344 0.089 0.068 0.073 0.039 0.209 0.178 0.000 0.000

Mean Valueof Usage 0.384 0.054 0.219 0.087 0.039 0.071 0.128 0.018 0.000

Table 3b. Proportion of habitat composition within the home range of each brood for 1993.

Brood No. OW S RF E EG GF WT AUD SM4009 0.469 0.040 0.008 0.033 0.009 0.038 0.356 0.047 0.0014025 0.701 0.083 0.150 0.020 0.005 0.000 0.041 0.000 0.0004367 0.232 0.096 0.125 0.195 0.032 0.169 0.150 0.000 0.0004422 0.529 0.123 0.139 0.097 0.021 0.009 0.082 0.000 0.0004692 0.854 0.102 0.032 0.004 0.004 0.002 0.000 0.002 0.0004782 0.465 0.151 0.113 0.098 0.035 0.009 0.073 0.056 0.0004826 0.718 0.089 0.104 0.029 0.008 0.001 0.049 0.001 0.0005759 0.883 0.060 0.028 0.000 0.000 0.014 0.010 0.005 0.0005774 0.932 0.017 0.000 0.000 0.000 0.017 0.000 0.034 0.0005855 0.632 0.087 0.183 0.047 0.011 0.006 0.033 0.001 0.0005958 0.408 0.056 0.018 0.095 0.030 0.082 0.081 0.229 0.0007243 0.460 0.124 0.094 0.156 0.033 0.049 0.084 0.000 0.000

Mean Valueof Usage 0.607 0.086 0.083 0.065 0.016 0.033 0.080 0.031 0.000

9

Location and home rangecomposition data for 1994 arerepresented in Tables 4a and 4b,respectively. Radio location data from1994 indicated that on average, habitatsused consisted of emergent vegetation(29%), open water (25%), and rootedfloating aquatic and submersed aquaticvegetation (16%; Table 5a). Habitatcomposition of the home ranges forbroods in 1994 indicated that open waterrepresented the largest portion of thehome range (64%; Table 5b).Submersed aquatic vegetationrepresented 14 percent of thecomposition with the remaining habitattypes representing less than 10 percentof the average home range composition.

Habitat selection

Habitat selection occurred if a habitatwas used more than what would beexpected if use among habitats occurredat random. While t-tests were used forpair-wise comparisons, MANOVA wasused to test for selection across all

habitats. Analysis indicated that second-order selection occurred in 1993 (p =0.0017). Selection was indicated by themean values in Table 6a and therankings in Table 6b. Comparing thesigns in the ranking matrix showed thatselection of submersed aquaticvegetation, which ranked highest amongall habitat types, differed from all otherhabitats except open water andemergents-grasses-forbs. Also, theselection for open water (rank 7) differedfrom that of grasses-forbs, sand-mud,woody terrestrial, and agriculture-urban-developed. Agriculture-urban-developed was selected significantly lessrelative to all other habitat types exceptemergent vegetation and woodyterrestrial. Other comparisons that wereindicative included selection of rootedfloating aquatic vegetation differingfrom submersed aquatic vegetation,woody terrestrial, and agriculture-urban-developed, and selection of emergentvegetation differing from submersedaquatic vegetation and emergents-grasses-forbs.

Table 4a. Location data showing the proportion of habitat types used in 1994.

BroodNo.

No. ofLocations

OW S RF E EG GF WT AUD SM

4208 14 6.54 0.12 3.02 1.46 0.24 0.03 2.58 0.01 04330 15 3.73 2.42 3.96 3.12 0.64 0.19 0.93 0.01 04537 6 1.05 1.16 1.20 1.81 0.71 0.02 0.05 0 04573 17 8.53 1.63 2.56 1.64 0.09 0.01 2.54 0 04689 2 1.14 0.11 0.56 0.10 0 0 0.09 0 04804 4 0.67 1.19 1.35 0.36 0 0 0.43 0 05074 24 0.06 2.57 1.33 18.51 1.53 0 0 0 05124 11 0.52 0.82 0.69 6.12 2.35 0 0.50 0 05235 7 0.50 1.12 1.33 2.95 0.44 0.09 0.57 0 05476 22 0.29 5.26 1.97 13.15 1.30 0.01 0.02 0 05494 31 9.20 2.11 7.59 9.34 1.08 0.13 1.55 0 05637 22 0.59 2.70 3.80 14.36 0.49 0 0.06 0 05816 3 1.51 1.36 0 0 0 0.13 0 0 05826 30 4.38 9.47 0.22 6.67 0.69 0.15 8.42 0 07030 10 3.86 1.44 0.78 0.46 0 0.03 3.43 0 07045 9 1.48 0.42 2.40 2.24 0.72 1.02 0.72 0 07211 10 4.28 0.99 1.79 1.86 0.15 0.03 0.90 0 0

10

Second-order selection alsooccurred in 1994 (p = 0.0001). Themean values (Table 7a) and the habitatselection rankings (Table 7b) indicated asignificant selection for submersedaquatic vegetation relative to all otherhabitat types. The selection of openwater (rank 7) differed significantly

from all habitat types except rootedfloating aquatic vegetation. Anotherobvious comparison was woodyterrestrial, grasses-forbs, and sand-mudbeing relatively equal in selection, butdiffering from all other habitats. Finally,selection of agriculture-urban-developed, which ranked lowest,

Table 4b. Minimum convex polygon home range habitat composition in hectares for 1994.

Brood No. OW S RF E EG GF WT AUD SM4208 515.27 52.63 14.74 15.76 10.31 3.71 78.71 0.79 04330 235.51 87.16 45.01 33.47 5.88 2.29 17.10 0.06 04537 546.56 130.55 59.45 42.85 8.16 3.27 14.68 0.84 04573 536.07 78.30 30.77 21.35 3.36 2.46 14.46 0.84 04689 154.13 23.85 15.63 1.80 0 0.05 2.45 0.06 04804 193.59 19.50 68.16 8.28 1.05 0.58 14.12 0.24 05074 94.72 23.19 7.52 24.30 3.55 0.002 0.23 0.005 05124 434.58 105.85 43.80 52.86 13.49 1.12 16.49 0.06 05235 422.93 125.56 121.56 140.38 28.26 12.23 58.67 0.11 05476 160.48 56.68 28.36 53.24 12.50 0.86 8.59 0.006 05494 180.12 47.33 31.86 20.45 4.06 3.11 14.86 0.57 05637 960.58 174.09 146.56 90.02 17.89 4.80 55.46 0.84 05816 35.67 16.29 1.41 0 0 1.08 0 0.54 05826 256.99 39.34 3.83 8.73 1.42 0.33 7.10 0.03 07030 280.69 26.97 7.26 6.94 7.2 3.69 54.17 0.68 07045 87.70 20.02 32.11 13.98 3.43 1.81 8.72 0.02 07211 452.08 99.33 109.67 47.69 10.45 4.09 50.31 0.54 0

Table 5a. Proportion of radio locations representing habitat use for 1994.

Brood No. OW S RF E EG GF WT AUD SM4208 0.467 0.009 0.216 0.104 0.017 0.002 0.184 0.001 0.0004330 0.249 0.161 0.264 0.208 0.043 0.013 0.062 0.001 0.0004537 0.175 0.193 0.200 0.302 0.118 0.003 0.008 0.000 0.0004573 0.502 0.096 0.151 0.096 0.005 0.001 0.149 0.000 0.0004689 0.570 0.055 0.280 0.050 0.000 0.000 0.045 0.000 0.0004804 0.168 0.298 0.338 0.090 0.000 0.000 0.108 0.000 0.0005074 0.003 0.107 0.055 0.771 0.064 0.000 0.000 0.000 0.0005124 0.047 0.075 0.063 0.556 0.214 0.000 0.045 0.000 0.0005235 0.071 0.160 0.190 0.421 0.063 0.013 0.081 0.000 0.0005476 0.013 0.239 0.090 0.598 0.059 0.000 0.001 0.000 0.0005494 0.297 0.068 0.245 0.301 0.035 0.004 0.050 0.000 0.0005637 0.027 0.123 0.173 0.653 0.022 0.000 0.003 0.000 0.0005816 0.503 0.453 0.000 0.000 0.000 0.043 0.000 0.000 0.0005826 0.146 0.316 0.007 0.222 0.023 0.005 0.281 0.000 0.0007030 0.386 0.144 0.078 0.046 0.000 0.003 0.343 0.000 0.0007045 0.164 0.047 0.267 0.249 0.080 0.113 0.080 0.000 0.0007211 0.428 0.099 0.179 0.186 0.015 0.003 0.090 0.000 0.000

Mean Valueof Usage 0.248 0.155 0.164 0.286 0.045 0.012 0.090 0.000 0.000

11

Table 5b. Proportion of habitat composition within the home range of each brood for 1994.

Brood No. OW S RF E EG GF WT AUD SM4208 0.745 0.076 0.021 0.023 0.015 0.005 0.114 0.001 0.0004330 0.552 0.204 0.106 0.078 0.014 0.005 0.040 0.000 0.0004537 0.678 0.162 0.074 0.053 0.010 0.004 0.018 0.001 0.0004573 0.780 0.114 0.045 0.031 0.005 0.004 0.021 0.001 0.0004689 0.779 0.120 0.079 0.009 0.000 0.000 0.012 0.000 0.0004804 0.634 0.064 0.223 0.027 0.003 0.002 0.046 0.001 0.0005074 0.617 0.151 0.049 0.158 0.023 0.000 0.001 0.000 0.0005124 0.650 0.158 0.066 0.079 0.020 0.002 0.025 0.000 0.0005235 0.465 0.138 0.134 0.154 0.031 0.013 0.064 0.000 0.0005476 0.500 0.177 0.088 0.166 0.039 0.003 0.027 0.000 0.0005494 0.596 0.157 0.105 0.068 0.013 0.010 0.049 0.002 0.0005637 0.662 0.120 0.101 0.062 0.012 0.003 0.038 0.001 0.0005816 0.649 0.296 0.026 0.000 0.000 0.020 0.000 0.010 0.0005826 0.809 0.124 0.012 0.027 0.004 0.001 0.022 0.000 0.0007030 0.724 0.070 0.019 0.018 0.019 0.010 0.140 0.002 0.0007045 0.523 0.119 0.191 0.083 0.020 0.011 0.052 0.000 0.0007211 0.584 0.128 0.142 0.062 0.013 0.005 0.065 0.001 0.000

Mean Valueof Usage 0.644 0.140 0.087 0.065 0.014 0.006 0.043 0.001 0.000

Table 6a. Habitat matrix of the means for 1993 by comparing proportions of habitats in thehome range with proportions of habitats available study area wide.

Habitat OW S RF E EG GF WT AUD SMOW 0.00 -0.15 0.95 1.67 0.97 1.40 2.58 4.63 2.51S 0.15 0.00 1.10 1.82 1.12 1.55 2.73 4.78 2.66RF -0.95 -1.10 0.00 0.72 0.03 0.45 1.63 3.68 1.56E -1.67 -1.82 -0.72 0.00 -0.70 -0.27 0.90 2.96 0.84EG -0.97 -1.12 -0.03 0.70 0.00 0.42 1.60 3.66 1.53GF -1.40 -1.55 -0.45 0.28 -0.42 0.00 1.18 3.24 1.11WT -2.58 -2.73 -1.63 -0.90 -1.60 -1.18 0.00 2.06 -0.07AUD -4.63 -4.78 -3.68 -2.96 -3.66 -3.24 -2.06 0.00 -2.13SM -2.51 -2.66 -1.56 -0.84 -1.53 -1.11 0.07 2.13 0.00

Table 6b. Ranking matrix for habitat selection in 1993 by comparing proportions of habitats in thehome range with the proportion of habitats available study area wide. A triple sign representssignificant deviation from random at P < 0.05.

Habitat OW S RF E EG GF WT AUD SM RankOW 0 - + + + +++ +++ +++ +++ 7S + 0 +++ +++ + +++ +++ +++ +++ 8RF - --- 0 + + + +++ +++ + 6E - --- - 0 --- - + + + 3EG - - - +++ 0 + +++ +++ +++ 5GF --- --- - + - 0 + +++ +++ 4WT --- --- --- - --- - 0 + - 1AUD --- --- --- - --- --- - 0 --- 0SM --- --- - - --- --- + +++ 0 2

12

differed significantly from all otherhabitats.

Third-order selection occurred in1993 (p = 0.01) with emergentvegetation ranking highest among allhabitat types. The mean values (Table8a) and the ranking matrix (Table 8b)indicated that selection for emergentvegetation differed significantly fromopen water, submersed aquaticvegetation, and agriculture-urban-developed. Rooted floating aquaticvegetation and emergents-grasses-forbswere equally selected relative to all otherhabitat types, and they both wereselected significantly more thansubmersed aquatic vegetation. Inaddition, submersed aquatic vegetation,the lowest ranked habitat, was selectedsignificantly less than rooted floatingaquatic vegetation, emergent vegetation,

emergents-grasses-forbs, woodyterrestrial, and sand-mud.

As in 1993, third-order selectionoccurred in 1994 (p = 0.0001). Themean value matrix (Table 9a) and thehabitat ranking matrix (Table 9b) bothindicated that emergent vegetation wasselected significantly more than all otherhabitat types. Rooted floating aquaticvegetation and emergents-grasses-forbswere equally selected relative to eachother, and they both were selectedsignificantly more than open water andagriculture-urban-developed habitats.Finally, the lowest ranked habitat, openwater, was selected significantly lessthan all other habitat types exceptgrasses-forbs and agriculture-urban-developed.

Analysis also indicated adifference in third-order habitat selection

Table 7a. Habitat matrix of the means for 1994 by comparing proportions of habitats in thehome range with proportions of habitats available study area wide.

Habitat OW S RF E EG GF WT AUD SMOW 0.00 -0.55 0.34 1.13 1.00 3.31 2.59 6.70 2.80S 0.55 0.00 0.89 1.68 1.55 3.86 3.14 7.25 3.35RF -0.34 -0.89 0.00 0.79 0.66 2.97 2.25 6.36 2.46E -1.13 -1.68 -0.79 0.00 -0.13 2.18 1.47 5.58 1.67EG -1.00 -1.55 -0.66 0.13 0.00 2.31 1.60 5.71 1.80GF -3.31 -3.86 -2.97 -2.18 -2.31 0.00 -0.72 3.39 -0.51WT -2.59 -3.14 -2.25 -1.47 -1.60 0.72 0.00 4.11 0.21AUD -6.70 -7.25 -6.36 -5.58 -5.71 -3.39 -4.11 0.00 -3.90SM -2.80 -3.35 -2.46 -1.67 -1.80 0.51 -0.21 3.90 0.00

Table 7b. Ranking matrix for habitat selection in 1994 by comparing proportions of habitats in thehome range with the proportion of habitats available study area wide. A triple sign representssignificant deviation from random at P < 0.05.

Habitat OW S RF E EG GF WT AUD SM RankOW 0 --- + +++ +++ +++ +++ +++ +++ 7S +++ 0 +++ +++ +++ +++ +++ +++ +++ 8RF - --- 0 +++ + +++ +++ +++ +++ 6E --- --- --- 0 - +++ +++ +++ +++ 4EG --- --- - + 0 +++ +++ +++ +++ 5GF --- --- --- --- --- 0 - +++ - 1WT --- --- --- --- --- + 0 +++ + 3AUD --- --- --- --- --- --- --- 0 --- 0SM --- --- --- --- --- + - +++ 0 2

13

(p = 0.0145) between 1993 and 1994.When the habitat ranking matrices for1993 (Table 8b) and 1994 (Table 9b)were compared, a few differencesbecame apparent. Selection foremergent vegetation was moresignificant in 1994 relative to all habitattypes. In 1993, emergent vegetation

significantly differed from only threehabitat types, open water, submersedaquatic vegetation, and agriculture-urban-developed. Also, in 1993, woodyterrestrial significantly differed fromopen water, submersed aquaticvegetation, grasses-forbs, andagriculture-urban-developed, but

Table 8a. Habitat matrix of the means for 1993 by comparing the proportions of radiolocations of each animal in each habitat with the proportion of each habitat within theanimal's home range.

Habitat OW S RF E EG GF WT AUD SMOW 0.00 1.33 -0.92 -1.08 -0.91 0.67 -0.88 0.84 -0.50S -1.33 0.00 -2.26 -2.41 -2.24 -0.67 -2.21 -0.49 -1.83RF 0.92 2.26 0.00 -0.16 0.02 1.59 0.05 1.76 0.43E 1.08 2.41 0.16 0.00 0.17 1.74 0.20 1.92 0.58EG 0.91 2.24 -0.02 -0.17 0.00 1.57 0.03 1.75 0.41GF -0.67 0.67 -1.59 -1.74 -1.57 0.00 -1.54 0.17 -1.16WT 0.88 2.21 -0.05 -0.20 -0.03 1.54 0.00 1.72 0.38AUD -0.84 0.49 -1.76 -1.92 -1.75 -0.17 -1.72 0.00 -1.34SM 0.50 1.83 -0.43 -0.58 -0.41 1.16 -0.38 1.34 0.00

Table 8b. Ranking matrix for habitat selection in 1993 by comparing the proportions of radiolocations of each animal in each habitat with the proportion of each habitat within the animal'shome range. A triple sign represents significant deviation from random at P < 0.05.

Habitat OW S RF E EG GF WT AUD SM RankOW 0 + - --- - + --- + --- 3S - 0 --- --- --- - --- - --- 0RF + +++ 0 - + + + + + 7E +++ +++ + 0 + + + +++ + 8EG + +++ - - 0 + + + + 6GF - + - - - 0 --- + - 2WT +++ +++ - - - +++ 0 +++ + 5AUD - + - --- - - --- 0 - 1SM +++ +++ - - - + - + 0 4

Table 9a. Habitat matrix of the means for 1994 by comparing the proportions of radiolocations of each animal in each habitat with the proportion of each habitat within theanimal's home range.

Habitat OW S RF E EG GF WT AUD SMOW 0.00 -1.42 -1.84 -2.91 -1.79 -0.96 -1.58 -0.46 -1.55S 1.42 0.00 -0.42 -1.49 -0.38 0.45 -0.16 0.95 -0.13RF 1.84 0.42 0.00 -1.08 0.04 0.87 0.26 1.37 0.29E 2.91 1.49 1.08 0.00 1.12 1.95 1.33 2.45 1.36EG 1.79 0.38 -0.04 -1.12 0.00 0.83 0.21 1.33 0.25GF 0.96 -0.45 -0.87 -1.95 -0.83 0.00 -0.62 0.50 -0.58WT 1.58 0.16 -0.26 -1.33 -0.21 0.62 0.00 1.12 0.03AUD 0.46 -0.95 -1.37 -2.45 -1.33 -0.50 -1.12 0.00 -1.08SM 1.55 0.13 -0.29 -1.36 -0.25 0.58 -0.03 1.08 0.00

14

significantly differed from only onehabitat type, open water, in 1994. Thestronger selection for woody terrestrialin 1993 compared to 1994 was expectedbecause important habitats for thebroods such as submersed aquatic,rooted floating aquatic, and emergentvegetation were either covered orwashed out by the high water levels in1993 (National Biological Service et al.1994, Spink and Rogers 1996). Oneother obvious difference between 1993and 1994 was the order of habitatrankings. Open water (rank 3 in 1993)and submersed aquatic vegetation (rank0 in 1993) switched rankings in 1994 torank 0 and rank 3, respectively. Second-order selection was not significantbetween 1993 and 1994 (p = 0.06).

Discussion

Several habitat selection studies havebeen conducted using radio-telemetrydata without accounting for the error thatcan occur with triangulation andmisclassification. More bias isintroduced into those studies that do notaccount for triangulation andmisclassification error (Samuel andKenow 1992). Many sources canintroduce error into triangulation

including unknown receiver locations,equipment malfunction, field conditions,and misuse of equipment (Samuel andFuller 1994). An error assessment wasdone for this study based on the methodsproposed by Lee et al. (1985), and thestandard deviation of bearings wascalculated to be 6.93 degrees. From thisstandard deviation an error polygon iscreated and used to minimize the effectsof misclassification errors. Ifmisclassification is not accounted for,two possible outcomes can occur. Therecan either be an underestimation of thetrue proportion of habitat used or anoverestimation of the true proportion ofhabitat used (Samuel and Kenow 1992).In this study, misclassification wasaddressed using methods proposed bySamuel and Kenow (1992). Taking asub-sample from the error distribution ofeach telemetry location reduced bias inthe habitat selection results, but thismethod only produces an estimate ofhabitat use and does not provide exactresults.

Data censoring is another way toincrease data quality from triangulationmethods. In habitat selection and usestudies, data collected by triangulationcan be reviewed to eliminate poor-quality locations. As suggested by

Table 9b. Ranking matrix for habitat selection in 1994 by comparing the proportions of radiolocations of each animal in each habitat with the proportion of each habitat within the animal'shome range. A triple sign represents significant deviation from random at P < 0.05.

Habitat OW S RF E EG GF WT AUD SM RankOW 0 --- --- --- --- - --- - --- 0S +++ 0 - --- - + - + - 3RF +++ + 0 --- + + + +++ + 7E +++ +++ +++ 0 +++ +++ +++ +++ +++ 8EG +++ + - --- 0 + + +++ + 6GF + - - --- - 0 - + - 2WT +++ + - --- - + 0 + + 5AUD + - --- --- --- - - 0 --- 1SM +++ + - --- - + - +++ 0 4

15

White and Garrott (1990), one possibleway to censor data is to incorporate aconfidence ellipse size less than somecutoff value. Data that generated a 95%error ellipse greater than 26 hectareswere eliminated in this study. Inaddition, all the data were censored toeliminate any erroneous or incompletereadings.

Although compositional analysisis accepted as a viable technique forhabitat selection analyses, one mainproblem exists related to missing habitatvalues. Two situations can occur.Either a habitat type is not available foruse or a habitat type is available but notutilized. Solutions to these situationsvary. When habitat availability is equalto zero habitat types could either becombined to reduce the number ofhabitats analyzed, or habitat types couldbe eliminated entirely from the analysis(Aebischer et al. 1993). In the situationwhere a habitat is available but not used,the zero implies that use was so small itwas not detected. A solution for a zerouse value is to replace the zero with asmall constant that is less than thesmallest recorded nonzero value(Aebischer et al. 1993). Zeros areinadmissible in log-ratio analysesbecause taking the log of zero ordividing by zero produce undefinedresults (Pendleton et al. 1998). For thisstudy, the compositional analysisprogram replaces all zeros with a smallconstant no matter what the zerorepresents. Other methods are availablethat are capable of performing habitatselection analysis. Alldredge and Ratti(1986) review four statistical methodsfor analysis of habitat selection based ontype I and type II error rates. Alldredgeand Ratti (1986) concluded from manysimulations, that not all analysis methodsproduce the same results. The error rates

and the overall results are affected by thenumber of habitats, the number ofanimals, the number of observations peranimal, and the magnitude of thedifferences to be detected (Alldredgeand Ratti 1986). It is important toreview many existing techniques and tochoose the one best suited for thedefined study.

The definition of availability isanother issue to be considered whenconducting habitat selection studies.Although we can define any arbitraryarea as available to an animal, the animalmay not perceive all of that area to beavailable. Availability boundaries canbe affected by territoriality of an animal,a disturbance located near an area orhabitat, or the presence of a competingor predator species (White and Garrott1990). The definition of availability is akey element in habitat selection studiesand can influence the results (McCleanet al. 1998). Therefore, how to defineavailability for a particular animal is anongoing problem with habitat use andselection analyses.

Land cover data for 1993 did notexist at the time of this study to use inconjunction with the 1993 radiotelemetry data. Instead, 1994 land coverdata were used with both 1993 and 1994telemetry data. This in itself introducesbias into the results for the 1993 data.To conduct a more accurate study of the1993 data, land cover data would need tobe generated for 1993. In addition, 1993was a flood year generating some of thehighest water levels recorded in the lastcentury. It has been noted thatsubmersed aquatic and emergentvegetation species decreased inabundance during the flood in July(National Biological Service et al.1994). Some species recovered veryquickly following the flood in August

16

and September and others did not(National Biological Service et al.1994). Also, the density and speciesrichness of invertebrate communitieswere influenced at different stages offlooding (National Biological Service etal. 1994). Talent et al. (1982) conducteda multiple year study with varying waterconditions in South Central NorthDakota and found different results forthe two years of their study. In 1976,more wetlands were available to themallards than in 1977. This resulted inhabitat selection occurring in 1976 withthe mallards preferring seasonal pondsdominated by whitetop rivergrass(Talent et al. 1982). Semi-permanentwetlands were used less than expected in1976, but were used exclusively in 1977when seasonal wetland basins were dry(Talent et al. 1982). In summary,mallard broods did not use all of thewetland sizes available to them in 1976,but in 1977, they did use all sizes ofsemi-permanent wetlands in proportionto their availability. According to Talentet al. (1982), water conditions have aprofound influence on wetland habitatuse by mallard broods. For this study onPool 7 of the UMR, the flood of 1993probably had an effect on habitatsselected by mallard broods due to theinfluence it had on aquatic vegetationand invertebrate communities, both ofwhich are imperative for adequatebrood-rearing habitat. Many studies(Spink and Rogers 1996, NationalBiological Service et al. 1994, andBhowmik et al. 1993) have examinedthe effects of the 1993 flood on variouscomponents in the UMR. Although it isimportant to consider the effects that theflood had on habitat in the UMR, it isbeyond the scope of this study. Oneassessment regarding the results fromthe 1993 analysis could be made. The

stronger selection for woody terrestrialhabitat in 1993 may reflect the brood'sadjustment to the high water levels,which damaged the aquatic habitatsnormally used by mallard broods. Also,the 1993 flood probably altered the stateof woody terrestrial habitat into anaquatic habitat state, which then becameadequate brood-rearing habitat.Unfortunately, an accurate assessment ofwhich habitats were used or selected in1993 can not be conducted until landcover data for 1993 is generated.

Another issue to consider whenperforming habitat selection analysis isthe method used to obtain an animal'shome range. In this study I used theminimum convex polygon (MCP)method due to its simplicity and ease ofcalculation. In actuality, home rangescan be difficult to estimate, and dependon the purpose of the study and whetherthe methods used to collect the datasatisfy this purpose (White and Garrott1990). The MCP method has itsdrawbacks, as do other existing homerange estimators. The size of thepolygon generated for an MCP homerange is a function of the number oflocations used to determine that polygon(White and Garrott 1990). Polygonsincrease in size as the number oflocations to create the polygon increase.This is attributed to the home range sizesranging from 2 hectares to 1,450hectares between 1993 and 1994.Another problem with the MCP methodis that some misrepresentation of classesused by an animal or available to ananimal may exist. Since the MCPmethod connects the outer-most points,it may include a class that is at the edgeof the home range. An example wouldbe including a small piece of agriculturalor urban land within an MCP homerange, when in reality, the animal does

17

not use that habitat or consider it to beavailable. This problem is apparentwhen reviewing the results from thesecond-order analysis. It wasdetermined that submersed aquaticvegetation and open water were selectedin the second-order analysis. The resultsof the second-order analysis may be anartifact of the MCP estimator. Whenestimating the home range using thenesting island and the telemetrylocations for each bird, the home rangeincluded large open water areas (Figure5) and took on a triangular shape. Thiswas due to the fact that the birds had totraverse the open water areas to reachbrood-rearing habitat. It is difficult toselect a home range estimator that isappropriate for habitat selection studies,especially since each estimator hasstrengths and weaknesses (White andGarrott 1990).

Based on the techniquespresented here, it was determined thatmallards did select certain habitats onthe UMR. Third-order selection for1993 and 1994 showed the top threeselected vegetation types were emergentvegetation, rooted floating aquaticvegetation, and emergents-grasses-forbs.The reason for hens selecting thesehabitats probably has to do with twoimportant aspects of brood-rearinghabitat; cover and food. The emergentvegetation contains species such asbulrush, cattail, and arrowhead, whichhelp protect the ducklings frompredators and provide a food resource.Water-lily, lotus, and duckweed arerooted floating aquatic species that alsoprovide food for the broods. Feedinghabits of young are different than thoseof adult birds. Diets of young aregenerally protein-rich (O'Connor 1984).It has been shown that a duckling'sgrowth rate is related to the amount of

# #################### ################ ###### ######## ## ### ##### ## ## ## # #### ###### ## # ##### ###### ### ## ### ## # ## ### ### ### # ####### ###### ##### ## # ## ### ## #### ## ###### # ### ## # ## ## ### ## ####### #### # ### # #### ## ##### ###### ########## ###### ## ##### ###### ## ##### ###### ## # ####### ##### ##### ####### ############# ################################### ########################################## #### ################### ####################################### ############## ########################### #################################

#

########### ## ####### ## ##### # ### ### #### ### # ##### ## #### ###### ## ### ######### ##### ### ## ## ## ## ## #### ### ### # ## ### #### #### ## ## # ## ### #### ### ### ## ## # ####### ## ## #### ##### ########## ####### # ########### ## #################################### ###### ####### #### ### ### ### ###### # ##### ## # ## ## ### #### # ##### #### #### #### ## ##### ###### ### ### ## #### #### ###### #### ## ############# #### # ####### #### ####################### #### ######

Figure 5. Example of a mallard brood homerange on Pool 7, calculated using the minimumconvex polygon method. Open water andsubmersed aquatic vegetation are shaded graywith 900 random point locations on top.

protein consumed and that invertebratesprovide the major source of proteinduring times of peak growth (O'Connor1984). In the emergent and rootedfloating aquatic vegetation, aquaticinvertebrates and insects are present forthe young to feed on (Voigts 1976).Voigts (1976) also found that theabundance of invertebrates increasedwhere submersed aquatic vegetationinterspersed with emergent vegetationstands. For third-order selection, theresults appear to be realistic based on thecover and food available in the selectedhabitat types.

As stated earlier, second-orderanalysis indicated selection ofsubmersed aquatic vegetation and openwater as primary habitats followed byrooted floating aquatic vegetation. Inorder to reach areas of brood-rearing

18

habitat, hens had to traverse large openwater and submersed aquatic vegetationareas. That is probably why these twoclasses appeared prominent in thesecond-order selection results.Submersed aquatic vegetation beds canprovide a vast amount of invertebratesand insects for feeding, but may notprovide enough protective cover for thebroods.

Information from this study canbe used by managers to assess habitatrequirements for mallard broods. Brood-rearing habitat is important for thesurvival of mallard populations. Ifbrood habitat is not adequately availablefor the ducklings, management planssuch as a water level reduction, can beimplemented to create or enhancerequired habitat. Once habitat isavailable, it needs to be protected fromwind and wave action. There are islandsthat exist along the main channel of theUMR called barrier islands. Theseislands help protect the brood-rearinghabitat from wind and wave action, sothey need to be created or maintained aswell. These habitat areas are alsoimportant resources for other waterfowlincluding migratory species. Theinformation from this study will assistmanagers in maintaining and protectinghabitat areas that are important to manyspecies on the UMR.

Several issues related to habitatselection studies were considered in thisstudy. It is imperative to devise aproject scope for any study addressingcertain issues mentioned here beforedata collection begins. Data availability,error and accuracy of data, and analysistechniques should be researched anddefined before beginning a project. Allof these issues will ultimately affect theresults obtained in a habitat use andselection study.

Acknowledgments

I would like to thank Kevin Kenow forall of his assistance on this project, SteveGutreuter for his statistical expertise, andGrey Pendleton for providing thecompositional analysis program. Iwould like to thank the members of mycommittee; Dr. David McConville, Dr.John Nosek, Kevin Kenow, and DeanMierau for reviewing the manuscript andproviding support as needed. I wouldalso like to mention the U. S. GeologicalSurvey, Upper Midwest EnvironmentalSciences Center for providing the data,programs, and other information used inthis study.

References

Aebischer, N. J., Robertson, P. A., andKenward, R. Compositional analysisof habitat use from animal radio-tracking data. Ecology 74:1313-1325.

Alldredge, J. R., and J. T. Ratti. 1986.Comparison of some statisticaltechniques for analysis of resourceselection. Journal of WildlifeManagement 50:157-165.

Arndt, L. and D. Olsen. 1995. LongTerm Resource Monitoring Programstandard operating procedures:Production ARCEDIT digitizing.National Biological Service,Environmental ManagementTechnical Center, Onalaska,Wisconsin, August 1995. LTRMP95-P008-3. 12 pp. + Appendixes A-K.

Bhowmik, N. G., et al. 1993. The 1993flood on the Mississippi River inIllinois. Illinois State Water Survey,Champaign, MiscellaneousPublication 151.

Dodge, W. E., and Steiner, A. J. 1986.

19

XYLOG: A computer program forfield processing locations of radio-tagged wildlife. U. S. Fish andWildlife Service, Fish and WildlifeTechnical Report 4. 22 pp.

Environmental Systems ResearchInstitute, Inc. 1998. ARC/INFOsoftware version 7.2 and ArcViewsoftware version 3.1. Redlands,California.

Gilmer, D. S., Ball, I. J., Cowardin, L.M., Riechmann, J. H., and Tester, J.R. 1975. Habitat use and home rangeof mallards breeding in Minnesota.Journal of Wildlife Management39:781-789.

Johnson, D. H. 1980. The comparisonof usage and availabilitymeasurements for evaluating resourcepreference. Ecology 61:65-71.

Korschgen, C. E., K. P. Kenow, W. L.Green, M. D. Samuel, L. Sileo. 1996.Technique for implanting radiotransmitters subcutaneously in day-old ducklings. Journal of FieldOrnithology 67:392-397.

Lee, J. E., White, G. C., Garrott, R. A.,Bartmann, R. M., and Alldredge, A.W. 1985. Accessing accuracy of aradiotelemetry system for estimatinganimal locations. Journal of WildlifeManagement 49:658-663.

Lenth, R. V. 1981. On finding thesource of a signal. Technometrics23:149-154.

Mauser, D. M., Jarvis, R. L., and Gilmer,D. S. 1994. Movements and habitatuse of mallard broods in northeasternCalifornia. Journal of WildlifeManagement 58:88-94.

McClean, S. A., Rumble, M. A., King,R. M., and Baker, W. L. 1998.Evaluation of resource selectionmethods with different definitions ofavailability. Journal of WildlifeManagement 62:793-801.

National Biological Service, IllinoisNatural History Survey, IowaDepartment of Natural Resources, andWisconsin Department of NaturalResources. 1994. Long TermResource Monitoring Program 1993Flood Observations. NationalBiological Service, EnvironmentalManagement Technical Center,Onalaska, Wisconsin, December1994. LTRMP 94-S011. 190 pp.(NTIS #PB95-181582)

O'Connor, Raymond J. 1984. TheGrowth and Development of Birds.John Wiley and Sons Ltd., New York.326 pp.

Owens, T. and K. D. Hop. 1995. LongTerm Resource Monitoring Programstandard operating procedures:Photointerpretation. NationalBiological Service, EnvironmentalManagement Technical Center,Onalaska, Wisconsin, July 1995.LTRMP 95-P008-1. 7 pp. +Appendixes A and B.

Owens, T. and L. Robinson. 1996.Long Term Resource MonitoringProgram standard operatingprocedures: Manual zoom transferscope. U.S. Geological Survey,Environmental ManagementTechnical Center, Onalaska,Wisconsin, December 1996. LTRMP95-P008-5. 3 pp. + Appendix.

Pendleton, G. W., K. Titus, E.DeGayner, C. J. Flatten, and R. E.Lowell. 1998. Compositionalanalysis and GIS for study of habitatselection by goshawks in southeastAlaska. Journal of Agricultural,Biological, and EnvironmentalStatistics 3:280-295.

Rotella, J. J., and Ratti, J. T. 1992.Mallard brood movements andwetland selection in southwestern

20

Manitoba. Journal of WildlifeManagement 56:508-515.

Samuel, M. D., and Fuller, M. R. 1994.Wildlife Radiotelemetry In Researchand management techniques forwildlife and habitats. Fifth ed. pp.370-418. Edited by T. A. Bookhout.The Wildlife Society, Bethesda,Maryland.

Samuel, M. D., and Kenow, K. P. 1992.Evaluating habitat selection withradio-telemetry triangulation error.Journal of Wildlife Management56:725-734.

SAS Institute, Inc. 1996. SAS softwareversion 6.12. Cary, North Carolina.

Spink, A., and S. Rogers. 1996. Theeffects of a record flood on theaquatic vegetation of the UpperMississippi River System: Some

preliminary findings in Hydrobiologia340:51-57. Reprinted by U.S.Geological Survey, EnvironmentalManagement Technical Center,Onalaska, Wisconsin, May 1998.LTRMP 98-R007. 7 pp.

Talent, L. G., Krapu, G. L., and Jarvis,R. L. 1982. Habitat use by mallardbroods in south central North Dakota.Journal of Wildlife Management46:629-635.

Voigts, D. K. 1976. AquaticInvertebrate Abundance in Relation toChanging Marsh Vegetation. TheAmerican Midland Naturalist 95:313-322.

White, G. C., and R. A. Garrott. 1990.Analysis of wildlife radiotrackingdata. Academic Press, San Diego,California.