Loxioides Bailleui)

THE AUK A QUARTERLY JOURNAL OF ORNITHOLOGY

VOL. 101 OCTOBER 1984 NO. 4

ANNUAL VARIATION IN THE DISTRIBUTION, ABUNDANCE, AND HABITAT RESPONSE OF THE PALILA (LOXIOIDES BAILLEUI)

J. MICHAEL SCOTT,• STEPHENMOUNTAINSPRING, 1 CHARLESVAN RIPER,III, 2 CAMERONB. KEPLER,3 JAMESD. JACOBI,1 TIMOTHY A. BURR,4 AND JON G. GIFFINs •U.S.Fish and WildlifeService, Patuxent Wildlife Research Center, Mauna LoaField Station, Hawaii National Park, Hawaii 96718 USA; 2CooperativeNational Park ResourcesStudy Unit, Departmentof Zoologyand Ecology, Universityof California,Davis, California 95616 USA; 3U.S.Fish and WildlifeService, Patuxent Wildlife Research Center, 248 KaweoPlace, Kula, Hawaii 96790 USA; 4HawaiiDepartment of Landand Natural Resources, Division of Forestryand Wildlife, Honolulu, Hawaii 96813 USA; and SHawaiiDepartment of Landand Natural Resources,Division of Forestryand Wildlife, Kamuela, Hawaii 96743 USA

ABSTRACT.--Westudied the distribution, population size, and habitat responseof the Palila (Loxioidesbailleui) during the 1980-1984 nonbreeding seasonsto infer factorsthat limit the population and to develop management strategies.Distribution was fairly constant from year to year. Palila were confined to the subalpinewoodland on Mauna Kea on the island of Hawaii, occurredbetween 2,000 and 2,850 m elevation, and reachedhighest densitieson the southwestslopes. The populationshowed large annual fluctuations,from 6,400 birds in 1981 to 2,000 in 1984. The width of woodland was the most important variable in determining habitat response.Palila were more common in areaswith greater crown cover, taller trees,and a higher proportion of native plants in the understory.Annual variation in Palila density within a habitat reflectedvariation in levels of their staplefood, mamanepods. The main limiting factorsof the population appearedto be the availability of good habitat and levels of their staple food. Palila had strongly depresseddensities in the Pohakuloa flats area.This low density could not be explainedby grosshabitat featuresor food levels. Site tenacity,thermal stress,disturbance, and diseasewere hypothesizedexplanations. Our study indicatedthat the mosteffective management strategies would be the removalof fetal un- gulatesand certain noxiousplants from Palila habitat and the extensionof the woodland zone to areasnow intensivelygrazed. Received 20 May 1983,accepted 9 March 1984.

DISTRIBUTION,population size, and habitat demonstrate the potential value of temporal responseof speciesthat may require manage- analysesof habitat responsein evaluationsof ment to ensuretheir survivalare ecologicalpa- factors that limit bird numbers. rameters of special interest. Annual variation In this study we examine annual variations in these parametersmay offer insight into the in distribution, population size, and habitat re- factorsthat limit the present population, yet sponse acrossthe entire range of the Palila relatively few workers have attempted to eval- (Loxioidesbailleui), a finch-billed Hawaiian hon- uate temporal variation in habitat response. eycreeperfound only in the subalpinewood- Studiesby Raitt and Pimm (1976), Rotenberry lands of Mauna Kea on the island of Hawaii and Wiens (1980a, b), and Rice et al. (1981) (Fig. 1) and one of 30 endangeredor threat- 647 The Auk 101: 647-664. October 1984 648 SCOTTET AL. [Auk, Vol. 101

GENERAL VEGETATION TYPES MAUNA KEA

FOREST RESERVE BOUNDARY CONTOURS IN ".J'Z'METERS

c/o Ac s/rs Ac 1900 s/rs So c/o So c/o My- So s/vsMY-So O:(ns) D:(xg) Pasture 0 I 2KI Burned Area

Fig. 1. The dry nativeforest on MaunaKea, Hawaii. General vegetation types: c/o Ac = closedto open canopykoa (Acacia koa); s/vs Ac = scatteredto veryscattered koa; s/vs So = scatteredto veryscattered mamane forest;c/o So = closedto open canopymamane; c/o My-So = closedto open canopymamane-naio; s/vs My-So= scatteredto very scatteredmamane-naio; D:(ns) = dry nativeshrub; D:(xg) = dry exoticgrass pasture;Burned Area = site of 1979fire.

ened birds in Hawaii (U.S. Fish and Wildlife nonbreedingseason, when population levels Service1983). The populationsize of the Palila were lowest in the annual cycle,food was most was estimated at 1,600 birds in the 1975 non- likely to be limiting, and bird distributionwas breedingseason (van Riper et al. 1978). most likely to reflect local variation in food The behavior and ecologyof the Palila were levels. intensively studied by Berger (1970) and van The objectivesof this studywere to (1) ex- Riper (1978,1980a). Palila are adaptedfor feed- amine annual variation in distribution, popu- ing on green seed pods of the mamanetree lation size, and habitat response during the (Sophorachrysophylla), although mamane flow- nonbreeding seasonacross the range of the ers, naio (Myoporumsandwicense) fruit, and in- Palila; (2) evaluate some of the factorsthat may sectsare alsoeaten, particularly when podsare limit presentdistribution and numbers;and (3) scarce.The breedingseason extends from April development managementstrategies to ensure to September and coincides with maximum the survival of the species.

productionof mamanepods. During the non- STUDY SITES AND METHODS breeding season,local and somelong-distance movement occursin responseto food levels. Study sites.--The entire range of the Palila lies We studied the speciestoward the end of the within the subalpine mamane and mamane-naio October1984] PalilaPopulation and Habitat 649

MAUNA KEA STUDY AREA

1700

1900

...... LIMIT, MAMANE-NAIO FOREST .... ' FOREST RESERVE BOUNDARY v STUDY AREA LIMITS 0 2KM 2•TRANSECT ROUTES • CONTOURS IN METERS 1900 1700

Fig. 2. Locationof study area and transectson Mauna Kea, Hawaii.

woodlandon Mauna Kea (Fig. 1). Our samplinguni- absence and to detect recolonization. The stations verse was the present range of Pallia, a 139 km 2 area were the sample points for monitoring plant phe- (Fig. 2). We also sampledthe north slopesof Mauna nology,counting birds, and verifying the vegetation Kea, where Pallia no longer occur(transects 116 and map. 117 in area 3), in order to document repopulation of Annual surveys were conducted in the nonbreed- the area. Data from this area were not included in ing seasonduring 7-21 February 1980, 20-29 January our analysis. 1981,22 February-4March 1982,25 February-4March The woodlandon Mauna Kea rangesfrom 1,800to 1983, and 6-9 February 1984. We used the variable 3,000 m elevation.The trees are generally short (3- circular-plotmethod with a 6-min samplingperiod 10 m), and canopy cover varies from very scattered (Reynolds et al. 1980). This period was determined (< 5%) to partly closed(60%). Rainfall averages35-75 to be long enough to hear or see Pallia near the sta- cm annually. tion, yet short enough to minimize the chancesof Mamane occurs around the entire mountain, but countingthe samebird twice or of birds moving into naio is mainly restrictedto the southwestslopes. A the count area (Scott and Ramsey 1981). During this detaileddescription of the vegetationis given in Hartt period, observersrecorded each Pallia heard or seen and Neal (1940). exceptfor birds flying high overhead(Reynolds et Samplingdesign.--Information from the 1975 sur- al. 1980). The countswere conductedfrom approxi- vey (van Riper et al. 1978) was used to stratify the mately 0730 to 1300,the period of greatestPallia ac- subalpinewoodland according to Pallia densityinto tivity (van Riper 1978).Observers were assignedto three areas:a high-density area (area 1), a low-den- transectsby random draw within a sampling area, sity area (area 2), and a supplemental area (area 3), and no countswere conductedwhen precipitation or where Pallia once occurredhistorically (Fig. 2). winds greaterthan 4 on the BeaufortScale (>29 km/ Transectswere randomly placed a minimum of 500 h) occurred.Transects 102 and 103, the area of high- m apart and approximately perpendicular to eleva- est density, were each sampled twice and a mean tional contoursin areas1 and 2. Stationswere placed density was computedfor those stations. 150 m apart from approximately 75 m below treeline The 20 observers that participated in the survey to the lower woodland boundary.In area 3 we placed were trained in identifying Pallia and estimatingde- 24 stationsalong two transectsrandomly located in tection distances(Kepler and Scott 1981). All but two the bestremaining woodland only to verify the bird's observersmet hearing standardssuggested by Emlen 650 SCOTTET AL. [Auk,Vol. 101 and DeJong(1981), and one of these did not partici- Canopy height (m); pate after 1980. Mamane biomass--mamane cover (%) times mamane Densityestimates.--Using the methods of Ramsey height (m); and Scott (1981), we calculateddensity estimatesfor Naio biomass--naio cover (%) times naio height (m); Palila. Using values pooled over 4 yr, we derived Total tree biomass--the sum of mamane and naio correction factors for each observer and vegetation biomass; type as samplesizes allowed. The effective area sur- Shrub cover (%); veyed at a stationvaried from 0.69 to 1.48 ha. Grasscover (%)--cover of grassesand herbs; Populationswere estimatedfor eacharea from the Mamane flower biomas--the averageprecentage of Pallia densities at the individual stations. To estimate mamane branches in flower on the 10 trees sam- the total population in the Palila universe, we pied at station multiplied by mamanebiomass; weighted the populations and variancesin areas 1 Mamame fruit biomass--same as for mamane flower; and 2 by the size of the area and pooled the results. Naio fruit biomass--same as for mamane flower; Densitieswere plotted by means of 5-point mov- Pohakuloaflats--a binary variable to indicate the Po- ing averages. Interpolation of densities was done hakuloa flats area and the immediately adjacent subjectively,with considerationgiven to vegetation slopes (bottom of transects 106, 107, and 108), 0 = typesand elevation. We assumedthat density varied absent, 1 = present; in a continuous manner from one point to another. Woodland width (km)--the width of mamane and Habitat description.--Vegetationstructure and com- mamane-naio woodland, measured perpendicular position were recorded at each station by observers to the contour along the transectline, but exclud- who had been intensively trained in estimating ing as unsuitablehabitat areasof scatteredto very vegetationvariables. The observersestimated the av- scattered trees and the areas in Pohakuloa flats. erage crown or foliage coverage of trees, shrubs, herbs, and grassesand the coveragefraction and av- Data analysis.--Alogarithmic transformation,x' = erageheight of mamaneand naio treesin the canopy log•0(x+ 1), wasapplied to stabilizethe varianceand within a 50-m-diameter circle centered on station. correctfor excessivelystraggling tails in Palila den- The percentageof branchesthat had yellow mamane sity and in the biomassof mamaneflower, mamane flowers,fully expandedgreen mamane pods with lit- fruit, and naio fruit. tle or no brown on them, or white naio fruits was To quantify the habitat responseof the Palila and estimated in order to index densities of items in the the annual variation in that response,we computed Palila diet (van Riper 1978). the Pearson correlation coefficient, r, between Palila Station locationswere plotted on high-resolution density and the 14 habitat variables for each year. aerial photographsand U.S. GeologicalSurvey 7.5' The five correlationsof a single Pallia/habitat vari- topographicmaps. Tree coveragesestimated in the able pair were tested for homogeneity by means of field were comparedwith photointerpretative esti- the X2 summation of Z-transformed correlations(Steel mates.This procedureensured standardizationof ob- and Torrie 1980: 280). Correlations were calculated servers and gave more accurate estimates for very between Palila densities in different years to quan- patchy woodlands. tify the tendencyfor birds to occuron the samesites. Vegetation maps of the study area were prepared Multiple regressionwas performed on the data of at a 1:24,000scale. Vegetation types were initially eachyear to identify the strongestvariables affecting interpretedon aerial photographsat a 1:45,000scale Palila response, given the relations among other with a mirror stereoscopeof 3 x and 6 x magnifica- variables(Draper and Smith 1981). Crown cover, can- tion. These preliminary map units were verified on opy height, and total tree biomasswere excludedbe- the ground,and a helicopterwas usedto resolvein- causeof their high correlationwith mamaneand naio terpretation problemsin someareas. The final map- biomass. We used a structured stepwise procedure unit boundarieswere compiled by meansof a Kern that worked down through a seriesof hierarchical PG-2 plotter. levels,adding significant(P < 0.05) variablesto the Habitatresponse.--We derived 14 independent vari- regressionequation and deleting variablesthat be- ablesfrom the habitatdata and graphicallyexamined cameinsignificant. Variables were organizedinto hi- univariate distributionsand Palila responseto each erarchicallevels to reflect increasinglydetailed hab- variable to determine necessarystandardizations and itat features(see Scott et al. in press).Variables were transformations. The independent variables in the consideredfor inclusion in this order: (1) elevation, analysiswere: (2) [elevation]2, (3) mamane and naio biomass, (4) shruband grasscover, (5) flower and fruit variables, Elevation (m)--from the USGS 1:24,000topographic and (6) Pohakuloa flats and woodland width. The last map; two variablesare nonlocal and extensive;we placed Elevationsquared--to representpotential ditonic re- them lastto avoid confoundingthem with localized sponse; responses. Crown (%); To compareyears further, we groupedstations by October1984] PalilaPopulation and Habitat 651

DISTRIBUTION and DENSITY of PALILA 1980 ;""'"•.,"•"'"'

170(

100 - 200 1900

'( lO 200-400 I 2KM BIRDS/ KM2

Fig. 3. Distribution and abundance of Palila in 1980.

the vegetation units used in mapping. The Palila patterns, variation at finer scalesis glossed over, and densities at the stations in each group were com- nonlinear interactions may be overlooked entirely paredby meansof Fisher'sprotected least significant (Jamesand Wamer 1982).To display someof the mul- difference(lsd) test for multiple comparisons(Steel tidimensional variation in response,we constructed and Torrie 1980:173). This is a two-part test:first, we contour graphs of Palila response to independent computed one-way analyses of variance to test variables by plotting each station, smoothing over whether or not there was significant variation be- cells of 2-5% of the total graph area, and drawing tween years; when this test was significant, we con- contoursby eye. The independent variablesused for structed the lsd test from s, the square root of the axes were chosen for their predictive power and poolederror variance,as lsd = 1.98s2%/•, where n management interest. is the number of stationssampled each year. Means differing by more than the lsd were significantat the 5% level. RESULTS The countstaken on consecutivedays on transects 102 and 103 were comparedto assessdaily variation Distribution.--Distributionalpatterns of the due to Palila movement.To determine whether daily Palila were rather constant during our 5-yr variation was more prominent between or within study (Figs. 3-7) and quite similar to those vegetation units, an analysis of variance was per- found in 1975 (van Riper et al. 1978). formed on stationsgrouped by vegetationunits. The highestdensities (>200 birds/kin2) were The adequacyof the samplesize for a quantifica- alwayscentered around the 2,300 m elevational tion of habitatresponse was checkedby computing contour near Puu Laau on transect 102 in area correlationsand regressionequations from randomly 1. This area lies immediately above an ecotone constructedsubsamples of 25% and 50% of the full between the mamane and mamane-naio wood- data set. Resultsfrom the 50% subsampleswere sub- lands.A secondmuch smaller high-density cell stantially identical to those from the full data set. was found 5 km southeast on transect 107; it Deviationsamong the 25%subsamples suggested that 100-150 stations were needed to allow us to make was especiallyprominent in 1981. Areas with statementsabout habitat responsein the sampling a density of more than 200 birds/kin 2 never universe. exceeded7% of the Palila's range, and areas Becausesummary statistics characterize general with fewer than 10 birds/kin 2 accounted for 652 SCOTTET AL. [Auk, Vol. 101

DISTRIBUTIONand DENSITYof " :•'• '• PALILA,1981 ...... /"'*'•'?'• ...... *• ? •'*"•'• •'"••;•' •/ .....\.....?..

1700

200-400 1900 '"?

lO-so c....?•'""**'; I •400 2KM

Fig. 4. Distribution and abundance of Palila in 1981.

DISTRIBUTION and DENSITY of ....,--,--•.-• P&LIL&,1982 /,.,...•..../'•. .?t-¾; •..i•:. ?....,...... •...•..,,•...... •

1700

100- 200 1900 50-100

• 10 I 2KM BIRDS/ KM2

Fig.5. Distributionand abundance of Palilain 1982. October1984] PalilaPopulation and Habitat 653

DISTRIBUTION and DENSITY of PALl LA, 1983 :."•"•....'":.

1900 50-100

10 -50

Burned Area

1 2KM BIRDS/ KM2

Fig. 6. Distribution and abundance of Palila in 1983.

more than 65% of the range every year (Figs. 1980, 1981, and 1982 estimates. In the 1975- 3-7). In 1983, a small shift in the Puu Laau 1981 period, therefore,Palila populationsap- density cell correspondedto higher resource pear to have increasednearly fourfold, but by levels in mamane-naiowoodland, as compared 1984 the population lost 68% of its 1981 num- with those in pure mamanewoodland. bers. The upper elevationallimit of Palila gener- From 1980to 1981the populationapparently ally coincidedwith treeline at about 2,850 m in increased91%, only to return in 1982 to the both areas 1 and 2. In area 1 the lower eleva- 1980 level. This points to a high potential for tional limit of the ?alila was approximately large annual fluctuationsin numbers,perhaps 2,000 m and generally marked a transition to resultingfrom optimal weather and high food scrub forest or grassland.In area 2 the lower stocksduring the May-September1980 breed- limit was approximately2,500 m at the bound- ing season,followed by high mortality (of ary of the Mauna Kea Forest Reserve and was youngbirds?) due to poor weatheror depleted characterizedby scatteredto very scatteredtrees food stockslater during 1981. in heavily grazedpastures. These range limits Rainfall records from Puu Laau indicate that are much higher than thosefound in the 1890's 1980 had the greatestrainfall of all yearsfrom (Perkins 1903). 1973 to 1983, whereas 1981 and 1983 had the Populations/ze.--The estimatedPalila popu- least rainfall of that period (R. Bachmanun- lation varied from 6,400 birds in 1981 to 2,000 publ. data). Van Riper (1980b) found that main 1984(Table 1). The differencesin population mane fruiting respondedstrongly to precipi- size were significant (P < 0.05) between 1981 tation; thus, there may have been a large and the otheryears, but not among1980, 1982, mamanepod crop in 1980 that supportedthe 1983,and 1984.In 1975,the Palila population boomobserved in 1981,followed by a cropfail- was estimated at 1,614 (95% CI = 1,057-2,171) ure in 1981 that led to the population "bust" during the nonbreedingseason (van Riper et in the 1982survey. The low populationin 1984 al. 1978).This value is significantlybelow the also seems to reflect adverse weather. Our data 654 SCOTTET AL. [Auk, Vol. 101

DISTRIBUTION and DENSITY of PALILA, 1984

1900

50 - 100 I I 0 1 2KM

BIRDS/ KM2

Fig.7. Distributionand abundance of Palliain 1984.

are inadequateto determinethe causesof an- Annualand daily correlation in occurrence.--The nual population fluctuations,because we re- correlationsbetween years for Palila densities cordedresource levels for only a short interval at the same station (Table 2) varied from r = of time and not the whole year. Apparent dou- 0.16 to 0.51 and were all significant. This in- bling or halvingof insularpasserine popula- dicates that Palila tended to occur at the same tionsfrom one yearto the next is not unusual. sitesthat they have in the past,although sub- J. L. Sincock(unpub,1. data) has documented stantial variation also occurred. wide fluctuations in the numbers of Laysan Two transectsin the areasof greatestPalila Finches(Telespyza cantans) and Nihoa Miller- densitywere replicatedon consecutivedays to birds (Acrocephalusfamiliaris kingi) on the lee- quantifythe extentof daily movementof the ward islands of Hawaii. population.The correlations for consecutiveday In area 1, populationsdiffered significantly counts were low, with a mean r of 0.36 (0.08 between1981 and the other years.The ratio of SE) and a range of -0.17 to 0.81 for the 10 the highest(1981) to the lowest(1984) popu- correlations. When the sample points were lation was 3.4. Densities in area 1 averaged 2.4 stratifiedby vegetationunit, correlationsim- timesas high as in area2, and, owing to the provedto a mean r of 0.74 (0.08 SE),with a greatersize of area1, from 79%to 90%of the rangeof 0.24to 0.99.All 10 correlationscom- total Palila populationoccurred there. putedfrom vegetation units were higher than In area2, populationsdid not differ signifi- thosecomputed from stations, suggesting either cantly betweenyears. This was partly due to that Palila movement between vegetation units the more scattered occurrence of Palila and was less than movement between stations or lower samplingintensity, resulting in higher that daily Palilamovement tended to resultin coefficientsof variation, but the ratio of highest the same relative numbers in each vegetation (1980)to lowest(1984) population, 2.2, is lower unit. than in area 1 and suggestsa more stablepop- Habitat correlatesand habitatresponse models.- ulation. The correlations between Palila density and October 1984] Palila Populationand Habitat 655

TABLE1. Summaryof populationestimates for the Palila.

Area 1 Area 2 Palila universe

Area (kin 2) 97 42 139 Number of stations 237 88 325 ?opulation estimate 1980 2,650 701 3,350 1981 5,795 615 6,410 1982 2,612 693 3,305 1983 1,882 386 2,268 1984 1,701 321 2,022 Population, 95% CI 1980 1,635-3,665 207-1,195 2,240-4,480 1981 4,120-7,470 221-1,009 4,690-8,131 1982 1,703-3,521 321-1,065 2,323-4,287 1983 1,283-2,480 58-714 1,584-2,951 1984 1,121-2,281 115-527 1,407-2,637 Number Pallia counted 1980 136 17 153 1981 229 14 243 1982 131 18 149 1983 117 10 127 1984 87 9 96 Frequency (%) 1980 19% 10% 17% 1981 27% 12% 23% 1982 22% 16% 20% 1983 19% 7% 16% 1984 16% 10% 15%

habitat variables provide an overview of Palila ally exoticgrasses and herbsindicative of a long habitat responseand its annual variation (Ta- history of heavy disturbance,was negatively ble 3). Every year, the width of subalpine associated with Palila in three models, but in woodland had the strongest correlation with the 1983 model shrub cover was entered in- Palila density. Palila also consistentlyrespond- stead.Shrub cover was indicative of lighter dis- ed strongly to total tree biomass,crown cover, turbance and was negatively correlated with and canopy height. Significant annual varia- ground cover (r=-0.58, P < 10 40). These tion occurred in the correlations with mamane models may therefore be interpreted as repre- biomass, naio biomass, and naio fruit. These senting avoidance of heavily disturbed vege- differencesseemed mainly due to phenological tation. Multiple regressionwas thus helpful in variation and resource shifting, as discussed identifying important variables that would be below. A negative responseto the Pohakuloa overlooked in a bivariate analysis. The corre- flats area occurredeach year, and only two Pali- lations of Palila with shrub and ground cover la were found on the 42 stations in the flats were modestuntil zone width, phenology, and over the 5-yr period. A multivariate perspective on response to TABLE2. Correlationsof Palila densitybetween years habitat was given by the regressionequations at the same stations. All correlations are signifi- (Table 4). From 17% to 36% of the variance was cant, P < 0.01. accounted for, these somewhat low values re- flecting the patchy distribution of Palila over 1980 1981 1982 1983 even short distances.The most important vari- 1981 0.36 able in every case was the width of mamane- 1982 0.51 0.37 naio woodland. Pohakuloa flats had a negative 1983 0.23 0.51 0.27 1984 0.42 0.27 0.35 0.16 term in four of the models. Ground cover, usu- 656 SCOTTET AL. [Auk,Vol. 101

TABLE3. Correlations between Palila density and habitat variables. An asterisk after the variable name indicatessignificant variation between years, P < 0.01. Significancevalues for the correlationsare I rl = 0.11, P = 0.05; Irl = 0.18, P = 0.001.

Variable 1980 1981 1982 1983 1984

Elevation - 0.01 0.04 0.02 - 0.10 - 0.08 Total tree biomass 0.24 0.27 0.13 0.25 0.23 Crown cover 0.16 0.20 0.11 0.27 0.11 Canopy height 0.20 0.20 0.09 0.12 0.22 Mamane biomass* 0.27 0.09 0.11 0.03 0.31 Naio biomass* 0.07 0.30 0.08 0.34 0.00 Shrub cover -0.17 -0.03 -0.11 -0.07 -0.26 Grass cover -0.07 -0.20 -0.04 -0.03 0.08 Mamane flower 0.20 0.08 0.21 0.03 0.03 Mamane fruit 0.24 0.26 0.07 0.21 0.14 Naio fruit* -0.05 -0.03 0.06 0.28 -0.03 Pohakuloa effect -0.17 -0.16 -0.17 -0.16 -0.12 Woodland width 0.37 0.36 0.33 0.37 0.35

other factors were taken into consideration, naio woodland units. In each series, units when it became apparent that Palila discrimi- ranged from taller, more forested to shorter, nated against heavily disturbed areas. Naio more open habitat. Additional contrastswere habitat componentsvaried widely among the made between native and exotic understories years and are discussedbelow. In the 1981 and between points lying within and outside model, the two naio variables were entered with Pohakuloa flats. opposite signs, indicating that Palila were us- Palila densities were highest in tall, more ing areas of mamane-naio woodland that had forested habitat in both the mamane and ma- relatively low naio fruiting intensity. mane-haloseries, the maximum density of 206 Variationin densityamong habitat types.--Palila birds/kin 2 occurring in 1981 in tall, partly densities showed striking variation among closed, mamane-naio woodland with a native habitat types (Table 5). The variation among understory (Table 5). Tall, partly closed, ma- habitat types and acrossyears generally paral- mane-naiowoodland had significantlyhigher leled the patternsin the previousanalysis. Palila density (basedon the 5-yr pooled values) The habitatsfell into two general series:pure than did otherwise identical open woodland mamane woodland units and mixed mamane- (t = 4.2, P < 10-4).

TABLE4. Habitat responsemodels constructedby stepwisemultiple regression.All equationsare highly significant(P << 0.001).Entries for variablesare t-statistics.All variablesentered are significantat P < 0.05; * indicates P < 0.001.

Variable 1980 1981 1982 1983 1984

R2 (variance accounted for) 0.236 0.361 0.171 0.300 0.178 Elevation ..... (Elevation) 2 ..... Mamane biomass 4.3* ------4.6* Naio biomass -- 8.1 * -- 3.0 -- Shrub cover ------2.4 -- Grass cover -3.4* -4.8* -2.7 -- -- Mamane flower -- -- 2.9 -- -- Mamane fruit 2.1 .... Naio fruit -- -5.0* -- 2.7 -- Pohakuloa flats -2.0 -2.4 -3.2 -6.2* -- Woodland width 5.7* 8.3* 5.7* 6.5* 5.6* October1984] PalilaPopulation and Habitat 657

T^BI•E5. Annual variation in Palila density by habitat type. Height: tall 5-10 m, short 2-5 m. Crown cover: partly closed50-60%, open 20-40%, scattered5-20%, very scattered0-5%. u/s = understory.Superscripts distinguishsignificantly different yearsaccording to the protectedlsd test. Habitattype Stations Paliladensities (birds/kin 2) (height,cover, understory) sampled 1980 1981 1982 1983 1984 Pooled Pure mamane woodland Tall, open, native u/s œ 36 87 b 69 b 72 b 13 a 41 a,b 57 SE 26 21 22 5 10 8 Tall, open, exotic u/s œ 21 36 14 21 0 53 25 SE 14 8 15 0 19 6 Short, open, native u/s œ 53 8 16 14 16 0 11 SE 5 7 5 6 0 2 Short, scattered, mixed u/s œ 35 13 21 25 1 14 15 SE 6 7 8 1 7 3 Short, very scattered,mixed u/s œ 14 0 0 0 0 0 0 SE 0 0 0 0 0 0

Mixed mamane-naio woodland Tall, partly closed,native u/s œ 35 50 a 206 b 37 a 73 ß 23 a 78 SE 15 38 9 15 7 9 Tall, open, native u/s (outside • 42 23 a 79 b 34 ' 18 ' 13 a 34 Pohakuloa flats) SE 7 20 13 6 6 5 Tall, open, native u/s (in œ 32 0 2 0 0 2 1 Pohakuloa flats) SE 0 2 0 0 2 1 Tall, open, exotic u/s œ 22 5 4 16 26 12 12 SE 4 4 6 10 6 3 Short, open, native u/s œ 13 12 0 0 6 6 5 SE 12 0 0 6 6 3 Short, scattered, mixed u/s • 22 0 4 5 0 0 2 SE 0 4 4 0 0 1

Tall open woodlands with native understo- mamane (t = 5.4, P < 10-?), 7.2 times in ma- ries had significantly higher Pallia densities mane-naio (t = 4.8, P < 10-6). Short scattered than did those with exotic understories (t = 3.2, woodlands with native or exotic understories P < 0.01 for mamane and t = 3.5, P < 0.001 for generate about the same level of Palila re- mamane-naio).This probably reflected greater sponse as short open woodlands (P > 0.30). food resourcesin native understory from ma- Short very scatteredmamane "woodlands" had mane and naio trees less than 2 m tall and from no Pallia, significantly fewer than short scat- Osteomeles. Exotic understories, which were tered mamane woodlands (t = 5.4, P < 10-7). comprisedalmost entirely of grassesand cer- Although highestPalila densitiesoccurred in tain rank herbs, may also provide less shelter. partly closed, mamane-naiowoodland, pure A glaring exception to the strong response mamanewoodland supportedhigher densities of Pallia to tall, open woodland with native than structurally similar mamane-naio wood- understory occurred within Pohakuloa flats, land. In tall open woodland with native under- where densitieswere significantly(t = 6.2, P < story, Palila densitieswere 1.7 times higher in 10-8 ) below densitiesin virtually identical hab- mamane than in mamane-naio (t = 2.3, P < itat with comparable phenology at about the 0.05); in tall open woodland with exotic under- same elevation. story, 2.0 times higher (t = 1.9, P < 0.07); in Next, we compared tall and short open short open woodland, 2.3 times higher (t = 1.6, woodlands with native understories. In both P < 0.12); and in short scattered woodland, 7.8 habitat seriestall woodlands had significantly times higher (t = 4.4, P < 10-5). Overall, Pallia higher densities:about 5.3 times as high in densitieswere higher in mamane than in struc- 658 SCOTTET AL. [Auk,Vol. 101

TABLE6. Annual variation in flowering and fruiting intensity in habitatswith significantannual variation in Palila density. Superscriptsdistinguish significantly different years accordingto the protectedlsd test. Habitattype Phenologyintensity (%branches inflower/fruit) (height, cover, understory) 1980 1981 1982 1983 1984 Pure mamane woodland Tall, open, native understory Mamane flowers 13 Mamane fruit 4 a 12 b 28 c 2 a 30 c

Mixed mamane-naio woodland Tall, partly closed,native understory Mamane flowers 27 c 20 b.c 55 d 15 ',b 8 a Mamane fruit 9 a,b 13 b 4 a 14 b 8 a,b Naio fruit 7 •,• 15 b.• 9 •.b 19 • 3 • Tall, open, native understory (outside Pohakuloa flats) Mamane flowers 15 ß 9 • 35 b 9 a 9 a Mamane fruit 2' 25 • 5 a 5' 3 • Naio fruit l0

turally similar mamane-naio(X 2 = 38.6, df = 8, partly closedmamane-naio with native under- P < 10-s). story coincidedwith high mamaneflower, ma- In general, Palila thus preferred (1) greater manepod, and naiofruit levels.Phenology val- crown cover, (2) greatercanopy height, (3) na- ues were alsohigh in 1983.At first glance,the tive instead of exotic understory, (4) locations 1983density value (73 birds/km2) does not seem outside Pohakuloa flats, and (5) pure mamane to corroboratethe patternof high density/high woodland over structurally comparablemixed phenology,but the Palila populationwas very mamane-naio woodland. low that year, and 73 birds/km2 was by far the Annual variation in habitat response.--Annual highestPalila densityin the 11 habitattypes in variation in Palila densitieswas significantly 1983. Comparedwith the mean density, den- greater than would be expectedif the variation sities in tall partly closed mamane-naio were were random in three habitat types: tall open 2.1 times the annual mean in 1980, 4.5 in 1981, mamanewoodland with native understory,(F = 1.6 in 1982, 4.5 in 1983, and 1.6 in 1984, sug- 2.6, P = 0.05); tall, partly closedmamane-naio gestingsimilar high responsesin 1981and 1983. with native understory (F = 13.4, P < 10-4ø); Finally, in tall open mamane-naio with na- and tall, open mamane-naiowith native under- tive understory,both Palila densitiesand ma- story (F = 5.1, P < 0.001).This annual variation mane pods were at significantly higher levels may be plausiblyattributed to variation in food- in 1981than in other years.In the 1981 habitat- resourcelevels, particularly mamane pods, the responsemodel (Table 4), the combination of staple of the Palila diet. positive naio biomassand negative naio fruit In tall open mamanewith native understory, termsmay indicatethat Palila were keying in Palila densitiesin 1983were significantlylow- on the mamanepods in this habitat and ignor- er than in 1980-1982 (Table 5), and both ma- ing naio fruit in other areas.Mamane flowers mane pods and mamane flowers were at their were significantly more abundant in 1982 than lowest level in 1983 (Table 6). Hence, in 1983 in otheryears, and thismay haveled to slightly Palila may have experiencedsevere food short- higher Palila densities in 1982 than in 1980, age in this habitat and emigrated, whereas in 1983, or 1984 for that habitat. the low pod year of 1980, pods were twice as Contourgraphs.--To give an impressionof an- abundant as in 1983, and mamane flowers were nual variation in habitat response, we con- also more abundant and may have been an al- structedcontour graphs(Fig. 8) of Palila den- ternative food. sity as a function of two important variables-- Extremelyhigh Palila densitiesin 1981in tall, woodland width and total tree biomass. The October1984] PalilaPopulation and Habitat 659

1980 )981 c

o 120 short

20050 -400- 100 []birds 10-50/ km 2

1 2 3 4 5 1 2 3 4 5

1982 1983 1984 l-Cll

tall

s•)ort 20 49ot::':.::.:.:.

WOO LA D WIDTH (KM)

Fig. 8. Habitat responseof Palila with respectto woodland width and total tree biomass.

latter variable was logarithmically scaledto in- bier (Dendroica kirtlandii) during its recent dex general forest development. population decline (May field 1972, Byelich et The graphic responsesmore or less paral- al. 1976). leled the correlations of Palila with these two variables (Table 3). The dependence of Palila DISCUSSION on wider woodlands and structurally more complex habitat appeared in each graph, but Comparabilityof methodswith 1975 survey.- annual variation was also apparent. Palila were Becauseour method of determining range and especially associatedwith wide woodlands in density differed from that of van Riper et al. 1980, 1982, 1983, and 1984, the years of reduced (1978), we should considerthe possibilitiesthat abundance,but in 1981 the Palila ranged fre- may accountfor the differencesin population quently into narrower woodlands.Similarly for size. The Emlen transect method that van Riper total tree biomass,the broadestrange of occur- et al. (1978) used has been directly compared rence was in 1981. with the variable circular-plot method in sev- The graphs also indicate that Palila appar- eral studies (Edwards et al. 1981, Anderson and ently tolerated very narrow woodlands more Ohmart 1981). The two methods provide com- readily than very sparsehabitat. This may rep- parable numbers when sample sizes exceed30. resent the occasional use of small, somewhat In addition, densities derived from variable cir- isolatedpatches of suitablehabitat and the gen- cular-plot estimatesand mapped plots are very eral avoidance of scatteredtrees (no matter how similar (DeSante 1981). Thus, we believe that much good habitat is nearby) and, hence, illus- the variable circular-plot method provides trates a degree of active habitat selection. densityestimates comparable to thoseobtained These results thus fit Diamond's (1975) "hot with the Emlen line transect. spot" model of a speciesexpanding out from Van Riper et al. (1978) defined range as that and contractingin to the best remaining habi- area bounded by a line connecting the outer- tat as the population fluctuates.The Palila pat- most sightings. This is a conservativemeasure tern is also similar to that of the Kirtland's War- of range,which includesless area with low bird 660 SCOTTET AL. [Auk,Vol. 101 densitiesthan doesa range basedon incidental Bachman, P. Banko, W. Banko, D. Jenkins, H. records, elevation, and presence of suitable Sakai, R. Walker pers. comm.). habitat. The density values are thus higher in Palila were extirpated from Kona between the more conservativedefinition of range, but 1896 and 1936, we believe most probably the population estimatesderived from the two around 1900-1910, contemporaneouswith the methods should be comparable. extinction of the three other finch-billed hon- Area of historicrange occupied.--Van Riper et eycreepers.Ranching activities in the late 19th al. (1978) estimated that in 1975 the Palila oc- century (e.g. water impoundment, forest clear- cupiedonly 10%of its historicalrange, defined ing) most likely acceleratedthe mosquitoinva- as equivalent to the range of mamane forests sion of higher elevationsby creatinghigh-ele- on Hawaii. We now have a more accurate es- vation breeding sites.The rapid disappearance timate of the area of mamane forests, 545 km 2, of the four speciesin Konamay havebeen due than was availableduring 1975.We found the to their lack of immunity to avian malaria or Palila in 139 km 2, or 25.6%, of the 545 km 2 of pox, asWarner (1968) and van Riper et al. (1982) extant mamane woodlands. suggested.In addition, Kona is subjectto strong The presentdistributional pattern suggestsa winter winds from the south and west that blow relict population. Indeed, as a result of hab- mosquitosup to the higher elevations. Water itat degradationand avian diseasesat lower el- troughsat 1,500rn elevationin Konaheld large evations,Palila may now be restrictedto areas populationsof mosquitolarvae in the summer that historicallywere lessthan optimal habitat of 1978 (J. M. Scott pers. observ.). but that are the best remaining. Role of feral ungulatesin rnamane-naiowood- Historical patterns of distributionand abun- lands.--Feral ungulates have had a severe im- dance.--Considering the present distribution pact on Palila habitat on Mauna Kea and ap- and abundance of the Palila in a historical con- pear to be the majorfactor that indirectly limits text gives perspective on the factors that pres- the present distribution of the Palila. Ungu- ently limit the population. Before the first lates generally prefer mamane, Osteomeles,and Polynesiansarrived about A.D. 400, the low- native grassesover naio and exotic herbs and lands of the main islands supported extensive grasses(Giffin 1976, 1982;van Riper 1980b). dryland foreststhat included mamane.Fossil In 1823, immense feral cattle (Bostaurus) herds evidence (Olson and James 1982) indicates that were reported on Mauna Kea, and by 1904 an Palilaapparentl_,v occurred down to sealevel in estimated 10,000 feral cattle were on Mauna Kea these•-rests on Oahuand probably on other (Ellis 1917). By 1931 feral cattle had been elim- islandsas well. The Polynesiansconverted dry inated from the upper elevationsof Mauna Kea lowland foreststo grasslandsby burning (Kirch (Tomich 1969). Damage to mamaneforests by 1982). By the time of Western contact(1778), feral cattle is incompletely documented,but Palila, along with several other finch-billed Rock (1913) and Warner (1960) describe some honeycreepers(Rhodacanthis fiaviceps, R. pal- effects. rneri,and Chloridopskona), were evidently re- Feral sheep (Ovis aries)did not becomeabun- strictedto the mid- and high elevation dry for- dant until after the decline of cattle, although ests of Kona and Mauna Kea. Scott et al. (in sheep carcasseswere found near the summit of press)estimated that Palila occupyless than 5% Mauna Kea in 1822 (Ellis 1917), and Bishop of their pre-Polynesianrange. (1852) estimated3,000 feral sheepon Hawaii in In 1891, Palmer (in Rothschild 1900) found 1851. Feral sheep numbers were limited for Palila abundant in Kona from 1,200 to 1,800 m many yearsby feral dogs(Canis familiaris) (Judd elevation. Perkins (1893) also found them quite 1936), but the dogs were eliminated from common in Kona in 1892, but in 1894 and in Mauna Kea around 1920by poisoning(Tomich 1896 he found only two birds (Munro 1944, 1969). By 1937, an estimated 40,000 sheep oc- Banko 1979). Munro (1944) visited Perkins' curred on Mauna Kea (Bryan 1937),and control Kona sites in 1936 and failed to find Palila, as measureswere initiated to protect the habitat. did Richards and Baldwin (1953) in 1946. Palila Only 200 sheepremained (Giffin 1976)in 1950, were found in Kona neither during the 1978 when managementof Mauna Kea changed to Hawaii ForestBird Survey (Scottet al. in press) sustainedyield hunting. Sheep gradually in- nor during recent searchesby other workers creasedin numbers,fluctuating between 1,000 (C. van Riper, III, J. G. Giffin pers. observ.; R. and 4,000, from 1955 to 1975 (Giffin 1976), but October1984] PalilaPopulation and Habitat 661 the subalpine woodland steadily deteriorated range of good mamane forest thus provide the with as few as 1,500 sheep on the mountain best opportunities for Palila to exploit shifts in (Warner 1960, Scowcroft and Giftin 1983). Ber- abundance of mamane pods and flowers. This ger (1981) estimated that treeline had receded is reflected by the strong correlation we found 200 m in elevation since the 1850's due to sheep between Palila density and woodland width. activity. Following a legal ruling under Section Relatively low densities of Palila on the south- 9 of the Endangered SpeciesAct, virtually all east slopes of Mauna Kea (transects 109-113) feral sheep were removed from Mauna Kea by may be due to the narrow width (1-2 kin) of July 1982. Small numbers of sheep still immi- the mamane woodland in that area. Allowing grate from adjacent lands, however. the heavily grazed pasture that lies below the Mouflon (Ovis musimon)were released in 1963 forest reserve fences to return to mamane on the east flank of Mauna Kea (Tomich 1969) woodland offers a good opportunity for in- and are potentially as destructive as feral sheep creasingPalila and re-establishingpopulations to the subalpine woodland (Giftin 1982). The in these areas. Planting trees may be necessary continued presence of sizable populations of in some areas. mouflon in areas2 and 3 portendsfurther hab- Branchsize potentially affectsPalila density, itat degradation there. becausePalila are fairly heavy birds (40 g) and We have noted a vigorous resurgenceof na- are unable to obtain pods hanging from small- tive understoryvegetation in area 1, from which er twigs (van Riper 1978). Trees with multiple feral sheep have been removed. A similar re- small trunks and correspondingly smaller generation of the understoryand mamane trees branchesoccur in someareas as a result of heavy in area 2 would be expected if mouflon were browsing pressure, and elsewhere small trees eliminated from the mountain. Several sight- indicate recent release from past browsing ac- ings of mouflon in area 1 since 1981 (J. G. Giftin tivity. A combinationof small treesand narrow unpubl. data) further indicate the need to re- elevational width of subalpine woodland may move mouflon. explain the absenceof Palila on the north and Feral õoats (Caurahircu$) fiõure prominently northwest slopes of Mauna Kea. These areas in the veõetationdestruction on Mauna Loa and are isolated from populations on the southwest Maui and apparently occurred on Mauna Kea and southeastslopes by large gaps in suitable in some numbers in the 1930's, althouõh they habitat and by very narrow habitat corridors were not as destructive as sheep (Judd 1936). on the east slopes, which makes colonization Goats persisted on Mauna Kea until their vir- difficult. tual extirpation durinõ the 1982 sheep remov- Pohakuloafiats.--The stationslying in Poha- al. kuloa flats exhibited an extraordinary depres- The deõradation and fraõmentation of the sion in Palila density compared with stations subalpine woodland by feral unõulates has outside that area in similar habitat (Table 5). contributed õreatly to the reduction in ranõe Reports from R. L. Walker and E. R. Kosaka, and population size of the Palila. The neõative who lived in the flats for a period of years in impact of feral unõulates on Palila habitat is the 1950's and 1970's, indicate that Palila have also made clear by the neõative responseof been absentfrom the flats for at least the past Palila to õrasscover and their positive response 30 yr. Although the absence of Palila in the to less heavily browsed shrub understories.It flatsremains unexplained at this point, the most is expected that the numbers of Palila will in- plausible hypothesesare site tenacity, thermal creasewith the removal of feral unõulates (Ber- stress, and disease, several other hypotheses õer et al. 1977), althouõh the responsewill be being lessprobable. delayed until mamane trees reach suitable sizes. The absence of Palila in the flats may be Presentlimiting factors.--Because mamane pods causedby some unspecified factor that operat- are the chief food of Palila, seasonal variation ed once, say, in the past 50 or 100 yr, such as in pod abundanceand distribution affectsPali- the destruction of vegetation by sheep 50 yr la abundance and distribution. Van Riper ago. According to this hypothesis, Palila could (1980b) found that mamane phenoloõy varied now inhabit the flats,but becauseof strongsite with elevation and exposure,so that edible pods tenacity there are virtually no potential colo- were available on the mountain at all times of nizers. Site tenacity might also work in con- the year. Areas with the õreatest elevational junction with a factor that operatesintermit- 662 Scott ETAL. [Auk,Vol. 101 tently every severalyears, such as infrequent (1978) found all three predatorspresent at Puu incursionsof malaria-infectedmosquitoes. Laau, where highest densitiesof Palila, Elepaio Palila are relatively intolerant of heat com- (Chasiempissandwichensis), and Common Ama- pared with other passerines,and their upper kihi (Hemignathusvirens) occurred. The absence criticaltemperature is only 31øC(Weathers and may be due to stochasticpopulation fluctua- van Riper 1982). Temperature recordsat the tions that resulted in local extinction, but the Pohakuloa Endangered SpeciesCaptive Prop- habitat, and presumably the population, was agationFacility indicated monthly highs in ex- continuouswith the high-density areas.In as cessof 33øCfor over half the months during much as the flats would support 800 Palila, as- 1978-1982 (R. Bachman unpubl. data). At a suming 35 birds/kin 2, stochasticfluctuation to nearby airfield weather station, highs above extinction seems improbable. Studies with 29øCwere recorded for 3 months of the year, cagedand radio-taggedbirds should prove use- and a high of 31øCfor 1 month, during 1979- ful in testing hypotheses. 1980 (BradshawAirfield, unpubl. data). At Puu Managementstrategies.--Our study suggests Laau, center of Palila distribution, the highest several managementstrategies to increasePali- temperature during 1973-1975 was 29øC(van la numbers and improve the survival pros- Riper 1980a). Thus, temperaturesin the flats pects.The most urgent need is control, or pref- may rise to levelsthat are unsuitablefor Palila erably elimination, of mouflon on Mauna Kea on sunny calm days. to prevent further degradationof mamane-naio The occasionalpresence of diseasein Poha- woodland and to encourage forest regenera- kuloa flats may kill Palila. This hypothesis is tion, particularly near timberline. Substantial plausible,because: (1) Palila are very suscepti- tree and understoryregeneration has already ble to avian malaria (van Riper et al. 1982) and occurred in area 1 following the removal of in all probability are susceptibleto avian pox; sheep in 1982. (2) night-biting mosquitoes,the vectorfor ma- Even with feral ungulatesremoved, the sub- laria, occasionallyoccur in the flats and breed alpine woodland is threatenedby exotic plants in water impoundmentsin that area; (3) large and fire. In 1979 a fire consumed 200 ha of ma- populationsof exotic birds, especiallygallina- mane-naiowoodland adjacentto the high-den- ceous gamebirds, House Sparrows (Passerdo- sity area. Fountain grass(Pennisetum setaceum), mesticus),House Finches( Carpodacusmexicanus), an aggressive,fire-adapted African tussockgrass Red-billed Leiothrix (Leiothrix lutea), and Com- that has recently establishedcolonies on the mon Mynas (Acridotherestristis), are established south slopes of Mauna Kea, could fuel future around sitesof human activity in the flats, and fires.Another potential problem is German ivy may serve as reservoirsfor pathogenic organ- (Seneciomikanioides), a choking, climbing liana isms; and (4) avian disease is believed to be that has recently been discoveredas a sporadic responsiblefor a large portion of the extinc- infestation over about 500 ha near Puu Laau. tions and range contractions of the native Effective control measures for these noxious in- Hawaiian forest birds that have occurred in the troductionsare needed before they expand be- pastcentury (Warner 1968,van Riper et al. 1982, yond their tenacioustoeholds. Scott et al. in press).Because human activities An appropriate long-term managementstrat- and some exotic passerinepopulations (House egy is to restore the very scattered mamane Sparrows,Common Mynas) are restrictedto the woodlands that lie below transects 110-114 on flats, the disease-affectedarea would be fairly the southeastslopes of Mauna Kea by remov- small and constitute the flats and immediately ing cattle and possiblyplanting seedlings.The adjacentslopes, exactly where Palila are inex- importance of woodland width in ensuring a plicably absent. good stock of mamane pods year round indi- Many other potential explanationscan be cates high potential for dramatic response in hypothesizedfor the absenceof Palila in the Palila density as these areas return to wood- flats. Disturbancefrom military activities--ar- land. Cattle grazing should also be barred on tillery practice,troop movements--maybe high the northwest and north slopesof Mauna Kea at times. Predation from feral house cats (Felis to re-establish Palila in area 3. domesticus),roof rats (Rattus rattus), or the small Finally, if diseaseproves to be responsible Indian mongoose(Herpestes auropunctatus) may for the absence of Palila on Pohakuloa flats, be excessivein the flats, although van Riper then strict measures are needed to control mos- October1984] PalilaPopulation and Habitat 663 quito breeding sites and exotic passerinepop- nities. Pp. 342-444 in Ecologyand evolution of ulations, particularly House Sparrows and communities (M. L. Cody and J. M. Diamond, Common Mynas. Eds.).Cambridge, Massachusetts, Belknap Press. DRAPER,N. R., & H. SMITH. 1981. Applied regression analysisßsecond ed. New York, J. Wiley. ACKNOWLEDGMENTS EDWARDS,D. K., G. L. DORSEY,& J. A. CRAWFORD. This study was made possiblethrough the help 1981. A comparisonof three avian censusmeth- and cooperationof numerousobservers who partic- ods. Pp. 170-176 in Estimating numbers of ter- ipated in data collection.We thank R. Bachman,C. restrial birds (C. J. Ralph and J. M. Scott,Eds.). Stud. Avian Biol. 6. Crivellone, S. Doyle, J. Engbring, P. Higashino, J. Jeffrey,E. Kosaka,P. H. McEldowney, M. Morin, M. ELLISßW. 1917. A narrative of a tour through Ha- Nakahara,R. Oshiro, E. Pung, H. Sakai,J. Sincock,P. waii, etc. (reprint of 1827 edition). Honolulu, Hawaii, Hawaiian Gazette. Stine, C. Stone, T. Sutterfield, P. W. Sykes, Jr., R. Walker, F. Warshauer,and J. Williams for their par- EMLEN,J. t., & M. J. DEJONG.1981. The application ticipation. Logistic assistancewas also provided by of song detection threshold distance to census the U.S. Fish and Wildlife Service, Hawaii Depart- operations. Pp. 346-352 in Estimating numbers ment of Land and Natural Resources, and the U.S. of terrestrial birds (C. J. Ralph and J. M. Scott, Eds.). Stud. Avian Biol. 6. Army at Pohakuloa Training Area. We thank J. Engbring, F. L. Ramsey,P. W. Sykes, Jr., J. Verner, GIFFIN,J. G. 1976. Ecologyof the feral sheep on and R. L. Walker for useful comments on the manu- Mauna Kea. Pittman-Robertson Proj. W-15-5, script.We are alsoindebted to W. Lantz for prepar- Study XI. Honolulu, Hawaii, Hawaii Dept. Land Nat. Resources. ing earlier drafts and the final manuscriptand to S. Doyle, R. Walker, and J. Williams for drafting illus- 1982. Ecology of the mouflon sheep on trations. Mauna Kea. Pittman-Robertson Proj. W-17-R, Study R-III. Honolulu, Hawaiiß Hawaii Dept. Land Nat. Resources. LITERATURE CITED HARTT,C. E., & M. C. NELL. 1940. The plant ecology ANDERSONßB. W., & R. D. OHMART. 1981. Compar- of Mauna Kea, Hawaii. Ecology21: 237-266. isonsof avian censusresults using variable dis- JAMES,F. C., & N. O. WAMER. 1982. Relationships tance transect and variable circular plot tech- between temperateforest bird communitiesand niques. Pp. 186-192 in Estimating numbers of vegetationstructure. Ecology 63: 159-171. terrestrialbirds (C. J. Ralphand J.M. Scott,Eds.). JVDD,C.S. 1936. Hawaii's forestswinning their bat- Stud. Avian Biol. 6. tle with wild animals. Honolulu Star-Bulletin, BANKO,W.E. 1979. Avian history report 2. History March 21, 3rd Sec.,pp. 1, 7. of endemicHawaiian bird specimensin museum KEPLER,C. B., & J. M. SCOTT.1981. Reducing count collections.Honolulu, Hawaii, Coop. Natl. Park variability by training observers.Pp. 366-371 in Res. Stud. Unitß Univ. Hawaii. Estimating numbers of terrestrial birds (C. J. BERGER,A.J. 1970. The eggsand young of the Palila, Ralph and J. M. Scott,Eds.). Stud. Avian Biol. 6. an endangeredspecies. Condor 72: 238-240. KIRCH,P. V. 1982. The impact of the prehistoric 1981. Hawaiian birdlife, second ed. Hono- Polynesianson the Hawaiian ecosystem.Pacific Sci. 36: 1-14. lulu, Hawaiiß Univ. Press Hawaii. ß E. KOSAKA, E. KRIDLER, J. M. SCOTT, P. MAYFIELD, H. F. 1972. Third decennial census of SCOWCROFT,C. WAKIDA, D. WOODSIDE, & C. VAN Kirtland's Warbler. Auk 89: 263-268. RIPERßIII. 1977. Palila recoveryplan. Portland, MUNROß G.C. 1944. Birds of Hawaii. Honolulu, Ha- OregonßU.S. Fish Wildl. Serv. waii, Tongg Press. BISHOP,C. R. 1852. Rept. Trans. Royal Hawaiian OLSON, S. L., & H. F. J^MES. 1982. Prodromus of the Agric. Soc. 1 (3): 91. fossil avifauna of the Hawaiian Islands. Smith- BRYAN,L. W. 1937. Wild sheepin Hawaii. Paradise sonJan Contrib. Zool. 365. Pacific 49 (3): 19, 31. PERKINS,R. C.L. 1893. Notes on collectingin Kona, BYELICH, J., W. IRVINE, N. JOHNSON, W. JAMES, H. Hawaii. Ibis Ser. 6, 5: 101-114. MAYFIELD, R. RADTKE, & W. SHAKEß 1976. Kirt- ß 1903. Vertebrata. Pp. 365-466 in Fauna Ha- land'sWarbler recoveryplan. Washington,D.C., waiiensis,vol. 1, part 4 (D. Sharp, Ed.). Cam- U.S. Fish Wildl. Serv. bridge, England, Univ. Press. DESANTE, D. F. 1981. A field test of the variable RAITT,R. J., & S. L. PIMM. 1976. Dynamics of bird circular-plotcensusing technique in a California communities in the Chihuahuan Desert, New coastalscrub breeding community.Pp. 177-185 Mexico. Condor 78: 427-442. in Estimating numbers of terrestrial birds (C. J. RAMSEY,F. L., & J. M. SCOTTß1981. Analysisof bird Ralph and J. M. Scott,Eds.). Stud. Avian Biol. 6. survey data using a modification of Emlen's DIAMOND,J. M. 1975. Assemblyof speciescommu- method. Pp. 483-487 in Estimatingnumbers of 664 SCOTTET AL. [Auk,Vol. 101

terrestrial birds (C. J. Ralph and J. M. Scott,Eds.). North American steppe vegetation: a multivari- Stud. Avian Biol. 6. ate analysis.Ecology 61: 1228-1250. REYNOLDS,R. T., J. M. SCOTT,& R. A. NUSSBAUM.1980. --, & . 1980b. Temporal variation in hab- A variable circular-plotmethod for estimating itat structure and shrubsteppe bird dynamics. bird numbers. Condor 82: 309-313. Oecologia 47: 1-9. RICE, J., R. D. OHMART, & B. W. ANDERSON. 1981. ROTHSCHILD,W. 1900. The avifauna of Laysan and Bird community use of riparian habitats:the im- the neighbouring islands, vol. 3. London, R. L. portance of temporal scale in interpreting dis- Porter. criminant analysis. Pp. 186-194 in The use of SCOTT,J. M., & F. L. RAMSEY.1981. Length of count multivariate statistics in studies of wildlife hab- period as a possiblesource of bias in estimating itat (D. E. Capen, Ed.). Fort Collins, Colorado, bird numbers. Pp. 409-413 in Estimating num- U.S. For. Serv., Gen. Tech. Rept. RM-87. bers of terrestrial birds (C. J. Ralph and J. M. RICHARDS,L. P., & P. H. BALDWIN. 1953. Recent rec- Scott, (Eds.). Stud. Avian Biol. 6. ords of some Hawaiian honeycreepers.Condor --, S. MOUNTAINSPRING,F. L. RAMSEY,& C. B. 55: 221-222. KEPLER.In press. Forest bird communities of VANRIPER, III, C. 1978. The breeding ecologyof the the Hawaiian Islands: their dynamics, ecology, Amakihi (Loxopsvirens) and Palila (P$ittirostra and conservation. Stud. Avian Biol. bailleui)on Mauna Kea, Hawaii. Unpublished SCOWCROFT,P. G., & J. G. GIFFIN. 1983. Feral her- Ph.D. dissertation. Honolulu, Hawaii, Univ. Ha- bivores suppress mamane and other browse waii. specieson Mauna Kea, Hawaii. J. Range Mgmt. ß 1980a. Observationson the breeding of the 36: 638-645. Palila P$ittirostra bailleui of Hawaii. Ibis 122: 462- STEEL,R. G. D., & J. H. TORRIE. 1980ß Principles and 475. proceduresof statistics:a biometrical approach, ß 1980b. The phenology of the dryland forest second ed. New York, McGraw-Hill. of Mauna Kea, Hawaii, and the impact of recent TOMICH,P.Q. 1969. Mammals in Hawaii: a synopsis environmental perturbations. Biotropica 1034 12: and notational bibliography. Spec.Pub. 57. Ho- 282-291. nolulu, Hawaii, BerniceP. BishopMus. , J. M. SCOTT,& D. M. WOODSIDE. 1978. Dis- UNITED STATES FISH AND WILDLIFE SERVICEß 1983. Re- tribution and abundance patterns of the Palila publication of the lists of endangered and on Mauna Kea, Hawaii. Auk 95: 518-527. threatened species.Fed. Reg. 48: 34182-34196. --, S. G. VAN RIPER, M. L. GOFF, & M. LAIRDß WARNER, R. E. 1960. A forest dies on Mauna Kea. 1982. The impact of malaria on birds in Hawaii PacificDiscovery 13: 6-14. VolcanoesNational Park. Tech. Rept. 47. Hono- --. 1968. The role of introduced diseases in the lulu, Hawaii, Coop. Natl. Park Res. Stud. Unit, extinction of the endemic Hawaiian avifauna, Univ. Hawaii. Condor 65: 3-23. ROCK,J.F. 1913. The indigenoustrees of the Hawai- WEATHERS,W. W., & C. VAN RIPER, III. 1982. Tem- ian Islands.Honolulu, Hawaii, privately printed. perature regulation in two endangered Hawai- ROTENBERRY,J. T., & J. A. WIENS. 1980a. Habitat ian honeycreepers:the Palila (Psittirostrabailleui) structure,patchiness, and avian communitiesin and the Laysan Finch (P$ittirostracantans). Auk 99: 667-674.