Pacific Science (1983), vol. 37, no. 2 © 1983 by the University of Hawaii Press. All rights reserved Tree Cover Changes in Miimane (Sophora chrysophylla) Forests Grazed by Sheep and Cattle! PAUL G. SCOWCROFT 2 ABSTRACT: Using aerial photographs taken in 1954, 1965, and 1975, per centage oftree cover was determined for three sections of the sheep- and cattle grazed mdmane (Sophora chrysophy lla) forest of Mauna Kea , Hawaii. In one section , the Ka'ohe Game Management Area, where grazing by sheep was judged light, tree cover increased slightly during the 21-yr period , and tree cover did not change significantly along an elevation gradient. This condition was probably the result of the predominance of naio (Myoporum sandwicense) trees, which are not as palatable as mdmane and, therefore, are less sensitive to browsing. In the Mauna Kea Forest Reserve, the most heavily sheep browsed of the three sections, a significant loss oftree cover was observed between 1965and 1975near tree line where feral sheep tended to concentrate their browsing. Ofthe three sections examined, Parker Ranch, which was grazed mainly by cattle, sustained the greatest loss of tree cover during the 21-yr period, reflecting the more destructive nature of cattle browsing as compared to sheep browsing. Increases of tree cover in areas relatively free of sheep within the Mauna Kea Forest Reserve indicated that habitat for the palila, an endangered bird that depends on the mdmane forest, willimprove slowly after feral sheep are removed. AT MAUNA KEA ON THE ISLAND of Hawaii the mdmane (Sophora chrysophy lla) forest has been adversely affected by feral sheep (Ovis aries) and other herbivores for more than 150yr (Kramer 1971). Damage from sheep was so severe that an effort to eradicate them, begun in 1937 (Bryan 1937), was continued through 1949. As a result introduced and native herbaceous and woody vegetation be came established and proliferated in many areas (Tomich 1969). At the same time, how ever, feral sheep became valued as a game species; consequently, the policy of eradica tion was replaced by sustained-yield game management in 1959 (Walker 1978). But in areas where the remaining sheep concen trated, especially near tree line, mdmane and other vegetation did not regenerate (Warner 1960), a condition prevalent even as recently 1 Manuscript accepted 30 December 1982. 2 Institute of Pacific Islands Forestry, Pacific South west Forest and Range Experiment Station , Forest Ser vice, U.S. Department of Agriculture , 1151 Punchbowl Street, Honolulu, Hawaii 96813. as 1980 (Scowcroft and Giffin, in press). Forest areas used by cattle (Bos taurus) for pasture also showed little evidence of success ful mdmane regeneration. The continued decimation of the mdmane overstory-from the death of old trees and from the failure to regenerate-was a signifi cant threat to the palila iPsittlrostra bailleui) , an endangered endemic Hawaiian bird re stricted to the mdmane forest of Mauna Kea (Berger 1972). Palila depend on mdmane and, to a lesser extent, on naio trees (Myoporum sandwicense) for food and nest sites (van Riper 1980a). Their breeding season coincides with the time ofyear when mdmane bear seed pods (van Riper 1980a). Palila were found most abundant and in highest densities near tree line (van Riper, Scott, and Woodside 1978). The birds probably were attracted to this area because it contained a high proportion of large mdmane trees capable of producing more food than smaller ones. If large trees are as necessary to the palila as these findings suggest, then death ofsuch trees could further restrict the bird's acceptable habitat (van 109 110 Riper 1980a), especially if the forest failed to regenerate. Because quantitative data were lacking, hunting advocates denied that tree cover had been lost. Some ofthose who acknowledged a loss suggested natural causes other than feral sheep. A study has shown that miimane re generation is suppressed by feral sheep in the vicinity of tree line (Scowcroft and Giffin, in press). Because of this suppression and the abundance of dead trees in upper portions of the forest , I hypothesized that there should be a net loss of tree cover in the upper fringe of the forest. This paper reports a study to determine changes in tree canopy cover from 1954 to 1975 within the Mauna Kea Forest Reserve (Mauna Kea FR), the Ka 'ohe Game Manage ment Area (Ka'ohe GMA), and adjacent Parker Ranch. STUDY AREA Site Description The study area, totaling about 4280 ha, was located between 1750 m and 2900 m on the west flank ofMauna Kea (Figure 1). This area was chosen because aerial photos of it were available for 1954, 1965, and 1975. It was also an area where palila were found in high den sities (van Riper, Scott, and Woodside 1978). Only forested land was included. Portions of the forest that were strip-bulldozed to en hance game-bird habitat were excluded. Three land divisions are represented in the study area-the Mauna Kea FR (2360 ha), Ka 'ohe GMA (890 ha) , both owned and managed by the State of Hawaii, and Parker Ranch (1030 ha), which is privately owned. Cattle used the ranch land during winter months; feral sheep used the other areas year round. Formost ofthe period, sheep browsing was lighter (fewer sheep per hectare ofutilized forest) in the Ka'ohe GMA than in the Mauna Kea FR (R. Walker, pers. comm.). The forest overstory in the study area was chiefly mdmane and naio with small stands ofEuphorbia olowaluana, na 'ena'e (Railliardia arborea), and coast sandalwood (Santalum PACIFIC SCIENCE, Volume 37, April 1983 ellipticum) scattered through portions at lower elevations. The relative abundance, density, and size of miimane and naio along the northern edge of the Ka'ohe GMA, as described by van Riper (1980b) , apply to the entire GMA study area. On pasture lands the forest is pure, open mdmane except near the boundary to the Ka 'ohe GMA where naio grows together with miimane. The forest ex tends from about 2870 m down to about 1750 m elevation. The forest to the east of Pu'u La'au, and at upper elevations in the Mauna Kea FR, is also open woodland of pure miimane. Elsewhere in the Mauna Kea FR, naio grows mixed with miimane, with the proportion ofnaio decreasing with increasing elevation. Forest density decreases as eleva tion increases. Climate in the study area is cool and dry (van Riper 1980b). Annual rainfall averages more than 500 mm , most of it falling between November and April. Temperature maxi mums rarely exceed 21°C, and the diurnal fluctuation is about 20°e. Fog frequently envelops the area by midafternoon. Soils in the area are very fine sandy loams and loamy fine sands formed from volcanic ash , sand, and cinders (U.S. Department of Agriculture 1973). They are well drained to excessively drained, with rapid permeability (up to 50 cm/hr) and with little runoff or erosion. Animal Populations The only estimates for the size of sheep populations in the Mauna Kea portion of the study area are those made by the Hawaii Division of Fish and Game." Sheep counts were obtained by aerial and ground surveys conducted by wildlife biologists. The counts were adjusted for duplicate sightings and for estimated number of animals missed due to 3 Job Progress Reports on Statewide Game Research and Surveys compiled by the Hawaii Division ofFish and Game and on file with the Hawaii Division of Forestry and Wildlife (HDFW), 1151 Punchbowl St., Honolulu 96813. Reports are available for most years since 1953. Authors , job titles, and project numbers vary. Tree Cover Chan ges in Mdmane Forests-SCOWCROFT I I I o o 2 (km) (mi) Elevation (m) Grazed by cattle: r">;] Parker Ranch lSS 'W FIGURE I. About 55 percent of the 4280-ha study area lay within the Maun a Kea Fo rest Reserve (FR), 21 percent within the Ka'ohe Ga me Management Area (GMA), and 24 percent within the Parker Ranch. Each is separated from the others by fences which converge near Pu'u La'au, 112 dense tree cover and fog. Estimate s were re ported separately for 6 to 10 topographically defined sections of the forest reserve. Thus , population estimates were available for the area between Pohakuloa and Kemole gulches where my study area lay. Sheep density was determined for the years 1954 thro ugh 1975 by dividing the estimated sheep population by the size of the forested area between Pohakuloa and Kemole (4700 ha). I assumed thatdensity within my 2360-ha study area was equal to the calculated density for the whole area between P6hakuloa and Kemole. Results showed that the feral sheep popu lation varied from an estimated low of 0.10 animals/ha in 1954 to a high of 0.39 animals/ ha of forest in 1968 (Table 1). In general, sheep densities were smaller during the first half ofthe period (1954-1965) than during the second half (1965-1975). A greater propor tion of the total population of feral sheep in the Mauna Kea FR was found between Poha kuloa and Kemo le gulches after 1965 than occurred before that date. The sheep were welldistributed throughout the area before 1965 (R. Walker, pers. comm.). Even as late as the early 1960sit was common to find sheep grazing near the lower fence boundary at 2130m elevation . This was the result of the maintenance of a relatively large number of feral sheep and of the game man agers' deliberate husbandry of them to obtain a uniform spatial distribution. About 1965, biologists realized that the large number of sheep in the Mauna Kea FR, although well distributed, were causing un acceptable damage to the vegetation. To re duce the population, the access road system was expanded and hunters were allowed to take both sexes instead of rams only. The increased hunting reduced the total sheep population in the Mauna Kea FR. In my study area, however, sheep populations were gen erally larger after hunting pressure increased. Increased hunting pressure also caused the survivors to keep to higher elevations, gen erally above 2400 m. Before 1965, therefore , browsing pressure was evenly distributed throughout the forest, but after 1965 brow sing pressure occurred mostly above 2400m. In the Ka'ohe GMA, census data on feral PACIFIC SCIENCE, Volume 37, April 1983 TABLE 1 ESTIMATED D ENSITY OF F ERAL SHEEP IN THE STUDY AREA* AND TH E E STIMATED SIZE OF THE SHEEP POPU LATION FOR THE ENTIRE FOREST RES ERVE, 1954 THROUGH 1975 DATE OF SHEEP DENSITY CENSUS IN STUDY ESTIMATED POPULATION (mo/yr) AREA (#/ha)! IN MAUNA KEA FRt 6/54 0.10 1560 3/55 0.11 2200 - /56 8/57 0.18 3200 10/58 0.19 3000 5/59 0.12 3500 5/60 0.20 4000 5/61 0.17 3500 3/62 0.13 2300 7/63 0.16 1250 7/64 0.17 1400 5/65 1840 5/66 0.34 2700 - /67 7/68 0.39 3250 7/69 0.35 2900 7/70 0.26 2500 7/71 0.19 2385 -/72 2/73 0.20 2125 5/74 0.13 1200 1/75 0.20 1735 • Within the Mauna Kea FR. tDensity was calculated by dividing the estimated number of sheep in the area between Pohakul oa and Kemole gulches by the size of the forested area. This figure was assumed to represent animal density in the study area. t From Giffin 1976. sheep existed only for 1961 (Walker 1961)and 1973 (1. Giffin, pers. comm.). These data and the recollections of the wildlife biologists on the Island of Hawaii during the 2l-yr period (R. Walker, E. Kosaka, and J. Giffin, pers. comm.) indicated that the resident feral sheep population in Ka 'ohe was less than one animal on 10ha (0.09/ha) of forest through the 1960s,and possibly as high as 1.4 animals on 10ha (0.14/ha) in the first half of the 1970s. Large concentrations of sheep were reported for 1965 in the upper, eastern quarter of Ka'ohe GMA where hunting pressure was low because of poor access (Bachman 1966). The sheep were more uniformly distributed thro ughout the area in the early 1970s (J. Giffin, pers. comm.). Additiona l browsing Tree Cover Changes in Miimane Forests-SCOWCROFT 113 pressure may have been exerted on less ac cessible areas when sheep from the Mauna Kea FR temporarily fled through broken fences into Ka 'ohe GMA to escape hunters (R. Walker, pers. comm.). In Parker Ranch, no estimate of cattle numbers was available. Use by cattle was sea sonal, however. An unknown number of feral sheep lived on the ranch land year-round. The sheep population there was probably larger before 1965 than after (Walker 1961). MATERIALS AND METHODS I estimated percentage of tree cover from aerial photographs taken in 1954, 1965, and 1975. These estimates allowed me to evaluate change in tree cover during a 21-yr period (1954-1975) and to look at the changes during the first (1954-1965) and second (1965-1975) halves of that period. The 1954 and 1965 photos were black and white and the 1975ones were color. The nom inal photo scales were 1 : 40,000 for 1954, 1 : 24,000 for 1965, and 1 : 12,000 for 1975. Image resolution was poor for the 1954 photos but good for the other two sets. The sample consisted of 870 circular sam ple plots: 480 plots lay within the Mauna Kea FR; 181 within the Ka 'ohe GMA; and 209 within Parker Ranch. Plot centers were first established on a 1 : 24,000 U.S. Geological Survey quadrangle map with a uniform grid. Each center was transferred to the 1975 photos by matching the topographic features of the map and photos. The same plot centers were then carefully located on the 1954 and 1965 photos with the aid of the marked 1975 photos. Each plot circumscribed an area on the photo equivalent to 0.2 ha on the ground. Because of differences in photo scale, it was necessary to vary the diameter of the photo plots to assure that the same size ground area (0.2 ha) was examined in every case. This involved first determining the photo scales across each photo, and then using an acetate overlay inscribed with the appropriate size circle. The 870 plots represented a 4 percent sample of the study area. Different scale grids of intersecting lines inscribed on a second clear acetate overlay were used to estimate cover. About 120inter sections fit within the smallest circle at the scale of 1 : 43,000, and more than 300 inter sections fit within the largest at the scale of 1 : 11,000. The proportion of points that overlaid the image of tree canopies was deter mined under 8 x to 20 x magnification. This value, to the nearest whole percent, was recorded for each sample plot along with elevation and land use. The change in percent tree cover for each of the 870 plots was calculated for each period by subtracting the percent cover at the start of the period from that at the end of the peri od. This difference was used in most of the analyses. Throughout the rest of the paper, I use the phrase average change in percent tree cover to mean the difference between cover at the start and end of a period averaged for all plots in the land-use category or elevation stratum under consideration. Changes in percentage of tree cover be tween land uses were compared for each ofthe three periods by regressing cover at the end of the period on cover at the beginning (linear model) and then calculating ajoint 95 percent confidence region for the slope and intercept (Draper and Smith 1966). This procedure allowed me to take into account differences in percent tree cover between land uses at the beginning of each period. Nonoverlapping confidence regions indicated that the change in cover was significantly different between the land uses being compared. For each land use, I used the Bonferroni t test to compare the average change in tree cover between 150-m elevation strata and be tween periods (Miller 1966).The Bonferroni t test is a multiple pairwise comparison test applicable to data where the sample sizes of factor levels are equal or unequal. RESULTS Mauna Kea Forest Reserve In 1954 the average percent tree cover for the sample area within the Mauna Kea FR, heavily grazed by sheep, was 27.3 percent. By 1965, average tree cover had declined slightly to 26.7 percent. A further decline to 25.6 percent occurred between 1965and 1975. 114 These means were not significantly different (Bonferroni t test, P~0.05). The loss of cover during the 21yr represented about one twentieth of the tree cover present in 1954. I hypothesized that tree cover losses should be greater at higher elevations in the Mauna Kea FR where damage to vegetation, and , by inference , browsing pressure, had been re ported as persistently high (Giffin 1976). Thi s proved to be true when data were analyzed by 150-m elevation strata (Figure 2A). For the entire period, the two upper-elevation strata lost cover but the three lower strata gained cover. Only the loss in the uppermost stratum (8.5 percent), however , was statistically dif ferent (Bonferroni t test, P < 0.01) from the changes at lower elevations. When the first half of the period was con sidered separately, average change in cover per plot was not statistically different among elevations (Figure 2A). During the second half, however, the changes in cover abo ve 2590 m elevation were significantly different (Bonferroni t test, P < 0.0 I) from those below 2590 m. During the second half of the period, the net change in cover per plot for the five elevations followed the pattern I expected, with greatest gains at the lowest elevation and greatest losses at the highest elevation. The mean change in tree cover during the first half of the period was not significantly different (Bonferroni t test, P~0.05) from that during the second half (see Figure 2A , dashed lines). Because errors in cover estimation may have resulted from poor image quality in the 1954 photos, I also compared changes in tree cover for only those plots that lost or gained at least 20 percent cover. When the entire 21-yr period was considered, 86 of the 480 plots in the Mauna Kea FR fell into this category: more than two-thirds of these lost cover , Analysis of the 86-plot subsample revealed that loss oftree cover increased at successively higher elevations, the same pattern I found when all plots were included in the analysis . Ka'ohe Game Management Area Tree cover in the Ka'ohe GMA averaged 54.5 percent in 1954,52.3 percent in 1965, and PACIFIC SCIENCE, Volume 37, April 1983 56.4 percent in 1975 (Table 2). Differences between these means were not significant (Bonferroni t test, P~0.05). Losses incurred during the first 11yr were compensated for during the last 10yr. The tendency for progressively greater losses oftree cover at higher elevations, which I observed in the Mauna Kea FR and which I attributed to increasing browsing pressure, was not evident in this dense portion of the forest (Figure 2B). If anything, the reverse was true : Cover losses became progressively great er at lower elevations. None ofthe differences between elevation strata were significant, however (Bonferroni t test, P~0.05). From 1954 to 1965 tree cover in Ka'ohe declined an average of2.2 percent (Figure 2B, dashed line). From 1965 to 1975 cover in creased an average of 4.1 percent. These mean changes were significantly different (Bon ferroni t test, P < 0.01). Compared with the heavily browsed Mauna Kea FR, only 9 percent of the 181 plots in the lightly browsed Ka'ohe GMA changed cover more than 20 percent during the 21-yr period . More than one-half ofsuch plots gained cover while the rest lost cover. Analysis of the 16 plot subsample revealed that average change in tree cover was not significantly different between elevation strata. Parker Ranch The remnant mdmane forest on Parker Ranch, grazed by cattle, lost one-fifth of its tree cover during the 21-yr period (Table 2). This loss was the greatest of those measured for the three land divisions. Almost all of this cover was lost before 1965 when the average percent tree cover fell from 18.9 percent in 1954 to 15.1 percent in 1965. The difference between cover in 1954 and 1965 was signifi cant (Bonferroni t test, P < 0.05). Average tree cover in 1975 was 15.0 percent. During the first halfofthe period, net losses were observed in all strata below 2290 m eleva tion (Figure 2C). But during the second half, net losses were noted only in the two strata between 1831 to 2130 m. The highest stratum (2291 to 2440 m) gained cover in both the first and second halves of the period. Some of the A MAUNA KEA FOREST RESERVE 10 1954-1965 1965-1975 1954-1975 5 o -5 I a I a Ia - 10 I 2210 I 2820 I 2210 I 2820 I 2210 I 2820 B KAOHE GAME MANAGEMENT AREA 1954-1965 T n-1 - - - --1a -- -re i --Ia··-------. 1965-1975 I 2515 1954-1975 I 2055 Ia --- ---J:a---- -- --- 1 a n-1 I 2515 I 2055 --- -·-··1~--Ia----- Ia I 2515 I 2055 o 5 -5 10 -10 t= z w (J a: w ~ a: w > o (J w w a: ~ z w CJ z -e :E: (J Ic 5 o -5 -10 C PARKER RANCH 1954-1965 IC Ib --·-r;~-----Iab----- ---- Ia 1965-1975 lab Ib ~~~-----'lrab-Ib~- Ia 1954-1975 -15 I 1755 I 2365 I 1755 I 2365 I 1755 I 2365 ELEVATION STRATUM MIDPOINT em) FIGUR E2. Average change in percent tree cover per plot ( ± 5i ) within the Mauna Kea FR (A), the Ka'ohe GMA (B), and the Parker Ranch (C), by 150-m-wide elevatio n strata and period. Elevation means for a given period and land division followed by a common letter are not significant ly different (Bonferro ni t test, P~0.05). The dott ed lines indicate the average change per plot for all strata combined for the given period. The uppermost elevation stratum in the Ka'ohe GMA had only one sample plot. 11 6 PACIFI C SCIENCE, Volume 37, April 1983 TAB LE 2 A VERAGE P ERCENT T REE COVER P ER P LOT FOR THREE G RAZED PORTIONS OF THE MAMANE F OREST OF M AUNA K EA IN 1954, 1965 AND 1975 ELEVATION PERCENT TREE COVER PER PLOT'" STRATUM SAMPLE (m) SIZE 1954 1965 1975 Mau n a K ea F R 2131-2290 6 24.2 (4.0)t 21.7 (4.2) 25.2 (3.7) 2291-2440 121 33.0 (2.0) 31.3 (1.8) 33.9 (1.8) 2441-2590 148 29.3 (1.9) 29.3 (1.8) 30.0 (1.8) 2591- 2740 III 24.9 (2.1) 25.3 (2.3) 22.4 (2.2) 2740 94 20.1 (2.2) a 18.7 (2.1) a 11.6 (1.6) b A ll 480 27.3 (1.0) 26.7 (1.0) 25.6 (1.0) K a 'ohe GMA 1981-2130 33 56.6 (3.8) 53.2 (3.8) 55.4 (3.8) 2131-2290 93 52.3 (1.8) 49.5 (1.8) 54.0 (1.9) 2291-2440 54 56.5 (3.3) 55.9 (3.2) 60.5 (3.0) 2441-2590 I 89.0 (-) 89.0 (-) 86.0 (-) A ll 181 54.5 (0.9) 52.3 (0.6) 56.4 (0.7) P a r ker R a n ch 1831 10 18.1 (3.6) 12.3 (1.8) 15.4 (3.5) 1831-1980 42 23. 1 (2.6) a 15.4 (1.7) b 13.2 (1.6) b 1981- 2130 73 18.8 (1.4) 14.7 (1.1) 14.4 (1.4) 2131-2290 68 17.8 (1.1) 15.9 (0.8) 16.1 (0.8) 2291-2440 16 13.7 (1.4) 14.5 (1.3) 17.4(1.6) A ll 207 18.9 (1.5) a 15.1 (l.5)b 15.0 (1.5) b • Values within a row followed by the same letter do not differ significantly (Bonferron i proced ure for pairwise comparisons, p~0.05). No significant differences exist in rows where letters do not appear. t Values in parentheses are standa rd erro rs of the mean . average changes in tree cover for the elevation strata were significantly different (Figure 2C). Areawide changes in cover for the first and second halves of the study period are shown in Figure 2C (dashed lines). The mean change during the first half (x = - 3.8 percent, SE = 0.5) was significantly different (Bon ferroni t test, P < 0.01) from that during the second half (x = - O.I, SE = 0.4). Like the Ka 'ohe GMA, 9 percent of the 209 pasture plots underwent a change in cover of 20 percent or more durin g the 2I- yr period. Fifteen of the 16plots lost cover. No trend was apparent in net cover change with elevation. All Three Land Uses I also compared the average change in tree cover per plot among the three land uses. The statistical analysis involved first regressing cover at the end of the period on cover at the beginning and then calculating a joint 95 percent confidence region for the slope and intercept. Differences in the change of cover were significant if confidence regions did not overlap. Before 1965, all three land divisions suffered a net loss of cover, but the loss for the forest grazed by cattle was significantly greater (P < 0.05) than the losses in the other areas (Table 3). After 1965, net losses continued in the Mauna Kea FR and in Parke r Ranch, both heavily grazed, but the Ka 'ohe GMA, lightly grazed, gained cover. Considering the entire 2I-yr period, the greatest net loss occurred in cattle-grazed areas (x = - 3.9 percent/plot). A smaller loss was observed for areas heavily grazed by sheep (x = - 1.7 percent/plot). The lightly sheep-browsed forest actually gained cover (x = + 1.9 percent/plot). Each of these changes was significantly different from the others (P < 0.05). Tree Cover Changes in Mdmane Forests-SCOWCROFT T ABLE 3 AVE RAGE C HANGE IN PERCENT T REE COVER P ER PLOT FOR THE F IRST HALF (1954-1 965), SECOND HALF (1965- 1975), AND E NTIRE STUDY P ERIOD (1954-1 975) ON GRAZED PORTIONS OF THE M AMANE F OREST OF M AUNAKEA PERCENT CHANGE IN TREE COVER PER PLOT" 117 LAND USE Hea vil y grazed by sheep (Mauna Kea F R) Lightl y g ra ze d by sheep (Ka'ohe G MA) Grazed b y catt le (Par k e r Ranch) SAMPLE SIZE 480 181 207 FIRST HALF (1954-1 965) - 0.6 (0.6)t a - 2.2 (0.7) a - 3.8 (0.5) b SECOND HALF (1965- 1975) - J.I (0.4) a 4.1 (0.6) a - 0.1 (0.4) a ENTIR E PERIOD (1954- 1975) - 1.7 (0.7) a 1.9 (0.8) b - 3.9 (0.7) c • Values within a column followed by the same letter do not differ significantly (P ;;:: 0.05). See text for description of statistical test. t Values in parenth eses are standa rd errors of the mean . DISCUSSION I suspect that the small scale and poor re solution of 1954 photos introduced some error into the data. Small-crowned trees may have been missed, with the result that cover was underestimated. Thi s situation most likely was true for areas of low to moderate tree cover where plant competition was low enough to permit establishment of trees during the late 1940s and early 1950s when sheep numbers were low. Underestimates of initial cover would have resulted in the over estimation of gains and the underestimation of losses in tree cover over time. Such errors probably affected data for those parts of the Mauna Kea FR where initial tree cover was low to moderate and browsing pressure was nil for a time. I do not believe such errors affected the significance of differences (Figure 2A). This view was supported when only plots that changed by 20 percent or more were compared. In areas with dense tree cover in 1954, shadows may have been viewed as tree canopy, leading to an overestimate of cover. The effect on cover change over time would have been an overestimate of losses and an undere stimate of gains in cover. This type of error probably affected data of the Ka 'ohe GMA and those areas of the Mauna Kea FR where stand density was high. Such errors would not have affected the outcome of the statistical tests, however. For instance, in Ka'ohe cover was uniformly dense, in general , and the means for all elevations would have been equally affected. So differences among means would still have been nonsignificant (Figure 2B). Supporting this view was the lack ofany statistical differences between elevation strata when the analysis included only plot s that changed by 20 percent or more . The average changes in percent tree cover per plot were generally on the order of 1 to 4 percent in the Mauna Kea FR and can be attributed to photo interpretation error. But in the uppermost tree-line stratum, tree cover declined 8.5 percent from 20.1 percent in 1954 to 11.6 percent in 1975 (Table 2). If 5 percent of this change is attributed to interpretation error, then the actual loss was 3.5 percent in 21 years. Thi s was a large loss considering that tree cover was initially sparse . Sustained losses of this magnitude in future years could result in a lowering of tree line, such as that alluded to by Berger (1972). In the ranch statistical differences in the average change in percent tree cover were detected, but they too were generally on the order of 1 to 5 percent. The exception was the stratum between 1830 to 1980m where cover declined from 23 to 13 percent in 21 yr (Table 2). If 5 percent of this change is attributed to interpretation erro r, then actual loss was 5 percent in 21 yr. As at tree line in the Mauna Kea FR, such a loss was large relative to the amount of cover present initially. Evidence from other studies (Scowcroft and Giffin, in press, Scowcroft and Sakai 1983) implicate feral sheep as a factor that " .., 118 could explain the reduction of tree cover near tree line in the Mauna Kea FR. These studies show that sheep suppressed mdmane regener ation in areas they used frequently. And be cause they used the upper fringe of the forest most (Giffin 1976), replacement of dying over mature mdmane could not occur. The pre dictable outcome was an increase of mdmane tree cover in lower areas and a decrease in upper areas. While the main effect of sheep browsing is suppression of mdmane regeneration, sheep can reduce tree cover directly by consuming foliage and twigs of established mdmane trees. Such losses are probably small because the browse height of sheep is less than 2 m. Similarly, sheep probably do not reduce tree cover by directly causing whole tree mor tality. Indeed, mortality would occur even without browsing pressure. But in a stable forest ecosystem, one would expect replace ment of dead trees and reestablishment of lost leaf area. One would not expect significant losses of tree cover in the absence of browsing pressure and catastrophic events. Thus, brow sing damage and decline in tree cover in the Mauna Kea FR are linked by the fate of mdmane reproduction. In the Ka'ohe GMA, the lack of change in tree cover during the 21-yr period beyond what could be attributed to interpretation error was not surprising. The area already supported dense stands of naio and mdmane in 1954, making large increases in cover un likely. Large decreases in cover were also un likely, if we assume that the principal factors causing a decrease were sheep browsing and old mdmane tree mortality. The sheep popu lation was relatively small and spatially well distributed during most of the 21-yr period (Walker, pers. comm.). And more important ly, naio, which is less palatable than mdmane and comprises but a small fraction of the diet of sheep (Giffin 1976),dominates the forest in this area. The regeneration of naio in openings and the expansion of crowns of larger trees, therefore, would be uninhibited by feral sheep. Furthermore, preferential browsing of any mdmane regeneration would give naio a competitive advantage (van Riper 1980b). PACIFIC SCIENCE, Volume 37, April 1983 Sheep browsing in the Ka'ohe GMA, there fore, even if heavy, would probably not be reflected in large losses of tree cover. The small changes of tree cover in the cattle-grazed area was surprising. Cattle are known to destroy vegetation by trampling and snapping twigs and branches as they maneuver in the vegetation , as well as by eating large quantities of plant material. Even established trees can be affected adversely. The relative youthfulness of a large portion of the mdmane trees in the ranch and the wide spacing of individuals may account for the small changes in cover-that is, even though regeneration may be suppressed, few estab lished trees would die. This study showed that the mdmane com ponent of the palila's habitat (perhaps the most critical component) has been decreasing in the tree-line area of the Mauna Kea FR (Figure 2A, Table 2). Because the palila de pends on the mdmane forest of Mauna Kea , the continued but gradual loss of mdmane tree cover can only affect this endangered species adversely. The remnant mdmane forest in the ranch apparently is not used by the palila , possibly because the surviving large trees are too widely scattered to support the bird. The palila apparently can not reoccupy the ranch as long as the food producing capacity of the mdmane forest remains low. And although tree cover increased slightly in the Ka 'ohe GMA, the increase probably resulted from regeneration and growth of naio, a species of much less value to palila. In 1979,the State of Hawaii was ordered by a federal judge to eliminate feral sheep and feral goats from the palila's critical habitat to comply with the Endangered Species Act of 1973.The Mauna Kea FR and Ka 'ohe GMA are part of the critical habitat. That effort has been completed. I expect mdmane to recover quickly in the higher elevations. Perhaps the losses sustained in the past will be erased and the forest secured for the palila . But mdmane grows slowly in the Mauna Kea FR, and , because large trees are necessary to palila, many years will pass before degraded areas again reach their full potential for supporting the bird. If i; S Tree Cover Changes in Miimane Forests-SCOWCROFT 119 LITERATURE CITED BACHMAN, R. E. 1966. Mauna Kea feral sheep survey, job completion report. Pittman Robertson Proj. No. W-5-R-16, Job No . 46(16), July 1964-June 1965 Unpubl, rept. Hawaii Dept. Land and Nat. Res., Div. of Fish and Game, Honolulu. BERGER, A. J. 1972. Hawaiian birdlife. Univ. Press of Hawaii, Honolulu, 270 pp . BRYAN, L. W. 1937. Wild sheep in Hawaii. Pac . Para. 49 : 19, 31. DRAPER, N . R. , and H. SMITH. 1966. Applied regression analysis. John Wiley and Sons, New York. 407 pp. GIFFIN, J. G. 1976. Ecology ofthe feral sheep on Mauna Kea, final report. Pittman RobertsonProj. No . W-15-5, Study No. IX, 1972-1975. Unpub!. rept. Hawaii Dept. Land and Nat. Res., Div . of Fish and Game, Honolulu. KRAMER, R. J. 1971. Hawaiian land mam mals. Charles E. Tuttle Co ., Tokyo, Japan. 347 pp. MILLER, R. G. 1966. Simultaneous statistical inference. McGraw-Hill Book Co., New York. 272 pp. SCOWCROFT, P. G., and J. G. GIFFIN. In press. Feral herbivores suppress mamane and other browse species on Mauna Kea, Hawaii. J. Range Mgt. SCOWCROFT, P. G., and H. F. SAKAI. 1983. Impact of feral herbivores on mamane for ests ofMauna Kea, Hawaii: Bark stripping and diameter class structure. J. Range Mgt. 36(4):495-498. TOMICH, P. Q. 1969. Mammals in Hawaii: A synopsis and notational bibliography. Bernice P. Bishop Museum Spec. Pub. 57, Bishop Museum Press, Honolulu. 238 pp . U.S. DEPARTMENT OF AGRICULTURE, SOIL CONSERVATION SERVICE. 1973. Soil survey of island of Hawaii, State of Hawaii. Soil Cons . Ser. in cooperation with the Univ. of Hawaii Agr. Exp. Sta . U.S. Govt. Printing Office, Washington, D.C. 115 pp . VAN RIPER, c., III. 1980a. Observations on the breeding ofthe Palila Psittirostra bailleui of Hawaii. Ibis 122:462 -474. - - . 1980b. The phenology ofthe dryland forest of Mauna Kea, Hawaii, and the impact of recent environmental perturba tions . Biotropica 12:282-291. VAN RIPER, c., III, J. M. SCOTT, and D. M. WOODSIDE. 1978. Patterns in the distribu tion and abundance ofthe Palila on Mauna Kea, Hawaii. Auk 95: 518-527. WALKER, R. L. 1961 . A brief report on the present status of game populations in the horse pasture game management area and recommendations for management and hunting. Unpubl. rept. Hawaii Dept. Agr . and Conserv., Div. of Fish and Game, Honolulu. 2 pp. - - -. 1978. A history of the Division of Fish and Game. Paper presented at the 20th Annual Hawaii Forestry Conference: Habitats, Energy, and Other Good Things, Kona, Hawaii, September 1978. WARNER, R. E. 1960. A forest dies on Mauna Kea. Pac. Dis. 13:6-14. ACKNOWLEDGMENTS The author thanks Carl Masaki for inter preting the aerial photos; C. W. Smith, R. A. Merriam, R. L. Walker, J. A. Baldwin, C. J. Ralph, C. van Riper, and J. G. Giffin for their critical reviews ofearlier drafts of this manu script; and J. A. Baldwin for his invaluable statistical guidance.