The Effects of Exotic Grasses on Litter Decomposition in a Hawaiian Woodland: The Importance of Indirect Effects

Transcription

1 Ecosystems (2003) 6: DOI: /s y ECOSYSTEMS 2003 Springer-Verlag The Effects of Exotic Grasses on Litter Decomposition in a Hawaiian Woodland: The Importance of Indirect Effects Michelle C. Mack * and Carla M. D Antonio Department of Integrative Biology, University of California Berkeley, Berkeley, California 94720, USA ABSTRACT Exotic grasses and grass-fueled fires have altered plant species composition in the seasonal submontane woodlands of Hawaii Volcanoes National Park. These changes have altered both structural and functional aspects of the plant community, which could, in turn, have consequences for litter decomposition and nitrogen (N) dynamics. In grass-invaded unburned woodland, grass removal plots within the woodland, and woodland converted to grassland by fire, we compared whole-system fluxes and the contributions of individual species to annual aboveground fine litterfall and litterfall N, and litter mass and net N loss. We assessed the direct contribution of grass biomass to decomposition and N dynamics, and we determined how grasses affected decomposition processes indirectly via effects on native species and alteration of the litter layer microenvironment. Grasses contributed 35% of the total annual aboveground fine litterfall in the invaded woodland. However, total litterfall mass and N were not different between the invaded woodland and the grass removal treatment because of compensation by the native tree Metrosideros polymorpha, which increased litter production by 37% 5% when grasses were removed. The 0.3 gnm 2 /y 1 contained in this production increase was equal to the N contained in grass litter. Litter production and litterfall N was lowest in the grassland due to the loss of native litter inputs. Decomposition of litterfall on an area basis was highest in the grass-invaded woodland. We attributed this effect to increased inherent decomposability of native litter in the presence of grasses because (a) the microenvironment of the three vegetation treatments had little effect on decomposition of common litter types and (b) M. polymorpha litter produced in the invaded woodland decomposed faster than that produced in the grass removal plots due to higher lignin concentrations in the latter than in the former. Area-weighted decomposition was lowest in the grassland due to the absence of native litter inputs. Across all treatments, most litter types immobilized N throughout the incubation, and litter net N loss on an area basis was not different among treatments. Our results support the idea that the effects of a plant species or growth form on decomposition cannot be determined in isolation from the rest of the community or from the direct effects of litter quality and quantity alone. In this dry woodland, exotic grasses significantly altered decomposition processes through indirect effects on the quantity and quality of litter produced by native species. Key words: biological invasion; exotic grasses; dry woodland; litter production; decomposition; nitrogen dynamics; litter quality; microclimate; species interactions. INTRODUCTION Received 28 January 2002; accepted 27 November 2002; published online 6 November *Corresponding author; mcmack@botany.ufl.edu Decomposition is a key ecosystem process that regulates carbon (C) and nutrient cycling and ultimately controls the availability of nitrogen (N), the 723

2 724 M. C. Mack and C. M. D Antonio nutrient most likely to limit primary production in terrestrial ecosystems (Vitousek and Howarth 1991). Because plant species and physiognomic growth forms differ in the chemical composition of their litter, decomposition is particularly sensitive to changes in plant species and growth form composition (Melillo and others 1982; Flanagan and Van Cleve 1983; Johnson and Damman 1991; Hobbie 1992). To date, most research that has examined the effects of plant species or growth forms on decomposition has focused almost exclusively on traits that are directly related to the inherent decomposability or quality of litter. The effect of a species litter on decomposition, then, is dependent on its N and phosphorus (P) content (Melillo and others 1982; Hobbie 1992; Berendse and others 1994, Berg and others 1996); the types and concentrations of C compounds, such as lignin (Stohlgren 1988; Hobbie 1996); and the relative availability of C and nutrients to decomposers (Melillo and others 1982; Pastor and others 1987; Taylor and others 1989; Parton and others 1994; Hobbie and Vitousek 2000). Although fewer studies have examined the effects of the amount of litter produced by a focal species relative to other species in the system (see for example, Hobbie 1996), it is clear that the impact of litter quality on total ecosystem decomposition is at least partially due to its quantity. In addition to litter quality and quantity, plant species and growth forms have been shown to have effects on community and ecosystem process that may affect decomposition indirectly. Competitive (Wardle and others 1998; Hooper 1998; Knops and others 2001) or facilitative (Callaway and Walker and Smith 1997; Pugnaire and others 1996) interactions among plant species may alter the quantity or quality of litter produced by dominant species. In addition, plant functional or structural traits provide unique microenvironments for decomposing organisms (Jones and others 1997; Walker and Smith 1997; Caldwell and others 1998; D Antonio and others 1998; Hector and others 2000) that may stimulate (for example, see Aplet 1990) or suppress (for example, see Vance and Chapin 2001) their activities. Although these types of indirect effects on decomposition have been identified, there are relatively few studies that have quantified their importance relative to the more widely recognized direct effects of litter quality and quantity on decomposition. In this study, our goal was to quantify and compare both direct (litter quantity and quality) and indirect (plant interactions and microenvironment) effects of exotic, invasive grasses on decomposition Figure 1. Hypothesized direct (solid lines) and indirect (dashed and dotted lines) effects of exotic grasses on decomposition in a Hawaiian dry woodland. Arrows indicate directions of effects. and N dynamics in a seasonal submontane woodland located in Hawaii Volcanoes National Park (HVNP). Biological invasions offer appropriate model systems for examining the effects of plant community composition on ecosystem processes because they involve a change in species composition; if an ecosystem process differs before and after invasion, it can be shown unambiguously that the inserted species or growth form has an impact on that process (Vitousek and Walker 1989). Grasses were rare in these evergreen woodlands until the 1960s, when exotic C 4 grasses invaded the area following an expansion of grazing land outside of the park boundary (Smith 1985). In a previous study, we showed that exotic grass litter had higher C:N ratios and lignin concentrations than the leaf litter of native species in these woodlands (Mack and others 2001). We also showed that the presence of grasses decreased soil mineral N pools (Mack and D Antonio 2003), reduced growth and tissue N concentrations of native species (D Antonio and others 1998; Mack and others 2001), and decreased soil surface temperatures (Mack and D Antonio 2002). These results led us to hypothesize that exotic grasses would reduce decomposition and N flux rates in this system directly via the production of large amounts of low quality litter and indirectly via a competitive reduction of native species litterfall and litter N content, as well as a reduction of the litter layer temperature (Figure 1). Finally, we predicted that these factors and the loss of native litter inputs would lead to greatly reduced decomposition and litter N loss after fire. We tested these hypotheses by comparing litterfall and decomposition across three states of community composition: (a) invaded unburned woodland, (b) grass removal plots within the woodland

3 Litter Decomposition in a Hawaiian Woodland 725 created to simulate conditions that may have existed prior to invasion, and (c) woodland converted to grassland by fire. First, we assessed the direct contributions of grasses and woody species to decomposition and N dynamics through measurements of litterfall and rates of in situ decomposition of dominant litter types across these three states of community composition. We then determined how grasses impacted decomposition processes indirectly by examining their effects on (a) the litter quantity and quality of native species, and (b) alteration of the litter layer microenvironment. To test for the effects of grasses on the initial tissue quality of native species, we decomposed leaf litter of the dominant native tree produced in either the removal or control treatment in a common site. We isolated grass alteration of microenvironmental factors by decomposing native and exotic litters from a common site in all treatments. METHODS Site Description We measured the effects of exotic grasses on fine litterfall and decomposition in the seasonal submontane woodland zone of Hawaii Volcanoes National Park (HVNP), Hawaii, near Kipuka Nene campsite (19 6'N, 'W; elevation, m). The soils, vegetation structure, and N pools in these woodlands have been described in detail elsewhere (Hughes and others 1991; D Antonio and others 1998; Mack and others 2001). Soils are young (less than 500 years old), shallow (mean depth, around 40 cm) ash deposits overlaying approximately 1000-year-old pahoehoe lava (Wolf and Morris 1996). Precipitation averages 1400 mm/y and is highly seasonal, with approximately 10% of annual precipitation falling between March and August (HVNP Resource Management Division Archives unpublished). Mean annual air temperature (T) is 20 C. Historically, these were open canopy woodlands of the native tree Metrosideros polymorpha (Myrtaceae) with an understory of native shrubs. Ground cover consisted of evergreen sedges, lichens, and bryophytes; native grasses were rare (Smith and Tunison 1992). Exotic C 4 grasses invaded these woodlands in the mid-1960s (Smith 1985), and today there are no submontane woodlands without these grasses in the understory (D Antonio and others 1998). Schizachyrium condensatum (Poaceae), a perennial bunchgrass native to Central and South America, is the most abundant grass in the understory. Melinis minutiflora (Poaceae), a stoloniferous perennial grass native to Africa, and Andropogon virginicus (Poaceae), a perennial bunch grass native to North America, are also present at low densities. Botanical nomenclature follows Wagner and others (1990). Since grasses invaded in the 1960s, these woodlands have experienced a three-fold increase in fire frequency and a 35-fold increase in hectares burned per fire (D Antonio and others 2000). The direct effects of fire on N cycling are short-lived; increased rates of soil N mineralization and nitrification return to ambient levels within 6 months (C. M. D Antonio, unpublished), and less than 1% of total ecosystem N is lost due to biomass burning (Mack and others 2001). Consequently, the most lasting effect of fire is a change in vegetation composition. The woodlands are converted from a functionally diverse assemblage of shrubs, graminoids, and trees to grassland dominated by M. minutiflora, with a few highly dispersed shrubs (Hughes and Vitousek 1993). Experimental Design Our study sites consisted of three experimental manipulations of vegetation type, hereinafter referred to as treatments. These consisted of (a) unburned woodland with grasses present (hereafter, Woodland Grass, or W G); (b) a grass removal treatment in the unburned woodland (hereafter, Woodland, or W), which simulated conditions that may have existed in the woodland prior to grass invasion; and (c) woodland burned in both 1970 and 1987 that converted to grassland (hereafter, Grassland, or G). In 1991, Woodland Grass and Woodland treatments were each randomly assigned to four m plots that had been selected to have similar native and exotic vegetation cover, soil depth, and aspect (D Antonio and others 1998). Standing aboveground biomass and necromass of exotic grasses were removed from Woodland plots by clipping grasses just below the root crown. Roots were left intact to minimize soil disturbance. New grass seedlings were weeded from plots semiannually between 1991 and We randomly selected four m Grassland plots from burned woodland in 1993 with the same criteria used to select plots in the unburned woodland, except that their locations were constrained to areas contiguous to the unburned woodland that burned in 1970 and Prefire aerial photographs and ground surveys show that all plots in this study were at one time part of a continuous forest belt dominated by an open-canopy of M. polymorpha (HVNP Resource Management Division Archives unpublished). For the duration of this study, vegetation cover in

4 726 M. C. Mack and C. M. D Antonio the Woodland treatment consisted of 100% native woody species cover and biomass, whereas in the Woodland Grass, grasses made up 70% of the understory cover and 30% of understory biomass (D Antonio and others 1998). In the Grassland, M. minutiflora made up 90% of the total cover and 70% of the total biomass (Mack and others 2001). Dodonaea viscosa is the only native shrub that recruits into these grasslands after fire. It made up less than 10% of biomass in the Grassland between 1992 and 1998 (D Antonio and others 2000). Aboveground Fine Litterfall We measured fine litterfall of native species in these vegetation treatments in by placing eight litter traps in a grid formation on the ground of each m plot. Traps consisted of 0.16-m 2, 10-cm deep plastic trays lined with 1-mm mesh fiberglass screening. They were emptied bimonthly between August 1994 and August Coarse woody debris (more than 3 cm in diameter) was discarded, and the remaining litter was pooled at the plot level, dried at 70 C for 48 h, sorted to species and dominant fractions (leaf, twig, or fruit litter), and weighed. From each litter trap collection, we analyzed % N on a subsample of each litter type (see Lab Analyses), except for leaves of the dominant species. For these, we sampled senesced but still attached leaf litter from at least 10 plants/plot at each trap collection time point. The mean plot N concentration of each litter type was then multiplied by dry mass to calculate litterfall N content at each time point. The dry weight and N content of total annual native litterfall was calculated per plot as the sum of species contributions per square meter across all time points. Because litter production of exotic grasses could not be estimated from litter traps, we measured this by harvesting grasses at peak live biomass in December We determined litter production as the fraction of live biomass produced during the previous year. We harvested biomass from two 1-m 2 plots in each replicate plot in Woodland Grass and Grassland; sorted it into live leaves, live stems, and dead material; and weighed it in the field. Subsamples were returned to the lab, weighed and dried at 70 C for 48 h and reweighed to convert field-moist weights to dry weights. To determine grass litter N concentrations and N pools, we harvested leaves and stems following scenescence and color change at the end of the wet season and analyzed % N (see Lab Analyses). The N concentration of each tissue type was multiplied by the dry mass of that tissue type per square meter. We then calculated total annual litterfall mass and N for each plot as the sum of native and exotic litter produced per square meter per year. We analyzed vegetation treatment effects on total fine litterfall using one-way analysis of variance (ANOVA). We analyzed treatment effects on individual species that occurred in two treatments using t-tests and species that occurred in three treatments using one-way ANOVA. We did not compare contributions among species because they were not independent (that is, increased growth and litter production in one species may have resulted in decreased growth of another). If necessary, data were in-transformed to meet the assumptions of homogeneity of variance; percentage data was transformed using an arc-sine square-root transformation. If assumptions were not met via transformation, we used a Mann-Whitney U test or a Kruskal-Wallis test. We used Bonferroni tests for post hoc multiple comparisons in all cases. This protocol was used for all analyses described hereafter. All statistical analyses were performed using Systat. 7.0 (Systat, Evanston, IL, USA). Litterbag Experiments We conducted three decomposition experiments to estimate litter mass and N loss rates and determine direct and indirect controls over decomposition in the three vegetation treatments. First, we decomposed leaf litter of each major species in its plot of origin (hereafter, in situ experiment) to determine in situ rates of decomposition. Second, we isolated vegetation treatment effects on initial litter quality of the dominant native tree, M. polymorpha, by decomposing leaf litter from the Woodland and Woodland Grass in a common site in W G (hereafter, litter quality experiment). Third, we isolated treatment effects on the microenvironment for decomposition by decomposing litter collected in a common site in all vegetation treatments (hereafter, microenvironment experiment). For all experiments, senesced but still attached leaf litter was collected in late December 1994, at the end of the wet season. Leaves were subsampled for leaf mass per square meter of leaf area (LMA), a factor hypothesized to covary with litter quality. The remainder of the field-moist leaves were dried at 30 C for 48 h; then 1-g subsamples were sewn into cm bags made from 1-mm mesh fiberglass screening. A final subsample was weighed, dried at 70 C, reweighed to obtain a dry weight conversion factor, then returned to the lab for analyses of initial tissue % C, N, and lignin (see Lab Analyses). Dry weight at 70 C was used to calculate initial dry mass of litter in each bag. Bags were individually labeled and four or six bags of

5 Litter Decomposition in a Hawaiian Woodland 727 each litter type were tied to nylon strings and placed at two random locations in the litter layer in replicate vegetation treatment plots in January One bag of each litter type was collected from each string at 6, 12, 18, and 24 months for most litter types. Leaf litter in the microclimate experiment was collected at a higher frequency: 3, 6, 8, 12, 18, and 24 months. See Lab Analyses for treatment of litter after collection. We analyzed percent initial mass and N remaining after 2 years as indices of decomposition rates. In the case of the in situ experiment, we analyzed years 1 and 2 seperately. All rate calculations and statistical analyses were done on the mean of the two bags collected from each plot at each time point. In the in situ experiment, where we decomposed leaf litter within its plot of origin, native leaf litter used in the Woodland and Woodland Grass vegetation treatments included M. polymorpha, Styphelia tameiameiae, Osteomeles anthyllidifolia, and D. viscosa. In the Woodland Grass and Grassland, we also used exotic leaf litter from S. condensatum and M. minutiflora. The only native leaf litter used in the Grassland was D. viscosa. To detect vegetation treatment effects and species differences, we analyzed decomposition and net N loss with separate twoway ANOVAs on three groups of species: (a) species common to the Woodland and Woodland Grass, (b) species common to Woodland Grass and Grassland and (c) species common to all three vegetation treatments. In these analyses, we also included S. condensatum stem litter (included in analyses of groups b and c) and D. viscosa twig litter (included analyses of groups a c) from the microenvironment experiment. To compare the effects of vegetation treatments on litter quality of the dominant native tree, M. polymorpha, we decomposed leaf litter from the Woodland and Woodland Grass in the latter vegetation treatment. We tested for treatment of origin effects with independent t-tests. To isolate vegetation treatment effects on decomposition via microenvironment, we decomposed litter collected in the unburned woodland surrounding the Woodland Grass plots in each vegetation treatment. We used leaf and twig litters from D. viscosa (the native species present in all treatments) and leaf and stem litters from S. condensatum (the dominant exotic species in W G). Stems used in this experiment were reproductive culms collected after flowering and color change, and were stripped of leaf bases. We cut culms into 4-cm long segments and mixed them so that each litterbag had a range of culm diameters. We collected D. viscosa twigs from dead but still attached branches, chopped them into 4-cm segments, and sorted them by diameter. We only used twigs that were between 0.75 and 1.5 mm in diameter (mean SE diameter cm, n 75). Twig and stem litters were bagged and placed into plots as described above. For either leaf litter or stem and twig litter, we analyzed species and vegetation treatment effects on percent initial mass or N remaining with two-way ANOVA. Area-weighted Mass and Net N Loss from Aboveground Litter We used annual litterfall and in situ decomposition rates to calculate an index of the flux of organic matter or N through incoming aboveground litterfall. For each species and litter type, we multiplied in situ mass and N loss rate at year 1 [1 (% initial mass or N remaining)] by litterfall mass or N to calculate mass or N lost per square meter during year 1. We then multiplied the remaining litter mass or N (detrital mass and N after year 1) by the in situ mass or N loss rate for year 2 to calculate mass or N loss per square meter during year 2. We summed both area-weighted loss terms to estimate annual mass and N loss from aboveground litter. Finally, we summed all species and litter types to estimate total mass or N flux per square meter from each plot. This term includes only the first 2 years of decomposition (approximately 50% total mass loss); therefore, it represents a relative measure that should be comparable among treatments, not an absolute value of mass or N flux from the litter compartment. For rates of mass or N loss from twigs and stems, we used plot values from the microenvironment experiment. For lack of a better estimate, we assumed fruit decomposition rates to be equal to leaf decomposition rates for each species. We assumed decomposition rates of the other fraction to be similar to the mean decomposition rate of all measured fractions in a plot. Lab Analyses For all leaf litter experiments, area of field-moist senescent leaves was measured on a leaf area meter ( T area meter; Delta Devices, Cambridge, England, UK). Leaves were weighed, dried at 70 C, and reweighed to determine LMA. Subsamples of senescent leaves used in litterbags and collected from each litter trap and subsamples of nonleafy fine litter in traps were returned to the University of California Berkeley, where they were ground in a Wiley mill (Thomas Scientific, Philadelphia, PA, USA) with a no. 40 screen, followed by a ball mill.

6 728 M. C. Mack and C. M. D Antonio Litter in bags was collected at each time point, gently washed in deionized water, and dried at 70 C. Soil and foreign litter was manually removed, and litter was weighed. Small samples were ground on a Wig-L-Bug amalgamator (Crescent Dental, Lyons, IL, USA). Ground materials were analyzed for % C and N on a Carlo Erba NA 1500 CHN analyzer (Fisons Instruments, Beverly, MA, USA). We measured % lignin with an acetyl bromide digest method (Iiyama and Wallis 1990). Absorbance of digested samples was measured spectrophotometrically and converted to % lignin using a constant (standard NIST pine needles 21.4) (B. Dewey personal communication). RESULTS Litterfall Mass and N Content Grasses contributed 35% of the total aboveground fine litterfall in the W G vegetation treatment (Figure 2). This was comparable to leaf and twig contributions by the native tree M. polymorpha (40%) and greater than the contributions of native shrubs (19%) (Figure 2). In the W vegetation treatment, where grasses had been removed, native species compensated for the absence of grasses by increasing litter production 54 5 g/m 2 (mean 1 SE) relative to natives in W G. The main species to respond to grass removal was M. polymorpha, which increased litter production by 37% in W relative to W G(t , P 0.05) (Figure 2). As a result, there was no difference in total fine litterfall between these two treatments. N delivered in fine litterfall did not differ between these two treatments either. N concentration in leaf litter was higher in W than in the W G for M. polymorpha (t , P 0.05), and for the shrubs D. viscosa (F 2,9 9.02, P 0.007) and O. anthyllidifolia (t , P 0.05) (Table 5). However, the increased N delivered in native leaf litter in W was equal to that contained in grass litter in W G (Figure 2). In the G vegetation treatment, total litter production (F 2, , P 0.001) and litterfall N (F 2, , P 0.001) were considerably lower than in the W or W G (Figure 2). Although native litter production was negligible in G (Figure 2), grass litter production was not different between W G and G, even though grass species dominance changed (Table 5). In situ Decomposition and N Dynamics Generally, native and exotic litter types decomposed more rapidly and immobilized less N in W Figure 2. Mean ( 1 SE) mass and N content of fine litterfall from the native tree Metrosideros polymorpha, native shrubs, exotic grasses, and total vegetation in three vegetation treatments within a Hawaiian dry woodland, Lower case letters indicate significant differences within a group at P 0.05 (n 4). G than in W or G. For native species litter common to W and W G, mass loss was somewhat faster in W G in year 1 (P 0.06) and significantly faster in year 2 (Table 1) due to greater mass loss from M. polymorpha and S. tameimeiae (Figure 3). Two native litter types, (S. tameiameiae and O. anthyllidifolia, also lost N more rapidly in W G than in W during year 2 (Figure 4). For species common to W G and G, mass loss was not different in year 1, but significantly faster in W G than in G in year 2 due to higher mass loss from stems and twigs in W G (Table 1). In year 2, most litter types lost more N in W G than in G (Figure 4). Dodonaea viscosa leaves and twigs, the only litter types present in all vegetation treatments, decomposed faster in W G than in W in year 2 and slowest in G (Figure 3 and Table 1). Species were a significant source of variation across all comparisons of in situ litterbag decompo-

7 Litter Decomposition in a Hawaiian Woodland 729 Table 1. Two-way ANOVA Results: In Situ Litter Decomposition Initial Mass Lost (%) Initial N Lost (%) Year 1 Year 2 Year 1 Year 2 Treatment Source df F P F P F P F P W and W G a Veg. treat. 1, Species 4, V. t. Sp. 4, W G and G b Veg. treat. 1, Species 4, V. t. Sp. 4, W, W G, G c Veg. treat. 2, Tissue 1, V. t. T. 1, W, Woodland; W G, Woodland and Grass; G, grassland; df, degrees of freedom. Boldface figures indicate significance at P 0.05 (n 4). a Species in analysis were Metrosideros polymorpha, Osteomeles anthyllidifolia, Styphelia tameiameiae, and Dodonaea viscosa leaves and twigs. b Species in analysis were Schizachyrium condensatum leaves and stems, Melinis minutiflora leaves, and D. viscosa leaves and twigs. c Tissues in analysis were leaves and twigs of D. viscosa. sition rates in year 1 (Table 1). Species common to W G and W fell into three groups during both years: (a) quickly decomposing (D. viscosa and S. tameiameiae), (b) significantly slower decomposing (shrub twigs), and (c) M. polymorpha and O. anthyllidifolia, which were not significantly different from either group (Figure 3). The vegetation treatment species interaction in the W versus W G comparison in year 1 occurred because S. tameiameiae decomposed more rapidly in W G than in W, whereas other species decomposed similarly in W and W G (Figure 3). In the comparison of W G versus G for both years, S. condensatum stems decomposed faster than all litters (P 0.05) except for D. viscosa leaves (Figure 3). S. condensatum leaves, by contrast, decomposed very slowly in year 1 (Figure 3). Species varied significantly in net N loss in all cross-species comparisons in year 1 (Table 1). In year 2, the W G versus W ANOVA comparison was the only one with significant differences between species net N loss (Table 1) because S. tameiameiae released more N than D. viscosa twigs (P 0.005) (Figure 4). All other species were similar. The species-by-vegetation treatment interaction was due to the fact that S. tameiameiae lost more N in W G than in W (P 0.04), whereas other species were not different between the treatments (Table 1 and Figure 4). Area-weighted Decomposition and N Flux from Litterfall When in situ decomposition rates are combined with litter production per species, it is clear that grass removal significantly decreased decomposition on an areal basis in the unburned woodland (F 2, , P 0.001) and variably increased area-weighted net N immobilization through litter (F 2,9 2.75, P 0.11) relative to treatments with grasses (Figure 5). Native litter pools (M. polymorpha plus native shrubs and other) showed net immobilization of N, with the W native pool variably immobilizing more N than the W G native pool (P 0.06) and significantly more N than the G native pool (F 2, , P 0.002) (Figure 5). Although M. polymorpha contributed more N mass in litterfall in W than in the other two treatments (Figure 2), it also immobilized more N on litter in W (Figure 4), resulting in more net immobilization in W than in W G(t , P 0.02) (Figure 5). The grass litter pool mineralized N in W G but immobilized N in G (Figure 5). This was due in part to the difference in grass species dominance between the vegetation treatments. Schizachyrium condensatum leaves had higher N release than the G- dominant M. minutiflora (Figure 4). The absence of natives in G led to low rates of area-weighted decomposition, but net N fluxes did not differ from the unburned woodland (Figure 5). Litter Quality of Exotic Grasses versus Native Species In the in situ experiment, we found that grass litter had decomposed similarly (leaves) or more quickly (stems) than most litter from native species by year 2 (Figure 3). Indices of litter quality showed that grass leaves were generally similar to natives except

8 730 M. C. Mack and C. M. D Antonio Figure 3. Mean ( 1 SE) % initial mass lost after 1 or 2 years from litterbags decomposed in situ in three vegetation treatments within a Hawaiian dry woodland. Bags contained leaves of dominant species, twigs of Dodonaea viscosa, or stems of Schizachyrium condensatum and were decomposed in plot of origin. Asterisk or lower-case letters indicate significant differences within species, between or among vegetation treatments at P 0.05 (n 4). See Table 2 for full names of species. for LMA and lignin content (Tables 2 and 3). Grass leaves were significantly thinner and higher in lignin than native species (Table 3). Percent lignin was the only significant predictor of in situ mean leaf decomposition rate across all species and treatments (leaves only; % initial mass remaining after 2 years % lignin, R , P 0.006, n 13). When regressed against mass loss or net N loss, none of the other litter quality indices (Table 3) or ratios of the indices (including lignin:n, lignin:c, or Figure 4. Mean ( 1 SE) % initial nitrogen (N) lost after 1 or 2 years from litterbags decomposed in situ in three vegetation treatments within a Hawaiian dry woodland. Bags contained leaves of dominant species, twigs of Dodonaea viscosa, or stems of Schizachyrium condensatum decomposed in plot of origin. Asterisk or lower-case letters indicate significant differences within species, between or among vegetation treatments at P 0.05 (n 4). See Table 2 for full names of species. lignin:lma) had regression coefficients significantly different from zero (data not shown). Effects of Vegetation Treatments on Litter Quality of M. polymorpha We isolated the effect of vegetation treatment on initial litter quality of the dominant native tree M. polymorpha by decomposing litter produced in W and W G in a common site (Figure 6). After 2 years, M. polymorpha litter from W G had lost more mass (t 6 3.5, P 0.01) and immobilized

9 Litter Decomposition in a Hawaiian Woodland 731 and stem litter decomposed more slowly and immobilized more N in G than in either treatment in the unburned woodland (Table 4 and Figure 7). Percent initial mass remaining at year 2 was significantly lower for W G(P 0.01) and W (P 0.02) compared to G for both litter types. Grass stems lost more mass than twigs within each vegetation treatment (P 0.001). DISCUSSION Figure 5. Mean ( 1 SE) area-weighted decomposition (mass loss and net nitrogen loss) from litterfall components and total litterfall in three vegetation treatments within a Hawaiian dry woodland. Lower-case letters indicate significant differences within a category at P 0.05 (n 4). less N (t , P 0.07) than M. polymorpha litter from W. Percent of initial mass remaining of litter from both origins were similar to in situ experiment values (litter quality W 75% 2% and W G 61% 3%, mean SE, in situ W 74% 5% and W G 63% 2%). Although M. polymorpha litter from W has significantly higher % N and lower C:N ratios than litter from W G, these variables were not correlated with decomposability. By contrast, lignin tended to be higher in W (Mann-Whitney U 2.0, P 0.08) (Table 2) and was positively related to % initial mass remaining after 2 years across both treatments (r , P 0.07). Effect of Microenvironment on Decomposition Microenvironment had no effect on leaf mass loss over 2 years (Table 4 and Figure 7). By contrast, twig Our study shows that both direct and indirect effects are important for interpreting the impact of exotic grasses on decomposition in this Hawaiian woodland. In fact, direct effects on litterfall and decomposition were counterbalanced by indirect effects that were opposite in sign and, in the case of decomposition, were greater in magnitude (Figure 8). Contrary to our original hypotheses (Figure 1), exotic grasses had a net positive effect on decomposition, primarily due to the surprising effect of grasses on the litter quality of native species (Figure 8). Grasses had no effect on net N loss from litter. Grasses directly contributed 35% of the total annual aboveground fine litterfall in the invaded woodland. This contribution, however, was counterbalanced by decreased litter production by native species in the invaded woodland, an indirect effect of exotic grasses on litterfall. Total litterfall mass and N were not different between the invaded woodland and the grass removal treatment because of compensation by the native tree Metrosideros polymorpha, which increased litter production by 37% 5% when grasses were removed (Figure 2). The 0.3 g N m 2 /y 1 contained in this production increase was equal to the N contained in grass litter in the invaded woodland. Litter production and litterfall N was lowest in the grassland. This pattern was driven by the loss of native litter inputs, since grass litter production was not different between the grassland and the grass-invaded woodland (Figure 2). Contrary to our original hypothesis, decomposition on an area basis was highest in the grassinvaded woodland, despite the fact that grass litter decomposed at a rate similar to the native litters that it replaced. Again, this direct effect of grasses on decomposition was countered by the indirect effect of grasses on the decomposition rate of native species litters. We attributed this effect to increased inherent decomposability of native litter in the presence of grasses because (a) microenvironment had little effect on decomposition of common litter types (Figure 7) and (b) M. polymorpha litter produced in the invaded woodland decomposed faster

10 732 M. C. Mack and C. M. D Antonio Table 2. Species Mean Leaf Litter Quality of Dominant Species % N C:N LMA (g/m 2 ) % Lignin W W G G W W G G W W G G W W G G Metrosideros polymorpha 0.34 a 0.27 b a b (0.04) (0.03) (19.7) (16.2) (12) (20) (1.1) (0.9) Osteomeles anthyllidifolia 0.40 a 0.36 b a b (0.01) (0.02) (2.1) (5.6) (15) (9) (0.7) (0.2) Styphelia tameiameiae 0.40 a 0.41 b a b (0.01) (0.01) (16.8) (7.0) (11) (8) (0.4) (0.3) Dodonaea viscose 0.46 a 0.39 b 0.37 b 97.1 a b c (0.03) (0.02) (0.03) (4.7) (4.9) (8.6) (11) (13) (23) (0.5) (0.6) (0.6) Twigs 0.36 * * 31.9 * Twigs 0.36 * * 31.9 * (0.003) (1.5) (0.4) Schizachyrium condensatum 0.29 a 0.35 b a b (0.02) (0.01) (9.7) (4.5) (18) (12) (1.5) (0.7) Melinis minutiflora a b (0.01) (0.01) (18.2) (11.3) (16) (8) (0.7) (0.4) W, Woodland; W G, Woodland Grass; G, Grassland; N, nitrogen; C, carbon; LMA, leaf mass per unit area. % lignin was determined with an acetyl-bromide assay (see Methods). Within species, means with different superscript letters are significantly different at P 0.05 and are shown in boldface (n 4 unless noted). *n 3 analytical replicates Table 3. Two-way ANOVA Results: Litter Quality Indices % N C:N LMA (g/m 2 ) % Lignin Treatment Source df F P F P F P F P W, W G a Veg. treat. 1, Species 3, V. t. S. 3, W G, G b Veg. treat. 1, Species 2, V. t. S. 2, W, W G, G c Veg. treat. 2, W, Woodland; W G, Woodland Grass; Grassland; df, degrees of freedom; N, nitrogen; C, carbon; LMA, leaf mass per unit area. % lignin was determined with an acetyl-bromide assay (see Methods). Boldface indicates significant differences at P 0.05 (n 4). a Species in analysis were Metrosideros polymorpha, Osteomeles anthyllidifolia, Styphelia tameiameiae, and Dodonaea viscosa. b Species in analysis were Schizachyrium condensatum, Melinis minutiflora, and D. viscosa. c Species in analysis was D. viscosa. than M. polymorpha litter produced in the grass removal plots (Figure 6), perhaps due to lower lignin concentrations in the presence of grasses (Table 2). Finally, area-weighted decomposition was lowest in the grassland, primarily because of the absence of native litter inputs (Figure 6). As is common in studies of the early stages of decomposition (Berg and Matzner 1997), litterfall tended to immobilize N across most species and treatments (Figure 4). Immobilization on an area basis tended to be lower in treatments with grasses (Figure 5), but it was not clear from these results if this will translate to higher rates of N mineralization in the later stages of decomposition or to the feedback between litter N mineralization and plant production that we presented earlier. Our results do clearly indicate that exotic grasses have altered the key ecosystem process of decomposition in this ecosystem. Litter C (as indicated by mass loss) dynamics were very responsive to the indirect effects of exotic grasses (Figure 5), which in turn has important consequences for C cycling and storage. These results indicate that changes in plant species composition that involve the introduction of

11 Litter Decomposition in a Hawaiian Woodland 733 Figure 6. Mean ( 1 SE) % initial mass or nitrogen (N) remaining in litterbags containing Metrosideros polymorpha leaves originating in either Woodland or Woodland Grass vegetation treatments and decomposed in a common site during ** P 0.01 and * P 0.05 (n 4). a new growth form can alter key ecosystem processes through indirect effects on preexisting species that are equal to or larger than more often studied direct effects. This illustrates that the effects of a plant species or functional group of species on decomposition cannot be determined in isolation from the rest of the community or from the direct effects of litter quality and quantity alone. Litterfall Grasses contributed 35% of the total annual aboveground fine litterfall in the invaded woodland (W G) (Figure 2), but evidence from the removal experiment suggests that this production was at the expense of native species. The native tree M. polymorpha was able to compensate litter production for the removal of grasses. This type of compensatory response should predictably buffer litter production against changes in community composition (Lawton and Brown 1995; Hooper 1998; Wardle and others 1999) and is characteristic of systems where resource competition is intense and niche overlap is high (Harper 1977; Tilman 1990). In a related study comparing growth of native species in W (removal treatment) and W G (control treatment) (D Antonio and others 1998), we established that grasses depressed growth and recruitment of native species through competition for N. We ruled out competition for water because soil moisture was consistently lower and plant-level indicators of water stress were consistently higher in W than in W G. It is unlikely that disturbance associated with removal contributed to increased litter production and N concentrations (see Aaressen and Epp 1990) because litterfall measurements were taken 4 years after the initial treatment. Furthermore, plant N availability was not significantly different between W and W G (Mack and D Antonio 2003). For the grassland site, the overall reduction in litterfall relative to the unburned woodland was attributable to the fire-driven loss of dominant native species. Grass litter production and litterfall N was the same whether grasses were alone (G) or with natives (W G) (Figure 2) despite a change in grass species dominance (Table 5). Surprisingly, competition between grasses and natives appears to be asymmetrical in this system; grasses decrease native growth, but natives do not appear to have effects on the growth of grasses. It is likely that some factor other than N was limiting grass litter production in the grassland; annual N mineralization was significantly greater in grassland than in the unburned woodland, and grass uptake only accounted for 17% of N available in the grassland (Mack and others 2001). Furthermore, compensation was not reciprocal between grasses and native species; natives were able to compensate litter production for the removal of grasses in the unburned woodland, but grasses were not able to compensate for the absence of native species in the woodland. This is consistent with the view that species, especially those that differ in physiognomy, are not equal in terms of their ecosystem-level effects (Chapin and others 1996; Power and others 1996; Wardle and others 1999). Decomposition Although litterfall was not different between W and W G, area-weighted decomposition was signifi-

12 734 M. C. Mack and C. M. D Antonio Table 4. Two-way ANOVA Results: Microenvironment Experiment Tissue type Source df Mass F P F P Leaves Veg. treat. 2, Species 1, V. t. Sp. 2, Sticks and stems Veg. Treat. 2, Species 1, V. t. Sp. 2, N N, nitrogen; df, degrees of freedom. Boldface indicates significance at P 0.05 (n 4). Figure 7. Mean ( 1 SE) % initial mass or nitrogen (N) remaining of leaves, stems, and twigs of Dodonaea viscosa (Dod). and Schizachyrium condensatum (Sch.) decomposed in the three vegetation treatments within a Hawaiian dry woodland. ** P 0.01 and * P 0.05 (n 4). cantly higher in W G, primarily because native species leaf litter decomposed more quickly and immobilized less N when grown in the presence of grasses. We attribute the higher area-weighted decomposition rate to an indirect effect of grasses on native species because (a) total litterfall was not different between the two treatments, and (b) M. polymorpha litter production compensated for grass litter production in W (Figure 2), and these litter types, on average, did not differ in decomposition rate when incubated in a common site (Figure 3). Thus, the addition of the grass litter component alone without stimulation of native decomposition would have resulted in no difference between W and W G in area-weighted decomposition. Why did native litter decompose more rapidly in the grass-invaded woodland than in the removal treatment? Our common litter experiment showed that microenvironment had little effect on decomposition rates despite the structural changes associated with either grass removal or loss of natives (Figure 7). The differences that we found in decom-

13 Litter Decomposition in a Hawaiian Woodland 735 Figure 8. Effects of exotic, invasive grasses on areaweighted decomposition in a Hawaiian dry woodland. Solid lines indicate direct effects; dashed or dotted lines indicate indirect effects; arrows indicate directions of effects. position rates were apparently due to treatment effects on litter quality. For M. polymorpha the litter type that made up 35% of native litter inputs litter produced in the absence of grasses (W) decomposed more slowly than that produced in the presence of grasses (W G) when both types were incubated in a common site. Metrosideros polymorpha litter produced in W had higher N concentrations, but it also tended to have higher lignin concentrations than litter produced in W G (Table 2). Litter N concentrations in these woodlands are extremely low (Killingbeck 1996) and are similar to litter from other M. polymorpha forests that are strongly N- limited (Crews and others 1994; Hobbie and Vitousek 2000). Increased N concentrations in litter originating in W may not have stimulated decomposition rates because the N was inaccessible to microbes due to the low availability of labile C substrates (Prescott 1995; Hobbie 2000), as indicated by higher lignin concentrations in this litter (Table 4). Alternatively, if microbes primarily decompose recalcitrant substrates such as lignin to mineralize N (Berg 1986; Fog 1988; O Connell 1994), increased N concentrations may have inhibited decomposition of litter since N was relatively more available. Our data do not allow us to reject either of these hypotheses conclusively. Although % N differed significantly between treatments (Table 2), neither its concentration nor its relationship to C or lignin was correlated with mass loss either within M. polymorpha or across all species. Lignin concentration, by contrast, was positively correlated with mass loss across all species, and treatment effects on M. polymorpha lignin concentration were marginally significant (P 0.08), with higher concentrations in litter produced in W than in W G (Table 2). Increased lignin concentrations may be indicative of acclimation to a drier microclimate (Heller 1995; Lange and others 1995; Leinhos and Bergmann 1995). Soil moisture was lower in the removal treatment than in the control (D Antonio and others 1998; Mack and D Antonio 2003), which is likely to be related to a combination of removal of the insulating grass canopy and increased transpiration by greater native biomass (Mack and D Antonio 2003). A more detailed study of plant stressors and litter nutrient and C compounds may be required to conclusively identify the mechanisms responsible for the increased decomposability of native litter produced in the presence of grasses. Biological Invasions and Ecosystem Processes Our study shows that the indirect effects of a plant species or plant growth form on an ecosystem process may be both predictable and surprising. Based on our previous study of growth (D Antonio and others 1998), we correctly predicted that grasses would reduce the amount of fine litterfall produced by native species. The simplicity of this observation makes it likely that this generalization can be applied to other systems. We were surprised, however, to find that an interaction between grasses and natives actually increased decomposition of native litter, despite a reduction in litterfall. Although it is unclear whether the mechanism underlying this interaction operates in other ecosystems, it is certain that plant interactions do occur in most ecosystems; our study shows that these interactions have the potential to impact decomposition. Hypotheses regarding the direct effects of plant species on ecosystem processes have dominated ecosystem-level research on the impacts of exotic species on native ecosystems. As with our study, these hypotheses have not identified exoticness per se but have focused on linking plant structural and functional traits with ecosystem processes. There are many examples of invasive species that have large and obvious effects on certain ecosystem processes, such as resource inputs via N fixation (Vitousek and others 1987; Stock and others 1995), water cycling (Holmes and Rice 1996; Le Maitre and others 1996; Caldwell and others 1998), and disturbance regimes (Vitousek 1986; D Antonio and Vitousek 1992; Mack and D Antonio 1998). Most work on the effects on exotic species on litter decomposition and N dynamics has been with N-fixers, which clearly increase N availability relative to that of natives (Vitousek and Walker 1989; Musil