Leaf Litter Samples: Recently abscised D. excelsa leaflets for each year were collected immediately prior to each study period from the forest floor near the El Verde field station (350 m elev.). Recently abscised C. racemiflora leaves were collected at or near the Icacos site for each year and also about 1.5 km south of the El Verde field station in 1990 (575 m elevation). Fresh (green) P. montana foliage was used due to the inconsistent and gradual way the species replaces fronds. Fronds were collected near the Icacos site, leaflets were stripped from the fronds and cut into 15-20 cm segments. The rhachi of the fronds were discarded. Foliage for the 1989 study was stored as collected for two days before being weighed. To improve our estimates of initial dry mass, the foliage collected for the 1990 study was air-dried for three days before being weighed. An unknown amount of moisture was lost from litter during storage in 1990. Foliage of a single species was weighed and placed in 20 cm by 30 cm litterbags (~1 mm fiberglass mesh). A sample of about one per six litterbags (randomly stratified across the processing interval) was retained and dried to later estimate the initial moisture, nitrogen, lignin, cellulose, and ash content of treatment samples. Between 8 and 14 g (oven-dry equivalent, Table 1) of foliage were placed in each litterbag, although the dry-mass per litterbag varied by less than one g for a given species and study year. Sampling Design: Plots, Sub-plots, and Litter Distribution: Two plots (each 8 x 16 m) were established at each of the Icacos and Bisley study sites in 1989 allowing us to make comparisons of decomposition among riparian and upland plots within watersheds and between watersheds. Similar plots were established in 1990 (riparian and upland, Bisley and Icacos). Since the Icacos upland plot (North side of stream) was only minimally disturbed (defoliated), a fifth plot, hereafter referred to as 1990 Icacos Upland Blowdown plot (data variable designated as South side), was established in an adjacent area that was severely disturbed. To summarize: 1989: (2 watersheds x 2 slope positions) = 4 plots 1990: (2 watersheds x 2 slope positions) + 1 blowdown plot = 5 plots Each 8 x 16 m plot was divided into a grid of 2 m by 2 m sub-plots. With three species and five replicates per species to distribute, fifteen sub-plots were randomly selected from the thirty-two total sub-plots. Assignment of species per sub-plot was also randomly selected. Litterbags were placed on the forest floor within the sub-plots and tethered to an anchor. The 1989 samples were distributed on 2 June and the 1990 samples were distributed on 7 July (Bisley) and 10 July (Icacos). A single litterbag was collected from each sub-plot after: i) 14, 35, 52 (Bisley), 55 (Icacos), and 78 days in 1989, ii) 12, 23, 45, 73, and 102 days at Icacos plots in 1990, and iii) 11, 22, 42, 69, and 99 days at Bisley plots in 1990. Samples were collected more frequently at the beginning of each study period to try and account for the rapid changes that occur in litter during this period. To summarize: 1989: (3 sp) x (4 plots) x (5 replicates) x (4 collections) = 240 litterbags 1990: (3 sp) x (5 plots) x (5 replicates.) x (5 collections) = 375 litterbags Mass and Quality Analyses: Foliage samples were oven-dried on the day of collection (~50°C) , shipped to Durham, New Hampshire, redried (65°C), removed from the litterbag, separated from loose soil, weighed, and milled to pass a 1 mm sieve. The ash content of each sample was estimated by combustion of a ~1 g sub-sample (500°C for 6 hours). Estimates of ash content were used to adjust the biomass and nitrogen values for soil contamination, therefore disappearance rates were calculated on an ash-free basis. Total nitrogen was estimated using a micro-Kjeldahl digestion method and Technicon Autoanalyzer (Bremner and Mulvaney 1973, Technicon 1983). Data Analysis: The fraction of the initial biomass remaining (FBMR) and the fraction of initial nitrogen remaining (FNR) were examined using several approaches. The decomposition rate is often described by the model: y = e-kt, where y = FBMR, t is elapsed time in years, and k is a litter-specific constant. We used simple least-squares regression [ln(FBMR) = -kt] to fit a model to our data for comparison of results of this study to other studies. However, application of this general model across all litter types and species has often been found to underestimate initial mass loss. Models that factor the variable mobility of litter constituents and the multi-stage nature of decomposition better describe mass loss. Forcing the y-intercept to zero as in the y= e-kt model can assert an undesirable leverage when modeling and other statistical procedures are applied. This artificial affect is particularly influential when applied to a study focusing on short-term decomposition dynamics. For example, immigration and incorporation of decomposer organisms may increase litter mass or biomass loss through leaching may not occur immediately upon placement on the forest floor. The single exponent model described above was modified to test for differences among species, site, and year categories. The rate component of the modified model was renamed k¢ and the model included a y-intercept term: ln(FBMR) = -k¢t + yint The y-intercept in this model may be viewed as an index of the initial leaching losses or delay in mass loss that may occur. The slope (k¢) may be viewed as the second stage in the multi-stage process of decomposition. We used regression techniques to fit this model with category indicator variables to simultaneously and efficiently test for differences in biomass loss among categories. Binary indicator variables (1 or 0) were factored for each category (e.g. year, watershed, slope position) along with elapsed time such that the influence of the category was reflected in a deviation in the rate and a deviation in the y-intercept. The pattern of the fraction of nitrogen remaining over time was curvilinear and the degree and shape of curvilinearity varied among sample groups. No single transformation technique would linearize the data for analyses using regression. Consequently, non-parametric Kruskal-Wallis rank-sum tests were used to compare nitrogen response over time and among sites for each study year. Statistical comparisons between study years were complicated by unequal sampling intervals. To address this, the data were divided into groups with year, species, and treatment interval in common. The fraction of nitrogen remaining at a given sampling time in 1989 was compared to the closest two sampling times (before and after) in 1990. For example, the 35 day samples from 1989 were compared to both 23 and 44 day samples from 1990.
