1The leaf litter of tropical forests serves as a carbon sink, slows soil erosion and leaching, and is home to a large fraction of the forest's biodiversity. Standing leaf litter reflects the balance of litterfall and decomposition; both rates may be limited by element availability. We propose a mechanism for this regulation that assumes phosphorus limits metabolism in tropical soils, and that phosphorus limitation is more pronounced in faster growing organisms.
2Leaf litter depth varied 16-fold (1·4–22·4 cm) across 28 forest stands in Panama and Peru and was deeper on sand vs. clay soils. Of five elements tested (N, P, K, Mg and Ca), the concentration of P in decomposing litter best predicted litter depth (r2 = 0·76, C : P1·90). This relationship broke down in the most impoverished sandy soils.
3These data are consistent with the hypothesis that the weathering of tropical soils limits the metabolism of microbes first and trees second, with decomposition and litterfall co-limiting litter depth in ecosystems with the least available phosphorus. This has implications for the dynamics of weathering: nutrient leaching may be regulated through negative feedback if deeper litter buffers soil from rainfall.
Litter depth reflects the balance of litterfall and decomposition. At the global scale, both rates increase with the co-availability of water and solar energy (Rosenzweig 1968; Gholz et al. 2000; Allen, Gillooly & Brown 2005). Within biomes of similar precipitation and temperature, nutrient limitation of litterfall and decomposition can map onto gradients of biogeochemistry (Vitousek 1984; Sterner & Elser 2002). Two such gradients include soil age and type. Younger soils are relatively rich in rock-born P, K, Mg, but can be poor in biologically fixed N. Weathering, in turn, depletes rock-born elements even as N-fixing symbioses accumulate (Walker & Syers 1976; Wardle, Walker. & Bardgett 2004b). Likewise, soil of different origins can also vary in nutrient availability. Much of northern Amazonia is a mosaic of red clay oxisols and ultisols and oligotrophic white sands (Cuevas & Medina 1988). These sands may have one-tenth of the soil nutrients (Anderson 1981) and understorey (Gentry & Emmons 1987) of clay soils.
Phosphorus, given its dearth in old tropical soils and relative abundance in the cells of plants and consumers (Sterner & Elser 2002) is a strong candidate to constrain rates of litterfall and decomposition. If it constrains both equally, then the effect of phosphorus gradients on litter depth should be nil. Elser et al. (1996), however, suggest this may not be the case if taxa with higher growth rates (and thus higher titers of P-rich ribosomes) generate a higher per-capita demand for phosphorus. Bacteria and fungal decomposers generally grow faster than trees (Peters 1983). In the Growth Rate Hypothesis (Elser et al. 1996) microbial decomposition should thus be limited by P at higher levels of availability, all else being equal, than arboreal litterfall.
There is some evidence for a discrepancy in phosphorus effects on decomposition and litterfall. In montane Hawai’i forests, decomposition was frequently enhanced on +P plots after 3 years (Hobbie & Vitousek 2000; Vitousek 2004) as it was after 6 years in a lowland Panama moist forest (Kaspari et al. 2008). A third experiment in a Costa Rican wet forest found no effect of P on leaf mass loss when the site was pre-fertilized for 1 year (Cleveland, Reed & Townsend 2006).
There is less evidence for P limitation of litterfall. In Vitousek's (1984) comparative study of 55 sites, a small positive correlation between litter P and litterfall rate arose only in the most P-impoverished litters (0·02–0·04%, see his Fig. 2). When P was use to fertilize tropical forests, litterfall failed to increase after 6 years in a relatively P-rich Panama forest (Kaspari et al. 2008). Similarly, it failed to increase after 4 years (Tanner, Kapos & Franco 1992), and 2 years (Vitousek et al. 1995) in two montane tropical forests.
Here we combine the hypotheses of Walker & Syers (1976: P limitation is widespread in old tropical forests) and Elser et al. (1996: P limitation expresses itself in decomposer microbes at higher levels of availability than trees). This produces the working hypothesis that as forest ecosystems lose phosphorus, standing crops of litter should increase. Furthermore, the rate of this increase should decline when P becomes sufficiently scarce to also limit litterfall. Here we evaluate this prediction with a geographical survey of litter chemistry and depth. We sample younger and older red clay soils (10 from the isthmus of Panama and 9 from Amazonian Peru) and 9 from Amazonian white sands of Peru. We compare C : P's ability to predict litter depth compared with C : X ratios of four other elements (N, K, Ca, Mg) that have been linked in other studies to decomposition and litterfall (Tanner et al. 1992; Vitousek et al. 1995; Hobbie & Vitousek 2000; Berg & Laskowski 2006; Hobbie et al. 2006). We show that gradients in the C : P of decomposing litter account for 76% of the variance in standing litter depth, and that this effect diminishes in high C : P sandy soils.
Materials and methods
Litter depth and chemistry were sampled from 28 lowland forests stands in Panama and Peru. We sampled three types of lowland forest. The Panama clays came from 10 stands on Barro Colorado nature monument (BCNM) in the Republic of Panama (9°06′N, 79°50′W). The species composition and stature of these forests are characteristic of a high canopy seasonal forest with c. 2400 mm annual rainfall (Leigh, Rand & Windsor 1996). A previous survey of BCNM (Knight 1975) was used to guide selection of most stands on eight clay and two gley soils suggested by Knight (1975) to show a range of soil fertility. Stands were sampled June and July 2004 near the beginning of the wet season. The minimum distance between stands was c. 0·5 km.
At each of the 28 forest stands we collected leaf litter from five sampling quadrats spaced 5 m apart on a 25 m transect. This was part of a larger project studying the trophic ecology of brown food webs (Kaspari and Yanoviak, unpublished data). The starting point for each transect was selected haphazardly, and all sampling points were > 50 m from a forest edge. Litter depth was measured in each corner of a PVC frame by inserting a surveyor's flag marked in 1-cm increments down to mineral soil. Leaf litter (excluding woody material > 4 cm diameter) inside the frame was collected by hand down to mineral soil. This litter was shaken over a 1 cm mesh for 30 s, generating a fine litter siftate (Agosti et al. 2000). Since litter was deeper in Peru, and to standardize siftate quantity, Peru quadrats were smaller (0·125 m2 vs. 0·25 m2 in Panama). Tullgren funnels (40 W incandescent bulb, 48 h) were used to separate arthropods from the siftate (Agosti et al. 2000). Once arthropods had been extracted, a 10 g subsample of the dried litter was analyzed for chemical composition at the Oklahoma State Soil, Water, and Forage Analytical Laboratory (OSU 2006). Briefly, nitrogen (as NO3) was extracted with calcium sulfate and measured on a Lachat™ flow injection autoanalyzer. Potassium, phosphorus, calcium and magnesium were all first extracted using Mehlich 3 (Mehlich 1984) and quantified with a Spectro CirOs™ ICP spectrometer (for P, the solution was additionally mixed with ascorbic acid colour complex and analyzed at 880 nm). Neutral detergent fibre (the percentage of lignin, cellulose and hemicellulose) was measured using an Ankom™ fibre analyzer. See (OSU 2006) for further details. C : X ratios were calculated based on mass. One Peru sand and clay forest sample was lost before analysis.
We first quantified the effects of geography and soil type comparing Panama with Peru, and clay and sandy soils within Peru using an anova. We tested for the predicted increase in litter depth by regressing log10 litter depth against log10 C : P (power law models accounted for 11% more variation than arithmetic models). We evaluated C : P as the best predictor using multiple regression (P < 0·05 for variable entry or exclusion Draper & Smith 1981), in which C : N, C : P, C : K, C : Ca and C : Mg were available for model selection. Finally, as sandy soils are more impoverished than clay soils (Coleman & Crossley 1996) we used ancova (on log10-transformed variables) to compare the predictive ability of C : P for litter of clay vs. sandy soils. We used sas 9·1 (SAS 2006) in all cases to perform these analyses.
Litter chemistry and depth differed across soils and regions (Fig. 1). Among the 140 individual sample plots, litter depth ranged from 0 to 25·5 cm. Among the 28 forest stands, litter depth varied 16-fold (1·37–22·4 cm), increased significantly from Panama to Peru (F1,26 = 77, P < 0·0001, Fig. 1), and was twice as deep on sandy vs. clay soils within Peru (F1,16 = 36, P < 0·0001, Fig. 1). The deeper litter on sandy soils had higher concentrations of celluloses and lignin (neutral detergent fibre: 72% on sand vs. 65% and 64% on Peru and Panama clay, F2,23 = 9·0, P = 0·0013).
The nutrient makeup of decomposing litter differed regionally (Fig. 1). C : P varied in a manner similar to litter depth (F2,23 = 32, P < 0·0001) with Peru clays 50% higher in C : P than Panama clays, and Peru sands 50% higher than Peru clays. C : K, in contrast, was twice as high in Peru sands than in either clay site (F2,23 = 31, P < 0·0001), as was C : Mg (F2,23 = 27, P < 0·0001). C : N ratios were highest on Panama clay and Peru sands (F2,23 = 7·4, P = 0·0033). C : Ca failed to vary significantly across the region-soil combinations (F2,23 = 2·5, P = 0·11).
Stepwise multiple regression selected C : P as the best predictor of litter depth across the 26 forest stands (x = 0·00002y1·90, r2 = 0·76, F1,24 = 77·2). An additional 5% of variation in litter depth was accounted for by the positive effect of C : Ca. This produced a model that accounted for 81% of the variation in litter depth (y = 0·00052E−4 (C : P1·27 + C : Ca0·25), C : P F1,23 = 77·2, P < 0·0001; C : Ca F1,23 = 5·73, P = 0·025).
The exponent of the relationship between C : P and litter depth differed between clay and sandy soils, however (Fig. 2). Litter depth increased with C : P (F1,22 = 7·19, P = 0·014) and was higher on sandy soils (F1,22 = 5·17, P = 0·033) although the litter-phosphorus relationship varied with soil type F1,22 = 4·76, P = 0·040). Clay soils, where C : P ranged from 313 to 830 yielded an exponent of 1·8 (F1,16 = 23·7, P = 0·0002) with C : P accounting for 60% of the variation in litter depth. Sandy soils, in contrast, where C : P ranged from 710–1228, failed to show a significant increase in litter depth along this gradient (F1,7 = 0·15, P = 0·71).
Understanding the factors underlying rates of production and decomposition remains a key goal of ecology. Litter depth is an easily measured variable that integrates across both processes. Here we combine theory and data to provide a simple hypothesis on the role of phosphorus in shaping the distribution of litter depth across tropical forests. The survey of 26 forest stands supports a key prediction: along a gradient of decreasing phosphorus on tropical soils, the rate of microbial decomposition is constrained before the rate of arboreal litterfall. This results in a biphasic curve linking litter depth to C : P of decomposing litter (see also Koide & Shumway 2000 for a similar increase phase across four temperate pine plantations). These data reinforce the hypothesis that P regulates the carbon cycle at a regional scale (McGroddy et al. 2004; Reich & Oleksyn 2004).
Any geographical analysis comes with caveats. First, these forest stands are not a random sample from the Neotropics, but are centred in two regions. However, nutrient availability varies at mesoscales in tropical forests (John et al. 2007). The forest stands sampled here represent c. twofold gradients of C : P for Panama clay (313–637), Peru clay (442–830) and Peru sand (710–1228). Litter depth also varied commensurately (5·6-, 3·7- and 2·6-fold respectively). Second, BCNM is more seasonal than Iquitos in climate, and likely, litterfall (Leigh et al. 1996). We sampled BCNM 1 -2 months after the onset of the wet season (when litter is at it's deepest, Leigh et al. 1996), and 4 months before the end of the wet season (when litter is most shallow). None-the-less, we do not interpret the exponent of the regression in Fig. 2 beyond the fact that it was positive, as predicted. Note, however, that the Peru forest stands of similar seasonality still generate an increase in litter depth with C : P (y = 0·0016*C : P1·3, F1,14 = 18·6, P = 0·0007, r2 = 0·57). Finally, low nutrient environments are predicted to favour plants with C-based defences (Coley, Bryant & Chapin 1985; Orians & Milewski 2007). White sand forests are characterized by sclerophyll leaves resistant to herbivory (Fine et al. 2004), and presumably, decomposition. We find higher levels of fibre in the litter of white sand forest stands, and average litter fibre content correlates with C : P (r = 0·63, P = 0·0004). This suggests that decreases in leaf quality may also contribute to the accumulation of litter shown here. Fertilization experiments in white sand forests (testing for the predicted increase in decomposition and litterfall rates) and common garden decomposition experiments (e.g. Hobbie & Vitousek 2002) are logical next steps toward evaluating these hypotheses.
The 16-fold variation in litter depth across forests of similar rainfall, and the role of P in generating this variation, has two implications for the dynamics of tropical forests. First, decomposition reflects the rate at which litter is converted into microbial biomass. This biomass, in turn, feeds the microbivores, predators, and top predators of the brown food web (Heneghan et al. 1998; Moore et al. 2003; Wardle et al. 2004a; Fitter et al. 2005). At the same time, decomposition destroys the physical habitat in which these organisms live. This ‘more food, less habitat’ dynamic of brown food webs (Koide & Shumway 2000; Scheu & Setala 2002) differs fundamentally from that of green food webs, where high productivity is associated with more food, more habitat, and higher abundance of consumers (e.g. Kaspari, Alonso & O'Donnell 2000). If food and habitat co-limit the populations of the brown food web, then abundance should show a nonlinear, and perhaps unimodal, relationship with litter depth.
Second, our results suggest a process by which mineral leaching in tropical forests is progressively slowed. Leaf litter slows soil erosion and mineral leaching (see review in Sayer 2005), particularly in tropical ecosystems where rainfall is measured in meters. As these warm wet ecosystems lose phosphorus and other mineral nutrients, a built-in negative feedback mechanism may slow further leaching: a thicker blanket of leaf litter.
This project was funded by the National Science Foundation under Grant No. 0212386 to M.K. Special thanks to S Bagwell, and J Morrison for assistance in the laboratory, to Y. Milton, W. Sanchez, E. Requena, and C. Valderrama for assistance in the field, and to the numerous private landowners who provided access to forest stands in Peru. Discussions with Brad Stevenson, Sarah Hobbie, Jennifer Powers and Peter Tiffin provided many insights and greatly increased the clarity of our thinking. The Smithsonian Tropical Research Institute, the Panamanian Autoridad Nacional del Ambiente (ANAM), and the Peruvian Instituto Nacional de Recursos Naturales (INRENA) provided permits.