Ecosystem structure and productivity of tropical rain forests along altitudinal gradients with contrasting soil phosphorus pools on Mount Kinabalu, Borneo


  • Kanehiro Kitayama,

    1. The Japanese Forestry and Forest Products Research Institute, PO Box 16, Tsukuba Norin Kenkyu Danchi, Ibaraki 305–8687, Japan, and
    Search for more papers by this author
    • Present address and correspondence: K. Kitayama, Center for Ecological Research, Kyoto University, 509–3 Ohtsuka, Kamitanakami Hirano-cho, Ohtsu, Shiga 520–2113, Japan (fax + 81 77 5498201; e-mail

  • Shin-Ichiro Aiba

    1. Faculty of Science, Kagoshima University, Korimoto, Kagoshima 890–0065, Japan
    Search for more papers by this author


  • 1We measured above-ground net primary productivity (ANPP) and ecosystem structure and processes in eight rain forest stands at four elevations (700, 1700, 2700 and 3100 m) and on two geological substrates (sedimentary vs. ultrabasic rock) on Mount Kinabalu, Borneo.
  • 2All ultrabasic sites had smaller pools of total soil phosphorus (P) and of labile inorganic P than did the sedimentary sites at the same altitudes. We predicted that the magnitude of altitudinal changes in ANPP would be less on ultrabasic than on sedimentary substrates, reflecting lower temperature dependency of ANPP under stronger P limitation.
  • 3Although ANPP declined with increasing altitude on both substrates, the slopes of the two regression lines were similar. The intercept was, however, marginally greater on sedimentary than on ultrabasic substrate.
  • 4Stand-level nutrient-use efficiencies (the ratio of litterfall mass to nutrient return) for N and P were only affected by altitude on ultrabasic substrate where they increased exponentially. Mean foliar N and P contents per unit leaf area of the canopy species increased with altitude on both substrates, but differed between substrates at the same altitude only for P (lower on ultrabasic).
  • 5Leaf area index (LAI) decreased upslope on both substrates. We assumed that half of primary production was allocated below-ground in order to evaluate stand level net assimilation rate (NAR). This was nearly constant on sedimentary substrate, but declined linearly with increasing altitude on ultrabasic substrate, where it may have to be added to LAI to explain ANPP patterns.
  • 6We suggest that on sedimentary substrate trees may be able to maintain NAR under colder environments by increasing foliar N and P per unit leaf area, but P deficiency prevents them from adjusting on ultrabasic substrate.


Tropical rain forests decrease in stature and above-ground biomass with increasing elevation (Whitmore 1984; Stadtmüller 1987). Above-ground net primary productivity (ANPP) also diminishes upslope (Weaver & Murphy 1990; Raich et al. 1997; Waide et al. 1998). Short-statured montane cloud forests occur at middle altitudes as a result of one or more of saturated soils (limiting root respiration), radiation attenuation by clouds, strong winds, high concentrations of phenolic compounds in soil organic matter, limited nutrient uptake (due to diminished transpiration), and limited nutrient supply associated with slow nutrient mineralization (Grubb 1977; Bruijnzeel & Proctor 1995; Bruijnzeel & Veneklaas 1998). No single factor can consistently explain altitudinal variation in this situation, but ground-level cloud is always a feature. Additional mechanisms are needed to explain why rain forests continue to decrease in stature (and ANPP) above the middle-slope cloud zone (Kitayama 1992).

Tanner et al. (1998) found that reported foliar and litterfall nutrients – particularly nitrogen (N) and to a lesser degree phosphorus (P) and potassium (K) – decreased with increasing elevation and suggested that upland tropical rain forests were constrained by low nutrient supply. In-situ fertilization experiments in montane forests have demonstrated that growth or litter production is limited by N or N and P (e.g. Tanner et al. 1992; Herbert & Fownes 1995; Raich et al. 1996; Vitousek & Farrington 1997) but comparable data are lacking from lowland forests (Vitousek 1998). Furthermore, it remains unclear how low concentrations of foliar nutrients diminish stand-level ANPP (Tanner et al. 1998). The major aim of our study was therefore to investigate the pattern of ANPP on a single mountain slope extending from the lowland to near the forest limit, relating any changes to stand structure, soil nutrient supply and foliar characteristics.

Raich et al. (1997) demonstrated that altitudinal gradients in ANPP on wet slopes of Mauna Loa, Hawaii, were correlated with temperature as well as with changes in soil nutrients. Although mountain slopes represent gradients of several interdependent factors, manipulation of temperature in tropical rain forests is not practical, and studying the control of ecosystem processes across an altitudinal gradient may be useful for inferring temperature effects.

Temperature may control plant physiology directly (Johnson & Thornley 1985) or indirectly via soil nutrient availability (Chapin 1983; Hobbie 1996; Raich et al. 1997; Hobbie & Chapin 1998). The availability of soil nutrients that are supplied through mineralization of organic matter is strongly temperature dependent (e.g. Myers 1975; Ellert & Bettany 1992; Kirschbaum 1995). Complex interactions between temperature and nutrients are indicated by a possible feedback between the quality of organic matter produced and the supply of nutrients (Hobbie 1996). Temperature manipulation in tussock tundra demonstrated that ANPP was constrained indirectly through soil processes (Hobbie & Chapin 1998).

Two different geological substrates (i.e. sedimentary vs. ultrabasic) occur in close proximity at a range of altitudes on Kinabalu, Borneo, thus enabling an investigation of temperature and substrate interactions. Soils derived from ultrabasic substrate were much poorer in P (and N) than those of sedimentary substrate at the same altitudes (Kitayama et al. 1998b; Aiba & Kitayama 1999; Kitayama et al. 2000). Impoverishment of bioavailable P is common in highly weathered lowland rain forest soils (Vitousek & Sanford 1986; Tiessen et al. 1994) and in aged soils elsewhere (Walker & Syers 1976; Lajtha & Schlesinger 1988; Crews et al. 1995). In such soils, P deficiency is suggested to limit carboxylation in photosynthesis and community-level ANPP.

Specifically, we consider whether: (i) ecosystem processes such as ANPP and decomposition are slower under more severe nutrient limitation (i.e. on ultrabasic than on sedimentary substrate at the same altitude); (ii) such processes are slower at cooler (higher) than at warmer (lower) altitudes on both substrates; and (iii) the magnitude of any altitudinal change is greater on sedimentary than on ultrabasic substrates because the temperature dependency of ecosystem processes is greater under relaxed nutrient limitation.

Materials and methods


Mount Kinabalu (4095 m, 6°5′ N, 116°33′ E) is non-volcanic, and is the highest mountain in SE Asia between the Himalayas and New Guinea. The main massif and adjacent areas are protected as the Kinabalu Park. Diverse pristine evergreen rain forests occur continuously from 300 m to the forest limit at 3700 m (Kitayama 1992). The climate is humid tropical with weak influences of the Asiatic monsoon. Climate has been monitored at 550, 1560, 2650 and 3270 m since 1995 (Kitayama et al. 1999). Figure 1 demonstrates climatic conditions during the study period (1996–97). Mean annual air temperatures decreased linearly with increasing altitude, with a mean lapse rate of 0.0055 °C m−1 as in previous reports (Kitayama 1992), with minimal month-to-month variation (generally < 2 °C). Mean atmospheric saturation deficits became less negative upslope as a function of decreasing air temperature. Mean annual rainfall at all sites (approximately 2300 mm year−1) was close to the 1975–83 mean at Park Headquarters (1560 m) (Kitayama 1992). Measurements during 1996–99 showed that although rainfall varied greatly from year to year (1800–3300 mm) it did not vary much among sites within years. Total photosynthetically active radiation (PAR) was reduced at mid-altitude (particularly 2650 m) due to clouds. Overall, only mean air temperature showed effects of altitude.

Figure 1.

Climate profile of Mt Kinabalu during 1996–97: (a) mean air temperatures (AT filled circles) and mean atmospheric saturation deficits (SD open circles); (b) mean annual rainfall; (c) mean annual total photosynthetically active radiation (400–700 nm, PAR), with lower and upper portions representing instantaneous PAR between 50 and 600 µmol m−2 s−1 and above 600 µmol m−2 s−1, respectively. Weaker PAR (< 50 µmol m−2 s−1), which is unlikely to contribute to net photosynthesis, was excluded. Instantaneous PAR was measured at 10-s intervals.

Below 3000 m, large areas of tertiary sedimentary substrate are interspersed with mosaics of ultrabasic rock (Jacobson 1970). Granite dominates above c. 3000 m. We established a matrix of eight study sites at 700, 1700, 2600 and 3100 m on each substrate (sedimentary vs. ultrabasic rocks) (Table 1). All eight sites were on gentle slopes to avoid effects of relief. Difference in elevation between paired sites was minimized but was up to 300 m (Table 1), corresponding to a temperature difference of 1.65 °C. The sedimentary rock consists largely of sandstone and mudstone of the Eocene Trusmadi Formation and the ultrabasic rock consists largely of serpentinized peridotite (Jacobson 1970). Peridotite is remarkably high in magnesium (Mg) and very low in P (Jacobson 1970). The composition of the sedimentary rock is unknown but such substrates generally contain > 50% quartz (Jacobson 1970). The ‘sedimentary’ site at 3100 m is actually underlain by granitic rock, but on the basis of the abundance of quartz we have classified this site as sedimentary rather than as ultrabasic.

Table 1.  Description of the eight study sites on Mount Kinabalu, Borneo. Canopy height data are after Aiba and Kitayama (1999). All sites correspond to those of Aiba and Kitayama (1999)
 Common altitude (m)Actual altitude (m)Plot area (ha)Slope (°)Canopy height (m)Forest type
  • *

    Actually underlain by granitic rock.

Sites on sedimentary rock 700 6501.001946.8Hill dipterocarp rain forest
 170015600.501730.0Lower montane rain forest
 270025900.252020.6Upper montane rain forest
 3100*30800.202715.0Subalpine forest
Sites on ultrabasic rock 700 7001.001165.4Hill dipterocarp rain forest
 170018600.202422.6Lower montane rain forest
 270027000.202214.2Upper montane rain forest
 310030500.0619 6.1Subalpine forest/scrub

The structure and floristics of the rain forests at these sites change considerably with altitude and substrate (Aiba & Kitayama 1999). However, the same plant life forms (i.e. evergreen broad-leaved trees) occur throughout. The family Dipterocarpaceae dominates the 700 m sites, while Myrtaceae and Podocarpaceae dominate at higher altitudes on both substrates.


At each site, we established a permanent rectangular plot. At upland sites, plot area was expanded until newly added canopy tree species became rare (Aiba & Kitayama 1999; see Table 1 for between site variation). At 700 m, plot area was arbitrarily set at 1 ha. Each plot was subdivided into 10 × 10 m subplots.

At the first census (September 1995 to February 1996), diameter at breast height (d.b.h., 1.3 m above the ground) was measured for trees ≥ 4.8 cm d.b.h. (15 cm girth) in plots at the six upland sites. At 700 m, all trees ≥ 10 cm d.b.h. were measured and two 10 × 100 m transects within each of the 1 ha plots were censused for trees ≥ 4.8 cm d.b.h. Girth was measured to the nearest mm at the closest point to breast height without any stem irregularities, and the position permanently marked with spray paint and an aluminium tag. The heights of trees were measured in a 10-m wide transect of varying length to include at least 90 trees per site. All marked trees were measured approximately 1.8 years later.

We used a simplified version of the rectangular hyperbola of Ogawa (1969) to fit the relationship between d.b.h. (D cm) and tree height (H m):

1/H = 1/AD + 1/H*

where A (m/cm) and H* (m) are regression constants. A is related to how densely plant biomass is packed in a unit volume and is therefore called the packing factor, whereas H* represents the asymptotic tree height. Our estimates of H* were close to the observed maximum height among the eight plots (r = 0.97, P < 0.001). We then calculated the standing above-ground biomass (AGB, kg m−2) using the allometric relationships obtained by destructive sampling of a lowland dipterocarp forest in West Kalimantan (Yamakura et al. 1986):

Ws = 0.02903(D2H)0.9813
Wb = 0.1192Ws1.059


Wl = 0.09146(Ws + Wb)0.7266

where Ws, Wb and Wl (kg) are the dry mass of the trunk stem, branches and leaves of a tree, respectively. AGB is the sum of Ws, Wb and Wl. These equations reasonably estimated AGB of a heath forest in Cambodia (Hozumi et al. 1969) and a temperate rain forest in Japan (Aiba & Kitayama 1999), as well as the stem biomass of a montane forest in Java (Yamada 1997), that were structurally and floristically similar to the montane forest of Kinabalu. The equations overestimated AGB of a New Guinean montane forest but this is probably due to a lack of tree species with heavy wood (Edwards & Grubb 1977) and does not invalidate their use here. We also calculated AGB density (kg m−3), defined as AGB/H* (cf. Kira & Shidei 1967). We estimated the amount of carbon (C) in live AGB as AGB/2.

We optically measured leaf area index (LAI) of each plot using an LAI-2000 plant canopy analyser (LI-COR, Nebraska, USA). We took 20–30 readings systematically at 2 m above the ground at the corners of 10 × 10 m subplots in September 1997 and again in February 1998, and calculated mean values. These optically measured values (LAIopt) were later compared with values estimated from allometries (LAIall): LAIall= Wl/LMA, where LMA (leaf mass per area) was determined for the dominant broad-leaved species at each plot (see below). Linear regression between LAIopt and LAIall values among the eight sites was highly significant, showing a slope of 1.00: LAIopt = 1.25 + 1.00 × LAIall (r2 = 0.80, P = 0.003). We therefore used mean LAIopt to represent stand-level LAI (cf. Harrington & Fownes 1995). The intercept (1.25) may reflect the LAI contributed by trees < 4.8 cm d.b.h.

We placed 0.5 m2 area litter traps, made of 1 mm mesh, 1 m above the ground at 10-m intervals along two transect lines at each plot (20 traps plot−1 at two lower altitudes, 10 traps at the higher site). Trapped fine litter was collected bi-weekly on the same day from all plots, and immediately oven-dried at 70 °C for at least 3 days before sorting by trap into leaves (including rachises of compound leaves), reproductive organs (fruits and flowers including pedicels and peduncles), twigs (≤ 2 cm girth, including bark), epiphytes (all parts combined), palms and bamboos (all parts combined) and dust (unidentifiable particles that pass through 2 mm mesh including insect detritus).

ANPP was calculated as the AGB increment of surviving trees between the first and second censuses plus fine litterfall. No attempt was made to estimate large detritus such as fallen branches > 2 cm girth and ANPP will therefore be a slight underestimate. Fine litterfall (g m−2 year−1) was determined as a mean of 2 years from February 1996 to February 1998. Details of the litterfall dynamics will be described elsewhere but rainfall fluctuations were associated with litterfall dynamics (Kitayama et al. 1998a).

Stand-level net primary productivity (NPP) is determined by two attributes:


where NAR is stand-level net assimilation rate (i.e. photosynthetic performance per unit of leaf area; Harrington et al. 1995). Given that only a proportion R of NPP is allocated above-ground:


As a first approximation, we assumed that R was constant across sites and set its value to 0.5.

In June 1996, fine roots were collected at each site with a 30-cm deep core sampler (diameter 37 mm) at 10 points at 10-m intervals along a transect started from a random point. Cores were subdivided into 0–5, 5–15 and 15–30 cm portions, gently rinsed with water to remove soil, and all live roots < 2 mm diameter were manually collected. Roots with dark coloured tissues and rotten bark were considered dead and were discarded. Roots were oven-dried at 70 °C for 3 days and dry weights were measured. Effects of depth and altitude on fine-root mass were analysed with one-way anova followed by a comparison of means using Tukey’s HSD at P = 0.05.


Net soil N transformation rates

Net soil N mineralization and nitrification rates were determined using the buried bag method in May 1995 (data reported in Kitayama et al. 1998b), except for the 1700 m sedimentary site analysed in February 1996. Four composite samples (each made up of 10 15-cm deep soil cores) from each site were incubated in situ for 10 days without the amendment of soil moisture. Subsamples were extracted before and after the incubation with 1.5 N KCl solution and NO3- and NH4-N determined colourimetrically. Net nitrification rate was defined as final minus initial NO3-N and net mineralization rate as final minus initial (NO3 + NH4)-N.

A portion of each composite was oven-dried at 105 °C for 48 h to determine gravimetric water content. Concentrations of exchangeable Ca and Mg were determined by atomic absorption spectrometry on each of the ‘initial’ subsamples. Organic C was determined on fresh soil by the Walkley-Black wet digestion method (Nelson & Sommers 1982). Total N was digested by the micro Kjeldahl procedure with concentrated sulphuric acid, and determined colourimetrically. Soil pH was measured on a 1 : 1 fresh soil to deionized water solution.

Total soil organic C

We excavated four soil pits down to saprolite (to 1 m at the 700-m sites) under closed canopy at each site in July 1996. From each horizon, we collected a bulk soil sample for chemical analyses, and duplicate soil cores of known volume for bulk density. We combined the same amounts of fresh soils from corresponding horizons at each site, manually homogenized them, and removed stones and roots. The composite samples were analysed for organic C and total N as above and contents determined on an area basis (to saprolite or to 1 m when deeper), based on concentration and bulk density.

Soil P pools

Soluble P was extracted from N mineralization samples with hydrochloric-ammonium fluoride solution, and determined colourimetrically (Murphy & Riley 1962). Kitayama et al. (2000) determined soil P fractions for each horizon at our sites according to the sequential extraction of Tiessen & Moir (1993) and we used their data to calculate the total amount of active soil P that could potentially enter biological cycles, as total P minus recalcitrant occluded inorganic P.

Buried ion-exchange resin extraction of Pi and mineral N

Moist anion-cation mixed bed resin (8 g of Bio-Rad, AG 501-X8, 20–50 mesh) was placed in a PVC ring (40 mm diameter, 5 mm high) wrapped in acid-washed nylon stocking. Five replicate rings at each site (widely spaced under closed canopy) were buried horizontally at 5-cm depth, taking care not to disturb the soil structure, and left for 30–40 days. We repeated this procedure five times at regular intervals from December 1996 to October 1997 at different locations on each occasion. Collected resins were rinsed with ultra-pure water, air-dried for 7 days, weighed, and shaken with 40 mL of 2 m KCl in 0.1 m HCl for 30 min. Pi, NO3-N and NH4-N were determined as above and expressed on an air-dry resin weight basis. We calculated the mean of five replicates for each time at each site.


To allow for the seasonality of litterfall, dried litterfall samples were ground separately by fraction and time of collection at about 2-month intervals from April 1996 to February 1997 (six collection times). Mean concentrations of nutrients were calculated as the means of these six collection times.

Intact sun leaves were collected from canopy trees of the most dominant and widespread taxa across altitudes and substrates, i.e. representative from genera Shorea (Dipterocarpaceae), Syzygium (Myrtaceae), Tristaniopsis (Myrtaceae), Schima (Theaceae) and a few other dominants from each site. At each site, we shot one sunlit shoot each from the top canopies of one to five trees in each species using a slingshot in July to August 1996. We collected only fully expanded mature leaves from between the terminal and senescent whorls, and combined leaves by individual tree. In the laboratory, leaves were wiped, measured for area using a portable area meter (CID Inc., Vancouver, Canada) within 24 h, oven-dried at 70 °C for 3 days and finely ground. Leaf mass per area (LMA) was calculated as oven-dried mass divided by area.

Each powdered litter or foliar sample was weighed in duplicate, and 0.2 g was digested in a block digester with 4.4 mL 1 : 1 concentrated H2SO4 and 30% H2O2 with selenium as catalyst for 45 min. Digestion was repeated until the solution became clear. Digests were filtered through Whatman 2 V filter paper and made up to 50 mL with deionized water. The concentrations of N (as NH4-N) and P in the digests were determined colourimetrically using the same methods as for soil nutrients. The concentrations of Ca, Mg and K were determined by inductively coupled plasma emission spectrometry. An internal standard was mixed in every batch throughout to check the quality of analysis. Duplicate results were within 10% of the mean for each pair for all elements.

The annual fluxes of N, P, Ca, Mg and K that return to the soil through litterfall were determined as the annual amounts of litterfall multiplied by the mean concentration of each element. Following the method of Vitousek (1982), we calculated stand-level nutrient-use efficiency (NUE) as the ratio of annual litterfall mass to the annual content of the respective nutrient in the litterfall.


The net amount of PAR that was instantaneously absorbed by the canopy was estimated as the net PAR absorbed (Qnet) at the top canopy layer minus PAR on the forest floor at 2 m above the ground (Qin). Qin relative to the open-air PAR (Qout) was measured by a quantum meter synchronized to a reference quantum meter set in a nearby open place. Instantaneous Qin readings were taken at 5-m intervals along lines that divided subplots until 100 or more readings were obtained at each plot. Qin readings were compared with synchronous Qout readings and converted to relative values (Qin/Qout). Annual Qin was estimated based on instantaneous Qin/Qout values and the annual Qout obtained from the nearby climate station, assuming a linear relationship between Qin/Qout and Qout. Qnet was approximated based on an albedo value of 0.15 and the annual Qout. Subsequently, PAR use efficiencies (QUE, mg mol−1) of ANPP were calculated as ANPP (mg m−2 year−1) divided by annual PAR absorbed (mol m−2 year−1).


We estimated the decomposition rates of litter at each site using the mass balance method. We collected standing fine litter in 10 circular frames (each 44 cm diameter) at 10-m intervals along a line at each site in July and November 1996, and February, June and September 1997; these represented one dry, one wet and three intermediate months in terms of rainfall. All structured standing litter (twigs ≤ 2 cm girth, leaf and other fractions ≥ approximately 1 cm2) inside the frames was collected, oven-dried at 70 °C for 3 days and weighed. At each time, we recorded the locations of frames to avoid repeated collections from the same points. We calculated the mean of five collections to represent a mean standing litter crop at each site. The decay constant (k) was approximated as annual litterfall (g m−2 year−1) divided by mean annual standing litter crop (g m−2).


Effects of altitude were examined by linear regressions using the exact altitude (Table 1) as the independent variable on each substrate. We used means when samples were replicated within a site. When appropriate, coefficients of determination were compared for two regressions with and without logarithmic transformation of dependent variables to determine whether the altitudinal changes were linear or exponential. Differences in the slopes and intercepts of the regressed lines were tested between substrates by ancova.

Differences between altitudes on the same substrate (with multiple comparison) were also tested by anova whenever samples were replicated at each site, to examine non-linear trends across altitudes. In these cases, anova was also applied to detect differences between substrates at the same altitudes.



The concentrations of mineral nutrients and organic matter in topsoils (0–15 cm) varied with altitude and substrate (Table 2). The concentrations of organic C and total N were always greater on sedimentary than on ultrabasic substrates at comparable altitudes (except for 700 m for N). On sedimentary substrate, both increased with altitude to a peak value at 2700 m and then declined on sedimentary substrate, while on ultrabasic substrate they varied little with altitude. The resulting C : N ratios were always lower at ultrabasic sites, whereas exchangeable Mg was consistently greater, reflecting richer primary mineral Mg in ultrabasic rocks.

Table 2.  Soil chemical properties, biomass, productivity, nutrient-use efficiency, decomposition rate (k) of litter, stand-level PAR-use efficiency (QUE) and net assimilation rate (NAR) of the eight sites on Mt Kinabalu. Soil properties include pH (H2O), and concentrations of organic-C, total-N, inorganic N (NH4-N and NO3-N) and exchangeable cations on an oven-dry weight basis for the top 15 cm; values are means (± SD). Biomass properties include wood, leaves, total above-ground vegetation (AGB), mean standing litter crop, optically measured leaf area index (LAI) and other allometric constants (A and H*, see text). Productivity includes above-ground net primary productivity (ANPP) by component (g m−2 yr−1). Nutrient-use efficiency is for phosphorus (PUE), nitrogen (NUE), calcium (CaUE), magnesium (MgUE) and potassium (KUE)
 Altitude (m)
  1. ND = not detectable. *Soils at this site are slightly anoxic, and reduced due to the presence of a cloud belt and to reduced water permeability. Probably this soil condition relates to the greater C : N and Mg : Ca ratios at this site than at the other sites.

Soil chemical properties
pH(H2O)4.1 (0.1)4.0 (0.2)3.4 (0.2)4.9 (0.1)4.5 (0.2)5.4 (0.1)5.1 (0.1)5.3 (0.1)
Organic-C (%)2.9 (0.1)4.4 (1.2)17.7 (4.0)8.6 (2.1)2.4 (0.2)3.4 (0.3)3.5 (0.6)3.5 (0.2)
Total-N (%)0.21 (0.01)0.32 (0.06)0.92 (0.18)0.60 (0.02)0.21(0.02)0.28 (0.02)0.35 (0.02)0.26 (0.02)
C/N ratio13.813.719.214.311.412.19.913.3
NH4-N (µg g−1)3.9 (0.6)23.0 (11.6)29.9 (14.3)4.8 (1.5)8.9 (2.1)26.6 (3.8)7.3 (1.5)5.2 (0.7)
NO3-N (µg g−1)10.2 (1.8)1.8 (0.6)2.9 (1.5)8.4 (0.9)7.2 (5.6)0.8 (0.7)0.8 (1.1)ND
Ex. Ca (µg g−1)17 (10)83 (13)61 (18)734 (336)29 (12)630 (132)299 (147)375 (147)
Ex. Mg (µg g−1)31 (21)87 (21)336 (119)*80 (22)84 (16)284 (108)401 (120)276 (34)
Biomass values
Wood biomass (kg m−2)43.228.930.321.054.823.211.93.5
Leaf biomass (kg m−2)0.480.490.540.540.560.560.350.17
Total AGB (kg m−2)43.729.430.821.555.423.812.23.7
Stand litter (kg m−2)0.660.680.530.730.670.880.740.37
AGB density (kg m−3)0.660.931.171.530.801.230.940.73
A (m cm−1)1.622.061.421.301.772.001.221.55
H* (m)66.631.726.314.169.319.313.05.1
LAI (m2 m−2)
Productivity (g m−2 yr−1)
Total litterfall11107995326311113628594164
Leaf litterfall726516295438619428470121
Wood increment80342324818560218513135
Nutrient-use efficiencies (g g−1)
Other values
Decomposition k(yr−1)1.691.181.000.871.670.710.800.44
QUE (mg mol−1)230.5184.2132.1113.9207.4123.2123.841.4
NAR (g m−2 yr−1)797643446680613452382248

Soil phosphorus

All indices demonstrated greater soil P pools on sedimentary substrates (Table 3), particularly at the three upper altitudes for soluble P and total active P. The effect of substrate on soluble P was more marked when compared by area than by weight and was then significant (P < 0.05), except at 700 m. Soluble P was significantly correlated with total active P when both substrates were combined (r2 = 0.54, P = 0.04, linear regression), indicating that supply of this most labile form increased with total active P.

Table 3.  Soil P availability indices at the eight study sites on Mount Kinabalu. Soluble P in top 15 cm (oven-dried weight, and area basis), IER exchanged P in top 5 cm, and total active P in top 30 cm (after Kitayama et al. 2000). Soluble P was extracted with hydrochloric-ammonium fluoride solution. Values are means (± SD). ‘Sed’ denotes sedimentary rock sites, and ‘Ult’ denotes ultrabasic rock sites. Sites sharing the same letters in each row do not significantly differ from each other at P = 0.05 (Tukey’s HSD). Effects of substrate: *P < 0.05, **P < 0.01, ***P < 0.001. anova was not applied to total active P due to the lack of replicates
 Altitude (m)
Soluble P on a weight basis (µg g−1)
Sed 1.56 (.56)a 2.70 (1.67)a20.93 (4.46)b 6.23 (.68)a
Ult 1.18 (.45)a 0.84 (.06)a 1.89 (1.15)a 0.80 (.42)a
anova  ******
Soluble P on an area basis (g m−2)
Sed 0.18 (.07)a 0.14 (.04)a 0.36 (.03)b 0.35 (.04)b
Ult 0.14 (.05)a 0.04 (.01)b 0.09 (.04)ab 0.07 (.04)ab
anova ********
IER exchanged P (µg g−110d−1)
Sed 0.17 (0.11)a 0.27 (0.24)a 1.00 (1.17)b 0.19 (0.14)a
Ult 0.09 (0.07)a 0.23 (0.32)a 0.31 (0.52)a 0.25 (0.46)a
anova  *** 
Total Active P on an area basis (g m−2)
Ult27.59 4.0411.36 8.95

The altitudinal pattern of soluble P differed between substrates. On sedimentary substrate, soluble P increased up to 2700 m and then declined (P = 0.05, one-way anova followed by Tukey’s HSD), whereas it did not vary significantly with altitude at ultrabasic sites (P > 0.1, one-way anova; P > 0.1, linear regression using means). There were no significant (P > 0.05) differences in IER-exchanged P between substrates or altitudes other than an elevation at the 2700-m sedimentary site (one-way anova, P = 0.00001) (Table 3). Considerable amounts of Pi were trapped at all three upper ultrabasic sites, implying that its supply is greater than expected from their small soluble P pools. The altitudinal pattern of IER-exchanged P on sedimentary substrate followed that of soluble P pools.

Soil nitrogen

Both in-situ net soil nitrification and N mineralization rates in topsoils (0–15 cm) varied significantly with altitude and substrate (P < 0.05, one-way anova; Table 4). On sedimentary substrate, these rates decreased from 700 to 1700 m (or to 2700 m) and then increased slightly at 3100 m. On ultrabasic substrate, these rates decreased linearly upslope (r2 = 0.84, P = 0.08 for nitrification, r2 = 0.95, P = 0.03 for mineralization). Net rates were significantly greater at sedimentary than at ultrabasic sites for three altitudes for nitrification, and for two altitudes for mineralization (P < 0.05, one-way anova).

Table 4.  In-situ net soil N transformation rates (0–15 cm) on an oven-dry weight basis per 10 days, and in-situ IER-exchanged Ni on an air-dry resin basis. Abbreviations and statistical conventions as in Table 3
 Altitude (m)
Nitrification (µg g−110d−1)
Sed22.2 (4.1)a 0.2 (0.05)c 7.7 (2.2)b 6.4 (2.1)b
Ult10.2 (5.3)a 1.9 (3.6)b 2.3 (1.8)b 0.2 (.2)b
anova* ***
Mineralization (µg g−110d−1)
Sed19.9 (3.8)a−0.01 (0.06)b−1.3 (8.8)b 5.5 (1.4)ab
Ult 8.5 (2.7)a 1.6 (2.1)b 0.4 (1.4)b−2.2 (.4)b
anova**  ***
IER exchanged NH4-N (µg g−1 10d−1)
Sed 1.53 (2.09)a 5.16 (8.32)ab 7.77 (15.12)b 1.51 (2.17)a
Ult 4.29 (7.08)a 7.39 (14.86)a 3.23 (5.72)a 2.17 (3.37)a
IER exchanged NO3-N (µg g−1 10d−1)
Sed67.99 (31.8)ab78.06 (138.7)a 2.38 (1.1)c24.92 (23.2)bc
Ult57.04 (41.0)a12.44 (11.6)bc18.61 (12.8)b 2.18 (0.9)c
anova *******
IER exchanged total inorganic N (µg g−1 10d−1)
Ult61.3319.8321.84 4.35


Total AGB decreased linearly with increasing altitude on both substrates (Fig. 2). The decrease was sharper on ultrabasic substrate, so that values were greater on sedimentary substrate at all altitudes except for 700 m. Table 5 demonstrates the results of linear regressions on this and other variables related to forest structure and function, and of ancovas based on data in Table 2.

Figure 2.

Above-ground biomass (AGB) of the eight plots on Mt Kinabalu. Open circles denote sedimentary sites and filled circles denote ultrabasic sites. Slopes of the two regression lines are different (ancova, P = 0.02).

Table 5.  Summaries of linear regressions against altitude on each of the two substrates, and ancova between the two substrates. Values under ancova are F (P) values. Dependent variables include AGB (above-ground biomass), A (packing factor), H* (asymptotic tree height), ANPP (above-ground net primary production), Δ wood (wood increment), NAR (net assimilation rate), LMA (leaf mass per area), QUE (ANPP/PARabsorbed; phtosynthetically active radiation-use efficiencies) and LAI (leaf area index). Nutrient-use efficiencies include PUE (phosphorus-use), NUE (nitrogen-use), CaUE (Ca-use), MgUE (Mg-use) and KUE (K-use). n = 4 for regression analyses, and n = 8 for ancova. *P < 0.05; **P < 0.01; ***P < 0.001. Bold letters indicate significant differences (P < 0.05) in slope or intercept
DependentSedimentaryUltrabasicancovaF (P)
AGB (kg m−2)46.3−0.007590.80268.3−0.02140.981**16.305 (0.02) 
Wood biomass (kg m−2)45.8−0.007590.80267.5−0.02130.979*15.560 (0.02) 
Leaf biomass (kg m−2)0.457−2.83E-50.917*0.733−1.56E-40.747 8.473 (0.04) 
Standing litter (kg m−2)0.656−3.147E-60.0020.851−8.964E-50.190 0.379 (0.57) 0.027 (0.88)
AGB density (kg m−3)0.418 3.321E-40.950*0.972−2.271E-50.012 5.166 (0.09) 0.714 (0.437)
A (m cm−1)1.97−1.87E-40.3692.02−1.85E-40.339 6.6E-5 (0.99) 0.070 (0.80)
H* (m)73.2−0.0200.88681.6−0.0260.907* 0.784 (0.43) 0.626 (0.46)
ANPP (g m−2 yr−1)2097−0.4640.914*2063−0.5780.921* 0.532 (0.51) 3.704 (0.11)
Litterfall (g m−2 yr−1)1194−0.2160.8571348−0.3480.879 1.431 (0.30) 1.142 (0.33)
Leaf litterfall (g m−2 yr−1)773−0.1420.733749−0.1630.667 1.431 (0.30) 1.142 (0.33)
Δ wood (g m−2 yr−1)903−0.2480.937*716−0.2300.923* 0.074 (0.80) 7.604 (0.04)
Litter/Δ wood0.80 7.15E-40.8071.02 0.00130.992** 4.088 (0.11)17.630 (0.01)
NAR (g m−2 yr−1)401−0.04060.361361−0.07140.949* 0.578 (0.49) 8.251 (0.03)
Ln(PUE)8.59 8.87E-60.0048.09 6.78E-40.855 9.156 (0.04) 
Ln(NUE)4.60 1.03E-40.6024.543 2.71E-40.915* 4.083 (0.11) 8.304 (0.03)
Ln(CaUE)5.83−1.55E-40.0805.594−2.35E-40.356 0.033 (0.87) 1.193 (0.32)
Ln(MgUE)6.16 1.13E-40.3456.035 2.08E-40.983** 0.721 (0.44) 0.455 (0.53)
Ln(KUE)5.78 1.36E-40.6395.895 5.15E-40.662 2.024 (0.23)10.829 (0.02)
LMA (g m−2)121 0.03450.974*128 0.05120.849 1.147 (0.34) 8.098 (0.04)
Foliar N (%)1.34−1.58E-40.6951.12−1.35E-40.612 0.040 (0.85) 3.957 (0.10)
Foliar P (%)0.024−1.67E-60.5230.0088 1.99E-70.019 1.722 (0.26)59.40 (0.0006)
Foliar N (g m−2)1.60 1.20E-40.4681.38 2.40E-40.905* 1.261 (0.32) 0.079 (0.79)
Foliar P (mg m−2)25.86 0.00490.868*10.40 0.00540.927* 0.094 (0.77)99.918 (0.0002)
LAI (m2 m−2)5.33−8.68E-40.907*6.51−0.00140.792 0.989 (0.38) 0.088 (0.78)
QUE (mg mol−1)0.26−0.0490.998**0.25−5.95E-50.843 0.372 (0.57) 5.256 (0.07)
Decomposition k (yr−1)1.82−3.20E-40.930*1.87−4.66E-40.839 0.885 (0.40) 2.810 (0.15)

AGB density varied much less than total AGB values. Since the packing factor (A) showed no systematic trend across altitudes or between substrates, increasing AGB is due to increasing stand stature. As expected, therefore, asymptotic tree height H* decreased with altitude on both substrates and was smaller on ultrabasic substrate except at 700 m (Table 2). Optically measured LAI values (LAIopt) decreased with increasing altitude on both substrates, but linear regressions were significant only on sedimentary substrate (Table 5).

Total fine-root biomass at 0–30 cm increased with altitude on both substrates (Table 6). The effects of soil depth on the distribution of fine roots were significant at three lower sedimentary and at 700-m ultrabasic sites (Table 6). At sites with lower P availability, fine roots penetrated evenly throughout the upper soil profile (30 cm). Where the effects were significant, the density of roots was greatest in the top soil layer, 0–5 cm (Table 6).

Table 6.  Mean (± SD) densities of fine-root (< 2 mm diameter) biomass (mg cm−3) in the upper (0–5 cm), middle (5–15 cm) and lower (15–30 cm) soil layers, and total fine-root biomass (0–30 cm) at the eight study sites on Mount Kinabalu. Significant effects of depth are expressed with P values (one-way anova). Values sharing the same letter(s) in each row do not significantly differ from each other at P = 0.05 (Tukey’s HSD)
SiteFine root biomass density (mg cm−3)PTotal (kg m−2)
700 6.71 (4.52)a1.55 (0.68)b 0.83 (0.46)b0.00010.56
1700 5.77 (2.99)a4.44 (2.96)ab 2.26 (1.61)b0.01850.95
2700 8.11 (4.23)a4.55 (2.49)b 1.76 (1.37)b0.01110.96
3100 6.82 (3.52)6.30 (2.54) 6.18 (2.71)0.90481.02
700 5.52 (3.01)a1.69 (0.70)b 1.36 (1.36)b0.00010.52
1700 7.06 (3.63)9.12 (3.42)10.0 (−)0.41930.90
2700 7.32 (3.27)4.61 (1.97) 4.75 (2.27)0.06860.96
310010.3 (4.00)9.49 (3.10)10.1 (4.56)0.90221.44

We estimated the allocation of C as the ratio of the amount in AGB vs. in AGB plus soil (excluding roots, down to saprolite or to 1 m). This index increased monotonically with altitude on both substrates (Fig. 3a; r2 = 0.93, P = 0.04 for sedimentary; r2 = 0.91, P = 0.04 for ultrabasic; linear regressions on arcsine transformed proportions), with remarkably similar slopes and intercepts between substrates (P > 0.05, ancova). However, the distribution of C within each ecosystem differed between substrates. On sedimentary substrate (Fig. 3b), total ecosystem C did not consistently vary with altitude (P > 0.05; linear regressions) and was rather constant, while soil C increased exponentially with increasing altitude (r2 = 0.96, P = 0.02 for log transformed soil C vs. r2 = 0.85, P = 0.08 for no transformation; linear regressions). By contrast, on ultrabasic substrate (Fig. 3c) total ecosystem C decreased linearly with increasing altitude (r2 = 0.98, P = 0.01; linear regression), while soil C did not change significantly across altitudes (P > 0.05).

Figure 3.

(a) Ratio of soil C to total ecosystem C (above-ground biomass C + soil C) for ultrabasic (left hand bars) and sedimentary substrates; amounts of soil C (lower portion) and above-ground biomass C in (b) sedimentary sites and (c) ultrabasic sites. Soil C-values are the totals to saprolite or to 1 m when the saprolite was deeper. C in roots was omitted entirely.


Total annual litterfall decreased linearly with increasing altitude on both substrates (Table 5). Litterfall was dominated by leaves, followed by twigs and then reproductive organs. The proportion of total litter consisting of leaves tended to increase with increasing altitude on ultrabasic substrate but was relatively constant on sedimentary substrate.

ANPP decreased linearly with increasing altitude on both substrates (Table 5; Fig. 4a). Contrary to our expectation that the slope would be greater on sedimentary substrate, there was no difference between substrates, although the intercepts were marginally different (Table 5). ANPP was always greater on sedimentary substrate due to greater wood increments compared with those on ultrabasic substrate.

Figure 4.

(a) Above-ground net primary productivity and (b) decomposition rate of standing litter of the eight plots on Mt Kinabalu. Open circles represent sedimentary sites, filled circles represent ultrabasic sites.

ANPP to AGB ratios decreased with increasing altitude on sedimentary substrate but increased on ultrabasic substrate. Litterfall to wood-increment ratios increased linearly with increasing altitude on both substrates, with significantly different intercepts between substrates (Table 5). The upslope increase of ANPP : AGB ratios on ultrabasic substrate can thus be in part explained by greater allocation to foliage than to wood.

Estimated stand-level net assimilation rate (NAR) did not vary with altitude on sedimentary substrate. NAR decreased linearly upslope on ultrabasic substrate from 308 to 124 g m2 year−1 (Table 5). This suggests that the controls of leaf-level assimilation differed between substrates.


The decomposition rate constant, k (annual litterfall divided by standing litter), decreased significantly with increasing altitude on both substrates (Table 5; Fig. 4b). As for ANPP, slopes were unexpectedly similar, although intercepts were marginally different on the two substrates (Table 5) and k was always greater on sedimentary substrates.


The number of canopy species from which we could collect sun leaves varied from site to site, being greater at the species-richer lowland sites. We always included the most abundant broad-leaved species at a site to ensure that stand-level foliar characteristics were represented. The most consistent pattern in foliar N and P concentrations (weight basis) was their greater values at sedimentary sites (Fig. 5). Foliar concentrations did not reflect the pattern of soil N mineralization rates nor labile soil P pools. Mean values did not vary with altitude other than distinctly elevated N-values at 700 m on both substrates; this was inconsistent with prevailing patterns of other tropical mountains where foliar N and P concentrations decreased with elevation (reviewed by Tanner et al. 1998).

Figure 5.

Mean leaf mass per area (LMA), and foliar N and P concentrations (weight basis) and contents (area basis) of each study site. Error bars indicate one SD. Means and SDs per site were based on all species analysed. Open circles represent sedimentary sites, filled circles ultrabasic sites.

Leaf mass per area (LMA), however, increased with increasing altitude on both substrates, with a significantly greater intercept on ultrabasic than on sedimentary substrates (Table 5). Due to the effects of LMA, area-based foliar P increased significantly with altitude on both substrates and foliar N on ultrabasic substrate (Table 5). ancova demonstrated no significant difference in either slopes or intercepts for foliar N. By contrast, much diluted foliar P concentrations on ultrabasic substrate could not be compensated for by the greater LMA, and mean foliar P contents (area basis) were therefore always lower and intercepts significantly different between substrates (Table 5).

To control for the effects of taxonomic differences, mean foliar N and P concentrations were calculated separately for Syzygium and Tristaniopsis (both belonging to Myrtaceae). Patterns were the same as for the means of all species combined, but the differences between substrates were no longer significant (P = 0.11 for LMA, P = 0.49 for N, P = 0.17 for P; ancova).


The nutrient-use efficiency of litter production did not significantly change across altitudes for any nutrient on sedimentary substrate, but increased significantly with altitude for N and Mg and marginally for P (P = 0.07) on ultrabasic substrate (Table 5). The rate of increase in P-use efficiency across altitudes was significantly sharper on ultrabasic than on sedimentary substrates. The invariant N-use efficiencies (i.e. constant litterfall N concentrations) across altitudes on sedimentary substrate occurred in spite of decreasing soil N availability (Table 4), and were inconsistent with the patterns found on other non-ultrabasic tropical mountains (see review by Tanner et al. 1998). The dilution of N in litter on ultrabasic substrate was consistent with the pattern found on the nearby ultrabasic mountain, Silam (Proctor et al. 1989).

QUE decreased with increasing altitude on both substrates, with a greater intercept on sedimentary substrate (Table 5).



Both ANPP and litter decomposition rates were consistently greater on sedimentary than on ultrabasic substrates at the same altitudes. This is in agreement with our hypothesis that low fertility limits ecosystem function on ultrabasic substrates. Their decline with altitude on both substrates is consistent with our prediction of the effect of temperature, but the pattern was independent of nutrient status.

Effects of substrate difference

P and N availabilities were greater on sedimentary than on ultrabasic substrates consistently at all altitudes. N is not derived from parent rock, and total soil P, the dynamics between various soil P fractions (and consequently the supply of P) are therefore the stronger indicators of substrate differences. The net supply rate of bioavailable P is determined by the synergistic effects of climate, weathering intensity (which determines the size and type of P fractions) and parental substrate. Although we do not know the geological age of each site, we can assume that weathering intensity is less in higher, cooler environments. Indeed, the difference in parent rock in terms of total active P is more strongly reflected at higher elevations (Table 3).

We suggest that soil P intimately regulates soil N transformation in P-deficient soils on Kinabalu. Net mineralization of organic P is dependent on the availability of inorganic P to microbes and consequently the dilution of inorganic P relative to organic C can affect subsequent decomposition processes (Tate 1984; Tate & Salcedo 1988). P fertilization experiments during soil incubation demonstrated that the mineralization of soil N was more strongly limited by P at all altitudes on ultrabasic than on sedimentary substrates (Kitayama et al. 1998b). The control of N mineralization by P was in agreement with Walker & Syers (1976), Cole & Heil (1981), Pastor et al. (1984) and Crews et al. (1995). N mineralization in sedimentary soils was controlled by temperature but not by P, while that in ultrabasic soils was less temperature-dependent due to P deficiency (Kitayama et al. 1998b).

Effects of altitude

We found profound linear effects of altitude on both ecosystem structure (AGB and soil C) and function (ANPP and decomposition). Although our altitudinal transects encompass multiple variables that may change independently of each other, these patterns suggest the importance of linearly decreasing air pressure or air temperature, or both. The partial pressure of atmospheric CO2 is proportional to air pressure (which decreases at 1 kPa 100 m−1, Henderson-Sellers & Robinson 1986) and although the molar concentration of atmospheric CO2 remains constant, the decreased partial pressure is associated with reduced photosynthetic capacity (e.g. Körner 1989). However, the diffusion coefficient of CO2 becomes greater as air pressure decreases, mitigating the effect of partial pressure (Gale 1972). Partial pressure of O2 will also fall, but this would be expected to depress photorespiration (and thus inhibition of carbon assimilation) particularly at cooler sites (oxygenation of RuBP is more temperature sensitive than carboxylation, Terashima et al. 1995). We therefore conclude that responses in ANPP (and AGB) are likely to reflect direct effects of air temperature.


Why is ANPP lower on P-deficient ultrabasic than on sedimentary substrate but there is no interaction between altitude (temperature) and substrate (nutrients)?

LAI decreases upslope on both substrates (although only marginally on ultrabasic substrate) and thus contributes to the reduction of ANPP. Stand-level NAR appears to be invariant (sedimentary) or decreases (ultrabasic) with increasing altitude (Table 5). This, however, assumes that the allocation factor R is constant, but the proportion of NPP allocated below-ground may actually increase with decreasing ANPP (Jordan 1989). Fine root biomass did increase with decreasing ANPP on both substrates and if this reflects below-ground productivity, NAR may be greater than the calculated values (Table 2) at higher altitudes.

The different altitudinal patterns of NAR may be associated with variations in soil nutrient supply and foliar nutrient contents between the two substrates. The carboxylation capacity of tropical rain forest trees may be controlled by foliar N content (Evans 1989; Tuohy et al. 1991; Reich et al. 1994). However, photosynthetic assimilation on deeply weathered acid soils that contain a small amount of labile P is controlled by foliar P (Raaimakers et al. 1995). On Mt Kinabalu, the potential increase in net assimilation capacity as foliar N and P increase upslope may be offset by decreasing air temperature. The amount of rubisco required for the same photosynthetic rate increases with decreasing temperature (Terashima et al. 1995).

Our finding is, irrespective of substrate, consistent with the fact that plants adjust foliar characteristics under colder environments by increasing N (Berry & Björkman 1980). Given that the rate of rubisco regeneration is controlled by P (Rao et al. 1986; Jacob & Lawlor 1992), plants may also adjust to colder environments by increasing foliar P. Mean foliar P on an area basis is lower on ultrabasic than on sedimentary substrate and P-use efficiency increases with increasing altitude only on ultrabasic substrates. When a mineral limits productivity, the efficiency of its use can become greater (Vitousek 1982; Vitousek 1984; Vitousek & Sanford 1986; Silver 1994). We therefore suggest that the canopy tree species on our ultrabasic substrate cannot adjust to decreasing air temperature because of the P deficiency in their leaves coupled with soil that is impoverished in labile P.

Conversely, less P is needed to assimilate a unit of C as temperature increases and net C assimilation will therefore be more sensitive to warming if P is limiting. This may increase the slope of the ANPP regression on P-limited ultrabasic substrate, obscuring the expected effect of greater sensitivity on sedimentary substrate.

Overall, ANPP decreases with altitude primarily because of the reduction of LAI on sedimentary substrate, and the reduction of LAI and NAR on ultrabasic substrate. Hence, our suggestions do not agree with the nutrient limitation hypothesis of Tanner et al. (1998), who suggested the dilution of foliar nutrients per se (particularly N) to explain the upslope reduction of ANPP. LAI, H* and AGB probably change in parallel and may be influenced by air temperature directly or through indirect effects via soil processes.

N and P concentrations in intact leaves are constant or increase, while in litterfall they decrease with altitude. The greater nutrient-use efficiency is thus achieved by a greater resorption capacity rather than by improved carboxylation capacity per unit nutrient. Assuming that the mean foliar nutrient concentrations in Fig. 5 represent mean values of entire canopies, P resorption capacity increases sharply from a low value to nearly 80% with increasing altitude on ultrabasic substrate (data not shown).

In addition to greater resorption, the extension of fine roots to deeper horizons may be an adaptation to supplement primary mineral Ca-phosphate that is relatively abundant in deep horizons (Kitayama et al. 2000). We are not aware of other cases where the extension of fine roots to deeper horizons occurs at P-deficient (or N-deficient) sites, because dense root mats near the surface are rather common on oligotrophic soils (Cuevas & Medina 1988; Jordan 1989; Tiessen et al. 1994).

Although we have stressed the role of N and P, the lower ANPP on ultrabasic substrate could also be due to the toxicity of other elements. Ultrabasic rocks generally contain toxic levels of Mg, chromium (Cr), cobalt (Co) or nickel (Ni) (Proctor et al. 1988). Exchangeable Mg was indeed greater on ultrabasic than on sedimentary substrates at all altitudes, but the highest concentration was only 400 µg g−1, comparable with other young volcanic mountains with a similar climate, where Mg toxicity is absent (e.g. Kitayama et al. 1997). Moreover, Ca is known to offset the effects of Mg and Ni (Proctor et al. 1988) and significant amounts of Ca were present in litterfall and topsoils on ultrabasic substrate in which Ca was rather scarce. Mg toxicity is therefore unlikely, and Ca accumulation could be a biological adaptation to toxicity in agreement with Proctor et al. (1988). The effects of Cr and Co need to be investigated.


We conclude that the four forests on sedimentary substrate are regulated by temperature either directly through leaf-level processes or indirectly through a soil-to-ecosystem cascade allowing them to be used as model. The four forests on ultrabasic substrate are more deficient in P, and additional strong biological adaptations appear to maintain ANPP.

Similarly contrasting patterns in rate constant between litter decomposition (an exponential increase) and litterfall (a linear increase) with increasing air temperature have been demonstrated by Vitousek et al. (1994) and Raich et al. (1997). We can extend this transient relationship between decomposition and production into steady-state ecosystem structure along the altitudinal gradient, i.e. the monotonic reduction of the proportion of soil C to ecosystem C with increasing temperature (Fig. 3). Our results suggest that ecosystem C storage will not respond to warming because the net release of soil C would be compensated by the net increase of AGB, if nutrients are not strongly limiting.

We speculate that there is a functional linkage between the pool of soil C per area and AGB. A large stock of soil C in relation to reduced ANPP suggests a slow rate of soil organic matter mineralization and vice versa. Thus, in a forest ecosystem where critical nutrients such as P are tightly recycled through the decomposition of organic matter, the supply rate of the critical nutrients may be negatively correlated with the soil C pool. This pattern is reinforced where the bulk of the critical nutrients is bound to soil organic matter. Empirically, greater supplies of critical nutrients are associated with more massive AGB for evergreen broad-leaved forests under the same climate (Crews et al. 1995; Kitayama et al. 1997). This antagonistic relationship between soil C and AGB appears to result in nearly constant total ecosystem C on sedimentary substrate. Similarly, the total ecosystem C varied less among life zones than each component did in Venezuelan forests (Delaney et al. 1997).

The downslope increase of ecosystem C on ultrabasic substrate can be explained from the patterns of AGB and soil C pools (Fig. 3). A sharp increase of AGB occurs on ultrabasic substrate, probably reflecting the increased allocation of C to wood with temperature (i.e. a sharper change in the ratio of litter to wood increment on ultrabasic than on sedimentary substrate), while its soil C pool is constant. The effects of temperature on the allocation of C between woods vs. short-lived organs appear to differ under contrasting P supplies. The causes need to be explored in the future.


This research was funded by grants from the Japanese Environmental Agency and the Science and Technology Agency to KK, and supplemented by a Research Fellowship of the Japan Society for the Promotion of Science for Young Scientists to SA and by the grant 08041148 from the Japanese MESSC to K. Kikuzawa, Kyoto U. We are grateful to the following persons: Datuk L. Ali, F. Liew, J. Nais, R. Repin, T. Kohyama, M. Ohsawa and T. Hirose for assisting with every aspect during the study; N. Majalap, Y.L. Lee and J. Chew for assisting soil analyses; K. Kimura, T. Shumiya and N. Nomura for assisting fieldwork; and J. Wright, J. Raich, D. Drake, P. Vitousek and E. Tanner for commenting on the manuscript. This is a contribution to the TEMA, a Japanese core research of IGBP-GCTE.