Site descriptions
The study was conducted in nine permanent intensive carbon plots located across the states of Sabah and Sarawak, Malaysian Borneo (Table S1). The plots are part of the Global Ecosystems Monitoring (GEM) network (see http://gem.tropicalforests.ox.ac.uk; Malhi et al., 2021). The vegetation type is a mixed dipterocarp lowland rainforest, which is the most extensive forest type in Malaysian Borneo (Palmiotto et al., 2004). Five plots, located in Kalabakan Forest Reserve, Sabah, had been previously logged at different intensities (Struebig et al., 2013), constituting a disturbance gradient from heavily to moderately logged forest plots (Malhi et al., 2022; Riutta et al., 2018). Two old-growth forest plots were located in Maliau Basin Conservation Area, Tongot District, Sabah, and the other two in Lambir Hills National Park, Miri District, Sarawak. Neither old-growth forest area had any record of either logging or other structural disturbance. The Sabah logged forest plots are part of the Stability of Altered Forest Ecosystems (SAFE) Project, one of the world’s largest experiments in human-modified forest landscapes (Ewers et al., 2011). All nine plots had a planimetric area of 1 ha, which was divided into 25 subplots of 20 m × 20 m. A description of all the plots and their productivity is given in Riutta et al. (2018).
The study plots in Sarawak harbour the highest recorded tree species richness in the Paleotropics (Ashton, 2005). In all plots, Euphorbiaceae and Dipterocarpaceae were the most species-rich families among trees > 10 cm DBH, with the Dipterocarpaceae dominating in overall biomass. In the old-growth plots, the most common genera were the dipterocarp genera Shorea and Parashorea in Maliau, Shorea and Dryobalanops in Lambir, while in the logged plots the most common genera were Macaranga (Euphorbiaceae), Shorea (Diptocarpaceae) and Syzygium (Myrtaceae). See Kho et al. (2013), Riutta et al. (2018), and Both et al. (2019) for detailed description of the species composition.
Soils around the study plots in Sabah have developed over rocks of the Kuamut formation and are mainly orthic Acrisol or Ultisols, formulated on sandstone and mudstone (Burghouts, 1993). In the study plots in Sarawak, soils vary between sandstone-derived sandy loam humult Ultisols on ridges, and shale-derived clay of clay-rich udult Ultisols (Kho et al., 2013) in valleys. For soil nutrient contents see Table S3. The mean annual temperature and precipitation in the region are 26.7 ± 1.9°C and 2,600-2,700 mm (Kumagai & Porporato, 2012). Although no distinct seasonality exists in typical years, some drier months occur in most years, and ecologically significant droughts occur during El Niño events (Malhi & Wright, 2004). As there is no regular dry season at any of these sites, this study controls for the effect of water stress on NPP, enabling us to isolate soil nutrient controls on NPP and nutrient cycling separate from water stress factors.
The short distance (typically a few km) between some plots in any one site could potentially raise issues of pseudoreplication (Malhi et al., 2015). However, the two adjacent plots both in Maliau and Lambir sites showed considerable differences in soil conditions and species composition, sufficient to treat them as independent sample points. In Lambir, one plot is on relatively fertile clay soil, while the other is on relatively infertile sandy soil (Table S3). Logged plots in close proximity can also be treated as independent due to high spatial heterogeneity in logging intensity. See also multivariate analyses in Riutta et al. (2018) and Both et al. (2019), showing that variation within forest types is larger than between forest types.
Sampling of tree components
Nutrient samples were collected between January and March in 2014. We selected 15 trees (≥ 10 cm diameter at breast height [DBH]), consisting of five tree species with three replicates each (n = 15) outside, but within 100 m of the plot boundary of each plot, representing the most common species of the plot (see Table S2). Sampled trees had no major visible damage (e.g. no insect damage, chlorotic leaves, bent stems). Four tree components were collected from each tree: branches, green leaves, senesced leaves and fine roots. The total number of samples was 540 (5 trees species × 3 replicates × 4 components × 9 plots). Collected voucher specimens of chosen trees were identified at the species level at the herbarium of the Forest Research Centre in Sabah, Malaysia.
Mature branches with fully expanded green leaves were obtained using a long pole with a saw blade, by tugging a rope cast over a branch with a fishing rod, or by climbing. Although the collection of sun leaves is often recommended by functional trait protocols for foliar nutrient analysis (Cornelissen et al., 2003), we collected shade leaves because the majority of trees and leaves (≥ 10 cm DBH) were shaded, particularly in old-growth plots. It was also to ensure a consistency in the leaf sampling across plots, and to match green leaves against freshly fallen senesced leaves for leaf resorption estimation assuming that the majority of fallen leaves were shade leaves. The selection of sun and shade leaves is unlikely to be influential. A study of 32 leaf traits across 284 species in Borneo, where the data were collected partly from the same plots as this study, found no systematic differences between sun and shade leaves (Both et al., 2019). Similarly, an analysis of leaf plasticity to light showed that the species ranking in mean concentrations of N and P was maintained across sun and shade leaves for 38 tropical tree species in Bolivia (Rozendaal et al., 2006). All sampled branches were ≥ 1 cm in diameter and were dissected into disks. The disks were taken in the midpoint of the branch (i.e. not near the trunk nor the tip of the branch). Newly shed senesced leaves were collected from the ground; all such collected leaves were intact, fully brown and identical in shape and size to the green leaves. Twenty green and senesced leaves from each individual tree were collected for chemical analysis, and were pooled, resulting in one sample of green leaves, and one sample of senesced leaves per tree.
Mature living fine roots (here defined as roots of ≤ 5mm in diameter) were excavated near to the coarse root of each sampled tree. They were sieved and washed with tap water. These samples contained not only fine roots of the target tree but also possibly of other plants as they were not able to be distinguished empirically. We, therefore, treated them as plot-level (i.e. plant community level) samples, not as individual tree samples in our study. Fine roots were, therefore, analysed for two replicates per species per plot only (n = 10 per plot).
A tree census was conducted, and 425 species (≥ 10 cm DBH) from 175 genera were identified from 2,155 trees across five logged forest plots. The species sampled for this study accounted for, on average, 26%, 41% and 53% of the species, genus and family basal areas, respectively, in the logged plots. This enabled us to cover a broad taxonomic and functional type range from light-demanding pioneer to shade-tolerant late-successional species.
In the two old-growth forest plots in Maliau, the total number of stems, 882, comprised 239 species from 118 genera. In the two old-growth forest plots in Lambir, the total number of stems, 809, comprised 360 species representing 118 genera. The species sampled for this study accounted for, on average, 16%, 37% and 65% of the species, genus and family basal areas, respectively, in the old-growth plots.
Plot-level values were estimated from the species-level data by a nested weighting approach: first, each species was assigned the weight of its basal area proportion. Then, unsampled species within the same genus were assigned a genus-level mean, and finally unsampled species within the same families were assigned a family-level mean. We only weighted according to sampled families – unsampled families were ignited in calculating plot-level weighted mean concentrations. To increase the proportion of the sampled species in each plot, we combined species-level estimates from the plots of the same area and disturbance class, that is, we combined species-level data among logged plots, and among the two old-growth plots in Maliau. We did not combine the species-level data in the two old-growth plots in Lambir due to their strongly contrasting soil types.
Chemical analyses of plant tissues
Concentrations of five macronutrient elements (N, P, K, Ca and Mg) and carbon (C) in each individual component were quantified at the chemistry laboratory in Forest Research Centre, in Sabah, Malaysia. First, all samples were oven dried at 50°C to constant weight for five days, and were then ground with a Thomas Wiley Mill (Thomas Scientific, Swedesboro, NJ, USA) to pass through a 100-mesh (212-µm) sieve. Branch samples including the bark were chipped to small pieces before grinding. Both green and senesced leaves included petioles.
Most of the protocols are described in more detail in Majalap and Chu (1992), and a summary of the methods used is as follows. The procedures included both references samples and procedural blanks. Each sample was digested following the sulphuric acid-hydrogen peroxide-lithium sulphate digest procedure for vegetation described in Allen (1989). Phosphorus in the digest was determined using the molybdenum-blue method described in Anderson and Ingram (1993) and read at 880 nm on a spectrophotometer (HITACHI UV-VIS, Tokyo, Japan), while K, Ca and Mg contents were measured on an atomic absorption spectrophotometer (GBC Scientific Equipment, Victoria, Australia). Total C and N contents (as total element contents) were determined by a dry combustion method at 900°C using an Elementar Vario Max CN analyzer (Elementar Analysensysteme, Hanau, Germany).
Soil sampling and analysis
Within each 1 ha plot, two soil cores to the depth of 100 cm were collected and divided into six depth layers (0–5, 5–10, 10–20, 20–30, 30–50, 50–100 cm), and one soil pit dug up to 300 cm, and samples collected from 10 layers divided into 10 depth layers (0–5, 5–10, 10–20, 20–30, 30–50, 50–100, 100–150, 150–200, 200–250, 250–300 cm) on the pit walls. We then defined the depth of 0–30 cm as topsoil layer (Table S3), and 30–300 cm as subsoil layer. We used the topsoil layer as a proxy for soil nutrient availability in this study, where fine roots were mostly limited to the top 30 cm of soil profile (Yoda, 1978). Soil sampling was done following the standard RAINFOR protocol (see http://www.rainfor.org/en/manuals; Quesada et al., 2010) and soil samples were air dried in the field. Following the removal of roots, detritus, small rocks and particles over 2 mm, samples were then sieved at 2 mm, and analysed for: total P (modified from Hedley et al., 1982)d and N (Nelson & Sommers, 1996), exchangeable cations by the silver-thiourea method (Pleysier & Juo, 1980), and pH in water at 1:2.5, along with bulk density, and soil texture (the proportion of sand, silt and clay) at the University of Leeds, United Kingdom.
Estimation of NPP
Net primary productivity (NPP) is the rate of biomass synthesis that is used to form organic compounds in plants (Roy & Saugier, 2001). It is expressed as the difference between gross primary productivity (GPP) (i.e. total ecosystem photosynthesis) and autotrophic respiration, and can be quantified at the scale of an individual plant or ecosystem (Malhi et al., 2011). The following NPP components were quantified: canopy (leaves, twigs, reproductive parts), woody (stem, coarse roots and branch), and fine roots. NPP data for seven plots in Sabah was collected during 2012 to 2014, over a 24 month period in each plot, and are described in detail in Riutta et al. (2018), while the data for two plots in Sarawak was collected for 15 months between 2008 and 2010, and tree census was carried out every five years during 1992 to 2008 (Kho et al., 2013). The data were collected using standardized GEM protocols (Malhi et al., 2021; Marthews et al., 2014). Methods are described in detail in a manual available on the GEM website (http://gem.tropicalforests.ox.ac.uk, and in the site-specific papers (Kho et al., 2013; Riutta et al., 2018).
Estimation of leaf nutrient and C resorption rate:
During senescence substantial leaf mass loss occurs, mainly due to the resorption of nutrients and C (Vergutz et al., 2012); thus, disregarding mass loss leads to an underestimation of resorption by ~ 10% (van Heerwaarden et al., 2003). Calcium has been reported not to be resorbed due to its immobility (Lambers et al., 1998). A global meta-analysis by (Vergutz et al., 2012) documented that Ca resorption was statistically indistinguishable from 0% resorption for all woody species. We thus adopted the approach depicted in Vitousek and Sanford (1986): senesced leaf nutrient content was corrected with the mass loss correction factor (MLCF), using Ca content as a reference value. Measured nutrient concentration was multiplied by the ratio of Ca concentration in green leaves and in senesced leaves averaged over the plot total leaf set.
Our mean leaf mass loss derived from the mean green to senesced leaf Ca ratio was 25.0 ± 3.5%, which was slightly higher than the mean leaf mass loss of tropical evergreen woody angiosperms of 22.6% averaged from 31 global data points (Vergutz et al., 2012), which did not include any observations from Bornean forests.
Nutrient (x) resorption rate (%) was computed for each sample as
\({R}_{x}= \left(1 - \frac{{x}_{senesced}}{{x}_{green}}MLCF\right) \times 100\) (Eq. 1)
where x is an element of either N, P, K, Ca, Mg or C, Rx is the resorption rate of x, xgreen is x concentration in green leaves, xsenesced is x concentration in senesced leaves, and MLCF is the green to senesced leaf Ca ratio. The MLCF was calculated separately for each species in each plot, and the mean MLCF was 0.75 ± 0.035 across plots.
Estimation of nutrient stoichiometry, uptake, requirement and use efficiency:
x stoichiometry was expressed as C:x ratio, that is, the ratio of C concentration to x concentration of nutrient x in each plot.
The following set of equations were used to calculate total nutrient requirement, total uptake (= requirement – resorption), nutrient use efficiency (NPP divided by requirement), and nutrient uptake efficiency (NPP divided by uptake) for each nutrient and each site (Eq. 1–5):
$$xRequirement \left(kg x {ha}^{-1} {year}^{-1}\right)$$
\(=\left[\left(\frac{{NPP}_{canopy}}{{C:x}_{green}}\right)\right]+\left(\frac{{NPP}_{wood}}{{C:x}_{wood}} \right)+\left(\frac{{NPP}_{fine roots}}{{C:x}_{fine roots}} \right)\) (Eq. 2)
II. Canopy II: Woody III: Fine roots
$$xUptake \left(kg x {ha}^{-1} {year}^{-1}\right)$$
\(=\left[\left(\frac{{NPP}_{canopy}}{{C:x}_{green}}\right)\times \left(1 -{R}_{x}\right)\right]+\left(\frac{{NPP}_{wood}}{{C:x}_{wood}} \right)+\left(\frac{{NPP}_{fine roots}}{{C:x}_{fine roots}} \right)\) (Eq. 3)
I. Canopy uptake II: Woody uptake III: Fine root uptake
$$xUse Efficiencies \left(kg C per kg x {ha}^{-1} {year}^{-1}\right)$$
\(= \frac{{NPP}_{green} + {NPP}_{wood} + {NPP}_{fine roots}}{xRequirement}\) (Eq. 4)
The above estimations depend on the assumption that nutrients used for NPP are offset by an equivalent amount of nutrients accrued from soil (and nutrient resorption) over the annual cycle. Nutrient use or requirement is the total stoichiometric requirement to build plant tissue, and nutrient uptake is the fraction of that demand that is not met through resorption and hence requires fresh uptake from soil or litter. This calculation assumed negligible resorption during the fine root senescence, or during conversion of live wood to heartwood, as very little is known about those two processes (see, however, (Inagawa et al., 2023).
The NPP wood was the sum of stem, branch, and coarse root production, while C:x wood was estimated solely using the branch C and x content. To reduce a systematic bias towards either underestimation or overestimation because C and x contents in wood differed in different parts of a whole tree, the correction factors of 0.792 (N), 0.796 (P), 0.847 (K), 1.531 (Ca), 1.226 (Mg) and 1.007 (C) were applied to C:x wood. These correction factors were derived by taking the mean of C and x content of all other woody components (sapwood at the bottom of the trunk, heartwood at bottom of the trunk, sapwood at the middle height of the trunk, heartwood at the middle height of the trunk, and coarse roots) except branches, and dividing this mean by the branch content (Inagawa et al., 2023).
Nutrient uptake can be considered as the sum of nutrient accrual in above- and belowground biomass, nutrient returns to the soil via litterfall, and foliar leaching from the canopy. Hiremath et al. (2002) reported that foliar leaching N loss is < 0.5% of total N uptake and P loss is 1.7–10.9% of total P uptake. Freschet et al. (2010) also documented leaching accounted for less than 1% of leaf N pool, and no more than 0.01% of the leaf P pool. The influence of leaching on leaf nutrient status appeared generally negligible compared with leaf resorption (Vergutz et al., 2012). We therefore excluded the potential effect of leaching in our study.
Statistical analyses:
To address the research questions, we ran simple linear regression models, separately for each nutrient. To address Q1, we ran models with NPP as the response variable and the soil nutrient concentration as the explanatory variables. For Q2, we ran models with nutrient uptake from soil as the response variable and soil nutrient content as the explanatory variable. For Q3, we ran models with resorption contribution to total nutrient requirement as the response variable and nutrient uptake from soil as the explanatory variable. For Q4, we ran models with nutrient use efficiency as the response variable and nutrient uptake from soil as the explanatory variable. And finally, for Q5, we ran models with woody contribution to total nutrient requirement as the response variable and nutrient uptake from soil as the explanatory variable.
First, we applied the regression separately for logged and old-growth forests. Then, if there was evidence for a common pattern across the forest types, we ran the model for the full dataset, with and without an interaction with forest type. If the model results confirmed that the interaction with forest type was not significant (p-value > 0.05), we accepted this overall model. If there was evidence of a threshold phenomenon, we fitted a segmented regression for the full dataset. The significance of the segmented regression was assessed with Davies’ test.
Statistical analyses were performed in the R programing language version 4.0.4 (R Core Team, 2021) and the packages 'ggplot2' (Wickham, 2016) and ‘segmented’ (Muggeo, 2016, 2017).