Development of allometric models for above and belowground biomass in swidden cultivation fallows of Northern Laos

Forest Ecology and Management 357 (2015) 104–116

Contents lists available at ScienceDirect

Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco

Development of allometric models for above and belowground biomass in swidden cultivation fallows of Northern Laos Iain M. McNicol a,⇑, Nicholas J. Berry a, Thilde Bech Bruun b, Kristell Hergoualc’h c, Ole Mertz d, Andreas de Neergaard b, Casey M. Ryan a a

School of GeoSciences, University of Edinburgh, Edinburgh, Scotland, UK Department of Plant and Environmental Sciences, University of Copenhagen, Denmark Center for International Forestry Research (CIFOR), Lima, Peru d Department of Geosciences and Natural Resource Management, University of Copenhagen, Denmark b c

a r t i c l e

i n f o

Article history: Received 12 March 2015 Received in revised form 13 July 2015 Accepted 28 July 2015 Available online 25 August 2015 Keywords: Tree allometry Swidden Shifting cultivation Roots Chronosequence South-east Asia

a b s t r a c t Shifting, or swidden, cultivation remains an important land use across Southeast Asia and other parts of the tropics, albeit under pressures from other land uses. The swidden cycle of cultivation and regrowth creates mosaic landscapes with regrowing fallows of various age interspersed with active fields and patches of mature forest. Quantifying tree biomass in these landscapes is limited by the availability of reliable allometric models, hindering accurate carbon stock estimation and thus quantification of GHG emission associated with land use transitions. We therefore developed new allometric models for the prediction of both aboveand below-ground woody biomass in swidden systems based on a destructive harvest of 150 trees in Luang Prabang Province, Laos People’s Democratic Republic (PDR). This study is the first to develop allometric models of root biomass for swidden landscapes in this region, which we hypothesised would be a major carbon pool given that resprouting, and associated high root biomass, is a common physiological/morphological trait in regularly disturbed ecosystems. We found that a general model including tree diameter (DBH, cm) and height (H, m) was best for estimating aboveground biomass (AGB, kg) for both resprouts and trees growing from seed (AGB = 1.09 + 0.027D2H) though the inclusion of H only resulted in a marginal increase in model performance meaning a DBH-only model is acceptable to use in the absence of height data (AGB = 0.1286DBH2.134). Tree height was less important for estimating root biomass, with models including only DBH performing best. Re-sprouting trees exhibited greater root biomass (BGB = 0.355DBH1.732) compared to those growing from seed (0.016DBH2.597) meaning different root allometric models were developed for each tree type. Thus, we suggest that field efforts should be directed towards checking resprouting status over the estimation of tree height. We also found that models fit using non-linear regression provided equally good fits to the data compared to the traditional approach of log-transforming the data. Our models were subsequently applied to 12 nearby plots spanning a chronosequence of fallows to examine the impact of re-sprouting allometry on biomass estimation. Root biomass stocks were on average 58% higher after accounting for the allometry of resprouting trees, resulting in an average 9% increase in total biomass stocks, highlighting the importance of choosing root allometrics based on growth form in swidden fallows. If re-sprouting status is unknown, a stand-level root:shoot ratio of 0.32 can be applied. Our analysis suggests that using our models will substantially improve the accuracy of estimates of tree biomass and its distribution among different pools in swidden fallows. This information is crucial in better quantifying carbon stock changes resulting from the conversion of swidden land to other land uses. Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction Reducing emissions from deforestation and forest degradation (REDD+) is a proposed policy mechanism through which ⇑ Corresponding author. E-mail address: [email protected] (I.M. McNicol). http://dx.doi.org/10.1016/j.foreco.2015.07.029 0378-1127/Ó 2015 Elsevier B.V. All rights reserved.

developing countries could potentially receive performance related payments for avoiding greenhouse gas emissions, or enhancing carbon stocks through land use and land cover change. Countries across Southeast (SE) Asia are currently undergoing some of the most rapid rates of deforestation and forest degradation globally (Hansen et al., 2013) and REDD+ is seen as way of encouraging more sustainable land uses.

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

Yet in the upland regions of SE Asia, defined as lands between 300 and 3000 m a.s.l (’Montane Mainland Southeast Asia’; Ziegler et al., 2009), changes in vegetation carbon stocks and the resultant emissions from common land use transitions remain poorly quantified (Bruun et al., 2009; Ziegler et al., 2012) leading to uncertainty as to whether REDD+ interventions in these areas will effectively reduce emissions (Mertz et al., 2012). Shifting, or swidden cultivation, has long been a principal cause of land cover change in these upland areas, but is rapidly being transformed as other, more profitable land uses, such as rubber cultivation, replace former swidden areas (Padoch et al., 2007; Müller et al., 2014; Vongvisouk et al., 2014). However, this agricultural practice still forms a major part of local livelihoods in many areas (Cramb et al., 2009; Hurni et al., 2012) and produces landscapes high in biodiversity (Rerkasem et al., 2009) that support a wide variety of ecosystem services (Bruun et al., 2009; Ziegler et al., 2009). Swidden cultivation involves clearing a small patch of land using a mixture of cutting and fire (‘slash and burn’), after which the land is cultivated for one or more years before eventually being abandoned and the natural vegetation allowed to regrow. Cultivation is commonly practiced between short (<5 years), intermediate (5–10) or long (20+ years) periods of fallow before the land is re-cultivated (Ziegler et al., 2012), resulting in a complex mosaic of land covers with agricultural fallows in various stages of regrowth interspersed with active fields and patches of mature forests. In Laos PDR – the focal country of this study – these dynamic landscapes make up approximately 30% of the total land area (Messerli et al., 2009). Their areal extent means that emission reductions from avoided degradation and carbon uptake from forest regeneration associated with changes in land management could make a significant contribution towards REDD+. However, in order to quantify potential emission reductions and removals, an accurate quantification of carbon stocks across these landscapes is required (Ziegler et al., 2011). The estimation of tree biomass ultimately relies on the use of allometric equations to convert forest inventory data (diameter, height) to its total biomass (in kg dry matter). The choice of allometric model is therefore a critical step in the estimation of forest biomass, and by extension, carbon emissions that occur via land use and land cover change (Chave et al., 2004; Skole et al., 2011; Picard et al., 2014). In order to generate accurate predictions of biomass, the chosen allometric model should ideally represent the growth form of trees in the area being surveyed, which often requires a site-specific or regionally developed model. However, allometric models for predicting aboveground woody biomass for swidden systems in SE Asia are still few in number (Ketterings et al., 2001; Chan et al., 2013), while none exist for estimating belowground root biomass, with the best alternatives based on harvest data collected from lowland tropical forests (Yuen et al., 2013). Harvesting trees and weighing their components is a time-consuming process meaning new projects may decide to apply one or more of the current regional models (e.g. Shanmughavel et al., 2001; Kenzo et al., 2009; Chan et al., 2013). However these models are typically developed from a small number of trees with a narrow diameter range, which may represent only a fraction of the allometric variability within a forest type and could result in potentially spurious predictions when applied to broader diameter ranges (Chave et al., 2004; Gotelli and Ellison, 2004; Picard et al., 2012; Rutishauser et al., 2013). To avoid any potential bias in the estimation of large tree biomass, some projects instead decide to use generic pan-tropical allometric models based on large datasets obtained from multiple regions across the tropics (e.g. Chave et al., 2005, 2014). Recent studies have shown that these generic models can result in more accurate predictions of biomass compared with regionally derived models

105

when applied to inventories of SE Asian dipterocarp (Rutishauser et al., 2013), and peat forests (Manuri et al., 2014). However, these global/pantropical allometric equations are largely parameterised using data from relatively undisturbed forests and include few secondary forest stands, introducing the potential for bias if applied to more disturbed forest types (Chave et al., 2014). Swidden fallows are subject to regular disturbance, thus trees growing in these areas are likely to have very different architecture, traits and biomass allocation strategies compared to those found in either undisturbed or logged forests. Highly intense and/or extremely frequent disturbance is often needed to completely eradicate individual trees and roots can persist over multiple cycles of disturbance and re-sprout when fields are left fallow (Bond and Midgley, 2001). This is a common physiological trait in ecosystems where disturbance is a regular occurrence, allowing rapid regeneration and trees to persist following the destruction of aboveground parts (Bond and Midgley, 2001; Clarke et al., 2012; Nzunda et al., 2014). Resprouting may also occur in places where nutrient availability and soil stability is a key constraint to growth, as may be expected in low input upland ecosystems (Brearley, 2011). The ability to re-sprout, however, may come at the detriment of height or canopy growth as more resources are held in reserve to support future regeneration (Clarke et al., 2012; Nzunda et al., 2014) leading to different allometries between trees which have re-sprouted and those which grow from seed. This suggests that aboveground tree biomass in fallows may be overestimated by equations developed in primary forests where disturbance is less common (van Breugel et al., 2011). In contrast, it is hypothesised that root biomass stocks in fallows will be considerably greater than is predicted by allometric models developed in non-swidden areas where re-sprouting trees are less common (Shanmughavel et al., 2001; Kenzo et al., 2009; Niiyama et al., 2010). The objective of this paper is to provide a way of better quantifying the carbon stock changes that are likely to result from the intensification of swidden agriculture and the conversion of swidden land to other land uses. The specific aims of this paper are to: (1) Develop allometric models for above- and below-ground tree biomass for swidden fallow trees in Northern Laos (2) Examine the impact of applying these new models to estimate woody biomass across a chronosequence of swidden fallows located in the same landscape, and compare these estimates to those derived from regional and pan-tropical allometric models This study is the first to develop allometric models of below-ground biomass for swidden cultivation landscapes of SE Asia. To construct our allometric models we used a combination of approaches involving non-linear regression, and linear regressions fit to log-transformed data. The conventional decision to log-transform the data has been the subject of considerable debate in the literature (Cunia, 1964; Xiao et al., 2011; Packard, 2013), yet many published papers fail to provide a rigorous examination of whether this is a necessary step (Návar, 2009; Mascaro et al., 2011) with most studies only applying the log-transformed version of the power-law function in order to try stabilise the residual variance, a key assumption when fitting regression models. The increasing availability of non-linear techniques have led to greater use of non-linear models fit to untransformed data (Litton and Kauffman, 2008; Ritz and Streibig, 2008; Návar, 2009; Mascaro et al., 2011; Mugasha et al., 2013), with studies also showing that log-transformation does not always result in a stabilisation of the residual variance (Packard, 2012), or necessarily result in a better fit compared to fitting non-linear models (Niklas, 2006; Litton and Kauffman, 2008; Návar, 2009; Packard, 2013). Therefore in is

106

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

study, multiple fitting methods are tested and the results compared. 2. Methods 2.1. Study area Sampling was carried out in and around the village of Moungmouay (20°160 1600 N, 103°20 5500 E) in Viengkham District, Luang Prabang Province, Laos PDR, the same 26 km  26 km study area used by Hett et al. (2012) (Fig. 1). The topography and land use in this area is typical of that found in northern Laos, with mountainous terrain (average elevation: 600 m a.s.l, average slope angle: 17° (0–63°); SRTM, Farr et al., 2007), a large part of which is used for shifting cultivation, with small areas of permanent rice paddy on flat areas in valley bottoms (Hett et al., 2012). The area has a humid subtropical climate with a wet season between May and September. The district has a mean annual temperature of 21 °C and a mean annual precipitation of 1550 mm (WorldClim: Hijmans et al., 2005), meaning the vegetation can be classified as ‘moist forest’ (Holdridge, 1967; Chave et al., 2005). Common species include the deciduous trees Quercus kerrii, Cratoxylum formosum, Aporosa villosa and the evergreen species Macaranga denticulate and Ficus drupacea, with bamboo common in the understory. All sample sites were located on Ultisols with sandy clay– clay loam texture. 2.2. Destructive harvest A total of 150 trees were harvested in fallows scheduled for clearance in the 2013 and 2014 growing seasons with candidate sites determined by interviewing local famers with the assistance of agricultural extension agents from the District Agriculture and Forestry Office. Fallows ranged in age from 2 to 35 years with harvested trees ranging in size from 1.7 to 36.2 cm DBH and in total dry mass from 1.5 to 685.7 kg. Trees were selected according to

whether they had re-sprouted from existing rootstocks after being cut (hereafter re-sprouting trees; n = 59) or whether they had regrown from seed (hereafter ‘simple’ trees; n = 91) (Fig. 2). Resprouting status was ascertained by checking whether scars from previously cut stems were present at the base of the stem, or below the soil surface. At each site a random start point was determined by choosing a compass bearing and a distance between 20 m and 100 m from the edge of the fallow. The closest 20 trees to the start point that fit the criteria (simple or resprouting) were numbered and a tree for measurement was selected at random. Sampling continued until a sufficient number of trees in each of the size classes 0–5 cm, 5–10 cm, 10–15 cm, 15–20 cm, and >20 cm were sampled. A new site was visited for each day of sampling. For each of the trees selected, the diameter (cm) of the stem at 1.3 m above ground level was first measured, after which the tree was felled and its total height (m) measured. All leaves were stripped from the branches which were then cut from the primary stem and all roots >2 cm diameter were excavated (Fig. 2). The fresh biomass of the leaves, branches, stems, and roots of each tree were weighed separately in the field. To estimate the moisture content of each tree compartment, and thus convert fresh mass measured in the field to dry mass, representative samples of around 1 kg of leaves, 5 kg of branches, 10 kg of stem wood, and 5 kg of roots were collected from each tree and were initially air dried to constant mass. Following this, a further representative subsample of 100–500 g of each air dried sample was then dried for 24 h in an oven at 102 °C for the final calculation of dry mass. For trees with more than one stem, the diameter of all stems was combined on a basal area basis (effective DBH: Ryan et al., 2011), which is defined as,

De ¼ 2 

p X ð pðD=2Þ2 Þ=p

ð1Þ

where De is the effective diameter of a multi-stemmed tree and D is the DBH of each of the s stems. This is needed as two stems of the same diameter do not have the same combined biomass as a single

Fig. 1. (a) Location of Laos PDR in SE Asia and (b) Viengkham district (shaded in red) where this study was conducted. Map (c) shows a more detailed map of our study area including the local topography, the locations of the harvested trees used to construct the allometric models and the 12 swidden fallows where the census data which we apply our models to was obtained. The age (years since cultivation) of each plot is also shown. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

107

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

Fig. 2. Root networks of re-sprouting trees that were harvested for the construction of the local allometric models.

stem double their size due to the non-linearity of the DBH–biomass relationship. Summing the basal area and back converting to get a single ‘effective DBH’ that has an identical basal area is way around this. 2.3. Allometric model development and evaluation The power-law relationship (Eq. (2)), or its linearised form (Eq. (3)) are commonly used as the underlying allometric models for predicting tree biomass (Brown et al., 1989; Chave et al., 2005; Návar, 2010; Picard et al., 2012);

y ¼ axb þ e

ð2Þ

lnðyÞ ¼ lnðaÞ þ b lnðxÞ þ e

ð3Þ

where y is the response variable, usually dry biomass in kg, x is the predictor variable, normally tree DBH and/or height, a and b are the scaling constant (or intercept) and the scaling exponent (or slope) respectively, which are estimated via ordinary least squares procedures, and e is the regression error. In this study both fitting methods (Eqs. (2) and (3)) were applied and the results compared (see Table 1). All statistical analyses were performed using R statistical software and associated packages where stated (R Core Team, 2014). As biomass data is always heteroscedastic – i.e. the variance in tree mass increases with tree size – directly fitting a non-linear model to the data (Eq. (2)) requires some form of weighting procedure so that rare, larger trees, which have a high intrinsic variability in biomass, and so tend not to fit the model well, have a lower leverage on

model coefficients. This has little impact on the parameter estimates of the model, however, the confidence intervals surrounding these estimates are likely to be biased (Picard et al., 2012). One way of approaching this is to weight each data point by the inverse of the diameter as implemented by Brown et al. (1989) who suggested 1/D4 to be an appropriate weighting. The destructive harvest datasets from South-East Asia included in Chave et al. (2014) were used to test this weighting approach by examining how the variance in measured biomass scaled with increasing DBH (1 cm bins) across the range of tree sizes sampled in this study (see also Cunia, 1964). Results indicated that variance is proportional to DBH to the power 3.51 (±1.88; ± indicates 1 standard error, SE) providing empirical support for the weighting used by Brown et al. (1989) (see also Picard et al., 2012). We therefore used the weighted least squares regression (WLS) approach from Brown et al. (1989) as the first fitting method used to construct allometric models (Model 1). An alternative approach to weighted regression is the method of variance modelling (VM), described in Ritz and Streibig (2008), Picard et al. (2012) and Packard (2013); where the variance of the i-th observation – var (ei) – depends on the corresponding mean DBH fitted through a power function with exponent 2h, however this time the coefficient h is now a parameter that needs to be estimated along with the other coefficients (a, b). Here, the variance structure of the data is fitted directly for each candidate model to untransformed data, as opposed to using an a priori value for the exponent to construct the weights (Model 2). These models were fit using the ‘nlme’ function and the built in ‘varPower’ variance model in the ‘nlme’ package (Pinheiro et al., 2014). The third and final approach was to fit models using ordinary least squares (OLS) linear regressions on log–log transformed data, with (Model 3) and without (Model 4) the correction factor (CF) applied (Sprugel, 1983). The CF is commonly recommend for back transforming logged data (Sprugel, 1983) and is applied by multiplying the right hand side of the allometric equation by a CF defined as exp[RSE2/2], where RSE is the Residual Standard Error on the regression model. Given the difficulties in measuring tree height in dense forests, all models were first parameterised using DBH as the sole predictor variable (e.g. M1DBH) after which tree height was included as an additional covariate to examine how it affected model performance (e.g. M1DBH+H). Height was first allowed to vary with its own exponent using weighted regressions fit to untransformed data, and log-transformed data (M1DBH+H–M4DBH+H), and also included as a compound variable with DBH (D2H: M5D2H and M6D2H), with all models fit separately to resprouts and simple trees (Table 1). The combination of different model forms and predictor variables meant there were 10 candidate models available for

Table 1 Description of the various model forms and fitting methods used to create models. Fitting methods include weighted least squares regression (WLS), non-linear regression (NLR), ordinary least squares regression (OLS), and variance modelling (VM). Models were categorised according to the independent variables (DBH, DBH + H, D2H) used. Independent (x) variable

Fitting method (R command)

Correction factor

Biomass-diameter models M1DBH Biomass M2DBH Biomass M3DBH log (Biomass) M4DBH log (Biomass)

Model ID

Dependent (y) variable

DBH DBH log (DBH) log (DBH)

WLS (nls) NLR + VM (nlme) OLS (lm) OLS (lm)

n/a n/a No Yes

Biomass-diameter-height M1DBH+H M2DBH+H M3DBH+H M4DBH+H M5D2H M6D2H

DBH + H DBH + H log (DBH) + log (H) log (DBH) + log (H) D2H D2H

WLS (nls) NLR + VM (nlme) OLS (lm) OLS (lm) WLS (nls) OLS + VM (nlme)

n/a n/a No Yes n/a n/a

models Biomass Biomass log (Biomass) log (Biomass) Biomass Biomass

108

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

selection, all of which were fitted separately for the aboveground (stem, branches and leaves combined), and belowground components of both simple and re-sprouting trees (Tables A1 and A2). Total biomass was calculated as the sum of these two pools. Allometric relationships between tree height and diameter were also constructed to examine whether there were any differences in this relationship between simple and resprouting trees (Fig. A1; Banin et al., 2012). We used ANCOVA to test for significant difference in allomteric parameters between tree types. Models were selected by first checking the residuals to assess whether the chosen statistical method (weighting or log-transformation) correctly accounted for heteroscedasticity. Residual homogeneity was tested statistically by converting the residuals to absolute values and checking for linear correlations (Kendall tau rank) with fitted values (Kutner et al., 2005). Commonly used metrics for model selection such as the AIC are not appropriate for comparing statistical models that are fit using different methods and response variables due to the way the log-likelihood constant used in the calculation is constructed, therefore we can only use it to compare nested models fit using the same command and to the same data (log or untransformed) (Burnham and Anderson, 2002; Picard et al., 2012; Packard, 2013) (Table 1). Model selection was therefore based on comparing the total, mean and standard deviation of the absolute errors (kg) for each model, defined as the difference between the measured biomass of an individual tree (Bobs), and the biomass predicted by each model (Best), and the mean and standard deviation (SD) of the bias, or relative % error (RE) for each model: P 100 ⁄ (Best  Bobs)/Bobs)/Nobs (Goodman et al., 2013; Chave et al., 2014). The decision of which model form is ‘best’ depends on whether the absolute or relative errors are used as main selection criteria (Tables A1 and A2). Even a small percentage discrepancy between measured and predicted biomass for large trees can have a disproportionate impact on absolute errors, although these stems are relatively rare in disturbed stands. In contrast, the relative error (RE) accounts for how well the model fits across all stem sizes, including smaller stems (<15 cm DBH) which made up on average 80% (35–100%) of the stems in each chronosequence plot. Therefore we used the mean and standard deviation of the RE as the main selection criteria. 2.4. Implications for biomass estimation in swidden fallows To explore the implications of using our new allometric models on biomass estimates, we estimated woody biomass stocks in 12  0.82 ha plots sampled across a chronosequence of swidden fallows. Four plots each were established in areas cultivated in either short (<5 years), intermediate (5–10 years) or long (>20 years) rotations. We used a nested sampling strategy with all stems >5 cm sampled within a central 80  5 m (0.04 ha) transect perpendicular to the slope of the field. In the old fallows only, medium sized trees P10 cm DBH were surveyed within a 0.32 ha plot surrounding the smaller transect, with larger trees >30 cm DBH sampled across the whole 0.82 ha plot. For the young and intermediate aged fallows all trees P10 cm DBH were sampled within the entire 0.82 ha plot. Tree height was not measured in the plots; therefore we use the best-fit models that includes only tree DBH as a predictor variable. Resprouting status was also not recorded in the chronosequence plots, therefore in order to estimate the potential impacts of any differences in the allometry of the different tree types we used a probability based method founded on data collected during the destructive sampling where we recorded the number of trees checked before a resprouting stem was found. This information was used to estimate the proportion resprouts in different diameter classes. We estimate that resprouts account for 29 ± 12% (SE) of

trees between 5 and 10 cm, decreasing to 27 ± 14% of those between 10 and 15 cm, 24 ± 13% of those 15–20 cm and 21 ± 12 of those >20 cm DBH. For trees <5 cm DBH the ratio was much higher at 46%, however these were not measured in our plots. We used a Monte-Carlo procedure to subsample trees within each plot into those that were likely to be re-sprouting and those that were not, repeating this 100 times to ensure an adequate number of potential combinations of trees. 2.5. Comparisons to existing models Both the tree- and plot-level estimates of biomass generated by our new allometric models were compared to two existing secondary forest AGB models developed in the same region (Kenzo et al., 2009; Chan et al., 2013); two pantropical studies of AGB (Chave et al., 2005: ‘moist forest’ equations, Chave et al., 2014) which use a wood density value of 0.54 g cm3, based on the average density of species identified across the chronosequence; and three regional models of BGB (Shanmughavel et al., 2001; Kenzo et al., 2009; Niiyama et al., 2010). For the plot-level comparisons, only those models including DBH were evaluated. The validity of the existing models was assessed by plotting the relative error (%) of the AGB estimates from each model against measured biomass across the range of tree sizes harvested. 3. Results 3.1. Destructive harvest and biomass allocation patterns For simple trees, the stem consistently accounted for 64% of the total tree biomass, whereas among resprouting trees this varied from around 15–35% for individuals <10 cm DBH, increasing to 50% for trees >10 cm DBH (Fig. 3). For resprouting saplings (0–5 cm) the average fraction of total tree biomass stored belowground was 66%, however this proportion tended to decrease as trees increased in size. The individual root:shoot (R:S) ratios for simple trees ranged from 0.04 to 0.72 (0.26 ± 0.13, mean ± SD), significantly lower than for the resprouts (ANCOVA, P < 0.01; 1.00 ± 1.2) where the ratio again varied depending on stem size, ranging from 0.19 to 5.78 (1.72 ± 1.68) for stems <10 cm, and 0.25 to 1.20 (0.54 ± 0.26) for stems P10 cm. The allocation of biomass to branches and leaves was broadly similar across both tree types, with the exception of saplings (0–5 cm) with simple trees exhibiting greater branch biomass and lower leaf biomass. On average, tree height was greater among simple trees compared to the resprouts for a given DBH with the best diameter–height allometric models reaching an asymptote at 16 m for the former, and 14.9 m for the latter (Fig. A1). 3.2. Allometric modelling As expected, the biomass of the harvested trees displayed clear evidence of heteroscedasticity, with the variance in both above(stem, branch and leaf) and below-ground biomass increasing with DBH (Figs. 5 and 6). Both of the non-linear methods and the log-transformation were largely successful in creating constant residual variance for all of the candidate models (Fig. A2); however the linear models fit to log-transformed data for resprouts (M3DBH) did not result in homoscedastic residuals for either pool (Kendall’s tau, P = 0.03 and P = 0.04; Fig. A2). The weighted least squares (WLS) regression models (M1DBH) where the weights were initially set in accordance with Brown et al. (1989) did not stabilise the residual variance for either tree type. Instead a weighting of 1/D5 was found to better characterise the error structure of simple trees, whereas for the resprouts, the variance in biomass at large stem

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

109

Fig. 3. Allocation of biomass to different tree compartments, separated by size class and resprouting status. The larger numbers within the sections of each bar indicate the mean values for trees in that diameter class while the smaller numbers below indicate the standard deviation.

Fig. 4. The relative error (% difference between predicted and measured biomass) with increasing tree size for all of the models fit to the AGB data (a–b, d–e) and BGB (c–f) data for simple and re-sprouting trees. The lines were fit using a lowess (locally weighted scatterplot smoothing) procedure.

sizes was relatively smaller – potentially reflecting the presence of a more homogenous group of disturbance-tolerant species with similar allometries – with a weighting of 1/D3 used instead (Fig. A2).

3.2.1. Aboveground biomass 3.2.1.1. Biomass–diameter models. The models including only DBH as a predictor (M1–M4DBH) yielded broadly similar mean estimates of AGB for both resprouts and simple trees across the range of tree

110

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

Fig. 5. Comparisons of our best fit AGB models for simple trees (black hatched lines), resprouts (red hatched lines) and all trees combined (solid black lines), and regionally developed and pan-tropical allometric models (see Table A1 for more information). The Chave models use a mean wood density of 0.54 g cm3 based on the average density of the trees sampled across the chronosequence plots. We also used a value of 0.252 for E in Chave et al. (2014) model. The right hand panels show the trend in relative error with increasing tree size (% difference between predicted and measured biomass) for the literature models and our general AGB model using a lowess (locally weighted scatterplot smoothing) procedure. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

sizes sampled (Fig. 4), with the exception of the models fit using log-transformed data with no back-transformed correction factor applied (M3) which under-predicted AGB. Averaged across all stem sizes, M1, M2 and M4 tended to overestimate the true biomass of both simple and resprouting trees by 21–27%, with a standard deviation of 56–62%, considerably higher than M3 which had mean bias of 10 ± 51% (Fig. 4; Tables A1 and A2). Model 3 was therefore considered to provide the best-fit for both simple and resprouting trees (Table A1), leading to the following ‘best’ equations for each tree type:

AGBSimple ¼ 0:1083  DBH2:37 AGBResprouting ¼ 0:1451  DBH2:27 However, an analysis of covariance clearly indicated there were no significant differences in the regression coefficients between models constructed for resprouting and non-resprouting trees

(ANCOVA; slope, P = 0.44, intercept, P = 0.75). The very similar fits for the two models further indicate that a single general allometric equation is suitable to use to estimate AGB from DBH alone (Fig. 5; Table A1):

AGBGeneral ¼ 0:1286  DBH2:31 3.2.1.2. Biomass–diameter–height models. The addition of tree height as an explanatory variable as part of the compound variable D2H (M5D2H and M6D2H) resulted in a linear relationship with AGB (Fig. 5). The models fit using the VM approach (M6) were almost identical to those fit using the WLS method (M5) (Fig. 4) with both models a substantial improvement on their comparable DBH-only models (M1 and M2) based on both their AIC values and a reduction in the overall (mean ± SD) bias (Table A1). The alternative models where height was allowed to vary with its own exponent (M1DBH+H and M2DBH+H) resulted in a poorer fit to the data

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

111

Fig. 6. Comparisons of our best fit models of BGB for each tree type and other regionally developed allometric models (see Table A2 for more information). The Shanmughavel et al. (2001) model uses D2H as a predictor meaning the individual estimates are shown instead. For each model the trend in relative error with increasing tree size was again represented using a lowess procedure.

compared to their equivalent model forms fit to untransformed data with D2H as a predictor (M5 and M6) (DAIC = 1.05  46.08; Table A1; Fig. 4). When fitting models using log-transformed data, the inclusion of tree height (M3DBH+H and M4DBH+H) again resulted in a statistically significant improvement to the model fit for both simple (ANOVA, P = 0.007; DAIC = 5.34) and resprouting trees (P = <0.001; DAIC = 11.43) compared to using the equivalent DBH only models, however its inclusion only resulted in a marginal increase in the explained variation (R2 increase of 0.01) and a minor reduction in the mean and standard deviation of the bias (Table A1). The models fit using log-transformed data without the CF applied (M3DBH+H) could again be considered to provide the best for fit AGB for both tree types based on a minimisation the overall bias (Fig. 4; Table A1). However, there was only a marginal increase (<5%) in model bias when using either of the models which used D2H as the independent variable (M5 or M6). For M3DBH+H a repeat of the ANCOVA revealed no significant difference between the scaling coefficients for tree height (P = 0.21), but a marginally significant difference, or interaction between DBH and tree type (b = 2.204 vs. 1.848, P = 0.04) suggesting separate allometric models would need to be applied. In contrast, for M5 and M6, both tree types exhibited broadly similar fits (ANCOVA; slope, P > 0.3) indicating that single general model could instead be used (Fig. 5; Table A1). We therefore favour the use of the models using D2H and untransformed biomass data over the log-transformed model for

estimating AGB when height data is available. Both of the fitting methods were considered valid given their very similar predictions of biomass, however we put forward the models constructed using the VM approach (M6D2H) given that the weightings were fitted directly to the data;

AGBSimple ¼ 0:027  D2 H AGBResprouting ¼ 1:26 þ 0:030  D2 H AGBGeneral ¼ 1:09 þ 0:027  D2 H 3.2.2. Belowground biomass For root biomass (BGB), the results of the model fitting and evaluation process closely mirrored that of AGB with three of the four DBH-only models (M1, M2 and M4) exhibiting broadly similar fits to the data (Table A2; Fig. 4). Model 3 was again found to fit the data best in terms of having the lowest error relative to measured biomass, particularly at smaller stem sizes;

BGBSimple ¼ 0:016  DBH2:597 BGBResprouting ¼ 0:355  DBH1:732 In agreement with the differences between simple and resprouting trees in terms of biomass allocation (Fig. 3), this time we found a significant difference in the diameter–biomass

112

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

relationship for the two types of trees, with root biomass shown to be significantly greater among the resprouts based the difference in the model intercepts (ANCOVA, intercept = P < 0.01), although the root growth rate (scaling exponent, b in Eqs. (1) and (2)) was greater for the simple trees (ANCOVA, slope = P < 0.01) (Fig. 5). Unlike the models fit to the AGB data, the inclusion of tree height did not improve the model fit so these models were removed from further consideration (Table A2). 3.3. Evaluating existing models The tree-level predictions of AGB generated by our general DBH-only model (M3DBH) are broadly similar to those derived from existing allometric models developed from trees in post-fire secondary forests in Malaysia (Kenzo et al., 2009) and swidden fallows in Myanmar (Chan et al., 2013). Compared to our model, both the Chan and Kenzo equations tended to underestimate biomass for smaller trees (<10 cm DBH) (Fig. 5), with the Chan model also comparatively overestimating the biomass of large trees (>30 cm DBH). In common with these studies, our biomass estimates were considerably lower than those generated by the generic tropical forest model developed by Chave et al. (2005) which systematically over-estimated AGB by an average of 78% (Fig. 5). The updated model by Chave et al. (2014) produced more accurate predictions of AGB compared to its predecessor with a mean RE of +21%, however the model still tended to overestimate AGB for trees P10 cm DBH (Fig. 5). In contrast, both of the pan-tropical models which included both DBH and height as predictors yielded very similar estimates of AGB to our model (Fig. 5) meaning when both DBH and height data are available either can be used. The DBH + H model developed by Chan et al. (2013) which also used D2H as a predictor performed poorly against our data, tending to under-predict biomass, especially for larger stature trees of which there were few in their study. For BGB, the relative differences in biomass estimates between our models and those developed regionally were more distinct than those for AGB (Fig. 6), with the root biomass model developed by Kenzo et al. (2009) under-estimating the biomass of simple trees by an average of 12%, while models developed in primary forests by Shanmughavel et al. (2001) and Niiyama et al. (2010) over-estimated their biomass by +14% and +49% respectively. However when compared to the resprouts, all three models tended to significantly under-estimate tree-level root biomass by an average of 68% (Kenzo), 64% (Shanmughavel) and 47% (Niiyama) across the range of tree sized sampled (Fig. 6). 3.4. Impacts of applying our new allometric models on estimates of forest biomass Applying our new general DBH-only allometric model to the data from the 12 chronosequence plots resulted in stand-level AGB stocks ranging from 2.7 Mg ha1 in a 3-yr old fallow, up to 67.7 Mg ha1 in a 25-year old plot (Table A3; Fig. 7). The two regional models (Chan and Kenzo) yielded very similar AGB estimates to our local models for the young (<5 years) to intermediate aged (5– 10 years) fallows. Similar estimates were produced by the Chave et al. (2014) model which both over- and under-estimated the AGB of some plots, a considerable improvement on the Chave et al. (2005) model which generated estimates up to 8 Mg ha1 higher in two 7-year old stands, a difference of 45–60%. Differences between our local and pan-tropical models were more pronounced in the older fallows with Chave et al. (2014) model over-predicting AGB by up to 11 Mg/ha1 in two 25-yr old plots, in one of which the Chave et al. (2005) model produced an estimate that was 54 Mg, or 80% higher than what was estimated using our model. Old fallow AGB stocks estimated by the Chan

model were again similar to those generated by our local model, while those by Kenzo were lower. The impacts of applying our new root allometric models on BGB stocks compared to using the existing regional models was pronounced. Our estimates of root biomass stocks were consistently higher than the estimates generated by the allometrics by Niiyama et al. (2010) (1–25%) and Kenzo et al. (2009) (45–51%) (Fig. 7), and 58% (22–85%) higher than those estimated using the allometric developed for simple trees only, equivalent to an average 9% (4–13%) increase in total biomass (TB) stocks (Table A3), thus highlighting the importance of resprouting allometry in swidden fallows. The stand-level BGB:AGB ratio averaged 0.32 and was largely consistent across all plots (±0.009 SD; 0.31– 0.34). Combining our AGB and BGB estimates resulted in total biomass stocks ranging from 3.7 to 89.3 Mg ha1 (Fig. 7), 18% higher (6– 27%) than using the regional models which ignore resprouting. Total biomass is estimated to accumulate at a rate of 2.88 Mg/ha/yr1 over a 25-year period of regrowth, corresponding to a 2.18 Mg/ha/yr1 increase in AGB, and a 0.67 Mg/ha/yr1 increase in BGB. 4. Discussion Destructive harvest data and associated allometric models from tropical forests are still relatively sparse given the extent of the biome, but they are also unevenly distributed with few models having been constructed for tropical secondary forests (Yuen et al., 2013; Chave et al., 2014). This issue is now of particular scientific interest given that secondary forests are projected to cover a large proportion of the tropical land mass (Hett et al., 2012) and are a major terrestrial carbon sink (Pan et al., 2011). Several studies have noted large differences in tree biomass estimates generated by pan-tropical models and models developed for a variety of disturbed ecosystems, exemplifying this issue (Williams et al., 2008; Skole et al., 2011; van Breugel et al., 2011). 4.1. Tree architecture in swidden fallows Our data revealed several different growth patterns between trees regrowing from seed and those resprouting from pre-existing rootstocks, with the latter found to be shorter for a given DBH (Fig. A1), and shown to store more biomass below ground compared to the simple trees (size-averaged R:S ratio = 1 vs 0.25). The R:S ratio of resprouts decreased as tree size increased (1.72 < 10 cm DBH vs 0.54 P 10 cm), which may be expected given that resprouting trees already have relatively well-established root networks which do not require much additional investment. This was ultimately reflected in the slope coefficients in each of the fitted models which were consistently lower compared to models for simple trees. The differing DBH + H relationships may well be explained by the fact that resprouting trees are more likely to hold resources in reserve to fund their rapid regeneration post any future disturbance, reducing resources for height growth (Nzunda et al., 2014). Resprouts also appear to allocate proportionally more resources to leaf biomass in the early stages of development which may also be interpreted as an adaptive response to regular disturbance, by allowing the plant to replenish non-structural carbohydrates in roots which in turn facilitates future regeneration (Clarke et al., 2012). Simple trees, on the other hand, tended to allocate more biomass to the stem and branches, and were taller for a given DBH, perhaps in order to attain a disturbance tolerant size, or reach a reproductively mature state (Bond and Midgley, 2001). This could result in simple trees having a competitive advantage over resprouts which may explain the gradual

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

113

Fig. 7. Estimates of total, aboveground, and belowground biomass for the 12 chronosequence plots sampled in the vicinity of the destructive harvest data. Stand-level estimates were generated using the best fit allometric models with DBH as the only predictor; and a combination of literature models (Table A3). The IPCC default R:S ratio for tropical secondary forests is 0.42 (Penman et al., 2003). For total and aboveground biomass a non-linear power-model best captured the pattern of regrowth (total biomass = age1.329; AGB = age1.242) and was selected over linear models on the basis of minimising the AIC and residual sum of squares. For BGB, a linear model provided the best fit. In the Figures the age of some plots was offset by 0.5–1 year to aid interpretation however models were fit using the real ages.

reduction with size in the likelihood of finding a resprouting individual (45% for saplings to 20% for trees >20 cm). Despite the differences in both diameter–height and root allometry, we found no clear differences in diameter–AGB allometry. This is likely the result of variations in wood density between simple and resprouting trees, with the simple trees likely have a lower wood density and thus comparable above-ground biomass than the shorter, denser resprouting trees. 4.2. Allometric modelling We adopted various approaches to construct our allometric models, eschewing the more conventional method of solely fitting linear models to log-transformed data. The issue of whether to log-transform biomass data has been the subject of considerable debate in the literature (Xiao et al., 2011; Packard, 2013), with some suggesting the decision is purely a matter of statistical convenience, while others highlight that the transformation is a necessary step as the allometry of many organisms is ‘multiplicative by nature’ (Kerkhoff and Enquist, 2009). We show here that there is no empirical justification to automatically log-transforming biomass data prior to the construction of allometric models, so long as the chosen non-linear alternatives are able to correctly reflect the error structure of the residuals. Studies which do not transform their data prior to analysis often fail to explain why they chose not to apply the more common log-transformation, or indicate whether the heteroscedasticity in the biomass data has been accounted for in the chosen model (e.g. Chan et al., 2013). Exemplifying this issue is that neither the weighted least squares method (M1), where the weights were initially set in accordance with Brown et al. (1989), nor the log-transformation of the biomass data for resprouting trees,

resulted in homoscedastic residuals. This highlights the limitations of setting the weights in advance as opposed to modelling the variance directly (M2), and also shows the need to critically examine the residual plots for each model to ensure that the assumptions of the regression analysis have been satisfied (Zuur et al., 2010; Picard et al., 2012). The models fit to log-transformed data with the correction factor applied (Model 4) resulted in broadly similar fits to both of the non-linear models (Models 1 and 2), whereas the log models without the CF applied (Model 3) unsurprisingly tended to underestimate biomass in comparison. Most studies automatically include the correction factor in order to account for the bias in the back transformation to real units (Chave et al., 2005; Mugasha et al., 2013), yet it was our model without the correction factor that ultimately resulted in the most accurate predictions for both AGB and BGB across the range of tree sizes sampled here. This helps to reinforce our earlier point of avoiding a ‘cookbook’ style of analysis and that simply using the conventional methods for constructing allomteric models may not always provide the best fit for the dataset in question (Niklas, 2006; Litton and Kauffman, 2008; Návar, 2009; Packard, 2013). However, the use of linear fitting methods (i.e. OLS) have additional benefits in that prediction intervals surrounding the estimated biomass values are simpler to construct within software packages like R, where there are in-built functions to calculate these rather than trying to calculate these manually (Gotelli and Ellison, 2004; Picard et al., 2012). This latter issue is important step in propagating the uncertainties on any new predictions to plot level estimates (Chave et al., 2004). We find that a model that ignores resprout status is suitable for estimating AGB; this is true both when DBH is used alone, and when it is combined with tree height, thereby simplifying forest inventories where the focus is solely on quantifying AGB, such as

114

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

for the calibration of earth observation data. Incorporating tree height into the allometric equations for AGB resulted in a statistically significant improvement in model performance; however for all comparable models forms the increase in explanatory power and reduction in the overall bias was marginal (Basuki et al., 2009; Ryan et al., 2011; Návar et al., 2013), contrasting several regional and pantropical studies which have shown the importance of height in explaining biomass variation (Chave et al., 2005; Feldpausch et al., 2012; Chan et al., 2013; Rutishauser et al., 2013; Manuri et al., 2014). This may reflect a lower competition for light in these relatively open secondary forests meaning trees have less incentive to grow tall, creating less spatial variability in diameter–height allometry compared to denser tropical forests where light limitation strongly controls tree community structure. Hence, the remaining variance in biomass between similarly sized trees that was not explained by DBH alone is therefore most likely modulated by differences in wood density, supporting the conclusion of Chave et al. (2005) that wood density is a more important predictor of AGB than tree height. Tree height was also found to be irrelevant for predicting BGB, being an insignificant parameter in all models where it was included as a separate covariate, whilst the compound variable D2H resulted in a poorer fit to the data than models fit using the same approach that used DBH alone. In contrast to AGB, the resprout status of the tree needs to be included in models of root biomass as simple and resprouting trees having markedly different belowground allometries (Figs. 3 and 6). Given that tree height did not result in major improvement in predictive power for estimating AGB, and is not required for estimating BGB, we can conclude that models including DBH as the only predictor variable are adequate in this system (Tables A1 and A2). Thus, we suggest that field efforts should be directed towards checking resprouting status, and away for height measurements for accurate biomass estimation, at least on a subset of trees in order to generate an estimate of the proportion in each stand that are resprouting as implemented here, 4.3. Comparisons with other models and implications for biomass estimations in swidden fallows The predictions of AGB generated by our best-fit DBH-only model were in broad agreement with the other secondary forest models developed in same region (Kenzo et al., 2009; Chan et al., 2013), at least across range of tree sizes sample here (Figs. 5 and 7). In common with these studies, we found that the Chave et al. (2005) pantropical model for moist forests grossly over predicted tree biomass in these secondary forests (van Breugel et al., 2011). In contrast, the Chave et al. (2014) model, which incorporated more datasets from dry and moist forests, including harvest data from other regularly disturbed ecosystems such as African savannah woodlands (Ryan et al., 2011; Mugasha et al., 2013) where resprouting is known to be common (Chidumayo, 2004; Luoga et al., 2004) resulted in more accurate estimates of tree biomass. Despite this, the mean relative error compared to measured biomass was still higher than both site-specific and regionally developed models, meaning these models should still be favoured when estimating tree AGB based on the tree diameter alone. Both the Chan et al. (2013) and Kenzo et al. (2009) models tended to underestimate AGB compared to our new model, mainly in the young to intermediate aged fallows where smaller stems (<15 cm) dominated. Although these overestimates in stand-level AGB estimates were typically small (<3 Mg/ha: 10–16% relative error), these minor discrepancies may have a major impact on landscape-level estimates of biomass storage with fallows of <10–15 years old estimated to make up 55% of our study area (Hett et al., 2012). When height data are available, we found that our best-fit model yielded very similar predictions of AGB to the

pan-tropical models that included height (Chave et al., 2005, 2014) meaning either can be used when both measures are available. These similarities indicate that the large differences in biomass found when using the DBH-only models are attributable to the trees in this study being smaller in stature relative to DBH compared those used to calibrate the pan-tropical models, further indicating the differences in allometry of trees growing in swidden areas. In terms of BGB, the expected finding of resprouting trees having larger roots in swidden fallows was shown to lead to a significant increase on estimates of root biomass in these systems. Previous estimates were biased low as they were based on models either from primary dipterocarp forest (Niiyama et al., 2010), or from post fire secondary forests (Kenzo et al., 2009), both of which have not been subjected to the same disturbance regime. Allometric models for below ground biomass in Southeast Asia are scarce (Yuen et al., 2013), and we were not able to locate any studies that described allometric relationships for BGB in swidden areas. At the stand level the ratio of above- to below-ground biomass was estimated at 0.32, lower than the IPCC default root-to-shoot ratio of 0.42 for secondary tropical forests (Penman et al., 2003), but greater than the estimated ratio in primary tropical forest ecosystems (0.20–0.24; Mokany et al., 2006) which is similar to the individual R:S ratio for simple trees in the study. Ascertaining the re-sprouting status of every tree is likely to be a very laborious task and classification may not be obvious based on an aboveground assessment alone; therefore we propose that our stand-level R:S ratio of 0.32, which was relatively consistent across all 12 plots, can be applied to the AGB data to estimate BGB stocks. However this could result in errors of unknown magnitude in areas where resprouts are either more, or less common than in our study plots. Our results in combination with previous studies suggest that the use of models that are not specific to swidden cultivation, or another disturbed system could risk greatly over-estimating biomass. This is an important consideration for studies that aim to understand emissions related to transition from relatively undisturbed forest to swidden cultivation as our results indicate that the use of the same generic allometric models for both land cover types would under-estimate emissions from this type of land use change. Conversely, applying either one, or a combination of the current regional models of both above- and below-ground biomass will likely result in an underestimation of the biomass stored in re-growing swidden fallows, thus undervaluing their role as a carbon sink (Pan et al., 2011). We therefore recommend the use of our allometrics for accurately estimating both the above- and below-ground biomass of trees in swidden fallows of upland Southeast Asia.

5. Conclusions This study is the first to develop allometric models of below-ground biomass for swidden cultivation landscapes of SE Asia. We also report a new model of above-ground biomass which supplements the small number of existing models for the region. Additional novelty was provided through our analysis of whether above- and below-ground allometries differed between trees re-growing from seed, and those resprouting from rootstocks. We found clear differences in the below-ground allometries of different tree types with the resprouting trees exhibiting significantly greater root biomass stocks. Resprouts were found to be less common than trees growing from seed; however the large differences in root biomass between the two tree types means that accounting for resprout allometry in forest inventories is a necessary step in accurately quantifying carbon stocks in swidden systems.

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

Acknowledgements This research is part of the project entitled Impacts of Reducing Emissions from Deforestation and Forest Degradation and Enhancing Carbon Stocks (I-REDD+). I-REDD+ is funded by the European Community’s Seventh Framework Research Programme. More information can be found on the web site: http://www.i-redd.eu. For assistance with fieldwork and processing samples we are grateful to Phaeng Xaphokhame and his team at the Viengkham District Agriculture and Forestry Office, Mouang-Muoy Agricultural Research Centre; Somvilay ChanThalounnavong from the National University of Laos, Department of Forestry; and farmers from Mouang-Muoy and surrounding villages. Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foreco.2015.07. 029. References Banin, L., Feldpausch, T.R., Phillips, O.L., Baker, T.R., Lloyd, J., Affum-Baffoe, K., Arets, E.J.M.M., Berry, N.J., Bradford, M., Brienen, R.J.W., Davies, S., Drescher, M., Higuchi, N., Hilbert, D.W., Hladik, a., Iida, Y., Salim, K.A., Kassim, a.R., King, D.a., Lopez-Gonzalez, G., Metcalfe, D., Nilus, R., Peh, K.S.-H., Reitsma, J.M., Sonké, B., Taedoumg, H., Tan, S., White, L., Wöll, H., Lewis, S.L., 2012. What controls tropical forest architecture? Testing environmental, structural and floristic drivers. Glob. Ecol. Biogeogr., n/a–n/a. Basuki, T.M., van Laake, P.E., Skidmore, a.K., Hussin, Y.a., 2009. Allometric equations for estimating the above-ground biomass in tropical lowland Dipterocarp forests. For. Ecol. Manage. 257, 1684–1694. Bond, W.J., Midgley, J.J., 2001. Ecology of sprouting in woody plants: the persistence niche. Trends Ecol. Evol. 16, 45–51. Brearley, F.Q., 2011. Below-ground secondary succession in tropical forests of Borneo. J. Trop. Ecol. 27, 413–420. Van Breugel, M., Ransijn, J., Craven, D., Bongers, F., Hall, J.S., 2011. Estimating carbon stock in secondary forests: decisions and uncertainties associated with allometric biomass models. For. Ecol. Manage. 262, 1648–1657. Brown, S., Gillespie, A.J.R., Lugo, A.E., 1989. Biomass estimation methods for tropical forests with applications to forest inventory data. For. Sci. 35, 881–902. Bruun, T.B., de Neergaard, A., Lawrence, D., Ziegler, A.D., 2009. Environmental consequences of the demise in Swidden cultivation in Southeast Asia: carbon storage and soil quality. Hum. Ecol. 37, 375–388. Burnham, K., Anderson, D., 2002. Model Selection and Multi- model Inference: A Practical Information-Theoretic Approach, second ed. Springer, New York. Chan, N., Takeda, S., Suzuki, R., Yamamoto, S., 2013. Establishment of allometric models and estimation of biomass recovery of swidden cultivation fallows in mixed deciduous forests of the Bago Mountains, Myanmar. For. Ecol. Manage. 304, 427–436. Chave, J., Andalo, C., Brown, S., Cairns, M.a., Chambers, J.Q., Eamus, D., Fölster, H., Fromard, F., Higuchi, N., Kira, T., Lescure, J.-P., Nelson, B.W., Ogawa, H., Puig, H., Riéra, B., Yamakura, T., 2005. Tree allometry and improved estimation of carbon stocks and balance in tropical forests. Oecologia 145, 87–99. Chave, J., Condit, R., Aguilar, S., Hernandez, A., Lao, S., Perez, R., 2004. Error propagation and scaling for tropical forest biomass estimates. Philos. Trans. Roy. Soc. Lond. Ser. B-Biol. Sci. 359, 409–420. Chave, J., Réjou-Méchain, M., Búrquez, A., Chidumayo, E., Colgan, M.S., Delitti, W.B., Duque, A., Eid, T., Fearnside, P.M., Goodman, R.C., Henry, M., Martínez-Yrízar, A., Mugasha, W.a., Muller-Landau, H.C., Mencuccini, M., Nelson, B.W., Ngomanda, A., Nogueira, E.M., Ortiz-Malavassi, E., Pélissier, R., Ploton, P., Ryan, C.M., Saldarriaga, J.G., Vieilledent, G., 2014. Improved allometric models to estimate the aboveground biomass of tropical trees. Glob. Change Biol., 1–14. Chidumayo, E.N., 2004. Development of Brachystegia-Julbernardia woodland after clear-felling in central Zambia: evidence for high resilience. Appl. Veg. Sci. 7, 237. Clarke, P.J., Lawes, M.J., Midgley, J.J., Lamont, B.B., Ojeda, F., Burrows, G.E., Enright, N.J., Knox, K.J.E., 2012. Resprouting as a key functional trait: how buds, protection and resources drive persistence after fire. New Phytol., 19–35. Cramb, R.A., Colfer, C.J.P., Dressler, W., Laungaramsri, P., Le, Q.T., Mulyoutami, E., Peluso, N.L., Wadley, R.L., 2009. Swidden transformations and rural livelihoods in Southeast Asia. Hum. Ecol. 37, 323–346. Cunia, T., 1964. Weighted least squares method and construction of volume tables. For. Sci. 10, 180–191. Farr, T., Rosen, P., Caro, E., 2007. The shuttle radar topography mission. Rev. Geophys. 45, 1–33. Feldpausch, T.R., Lloyd, J., Lewis, S.L., Brienen, R.J.W., Gloor, M., Monteagudo Mendoza, A., Lopez-Gonzalez, G., Banin, L., Abu Salim, K., Affum-Baffoe, K.,

115

Alexiades, M., Almeida, S., Amaral, I., Andrade, A., Aragão, L.E.O.C., Araujo Murakami, A., Arets, E.J.M.M., Arroyo, L., Aymard, C.G.A., Baker, T.R., Bánki, O.S., Berry, N.J., Cardozo, N., Chave, J., Comiskey, J.a., Alvarez, E., de Oliveira, A., Di Fiore, A., Djagbletey, G., Domingues, T.F., Erwin, T.L., Fearnside, P.M., França, M.B., Freitas, M.A., Higuchi, N., Iida, Y., Jiménez, E., Kassim, A.R., Killeen, T.J., Laurance, W.F., Lovett, J.C., Malhi, Y., Marimon, B.S., Marimon-Junior, B.H., Lenza, E., Marshall, A.R., Mendoza, C., Metcalfe, D.J., Mitchard, E.T.A., Neill, D.A., Nelson, B.W., Nilus, R., Nogueira, E.M., Parada, A., Peh, K.S.-H., Pena Cruz, A., Peñuela, M.C., Pitman, N.C.A., Prieto, A., Quesada, C.A., Ramírez, F., RamírezAngulo, H., Reitsma, J.M., Rudas, A., Saiz, G., Salomão, R.P., Schwarz, M., Silva, N., Silva-Espejo, J.E., Silveira, M., Sonké, B., Stropp, J., Taedoumg, H.E., Tan, S., ter Steege, H., Terborgh, J., Torello-Raventos, M., van der Heijden, G.M.F., Vásquez, R., Vilanova, E., Vos, V.A., White, L., Willcock, S., Woell, H., Phillips, O.L., 2012. Tree height integrated into pantropical forest biomass estimates. Biogeosciences 9, 3381–3403. Goodman, R.C., Phillips, O.L., del Castillo Torres, D., Freitas, L., Cortese, S.T., Monteagudo, A., Baker, T.R., 2013. Amazon palm biomass and allometry. For. Ecol. Manage. 310, 994–1004. Gotelli, N.J., Ellison, A.M., 2004. A Primer of Ecological Statistics. Sinauer, Sunderland, Massachusetts, USA. Hansen, M.C., Potapov, P.V., Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau, D., Stehman, S.V., Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L., Justice, C.O., Townshend, J.R.G., 2013. High-resolution global maps of 21st-century forest cover change. Science (New York, N.Y.) 342, 850–853. Hett, C., Castella, J.-C., Heinimann, A., Messerli, P., Pfund, J.-L., 2012. A landscape mosaics approach for characterizing swidden systems from a REDD+ perspective. Appl. Geogr. 32, 608–618. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978. Holdridge, L.R., 1967. Life Zone Ecology. Tropical Science Center, San Jose, Costa Rica. Hurni, K., Hett, C., Heinimann, A., Messerli, P., Wiesmann, U., 2012. Dynamics of shifting cultivation landscapes in Northern Lao PDR between 2000 and 2009 based on an analysis of MODIS time series and landsat images. Hum. Ecol. 41, 21–36. Kenzo, T., Ichie, T., Hattori, D., Itioka, T., Handa, C., Ohkubo, T., Kendawang, J.J., Nakamura, M., Sakaguchi, M., Takahashi, N., Okamoto, M., Tanaka-Oda, A., Sakurai, K., Ninomiya, I., 2009. Development of allometric relationships for accurate estimation of above- and below-ground biomass in tropical secondary forests in Sarawak, Malaysia. J. Trop. Ecol. 25, 371. Kerkhoff, A.J., Enquist, B.J., 2009. Multiplicative by nature: why logarithmic transformation is necessary in allometry. J. Theor. Biol. 257, 519–521. Ketterings, Q.M., Coe, R., van Noordwijk, M., Ambagau’, Y., Palm, C.a., 2001. Reducing uncertainty in the use of allometric biomass equations for predicting aboveground tree biomass in mixed secondary forests. For. Ecol. Manage. 146, 199– 209. Kutner, M., Nachtsheim, C., Neter, J., Li, W., 2005. Applied Linear Statistical Models, fifth ed. McGraw-Hill/Irwin, New York, NY, USA. Litton, C.M., Kauffman, J.B., 2008. Allometric models for predicting aboveground biomass in two widespread woody plants in Hawaii. Biotropica 40, 313–320. Luoga, E.J., Witkowski, E.T., Balkwill, K., 2004. Regeneration by coppicing (resprouting) of miombo (African savanna) trees in relation to land use. For. Ecol. Manage. 189, 23–35. Manuri, S., Brack, C., Puji, N., Hergoualch, K., Novita, N., Dotzauer, H., Verchot, L., Agung, C., Putra, S., Widyasari, E., 2014. Tree biomass equations for tropical peat swamp forest ecosystems in Indonesia. For. Ecol. Manage. 334, 241–253. Mascaro, J., Litton, C., Hughes, R., 2011. Minimizing bias in biomass allometry: model selection and log-transformation of data. Biotropica 43, 649–653. Mertz, O., Müller, D., Sikor, T., Hett, C., Heinimann, A., Castella, J.-C., Lestrelin, G., Ryan, C.M., Reay, D.S., Schmidt-Vogt, D., Danielsen, F., Theilade, I., Van Noordwijk, M., Verchot, L.V., Burgess, N.D., Berry, N.J., Pham, T.T., Messerli, P., Xu, J., Fensholt, R., Hostert, P., Pflugmacher, D., Bruun, T.B., De Neergaard, A., Dons, K., Dewi, S., Rutishauser, E., Sun, Z., 2012. The forgotten D: challenges of addressing forest degradation in complex mosaic landscapes under REDD+. Geogr. Tidsskr.-Dan. J. Geogr. 112, 63–76. Messerli, P., Heinimann, A., Epprecht, M., 2009. Finding homogeneity in heterogeneity – a new approach to quantifying landscape mosaics developed for the Lao PDR. Hum. Ecol.: Interdiscipl. J. 37, 291–304. Mokany, K., Raison, R.J., Prokushkin, A.S., 2006. Critical analysis of root: shoot ratios in terrestrial biomes. Glob. Change Biol. 12, 84–96. Mugasha, W.A., Eid, T., Bollandsås, O.M., Malimbwi, R.E., Chamshama, S.A.O., Zahabu, E., Katani, J.Z., 2013. Allometric models for prediction of above- and belowground biomass of trees in the miombo woodlands of Tanzania. For. Ecol. Manage. 310, 87–101. Müller, D., Sun, Z., Vongvisouk, T., Pflugmacher, D., Xu, J., Mertz, O., 2014. Regime shifts limit the predictability of land-system change. Glob. Environ. Change 28, 75–83. Návar, J., 2009. Biomass component equations for Latin American species and groups of species. Ann. For. Sci. 66. Návar, J., 2010. Measurement and Assessment Methods of Forest Aboveground Biomass: A Literature Review and the Challenges Ahead. Biomass. Sciyo Open Access to Knowledge, Intechopen, pp. 27–64. Návar, J., Ríos-Saucedo, J., Pérez-Verdín, G., Rodríguez-Flores, F.D.J., DomínguezCalleros, P.a., 2013. Regional aboveground biomass equations for North American arid and semi-arid forests. J. Arid Environ. 97, 127–135.

116

I.M. McNicol et al. / Forest Ecology and Management 357 (2015) 104–116

Niiyama, K., Kajimoto, T., Matsuura, Y., Yamashita, T., Matsuo, N., Yashiro, Y., Ripin, A., Kassim, A.R., Noor, N.S., 2010. Estimation of root biomass based on excavation of individual root systems in a primary dipterocarp forest in Pasoh Forest Reserve, Peninsular Malaysia. J. Trop. Ecol. 26, 271. Niklas, K.J., 2006. A phyletic perspective on the allometry of plant biomasspartitioning patterns and functionally equivalent organ-categories. New Phytol. 171, 27–40. Nzunda, E.F., Griffiths, M.E., Lawes, M.J., 2014. Resource allocation and storage relative to resprouting ability in wind disturbed coastal forest trees. Evol. Ecol. 28, 735–749. Packard, G.C., 2013. Is logarithmic transformation necessary in allometry? Biol. J. Linn. Soc. 109, 476–486. Packard, G.C., 2012. Is non-loglinear allometry a statistical artifact? Biol. J. Linn. Soc. 107, 764–773. Padoch, C., Coffey, K., Mertz, O., Wadley, R., Leisz, S., Fox, J., 2007. The demise of swidden in Southeast Asia? Local realities and regional ambiguities. Geogr. Tidsskr. Dan. J. Geogr. 107, 29–41. Pan, Y., Birdsey, R.a., Fang, J., Houghton, R., Kauppi, P.E., Kurz, W.a., Phillips, O.L., Shvidenko, A., Lewis, S.L., Canadell, J.G., Ciais, P., Jackson, R.B., Pacala, S.W., McGuire, a.D., Piao, S., Rautiainen, A., Sitch, S., Hayes, D., 2011. A large and persistent carbon sink in the world’s forests. Science (New York, N.Y.) 333, 988– 993. Penman, J., Gytarsky, M., Hiraishi, T., Krug, T., Kruger, D., Pipatti, R., Buendia, L., Miwa, K., Ngara, T., 2003. Good Practice Guidance for Land Use, Land-Use Change and Forestry, Intergovernmental Panel on Climate Change National Greenhouse Gas Inventories Programme. Institute for Global Environmental Strategies (IGES) for the IPCC, Hayama, Kanagawa, Japan. Picard, N., Boyemba Bosela, F., Rossi, V., 2014. Reducing the error in biomass estimates strongly depends on model selection. Ann. For. Sci. Picard, N., Saint-André, L., Henry, M., 2012. Manual for building tree volume and biomass allometric equations: from field measurement to prediction. Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., 2014. Linear and nonlinear mixed effects models. R Core Team, 2014. R: A Language and Environment for Statistical Computing. Rerkasem, K., Lawrence, D., Padoch, C., Schmidt-Vogt, D., Ziegler, A.D., Bruun, T.B., 2009. Consequences of swidden transitions for crop and fallow biodiversity in Southeast Asia. Hum. Ecol. 37, 347–360. Ritz, C., Streibig, J., 2008. Nonlinear regression with R.

Rutishauser, E., Noor’an, F., Laumonier, Y., Halperin, J., Hergoualc’h, K., Verchot, L., 2013. Generic allometric models including height best estimate forest biomass and carbon stocks in Indonesia. For. Ecol. Manage. 307, 219–225. Ryan, C.M., Williams, M., Grace, J., 2011. Above- and belowground carbon stocks in a Miombo woodland landscape of Mozambique. Biotropica 43, 423–432. Shanmughavel, P., Zheng, Z., Liqing, S., Min, C., 2001. Floristic structure and biomass distribution of a tropical seasonal rain forest in Xishuangbanna, southwest China. Biomass Bioenergy 21, 165–175. Skole, D.L., Samek, J.H., Smalligan, M.J., 2011. Implications of allometry. Proc. Natl. Acad. Sci. U.S.A. 108, E12 (author reply E13-4). Sprugel, D., 1983. Correcting for bias in log-transformed allometric equations. Ecology, 209–210. Vongvisouk, T., Mertz, O., Thongmanivong, S., Heinimann, A., Phanvilay, K., 2014. Shifting cultivation stability and change: contrasting pathways of land use and livelihood change in Laos. Appl. Geogr. 46, 1–10. Williams, M., Ryan, C.M., Rees, R.M., Sambane, E., Femando, J., Grace, J., 2008. Carbon sequestration and biodiversity of re-growing miombo woodlands in Mozambique. For. Ecol. Manage. 254, 145–155. Xiao, X., White, E., Hooten, M., Durham, S., 2011. On the use of log-transformation vs. nonlinear regression for analyzing biological power laws. Ecology 92, 1887– 1894. Yuen, J.Q., Ziegler, A.D., Webb, E.L., Ryan, C.M., 2013. Uncertainty in below-ground carbon biomass for major land covers in Southeast Asia. For. Ecol. Manage. 310, 915–926. Ziegler, A.D., Bruun, T.B., Guardiola-Claramonte, M., Giambelluca, T.W., Lawrence, D., Thanh Lam, N., 2009. Environmental consequences of the demise in swidden cultivation in montane mainland Southeast Asia: hydrology and geomorphology. Hum. Ecol. 37, 361–373. Ziegler, A.D., Fox, J.M., Webb, E.L., Padoch, C., Leisz, S.J., Cramb, R.a., Mertz, O., Bruun, T.B., Vien, T.D., 2011. Recognizing contemporary roles of swidden agriculture in transforming landscapes of Southeast Asia. Conserv. Biol. 25, 846–848. Ziegler, A.D., Phelps, J., Yuen, J.Q., Webb, E.L., Lawrence, D., Fox, J.M., Bruun, T.B., Leisz, S.J., Ryan, C.M., Dressler, W., Mertz, O., Pascual, U., Padoch, C., Koh, L.P., 2012. Carbon outcomes of major land-cover transitions in SE Asia: great uncertainties and REDD+ policy implications. Glob. Change Biol. 18, 3087–3099. Zuur, A.F., Ieno, E.N., Elphick, C.S., 2010. A protocol for data exploration to avoid common statistical problems. Methods Ecol. Evol. 1, 3–14.