Developing biomass estimation models for above-ground compartments in Eucalyptus dunnii and Corymbia citriodora plantations

Highlights

• Power-law and combined models were significant and well fitted (p<0.05, R²>0.9).

• Multiple linear and polynomial models had unreliable and collinear coefficients.

• Stem biomass relationships were stronger when using the compound variable DBH²Hp.

• DBH alone was the best predictor variable for bark and crown biomass estimations.

• High-predictability models for branch biomass were derived from branch diameter.

Abstract

Biomass has been widely studied in terms of ecosystem ecology, timber production profitability, bioenergy (biofuels) and greenhouse gas emission reduction mechanisms. However, uncertainty in biomass estimation is still a current concern. In this study, direct and indirect methods were used to develop species-specific biomass estimation models (BEMs) for stem, bark, branch and crown compartments in 16-year old plantations of Eucalyptus dunnii and Corymbia citriodora. A total of 93 trees were destructively sampled. An analysis of covariance (ANCOVA) assessed the effect of species on biomass prediction. Our results indicated that equations developed by using parameters or predictors such as diameter at breast height (DBH), height (H), wood density (p) and branch diameter were generally significant (p < 0.05) and their regression lines fitted well the data (R² > 0.84). After a rigorous process that included testing hypotheses, checking diagnostic statistics, assessing model coefficients and model functionality, the most suitable stem BEMs corresponded to those ones derived from the compound variable DBH²Hp. The most reliable branch and crown BEMs used DBH and branch diameter respectively as single variable (simple linear models). Bark BEMs differ between species as DBH was the best predictor for E. dunnii whilst the compound variable DBH H predicted better for C. citriodora. The BEMs with multiple predictors, and in particular polynomial models, produced wider confidence intervals, unreliable coefficients, multicollinearity and higher proportion of outliers and leverage points. In conclusion, appropriate model diagnosis can reduce pitfalls and ensure selection of valid BEMs.

Introduction

Biomass, and particularly above-ground biomass (AGB), is one of the most measured variables in vegetation systems, widely used to understand ecological and management processes in ecosystems, and to quantify timber products in the forestry industry [1,2]. Biomass has been also studied to derive estimates of quantities of biofuels and biogas as part of renewable energy alternatives [3], and to quantify carbon stocks under the greenhouse gas (GHG) emission reduction mechanisms [[4], [5], [6]]. Tree biomass estimation includes direct and indirect methods. Direct methods can involve field measurements and destructive procedures whilst indirect methods include statistical modelling (regression analyses) and remote sensing techniques [[7], [8], [9], [10], [11]]. Undertaking field measurements in conjunction with destructive sampling is a crucial step to yield precise data to develop biomass estimation models (BEMs) [7,15,16]. Destructive sampling consists of harvesting trees, and separating and weighing their compartments, being an expensive and time consuming procedure. However, it has been considered the most accurate method of estimating biomass [3,[12], [13], [14]]. Data produced from destructive sampling is used in regression analyses to develop biomass estimation models (BEMs) [15,16] and is also widely used to validate results from existing BEMs and biomass estimates obtained from remote sensing analyses [12,23].

The use of BEMs is the basis for estimating tree biomass, by using easy-to-measure parameters (e.g. DBH, height) [13,17,18]. These BEMs are mathematical expressions that quantify proportionality relationships between dimensions of tree characteristics as responses to ontogenic development [[19], [20], [21]]. Developing BEMs becomes a key process to predict biomass of individual trees sampled in forest inventory plots [21]. Tree biomass can be used to obtain plot level estimates, then to extrapolate mean values to a whole population [6,12]. These BEMs can be 1) generalised or 2) specific equations (species-, site-, age-specific) [13]. The use of generalised equations on different species and ecosystem types can be problematic, yielding around 15–25% error on biomass estimations [1,6,17,21,22]. Therefore, species-specific equations are preferable to produce reliable estimates [5,12,14,22,23].

Reducing uncertainty and bias during the development of BEMs can improve the accuracy of biomass estimates [24,25]. Undertaking model diagnosis and the assessment of appropriate statistics to select suitable equations are required to reduce sources of errors, and therefore, produce statistically reliable and biologically realistic models [26]. Generalised as well as species-specific BEMs have been developed for some hardwood species belonging to Eucalyptus and Corymbia genera [16,18,22,27,28], but not to the same extent as for softwood species. In addition, existing equations commonly focus on the prediction of stem biomass given that this is the tree compartment that is merchantable, and little information is available for the remaining compartments, and therefore total AGB [3,17,29]. The stem can represent a large proportion of the total tree biomass, however, the inclusion of other components such as bark and in particular crown, allows assessing the manner in which trees allocate biomass to each compartment. The biomass partitioning reflects the distribution of the net primary productivity (NPP) in response to the different species genetic characteristics, physiological processes and environmental conditions [12,30].

In Australia, the hardwood plantation state is dominated by several eucalypts species [31], and two of them, Dunn's White Gum (Eucalyptus dunnii Maiden) and Spotted Gum (Corymbia citriodora subsp. Variegata (F.Muell.) A.R.Bean & M.W.McDonald, hereafter referred to as C. citriodora) are of particular interest for this research. In subtropical regions of Australia, E. dunnii has been primarily planted for sawlog and pulp production whilst C. citriodora has been widely used for structural timber [31]. Although E. dunnii is naturally distributed in Australia [32], this species has been successfully grown in China, Africa and South America (Brazil and Argentina) to produce timber suitable for a range of uses [33]. With a wider distribution range, C. citriodora occurs naturally in Australia and New Zealand, and has also been established for timber and essential oil production in the Mediterranean area, southern China, southern Africa, South-East Asia, South America and North America [32,34,35]. Research is increasing in fields such as the production of biofuels and electricity from thinning residues (low-value and defective stems) [36], and the estimation of biomass and sequestered carbon [6] for these species and other subtropical eucalypts.

This research was carried out in 16-year old hardwood plantations of E. dunnii and C. citriodora established in 2000/2001 in north-eastern New South Wales (NSW), Australia. In the study area, thinning practices undertaken at eight years after planting to reduce competition between trees and increase yield of high-quality timber products, resulted in stands of residual stocking densities of 300 (heavily thinned) and 600 stems ha⁻¹ (moderately thinned). Unthinned areas (1200 stems ha⁻¹) represented the “Control”. As no BEMs are available for these two hardwood species, this study aimed to 1) develop equations from harvested data to predict biomass in stem, bark, branch and crown compartments and 2) select the most suitable BEMs for each species and tree compartments through a rigorous process of testing hypotheses, checking diagnostic statistics, confirming reliability of model coefficients, and assessing model functionality.