Variations in the biomass of Eucalyptus plantations at a regional scale in Southern China

Quanyi Qiu•Guoliang Yun•Shudi Zuo•Jing Yan•Lizhong Hua•Yin Ren•Jianfeng Tang•Yaying Li•Qi Chen

Abstract We quantified deviations in regional forest biomass from simple extrapolation of plot data by the biomass expansion factor method(BEF)versus estimates obtained from a local biomass model,based on large-scale empirical field inventory sampling data.The sources and relative contributions of deviations between the two models were analyzed by the boosted regression trees method.Relative to the local model,BEF overestimated accumulative biomass by 22.12%.The predominant sources of the total deviation(70.94%)were stand-structure variables.Stand age and diameter at breast height are the major factors.Compared with biotic variables,abiotic variables had a smaller overall contribution(29.06%),with elevation and soil depth being the most important among the examined abiotic factors.Large deviations in regional forest biomass and carbon stock estimates are likely to be obtained with BEF relative to estimates based on local data.To minimize deviations,stand age and elevation should be included in regional forest-biomass estimation.

Keywords BEF ·Boosted regression trees·Eucalyptus plantations·Local biomass model·Regional biomass estimation ·Biotic versus abiotic factors·Uncertainty analysis

Introduction

Rapid population growth begets an increased demand for timber,which degrades forest resources.To alleviate timber-supply shortages,many countries—including Australia,Brazil,Spain,Portugal,China,Congo,India,and New Zealand—have developedEucalyptusas a major economic planation type(Jing et al.2012).However,expansiveEucalyptusplantations have caused soil hardening,deterioration of soil fertility,and reduction of biodiversity.These new problems have made fast-growthEucalyptusplantations highly contentious(Arias et al.2011).It is hoped that precise prediction ofEucalyptusbiomass at various temporal and spatial scales through scale-transformation methods can provide a scientific base for sustainable management of fast-growth plantation forestry.

There has been an international surge inEucalyptusbiomass estimation research in the literature since the year 2000,with tree-level,stand-level,and regional-level analyses being the major spatial scales used(le Maire et al.2011).Around half of these studies focused onEucalyptusonly plantations,while the remainder examined plantations ofEucalyptusmixed withAcacia,Pinus(pines),andQuercus(oaks).

Researchers distinguish above-ground biomass(stem,branch,and leaf)and under-ground biomass(root)from whole-plant biomass(Dube and Mutanga 2016;Sochacki et al.2017).Most studies have focused on(1)cost ef ficiency and the representativeness of samples(Raimundo et al.2017);(2)biomass mapping ofEucalyptusplantations of different ages(Silva et al.2015);and(3)comparing the timber production and carbon-sequestration potential of different forest management scenarios(Shankhwar and Srivastava 2015).Most of the studies included error and precision estimates.Error is generally represented by indicators,while precision is reflected by root-mean-square error,prediction error,and sum of squares for error(Gama et al.2016;Padalia and Yadav 2017;Paul et al.2013).However,little quantitative research exists on the multiplicity of biotic and abiotic factors that contribute to detected errors.

The three biomass-estimation methods employed forEucalyptusplantations are field survey,remote-sensing monitoring,and ecological process models.Generally,small-scaleEucalyptusforest-biomass estimation is conducted by way of the harvest method,which is appreciated for its high accuracy.However,the harvest method is time consuming and requires too much manpower and material resources to be suitable for large-scale forest biomass estimation.

Remote sensing and model simulation are used to estimateEucalyptusforest biomass on larger scales.However,the results of remote sensing provide only an indirect estimation of biomass because researchers cannot ensure that such data are always obtained on cloudless days,and the effects of undergrowth biomass are not accounted for.Although models and simulations are easy to manage,the great number of model parameters that need to be inputted without empirical verification will introduce substantial uncertainty(Basuki et al.2009;Teobaldelli et al.2009;Vieilledent et al.2012;Kalboussi and Achour 2017).

Going forward,biomass estimate methods forEucalyptusplantations are expected to involve integration of technology with surface investigations,remote-sensing monitoring,and ecological process models.Field survey methodology that involves collection of tree destruction and harvest data to inform biomass-estimation parameters is fundamental for technological integration into biomass calculation methodology.Innovation in this realm represents an opportunity to verify remote-sensing monitoring and ecological process models at multiple spatial scales(Chen et al.2016;Paul et al.2016).

The biomass expansion factor method(BEF)provides a bridge for the estimation of forest biomass between large and small scales and is especially suitable for midscale estimates(regional scale)(Ren et al.2011,2013).Forest biomass estimation on a regional scale that is done by combining forest-inventory data and field-sampling data can serve as an effective approach to achieve scale transformation from small to large scales(Petersson et al.2012).Generally,BEF uses a continuous ratio of stand biomassto-volume,multiplied by the total volume of a forest,to calculate total biomass of a specific forest type.

Alternatively,BEF can be conducted based on wood density multiplied by total volume and a conversion coefficient(Basuki et al.2009;Jalkanen et al.2005).Thus,BEF provides a simple method of enabling the transfer of information from stand-level field surveys to regional-level estimation and allows forest biomass to be calculated on a regional scale(Stinson et al.2011).On a sample scale,local values are obtained by way of the regional harvest method;then this information is combined with forest-resource inventory data to calculate forest biomass on a regional scale with BEF(Ren et al.2012).

For example,by using 758 sets of volume-biomass field dataand plot orstand-based forest inventory data,researchers have estimated forest biomass in China using a linear formula(B=a+b V)to establish the relationship between volume(V)and biomass(B)and then conducted a simple scale transfer to calculate regional forest biomass.Such an approach has been widely applied to the estimation of forest biomass or carbon stock on landscape and regional levels(Fang et al.2001;Goodale et al.2002;Pajtik et al.2008;Petersson et al.2012;Singh et al.2011;Teobaldelli et al.2009;Zhang et al.2013).Using data from one spatial scale to estimate biomass at another provides a means of achieving data conversion.However,in recent decades,caution has been recommended repeatedly in the application of this method because of the potential for large deviations(Luo et al.2013).

BEF may be prone to various sources of deviation.For example,BEF does not account for the impact of numerous biotic(e.g.,diameter at breast height(DBH)and stand density)and abiotic(e.g.,climate,topography,and soil character)factors,nor of man-made disturbances(Asner et al.2009;Axmanova et al.2011;Chapin et al.1987).To date,researchers have conducted studies related to the effects of several variables related to aboveground biomass,including the effects of climate(e.g.,temperature and precipitation)(Malhi et al.2006;Stegen et al.2009),topography(e.g.,elevation and slope)(Asner et al.2009;de Castilho et al.2006;Ferry et al.2010;Laumonier et al.2010),and soil properties(e.g.,texture and nutrients)(DeWalt and Chave 2004;Laurance et al.1999;Paoli et al.2008).

However,few studies have analyzed the relative contributions of such variables to biomass deviation synergistically(Asner et al.2009;Chapin et al.1987;Mascaro et al.2011).Current BEF methodology may not meet the demands of Intergovernmental Panel on Climate Change(IPCC)guidelines(IPCC 2003)due to the risks for large deviations.Though the existence of probable deviations with BEF at the national or regional scales have been acknowledged in many studies,these deviations have rarely been quantified(Jalkanen et al.2005).Quantification of these potential deviations(secondary to BEF uncertainties)is urgently needed to allow researchers to assess the actual uncertainties associated with estimation of regional or national forest biomass and carbon stock(DeWalt and Chave 2004;Laurance et al.1999;Paoli et al.2008;Miranda-Aragon et al.2012;Shahbudin et al.2012).

The objectives of this study were twofold.The first objective was to assess potential deviations associated with BEF by comparing BEF-derived data with a local biomass model.The second objective was to quantify the relative contributions of biotic and abiotic factors to deviations in biomass estimation.A large dataset onEucalyptusin Nanjing County,China,was analyzed in this study.We addressed two specific research questions:(1)Where do the differences in the spatial distribution of deviations in biomass estimation come from?(2)What factors are dominant in influencing dissociation between BEF-derived data and a local biomass model?We hypothesize that both biotic and abiotic factors have a significant influence on BEF versus local model differences in forest biomass estimates on a regional scale.If biotic factors are dominant,then researchers should pay more attention to factors such as stand age,DBH,and density during field sampling.If a biotic factors affect forest biomass estimates,then multiple sources of data and approaches,such as remote sensing,should be combined to map forest biomass on a regional scale.

Materials and methods

Study site

The 1962-km2Nanjing County study area(117°00′–117°36′E,24°26′–25°00′N)lies in a region of southeastern Fujian,China(Fig.1)that is often referred to as a sea of trees or bamboo ocean because forests cover 74%of the landscape.Nanjing County has a South Asian tropical monsoon climate with an annual average temperature of 21.1°C,an average annual precipitation of 1700 mm,and a frost-free period of more than 340 days.The main soil type is latosolic red soil.Forests cover 146,130 ha-1,of which 79,346 ha-1are plantations.The major forest types within the study areEucalyptusspp.,Pinus massonianaandCunninghamia lanceolata.

Eucalyptusspecies are the most commonly planted trees and now cover the largest number of hectares of any fast growing tree genus worldwide(Perez-Cruzado et al.2011).They are planted owing to their numerous advantages including high adaptability to different environments,a short cultivation rotation,high wood productivity,and usefulness for a wide variety of commercial purposes.In China,researchers have focused primarily on(1)the relationship between biomass and biodiversity at the stand and regional level,(2)changes in soil carbon dynamics,and(3)comparing biomass with productivity.

According to the Seventh National Forest Inventory of China(2004–2008),theEucalyptusplantation area exceeds 2.67×106ha-1,accounting for 4.2%of the total plantation area in China,80%of which is in southern regions,such as the Guangdong,Hainan,and Fujian provinces.TheEucalyptusplantation area in Nanjing has been increasing rapidly by 18.1×103ha-1per year for the past 10 years.These widespreadEucalyptusplantations provide an appropriate experimental setting for evaluation of biomass estimation.

Data sources

Fig.1 Maps of Nanjing County,Fujian,China;a vicinity map showing Fujian Province within China,b inset map of Fujian Province showing the vicinity of Nanjing County,c outline map of Nanjing County

China’s State Forestry Administration has been engaged in forest resource inventory and monitoring for over 40 years,and has established a forest resource geographic information system.Forest inventory data,collected by the State Forestry Administration,record detailed attribute data for each forest class and represent true forest conditions on a regional scale.This data compilation allows for systematic,longitudinal tracking of information,including forest growth,site index,human activity,and stand age,based on spatial sampling.Forest inventory data can be used to analyze the relationship between the dynamics of forest carbon stock and various environmental factors,and provides excellent carbon stock data on a regional scale when BEF is applied(Ren et al.2016).

The forest inventory data used in this paper come from 4176 pureEucalyptusplantation plots with data collected from 34,075 plots in Nanjing County,Fujian,China in 2009(Table 1),including data on stand area,species composition,stand age,average DBH,average height,stand density,volume,elevation,and site index.The 4176 plots ofEucalyptusin Nanjing County,represented in Fig.2 as polygons,include primarily the following species:Eucalyptus huges(3527);Eucalyptus grandis(358);Eucalyptus exserta(2);lemon-scented gum(12);Eucalyptus urophylla(162);and otherEucalyptusspecies(115).

Methods

This study employed data from sample plots that had been collected as a part of the Chinese Forest Management Planning and Inventory Program,specifically for Nanjing County,Fujian,China.These data describe the basic characteristics and land area of each stand type,and provide an excellent opportunity to assess the deviations associated with direct scale-transferring from plot to regional levels by comparing BEF-based estimates with standard trees from the Chinese Forest Management Planning and Inventory Program.In addition,we applied boosted regression tree(BRT)analysis to assess the relative contributions of various biotic and abiotic factors to any observed deviations.

Establishment of the local forest biomass model

A standard table of age-group and age-class divisions for short-rotation period timber forest(based on forest resourceand management data from Fujian Province)is provided(see the appendix in Ren et al.2017).Age group and class depended mainly on wood growth and cutting cycle during the application of forestry techniques.Eucalyptusforests were classified by age:young,≤2 years;middle-aged,3–4 years;near-mature,5–6 years;mature,7–10 years;and over-mature,≥11 years.Eucalyptusis considered a short rotation period timber in China and trees in the study region are harvested at 5–10 years old.Consequently,here,we consideredEucalyptusforests to be mature for harvest at 5–10 years old(near-mature or mature age).

Table 1 Descriptive statistics from the Chinese National Forest Inventory Program

Based on field surveys,30 typical 20 m×20 m sample plots ofEucalyptusplantations from 1 to 10 years old were selected for analysis.These 30 sample plots should reflect the general growth conditions for a given age,including topographical(elevation,degree,position,and direction of slope)and soil(soil depth,site index)characteristics.Each individual selected tree was wood gauged for a sample based on the diameter class;then the average DBH and height of standard trees was calculated.In all,three standard-sized trees were selected for each plot.A total of 90 trees(3 trees×30 samples)were selected and sampled in Nanjing County.

All selected trees were cut at ground level and all roots of each cut tree were carefully excavated and removed from the soil.For each tree,four representative samples of roots,stems,branches,and leaves were collected for estimation of dry mass in the laboratory.Moisture content of each part was also measured.We obtained the total biomass of only 90 sample trees because the trees all had the same moisture content allowing us to minimize the effort required;had there been differences in moisture content,then measurements from additional trees would have been needed.Finally,we calculated the total biomass of each sample by multiplying the standard tree biomass with each tree’s density,based on the records collected during field surveys.For detailed information about the process,see Ren et al.(2017).

The volume of each individual tree was calculated based on a Binary List of forest resources and management in Fujian Province(Table 2);then we used the individual tree volumes to obtain the total volume of each sample.Table 2 provides descriptive biomass data and statistics of the 30 plots,and Ren et al.(2017)provides detailed information for sample trees.

A local biomass model was then established based on Formula(1):

whereB(t hm-2)is total stand biomass(including above and below-ground biomass),V(m3ha-1)is stand volume,and a and b are biomass model parameters(Table 3).BEFwas conducted with the following commonly applied formula in China(2):

Fig.2 Spatial distribution of biomass differences(percentage,%)in Nanjing County,Fujian,China,using two different models.This map shows three main areas:(1)gray,non-forest land,(2)green,other forest land,and(3)Eucalyptus plantations.Plantations are indicated as:(1)red,forest biomass estimation in our study is lower than that measured using the previous BEF,and(2)blue,Eucalyptus plantation biomass estimation is higher(darker colors indicate extremes).Forest biomass estimation is indicated by light blue(0–40%),sky blue(40–80%)and deep blue(＞80%).In addition,other forest land includes several main types of forested land,such as pine and fir trees,plantation forests,bamboo forests,open forest land,shrubland,and land used for tree nurseries;non-forest land includes farmland,meadowland,open water,roads,urbanized areas and other non forest land

Table 2 Descriptive statistics of sampled biomass data

Table 3 Localized stand biomass equations

In this case,r2=0.8020 andn=20.Forest biomass in each plot was estimated in two ways,with a local biomass model and with a regional biomass model or BEF.

The volume of 4176Eucalyptusplantation plots was obtained directly from forest resource inventory data.The biomass of each plot was calculated in several steps.First,the 4176 plots were divided into three stand age groups;then,we adopted the corresponding biomass expansion factors to calculate the biomass of theEucalyptustrees within the plantations of each plot(Table 3); finally,we added the biomass of 4176 plots and extrapolated the totalEucalyptusplantation biomass in Nanjing County.

Spatial visualization of differences in biomass estimation

A geographic database was established by vector quantization.We scanned basic paper stock maps to rectify the borders of plots in an ArcGIS software platform.We vectorized the basic 2009 map of forest resources of the study area,structured a topological relationship,and then saved the final map with vector quantization in a geodatabase.We estimated biomass estimation differences by finding the difference between the actual biomass and estimated biomass,multiplying that difference by 100%,and dividing that product by the actual biomass.

To establish an attribute database,attribute data were linked to each site after vector quantization.Attribute data included location,area,forest type,species composition,age stage,plot volume,plot biomass,plot carbon stock,and plot carbon density,and＞20 additional items.The attribute database was linked with a geographic database in ArcMap software to achieve a good spatial match and allow the search function to be used with both spatial polygon maps and attribute information for statistical analysis.

BRT analysis

BRT analysis,the integration of boosting techniques and regression trees in data analysis,makes it possible to handle different types of predictor variables without data transformation or outlier elimination.The method is superior to conventional modeling approaches because it can overcome the disadvantages of having a poor predictive function if a single-tree model is used(Jalabert et al.2010;Martin et al.2011).The specific procedure of the BRT method was as follows:install R software(version 2.15.1)packages(including gbm package),transfer to BRT add-ins,edit Excel file,input BRT coding,and run the model.

The effects of multiple factors on deviations in biomass estimation and their relative contributions were quantified with the BRT model.The 10 studied variables were stand age,DBH,stand density,elevation,degree,position,direction,humus depth,soil depth,and site index.This procedure has been described in greater detail elsewhere(Elith et al.2008).

Results

Estimation of forest biomass by two methods

As shown in Table 4,the estimated 2009Eucalyptusplantation biomass obtained using the regional biomass model(12.53×106t)was 22.12% greater than that obtained using the local sampling model(10.26×106t).The local and regional models indicated that the biomass in each of the three age groups increased as the stand age increased.As shown in Table 5,the deviations in biomass estimates between the two models for 4176 plots were lowest in the mature group(15.94%)and largest in the middle-age group(62.63%).

Spatial distribution of deviations in biomass estimation

Differences in the spatial distribution of deviations in biomass estimation in Nanjing County,China were mainly concentrated in the towns of Longshan and Nankeng,which supported the two largestEucalyptusplantation areas in 2009.The negative values in the differences indicate where estimates obtained with the local model were greater than those obtained with regional BEF,while positive values show the opposite trend.Areas with negative values accounted for 18.49%of the total area.The remaining plantation areas with positive values were constituted as follows:43.13%of the landscape area had 0–40%deviation,26.98%had 40–80%deviation,and 11.40%had＞80%deviation(Fig.2).

Our BRT analysis hypothesized that 10 variables accounted for 100%of the total deviation in biomass estimation,including three forest stand condition indicators(stand age,DBH,and stand density),four topographical indicators(elevation,degree,direction,and position),and three soil indicators(humus depth,soil depth,site index).The three dominant determinants of biomass estimation(stand age,elevation and DBH)accounted for 76.43%of the deviation in total(Fig.3).

Table 4 The differences in forest biomass estimates between the local and regional biomass models

Table 5 The difference in biomass estimates using the local and regional biomass models among three age groups

Fig.3 Relative contributions of predictor variables to deviations based on the boosted regression tree models.The relative influence(expressed as a percentage)of each predictor variable was calculated by partitioning the total differences explained by each predictor variable

Forest stand factors as a group(including stand age,DBH,and stand density among others)were the largest contributors to biomass estimation and accounted for 70.94%of the total deviation between the two models(Fig.3).Of these three variables,stand age and DBH contributed to the large part of the difference.Stand density contributed the least.The deviation declined as stand age increased.Biomass estimate deviations increased as both DBH and stand density increased,but tended to stabilize over time(Fig.4).

Four topographical variables together accounted for 18.65%of the total deviation in biomass estimation,with the following contributions(Fig.3):elevation(10.91%),degree(4.70%),direction(2.29%),and position(0.75%).Elevation emerged as the most important topographic factor in estimate deviation,with the deviations initially declining and then surging to the stabilization as elevation increased(Fig.4).In terms of slope position,the largest differences in biomass estimation between the two models were obtained on the middle slopes and on the downslope areas(Fig.4).Regarding direction,the deviations were smallest on sunning aspects and largest on semi-shady slopes(Fig.4).

Soil variables accounted for 10.41%of the total deviations in biomass estimation between the two models;the order of importance of the three variables from high to low were soil depth(4.82%),humus depth(4.39%),and site index(1.2%)(Fig.3).No significant statistical relationship was observed between deviations in biomass estimation and soil depth.However,the deviations in biomass estimation increased with increases in humus depth(Fig.4).

Discussion

Significance of this study

Fig.4 Partial dependence plots of the predictor variables using boosted regression trees.Position:ds(downslope),ms(middle slope)and us(upslope);Direction:a(sunny),b(semi-sunny),c(semi-shady)and d(shady);Site Index:I(very fertile),II(fertile),III(moderately fertile)and IV(infertile).The Y-axis indicates the differences in biomass estimation caused by each factor.The type of data were indicated by divided lines as in the figures of position,Direction and Site Index,numeric data by continuous lines as in the figures of DBH,density and others

Our study analyzed differences in the spatial distribution of biomass estimates on a regional scale and quantified the relative contributions of biotic and abiotic factors to those estimation differences.These results are important for accurate mapping of forest biomass on a regional scale,which is needed to enable estimation of carbon sequestration capability as well as for elucidating the potential effects of estimation deviations on regional-scale quantitative assessments.This information can be used to reduce uncertainty in regional carbon balance estimates in support of providing a theoretical basis for the sustainable management ofEucalyptusplantations.It also lays the foundation for future data assimilation and model calibration for scale transformation.

Some researchers have indicated that data assimilation could be used to estimate forest biomass accurately during scale transformation.The core idea underlying this approach is that ecological process-based model simulation results can be integrated with background information by combining BEF-derived and forest resource inventory data.This integration should be done to optimize the parameters of ecological process-based models by minimizing the objective function(i.e.,differences in values of analysis and background fields)(Barker et al.2012;Lorenc 2003;Tickle et al.2001).

This study describes differences in the spatial distribution of biomass inEucalyptusplantations obtained by BEF versus a local biomass model on regional scale.Stand age and elevation emerged as the most influential biotic and abiotic factors affecting estimate deviation,respectively.Moreover,our findings indicate that BEF,as it is currently applied,does not account for the contributions of stand age and elevation adequately.Because regional estimation of forest biomass from a background field can result in a large deviation,these highly influential variables should be accounted for in future BEF equations to minimize such deviations.In addition,the results of this study have great implications for determination of regional carbon stocks.

Those tasked with designing strategies for mitigating the impacts of climate change should take note because the data presented here are closely related to the reporting of greenhouse gas inventory statistics at the country level under the United Nations Framework Convention on climate change(Bishop and Hodyss 2011;Carrassi et al.2008;IPCC 2003;Volkova et al.2015).Although our research results represent relatively young(＜10-year-old)Eucalyptusforests within a limited region,the findings are informative for the broader management ofEucalyptusplantations.

The magnitude of BEF deviations revealed here could have significant ramifications on national-level carbon inventory and reporting and,consequently,affect continuous IPCC assessments.It has been suggested that the application of remote-sensing data with simulations could help overcome problems associated with data transfer from one spatial scale to another in regional-level forest biomass research.However,applications of remote-sensing data and computer simulations require verification with numerous field samples,which has been proven to be difficult(Wulder et al.2008).Thus,BEF is expected to remain popular for estimating regional forest biomass(Albaugh et al.2009;Basuki et al.2009)despite the method’s shortcomings.

Comparisons of biomass deviations

BEF may be susceptible to various sources of error.First,the ratio of crown and stem biomass-to-total biomass varies widely across regions due to the spatial–temporal heterogeneity of stand growth in different regions.For example,in the UK,Levy et al.(2004)showed that stand-level forest biomass depended strongly on tree height.Meanwhile,in a study in China,Fang et al.(2001)assumed a linear relationship between biomass and volume.In Finland,Lehtonen et al.(2004)employed BEF with stand age groups,rather than using a single constant coefficient,and found a significant relationship between stand age and leaf/stem biomass.

When applying BEF,the accuracy of the results can vary depending upon which formulas and parameters are used.Second,simple extrapolation of site-specific forest biomass information onto a larger spatial scale entails considerable uncertainty,especially when environmental variables have substantial temporal and spatial variations(Guo et al.2010).Analysis of forest biomass on large,even regional,spatial scales often requires a large dataset and considerable calculations due to the complexity and variety of forest types encompassed as well as database insufficiencies,which lead to large potential deviations or uncertainties(Hilker et al.2009).

Finally,BEF relies mainly on experts to analyze the method’s uncertainties and to make assumptions for new study areas based on information about other locations in the literature.Few comparative studies use empirical parameters obtained for a specific subzone or from typical wood parsing analysis(Albaugh et al.2009).Despite the aforementioned risks,for substantial deviations,BEF is still considered to be the most effective method for estimating biomass based on forest resource inventory at regional and national scales.

The present finding of a 22.12%overestimation of the regional forest biomass in 2009 in Nanjing County with BEF relative to local biomass model results exceeds the 4–13%deviation range observed by Jalkanen et al.(2005)using age-dependent BEF to estimate conifer-dominated biomass on a regional scale in Sweden.But it is more modest than the 83–142%range reported by Albaugh et al.(2009),who applied site-specific analysis in two forest settings,also in Sweden.

The greater present deviation relative to those reported by Jalkanen et al.(2005)could be due to species differences:we examined broad-leavedEucalyptus,whereas Jalkanen and colleagues analyzed conifer forest sites dominated by pine,spruce,and birch species.Indeed,broad-leaved forests have been reported previously to have higher variations in forest biomass than coniferous forests(Simpson et al.1995;Teobaldelli et al.2009).Additionally,our study had a more refined classification of the components of tree biomass.That is,we calculated the biomass contributions of stems,branches,leaves,and roots separately because of their differing moisture contents;Jalknen and colleagues did not.

Our lower deviations relative to the work of Albaugh et al.(2009)may also be related to methodological considerations,namely the fact that our study considered stand age explicitly but not management practices,whereas Albaugh et al.considered management practices,but not stand age.Specifically,we analyzed young,middle-aged,and mature groups separately.Conversely,in their analysis,Albaugh et al.(2009)considered the management strategies employed at samples sites(control,irrigated,fertilized,or irrigated plus fertilized).Notwithstanding,our findings are in agreement with these prior studies in indicating that reliance on published parameters,rather than local empirical parameters,impedes the reliability of biomass estimation(Luo et al.2013).

Partitioning sources of deviation in biomass estimation is important for the development of strategies to improve the accuracy of forest biomass estimation at the regional level and therefore is urgently needed(IPCC 2003;Lehtonen et al.2007;Stinson et al.2011).Sources of deviation in biomass estimation include mainly:(1)the model itself,(2)input data,and(3)model parameters(Bottcher et al.2008;Larocque et al.2008).Model parameter deviations have been described as the principle source of uncertainty in forest carbon stock estimation(van Oijen et al.2005,2011).

Growing efforts to include consideration of biotic and abiotic factors in regionalEucalyptusforest biomass estimation may improve accuracy to some extent(Antonio et al.2007).Understanding the relative contributions of various biotic and abiotic variables to forest biomass estimation would aide in the design of future field sampling and monitoring techniques such that those variables that are more important could be given greater weighting,thereby reducing uncertainties (Stegen et al.2009;Vieilledent et al.2012).

Relative contributions of biotic factors to deviations in forest biomass estimation

We found that biotic factors were dominant over abiotic factors in affecting the estimation of forest biomass.We found that stand age was the dominant explanatory biotic factor,consistent with Montagu et al.(2005)and Saint-Andre et al.(2005) findings indicating that stand age is a dominant factor in estimatingEucalyptusforest biomass on the regional scale with strong heterogeneity.Forest biomass fluctuated with stand age mainly because the proportion of biomass accumulating in each biomass component changed with stand age.

Our observation that stand age-related deviations decreased with increasing stand age,with large deviations being derived from the young-and middle-age groups,is consistent with the findings of Baraloto et al.(2011).The emergence of stand age as a particularly important factor in forest biomass estimation in this study may be consequent to our examination of a range of stand ages and sample sizes,which also yielded ten times the variation documented previously(Baghdadi et al.2015).Pan et al.(2004)described the correspondence of extensive morphological age-based differences with various environmental conditions,as well as maturation-related increases in the ratio of volume to biomass.Their results were closely associated with tree morphology because as trees grow,leaf biomass and the sapwood area of the canopy tend to stabilize.

It is our recommendation that stand age be considered when establishing any new allometric biomass equations.Consistent with this recommendation,Ju et al.(2006)found that biomass estimates would deviate widely if the effects of stand age were neglected.Despite the difficulty involved in obtaining forest age information on a large spatial scale,recent regional forest carbon research studies have accounted for stand age in both Canada and China(Wang et al.2007).Moreover,precise quantification of the relationships among stand age,volume,and biomass is essential for enabling accurate estimations of biomass on a regional or larger scale.

Relative contributions of abiotic factors to deviations in forest biomass estimation

Topographical and soil factors accounted for 18.65 and 10.41%of the total deviation in biomass estimation,respectively.Among all abiotic factors,elevation and soil depth were the most important variables affecting biomass estimation.Alves et al.(2010)also found that elevation played the most significant role in biomass accumulation among topographical variables.Our estimate deviations decreased initially with increases in elevation,but then increased as elevation increased.Our finding that the influence of elevation on biomass was particularly pronounced for mature stands is consistent with the conclusion of Asner et al.(2009).Baraloto et al.(2011)argued that the relationship between soil properties and biomass depended heavily on climate,such that less-weathered soils had higher turnover rates,and thus less accumulation of aboveground biomass,whereas forests on weathered soils can accumulate both standing biomass and soil nitrogen steadily(Baraloto et al.2011).Aboveground biomass has also been correlated to precipitation(Stegen et al.2009)and may decrease during long periods of drought(Du et al.2015;Malhi et al.2006).

Recommendations and further research

Given that stand age was the dominant variable affecting deviation of BEF biomass estimation from local modelling data,we suggest that in future field sampling,researchers should consider stand age fully when establishing biomass models to reduce biomass overestimation.For longitudinal,regional-scale BEF studies,we should also recognize that biomass modelling parameters change with stand age within a given species.The complex effects of multiple biotic and abiotic variables must be considered when conducting field sampling and inventory collection,as well as in the development of simple statistical models.

Methodological limitations vary depending on study scale.When estimating forest biomass on a small scale,the harvesting method may only represent a certain stage of the entire process of vegetative growth,and age-dependent shifts in biomass vary across different species.Moreover,measurement and mathematical methods each have some inherent degree of uncertainty.For large-scale research,environmental effects should be stressed because the mechanism of impact may differ across different areas and species.In addition,identifying optimal parameters for modeling is challenging because forest ecological processes have nonlinear response characteristics,driving forces exhibit spatial–temporal heterogeneity,and it is difficult to obtain expansive local physical and characteristic parameters,including those related to soil,terrain,and climate.Therefore,as we do the necessary work of integrating and comparing various methods,we need to optimize the benefits of each method and develop quantitative statistical models that reflect difference mechanisms.

Eucalyptusplantations play an important role in both carbon fixation and timber production owing to their short rotation times and rapid growth.Outside of China,Eucalyptusforests are usually harvested at 20–30 years old.However,in China,they are harvested at 5–10 years old because of the strong demand for timber,in an attempt to balance the ecological benefits with the need for economic growth.

The practical application of forestry should be emphasized in China.Knowing how to integrate different forms of rotation,various methods,and multi-sources of data to achieve scale transformation has become a universally difficult problem when researchers attempt to estimate forest biomass accurately.Accurate estimation ofEucalyptusforest biomass on regional scale is of great significance for local forest resource management aimed at balancing ecological benefits with economic growth.

We recommend that future research focus on the following issues:

1. Verification of whether the results obtained here withEucalyptusforests are replicated with other species and in other regions;

2. Given the importance of biotic factors in regional forest biomass estimation,more detailed sampling criteria are needed,including consideration of plot size and the accuracy of stand age-group spatial extents;

3. Adoption of process-based models that incorporate additional ecological and biological factors;

4. Assimilation of data from multiple sources with modified model parameters in forest biomass estimation;

5.Combining biomassestimation based on forest resource inventories with complementary methods,such as the eddy covariance microclimate method,simulation,and remote sensing.These methods can improve data conversion across spatial scales to enable calculation of forest vegetative biomass and carbon on a large scale,and eventually on a global scale,in the future.

Conclusions

By combining forest inventory data and field sampling in this study,we quantified the magnitudes and sources of deviations inEucalyptusplantation forest biomass estimates in Nanjing County,Fujian,China in 2009.Our results show that use of scale-transferring BEF overestimated accumulated biomass by 22.12%,relative to a local model,and the largest source of deviations was stand structure variables,which accounted for 70.94%of the total deviation.We conclude that information on forest stand age,elevation,and DBH should be explicitly included in regional forest biomass/carbon estimation to minimize deviations.Researchers should pay more attention to biotic factors such as stand age and DBH during field sampling.

AcknowledgementsWe are grateful to Professor Li Hu for his helpful suggestions.We thank anonymous reviewers for their helpful comments on the earlier manuscript.

Authors’contributionsQQ and GY contributed equally to this work and should be considered co- first authors.