Forestry An International Journal of Forest Research

Forestry 2014; 87, 165 – 176, doi:10.1093/forestry/cpt045

Advance Access publication 27 November 2013

A dynamic volume and biomass growth model system for even-aged

downy birch stands in south-western Europe
Esteban Gómez-Garcı́a1*, Felipe Crecente-Campo1, Brian Tobin2, Michael Hawkins2, Maarten Nieuwenhuis2
and Ulises Diéguez-Aranda1

Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

1
Departamento de Ingenierı́a Agroforestal, Universidad de Santiago de Compostela. Escuela Politécnica Superior, C/ Benigno Ledo, Campus
Universitario, 27002 Lugo, Spain
2
UCD Forestry, University College Dublin, Belfield, Dublin 4, Ireland

*Corresponding author. Email: esteban.gomez@usc.es

Received 18 January 2013

This study develops some components to complete a management-oriented dynamic model system for simulat-
ing current and future volume and biomass of even-aged stands of managed downy birch (Betula pubescens Ehrh.)
in Galicia (NW Spain). The specific components developed are a mortality equation, a basal area growth equation, a
diameter distribution prediction system and equations for total and component biomass prediction. Functions to
predict decreases in number of trees per hectare and basal area growth were simultaneously fitted using Nonlinear
Seemingly Unrelated Regression (NSUR) and a base-age invariant dummy variable method; critical errors of 19 and
16 per cent were obtained, respectively, when projections were made. The Weibull function modelled successfully
all but four of the 198 diameter distributions examined, using the parameter recovery method through moments.
To predict components and total aboveground biomass, a system of additive equations was simultaneously fitted
using the generalized method of moments (GMM), which takes into account heteroscedasticity and inherent cor-
relations among the biomass components. A two-step fitting procedure was used because of the different number
of observations for the components considered. The biomass equations explained more than 85 per cent of
the observed variability. The Galician birch stands which were surveyed were found to have grown by
3 –10 m3 ha21 year21 in volume and by 2 –8 Mg ha21 year21 in aboveground total biomass, with rotations
between 30 and 60 years. The model system now forms an accessible decision support system for forest manage-
ment and land development in the region.

Introduction (Castedo Dorado et al., 2007a). The re-measurement of such

The Betula genus is distributed throughout most of Europe, where
it is represented by four species, two of which are trees: downy birch
(Betula pubescens Ehrh.) and silver birch (B. pendula Roth.).
In Spain, these two taxa freely hybridize and their taxonomy is
somewhat confused, with each one possessing a number of differ-
ent nomenclatural identities (Castroviejo et al., 1990). Although
B. pubescens (also referred to as B. alba L. and B. pubescens
subsp. celtiberica Rothm. & Vasc.; Castroviejo et al., 1990) is an im-
portant fast growing pioneer tree species in Galicia (north-western
Spain), there is a lack of reference information regarding its silvicul-
ture and management. In contrast, growth and yield research on
birch has been relatively active in northern Europe (Hynynen
et al., 2010).
With data from a first inventory of a network of plots covering a
wide range of ages, densities and sites in the distribution area of the
species within this region, it has been possible to develop static
growth models, which may work well in unthinned or infrequently
thinned stands (Garcı́a, 2001), but they do not reflect accurately
the evolution under more intensively managed conditions
plots has provided information about the trajectories of the main
stand variables over time, which allows the construction of
dynamic growth equations. These equations (also referred to as
transition functions in this study) have the general form (omitting
the vector of model parameters) of Y2 ¼ f(t2, t1, Y1), where Y2 is the
value of the function at age t2, and Y1 is the reference variable
defined as the value of the function at age t1. These dynamic
models will offer a better description of the development (accord-
ing to specified management intensities) of a stand over time than
static models (Garcı́a, 1988), and are therefore preferable for simu-
lation purposes.
Although dynamic whole-stand growth models represent a
good compromise between generality and accuracy of estimates,
and directly project information that is easily obtained from inven-
tory data (Vanclay, 1994; Gadow, 1996; Garcı́a, 2003), modern
forest management decisions increasingly require more detailed
information at the tree level. For this purpose, stand estimates at
a particular point in time can be disaggregated mathematically
in order to estimate tree variables, and then total tree volume
(ground to tip) or merchantable volumes (i.e., the merchantable

# Institute of Chartered Foresters, 2013. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.

165
Forestry
part of the tree can be subdivided into logs for transport, use or pro-
cessing), and total tree or tree component biomass. Such estima-
tion of forest biomass and carbon stocks has gained importance as
a result of reporting requirements to the United Nations Framework
Convention on Climate Change and the Kyoto Protocol. In addition,
the need to reduce fossil fuel dependence has resulted in the con-
sideration of otherwise non-merchantable parts of the tree and
even the whole tree as a useful renewable source of biomass, par-
ticularly during early stages in stand development. Zianis et al.
(2005) carried out a review of the volume and biomass equations
developed for forest species in Europe, and found 607 biomass
equations, but very few of these had been developed for species
common to the south of the continent. Montero et al. (2005) devel-
oped power equations for tree components’ biomass estimation
for 32 forest species (including birch) in Spain, which were intended
to quantify the carbon stocks and the potential of Spanish forests
as carbon sinks. Because these equations only depended on diam-
eter at breast height and did not fulfil the additivity property (the
sum of the estimated biomass of the different components
equals the estimated biomass of the total; Cunia and Briggs,
1984; Parresol, 1999), new equations (except for birch) were devel-
oped that incorporated this desirable characteristic and included
also total height as an independent variable (Ruiz-Peinado et al.,
2011, 2012).
Thus, to equip forest management with the precise quantifica-
tion and projection of stocks to satisfy the increasingly diverse array
of demands for goods and services from forests, this study aims to
complete a management-oriented, dynamic, disaggregated
model system for the simulation of the growth of even-aged
birch stands in Galicia. In this study, we specifically developed
dynamic equations to estimate the number of trees and basal
area per hectare, the diameter distribution function and the
output functions for biomass prediction (see Figure 1). For the
purpose of completeness, the system equations previously devel-
oped are included in the Appendix.
Materials and methods
Data
The data used were obtained from two different sources. Summary statis-
tics of the main stand and tree variables used in model development are
shown in Table 1. For the production of the first of these datasets, a
network of 137 plots was established in even-aged, birch-dominated
stands (85 per cent or more of the standing basal area consisted of birch)
during the winters of 1997 –1998, 1998– 1999 and 2000– 2001. The plots
were located throughout the area of downy birch distribution in Galicia,
and were subjectively selected to represent the existing ranges of ages,
stand densities and sites. The age in planted stands was established
using the planting records, while in natural stands it was determined by
counting the number of rings at stump height in dominant trees. All trees
in each sample plot were numbered. Two measurements of diameter at
breast height (1.3 m above ground level) were made (at right angles to
one another) to the nearest 0.1 cm using callipers, and a mean was calcu-
lated. Total tree height in each plot was measured to the nearest 0.25 m
with a hypsometer in a random sample of 30 trees (all plots contained
more than 30 trees). Additionally, undamaged, dominant trees represent-
ing the 100 largest-diameter trees per hectare were selected for site index
determination. Other descriptive variables for each tree were also recorded
(e.g. form, quality, living or dead, etc.). Because many plots were disturbed
by forest fires or tree cutting, only a subset of 50 of the initially established
plots was re-measured in the winter of 2007– 2008. Additionally, 11 new
166
plots were established in the winter of 2009– 2010 to augment the data-
base for this and future studies. The mean size of the 148 established
plots was 537 m2 (min. 200 m2, max. 1000 m2 and S.D. 224 m2). The
data from these temporal or permanent plots were used to develop the
number of trees per hectare and stand basal area transition functions
and the diameter distribution function.
The second dataset contained data on 50 trees felled in the winter of
2002– 2003 and in the summer of 2009, which were used to develop tree
biomass equations. The trees were selected outside the research plots,
but in the same stands. Trees of different sizes and crown classes were
included, but trees with deformities or situated close to stand edges were
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
rejected. Total tree biomass was divided among four components:
biomass with diameter .7 cm (including stem and branches), branches
with diameters from 2 to 7 cm, branches with diameters ,2 cm and
leaves. The foliage component was only present in the 34 trees sampled
in the summer of 2009. The wet weight of components and the differences
between the wet and dry weights of samples (dried in an oven at 65 8C until
constant weight) were used to determine total dry biomass for each com-
ponent of each tree. Data for branches with diameter ,2 cm collected in
the winter of 2002– 2003 showed significant differences in dry density to
those collected in the summer of 2009. The conclusion reached was that
the sample of branches with diameters ,2 cm collected in 2002 –2003
may not have been sufficiently dried. Therefore, the fit of the model for
this component was carried out using the data of the 34 birch trees felled
in the summer of 2009.
Model system structure
The basic structure of our model system, outlined in Figure 1, can be classi-
fied as a statistical, disaggregated, non-spatial, deterministic, stand model
(Weiskittel et al., 2011), which was based on the state–space approach
(Garcı́a, 1994). The model developed is similar in structure to those devel-
oped by Diéguez-Aranda et al. (2006a) and Castedo-Dorado et al. (2007a)
for volume estimation for radiata and Scots pine, respectively. In compari-
son to these models, an advantage of this model system is its ability to
predict biomass as well as volume. The present methodology also advances
the work by those authors through the simultaneous fitting of transition
functions and the utilization of a base-age invariant dummy variable
method. This work also includes innovations to biomass estimation meth-
odology in the form of a two-step fitting procedure and a simultaneous
fit with the generalized method of moments (GMM) that takes into
account heteroscedasticity and inherent correlations among the biomass
components.
In this model system, three state variables (dominant height, number of
trees per hectare and stand basal area) define the initial stand condition at
any point in time. Therefore, three transition functions are used to predict
the growth by updating the state variables. In accordance with Garcı́a
(1994), the transition functions possess certain desirable properties: (i) con-
sistency, which implies no change for zero elapsed time; (ii) path-invariance,
where the result of projecting the state first from t1 to t2, and then from t2 to
t3, must be the same as that of the one-step projection from t1 to t3; and (iii)
causality, in that a change in the state can only be affected by inputs within
the relevant time interval. Transition functions generated by integration of
differential equations (or summation of difference equations when using
discrete time intervals) satisfy these conditions and allow computation of
the future state trajectory.
In the disaggregation system, a distribution function estimates the
number of trees in each diameter class, once the state variables are
known for a specific age. A generalized height– diameter relationship is
used to estimate the height of the average tree in each diameter class. It
is then possible to estimate total or merchantable volumes (which
depend on specified log dimensions) by using a taper equation that uses
the above predicted diameters and heights. Finally, biomass by diameter
class and for different biomass components can be estimated using
 A dynamic volume and biomass growth model system

Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

Figure 1 The basic structure of the model system presented in this study. H1, H2, N1, N2, G1 and G2 ¼ dominant height, number of trees per hectare and
stand basal area at ages t1 and t2, respectively. The symbol represents the input variables. The functions to transform input variables in intermediate
or in output variables are indicated by the symbol . Finally, the symbol represents intermediate variables, and the result or output variables
are represented by the symbol . White background symbols represent variables considered and functions developed in this study.

167
Forestry

Table 1 Summarized data from the plot inventories and the felled trees used to fit the equations developed in this study

Variable Mean Min. Max. SD Mean Min. Max. SD

First inventory (148 plots) Second inventory (50 plots)

t (years) 32.0 12.0 94.0 11.9 38.0 22.0 56.0 9.8
H (m) 15.3 7.2 24.4 3.6 18.1 11.0 24.5 3.1
N (trees ha21) 1750 390 6000 1099 1430 350 4480 865
G (m2 ha21) 25.1 3.3 66.5 10.5 30.1 9.2 71.8 11.2
Fifty felled trees

Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

d (cm) 16.0 7.3 34.0 6.7
h (m) 14.6 8.7 26.2 3.2
hst (m) 0.07 0 0.15 0.04
wfb (kg) 98.9 6.3 425.9 109.0
wb27 (kg) 20.1 0.5 71.9 17.7
wb02 (kg)* 13.2 2.0 57.3 11.8
wl (kg)* 3.1 0.1 14.8 3.0

t ¼ stand age; H ¼ dominant height, defined as the mean height of the 100 largest-diameter trees per hectare; N ¼ number of trees per hectare; G ¼ stand
basal area; d ¼ diameter at breast height over bark; h ¼ total tree height; hst ¼ stump height; wfb ¼ biomass .7 cm diameter; wb27 ¼ biomass of branches
from 2 to 7 cm diameter; wb02 ¼ biomass from branches ,2 cm; wl ¼ leaf biomass.
*These components were only available for 34 trees felled in the summer of 2009.

individual-tree biomass equations (which depend on the above predicted Several methods for parameter estimation were preliminarily tested
diameters and, in some cases, heights). and compared for their goodness of fit (Cao, 2004). A parameter recovery
The variables studied and the functions developed in this work are repre- method through moments proved best. Moreover, this method guarantees
sented with white background symbols in Figure 1. The functions already that the sum of the disaggregated basal area obtained by the Weibull func-
developed in previous studies are summarized in the Appendix. tion equals the stand basal area provided by an explicit growth function of
the variable. The function parameters were recovered from the first raw
moment, which was the arithmetic mean diameter (d), and the second
Number of trees and stand basal area transition functions central moment, which was the variance of the distribution (var), estimated
by the arithmetic and the quadratic mean diameters (dg) (var = d2g − d 2 ),
The reduction in the number of trees due to mortality is mainly the result of
competition for light, waterand soil nutrients within a stand (Peet and Chris- using equations 2 and 3 (Cao et al., 1982).
tensen, 1987). The models reported by Diéguez–Aranda et al. (2005a) were     
2
d 2 1
the starting point for developing the transition function for number of trees var = G 1 + − G2
1 + (2)
per hectare. The stand basal area growth of an even-aged stand depends G2 (1 + (1/c)) c c
on age, number of trees, stand basal area and site productivity. While devel- 
d
b= (3)
oping the transition function for stand basal area, an evaluation of the G(1 + (1/c))
equations reported by Diéguez-Aranda et al. (2005b), as well as the GADA
(Cieszewski and Bailey, 2000) formulations tested by Castedo-Dorado where G is the Gamma function.
et al. (2007b), was carried out. Once the mean and the variance of the diameter distribution are known
Initially, the transition functions for the number of trees and stand basal at any specific time, and taking into account that equation 2 only depends
area were separately fitted to the re-measured plots (50 plots measured on parameter c, the latter could be obtained using an iterative procedure.
twice) using the base-age-invariant dummy variables method (Cieszewski Parameter b could then be calculated directly from equation (3). As this
et al., 2000) and the SAS/ETSw MODEL procedure (SAS Institute Inc., 2008). model system allowed prediction of quadratic mean diameter directly
Because a significant correlation between the errors of both equations was from number of trees per hectare and stand basal area, the arithmetic
observed, the functions were fitted again simultaneously with Nonlinear mean diameter was the only variable to be modelled through a relationship
Seemingly Unrelated Regression (NSUR) and the same base-age-invariant with the quadratic mean diameter and other stand-level variables using fol-
dummy variables method, using the SAS/ETSw MODEL procedure. lowing equation.

Diameter distribution function
The diameter distribution function was required for the disaggregation
system. A two-parameter Weibull function (equation 1) was used to
model the diameter distribution of the 198 plot-inventory combinations
(first inventory of 148 plots plus second inventory of 50 plots).
 = dg − exp(Xb)
d (4)
where X is a vector of known stand variables and b is a vector of parameters
to be estimated.
This equation was fitted analyzing different combinations of stand vari-
ables as predictors and by ordinary nonlinear least squares (ONLS) using the
SAS/ETSw MODEL procedure.

 c  x c−1   c 
x
f (x) = exp − (1) Biomass estimation
b b b
where x is the random variable, b the scale parameter of the function and c For the biomass estimation models, linear and power functions were first
the shape parameter that controls the skewness. fitted for the different biomass components and the best one selected for

168
 A dynamic volume and biomass growth model system

each component. Then a system of equations with cross-equation con- root mean square error (RMSE) (equation 7); and Schwarz’s Bayesian
straints on the structural parameters and cross-equation error correlation Information Criteria (BIC) (equation 8). The RMSE was useful because it is
was defined for predicting components and total aboveground biomass expressed in the same units as the dependent variable, giving an idea of
with additivity (Parresol, 2001; Bi et al., 2010). The system had the following the mean error when using the model. Both RMSE and BIC penalize
general formulation: models with more parameters, according to the general principle of
scientific simplicity.
b
wi = bi0 xj ij + 1i i=n
(Yi − Ŷi )2

n (5) R2 = 1 − i=1 (6)
i=n 2
wtotal = wi + 1t i=1 (Yi − Y)
i=1


Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

where wi represents the biomass estimation for the ith component, wtotal is i=n
i=1 (Yi − Ŷi )
2
the total aboveground tree biomass, xj are independent variables, bi0 and bij RMSE = (7)
are parameters to be estimated in the fitting process, and 1i and 1t are inter- n−p
correlated error terms.
Because some components were not available for all the trees
(branches ,2 cm in diameter and foliage), a two-step fitting procedure
was used. First, the two equations for estimation of foliage and branch
biomass ,2 cm in diameter, and a third combining both foliage and
branch biomass, were simultaneously fitted. Second, a system of equations
to estimate the remaining biomass components as well as total biomass
(including estimations for leaves and branches ,2 cm in diameter carried
out with the combined equation) were simultaneously fitted.
Equations were fitted using the GMM of the MODEL procedure of SAS/
ETSw. This method produces efficient parameter estimates under hetero-
scedastic conditions, without requiring any specification of the nature of
the heteroscedasticity (Greene, 2000; SAS Institute Inc., 2008). One
problem encountered in estimating the system of equations for biomass es-
timation was that the error term of the equation for total aboveground tree
biomass was a linear combination of the error terms of the biomass com-
ponent equations, which led to a singular cross-equation variance– covari-
ance matrix. In SAS, this problem may be overcome by computing a
generalized inverse of the variance–covariance matrix, by setting the part
of the matrix for total biomass to zero. Although this procedure avoided
the problem of singularity by ignoring the correlations between the error
term for the total biomass and the error terms for other biomass compo-
nents, the cross-equation constraints on the structural parameters were
still applied in the parameter estimation (Bi et al., 2004; 2010). The para-
meters estimated by fitting the biomass component equations simultan-
eously, without considering the equation for total biomass, are therefore
not the same.
The predictions of aboveground woody biomass (excluding leaves) for
trees of different diameters were compared with the results obtained
with equations developed for birch elsewhere in Europe: for Betula species
in UK (Bunce, 1968), Sweden (Marklund, 1988) and Finland (Repola,
2008); and for B. pubescens in Iceland (Snorrason and Einarsson, 2006).
The reference functions all used tree diameter as an independent variable.
The diameter range used to develop these equations was similar in all cases
(from 3 cm to 35 cm), and this corresponded with the ranges used in this
study. The equation for total aboveground biomass in Iceland included a
leaf component and, therefore, the value of woody biomass should be
slightly lower than the model prediction. Moreover, this equation used
diameter at 0.5 m aboveground level, rather than at 1.3 m, as an independ-
ent variable. This diameter was estimated using the taper equation in the
Appendix.
Selection and evaluation of models
The comparison of the estimates of the fitted models was based on numer-
ical and graphical analyses. The graphical analyses included plots of resi-
duals against the estimated values, and plots of the fitted curves overlaid
on the trajectories of different variables. The numerical analysis consisted
of a comparison of three statistical criteria obtained from the residuals:
the coefficient of determination (R 2) (equation 6), which indicated the
proportion of the total variance that was explained by the model; the

i=n
(Yi − Ŷi )2
BIC = n log i=1
+ p log(n) (8)
n−p
where Yi , Ŷi and Y are the measured, estimated and average values of the
dependent variable, respectively, n is the total number of observations
used to fit the model, p is the number of model parameters, and log is
the natural logarithm.
As new independent data were not available to carry out a real valid-
ation of the models (Kozak and Kozak, 2003; Yang et al., 2004), the observed
state variables for the first inventory of the 50 plots measured twice were
used to evaluate projections of the developed transition functions from
the first to the second inventory. The projection length ranged from 7 to
10 years. Estimates of number of trees per hectare and stand basal area
were evaluated in terms of the critical error (Huang et al., 2003) computed
by re-arranging Freese’s statistic (Freese, 1960):

t2 ni=1 (Yi − Ŷi )2/x2crit.
Ecrit. = (9)
Y
where Ecrit. is the critical error, expressed as a percentage of the observed
mean, t a standard normal deviate at the specified probability level
(t ¼ 1.96 for a ¼ 0.05), x2crit. is the value for the x2 distribution obtained
for a ¼ 0.05 and n degrees of freedom, and the remaining variables are
as previously defined.
If the specified allowable error expressed as a percentage of the
observed mean was within the limit of the critical error, the x2n test could
indicate that the model did not provide satisfactory predictions; otherwise,
predictions were acceptable.
The Kolmogorov– Smirnov test was used to assess if the Weibull func-
tion was appropriate to describe the diameter distributions (Cao, 2004).
Finally, the contrast of White (1980) and the Condition Number (Belsley,
1991) were used to evaluate heteroscedasticity and multicollinearity, re-
spectively, in the development of biomass equations.
Results
Number of trees and stand basal area transition functions
A modification of the model of Clutter and Jones (1980) (Equation
10) and a GADA formulation derived from the model of Korf (cited
in Lundqvist, 1957) (Equation 11) were selected as the most appro-
priate functions to predict tree number reduction and stand basal
area growth, respectively. The Pearson product moment correl-
ation between the residuals from the separate fitting of these
two models was 0.28, indicating a relationship between them.
Simultaneous fitting of these two models using NSUR dealt with
this problem. The goodness-of-fit statistics were R 2 ¼ 0.98 and
169
Forestry
RMSE ¼ 139 trees ha21 for the reduction in number of trees func-
tion, and R 2 ¼ 0.97 and RMSE ¼ 1.96 m2 ha21 for the stand basal
area growth function. Trajectories of number of trees and stand
basal area over time (Figures 2 and 3) showed similar behaviour
between observed and predicted values. All parameters were sig-
nificant at the 5 per cent level. The plots of residuals against esti-
mated values (Figure 4) showed a random pattern of residuals
around zero, with homogeneous variance and no detectable
trends. Critical errors of 19 and 16 per cent were obtained, respect-
ively, for number of trees and stand basal area predictions when
projecting the variables from the first to the second inventory in
the twice-measured 50 plots.
around zero, with homogeneous variance and no trends.
d = dg − exp( − 1.31 + 0.08459H − 0.04192 S) (12)
where d is the predicted arithmetic mean diameter (cm), dg the
quadratic mean diameter (cm), H the dominant height (m) and S
the site index (m). The goodness-of-fit statistics were R 2 ¼ 0.99
and RMSE ¼ 0.31 cm.
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
Biomass estimation
In the selection of models for each biomass component, diameter
 
−1.581 and height provided the best results as independent variables in
  
t2 1.581 t1 1.581 the equations for predicting biomass with diameter .7 cm and
N2 = N1−1/1.581 + 0.004637 log S − leaf components. For the other components, the parameters asso-
100 100
ciated with height were not significant, or models including this
(10) variable were less accurate. Because of observed heteroscedasti-
city (see Figure 4), the GMM was used in model fitting. The
G2 = exp(XG ) exp[(136.4 − 642.8/XG ) t−0.9385
2 ] with system of tree biomass equations was as follows:

XG = 0.5t−0.9385
1 −136.4 + t0.9385
1 log(G1 ) (11)


 wfb = 0.01792d2.396 h0.6126

+ 4 × 642.8 t0.9385
1 + (136.4 − t0.9385
1 log(G1 ))2 wb27 = 0.06712d1.993

where N1, G1, S and t1 represent the predictor number of trees per wb02 = 0.02472d2.346
(13)
hectare, stand basal area (m2 ha21), site index (m, see Appendix) wl = 0.001042d2 h
and age (years), respectively; and N2 and G2 are, respectively, the
predicted number of trees per hectare and stand basal area wtotal = 0.01792d2.396 h0.6126 + 0.06712d1.993
(m2 ha21) at age t2 (years).
+ 0.02472d2.346 + 0.001042d2 h

where wfb is biomass .7 cm (stem and branches with diameter

Diameter distribution function
The Weibull function modelled successfully all but four of the 198
diameter distributions, based on the Kolmogorov –Smirnov test
at 5 per cent level. Equation 12 was selected for predicting arith-
metic mean diameter and for use in the parameter recovery ap-
proach for the diameter distribution. All parameters were
significant at 5 per cent level and the plot of residuals against esti-
mated values (Figure 4) showed a random pattern of residuals
.7 cm) (kg tree21), wb27 is biomass for branches with diameter
from 2 to 7 cm (kg tree21), wb02 is biomass for branches with diam-
eter ,2 cm (kg tree21), wl is biomass for leaves (kg tree21) and
wtotal is total aboveground tree biomass (kg tree21).
All parameters were significant at the 5 per cent level and the
plots of residuals against estimated values showed a random
pattern of residuals around zero (see Figure 4). The best fits were
obtained for the equations of biomass with diameter .7 cm and

Figure 2 Trajectories of observed and predicted tree number per hectare

over time. Model projections for spacing conditions of 500, 1500, 3000 Figure 3 Stand basal area growth curves for stand basal areas of 8, 18, 32
and 5500 trees ha21 at 20 years and site indices of 5, 9, 13 and 17 m, and 50 m2 ha21 at 20 years overlaid on the trajectories of observed values
respectively, (solid lines) are shown. over time.

170
 A dynamic volume and biomass growth model system

Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

Figure 4 Residuals plotted against predicted values from the functions developed in this study.

171
Forestry
Figure 5 Comparison of predicted total aboveground woody individual-
tree biomass (stem and branches) against diameter at breast height
using the birch biomass equations developed for NW Spain (in this study),
UK (Bunce, 1968), Sweden (Marklund, 1988), Finland (Repola, 2008) and
Iceland (Snorrason and Einarsson, 2006). *The equation for Iceland
includes the leaf component.
total biomass. All the equations performed well and explained
between 85 and 98 per cent of the observed variability. The RMSE
values (kg tree21) were 21.0, 6.9, 3.7, 0.9 and 22.9 for wfb, wb27,
wb02, wl, and wtotal, respectively. The Condition Number obtained
in the fitting of the system of equations was 64.6, indicating no
serious problems associated with multicollinearity.
In Figure 5, a comparison is presented of the predictions of
aboveground woody tree biomass using equations developed in
this study and other equations developed for birch in Europe.
Average height was estimated for each diameter class because
this variable was included in some of the equations developed in
this study. To estimate individual tree heights, an average popula-
tion h–d relationship (i.e. only diameter at breast height was used
as an explanatory variable) was fitted to the dataset, using ONLS
(equation 14).
 
−6.962
h = 1.3 + 21.55 exp (14)
d density is reduced so severely by thinning that the site is clearly
The equations developed in this study predicted the highest in-
dividual tree, aboveground woody biomass compared with esti-
mates produced using equations developed in northern Europe.
Discussion
The model system presented in this study is a disaggregated stand
model and requires four state variables for simulation: stand age,
dominant height, number of trees per hectare and stand basal
area. These variables are all normally measured in Spanish forest
inventories. In this model system it was assumed that the domin-
ant height was not altered by silvicultural treatments; therefore,
the model should only be used if thinnings from below or system-
atic thinnings are applied.
The equation selected for estimating the reduction of trees per
hectare (equation 10) included site index in its formulation. This
result was consistent with previous studies in which higher
172
mortality was related to better productivity (Eid and Tuhus, 2001;
Yao et al., 2001; Diéguez-Aranda et al., 2005a). Highest productivity
usually increases the growth rate of the largest (dominant) trees.
These large, rapidly growing trees quickly induce mortality in sub-
dominant trees, through asymmetric competition for available
resources such as light. The net effect is a higher mortality rate
(concentrated in the suppressed trees) and more rapid stand devel-
opment, a phenomenon called the Sukatschew effect (Harper,
1977; Bi, 2004). However, the impact of site quality on mortality
is not straightforward. In some studies, lower mortality was
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
related to better productivity (Woollons, 1998), and Zhao et al.
(2007) found opposite results for loblolly pine (Pinus taeda
Engelm.) growing in different regions. Moreover, good sites are
expected to support higher stocking rates than sites of lower prod-
uctivity (Vanclay, 1994). Bi (2001) related mortality with site occu-
pancy (Bi et al., 2000),which is a relative measure that indicates the
degree to which a stand has occupied the growing space and uti-
lizes the available resources for growth. Both the site occupancy
of a stand and the size of individuals that compose it need to be
considered for accurate predictions of mortality of trees in
even-aged, pure stands (Bi, 2001), and both variables are basically
taken into account in equation 10 by including a measure of site
productivity and a measure of stock at a specific age.
The highest stand basal areas in the dataset (i.e., stand basal
area over 40 m2 ha21) are associated with the smallest plots and
probably do not represent the full stands in which they were
obtained. However, it is important to include these extreme data
in order to be able to specify the behaviour of the model system
over the full range of stands it may be used for in the future.
Some studies (Pienaar and Shiver, 1984; Amateis, 2000; Álvarez-
González et al., 2010) concluded that there was a slight difference
in the stand basal area growth in thinned and unthinned stands of
the same initial age, site index and stand basal area. However,
Clutter et al. (1992) concluded that in thinning experiments involv-
ing even-aged stands, the effects of stand density variation on
stand basal area or volume growth have been inconsistent. They
stated that the results of most studies support the general conclu-
sion that thinning treatments do not significantly affect growth,
except where severe overcrowding would greatly restrict root and
crown development in an unthinned stand, or where stand
underutilized for some time. Clutter et al. (1992), Barrio Anta
et al. (2006) and Castedo –Dorado et al. (2007b) proposed the
same stand basal area growth equation for thinned and unthinned
stands, assuming that the thinning effect was an inherent part of
the model. These authors concluded that any contradictory
results relating to the ‘thinning effects theory’ may be at least
partly attributable to the specific structure of the experimental
datasets. Moreover, as in this study, heavy thinning treatments
were not considered by Barrio Anta et al. (2006) and Castedo-
Dorado et al. (2007b), and the effect of low thinning does not
seem to be very important (Garcı́a, 1990).
Considering the accuracy required for forest growth modelling,
where a mean prediction error of the observed mean at 95 per
cent confidence intervals within+10220 per cent is generally real-
istic and reasonable (Huang et al., 2003), it can be stated, on the
basis of the critical error statistics obtained, that the density and
stand basal area transition functions provided satisfactory predic-
tions. The disaggregated stand model presented in this study
should be used with caution in young stands (,10 years),

because at young ages erratic height growth may lead to errone-
ous site classifications. Moreover, the plots used in fitting the tree
number reduction and stand basal area growth transition func-
tions had a minimum age of 12 years. In general, it is recom-
mended that the ages for the simulation range between the
minimum and maximum values of the data used in the develop-
ment of the model system, because accuracy of predictions
decreases as projections are made outside this range (Weiskittel
et al., 2011). However, the predicted trajectories of the state vari-
ables (H, N and G) showed appropriate trends, logical asymptotes
and a similar behaviour to observed values (Diéguez-Aranda
et al., 2006b; Figures 2 and 3, respectively). Therefore, predictions
outside the range of the development dataset could be useful
within reasonable limits (e.g. in order to compare with growth
and yield in other countries, it might be justifiable to make predic-
tions at 70 –80 years).
The developed system of additive biomass equations elimi-
nated the logical inconsistency between the sum of predicted
values for the components and a prediction of total biomass
(Kozak, 1970). In addition, since the inherent correlation among
biomass components was taken into account, the equations
were more efficient statistically (Parresol, 1999, 2001). The equa-
tions developed in this study predicted the highest individual
tree, aboveground woody biomass compared with results
obtained from equations developed in northern Europe (Figure 5).
Since the range of diameters in the datasets used to develop the
equations was similar, the differences might be caused by other
tree characteristics or stand characteristics. For instance in relation
to the trees, in Iceland the maximum tree height was 11 m and in
this study it was 26 m, while the equations developed in Sweden
estimated the proportions of stem vs branch biomass at 80 –20
per cent, for the range of diameters shown in Figure 5, and the
equations developed in this study estimated these proportions as
70 –30 per cent. In terms of stand characteristics, the differences
might for instance be caused by the fact that in the UK and
Finland, the datasets were from mixed stands, or the higher
density of stands in northern Europe compared with those in
Galicia. Moreover, in contrast with the other studies, this study
was developed in a temperate zone. The general equation devel-
oped by Schroeder et al. (1997) for estimate aboveground tree
biomass for temperate broadleaf species (including birch) in the
USA generates similar results as the biomass equations developed
in this study. At the same time, the model system predicts a mean
aboveground total biomass growth of 2 –8 Mg ha21 year21 for the
data used in this study. Considering an average of 50 per cent of
carbon, the whole stand C sinks would be 1 –4 Mg ha21 year21,
which is consistent with the values reported by Nabuurs and Schel-
haas (2003), i.e. whole stand C sinks between 2– 3 Mg ha21 year21
for deciduous species in Galicia. Finally, the results underline the
importance of parameterizing equations for the specific region,
and that equations developed in northern Europe should be used
with caution in southern Europe.
The model system presented in this study predicts a maximum
mean annual increment (MAI) of 3–10 m3 ha21 year21 with
optimal volume rotations between 30 and 60 years. Higher volume
yield over shorter rotations was observed than in northern and
central Europe. According to the Finnish yield tables (Koivisto, 1959),
MAI varies between 4 and 6.75 m3 ha21 year21 in naturally regener-
ated birch stands and according to the yield tables for silver birch in
Central Europe (Schwappach, 1903), MAI is 4.9 m3 ha21 year21,
A dynamic volume and biomass growth model system
both with rotations of 80 years. However, similar volume yields
were observed in this study as those from managed silver birch plan-
tations in Finland (MAI 6–9.3 m3 ha21 year21 and rotations of 60
years; Oikarinen, 1983), for silver birch stands on good sites in
Sweden (MAI 10 m3 ha21 year21 and rotations of 30–60 years;
Dahlberg et al., 2006), and for downy birch stands in Ireland
(MAI 8 m3 ha21 year21 in unmanaged stands and rotations of
50–60 years; Nieuwenhuis and Barrett, 2001, 2002).
In northern Europe, the main aim of growing downy birch has
been to produce low-cost pulpwood and fuelwood; therefore,
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
heavy pre-commercial and commercial thinnings are not recom-
mended (Hynynen et al., 2010). If the management goal is the pro-
duction of fuelwood, these authors recommend carrying out only
one thinning when dominant height is 10 –11 m, and for pulpwood
production to carry out two thinnings (the first when dominant
height is 4 –6 m to a density of 2000 –2500 trees ha21 and the
second when dominant height is 13 –14 m to a density of
1000 trees ha21). In southwest Europe there are other species
more appropriate for pulpwood production; therefore, downy
birch should be considered on poor sites for improvement and fuel-
wood production and on good sites to grow veneer or sawlog.
According to Álvarez Álvarez et al. (2000), to produce quality
timber in birch, four thinnings are necessary to sufficiently reduce
stand density: the first at 2– 3 years to 1300 trees ha21 (not neces-
sary in plantations with an initial density of 1100 trees ha21, i.e.
trees at 3 × 3 m), the second at 10–15 years to 800 trees ha21,
the third at 20 years to 500 trees ha21, and the final thinning at
25 years to 300 trees ha21. The final cutting takes place at the
age 40– 45 years. Figure 6 shows the changes in stand volume pre-
dicted by this model system for this silvicultural regime, for an
average to high site index stand (H ¼ 15 m at 20 years).
The relatively simple structure of the model system completed
in this study (Figure 1) makes it suitable for use with decision
support systems that enable forest managers to optimize man-
agement strategies (Gadow and Pukkala, 2008), taking into
account economic aspects (timber production) and environmental
considerations related to climate change (biomass production).
Nevertheless, because of the large number of calculations
needed to obtain outputs (especially those involving the use of
the disaggregation system), the model has been integrated into
Figure 6 Simulation of changes in stand volume for the silvicultural regime
proposed by Álvarez Álvarez et al. (2000) and for an average to high site
index stand (H ¼ 15 m at 20 years).
173
Forestry
a forest growth simulator called GesMO# 2.1 (GesMO Support,
2012) to facilitate its practical use by forest managers.
Acknowledgements
This study was completed during a stay by the first author at UCD Forestry in
the School of Agriculture and Food Science, University College Dublin
(Ireland).
Conflict of interest statement
None declared.
Funding
This work was supported by the Spanish Ministry of Science and Innovation
through the research project ‘Modelos de evolución de bosques de frondo-
sas autóctonas del noroeste peninsular’ (AGL2007-66739-C02-01),
co-financed by the European Union through ERDF (European Regional De-
velopment Fund).
References
Álvarez Álvarez, P., Barrio Anta, M., Dı́az Varela, R.A., Riesco Muñoz, G.,
Rigueiro Rodrı́guez, A., Rodrı́guez Soalleiro, R.J. and Villarino Urtiaga, J.J.
2000 Manual de silvicultura de frondosas caducifolias. Escuela Politécnica
Superior. Universidad de Santiago de Compostela, Santiago de
Compostela, Spain. 113 pp.
Álvarez-González, J.G., Zingg, A. and Gadow, K.V. 2010 Estimating growth in
beech forests: a study based on long term experiments in Switzerland. Ann.
For. Sci. 67, 307p1–307p13.
Amateis, R.L. 2000 Modeling response to thinning in loblolly pine
plantations. South. J. Appl. For. 24, 17 –22.
Barrio Anta, M., Castedo Dorado, F., Diéguez-Aranda, U., Álvarez González,
J.G., Parresol, B.R. and Rodrı́guez, R. 2006 Development of a basal area
growth system for maritime pine in northwestern Spain using the
generalized algebraic difference approach. Can. J. For. Res. 36, 1461– 1474.
Belsley, D.A. 1991 Conditioning Diagnostics, Collinearity and Weak Data in
Regression. Wiley, 396 pp.
Bi, H. 2001 The self-thinning surface. For. Sci. 47, 361– 370.
Bi, H. 2004 Stochastic frontier analysis of a classic self-thinning experiment.
Austral Ecol. 29, 408–417.
Bi, H., Wan, G. and Turvey, N.D. 2000 Estimating the self-thinning boundary
line as a density-dependent stochastic biomass frontier. Ecology. 81,
1477– 1483.
Bi, H., Turner, J. and Lambert, M. 2004 Additive biomass equations for native
eucalypt forest trees of temperate Australia. Trees 18, 467–479.
Bi, H., Long, Y., Turner, J., Lei, Y., Snowdon, P., Li, Y., Harper, R., Zerihun, A. and
Ximenes, F. 2010 Additive prediction of aboveground biomass for Pinus
radiata (D. Don) plantations. For. Ecol. Manage. 12, 2301 –2314.
Bunce, R.G.H. 1968 Biomass and production of trees in a mixed deciduous
woodland. I. Girth and height as parameters for the estimation of tree dry
weight. J. Ecol. 56, 759–775.
Cao, Q.V. 2004 Predicting parameters of a Weibull function for modelling
diameter distributions. For. Sci. 50, 682–685.
Cao, Q.V., Burkhart, H.E. and Lemin, R.C. 1982 Diameter distributions and
yields of thinned loblolly pine plantations. School of Forestry and Wildlife
Resources, VPI and SU, Publication No. FSW-1-82.
174
Castedo Dorado, F., Diéguez-Aranda, U. and Álvarez González, J.G. 2007a A
growth model for Pinus radiata D. Don stands in north-western Spain. Ann.
For. Sci. 64, 453–465.
Castedo Dorado, F., Diéguez-Aranda, U., Barrio-Anta, M. and Álvarez
González, J.G. 2007b Modelling stand basal area growth for radiata pine
plantations in Northwestern Spain using the GADA. Ann. For. Sci. 64,
609–619.
Castroviejo, S., Laı́nz, M., López González, G., Monteserrat, P., Muñoz
Garmendia, F., Paiva, J. and Villar, L. 1990 Flora ibérica: Plantas vasculares
de la Penı́nsula Ibérica e Islas Baleares. Volumen II: Platanaceae-
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
Plumbaginaceae (partim.). CSIC, 897 pp.
Cieszewski, C.J. and Bailey, R.L. 2000 Generalized algebraic difference
approach: theory based derivation of dynamic site equations with
polymorphism and variable asymptotes. For. Sci. 46, 116–126.
Cieszewski, C.J., Harrison, M. and Martin, S.W. 2000 Practical methods for
estimating non biased parameters in self-referencing growth and yield
models. University of Georgia PMRC-TR 2000– 7.
Clutter, J.L. and Jones, E.P. 1980 Prediction of growth after thinning in
old-field slash pine plantations. USDA Forest Service, Res. Pap. SE-217.
Clutter, J.L., Fortson, J.C., Pienaar, L.V., Brister, G.H. and Bailey, R.L. 1992
Timber management: a quantitative approach. 2nd edn. Krieger, 333 pp.
Cunia, T. and Briggs, R.D. 1984 Forcing additivity of biomass tables: some
empirical results. Can. J. For. Res. 14, 376–384.
Dahlberg, A., Egnell, G., Bergh, J., Rytter, L. and Westling, O. 2006 Miljöeffekter
avskogsbränsleuttag och askåterföring i Sverige – En syntes av
Energimyndighetens forskningsprogram 1997 till 2004. Energimyndigheten,
Rapport ER2006:44, Eskilstuna, pp. 1–211.
Diéguez-Aranda, U., Castedo-Dorado, F., Álvarez González, J.G. and
Rodrı́guez Soalleiro, R. 2005a Modelling mortality of Scots pine (Pinus
sylvestris L.) plantations in the northwest of Spain. Eur. J. Forest Res. 124,
143–153.
Diéguez-Aranda, U., Castedo-Dorado, F. and Álvarez González, J.G. 2005b
Funciones de crecimiento en área basimétrica para masas de Pinus
sylvestris L. procedentes de repoblación en Galicia. For. Syst. (formerly
Investigación Agraria: Sistemas y Recursos Forestales) 14, 253–266.
Diéguez-Aranda, U., Castedo-Dorado, F., Álvarez-González, J.G. and
Rodrı́guez-Soalleiro, R. 2006a Dynamic growth model for Scots pine
(Pinus sylvestris L.) plantations in Galicia (north-western Spain). Ecol.
Model. 191, 225–242.
Diéguez-Aranda, U., Grandas-Arias, J.A., Álvarez González, J.G. and Gadow,
K.v. 2006b Site quality curves for birch stands in North-Western Spain. Silva
Fenn. 40, 631– 644.
Eid, T. and Tuhus, E. 2001 Models for individual tree mortality in Norway. For.
Ecol. Manage. 154, 69 –84.
Freese, F. 1960 Testing accuracy. For. Sci. 6, 139–145.
Gadow, K.v. 1996 Modelling growth in managed forests – realism and limits
of lumping. Sci. Total Environ. 183, 167–177.
Gadow, K.v. and Pukkala, T. 2008 Designing Green Landscapes. Managing
Forest Ecosystems Vol. 15. Springer.
Garcı́a, O. 1988 Growth modelling – a (re)view. N. Z. J. For. 33, 14– 17.
Garcı́a, O. 1990 Growth of thinned and pruned stands. In Proceedings of a
IUFRO Symposium on New Approaches to Spacing and Thinning in
Plantation Forestry. James, R.Ñ. and Tarlton, G.L. (eds). Rotorua, New
Zealand, Ministry of Forestry, FRI Bulletin No. 151, 84–97.
Garcı́a, O. 1994 The state-space approach in growth modelling. Can. J. For.
Res. 24, 1894– 1903.
Garcı́a, O. 2001 On bridging the gap between tree-level and stand-level
models. In Proceedings of IUFRO 4.11 Conference. Forest Biometry,

Modelling and Information Science. Rennolls, K. (ed). University of
Greenwich.
Garcı́a, O. 2003 Dimensionality reduction in growth models: an example.
For. Biometry Model. Inf. Sci. 1, 1– 15.
GesMO support. 2012 Descargar GesMO# 2009. http://www.usc.es/uxfs/
Libros,44 (accessed on 14 February 2012).
Gómez-Garcı́a, E., Diéguez-Aranda, U., Castedo-Dorado, F. and
Crecente-Campo, F. 2013a A comparison of model forms for the
development of height– diameter relationships in even aged stands. For.
Sci. (in press).
Gómez-Garcı́a, E., Crecente-Campo, F. and Diéguez-Aranda, U. 2013b
Selection of mixed-effects parameters in a variable exponent taper
equation for birch trees in northwestern Spain. Ann. For. Sci. 70, 707–715.
Greene, W.H. 2000 Econometric Analysis. 4th edn. Prentice Hall, 1004 pp.
Harper, J.L. 1977 Population Biology of Plants. Academic Press, 900 pp.
Huang, S., Yang, Y. and Wang, Y. 2003 A critical look at procedures for
validating growth and yield models. In Modelling Forest Systems. Amaro,
A., Reed, D. and Soares, P. (eds). CAB International, pp. 271–293.
Hynynen, J., Niemistö, P., Viherä-Aarnio, A., Brunner, A., Hein, S. and Velling,
P. 2010 Silviculture of birch (Betula pendula Roth and Betula pubescens
Ehrh.) in northern Europe. Forestry 83, 103–119.
Koivisto, P. 1959 Growth and yield tables. Commun. Inst. For. Fenn. 51, 1 –49.
Kozak, A. 1970 Methods of ensuring additivity of biomass components by
regression analysis. For. Chron. 46, 402–404.
Kozak, A. and Kozak, R.A. 2003 Does cross validation provide additional
information in the evaluation of regression models? Can. J. For. Res. 33,
976–987.
Lundqvist, B. 1957 On the height growth in cultivated stands of pine and
spruce in Northern Sweden. Medd. Fran Statens Skogforsk. Band 47, 1– 64.
Marklund, G. 1988 Biomassafunktioner för tall, gran och björk i Sverige.
Sveriges lantbruksuniversitet, Institutionen för skogstaxering, Rapport 45,
1 –73.
Montero, G., Ruiz-Peinado, R. and Muñoz, M. 2005 Producción de biomasa y
fijación de CO2 por los bosques españoles. Monografı́as INIA, Serie Forestal
13. Instituto Nacional de Investigación y Tecnologı́a Agraria y
Alimentaria, Ministerio de Educación y Ciencia. Madrid, Spain, 270 pp.
Nabuurs, G.J. and Schelhaas, M.J. 2003 Spatial distribution of whole-tree
carbon stocks and fluxes across the forests of Europe: where are the
options for bio-energy? Biomass Bioenerg. 24, 311–320.
Nieuwenhuis, M. and Barrett, F. 2001 Volume production patterns in six
downy birch stands in Ireland. Ir. For. 58, 46– 58.
Nieuwenhuis, M. and Barrett, F. 2002 The growth potential of downy birch
(Betula pubescens (Ehrh.)) in Ireland. Forestry 75, 75 –87.
Oikarinen, M. 1983 Growth and yield models for silver birch (Betula pendula)
plantations in southern Finland . Commun. Inst. For. Fenn. 113, 1– 75.
Parresol, B.R. 1999 Assessing tree and stand biomass: a review with
examples and critical comparisons. For. Sci. 45, 573– 593.
Parresol, B.R. 2001 Additivity of nonlinear biomass equations. Can. J. For. Res.
31, 865– 878.
Peet, R.K. and Christensen, N.L. 1987 Competition and tree death. Bioscience
37, 586– 595.
Pienaar, L.V. and Shiver, B.D. 1984 An analysis and models of basal area
growth in 45-year-old unthinned and thinned slash pine plantation plots.
For. Sci. 30, 933–942.
A dynamic volume and biomass growth model system
Repola, J. 2008 Biomass equations for birch in Finland. Silva Fenn. 42,
605–624.
Ruiz-Peinado, R., Rı́o, M. and Montero, G. 2011 New models for estimating
the carbon sink capacity of Spanish softwood species. For. Syst. (formerly
Investigación Agraria: Sistemas y Recursos Forestales). 20, 176– 188.
Ruiz-Peinado, R., Montero, G. and Rı́o, M. 2012 Biomass models to estimate
carbon stocks for hardwood tree species. For. Syst. (formerly Investigación
Agraria: Sistemas y Recursos Forestales). 21, 42– 52.
SAS Institute Inc. 2008 SAS/ETSw 9.2. User’s Guide. SAS Institute Inc., Cary,
NC,2861 pp.
Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018
Schroeder, P., Brown, S., Mo, J., Birdsey, R. and Cieszewski, C. 1997 Biomass
estimation for temperate broadleaf forests of the United States using
inventory data. For. Sci. 43, 424–434.
Schwappach, A. 1903 Beiträge zur Kenntnis der Wuchsleistung von
Birkenbestände n. Z. Forst- Jagdwes. 35, 479– 484.
Snorrason, A. and Einarsson, S.F. 2006 Single-tree biomass and stem
volume functions for eleven tree species used in Icelandic forestry.
Icelandic Agric. Sci. 19, 15 –24.
Vanclay, J.K. 1994 Modelling forest growth and yield. Applications to mixed
tropical forests. CAB International, 312 pp.
Weiskittel, A.R., Hann, D.W., Kershaw, J.A. Jr. and Vanclay, J.K. 2011 Forest
Growth and Yield Modeling. Wiley-Blackwell, 415 pp.
White, H. 1980 A heteroskedasticity-consistent covariance matrix estimator
and a direct test for heteroskedasticity. Econometrica 48, 817–838.
Woollons, R.C. 1998 Even-aged stand mortality estimation through a
two-step regression process. For. Ecol. Manage. 105, 189– 195.
Yang, Y.Q., Monserud, R.A. and Huang, S. 2004 An evaluation of diagnostic
tests and their roles in validating forest biometric models. Can. J. For. Res.
34, 619– 629.
Yao, X., Titus, S. and MacDonald, S.E. 2001 A generalized logistic model of
individual tree mortality for aspen, white spruce, and lodgepole pine in
Alberta mixedwood forests. Can. J. For. Res. 31, 283– 291.
Zhao, D., Borders, B., Wang, M. and Kane, M. 2007 Modeling mortality of
second rotation loblolly pine plantations in the Piedmont/Upper Coastal
Plain and Lower Coastal Plain of the southern United States. For. Ecol.
Manage. 252, 132–143.
Zianis, D., Muukkonen, P., Mäkipää, R. and Mencuccini, M. 2005 Biomass and
stem volume equations for tree species in Europe. Silva Fenn. Monogr. 4,
63 pp.
Appendix. Components of the model system
developed in previous studies
Dominant height transition function (Diéguez-Aranda et al.,
2006b)
19.80 + XH
H2 = ,
1 + 758.0/XH t−1.398
2

H1 − 19.80 + (19.80 − H1 )2 + 3032H1 t−1.398
1
with XH =
2
(15)
where H1 and t1 represent the predictor dominant height (m)
and age (years), respectively; and H2 is the predicted dominant
175
Forestry

height (m) at age t2 (years). R 2 ¼ 0.99 and RMSE ¼ 0.51 m. Taper equation (Gómez-Garcı́a et al., 2013b)
Reference age for site index estimation ¼ 20 years.
Generalized height –diameter relationship (Gómez-Garcı́a et al., di = 0.9652d0.9421 h0.08482 xc
with
2013a)  
1
c = 4.094 − 0.3939q4 − 0.4112
exp(d/h)
h = 1.3 + (H  
1
   − 4.153x0.1 + 2.813 (17)
1 1 d
− 1.3) exp (2.461 − 0.2336H − 0.2737dg ) − (16)
d d0
+ 0.05345hw + 0.3332x

Downloaded from https://academic.oup.com/forestry/article-abstract/87/1/165/600011 by guest on 13 October 2018

w hi
x= , w = 1 − q1/3 , q=
where h is the predicted total height (m), d is the diameter at breast [1 − (1.3/h)1/3 ] h
height (cm), H is the dominant height (m), d0 is the dominant diam-
eter (cm, the mean diameter of the 100 largest diameter trees per where di is the predicted diameter over bark (cm) at height above
hectare) and dg the quadratic mean diameter (cm). R 2 ¼ 0.78 and ground level hi (m), d is the diameter at breast height (cm) and h
RMSE ¼ 1.76 m. is the total height (m). R 2 ¼ 0.95 and RMSE ¼ 1.56 cm.

176