Madera y Bosques
ISSN: 1405-0471
publicaciones@ecologia.edu.mx
Instituto de Ecología, A.C.
México
Návar, José de Jesús; González, Nicolás; Graciano Luna, José de Jesús; Dale, Virginia; Parresol,
Bernard
Additive biomass equations for pine species of forest plantations of Durango, Mexico
Madera y Bosques, vol. 10, núm. 2, otoño, 2004, pp. 17-28
Instituto de Ecología, A.C.
Xalapa, México
Available in: http://www.redalyc.org/articulo.oa?id=61710202
How to cite
Complete issue
More information about this article
Journal's homepage in redalyc.org
Scientific Information System
Network of Scientific Journals from Latin America, the Caribbean, Spain and Portugal
Non-profit academic project, developed under the open access initiative
Madera y Bosques 10(2), 2004:17-28
17
ARTÍCULO DE INVESTIGACIÓN
Additive biomass equations for pine species
of forest plantations of Durango, Mexico
José de Jesús Návar Cháidez1
Nicolás González Barrientos2
José de Jesús Graciano Luna3
Virginia Dale4
Bernard Parresol5
ABSTRACT
Statistical analysis between three weighted additive biomass equations are presented for
planted pine species typical of the coniferous forests of the Western Sierra Madre mountain range of
Durango, Mexico. Statistical and graphical analyses were used to select the best single and multiple
individual biomass component equation. Linear equations better fitted the biomass components.
Therefore, three linear additive procedures were tested: (i) the conventional, (ii) a harmonization, and
(iii) the seemingly-unrelated regression in two types of equations of component biomass estimation
using both simple regression and multiple regression techniques. These tests were performed at two
scales: (a) each of three pine species and (b) all three species. For both the simple linear and best
multiple regression equation, the seemingly-unrelated equations provided more precise biomass
component estimates, with tendencies consistent with the conventional non-additive non-linear regression procedures, and provided average biomass component estimates when equations were applied
to a data set of 23 sample quadrants.
KEY WORDS:
Biomass additivity, P. cooperi, P. durangensis, P. engelmannii, seemingly unrelated regression.
RESUMEN
Se presentan análisis estadísticos entre tres ecuaciones aditivas ponderadas para el cálculo de
la biomasa en plantaciones de especies de pino típicas de los bosques de coníferas de la Sierra
Madre Occidental en el Estado de Durango, México. Se usaron análisis estadísticos y gráficos para
seleccionar las mejores ecuaciones para los componentes de la biomasa, de manera individual o
múltiple. Por lo tanto, se ensayaron tres procedimientos aditivos lineales: (i) el convencional, (ii) una
armonización, y (iii) la regresión generalizada no relacionada en dos tipos de ecuaciones para estimación de los componentes de la biomasa utilizando técnicas tanto de regresión simple lineal como
múltiple. Estas pruebas se llevaron a cabo en dos escalas: (a) cada una de las tres especies de pino
y (b) las tres especies en conjunto. Tanto para las ecuaciones de regresión lineal simple como para
las de regresión múltiple, las ecuaciones generalizadas no relacionadas dieron estimaciones más
precisas de los componentes de la biomasa, con tendencias consistentes con los procedimientos
convencionales no aditivos de regresión múltiple no lineal y arrojaron estimaciones promedio de los
componentes de la biomasa cuando las ecuaciones se aplicaron a un conjunto de datos de una
muestra de 23 cuadrantes.
PALABRAS CLAVE:
Aditividad de la biomasa, P. cooperi, P. durangensis, P. engelmannii, regresión generalizada no relacionada.
1
2
3
4
5
Universidad Autónoma de Nuevo León. Facultad de Ciencias Forestales. Km 145 Carretera Nacional.
Linares 67700 N.L., México. E-mail: jnavar@ccr.dsi.uanl.mx.
Centro de Bachillerato Tecnológico Forestal. Durango, Mexico.
Instituto Tecnologico Forestal No 1. El Salto, P.N., Durango. Durango, Mexico.
Environmental Science Division, Oak Ridge National Laboratory. Oak Ridge, TN. E-mail: dalevh@ornl.gov.
USDA Forest Service Southern Research Station. P.O. Box 2680 Asheville, NC 28802. E-mail:
bparresol@fs.fed.us.
18
Aditive biomass equations for pine species. Návar et al.
INTRODUCTION
Estimating
the
environmental
services provided by carbon sequestration requires the use of biomass estimates (Houghton, 1991; Brown, 1997).
Several models estimate biomass or
biomass-related parameters to assess the
fate of CO2 in the biosphere ( Mohren and
Goldewijk, 1990). Given this importance,
biomass equations have been compiled
or developed for tropical, temperate,
boreal, and semi-arid, sub-tropical trees
(Schroeder et al., 1997; Ter-Mikaelian and
Korzukhin, 1997; Návar et al., 2002).
A variety of regression models have
been used for estimating total tree
biomass or tree biomass components
(Clutter et al., 1983), and they fall into
three main forms: linear additive, nonlinear additive, and nonlinear multiplicative
error (Parresol, 1999). A desirable feature
of tree component regression equations is
that the sum of predictions for the components equals the prediction for the total
tree (Cunia and Briggs, 1984, 1985;
Parresol, 1999). Procedures for forcing
additivity have been proposed. Total
biomass is estimated by (1) using the
same independent variables for estimating each component, (2) adding the most
appropriate regression functions of each
biomass component, and (3) forcing the
regression coefficients to estimate total
biomass (Cunia and Briggs, 1985;
Parresol, 1999). These procedures have
been applied widely for temperate trees
(Cunia and Briggs, 1985; Parresol, 1999)
and semi-arid, subtropical trees (Návar et
al., 2002). However, biomass component
equations for young pine plantations that
meet the additive requirements are rare in
the scientific literature, when currently
there is a wide range of forest plantations
scattered over the Western Sierra Madre
of Mexico. For the period between 1993
and 1998 approximately 5 000 hectares
were annually planted only in the State of
Durango. Regardless of their importance,
no attempts have been made to quantify
the environmental services provided by
these forest plantations. The projection of
carbon stocks in the early stages of forest
plantations requires established additive
biomass equations.
OBJECTIVES
The aim of this study was (1) to
select the best single and multiple regression equation to estimate biomass
components, (2) to fit three biomass additivity procedures in two different forms, (2)
to compare efficiency in biomass estimates between two scales, and (3) to
contrast biomass estimates among additive procedures in 23 planted forest
stands of Durango, Mexico.
MATERIALS AND METHODS
This research was conducted in
forests managed by the Unidad de
Conservacion y Desarrollo Forestal No 6
of the Western Sierra Madre mountain
range of Durango, Mexico. Forest plantations of several community-based land
ownership, ejidos, including La Campana,
San Pablo, La Ciudad, Los Bancos, La
Victoria, located in the municipality of
Pueblo Nuevo, Durango, were sampled.
The ejidos are within the coordinates
105º36’19’’W and 105º51’48’’W and
24º19’05’’N and 24º30’16’’N and lie
between 2 000 meters and 2 900 meters
above sea level (masl). The area is
characterized by a cold-temperate climate
with average annual long-term precipitation and temperature of 900 mm and 15
ºC, respectively. Plant cover is dominated
by native, uneven aged, mixed coniferous
forests. Pines and oaks are the most
abundant genera in the western Sierra
Madre mountain range.
Madera y Bosques 10(2), 2004:17-28
19
dure ii. It causes the equations of individual biomass components to behave
harmoniously and never exceeding
biomass estimates of the upper compartment (i.e., bark would never exceed total
stem biomass estimates). Procedure iii is
an extension of procedures CON and
HAR. Cunia and Briggs (1984; 1985) and
Parresol (1999) reported examples of
SUR applied to additive best regression
equation. Recently, Parresol (2001)
reported two different examples of SUR
by developing further into non-linear
regression.
Methodology. Biomass components of
56 trees of P. durangensis (25), P. cooperii
(19), and P. engelmannii (12) were
measured. In each of 23 forest plantations, at least two trees were selected
from each forest stand for biomass
measurements. Trees were felled and
separated into biomass components leaf,
branch and stem. Biomass components
were weighed fresh, and samples of 15 %
of each component were collected for
ovendry analysis. All trees of each of the
twenty-three forest plantations were
measured in basal diameter (D), top
height (H), and canopy cover (CT) (Table
1) in quadrants of 20 m x 30 m.
Preliminary exploration of biomass
component equations. Scatterplots of
the biomass data, stepwise regression
procedures, and residual analysis were
used to select the best single and multiple
regression equations. The dependent
variables were the biomass components
leaf, branch, and stem and their log-transformations. The explanatory independent
variables used were D, H, the combined
variable D2H and their log-transformations. Goodness of fit statistics were used
to select the most appropriate single and
multiple variable models. Once the
regression equations were selected on
the basis of the goodness of fit statistics,
Additive biomass equations. Three
procedures were used to develop the
additive biomass component equations:
(i) the conventional procedure (CON), (ii)
a harmonization procedure (HAR), and
(iii) seemingly-unrelated regression
(SUR). Clutter et al. (1983) discuss a wide
range of multiple equations that use top
height, basal diameter, and interactions of
these variables. These equations are
classified as best regression models and
they fall into the category of procedure i.
Cunia and Briggs (1985) defined harmonization, which falls within additive proce-
Table 1. Statistics for 56 sample trees used for developing additive biomass
equations for three pine species planted in Durango, Mexico
SPECIES
TREE PARAMETERS
AGE
(years)
BASAL
DIAMETER
(cm)
TOP
HEIGHT
(m)
COVER
2
(m )
M
S
M
S
M
S
M
S
P. cooperii
13,9
5,8
12,4
3,8
6,3
3,0
4,2
2,5
P. durangensis
14,1
4,8
13,4
4,2
5,9
2,3
5,4
2,8
P. engelmannii
14,8
5,2
12,5
3,8
4,4
1,3
5,0
3,4
All Species
14,3
5,3
12,8
3,9
5,5
2,2
4,8
2,9
20
Aditive biomass equations for pine species. Návar et al.
coefficient of variation of residuals, CV,
mean percent standard error, S(%), and
percent error, Pe. Information on these
statistics is given in Parresol (1999). S(%)
is based on the ratio between the absolute error and the biomass estimate, while
Pe is based on the ratio between the estimated and observed biomass data minus
1. Therefore, S(%) and Pe are estimates
of the model deviance.
scatterplots of squared residuals and the
explanatory variables were regressed to
obtain the weigthing equations. Weighting
is a statistical procedure required to
reduce the increasing biomass variance in
larger trees. It is conducted on developing
equations to predict the biomass error as
a function of the independent variables.
Three additive procedures were tested:
CON, HAR, and SUR. Examples using
one and multiple variables are given in
equations (1) and (2), equations (3) and
(4) are for CON and HAR additive procedures, respectively. SUR was developed
with equations (1) and (2) and the parameters of biomass component equations
were restricted to produce total biomass.
The goodness of fit statistics were
estimated for each biomass component
separately. The goodness of fit statistics
of the log-transformed linear square
procedures are not compatible. Hence,
when variables were log-transformed, first
the parameters were estimated with the
log-transformation procedure and then
biomass components were computed by
the commonly used correction factor
(y = exp(g(y)+σ2/2)), where σ2 is the variance
The goodness of fit statistics used to
select the single and best regression
equation were the coefficient of determination or fit index, r2, standard error, Sx,
yˆ leaf = b10 + b11(D 2H )
yˆ branch = b20 + b21(D 2H )
(1)
yˆ stem = b30 + b31(D 2H )
yˆ total = yˆ leaf + yˆ branch + yˆ stem = (b10 + b20 + b30 ) + (b11 + b21 + b31)(D 2H )
yˆ leaf = b10 + b11(D ) + ... + b1k (D 2H ) + b12 (ln D ) + ... + b1n (ln D 2H )
yˆ branch = b20 + b21(D ) + ... + b2k (D 2H ) + b2k +1(ln D ) + ... + b2n (ln D 2H )
yˆ stem = b30 + b31(D ) + ... + b3k (D 2H ) + b3k +1(ln D ) + ... + b3n (ln D 2H )
(2)
yˆ total = yˆ leaf + yˆ branch + yˆ stem
(3)
yˆ total = b0 + b11(D 2H ) + b12 (D 2H ) + b13 (D 2H )
[
]
yˆ total = b0 + b11(D ) + ... + b1k (ln D 2H )
[
2
] [
+ b21(D ) + ... + b2n (ln D H ) + b31(D ) + ... + b3m (ln D 2H )
]
(4)
Madera y Bosques 10(2), 2004:17-28
21
of the log-transformed regression).
Observed and computed biomass in
original units provided information to estimate the goodness of fit statistics in
compatible units. Least square techniques in nonlinear, linear, multiple regression, multiple regression with dummy
variables, linear and ln-transformed
system of equation procedures was used
to compute parameters. The Newton’s
method in SAS package was employed in
the nonlinear procedures.
D2H as independent variable, provided the
best goodness of fit statistics for stem and
total biomass for all species, as well as for
each species (Table 2). Therefore, the
linear regression using D2H as independent variable was used to further explore
the additive procedures CON, HAR, and
SUR. For all three additive procedures,
the biomass component equations are:
Comparisons between biomass estimates of independent data sets. The
performance of additive procedures has
been assessed on the standard error of
the parameter estimates. In this report the
t values were contrasted among additive
procedures since parameter estimates
also change with the additive procedure.
Additive equations were also applied to all
standing trees of each of the 23 quadrants
to contrast biomass estimates and confidence intervals for each biomass component for each additive procedure for both
species scales.
y stem = b30 + b31(D 2H )
RESULTS
Selection of equations for additive
procedures. Different goodness of fit
statistics resulted from the regression
procedures employed to estimate
biomass components in simple and
multiple regression procedures. Clearly
the simple and multiple linear equations fit
better biomass components and total. The
non-linear (i.e. B = α Dβ) and the log transformed (i.e. lnB = α + β ln (D2h))
regression equations had the poorest
goodness of fit statistics for branch, stem,
and total biomass for all species, as well
as for each species but because of poor
performance the coefficients are not
presented in the manuscript. For the
simple regression approach, the linear
equation, using the combined variable
y leaf = b10 + b11(D 2H )
y branch = b20 + b21(D 2H )
y total = y leaf + y branch + y stem
= b40 =(10 + 20 + 30 ) + b41=(11+ 21+ 31) (D 2H )
The multiple linear regression fit the
biomass data better for all biomass
components with the exception of the leaf
component, in contrast to the log-transformed multiple regression model.
Goodness of fit statistics improved by 4,
9, 6, -114, and -30 % for the r2, Sx, CV,
S(%), and Pe, when using multiple linear
as compared to using log-transformed
multiple regression equations to estimate
biomass components. Therefore, for the
best regression model, the multiple linear
regression was used to further develop
the CON, HAR, and SUR additive procedures. The final regression equations for
each biomass component, for each
species, and for all species were the following:
P. duranguensis:
y leaf = b10 + b11D + b12 ln D 2H
y branch = b20 + b21D 2H
y stem = b30 + b31D 2H
y total = b40 + b41D + b42 ln D 2H + b43D 2H
22
Aditive biomass equations for pine species. Návar et al.
P. cooperii:
y leaf = b10 + b11D 2H + b12 ln D + b13 ln D 2H
y branch = b20 + b21D + b22 ln D
y stem = b30 + b31D 2H
y total = b40 + b41D 2H + b42 ln D + b43D
+ b44 ln D 2H
P. engelmannii:
y leaf = b10 + b11D + b12 ln D 2H
y branch = b20 + b21D 2H
y stem = b30 + b31D + b32 ln D 2H
2
y total = b40 + b41D + b42 ln D H
+ b43D 2H
All Species:
y leaf = b10 + b11D + b12 ln H
y branch = b20 + b21D + b22 ln D
y stem = b30 + b31D 2H
y total = b40 + b41D + b42 ln H + b43 ln D
+ b44D 2H
In general, and as expected, the
multiple regression procedures improved
predictions of biomass components in
contrast to the simple regression equations (Table 2). Stem biomass was consistently well predicted by using the single
independent variable (D2H) in both the
simple and multiple regression equations.
For the rest of biomass components, the
goodness of fit statistics improved the
coefficient estimates by 19 %, 44 %, 10
%, and 28 % for the r2, Sx, Cv, and Pe,
respectively, when using multiple linear
equations. An equation for each species
improves the coefficient estimates when
computing total biomass, in contrast to
using a single biomass equation for all
species. When using the best multiple
linear equation, the coefficient estimates
improves on the average by 14 %.
The weighing equations that
predicted the error as a function of the
independent variables, resulted in power
functions (i.e. e2 = α (D2H)β)(37,5 %) and
linear (i.e. e2 = α + β (D2H)) (62,5 %) functions of the independent variables, stressing the heteroscedasticity of the biomass
data (Cunia and Briggs, 1984; Parresol,
1999). This variation was functionally
related to the predictor variables in the
regression, although the coefficients of
determination hardly surpassed 0,50.
Biomass estimates for trees with equations developed. Additive procedures
CON and SUR computed unbiased total
biomass when using one independent
variable. In contrast the HAR additive
procedure (ii) underestimated leaf but
overestimated stem biomass components
when using one single independent
variable for one equation for all species.
However, the HAR procedure provided
unbiased total biomass estimates for each
species. For the best multiple regression
models, the SUR procedure slightly
overestimated total biomass but for only
P. engelmannii. The potential causes of
this deviation could not be determined.
However, a poor regression fitting may
have explained this deviation.
Contrasting additive procedures
Standard errors on the parameter estimates. The standard errors of the parameter estimates are smaller, often by
several orders of magnitude in SUR in
contrast to CON and HAR procedures
(Table 3). Therefore, the t values are also
several orders of magnitude larger when
using SUR procedures. For the simple
variable equation, the average t values
increase approximately 27 % and 1 500 %
in additive procedure SUR in contrast to
Madera y Bosques 10(2), 2004:17-28
23
and 4 279 % in contrast to the t values
estimated for the conventional and
harmonization procedures, respectively.
The harmonization procedure again
consistently had higher t values, on the
average 1 057 % in contrast to additive
procedure CON.
additive procedures CON and HAR,
respectively. The harmonization procedure also consistently increased the t
values in contrast to the CON procedure
by an average of 38 %. For the best
regression equation, the t values increased in additive procedure SUR by 418 %
Table 2. Statistics of goodness of fit for selecting a single and a multiple equation
to test three additivity procedures for total biomass for pine trees planted in
Durango, Mexico
SPECIES
EQUATION
GOODNESS OF FIT STATISTICS
2
All species
P.durangensis
P. cooperii
P. engelmannii
r (%)
Sx(kg)
CV(%)
S(%)
Pe(%)
LS
82
6,2
34
22
80
NLS
68
8,1
45
37
104
LTS
81
6,2
35
24
85
ML
86
5,6
31
22
81
MNL
83
6,1
34
20
77
LS
88
5,4
28
22
69
NLS
62
9,9
52
42
289
LTS
80
7,1
37
27
62
ML
89
5,6
29
24
67
MNL
86
6,2
32
22
66
LS
89
4,2
25
16
31
NLS
73
6,8
40
35
148
LTS
89
4,3
26
20
28
ML
94
3,6
21
14
22
MNL
93
3,9
23
13
20
LS
86
4,9
32
39
69
NLS
52
9,1
59
43
110
LTS
68
7,4
48
24
36
ML
88
5,1
33
34
40
MNL
85
5,6
37
18
27
LS = Simple Linear
NLS = Simple Non-linear
LTS = Simple Log-transformed
ML = Multiple Linear
MNL = Multiple non-linear regression
24
Aditive biomass equations for pine species. Návar et al.
Table 3. Parameter (E), standard error (Sx), and changes in t values from fitting three
additive techniques for multiple regression models for pine biomass components and
total for all species
S
All
Species
P
CONVENTIONAL
HARMONIZATION
SEEMINGLY
UNRELATED
REDUCTION IN t
(%)
E
Sx
E
Sx
E
Sx
1vs2
1vs3
2vs3
b10
-0,43
0,72
0,287
0,84
-1,13
0,35
-157
445
*
b11
0,575
0,06
0,554
6,33
0,353
0,05
-99
-19
*
b12
-2,82
0,52
-3,05
50,3
-0,54
0,3
-99
-67
*
b20
14,95
13,3
0,287
0,84
9,413
2,28
-70
267
*
b21
2,009
0,61
1,357
0,83
1,605
0,19
-50
157
417
b22
-14,7
8,33
-5,57
4,88
-10,3
1,86
-35
216
387
b30
0,285
1,45
0,287
0,84
0,093
0,16
74
196
71
b31
0,009
0,00
0,009
0,00
0,009
0,0003
76
98
13
b40
14,81
6,13
0,862
0,84
8,367
2,14
-58
62
281
b41
2,585
0,83
1,912
1,08
1,959
0,2
-43
215
456
b42
-2,82
0,8
-3,05
50,3
-0,54
0,3
-98
-49
*
b43
-14,7
6,23
-5,57
4,88
-10,3
1,86
-52
136
387
b44
0,009
0,00
0,009
0,00
0,009
0,0003
457
528
13
S = Species
P = Parameter
* values greater than 1000
t (%) = [(E1 / Sx1) - (E2 / Sx2)] / (E2 / Sx2)*100
Estimates of total stand biomass. The
additive procedures computed different
biomass component estimates when
equations were applied to tree data from
23 planted stands. For a single equation
for each pine species, all additive procedures computed similar biomass for the
branch, stem, and total components (Fig
1a). For leaf biomass, for the single independent variable and for the multiple
regression, all additive procedures
computed different biomass estimators.
However, the additive procedures SUR
and HAR computed similar leaf biomass
estimates across single and multiple
regression equations. Biomass computations are biased when developing a single
equation for all species (Fig 1b). For the
single regression equations, the HAR
procedure overestimated average leaf
biomass.
For multiple regression equations,
the SUR regressions overestimated leaf
biomass. For the simple equations, the
SUR procedure underestimated branch
biomass, while for multiple regression
equations, the HAR procedure underestimated branch biomass. For the simple
regression equations, the HAR procedure
overestimated stem biomass. The SUR
and the HAR procedures developed in
multiple regression underestimated total
biomass in contrast to the SUR and HAR
developed in simple regression, when developing a single equation for all species.
Madera y Bosques 10(2), 2004:17-28
25
Average Dry Weight (kg)
20
1a
Each Species
18
16
14
12
10
8
CON1
HAR1
SUR1
CON2
HAR2
SUR2
6
4
2
0
Leaf
Branch
Stem
Total
Average Dry Weight (kg)
20
1b
All Species
18
16
14
12
10
8
6
4
2
0
Leaf
Branch
Stem
Total
Biomass Components
Figure 1. Average biomass and confidence intervals estimated with six different equations in three additive procedures, for simple and multiple regression equations for pine
species planted in Durango, Mexico. CON1 = Conventional, HAR1 = harmonization,
SUR1 = seemingly-unrelated in simple linear regression, CON2 = Conventional, HAR2 =
harmonization, and SUR2 = seemingly-unrelated in multiple linear regression. a) For a
single equation developed for each pine species. b) For a single equation developed for
all pine species.
26
Aditive biomass equations for pine species. Návar et al.
DISCUSSION
Observed Data
Non-linear Regression
Best Regression SUR
60
Total Dry Biomass (kg/tree)
Total Dry Biomass (kg/tree)
Simple and multiple linear regression
equations better fit the biomass component data in contrast to the log-transformed and the non-linear regression
equations. This result contrasts with the
traditional use of simple non-lineal equations developed in most non-additive allometric studies (e.g., Ter Mikaelian and
Korzukhin, 1997; Schroeder et al., 1997).
However, the seemingly-unrelated regression equations developed for this study
resulted in a best regression equation
while the conventional non-linear equations match in a similar way observed
total biomass data for each species and
50
40
2a
P.durangensis
30
20
10
0
0
5
10
15
for all species for the range of measured
data (Fig 2a, 2b, 2c, and 2d). When
biomass projections are required below
the minimum observed basal diameter (5
cm), allometric equations developed in
SUR overestimate total biomass in trees
of the species P. engelmannii. Equations
developed in SUR for all species also
overestimate total biomass in trees with
basal diameter less than 2 cm. This was
observed when projecting the additive
equations in range of 0 cm to 60 cm of
diameter. However, for the range of
observed data, the linear regression
equations mimic the non-linear tendency
well.
20
Observed Data
Non-linear Regression
Best Regression SUR
60
50
40
2b
P.cooperii
30
20
10
0
0
25
Total Dry Biomass (kg/tree)
Total Dry Biomass (kg/tree)
Observed Data
Non-linear Regression
Best Regression SUR
50
40
2c
P.engelmannii
30
20
10
0
0
5
10
15
20
Basal Diameter (cm)
10
15
20
25
Basal Diameter (cm)
Basal Diameter (cm)
60
5
25
60
Observed Data
Non-linear Regression
Best Regression SUR
50
40
2d
All Spp
30
20
10
0
0
5
10
15
20
25
Basal Diameter (cm)
Figure 2. Observed and computed biomass by the conventional non-linear procedures
and the resulting equations in seemingly-unrelated regression using the best model for
each of three pine species and for all pine species planted in Durango, Mexico.
Madera y Bosques 10(2), 2004:17-28
The seemingly-unrelated regression
clearly is the method of choice for fitting
additive biomass component equations.
This additive procedure achieves lower
variance as seen by the confidence
bounds of equation coefficients. Therefore
SUR is a more efficient estimator
(Parresol, 1999). It is more flexible and
takes into account statistical dependencies among sample data by setting constraints on the coefficients (Cunia and
Briggs, 1985; Parresol, 1999). Other
advantages of using SUR include (a)
predictions of the components add up to
the prediction for the total tree and (b) the
coefficients are more consistent (Cunia
and Briggs, 1984; Parresol, 1999).
Moreover, there is an increasing
need for estimating biomass compartments for environmental-related issues,
productivity, and economic values.
However, this procedure is the most difficult additive method to calculate in this
analysis, and predictions beyond the
characteristics of measured trees are
uncertain (Cunia and Briggs, 1984).
Extrapolation uncertainty is also a feature
of all regressions developed in this study.
Should additive procedures be used
to estimate total biomass, attention must
be paid to selecting the right equation
when quantifying total biomass at the
stand scale. In this research, it is shown
that different additive equations might
provide different biomass estimates, and
this finding could also apply to non-additive procedures. Indeed, Woods et al.
(1991) found that most of the error in estimating biomass density was attributed to
the method of parameter estimation.
Another potential sources of error
frequently mentioned is the component
due to the random selection of the sample
unit (Woods et al., 1991; Parresol, 1999).
It is a function of the sampling design, the
sample size, the type of estimator used,
and the inherent variation between the
sample units. The exploration of the latter
27
error sources is of paramount importance
and should be addressed accordingly to
provide reliable stand biomass quantification.
CONCLUSIONS
Equations to estimate aboveground,
standing biomass were developed for
three single species and for the set of
three, typical of small-scale forest plantations of Durango, Mexico. Equations
developed using additive procedures in
seemingly-unrelated regressions computed biomass components with the largest
efficiency and they were consistent with
equations developed in non-additive nonlinear regression. Since they provided
average estimates, additive equations in
seemingly-unrelated multiple linear
regression for each species or all species
are recommended to quantify biomass
components in biomass inventory of the
coniferous forest plantations of Durango,
Mexico.
ACKNOWLEDGMENTS
The Mexican Foundation for Science
and Technology and University Fund for
Science and Technology funded this
project through research grants 28536-B
and CN 323-00, respectively.
REFERENCES
Brown, S. 1997. Los bosques y el cambio
climatico: el papel de los terrenos
forestales como sumideros de
carbono. In: Actas del XI Congreso
Mundial
Forestal:
Recursos
Forestales y Arboles. Vol 1. Antalya
Turky 13-22 October of 1997.
Clutter, J.L.; J.C. Forston; L.V. Pienaar;
G.H. Brister and R.L. Bailey. 1983.
28
Aditive biomass equations for pine species. Návar et al.
Timber management: A quantitative
approach. Wiley, New York. 333 p.
northeastern Mexico. Journal of Arid
Environments 52:167-179.
Cunia, T. and R.D. Briggs. 1984. Forcing
additivity of biomass tables - some
empirical results. Canadian Journal
of Forest Research 14: 376-384.
Parresol, B. 1999. Assessing tree and
stand biomass: a review with examples and critical comparisons. For.
Sci. 45:573-593.
Cunia, T. and R.D. Briggs. 1985. Forcing
additivity of biomass tables - use of
the
generalized
least-square
method. Canadian Journal of Forest
Research 15: 23-28.
Schroeder, P.; S. Brown; J. Mo; R. Birdsey
and C. Cieszewski. 1997. Biomass
estimation for temperate broadleaf
forest of the United States using
inventory data. For. Sci. 43:424-434.
Houghton, R.A. 1991. Tropical deforestation and carbon dioxide. Climate
Change 19: 99-118.
Ter-Mikaelian, M.T. and Korzukhin, M.D.
1997. Biomass equations for sixty
five North American tree species.
Forest Ecology and Management
97:1-24.
Mohren, F. and K. Goldewijk. 1990. CO2
Fix Model. Institute of Forestry and
Nature Research. Wageningen,
Holanda.
Návar C., J.J.; J. Nájera, and E. Jurado.
2002. Biomass estimation equations
in the Tamaulipan thornscrub of
Woods, K.D.; A.H. Feiveson and D.B.,
Botkin. 1991. Statistical analysis for
biomass density and leaf-area index
estimation. Canadian Journal of
Forest Research 21:974-989.±
Manuscrito recibido el 30 de octubre de 2003.
Aceptado el 12 de julio de 2004.
Este documento se debe citar como:
Návar Ch., J.J.; N. González B.; J.J. Graciano L.; V. Dale y B. Parresol. 2004. Additive biomass equations for
pine species of forest plantations of Durango, Mexico. Madera y Bosques 10(2):17-28.