Page 1 of 9
Discoveries in Agriculture and Food Sciences - Vol. 12, No. 5
Publication Date: October 25, 2024
DOI:10.14738/dafs.125.15549.
Egonmwan, Y. I., & Orukpe, W. O. (2024). Comparison of Non-Linear Height-Diameter Models of Tectona grandis L.f. in a Rainforest
Site in Southern Nigeria. Discoveries in Agriculture and Food Sciences, 12(5). 20-28.
Services for Science and Education – United Kingdom
Comparison of Non-Linear Height-Diameter Models of Tectona
grandis L.f. in a Rainforest Site in Southern Nigeria
Y. I. Egonmwan
ORCID https://orcid.org/0000-0001-6516-8881
Department of Forest Resources and Wildlife
Management, University of Benin, Benin City, Nigeria
W. O. Orukpe
Department of Forest Resources and Wildlife
Management, University of Benin, Benin City, Nigeria
ABSTRACT
This study was carried out to select the best model for predicting height for effective
management of Tectona grandis stand in the study area. Five 3-parameters well- established non-linear models were used to compare their suitability in predicting
height, evaluate and select the best H-D model based on test statistics criteria such
as R2, RMSE AIC, and BIC. Models with the lowest RMSE, AIC and BIC and highest R2
values are judged the best. One hundred (100) trees were enumerated and were
subjected for analysis in the study. A plot size of 0.04ha (20m x 20m) with an area
of 0.64ha was chosen for this study and 10 sample plots were purposively selected
from the study area. The result indicated that schnute model had the highest R2
value of 99.9% but it also had the highest RMSE, AIC and BIC values, hence the worst
candidate in predicting height in the study area. Weibull model although did not
have the largest R2 of 97.2% when compared to schnute, had the smallest RMSE, AIC
and BIC (1.37596E-14, -832.6836 and -823.9054 respectively) and hence is the best
candidate model. The Chapman-Richards, Logistic and Korf models also performed
well based on the evaluation criterion. Weibull function with 3-parameters has
been confirmed to provide a secure estimate of total tree height for Tectona grandis
in the rainforest site of the study area of Southern Nigeria.
Keywords: Comparison, fit statistics, Weibull model, parameter-estimates.
INTRODUCTION
The relationship between tree height and diameter at breast height (DBH) is an important
element of forest structure that has long been used to characterize important aspects of tree
growth, stand development and that helps define relationships among individuals. Height
diameter models are used to calculate measurable quantities needed to characterize forest
growth and yield [1] especially volume, weight and biomass. For inventory purposes, it is
common practice to measure DBH on all trees in a plot while limiting the measurement of height
to a subsample of trees because measurements of height can be more difficult and time
consuming [2]. The relationship between tree DBH and total height is a structural characteristic
of a tree that describes key elements of stem form, and thus the volume of the harvestable stem.
[3]
Page 2 of 9
21
Egonmwan, Y. I., & Orukpe, W. O. (2024). Comparison of Non-Linear Height-Diameter Models of Tectona grandis L.f. in a Rainforest Site in Southern
Nigeria. Discoveries in Agriculture and Food Sciences, 12(5). 20-28.
URL: http://dx.doi.org/10.14738/dafs.125.15549
The height-diameter relationship describes the correlation between height and diameter of the
tree in a stand on a given site and this can be represented by a linear or non-linear mathematical
model. Developing forest inventory estimates often involves predicting tree volumes from only
diameter at breast height (dbh.) and/or merchantable height. [3]. Tree height and diameter at
breast height (DBH; that is, diameter outside of the bark of tree taken at 1.3m above ground
level) are fundamental tree characteristics used in forest measurements in order to come up
with estimates for timber volume, site index and other important variables related to forest
growth and yield, succession and carbon budget models [4]
DBH can be determined easily and accurately at little cost and time using such instruments as:
diameter tapes, callipers, Biltmore stick and bark gauge. “Information on tree heights is
essential in forest inventories for computing tree volumes. However, compared to measuring
tree diameter, field measurement of tree height is rather tedious. This is why many forest
inventories save time and effort by predicting tree heights using height–diameter (H–D) models
instead of direct measurements. To improve prediction of the local H–D curve, height
measurements from a subsample of trees on each sample plot or sample plot cluster may be
utilized. Predicting total tree-height based on observed diameter at breast height outside bark
is routinely in practical management and silvicultural research work” [5]
The Tectona grandis L.f. is a fast-growing tropical hardwood tree species belonging to the family
of Verbenaceae. Teak is one of the most widely cultivated exotic species in Nigeria because of
its good anatomical and physical properties [6]. It is also multipurpose tree species and as such,
there is continuous demand for its products [6]. The teak plantation in the study area is being
managed for research purpose. There are dearth of information on non-linear H-D models for
teak plantation in Nigeria which is important for the management of the plantation. The main
aim of the study therefore was to compare five non-linear height-diameter models for height
prediction and prescribing silvicultural option for the effective management of the teak
plantation stand.
METHODOLOGY
Study Area
The study area is located in the University of Benin which was formerly called the institute of
technology. It was founded on Saturday 23rd of November 1970. It occupies an area of about
13000 hectares. The study area is located between latitudes 6o24’08 N and longitudes 5o37’38
E. It is under the Ovia North East Local Government Area, Edo State, Nigeria. The climate
condition around the area varies all year round. In the wet season, it is warm, oppressive and
overcast and the dry season is hot and muggy. Over the course of the year, the temperature
typically varies from 67oF to 88oF and is rarely below 60oF or above 91oF. It has a tropical
climate. During most months of the year, there is significant rainfall. The average annual
temperature is 25.7oC. About 2679mm/ 105.5 inch of precipitation falls annually.
Data Collection
A total of 30 sample plots of 0.04ha (20m x 20m) with an area of 0.64ha were chosen for this
study. Ten sample plots were purposively selected from the total sample plots. Within each plot,
the following measurements were taken; diameter at breast height (dbh), diameter at top (dt),
Page 3 of 9
22
Discoveries in Agriculture and Food Sciences (DAFS) Vol 12, Issue 5, October- 2024
Services for Science and Education – United Kingdom
midpoint diameter (dm), total height (tht), merchantable height (mht) were taken with spiegel
relaskop. Observed volume by Newton and basal area were computed.
Data Cleaning
Data cleaning is the process of detecting and correcting inaccurate records from a data set. It
involves fixing or removing outliers, incorrectly formatted, duplicate or incomplete data within
a dataset. The duplicated data were cleaned and incorrect computations were adjusted. After
inputting the data into the system, each datum was crosschecked to ensure that there no wrong
data was inputted and also to ensure that there were no duplicate data.
The Applied Nonlinear Models
From past published studies on H–D modeling (Table 1), five (5) sets of 3- parameters nonlinear
models were chosen for this study. In other to avoid problems with back transformation bias,
only models that expressed tree height without transformations were used. The models were
parameterized so that parameter b0 defines the scale of the H–D models, which is the major
dimension of variability among sample plots. To generate parameter estimates, the datasets
were handled with R-script using package ‘nlme’ of R-environment [7].
Table 1: The five non-linear 3-parameters height-diameter models chosen are;
Model Form Reference Eq.
Chapman Richards HT = 1.3 + bo(1 − e
−b1.DBH)
b2 Chapman 1961, Richards 1959 [1]
Weibull HT = 1.3 + bo (1 − e
−b1.DBHb2
)
Yang et al. 1978 [2]
Schnute; HT = [1.3
b1 + (b2
b1 − 1.3
b1
)
1−e
−bo(DBH−DBH0)
1−e
−bo(DBH2−DBH0)
]
1
b1
Schnute 1981 [3]
Logistic HT = 1.3 +
bo
(1+b1
−1
.DBH−b2)
Hung et al., 1992 [4]
Korf/Lundqvist; HT = 1.3 + bo. e
(−b1.DBH−b2
) Stage 1963, Zeide 1989 [5]
bo, b1, b2 = model parameters to be estimated, e = base of the natural logarithm, HT = total tree height (m); DBH =
diameter at breast height (cm)
Evaluation Statistics
The models used in this were adjudged base on four test statistics; Root Mean Square Error
(RMSE), Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). The best
model was adjudged on the criterion that the smaller the values of the evaluation statistics, the
better the fitting model. The model evaluation criterion is indicated below:
R
2 = 1 −
SSres
SStot
[6]
RMSE = √
∑ (Yi−Ŷ
i
n )2
i=1
n
[7]
AIC = n ln (
RSS
n
) + 2p [8]
Page 4 of 9
23
Egonmwan, Y. I., & Orukpe, W. O. (2024). Comparison of Non-Linear Height-Diameter Models of Tectona grandis L.f. in a Rainforest Site in Southern
Nigeria. Discoveries in Agriculture and Food Sciences, 12(5). 20-28.
URL: http://dx.doi.org/10.14738/dafs.125.15549
BIC = n ln(
RSS
n
) + p ln n [9]
Map of the Study Area
Plate 1: Satellite image of the study area.
RESULTS
The estimated parameters of the five fitted H-D models and their corresponding evaluation test
statistics criterion are summarized in table 2 below. Models with the lowest RMSE, AIC and BIC
and highest R2 values are judged the best. Schnute model had the highest R2 value of 99.9% but
it also had the highest RMSE, AIC and BIC values (see table 2), hence performed poorly in
predicting height in the study area. Weibull model although did not have the largest R2 of 97.2%
when compared to schnute, it had the smallest RMSE, AIC and BIC (1. 37596E-14, -832.6836
Page 5 of 9
24
Discoveries in Agriculture and Food Sciences (DAFS) Vol 12, Issue 5, October- 2024
Services for Science and Education – United Kingdom
and -823.9054 respectively) and hence is the best fit model. The Chapman Richards, Logistic
and Korf models also performed well based on the evaluation criterion.
Table 2: Parameters and fit indices of the height prediction models for the teak data
Parameters Fit statistics
H-D Models b0 b1 b2 R
2 RMSE AIC BIC
Chapman- Richards
27.846 0.636416 18311.12 0.911 3.149 35.894 29.894
Korf 30.4928 0.063406 9.066469 0.9219 5.3368 49.6369 45.4151
Logistic 0.0000004 0.075 0.635506 0.9953 16.2201 78.6031 74.3813
Weibull 27.846 0.0000003 0.612322 0.9719 1. 37596E-14 -832.6836 -823.9054
Schnute 27.846 0 1.0635 0.999 2.23593E+15 15.3494 922.7465
bo, b1, b2 – parameter estimates, R2 – coefficient of determination, RMSE – root mean square error, AIC – Akaike
information criteria, BIC- Bayesian information criterion.
Fig.1: Scatter plot of total height (HT) against diameter at breast height (DBH) for Tectona
grandis
Figure 2: Scatter plot and fitted height trajectories of the five H-D models.
Page 6 of 9
25
Egonmwan, Y. I., & Orukpe, W. O. (2024). Comparison of Non-Linear Height-Diameter Models of Tectona grandis L.f. in a Rainforest Site in Southern
Nigeria. Discoveries in Agriculture and Food Sciences, 12(5). 20-28.
URL: http://dx.doi.org/10.14738/dafs.125.15549
Figure 3: Residual plot of the five H-D models.
Table 3: Summary statistics of the data
STATISTICS DBH (cm) THT(m) BA(m2) VOL.(m3)
Mean 44.17 33.10 0.17 4.34
Standard Error 1.38 0.52 0.01 0.26
Standard Deviation 13.78 5.20 0.10 2.60
Kurtosis -0.57 4.32 0.02 0.61
Skewness 0.38 -1.36 0.91 0.83
Minimum 15 8 0.0177 0.136
Maximum 75.5 40.5 0.4478 13.609
DISCUSSION
Summary of the growth variables are listed in table 3. The minimum and maximum diameter at
breast height was 15 cm and 75 cm respectively. The minimum height was 8m while the
maximum height was 40.5m. The standard error for the various parameters ranged from 0.01
to 1.69. The result revealed a strong relationship between the growth parameters for the
pooled data.
The results of goodness for predicted height are shown in table 2 along with the performance
criteria (R2, RMSE, AIC and BIC) for the five models. Models with the lowest RMSE, BIC and AIC
values closest to unity are known to perform best (Ahmadi et al. 2013). The R2 values as
indicated in table 2 showed that Logistic, Weibull and Schnute models had the highest R2 values
of 99.5%, 97.2% and 99.9% respectively, hence they were selected as the best models for the
H-D prediction which also agrees with the findings of [8]. This implies that 99.5%, 97.2% and
99.9% variation in height were explained by the diameter. Also, the curve was best captured by
the Chapman-Richards, Logistics and Schnute fit (figure 3). The predicted value follows the
same nonlinear pattern and fall close to the line of best fit without much outliers.
The Weibull model however, did not have the largest R2 but had the smallest RMSE, AIC and
BIC (1.37596E-14, -832.6836 and -823.9054 respectively) and hence was chosen as the best fit
model. Although Weibull model did not produce the highest R2 value, it was still chosen as the
best model because according to [9], R2 is a poor tool for model selection as it almost always
Page 7 of 9
26
Discoveries in Agriculture and Food Sciences (DAFS) Vol 12, Issue 5, October- 2024
Services for Science and Education – United Kingdom
favors the most complex models. Also [10] stated that R2 cannot be used to select the best model
because it increases as predictors are added i.e. overfitting is guaranteed. Hence, R2 cannot be
used as a sole indicator for determining the best model. The Chapman Richards and the Korf
models were also well fitted to the tree H-D data.
The Logistics model yielded the maximum RMSE which implies that the estimator is not as
efficient compared to other growth curves used for the study, this is in sharp contrast in the
work by [8], who compared ten nonlinear H-D models and Logistic model out-performed all the
other models used in the study with the lowest RMSE. According to [11] determination of fixed
and random effects parameters in a model is a flexible decision subject to debate. While [12]
were of the opinion that all parameters in a model should first be considered mixed if
convergence is possible. However, [13] suggest those parameters with high variability and less
overlap in confidence intervals obtained by fitting at each plot should be considered mixed if
the convergence is not achieved when considering all the parameters. Hence the results of this
study suggest that the ability of DBH in determining height is strong enough as observed among
Chapman-Richards, Logistics and Schnute model’s ability for predictive purposes.
Weibull was the most suitable model for the teak plantation in this study, agreeing with the
work of [14]. The fitness of Weibull model in predicting tree height diameter distribution is well
documented. For example, [4] found the Weibull model to be adequate in predicting tree height
diameter data of nine major tree species in Ontario’s boreal forest. However, some researchers
such as [15] and [16] in their study had Weibull model as the least fit due to its inconsistency
in describing the tree height diameter relationship of trees. The analysis of the plot of residual
and predicted values demonstrated that there was little or no systematic bias towards over-or
underestimation of the tree total height (Fig. 3). Plotting the residuals also showed that the
models were randomly distributed and had heterogeneous residuals. The five H-D models
produced tree height trajectories that are consistent with biological realism: monotonic
increment, inflection point and asymptote (Fig. 2). A number of researchers found that adding
stand variables to the height-diameter equations and using the generalized height-diameter
models increase the precision [17]; [18]. Although Weibull model out-performed the others H- D models in this study, Chapman-Richards, Schnute and Logistics models explained over 90%
of the total variance in their corresponding response variables for the data sets. Hence, five
models used in this study have shown to be very flexible and have been used extensively in
growth and yield studies for describing height-age, diameter-age, and volume-age
relationships” [19]
CONCLUSIONS
From this study, we found out that the Weibull model was best suitable for the teak plantation
in the study site. When compared to previous studies, this H-D model was found to be more
accurate and applicable for predicting height-diameter relationship. However, height-diameter
relationship varies within a region, depending on local environmental conditions [20], [21]. The
models used in this study do not take into account the effects of climatic and ecological factors
on height-diameter relationships within different ecological sites. Hence further development
of ecoregion-based individual tree height-diameter models is critical for accurate models on
which to base forest management decisions [20], [21].
Page 8 of 9
27
Egonmwan, Y. I., & Orukpe, W. O. (2024). Comparison of Non-Linear Height-Diameter Models of Tectona grandis L.f. in a Rainforest Site in Southern
Nigeria. Discoveries in Agriculture and Food Sciences, 12(5). 20-28.
URL: http://dx.doi.org/10.14738/dafs.125.15549
Competing Interest
The authors have declared that no competing interests exist.
Financial Disclosure
The authors declared that this study has received no financial support.
References
[1]. Ahmadi K., Alavi S.J., Kouchaksaraei M.T., Aertsen W. (2013): Non-linear height-diameter models for oriental
beech (Fagus orientalis Lipsky) in the Hyrcanian forests, Iran. Biotechnology, Agronomy, Society and
Environment, 17: 431–440.
[2]. Egonmwan I.Y., Ogana F.N. Application of diameter distribution model for volume estimation in Tectona grandis
Lf stands in the Oluwa Forest Reserve, Nigeria. Tropical Plant Research. 2020;7(3):573-80.
[3]. Fang, Z, Bailey R, L. (2001). Nonlinear mixed effects modelling for slash pine dominant height growth following
intensive silvicultural treatments. For. Sci. 47:287–300.
[4]. Gregoire T. G., Schabenberger O., Barrett. J. P. (1995). Linear modelling of irregularly spaced, unbalanced,
longitudinal data from permanent plot measurements. Can. J. For. Res. 25:137.
[5]. Huang, S., Titus, S., J and Wiens, D. P. (1992). Comparison of nonlinear height-diameter functions for major
Alberta tree species. Canadian Journal of Forest Research 22: 1297 - 1304.
[6]. Huang, S., Price, D., and Titus, S.J. 2000. Development of ecoregion-based height diameter models for white
spruce in boreal forests. For. Ecol. Manage. 129: 125–141. doi:10.1016/S0378-1127(99)00151-6.
[7]. Krisnawati, H., Wang Y., and Lui L. (2010). Generalized height-diameter models for Acacia mangium Wild.
Plantations in Sumatra. Journal of Forestry Research 7 (1):1 - 19.
[8]. Meyer HA. A mathematical expression for height curves. Journal of Forestry. 1940; 38:415–420.
[9]. Miranda I, Sousa V and Pereira H (2011) Wood properties of teak (Tectona grandisL.f.) from a mature
unmanaged stand in East Timor. Journal of Wood Science 57: 171–178.
[10]. Mkt updated July 1, 2019: The best model selected with AIC have lower R2 than the full/ global model.
[11]. Nugroho, B. (2014). Menuju KPH mandiri, apa yang harus dilakukan? In Sugiharto, Strategi pengembangan KPH
dan perubahan struktur kehutanan di Indonesia. Jakarta: Direktorat Jenderal Planologi Kehutanan Kementerian
Lingkungan Hidup dan Kehutanan
[12]. Osman E. l., Mamoun H., Idris El Zein A., Ibrahim E. l and Mugira, M. (2013). Height- Diameter Prediction
Models for Some Utilitarian Natural Tree species. Journal of Forest Products and Industries, 2013, 2(2), 31-39
ISSN 2325-453X.
[13]. Peng, C., Zhang, L., and Liu, J. (2001). Developing and validating nonlinear height-diameter models for major
tree species of Ontario’s boreal forests. Northern Journal of Applied Forestry 18:87 - 94.
[14]. Pinheiro J. C, Bates, D. M. (1998). Model building for nonlinear mixed effects model. Madison (WI): University of
Wisconsin.
Page 9 of 9
28
Discoveries in Agriculture and Food Sciences (DAFS) Vol 12, Issue 5, October- 2024
Services for Science and Education – United Kingdom
[15]. R Core Team (2017) R: A language and environment for statistical computing. R Foundation for Statistical
Computing, Vienna, Austria. Available from: http://www.R-project.org/. (accessed: 30 Jun. 2017).
[16]. Russel Pierce (2014): The best model selected with AICc have lower R2 than the full/ global model.
[17]. Scaranello, M. A da Silva, Luciana F.A., Simone A.V., Plinio B de Camargo, C.A J. and Luiz, A.M. (2012). Height- diameter relationship of Tropical Atlantic Moist Forest trees in south eastern Brasil Sci. Agric. 69 (1):26-37.
[18]. Somers GL, Farrar RM. Bio-mathematical growth equations for natural longleaf pine stands. Forest Science.
1991; 37:227–244.
[19]. Temesgen G, Amare B and Abraham Mahari (2014). Population dynamics and land use/land cover changes in
Dera District, Ethiopia. Global Journal of Biology, Agriculture and Health sciences. 3(1):137-140.
[20]. Tewari V.P., Álvarez-González J.G., García O. (2014): Developing a dynamic growth model for teak plantations in
India. Forest Ecosystems, 1: 1–9.
[21]. Y. I. Egonmwan (2022): Height-Diameter Models for Prediction of Teak Stand in Western Nigeria Asian Journal
of Research in Agriculture and Forestry AJRAF, 8(4): 293-300, 2022; Article no. AJRAF.93599.