Content uploaded by Ali Jahani
Author content
All content in this area was uploaded by Ali Jahani on Apr 21, 2015
Content may be subject to copyright.
This article was downloaded by: [Ali Jahani]
On: 06 April 2015, At: 05:16
Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Journal of Environmental Planning and
Management
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/cjep20
Optimized forest degradation model
(OFDM): an environmental decision
support system for environmental
impact assessment using an artificial
neural network
Ali Jahania, Jahangir Feghhib, Majid F. Makhdoumb & Mahmoud
Omidc
a Environment and Natural Resources Sciences Department,
University of Environment, Karaj, Iran
b Department of Forestry and Forest Economic, Faculty of
Natural Resources, University College of Agriculture and Natural
Resources, University of Tehran, Karaj, Iran
c Department of Agricultural Machinery Engineering, Faculty of
Agricultural Engineering and Technology, University of Tehran,
Karaj, Iran
Published online: 11 Mar 2015.
To cite this article: Ali Jahani, Jahangir Feghhi, Majid F. Makhdoum & Mahmoud Omid (2015):
Optimized forest degradation model (OFDM): an environmental decision support system for
environmental impact assessment using an artificial neural network, Journal of Environmental
Planning and Management, DOI: 10.1080/09640568.2015.1005732
To link to this article: http://dx.doi.org/10.1080/09640568.2015.1005732
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Ali Jahani] at 05:16 06 April 2015
Optimized forest degradation model (OFDM): an environmental
decision support system for environmental impact assessment using an
artificial neural network
Ali Jahani
a
*, Jahangir Feghhi
b
, Majid F. Makhdoum
b
and Mahmoud Omid
c
a
Environment and Natural Resources Sciences Department, University of Environment, Karaj,
Iran;
b
Department of Forestry and Forest Economic, Faculty of Natural Resources, University
College of Agriculture and Natural Resources, University of Tehran, Karaj, Iran;
c
Department of
Agricultural Machinery Engineering, Faculty of Agricultural Engineering and Technology,
University of Tehran, Karaj, Iran
(Received 1 May 2014; final version received 6 January 2015)
The purpose of this article is Artificial Neural Network (ANN) modeling using
ecological and associated factors with forest degradation to predict the degradation of
ecosystem, thereby enabling us to assess the environmental impacts of forest projects
as an Environmental Decision Support System (EDSS). Results of the Multi-Layer
Feed-Forward Network (MLFN), trained for Optimized Forest Degradation Model
(OFDM), indicate that the performance of OFDM is more than other degradation
models. Changes in forest management activities with higher value in sensitivity
analysis help forest managers to decrease OFDM entity and environment impacts. The
system is an intelligent EDSS, which allows the decision-maker to model criteria in
forest degradation in order to reach and employ the optimal allocation plan.
Considering results, multi criteria decision analysis (MCDA) approaches based on
ANN, is an encouraging and robust method for solving MCDA problems.
Keywords: EIA; EDSS; ANN; OFDM; MCDA
1. Introduction
Environmental Impact Assessment (EIA) has been recognized as a basic tool for
environmental management and sustainable development (Canter 1995; McDonald and
Brown 1995;George1999;Eccleston2000). But when quantitative entities are needed for
decision-making or where the environmental cumulative effects are concerned, the
uncertainty and inconsistency between the decision-makers in determining development
programs and policy making in impact assessment of development alternatives elucidate
(Burris and Canter 1997; MacDonald 2000; Leknes 2001). EIA, as a substantial instrument
for environmental management and sustainable development, has long been recognized
(Canter 1995; McDonald and Brown 1995;George1999;Eccleston2000), but it is
hampered when quantitative measures for decision-making are needed, and where
cumulative effects of environmental impact are concerned (Burris and Canter 1997;
MacDonald 2000; Leknes 2001). Activities, which are implemented in the most forest
projects, often contain inappropriate environmental impacts (Sepp€
al€
aet al. 1998;Gumus
Acar and Toksoy 2008; Michelsen, Solli and Strømman 2008;Hanna,P
€
ol€
onen and Raitio
2011;Koskela2011). Environmental impact assessment in forest projects identifies
*Corresponding author. Email: ajahani@ut.ac.ir
Ó2015 University of Newcastle upon Tyne
Journal of Environmental Planning and Management, 2015
http://dx.doi.org/10.1080/09640568.2015.1005732
Downloaded by [Ali Jahani] at 05:16 06 April 2015
environmental impacts of forest project activities and whether these activities contain serious
environmental consequences? Although EIA of forest projects is the integral part of projects,
it has not been implemented widely in developing countries (Knowler and Lovett 1996;
Hanna, P€
ol€
onen and Raitio 2011) such as Iran. The essential changes in decision-making
methods caused environment managers to introduce Environmental Decision Support
System (EDSS) as a new tool which often consists of various combined environmental
models, databases, and assessment tools for the comparative evaluation and alternatives
selection (Znidarsic, Bohanec and Zupan 2006). As a first attempt, the concept of DSS was
defined by Gorry and Scott Morton (1971) using Simon’s (1960) research on organizational
decision-making (Oliver and Twery 1999; McIntosh et al. 2005). After a while, Oliver and
Twery (1999) defined the experimental and practical circumstance of EDSS application in
forest management. A glance at the research reveals many interesting studies in EDSS
design in the real world (Cort
es et al. 2000; Twery, Peter and Scott 2005; Argent et al. 2009;
Elmahdi and McFarlane 2009), but these are less than the number of articles focused on
problems and methods to solve environmental issues (Segura, Ray and Maroto 2014).
Considering the shortcoming of EIA in the quantity impact evaluation (Rothman
2000), Artificial Neural Networks (ANNs) are more practical. ANNs have their
origins in the study of the complex behavior of the human brain. Historically,
McCulloch and Pitts (1953) introduced simple models with binary neurons. Then,
Rosenblatt (1958) proposed the multilayer structure with a learning mechanism
based on the work of Hebb (1949), the so-called perceptron, and the first neural
networks applications began with Widrow (1959). ANNs have been applied in many
studies (Maier and Dandy 2000;Maieret al. 2010) as an EIA tool in environmental
management (Tayebi, Tangestani and Roosta 2010; Yijun et al. 2010;Valiet al.
2012). The ANN application, in exploring the relationship between components of
the ecosystem, quantifies them and their role in recognizing the degradation of
ecosystem in the development process, reduces risk-taking in decision-making, and
it is recognized as one of the EDSSs in EIA in forest management.
The analysis of the state of the art in the domain of the ANN application in the field of
environmental decision problems reveals several preliminary attempts towards a wider
use of ANNs in water resources management (Iliadis and Maris 2007; Fernandez et al.
2009; Arsene, Gabrys and Al-Dabass 2012), forest management (Mas et al. 2004; Elmas
and Sonmez 2011), energy management (Asadi et al. 2014; Yeo and Yee 2014), and
landslide susceptibility analysis (Pradhan and Lee 2010). Considering research on ANN
application in EDSS design, several kinds of ecological data, as inputs and outputs, were
used to design ANN models with various structures. The studies present EDSS models to
assist decision-makers in environmental management of natural resources and indicate
that ANN results are better than the earlier methods.
The purpose of this article is ANN modeling for temperate broadleaf and mixed
forests using ecological and associated factors with forest degradation to predict the
degradation of temperate broadleaf forest ecosystems, thereby enabling to assess the
environmental impacts of forest project as an EDSS.
2. Materials and methods
2.1. Study area
The Hyrcanian temperate broadleaf forests are located in the northern hillside of the
Alborz Mountain of Iran. The Hyrcanian forests are one of the remnant virgin forests of
temperate regions in the world (remained of Tertiary). Pure and mixed beech forests are
2A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
one of the richest forests, generally lying on the northern slopes of the Alborz Mountains
adjacent to the Caspian Sea (Saghebtalebi, Sajedi, and Yazdian 2003). The study area is
3000 hectare of the Hyrcanian temperate broadleaf forests, 7 km southeast of Nowshahr
city, in the north of Iran (363003000 to 363203000 N latitude and 514000000 to 514103000 E
longitude). The main anthropogenic influence is animal breeding in eight animal
husbandries. The study area has been influenced by grazing for 30 years. Natives of the
study area graze livestock in the forest as a traditional activity. Three districts of the
study area have been managed by Watershed, Rangelands and Forests Organization of
Iran just for harvesting objective for 30, 20 and 10 years, respectively. Managers
harvested trees and logs to achieve the expected income from the forest project while the
harvesting cannot provide all required income. The varieties of forest project activities
have been implemented in three districts of the study area to harvest logs, such as road
construction, logging, skidding, logs depot construction, and so on. As a result, the
pressure of forest project activities caused degradation of the ecosystem. Hence, today
managers are looking for new methods in EIA for forest project to find the balance
between income and land capability protection.
2.2. Methods
2.2.1. Mapping
The susceptibility (vulnerability or ecologically inability to resist a particular external
influence) of the landscape depends upon the ecological characteristics of the area and the
type of ecosystems encountered (Makhdoum 2002; Potter, Cubbage and Schaberg 2005)so,
in this study, Land Management Units (LMUs) were formed in the region considering
ecological characteristics of land. LMUs were mapped out based on Ian McHarg’s overlay
technique (McHarg 1969) by ARC GIS 9.3 software. The maps of ecological factors were
overlaid and the new map with boundaries of the specific ecological factor classes (classes of
altitude, slope, aspect, vegetation type, depth and drainage of soil, erosion vulnerability of
soil, geology type, the mean of annual temperature, and the mean of annual precipitation)
was produced. It means that ecological factor classes of an LMU differ from ecological
factor classes of adjacent LMUs (at least in one ecological factor class). Ecological factors,
which influence vulnerability of ecosystem, have been identified by the scientific literature
review (i.e., Makhdoum 2002; Potter, Cubbage and Schaberg 2005). The study area was
divided into 130 LMUs with various ecological conditions (i.e., various susceptibility).
2.2.2. Degradation models
Previously, Makhdoum (2002) used Equation (1):
HDXICDP=V;(1)
where Hrepresents the degradation coefficient of the habitat, Iis cumulative past to
present impact, DP indicates the physiographic density of the population, and V
represents the habitat susceptibility. This model successfully explained the shortage of
data (Azari Dehkordi and Nakagoshi 2003).
Makhdoum (2002) defines DP as a pressure factor on the environment. Considering
forest ecosystem, without any inmate population, we replaced DP with livestock density
(LD) as a pressure factor on the environment of the forest ecosystem. The degradation
model was calculated by taking two factors which are population density (DP) and (LD)
Journal of Environmental Planning and Management 3
Downloaded by [Ali Jahani] at 05:16 06 April 2015
(Equations (1) and (2)).
Hl DXICLD=V;(2)
where Hl represents the degradation coefficient of the habitat, Iis cumulative past to
present impact, LD indicates the physiographic density of the livestock, and Vrepresents
the habitat susceptibility.
To optimize the degradation model for forest ecosystems, indicators of forest
degradation were determined and evaluated. The rate of erosion in forest ecosystems is
directly associated with the forest project activities. Therefore, it was calculated by two
factors (Zand E) which are evaluated by a specific method. The rate of erosion index, Z,
is one of the EPM method’s indices which are applied to calculate the volume of
sediment production Equation (3) (Gavrilovic 1998).
ZDXa:YðfCI0:5Þ;(3)
where Zrepresents the rate of erosion index, X
a
is land use index, Yindicates the erosion
of soil susceptibility, frepresents erosion index, and Irepresents the mean slope in
percent.
According to Gavrilovic (1998), fequals 0.1 and Yequals 0.8 for LMUs in forest
ecosystems. Considering different land uses in LMUs of forest, X
a
equals 0.1 for LMUs
which have a concentration of forest project activities for harvesting, and 0.2 for the rest
of LMUs with local uses.
The results of Zindex are categorized in five classes of the rate of erosion (Table 1).
Considering the forest ecosystem of LMUs, the rate of surface litter erosion (E) as the
most common erosion was evaluated in LMUs based on Table 2.
2.2.3. Artificial neural network model
ANN is now generally accepted as a main tool in various fields of science and engineering
in the development of intelligent systems. ANN has been recently developed for data
mining, pattern recognition, quality control, and has gained wide popularity in modeling
of many processes in environmental sciences and engineering (Jambunathan et al. 1996;
Sreekanth, Ramaswamy and Sablani 1998; Paliwal, Visen and Jayas 2001; Hussain,
Safiur and Rahman 2002; Sablani, Baik and Marcotte 2002; Boillereaux, Cadet and Le
Bail 2003). ANN learns by examples and it can combine a large number of variables
(Haykin 1999). An ANN is a computing tool which consists of many interconnected
Table 1. The classes of Zindex.
Zclass Zindex The rate of erosion
1>1 Very intensive
210.71 Intensive
3 0.70.41 Moderate
4 0.40.2 Weak
5<0.19 Very weak
4A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
simple processing elements (PEs), and they operate in a parallel way. Each processor
(neuron) is only aware of signals that it periodically receives and sends to other
processors (Callan 1999). In this study, an ANN is considered as a computer program
capable of learning from samples, without requiring a prior knowledge of the
relationships between parameters (Callan 1999). ANN learns to solve problem by
adequately adjusting the strength of its interconnections (weights), considering the input
data (Nasr et al. 2012). It can also adapt easily to new environments by learning more and
it can deal with complicated, vague, or probabilistic data (Leondes 1998). The concept of
ANN was inspired from biological nervous systems (Picton 2000). In fact, neural
networks are an attempt to build systems that operate in a similar way to the human brain.
The human brain contains tens of billions of neurons densely interconnected. The
function of an ANN is to create an output pattern using received inputs. Output signals
sent to other units along connections (known as weights), which excite or inhibit the
signal that is being connected. Learning is the process of adapting the connection weights
in response to stimulus being presented at the inputs and optionally to the output buffer.
A stimulus exhibited at the output buffer corresponds to a desired response to a specific
input. In this case the learning is called “supervised learning.” ANN consists of units that
have a limited computing capability, and the complete network is capable of solving very
complicated problems, when many of units are linked together. Generally, the
architecture of an ANN contains the input layer, the hidden layer and the output layer
(Figure 1). The data are being processed in the hidden layer and it may have one or more
sub-layer(s) based on the designer’s view.
The reasons for the interest in ANN application for data analysis are its successful
application in classification, function estimation and optimization and pattern
recognition, in addition to its wonderful ability for parallel calculation and generalization
which are specific function of ANN in data processing. Back Propagation (BP) technique
is considered as one of the most popular algorithms of ANN so it is a dominant method
for node weight arrangements (Elmas et al. 1994). For this reason, BP network structure
has been used for determination of environmental danger rating.
A BP network is made up of a minimum of three layers involving a series of
propagation errors (PEs), a specific transfer function for each PE in layers, an input, a
hidden and an output layer, weights of the layers and a specific learning rule
(Figure 1) (Nasr and Zahran 2014). At the beginning of data fusion, random low
values are allocated to the weights that are between the layers. This situation allows
Table 2. The characteristics of the rate of surface litter erosion (E).
Characteristics Rate of surface litter erosion
Very high surface litter erosion (the replacement of litter and plant
residues on the earth surface is detectable widely)
1
High surface litter (the replacement of litter and plant residues on the
earth surface is detectable clearly)
2
Moderate surface litter erosion (the replacement of litter and plant
residues on the earth surface is detectable)
3
Low surface litter erosion (the replacement of litter and plant residues
on the earth surface is detectable barely)
4
Very low surface litter erosion (the replacement of litter and plant
residues on the earth surface is not detectable)
5
Journal of Environmental Planning and Management 5
Downloaded by [Ali Jahani] at 05:16 06 April 2015
an ANN network to be taught more easily (Elmas et al. 1994). The learning process is
known as a generalized delta rule. In this process, input signals are presented to the
network in a specific order. For each input sample, the network feed-forward and the
output are calculated (multi-layer feed-forward network, MLFN). An error value is
calculated for each output, considering comparison between calculated output and
expected value. Through the hidden layers, error signals propagate from the output
layers backward toward the input layers. The weight adjustments of related layers are
carried out during this process. This process is repeated until the minimum total error
value is obtained and then the learning is ended. The numbers of PEs in input and
output layers are approximately the same as the number of inputs and outputs of the
model. The number of PEs in the hidden layers is up to the user. In MLFN, an input
set comes forward from the input layer to the hidden layer. Each PE on the hidden
layer sums its inputs and it transfers to the next layer by passing through the
activation function. The final hidden layer transfers the activation value into the
output layer and hereby the output values are acquired. In this section of the training,
Xdefines the input vector (ecological and associated factors with forest degradation)
and Ydefines the desired output vector (degradation models and indicators of forest
degradation). The values of Xand Yin a BP network with ninputs and moutputs are
presented in Equation (4) (Elmas and Sonmez 2011):
XDfðX1;X2;X3;...;XnÞYDfðC1;C2;C3;...;CmÞ:(4)
When the real output of the network is defined with Y
net
, the purpose of BP is to
minimize the error value between Yand Y
net
.WhenXand Yare given to the network, an
output of jth PE on kth layer (PEk
j
) is calculated in Equation (5) (Elmas and Sonmez 2011):
netk
jDX
n
iD0
WjiXji :(5)
This output of PE is calculated with a specific function of net
j
which is presented in
Equation (6). This function is known as a transfer or threshold function. In this study, we
Figure 1. The structure of designed MLFN.
6A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
have used the sigmoid tangent transfer function, presented in Equation (7).
Ynet DfkðnetjÞ;(6)
fkðYnetjÞD1
1Cexpð¡YnetjCTÞ:(7)
In Equation (7), Tis the threshold value and Y
net
is the sum of weighted inputs
for PEk
j
. The function presented in Equation (7) is used only by the hidden layers.
The input and output layers use the linear activation function. After the end of the
feed-forward process on the output layer, the BP process starts. In the BP process,
the weights are permanently in change while the difference between the real and
desired values are minimizing. According to the generalized delta rule, if tnumbers
of input/output parts (xand y) are used, then the weights would be changed as in
Equation (8) (Elmas and Sonmez 2011).
Wt
ji DWðt¡1Þ
ji CDWt
ji:(8)
In the MLFNs, error minimization can be obtained by a number of procedures
including gradient descent (GD), LevenbergMarquardt (LM), and conjugate gradient
PolakRibiere (CGP). MLFNs are normally trained with an error BP algorithm. BP
uses a GD technique that is very stable when a small learning rate is used but has
slow convergence properties. Several methods for speeding up BP have been used,
including adding a momentum term or using a variable learning rate. In this paper,
CGP algorithm which is one of the fastest batch training algorithms with more
calculations in any repetition and faster converging was used (Omid, Baharlooei and
Ahmadi 2009). The next important step is known as the validation step which is
performed in order to generalize ANN. The ANN is applied to a data set which
differs from the training data set. High number of epochs or large training data set
can probably cause over-fitting or memorizing. The validation step must be applied to
ensure the generalization of ANN (Hagan, Demuth and Jesus 2002). The final step is
the test step, in which a data set that is different from training and validation data sets
is applied for testing. This step is similar to the validation step except in one
difference. The validation step is the criteria in order to stop the training process. On
the other hand, test step is applied to a trained and validated ANN to measure its
performance (Hagan, Demuth and Jesus 2002). Finally, the training process must be
stopped with respect to a criterion. This criterion can be a specified error and
specified error gradient, number of epochs, or validation checks. All these parameters
are user defined and finding the optimum parameters usually depends on trial and
error (as the number of hidden layers and the number of neurons in each layer).
2.2.4. Model selection
A program was developed in MATLAB R2012a package for the feed-forward and BP
network. To objectively evaluate the performance of the network, two different statistical
indicators were used. These indicators are Mean-Squared Error (MSE), Root-Mean-
Squared Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination
Journal of Environmental Planning and Management 7
Downloaded by [Ali Jahani] at 05:16 06 April 2015
(R
2
) (Zangeneh, Omid and Akram 2010):
MSE DXn
iD1ðyi¡^
yiÞ2
n;(9)
RMSE Dffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Xn
iD1ðyi¡^
yiÞ2
n
s;(10)
MAE DXn
iD1jyi¡^
yij
n;(11)
R2DXn
iD1ð^
yi¡yiÞ2
Xn
iD1ðyi¡yiÞ2;(12)
where yiand ^
yiare the actual and estimated values, respectively, yiis the mean of actual
values, and nis the number of observations.
The relationship between the degradation models and indicators of forest degradation,
with ecological and associated factors with forest degradation, was distinguished using
the designed ANN.
Considering ecological condition of land and associated factors with forest
degradation (refer to 2.2.1. Mapping), the study area was divided into 130 LMUs with
specific ecological condition and associated factors with forest degradation in each LMU
(130 samples for ANN modeling). In order to have an estimation of predictive ability of
the model with data that were not used in the learning process, the total collected data
were randomly divided into three subsets, which are: training data subset (60% of total
samples), validation data subset (20% of total samples), and test data subset (20% of total
samples). Based on the designed ANN accuracy (R
2
) of test data group, Optimized Forest
Degradation Model (OFDM) was introduced using combination of degradation models
with forest degradation indicators.
In computing, graphical user interface (GUI) is a type of user interface which allows
users to interact with computer software through graphical icons and visual indicators
such as secondary notation, as opposed to text-based interfaces, typed command labels,
or text navigation. As an EDSS tool, the GUI was designed in MATLAB2012a software
to allow users of OFDM to interact through a graphical icon.
In order to provide the decision-makers with a set of quantitative measures to observe
affected areas (critical and noncritical) for future development plans, the nominal figures
of all LMUs (130) were classified into categories and criteria using a fuzzy set theoretic
approach as described by Regan, Colyvan and Burgman (2000). Accordingly, nominal
figures of sets of the LMUs can have degrees of membership, by avoiding borderline
cases. Moreover, as Regan, Colyvan and Burgman (2000) have pointed out, fuzzy
boundaries are more forgiving with imprecise data; the OFDM may have resulted from
imprecise data, by which the OFDM nominal figure may not be precisely correct.
3. Results
3.1. Modeling
The forest management activities causing an impact on the environment of the LMUs
consist of: overgrazing, trails construction, tree cutting, road building, logging, skidding,
8A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
soil compaction, recreational overuse, waste pollution, tubing under roads, road damages,
and high-voltage cabling. By computing the assigned weighted values, and physiographic
population or livestock density, a nominal figure was resulted for each degradation
model. Indicators of forest degradation (Zand E) were calculated and assessed in
categories in LMUs.
Various MLFNs were designed and trained as two and three layers to find an optimal
model prediction for the degradation models and indicators. Training procedure of the
networks was as follows: different hidden layer neurons and arrangements were adapted
to select the best production results. Altogether, many configurations with different
number of hidden layers (varied between one and two), different number of neurons for
each of the hidden layers, and different inter-unit connection mechanisms were designed
and tested. For illustration purposes, in Figure 1 a network having one hidden layer and
one output layer is designed. We named this network a (30-3-3-1)-MLFN topology; that
is, the number of inputs is 30, the number of PEs in the first hidden layer is 3, the number
of PEs in the second hidden layers is 3, and the number of outputs is 1.
The simulated data (degradation models and indicators) were used to train the feed-
forward neural networks. The ecological and associated factors with degradation, a total
of 30 factors containing: altitude, slope, aspect, vegetation type, depth, drainage and
erosion vulnerability of soil, geology type, the mean of annual temperature, the mean of
annual precipitation, overgrazing, trails construction, tree cutting, road construction,
logging, skidding, soil compaction, recreational overuse, waste pollution, tubing under
roads, high-voltage cabling, reforestation, distance to road, distance to trail, distance to
livestock husbandry, distance to logs depot, density of livestock, reforested areas, seed
sown areas, and the volume of tree cutting. Considering the highest value of R
2
, the best
topology of ANN is (30-5-5-1)-MLFN for H, (30-20-20-1)-MLFN for Hl, and (30-11-11-
1)-MLFN for Zand (30-6-6-1)-MLFN for E. The sigmoid tangent transfer function which
is presented in Equation (7) is used by the hidden layers in all designed networks. The
input and output layers use the linear activation function.
Results of the MLFN trained for degradation models and indicators are presented in
Table 3.
Considering results of ANNs, Hl model with higher amount of R
2
, makes more
accurate correlation between inputs and outputs. Two degradation indicators (Zand E)
make an acceptable correlation too. Therefore, Hl model, using two indicators of
degradation, can be a more accurate degradation model for forest ecosystems. Two
indicators of degradation are known as indicators of ecosystem susceptibility (which
cause an increase in the susceptibility of the ecosystem), so their function is similar to the
function of Vin Hl model. Equation (13) is proposed as OFDM in managed forest
Table 3. Results of the designed MLFNs for degradation models and indicators of degradation.
R
2
test MAE RMSE MSE Models and indicators
0.744 4.72 0.516 0.266 H
0.783 4.12 0.501 0.251 Hl
0.752 4.55 0.507 0.257 Z
0.821 3.23 0.491 0.241 E
Note: MSE: mean square error; MAE: mean absolute error; RMSE: root mean square error; R
2
: the coefficient of
determination.
Journal of Environmental Planning and Management 9
Downloaded by [Ali Jahani] at 05:16 06 April 2015
ecosystems:
OFDM DXICLD=VCZCE;(13)
where OFDM is an optimized forest degradation model, Iis cumulative past to present
impact, LD indicates the physiographic density of the livestock, Vrepresents the habitat
susceptibility, Zis class of the rate of erosion index, and Erepresents the rate of surface
litter erosion.
Results of the MLFN trained for OFDM are presented in Table 4. Considering the
highest value of R
2
, the best topology of ANN is (30-11-11-1)-MLFN for OFDM. The
sigmoid tangent transfer function, which is presented in Equation (7), is used by the
hidden layers in network. The input and output layers use the linear activation function.
The error on the validation set is checked during the training process. The validation
error decreases during the initial phase of training, as does the training set error. When
the network starts to over fit the data, the error on the validation set typically starts to rise.
The network weights and biases are saved at the minimum of the validation set error
(Figure 2). Also the performance of the MLFN model meaning the MSE, MAE, RMSE
and the coefficient of determination (R
2
) which are calculated based on the differences
between the output of the ANN model (estimated) and the actual (calculated) value of
OFDM is presented in Table 4.Figure 3 illustrates the correlation between the ANN
model (estimated) outputs in the training, validation, test and all data based on the inputs
and actual OFDM outputs (Target).
The comparison R
2
of degradation models in Figure 4 indicates that OFDM makes the
most accurate correlation between inputs (30 ecological and associated factors with
degradation) and output (OFDM). The main ability of OFDM is in forest degradation
prediction based on susceptibility of forest and planned management activities. In EDSS
tool design, OFDM is applied for EIA decision-making in forest management activities
planning to reduce environmental impacts of forest management plan.
3.2. Sensitivity analysis of OFDM
Actually, sensitivity analysis consists of the study of how the uncertainty in the output of a
mathematical model can be apportioned to different sources of uncertainty in its inputs
(Saltelli et al. 2008). The “Analysis” infers that the scrutiny occurs a posteriori, as is the
case with sensitivity analysis. On the other hand, “Robustness” involves concerns that
must be taken into account a priori, at the time that the problem is formulated. Certainly,
the use of a sensitivity analysis to respond to such concerns is not excluded, if necessary
(Roy 2010). Ideally, robustness and sensitivity analysis should be run in tandem. Thus,
sensitivity analysis is essential in order to examine the robustness of empirical variables to
the possible existence of an unmeasured confounder (Imai, Keele and Yamamoto 2010).
Table 4. Results and structure of the designed MLFN for OFDM model.
R
2
test MAE RMSE MSE Models and indicators
0.897 5.52 0.55 0.303 OFDM
Note: MSE: mean square error; MAE: mean absolute error; RMSE: root mean square error; R
2
: the coefficient of
determination.
10 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
The sensitivity of the model was determined by examining and comparing the output
produced during the validation stage with the calculated values. According to the results in
Figure 5, the share of each input item of the developed MLFN model on desired output
can be seen clearly. Sensitivity analysis provides insight into the usefulness of individual
variables. With this kind of analysis, it is possible to judge what parameters are the most
significant (with sensitivity value close to 1) and the least significant (with sensitivity
value close to 0) during generation of the satisfactory MLFN. According to sensitivity
analysis, the degree of trails building, with a value of 0.22, is the most significant
parameter in OFDM. Less significant appear to be skidding, tree cutting, and distance to
road as 0.2, 0.17, and 0.16, respectively. Forest road networks aimed at enhancing wood
production must be implemented to meet the targets of the management plan. Trails,
skidding, and roads are directly related to the forest road network, so it is clear that the
designed road network is the main factor in forest degradation and higher OFDM entity.
Therefore, the designing of forest roads has been the main debating point in discussions
between forest managers and environmentalists (Gumus, Acar and Toksoy 2008). These
results were obtained from the wood-harvesting road network planning method.
Considering OFDM entity, environmentally positive changes in forest road network
designing, such as lower density of roads or other road directions alternatives far from the
Figure 2. The performance of validation in OFDM modeling. (See online colour version for full
interpretation.)
Journal of Environmental Planning and Management 11
Downloaded by [Ali Jahani] at 05:16 06 April 2015
Figure 3. Correlation between the ANN (estimated) outputs and actual OFDM outputs.
Figure 4. R
2
of degradation models and degradation indicators.
12 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
vulnerable LMUs of forest, lead to lower OFDM entity or forest degradation. On the other
hand, tree cutting is known as the other important factor which affects OFDM. Consider
that any changes in the wood production plan will directly affect tree cutting and OFDM
entity. Unfortunately, we are not able to change ecological factors of ecosystem, but with
manipulation of forest management activities, it is possible to change OFDM entity.
Therefore, changes in forest management activities with higher value in sensitivity
analysis help forest managers to decrease OFDM entity and environment impacts.
3.3. EDSS tool
Designed GUI (Figure 6), as a user-friendly EDSS tool, allows users to apply ANN to
predict OFDM entity in spatial units of forest (LMUs). The results of prediction are
presented in a table as shown in Figure 6. Designed EDSS tool helps forest managers to
predict OFDM entities which have resulted from forest management activity
manipulation. When users import a database into MATLAB software and run designed
EDSS tool by clicking on designed GUI icon, four steps, consisting of preprocessing,
simulation, post-processing and output calculation (Figure 7) will be implemented
systematically and OFDM entities are created in the output table. Designed EDSS tool
uses data and models, provides an easy, user-friendly interface, and can incorporate the
decision-makers’ own insights. Figure 7 illustrates schematically the structure of EDSS
in EIA of forest projects.
The classification of nominal figures of OFDM in all LMUs of forest, as computed by
the fuzzy set theoretic approach, is presented in Table 5 and Figure 8. Based on fuzzy set
theoretic approach, Figure 8 illustrates a membership degree of current forest project to
the classes of decision-making.
Considering EIA, Table 5 also demonstrates the standing of acceptable forest project,
needed rehabilitation and mitigation actions and unacceptable forest project. Considering
categories of OFDM entities, forest manager should manipulate forest management
Figure 5. The results of sensitivity analysis of MLFN model.
Journal of Environmental Planning and Management 13
Downloaded by [Ali Jahani] at 05:16 06 April 2015
activities and recalculate OFDM using EDSS tool. This way, the alternative forest project
with low and acceptable environmental impacts will be provided.
Considering result of fuzzy classification of OFDM, the map of EIA decision-making
in forest project was prepared. Table 5 indicates the overlap between the classes of
decision-making based on the membership functions. The forest planning, in most of the
study area, belongs to the two acceptable classes. Based on the sensitivity analysis, some
parameters, which are more significant, were manipulated, as an alternative project, to
reduce the OFDM outputs (Table 6).
4. Discussion
In general, the forest impact assessment methods aim to support decision-making by
ensuring that potential management options are environmentally sound. In this paper, we
presented a new tool to assess environmental impacts in the forest ecosystems. OFDM is
applied in the structure of the EDSS tool to predict the intensity of negative
environmental impacts in forest management alternative plans. Elmahdi and McFarlane
(2009) describe EDSS as an intelligent analysis and information system that pulls
together in a structured but easy-to-understand platform (i.e., DSS), the different key
aspects of the problem and system, i.e., EDSS should combine database engineering and
modeling. We propose an EDSS architecture based on four levels in EDSS design for
EIA in managed forest ecosystems. The first level of the EDSS encompasses the severity
of degradation in forest ecosystem and its associated factors identification as a challenge
Figure 6. Simulation of OFDM in forest using EDSS tool.
14 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
in EIA. The second level encompasses data gathering and classification which contains
forest ecosystem condition data, forest management activities, and the severity of
degradation of forest. This level comes to end by modeling using mathematical methods.
The results of ANN modeling and its high accuracy in degradation modeling in natural
ecosystems introduced OFDM as a comparative model of forest degradation. OFDM
indicates the strong relationship with associated factors with degradation which can be
used to compare the level of degradation of forest spatial sections and their development
potential in future. In the third level, a user-friendly tool is designed for decision-makers
to deal with a computer system. Considering this level, the EDSS tool was designed for
forest managers or other decision-makers to compare the results of planned activities on
Figure 7. Flow diagram for development of EDSS in EIA of forest project.
Table 5. Classes for decision-making.
Class Categories of OFDM entities EIA decision-making
1 0.031.16 Acceptable
2 0.032.35 Acceptable and marginally acceptable
3 1.173.5 Some rehabilitation and mitigation actions needed
4 2.364.66 Considerable rehabilitation and mitigation actions needed
5 3.515.82 Unacceptable and marginally unacceptable
6 4.675.82 Unacceptable
Journal of Environmental Planning and Management 15
Downloaded by [Ali Jahani] at 05:16 06 April 2015
forest degradation in temperate broadleaf forests. The output of EDSS tool illustrates the
predicted OFDM in LMUs using ANN. OFDM, such as other EDSSs, solves very
specific environmental problems in study area and it is hardly used in other regions.
Considering capability of ANN models in retraining, if the designed ANN model is
retrained in new LMUs of other forest regions, EDSS will be verified with new specific
accuracy (R
2
). The implementation of EDSS is in the fourth level. Considering the
susceptibility of LMUs (categories of OFDM), the rehabilitation and mitigation actions
were taken.
Figure 8. Membership degree map of studied forest to the classes of decision-making.
16 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
McCarter et al. (1998) primarily provided landscape management system (LMS) as a
decision support at the project level. LMS integration with ArcView software affords
opportunities for spatial analysis of management alternatives, although LMS does not
interpret the meaning of predicted values. In other words, LMS presents results, but does
not attempt to evaluate them, whereas the results of OFDM were classified in three
classes of decision-making by fuzzy logic. An interesting feature of LMS to planning is
that actions in the implementation phase of the adaptive management process are implicit
in the design of alternatives. That is, the set of actions to pursue is determined by the
selection of a preferred alternative, whereas OFDM lets decision-makers change forest
management actions for achieving optimized alternative.
Twery, Peter and Scott (2005) introduced NED-2 as a DSS with data management,
artificial intelligence, and simulation models to improve project-level planning and
decision-making by integrating treatment prescriptions and alternative comparisons with
evaluations of multiple resources. NED-2 was unable to predict land degradation based
on every forest plan activity while OFDM identifies the most effective ecological and
human factors on forest land degradation. In NED-2, knowing which objectives have or
have not been achieved allows decision-makers to design alternative action that may
improve the situation. In OFDM, as an EIA model for forest management, the main goal
is the level of degradation of land and hence OFDM is looking for forest management
activities which are influencing forest degradation. In NED-2, such as OFDM, various
simulation models are used to predict the future state of the forest ecosystem, given a set
of management activities and the natural change of the existing forest ecosystems. Given
several possible, simulated future forest ecosystem conditions, it is then possible to assess
how well each future state of the forest satisfies the original objective. NED-2 and OFDM
provide several tabular and graphical reports to help in this evaluation, but alternative
selection is done ad hoc by a professional forester in consultation with the landowner and
other stakeholders. Hence, NED-2 used primarily by private consulting foresters to
manage forest lands, but also by public land managers to evaluate alternatives for large
holdings without paying attention to each forest management activities of each
alternatives. Reynolds (2005) introduced ecosystem management decision support
(EMDS) which provides integrated decision support for environmental evaluation and
planning at multiple spatial scales. The EMDS system (Version 3.0), developed by the
U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station, is an
extension to Arc Map that provides integrated decision support for environmental
Table 6. The results of decision-making in current forest project and alternative forest project.
Class
Area (%) in current
forest project
Area (%) in alternative
forest project EIA decision-making
1 42.89 51 Acceptable
2 75.6 98 Acceptable and marginally acceptable
3 51.1 49 Some rehabilitation and
mitigation actions needed
4 23.51 0 Considerable rehabilitation and
mitigation actions needed
5 5.2 0 Unacceptable and marginally unacceptable
6 0.9 0 Unacceptable
Journal of Environmental Planning and Management 17
Downloaded by [Ali Jahani] at 05:16 06 April 2015
evaluation and planning at multiple spatial scales. Current planning for a subsequent
version of EMDS and NED includes implementation of the analytical hierarchy process
(Saaty 1992) as a way to objectively select an alternative and to perform sensitivity
analyses on goals, whereas OFDM uses ANN modeling, as a computer-based system, to
simulate forest ecosystem in different forest management alternatives.
Impact assessment management and DSSs recognize the complexity of natural
ecosystems and human interactions with them (Jakeman and Letcher 2003). Any
assessment on this basis is presently very difficult due to the need to assess independently
and integrate possibly deficient and conflicting data from a wide variety of sources
(Hepting 2007). Sustainability of forest management depends on maintaining soil quality
and proper forest management. Soil erosion from productive forests decreases soil
quality, diminishes on-site land value, and causes off-site environmental damage.
Therefore, OFDM, which consists ecological and erosion index, is introduced to assess
the complexity of forest ecosystems and human interactions with them. Similarly, Mas
et al. (2004) prepared deforestation area maps which were overlaid with spatial variables
which are proximity to roads and settlements, forest area fragmentation, slope, elevation,
and soil type to determine the relationship between deforestation and these variables.
ANN was trained to estimate the propensity to deforestation and was used to develop
deforestation risk assessment maps (Mas et al. 2004).
In this paper, the larger size of the area of the forest project shows a higher
environmental impact due to intensive road network and the larger cutting level
(Table 6). The recent main criticism is that the construction of forest roads is destroying
the environment, causing soil erosion, habitat loss, scenic impacts, etc. (Gumus, Acar and
Toksoy 2008). Using EDSS tool, the density of the forest road network and the volume of
wood production were diminished to achieve OFDM entities in the range of acceptable
and marginally acceptable categories. Other forest activities such as tree cutting,
recreation use, or overgrazing influence the OFDM entities and have potential to be
planned based on OFDM as mitigation actions in alternative forest project. Rehabilitation
actions decrease the susceptibility of ecosystem so some forest activities, such as
reforestation, are suggested as a rehabilitation action in alternative forest project.
Multiple-criteria decision analysis (MCDA) is used extensively in real-world
applications. Despite the diversity of MCDA approaches, methods, and techniques, the basic
ingredients of MCDA are very simple: a finite or infinite set of actions (alternatives,
solutions, courses of action,...), at least two criteria and, obviously, at least one decision-
maker (Figueira, Greco and Ehrgott 2005). Hence, we determined criteria which indicate
degradation of land along forest management plan implementation. Solving MCDA
problems has been used as a popular research topic for many decades. Different approaches
have been proposed (and applied) by famous researchers, such as the analytic hierarchy
process (AHP) (Saaty 1992), for a survey of MCDA methods and applications see Stewart
(1992); Gardiner and Steuer (1994); Siskos and Spyridakos (1999). The main goal for
solving an MCDA problem is to determine the preference structure of the decision-maker,
and many approaches have been developed (Jian and Song 2002). Many different
techniques, described in the literature as ANN, have been developed to evaluate the fitness
of solutions in a multi-criteria context and guarantee enough diversity to achieve a uniform
distribution of solutions. The advantages make ANN a promising tool in solving MCDA
problems and representing the preference of the decision-maker (Malakooti and Zhou 1998;
Malakooti and Subramanian 2000). In this research, to solve MCDA problems using ANN,
the most important issues to be considered are how to formulate forest degradation,
considering associated factors with degradation as decision-makers’ criteria to guide the
18 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
search for the most desirable preference. In the field of EIA in this research, decision-makers’
preferences are based on the numerical results of degradation models (Tables 5 and 6). The
AHP-like methods (those methods that calculate the principal eigenvector of the pairwise
comparison matrix AHP-like methods) are able to accommodate and handle the imprecise
and conflictive results of criteria pairwise comparison which is done by experts. However,
we used ANN to capture the degradation results to avoid any prior assumptions of the
structure of the preference of degradation criteria while applying AHP requires some
specific conditions. The system is an intelligent EDSS, which allows the decision-maker to
model criteria in forest degradation in order to reach and employ the optimal allocation plan.
Considering results, MCDA approaches based on ANN, is an encouraging and robust
method for solving MCDA problems.
5. Conclusions
An accurate forest management plan aimed at maximizing the environmental
performance while minimizing the environmental impact of forestry activities
(Buonocore et al. 2014). Our approach adopts a structured analysis of EIA issues in the
forests that can measure ecological impacts comparatively, which is fostering
environmental protection. Considering uncertainties, direct and indirect impacts,
cumulative impacts, short-term, mid-term, and long-term impacts as a challenge in EIA,
the feedback of ecosystem in response to the development plan is crucially significant.
The results of research proved the capability of ANN in prediction of forest ecosystem
feedback after forest project implementation. Thereby, the EIA of forest projects is
considered as the most reliable way to reduce the environmental impacts. In this way,
OFDM, as an EDSS in the hands of decision-makers, predicts the feedback of temperate
broadleaf forests in respond to the forest project activities and it helps to find the most
logical solution using the rehabilitation and mitigation actions. Such a tool can support
local managers and policy makers committed to implement an environmentally sound
management of forestry activities. Moreover, the calculated indicators provide a
benchmark for future comparisons with similar systems and for monitoring the trend of
the investigated forestry system over time.
In the research references, the importance of comprehension and proper use of EDSS
and the related technologies have been highlighted (De la Rosa et al. 2004; Oxley et al.
2004; Elmahdi, Kheireldin and Hamdy 2006; Znidarsic, Bohanec and Zupan 2006;D
ıez
and McIntosh 2009; Lautenbach et al. 2009). Future studies could be oriented to integrate
the applied assessment method framework with other methods capable of capturing
economic and social aspects related to forest use and management, thus fully exploring
constraints and potentialities of forest management across environmental, economic, and
social dimensions. Obviously, the development of EDSS is avoidable and in the near
future, the extended progress in EDSS, especially spatial database, long-term database,
modeling and advanced calculation technics, will be taken (Denzer 2005).
Acknowledgements
I am grateful to the University of Tehran for cooperation in part of this research.
Disclosure statement
No potential conflict of interest was reported by the authors.
Journal of Environmental Planning and Management 19
Downloaded by [Ali Jahani] at 05:16 06 April 2015
ORCID
Ali Jahani http://orcid.org/0000-0003-4965-3291
Mahmoud Omid http://orcid.org/0000-0003-2616-4903
References
Argent, R.M., J.M. Perraud, J.M. Rahman, R. Grayson, and G.M. Podger. 2009. “A New Approach
to Water Quality Modelling and Environmental Decision Support Systems.” Environmental
Modelling & Software 24 (7): 809818.
Arsene, C.T.C., B. Gabrys, and D. Al-Dabass. 2012. “Decision Support System for Water
Distribution Systems Based on Neural Networks and Graphs Theory for Leakage Detection.”
Expert Systems with Applications 39: 1321413224.
Asadi, E., G. da Silvab, M. Henggeler Antunes, C. Dias, and L. Glicksman. 2014. “Multi-objective
Optimization for Building Retrofit: A Model Using Genetic Algorithm and Artificial Neural
Network and an Application.” Energy and Buildings 81: 444456.
Azari Dehkordi, F., and N. Nakagoshi. 2003. “Rehabilitation of Shibateranthis Pinnatifida Maxim, a
GIS Approach.” Hikobia 14: 914.
Boillereaux, L., C. Cadet, and A. Le Bail. 2003. “Thermal Properties Estimation via Real Time
Neural Network Learning.” Journal of Food Engineering 57: 1723.
Buonocore, E., T. H€
ayh€
a, A. Paletto, and P.P. Franzese. 2014. “Assessing Environmental Costs and
Impacts of Forestry Activities: A Multi-method Approach to Environmental Accounting.”
Ecological Modelling 271: 1020.
Burris, R.K., and L.W. Canter. 1997. “Cumulative Impacts Are Not Properly Addressed in
Environmental Assessments.” Environmental Impact Assessment Review 17 (1): 518.
Callan, R. 1999. The Essence of Neural Networks. Upper Saddle River, New Jersey: Prentice Hall.
Canter, L. 1995. Environmental Impact Assessment. New York: McGraw-Hill.
Cort
es, U., M. S
anchez-Marr
e, L. Ceccaroni, I. R-Roda, and M. Poch. 2000. “Artificial Intelligence
and Environmental Decision Support Systems.” Applied Intelligence 13 (1): 7791.
De la Rosa, D., F. Mayol, E. Diaz-Pereira, M. Fernandez, and D. de la Rosa Jr. 2004. “A Land
Evaluation Decision Support System (MicroLEIS DSS) for Agricultural Soil Protection: With
Special Reference to the Mediterranean Region.” Environmental Modelling & Software 19
(10): 929942.
Denzer, R. 2005. “Generic Integration of Environmental Decision Support Systems State-of-the-
Art.” Environmental Modelling & Software 20 (10): 12171223.
D
ıez, E., and B.S. McIntosh. 2009. “A Review of the Factors which Influence the Use and
Usefulness of Information Systems.” Environmental Modelling & Software 24 (5): 588602.
Eccleston, C. 2000. Environmental Impact Assessment. New York: John Wiley & Sons.
Elmahdi, A., K. Kheireldin, and A. Hamdy. 2006. “GIS and Multi-criteria Evaluation: Robust Tools
for Integrated Water Resources Management.” Water International 31 (4): 440447.
Elmahdi, A., and D. McFarlane. 2009. A Decision Support System for Sustainable Groundwater
Management Case study:Gnangara Sustainability Strategy – Western Australia.
International Congress on Modeling and Simulation Cairns, Queensland. Australia Modeling
and Simulation Society of Australia and New Zealand.
Elmas, C., S. Sagıroglu, I. Colak, and G. Bal. 1994. “Modeling of a Nonlinear Switched Reluctance
Drive Based on Artificial Neural Networks.” In IEEE conference on power electronics and
variable-speed drives, 712. London: IET Press.
Elmas, C., and Y. Sonmez. 2011. “A Data Fusion Framework with Novel Hybrid Algorithm for
Multi-agent Decision Support System for Forest Fire.” Expert Systems with Applications 38:
92259236.
Fernandez, F., J. Seco, A. Ferrer, and M.A. Rodrigo. 2009. “Use of Neurofuzzy Networks to
Improve Wastewater Flow-rate Forecasting.” Environmental Modelling & Software 24:
686693.
Figueira, J., S. Greco, and M. Ehrgott. 2005. Multiple Criteria Decision Analysis: State of the Art
Surveys. Boston: Springer Science.
Gardiner, L.R., and R.E. Steuer. 1994. “Unified Interactive Multiple Objective Programming.”
European Journal of Operational Research 74 (5): 235244.
20 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
Gavrilovic, Z. 1998. “The Use of an Empirical Method (Erosion Potential Method) for Calculating
Sediment Production and Transportation in Unstudied or Torrential Streams.” International
Conference of River Regime 1: 2329.
George, C. 1999. “Testing for Sustainable Development Through Environmental Assessment.”
Environmental Impact Assessment Review 19 (2): 175200.
Gorry, G., and M. Scott Morton. 1971. “A Framework for Management Information Systems.”
Sloan Management Review 13 (1): 5565.
Gumus, S., H.H. Acar, and D. Toksoy. 2008. “Functional Forest Road Network Planning by
Consideration of Environmental Impact Assessment for Wood Harvesting.” Environmental
Monitoring and Assessment 142 (13): 109116.
Hagan, M.T., H.B. Demuth, and O.D. Jesus. 2002. “An Introduction to the Use of Neural Networks
in Control Systems.” International Journal of Robust and Nonlinear Control 12 (11): 959985.
Hanna, K.S., I. P€
ol€
onen, and K. Raitio. 2011. “A Potential Role for EIA in Finnish Forest Planning:
Learning from Experiences in Ontario, Canada.” Impact Assessment and Project Appraisal 29
(2): 99108.
Haykin, S. 1999. Neural Networks: A Comprehensive Foundation. New York: Mcmillan College
Publishing Company.
Hebb, D. 1949. “The Organization of Behavior.” New York Science11: 3337.
Hepting, D.H. 2007. “Decision Support for Local Environmental Impact Assessment.”
Environmental Modelling & Software 22 (4): 436441.
Hussain, M.A., M. Safiur, and C.W. Rahman. 2002. “Prediction of Pores Formation (Porosity) in
Foods During Drying: Generic Models by the Use of Hybrid Neural Network.” Journal of Food
Engineering 51: 239248.
Iliadis, L.S., and F. Maris. 2007. “An Artificial Neural Network Model for Mountainous Water-
Resources Management: The Case of Cyprus Mountainous Watersheds.” Environmental
Modelling & Software 22: 10661072.
Imai, K., L. Keele, and T. Yamamoto. 2010. “Identification, Inference and Sensitivity Analysis for
Causal Mediation Effects.” Statistical Science 25 (1): 5171.
Jakeman, A.J., and R.A. Letcher. 2003. “Integrated Assessment and Modelling: Features, Principles
and Examples for Catchment Management.” Environmental Modelling & Software 18 (6):
491501.
Jambunathan, K., S.L. Hartle, S. Ashforth-Frost, and V.N. Fontana. 1996. “Evaluating Convective
Heat Transfer Coefficients Using Neural Networks.” International Journal of Heat Mass
Transfer 39 (11): 23292332.
Jian, C., and L. Song. 2002. “An Interactive Neural Network-Based Approach for Solving Multiple
Criteria Decision-Making Problems.” Decision Support Systems 36: 137146.
Knowler, D., and J. Lovett. 1996. Manual for Environmental Assessment in Forestry, Prepared for
FAO Regional Project “Forestry Planning and Policy Assistance in Asia and the Pacific.”
New York: New York University Press.
Koskela, M. 2011. “Expert Views on Environmental Impacts and Their Measurement in the Forest
Industry.” Journal of Cleaner Production 19: 13651376.
Lautenbach, S., B. J€
urgen, N. Graf, R. Seppelt, and M. Matthies. 2009. “Scenario Analysis and
Management Options for Sustainable River Basin Management: Application of the Elbe DSS.”
Environmental Modelling & Software 24 (1): 2643.
Leknes, E. 2001. “The Roles of EIA in the Decision-Making Process.” Environmental Impact
Assessment Review 21 (4): 309334.
Leondes, C. 1998. Fuzzy Logic and Expert Systems Applications. Los Angeles: Academic Press.
MacDonald, L.H. 2000. “Evaluating and Managing Cumulative Effects: Process and Constraints.”
Environmental Management 26 (3): 299315.
Maier, H.R., and G.C. Dandy. 2000. “Neural Networks for the Prediction and Forecasting of Water
Resources Variables: A Review of Modelling Issues and Applications.” Environmental
Modelling & Software 15 (1): 101124.
Maier, H., R.A. Jain, G.C. Dandy, and K.P. Sudheer. 2010. “Methods Used for the Development of
Neural Networks for the Prediction of Water Resource Variables in River Systems: Current
Status and Future Directions.” Environmental Modelling & Software 25 (8): 891909.
Makhdoum, M.F. 2002. “Degradation Model: A Quantitative EIA Instrument, Acting as a Decision
Support System (DSS) for Environmental Management.” Environmental Management 30 (1):
151156.
Journal of Environmental Planning and Management 21
Downloaded by [Ali Jahani] at 05:16 06 April 2015
Malakooti, B., and S. Subramanian. 2000. “Multiple Criteria Approach for Integrated Matching
Supervision, Machinability, and Tool Performance with Polynomial Utility Functions.”
Engineering Valuation and Cost Analysis 2 (6): 238248.
Malakooti, B., and Y.Q. Zhou. 1998. “Approximating Polynomial Functions by Feedforward
Artificial Neural Networks: Capacity, Analysis, and Design.” Applied Mathematics and
Computation 90 (1): 55645577.
Mas, J.F., H. Puig, J.L. Palacio, and A. Sosa-L
opez. 2004. “Modelling Deforestation Using GIS and
Artificial Neural Networks.” Environmental Modelling & Software 19 (5): 461471.
McCarter, J.B., J.S. Wilson, P.J. Baker, J.L. Moffett, and C.D. Oliver. 1998. “Landscape
Management Through Integration of Existing Tools and Emerging Technologies.” Journal of
Forestry 96 (6): 1723.
McCulloch, W., and W. Pitts. 1953. “A Logical Calculus of the Ideas Immanent in Nervous
Activity.” Bulletin of Mathematical Biophysics 5: 115133.
McDonald, G.T., and L. Brown. 1995. “Going Beyond Environmental Impact Assessment:
Environmental Input to Planning and Design.” Environmental Impact Assessment Review
15 (6): 483495.
McHarg, I., 1969. Design with Nature. New York: Natural History Press.
McIntosh, B., P. Jeffrey, M. Lemon, and N. Winder. 2005. “On the Design of Computer-Based
Models for Integrated Environmental Science.” Environmental Management 35 (6): 741752.
Michelsen, O., C. Solli, and A.H. Strømman. 2008. “Environmental Impact and Added Value in
Forestry Operations in Norway.” Journal of Industrial Ecology 12 (1): 6981.
Nasr, M., M. Moustafa, H. Seif, and G. El Kobrosy. 2012. “Application of Artificial Neural
Network (ANN) for the Prediction of ELAGAMY Wastewater Treatment Plant Performance-
EGYPT.” Alexandria Engineering Journal 51 (1): 3743.
Nasr, M., and H.F. Zahran. 2014. “Using of pH as a Tool to Predict Salinity of Groundwater for
Irrigation Purpose Using Artificial Neural Network.” Egyptian Journal of Aquatic Research 40:
111115.
Oliver, C., and M. Twery. 1999. Decision Support Systems/Models and Analysis. New York: Oxford
University Press.
Omid, M., A. Baharlooei, and H. Ahmadi. 2009. “Modeling Drying Kinetics of Pistachio Nuts with
Multilayer Feed-Forward Neural Network.” Drying Technology 27 (10): 10691077.
Oxley, T., B.S. McIntosh, N. Winder, M. Mulligan, and G. Engelen. 2004. “Integrated Modelling
and Decision-Support Tools: A Mediterranean Example.” Environmental Modelling &
Software 19 (11): 9991010.
Paliwal, J., N.S. Visen, and D.S. Jayas. 2001. “Evaluation of Neural Network Architectures for
Cereal Grain Classification Using Morphological Features.” Journal of Agricultural
Engineering 79 (4): 361370.
Picton, P. 2000. Neural Networks, 2nd ed. New York: Palgrave.
Potter, K.M., F.W. Cubbage, and R.H. Schaberg. 2005. “Multiple-Scale Landscape Predictors of
Benthic Macroinvertebrate Community Structure in North Carolina.” Landscape and Urban
Planning 71 (24): 77—90.
Pradhan, B., and S. Lee. 2010. “Landslide Susceptibility Assessment and Factor Effect Analysis:
Backpropagation Artificial Neural Networks and Their Comparison with Frequency Ratio and
Bivariate Logistic Regression Modeling.” Environmental Modelling & Software 25: 747759.
Regan, H.M., M. Colyvan, and M.A. Burgman. 2000. “A Proposal for Fuzzy International Union for
the Conservation of Nature (IUCN) Categories and Criteria.” Biological Conservation 92 (1):
101108.
Reynolds, K. 2005. “Integrated Decision Support for Sustainable Forest Management in the United
States: Fact or Fiction?” Computers and Electronics in Agriculture 49: 623.
Rosenblatt, F. 1958. “The Perceptron: A Probabilistic Model for Information Storage and
Organization in the Brain.” Psychological Review 62 (11): 386408.
Rothman, D. 2000. “Measuring Environmental Values and Environmental Impacts: Going from the
Local to the Global.” Climatic Change 44 (3): 351376.
Roy, B. 2010. “Robustness in Operational Research and Decision Aiding: A Multi-faceted Issue.”
European Journal of Operational Research 200: 629638.
Saaty, T.L. 1992. Multicriteria Decision Making: The Analytical Hierarchy Process. Pittsburg:
RWS Publications.
22 A. Jahani et al.
Downloaded by [Ali Jahani] at 05:16 06 April 2015
Sablani, S.S., O.D. Baik, and M. Marcotte. 2002. “Neural Networks for Predicting Thermal
Conductivity of Bakery Products.” Journal of Food Engineering 52: 299304.
Saghebtalebi, K., T. Sajedi, and F. Yazdian. 2003. Forest of Iran. Tehran: Research Institute of
Forest Publication.
Saltelli, A., M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Saisana, and S.
Tarantola. 2008. Global Sensitivity Analysis. New York: John Wiley & Sons.
Segura, M., D. Ray, and C. Maroto. 2014. “Decision Support Systems for Forest Management: A
Comparative Analysis and Assessment.” Computers and Electronics in Agriculture 101:
5567.
Sepp€
al€
a, J., M. Melanen, T. Jouttij€
arvi, L. Kauppi, and N. Leikola. 1998. “Forest Industry and the
Environment: A Life Cycle Assessment Study from Finland.” Resources, Conservation and
Recycling 23 (12): 87—105.
Simon, H. 1960. The New Science of Management Decision. New York: Harper Brothers.
Siskos, Y., and A. Spyridakos. 1999. “Intelligent Multi-criteria Decision Support: Overview and
Perspective.” European Journal of Operational Research 113 (2): 2538.
Sreekanth, S., H.S. Ramaswamy, and S.S. Sablani. 1998. “Prediction of Sychrometric Parameters
Using Neural Networks.” Drying Technology 16 (35): 825837.
Stewart, T.J. 1992. “A Critical Survey on the Status of Multiple Criteria Decision Making Theory
and Practice.” Omega 20 (5 and 6): 654668.
Tayebi, M., M. Tangestani, and H. Roosta. 2010. Environmental Impact Assessment Using Neural
Network Model: A Case Study of the Jahani, Konarsiah and Kohe Gach Salt Plugs, Shiraz,
Iran. Shiraz: ISPRS TC VII Symposium.
Twery, J., D. Peter, and A. Scott. 2005. “NED-2: A Decision Support System for Integrated Forest
Ecosystem Management.” Computers and Electronics in Agriculture 49: 2443.
Vali, A., M. Ramesht, A. Seif, and R. Ghazavi. 2012. “An Assessment of the Artificial Neural
Networks Technique to Geomorphologic Modeling Sediment Yield (Case Study Samandegan
River System).” Geography and Environmental Planning Journal 22 (4): 1934.
Widrow, B. 1959. Adaptative Sampled-Data Systems, A Statistical Theory of Adaptation Ire Wescon
Convention Record. New York: New York Institute of Radio Engineers.
Yeo, I., and J.J. Yee. 2014. “A Proposal for A Site Location Planning Model of Environmentally
Friendly Urban Energy Supply Plants Using an Environment and Energy Geographical
Information System (E-GIS) Database (DB) and an Artificial Neural Network (ANN).” Applied
Energy 119: 99117.
Yijun, L., T. Jiali, J. Hongfen, Z. Guangping, and Y. Zhimin. 2010. “Artificial Neural Networks
Applied in Environmental Quality Assessment.” Chengdu: 3rd IEEE International Conference
on Computer Science and Information Technology (ICCSIT).
Zangeneh, M., M. Omid, and A. Akram. 2010. “Assessment of Machinery Energy Ratio in Potato
Production by Means of Artificial Neural Network.” African Journal of Agricultural Research
5 (1): 993998.
Znidarsic, M., M. Bohanec, and B. Zupan. 2006. “ProDEX A DSS Tool for Environmental
Decision-Making.” Environmental Modelling & Software 21 (2): 15141516.
Journal of Environmental Planning and Management 23
Downloaded by [Ali Jahani] at 05:16 06 April 2015
