Ecological Indicators 88 (2018) 439–453

Contents lists available at ScienceDirect

Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind

Original Articles

Modeling land use change using Cellular Automata and Artificial Neural T
Network: The case of Chunati Wildlife Sanctuary, Bangladesh
⁎
Kamrul Islama, , Md. Farhadur Rahmanb, Mohammed Jashimuddinc
a
Graduate School of Engineering, Department of Systems Innovation, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan
b
Graduate School of Agriculture, Division of Forest and Biomaterials Sciences, Kyoto University, Kyoto 606-8502, Japan
c
Institute of Forestry and Environmental Sciences, University of Chittagong, Chittagong 4331, Bangladesh

A R T I C L E I N F O A B S T R A C T

Keywords: Land use changes generally affect the integrity of an ecosystem. The effect of this change can be very severe if the
Chunati Wildlife Sanctuary conversion disrupts a crucial habitat of major plants and animals. The degraded Chunati Wildlife Sanctuary is
Markov chain one such area of Bangladesh which is facing a serious problem of rapid land use change. In this study, the future
Cellular Automata trend of land use change of the area was modelled using Artificial Neural Network. Several driver variables were
Artificial Neural Network
also incorporated to determine their effect on land use change. Binary logistic regression was used to assess the
Binary logistic regression
significance of the drivers of land use change for this region. The analysis shows that nearly 76% of the total land
Land use and land cover change
area (8258 ha) was covered by vegetation during 2005. After 2005, that was reduced to 61% (6637 ha) in 2015,
a 15% decline from 2005. On the other hand, the coverage of vacant land increased from nearly 10% in 2005 to
22% in 2015. This is indeed a matter of real concern. The critical analysis suggests that Cellular Automata is not
a good fit to simulate the future land uses as it misdirects the analysis both spatially and numerically. The
incorporation of driver variables gives strength to the Artificial Neural Network to predict the future. The chi-
square value for the prediction of land use of the area found from the neural network was 7.815 which was
greater than the critical value (3.316). The neural network was found to be a good fit for future land use
prediction. The kappa index of variation shows that the overall accuracy of the prediction using neural network
was above 90%. Elevation, slope, and distance to the road were the three driver variables which were found
statistically significant while predicting the probability of forest land use change. The accuracy of the binary
logistic regression was about 61% which was quite satisfactory. The simulation result shows that almost 5732 ha
of the total land will be in the forest category of land use during 2020 and it will be further decreased to 5128 ha
in 2025. The vacant area will increase from 24% to 31% from 2020 to 2025. Based on the findings and simulated
land use map of 2020–2025, the study will help the management authority of this critical habitat to take proper
action before further degradation occurs.

1. Introduction that LULCC is happening due to some factors or drivers. Biophysical

Information regarding land use and land cover change (LULCC) has
been a great interest for environmentalists because of the impact of it
on the global environment (Abuelaish and Olmedo, 2016; Subedi et al.,
2013). Anthropogenic utilization of land is referred to as ‘land use’
(Islam et al. 2017), while unaltered surfaces are described as having a
land cover (Islam et al. 2016). The major causes for LULCC vary de-
pending on the nature and extent of the area (Lambin et al., 2001) but
massive deforestation, grassland modification, agricultural intensifica-
tion and excessive extraction ends the list of LULCC causes (Goldewijk,
2001). Apart from the causes, some researchers (Araya and Cabral,
2010; Dahdouh-guebas, 2002; Gao et al., 2015; Han et al., 2015) opined
(López et al., 2001), physical (Rizk Hegazy and Kaloop, 2015), and
socio-economic (Megahed et al., 2015) factors are the most influential
drivers of LULCC.
With the massive improvement of satellite based technologies, sci-
entists are now continually using Remote Sensing (RS) and Geographic
Information System (GIS) in the broader aspect of land use science (Le
Hégarat-Mascle et al., 2005; Mallick et al., 2008; Nguyen et al., 2016;
Raghuvanshi et al., 2015; Ramachandra et al., 2016). Until 2005, re-
searchers of land use science were confined to figuring out past and
present trends of LULCC (Aspinall, 2004). However, with recent im-
provements and availability of data sources and sophisticated tech-
nologies, researchers are now trying to predict the future LULCC

⁎
Corresponding author.
E-mail addresses: kamrul-islam@g.ecc.u-tokyo.ac.jp, ksujonifescu@gmail.com (K. Islam).

https://doi.org/10.1016/j.ecolind.2018.01.047
Received 17 May 2017; Received in revised form 30 December 2017; Accepted 24 January 2018
Available online 02 February 2018
1470-160X/ © 2018 Elsevier Ltd. All rights reserved.
K. Islam et al.
(Ahmed and Ahmed, 2012; Beaumont and Duursma, 2012; Fang et al.,
2006). Modeling LULCC in this aspect is becoming more important
(Herold et al., 2002) as there will be time to address anticipated
changes at various spatial and temporal scale scenarios (Daye and
Healey, 2015). Markov Chain Analysis (MCA) or Markov Model
(Sivakumar, 2014), Cellular Automata (Sinha et al., 2015), Cellular
Automata-Markov Model (CA-Markov) (Subedi et al., 2013), Artificial
Neural Network (ANN) (Schneider and Pontius, 2001), Binary Logistic
Regression (Zeng et al., 2008; Zhan et al., 2002) are some of the most
commonly used model for the prediction and simulation of future
LULCC. The Chunati Wildlife Sanctuary (CWS), located at the southeast
corner of Bangladesh, is considered as one of the most important Pro-
tected Areas (PA) as it is the habitat of endangered Asian elephants
(Islam et al., 2017; Rahman et al., 2016). However, due to excessive
extraction of its natural resources, this nature reserve is now in a fragile
condition (Islam et al., 2016; Nath et al., 2016a). Rahman et al. (2016)
showed that the forested landscape of CWS is highly fragmented,
having a tremendous impact on ecosystem functioning. The rich natural
forest of the area is now disappearing at an increasing rate (Islam et al.,
2017). Nath et al. (2016a) described the background causes of forest
loss while Islam (2015) depicted the trend of LULCC for the year 2005,
2010, and 2015. It is evident from these findings that this forested
landscape is rapidly degrading. Moreover, the future scenario is still
unknown. Some comprehensive studies on the quantitative assessment
of vegetation condition can be found from GIZ (2015); Nath et al.
(2016a,b); NSP (2003,2004). Most of the studies are performed to find
out the impact of co-management (Tamima, 2016), resource extraction
from the sanctuary (Miah, 2009), and status of biodiversity (Nath et al.,
2016b). However, very little information is available regarding LULCC
in CWS and there was no such attempt to assess the future land use
change of the area. This research gap must be addressed for the sus-
tainable management of CWS. If such predictions are possible, they will
be of great use considering the conservation of this nature reserve.
Hence, this study aimed to predict the future LULCC of CWS based on
the study of Islam (2015). ANN and binary logistic regression model are
found best fitted for this purpose, and are used to simulate the LULCC of
CWS for the year 2020 and 2025. These models were also found suc-
cessful to predict the future LULCC of the forested landscape (Beaumont
and Duursma, 2012; Beuchle et al., 2015; Dale et al., 2009). Attempts
are also made to determine the drivers of the land use change. The
findings from this study will be useful for the conservation planning of
this important protected area.
1.1. Objectives of the study and research questions
This study will answer some of the important research questions
regarding the trend of LULCC of the CWS. The research questions are
(a) what is the current spatial trend of LULCC? (b) what is the rate of
this change? (c) where is the location of this change? (d) what are the
drivers of this change? (e) which models are useful to simulate and
predict this change with good accuracy? and (f) what will be the future
scenarios regarding LULC?
The overall objectives of the study are:
1. To predict the land use and land cover (LULC) scenario of CWS for
the year 2020 and 2025.
2. To find out the influence of drivers of land use and land cover
change (LULCC) of CWS.
2. Materials and methods
2.1. Study area
The CWS is a tropical semi-evergreen forest in south-eastern
Bangladesh that lies between 21° 40′ N latitude and 92° 07′ E longitude
(Fig. 2.1). It was officially established as a wildlife sanctuary through a
440
Ecological Indicators 88 (2018) 439–453
gazette notification in 1986 under the provision of Bangladesh Wildlife
(Preservation) (Amendment) Act, 1974 (Act No. XVII of 1974). It is
divided into 7 beats (beat is the smallest administrative unit of a large
forest area) namely Chunati, Aziznagar, Harbang, Jaldi, Chambal, Na-
pura, and Puichari. It is one of the most vital protected areas of Ban-
gladesh because it is an important habitat of the globally threatened
Asian Elephant (Elephas maximus). This sanctuary is also rich with di-
versified flora and fauna, but due to over-exploitation of natural re-
sources, it is now in a poor condition (Islam et al., 2016; Nath et al.,
2016b).
The physiography of CWS is hilly to mountainous with shallow to
deep gullies spreading over the entire area. The average elevation of the
area is 30–90 m above mean sea level. The area is also crisscrossed with
creeks, which offers good drainage and supply clean water to both
people and wildlife and are also used for irrigation for roughly
3–4 months in a year (Nath et al., 2016b).
2.2. Data processing
This study uses both the remotely sensed satellite data provided by
National Aeronautics and Space Administration (NASA) and some other
ancillary data. The classified images of 2005, 2010 and 2015 of CWS
(Islam, 2015) are used as the main input data and again reclassified into
four broad categories i.e. Vegetation, Agricultural land, Vacant land
and Others (Table S1). Several preprocessings steps were taken to use
these classified imageries. The 16 land use classes of Islam (2015) are
reclassified to 4 major classes to narrow down the focus of the study.
The input classified images came from two different sensors Landsat
TM (Thematic Mapper) and Landsat OLI/TIRS (Operational Land
Imager/Thermal Infrared Sensor). As the sensors are different, it could
affect the DN (Digital Number) value of pixels. So, the three imageries
are resampled to 30 m resolution each having the same extent and same
data type (8 bit unsigned) based on the study of Ghebrezgabher et al.
(2016). The imageries are then reprojected to UTM (Universal Trans-
verse Mercator) 46 N coordinate system.
Some studies (Ghebrezgabher et al., 2016; Kantakumar and
Neelamsetti, 2015; Lin et al., 2015) suggested that the satellite data
should be passed through atmospheric correction process. Although
Song et al. (2001) described that it would not affect the result sub-
stantially. So, atmospheric correction is not applied. Beuchle et al.
(2015) deciphered that if images (whether classified or raw) from dif-
ferent satellites are used for the study, it should be co-registered to keep
the extent of the imageries same. The same method is also applied to
this research to co-register all the used imageries keeping the 2005
classified image as a master image and the rest two (2010 and 2015) as
slave images (Beuchle et al., 2015).
Apart from the classified imageries, this study also takes some other
ancillary data as input for future land use simulation and modeling
purpose. Land use change modeling experts showed that some variables
like elevation, slope, population, economic proxy indicators, distance
from roads, distance from urban centers and so on are the major drivers
for change in land use (Han et al., 2015; Kim et al., 2014; Lin et al.,
2008; Millington et al., 2007; Raghuvanshi et al., 2015; Schneider and
Pontius, 2001; Zeng et al., 2008; Al-sharif and Pradhan, 2013). The
driver variables used in this study are slope, elevation, distance to brick
field, distance to highways, distance to local roads, distance to locality,
and distance to natural water sources (Table S2 and Fig. S1). SRTM
DEM (Shuttle Radar Topography Mission; Digital Elevation Model) of 1
arc second from NASA are extracted based on the extent of the study
area and processed. The raw DEM are resampled to 30 m resolution
along with UTM 46 N re-projection. The elevation and slope images are
then prepared using ArcGIS 10.2.2. The unit used for elevation is meter
(m) and for slope is percentage (%). These two driver variables are used
in the modeling without any further processing operation. Due to un-
availability of the valid database for roads, highways, brick fields, lo-
calities, water sources, this study uses Google Earth to create the
K. Islam et al.
Fig. 2.1. Study area.
necessary layers. The said features are digitized in Google Earth and
exported as KML (Keyhole Markup Language) file in ArcGIS. It is no-
teworthy that these features are in vector format. So, to match these
datasets with the previous input; rasterization of these data is required.
These datasets are rasterized keeping 30 m pixel resolution and re-
projected to UTM 46 N coordinate system. Elevation and slope data
come from a valid source with known units. However, the manually
created proximity datasets, as stated above, do not have a valid unit in
them. So, to use these as driver variables, Euclidean distance approach
is adopted. This process analyses the cell’s relationship to a source or set
of sources based on the straight-line distance. It measures the distance
from each cell in the raster to the closest or nearest source. The closer
the distance from a specific source the higher the impact for that source
to that specific cell or area. The driver variables used in this study are
shown in Fig. S1. The prepared land use images of the studied time
frame and the explanatory variables are exported to IDRISI Selva for
further simulation. Classified images of 2005 and 2010 were used to
model the land use change in 2015 using Markov model, Cellular Au-
tomata-Markov (CA-Markov) model, Artificial Neural Network (ANN)
model in IDRISI. The classified image of 2015 is used to validate the
model performance using the said image as a reference.
2.3. Land use change modeling
In this study, Cellular Automata-Markov (CA-Markov) model,
Artificial Neural Network (ANN) model and binary logistic regression
model are used and compared.
2.3.1. Cellular Automata-Markov
Cellular Automata-Markov is a combined Cellular Automata-
Markov Chain LULCC prediction procedure. The transition areas pro-
duced from Markov model of the land use map 2005 and 2010 establish
the quantity of expected land use change from each existing category to
441
Ecological Indicators 88 (2018) 439–453
each other category in the next time period (i.e. 2015). The base land
use image (the land use image of 2010) is used as the starting point for
change simulation. The suitability image produced in the Markov
model establishes the inherent suitability of each pixel for each land
cover type. The predicted land use image of 2015 from this model is
compared with the actual 2015 image. A chi-square test is performed
between the actual 2015 image with the predicted image from CA-
Markov model. This model is run using the CA-Markov module of
IDRISI and validated using the validation module of it. The ROC
(Relative Operating Characteristic) is calculated following the metho-
dology of (Pontius and Parmentier, 2014). The flowchart of metho-
dology and the validation result for CA-Markov model are shown in Fig.
S2 and Table S3 respectively. In this study, it is found that CA-Markov
model is not a good fit for the study site and henceforth only best fitted
ANN, and binary logistic regression models are thoroughly analysed,
depicted and discussed.
2.3.2. Artificial Neural Network (ANN)
ANN is also known as Multi-Layer Perceptron (MLP) that undertakes
the classification of remotely sensed imagery through a Multi-Layer
Perceptron neural network classifier using the Back Propagation (BP)
algorithm. Land Change Modeler (LCM) of IDRISI is used to perform the
MLP. The driver variables are inserted as the independent variables and
land use change image from 2005 to 2010 as the dependent variable.
The explanatory power of driver variables is tested using Cramer’s V
value. A high Cramer’s V value indicates that the potential explanatory
power of the variable is good. However, it does not guarantee a strong
performance since it cannot account for the mathematical requirements
of the modeling approach used and the complexity of the relationship.
MLP first calculates the pixel of land use class based on persistence and
change. The input images of 2005 and 2010 along with transitional
probabilities calculated in Markov model are used to predict the sce-
nario of 2015. The change analysis tab of LCM calculates the trend of
K. Islam et al.
change and the persisted changed pixel of different land use category. It
also produces the change map based on the previous (2005) and later
(2010) image. Around 50% of the changed pixels were taken for
training the MLP, and the remaining 50% pixel were kept for valida-
tion. The training accuracy for MLP in this study was 80%. After
achieving such precision, the simulation is done to predict the land use
of 2015. The validation of 2015 simulated image is similar to the
process as described in CA-Markov model. After successful validation,
simulation is done to predict the land use of the study area for the year
2020, and 2025. The flowchart of ANN method is shown in Fig. 2.2.
2.3.3. Binary logistic regression
Binary logistic model is quite useful for LULCC modeling where the
dependent variable may be a categorical or binary variable. After the
validation of simulated 2015 image with the actual 2015 classified
image, binary logistic regression is applied to model the LULCC through
incorporating driver variables as the predictor. The best-predicted land
use image of 2015 from the CA-Markov and ANN model is used in this
method. This image is further reclassified into forest and non-forest
classes and is assigned a value of ‘1’ if the pixel is in the forest category
or ‘0’ if non-forest. The total area of CWS is covered by approximately
120244 pixels (or cells). Taking a 10% sampling intensity, 12024 ran-
domly selected points representing a pixel (or cell) (Fig. S3) were then
used to extract background values of the driver and dependent variables
from the imagery. These randomly generated points are used to extract
the background value of driver variables and dependent variable ima-
geries. The ‘Extract multiple values to points’ tools from ArcGIS was
used to perform this operation. The database file (.dbf extension) from
this operation is then exported to CSV (Comma Separated Value) format
and fed into ‘R’. Binary logistic regression is then applied using the GLM
function of ‘R’.
The GLM has three components; (a) a dependent variable (binary in
this case), (b) the predictor variable (driver variables used in this case),
and (c) a link function that links (a) and (b). The equations used for the
logistic regression are shown below.
442
Ecological Indicators 88 (2018) 439–453
Fig. 2.2. Flowchart of methodology for Artificial Neural
Network (ANN) model.
P
log ⎛ i ⎞ = β0 + β1x1 + β2 x2 + …+βi xi
⎜ ⎟
⎝ Pi ⎠
1−
eα + β1x1+ β2x2+…+ βi xi
P(x) =
1 + eα + β1x1+ β2x2+…+ βi xi
where, P(x) is the probability of the dependent variable (probability of
a pixel being forest or non-forest), βi are the regression coefficient of the
predictor variable. The exponential of the regression coefficient de-
termines whether a predictor variable is a risk or protective factor in
explaining the dependent variable.
3. Results
3.1. Temporal land use change in CWS from the period 2005–2015
During 2005, the whole sanctuary area of Chunati was rich in dif-
ferent vegetation. More than 75% (8257.77 ha) of the area was covered
by vegetation during that time. There was about 839 ha of agricultural
land which shared almost 8% of the total area of CWS. The other land
use classes covered 1826 ha, of which 1053 ha was vacant and 673 ha of
land was used in other ways (Fig. 3.1). Within this five years’ gap
(2005–2010), around 2200 ha of vegetation had been lost and con-
verted to other human induced land uses. Thus, it was found that in
2010, the vegetation reduced to 56% (6012.18 ha) while there was a
sharp increase in vacant areas sharing almost 26% (2812.68 ha) of the
total area of this sanctuary. Meanwhile, there was also a steady increase
in the agricultural land and others classes of land use during 2005–2010
periods.
After the devastating deforestation period of 2005–2010, the sce-
narios started to change a bit due to the initiation of different con-
servation projects in the area. In 2015, the vegetation area was found
covering about 6636.78 ha (61.32%) of the whole area. It is indeed a
significant positive change compared to 2005–2010 time period.
Agricultural land use also increased compared to the previous period
(1159.65 ha vs. 1083.15 ha). One important thing which happened
K. Islam et al. Ecological Indicators 88 (2018) 439–453

Fig. 3.1. LULC map of CWS 2005–2015.

between 2010 and 2015 was the reduction of vacant areas. About
400 ha of vacant areas were reduced due to plantation activities which
could be considered as a positive land use change for this sanctuary.
The change of the rest land use category was not so significant
(Fig. 3.1).
The trend of land use change of the area from 2005 to 2015 time
period is better understood from Fig. S4 and Fig. 3.2. Almost all the
land use categories followed a linear trend. Vegetation land use
followed a sharp fall in 2010 and then a steady rise till 2015. However,
the linear trend line suggested that there is a continuous decrease in this
category from 2005 to 2015 years with an R2 value of 0.49. Agricultural
land use showed a constant increasing trend and its association with
time period is found quite strong (R2 = 0.92) (Fig. S4).
Fig. 3.2 shows the spatial trend of change in land use in the study
area and provides the scale of change along with the direction. It is
helpful to identify the critical zone which changes a lot during the

Fig. 3.2. Spatial trend of change in land use (from vegetation to rest classes).

443
K. Islam et al.
studied time frame.
Vegetation is considered the most significant land use in this con-
text. Henceforth, the trend of change from this class to other is shown in
the spatial trend of change map. Fig. 3.2a shows the spatial trend of
change in vegetation to other land use classes for 2005–2010 time
period, while Fig. 3.2b shows the change from 2010 to 2015. The value
of the change is found after applying third-degree polynomial order
from the land change modeler of IDRISI Selva version. The positive
value indicates conversion from the said category to other whereas the
negative value indicates the gain. It is noteworthy here that the ideas of
this spatial trend of change map is to provide a means of generalizing
idea about the pattern of change. The numeric values do not have any
special significance. The surface is created by coding areas of change
with ‘1’ and areas of no change with ‘0’ and treating them as if they
were quantitative values.
The dark pitch color represents the severity of change. Pitch color of
Fig. 3.2a is darker than that of Fig. 3.2b considering the spatial cov-
erage of change. This signifies that change in 2005–2010 was massive
compared to 2010–2015.
Although conversion started from the northern margin of the area,
the change in the south-eastern outweighed the other areas of the
sanctuary. Thus, it can be inferred that the south-east margin of the area
is highly sensitive to conversion to other land uses from vegetation.
3.2. LULCC in CWS from 2005–2015 based on ANN model
ANN is a machine-learning technique which is capable of capturing
non-linear relationships among different land use transformations
through formulating the relationships between input and output vari-
ables (Azari et al., 2016). The Land Change Modeler (LCM) module of
IDRISI can perform the ANN method for the prediction of future land
use change through incorporating Markov chain with Multi-Layer
Perceptron (MLP). There is a three-step procedure for performing MLP
as shown in Fig. 2.2. The change analysis step shows the shift in land
uses from one period to another. The transition potential determination
helps to find out the driving factors that impact the land use change.
The change prediction step helps to figure out the future land use of the
particular area. But in order to use the prediction, the model needs to be
validated first (Losiri et al., 2016).
3.2.1. Change analysis
The quantification of shift or change in different land uses is im-
portant. The shift in land use from 2005–2010 time period is displayed
in Fig. 3.3a while Fig. 3.3b shows the spatial distribution of land use
shift from 2010 to 2015 time period. The shift in land use from one
period to another shows difference in their nature of change. The
conversion between vegetation and vacant areas was the most dis-
tinctive change considering the study period. During 2005–2010 time
period, around 2032 ha of vegetated area had been converted to barren
land whereas only 431 ha of vacant land had been converted into ve-
getation in the same time period. Again nearly 807 ha of vegetation had
been converted to agricultural land within this five-year period
(Fig. 3.4).
It is clear that much of the deforestation occurred in the sanctuary
during the 2005–2010 time period. During the 2010–2015 period, the
conversion of land uses showed a different trend compared to the
previous period. There was a gain of around 1876 ha of vegetation from
the vacant areas which was quite promising. But still there was 1424 ha
of vacant areas that were converted from the vegetation (Figs. 3.3 and
3.4).
3.2.2. Driving factors for LULCC of CWS
ANN model takes the advantage of incorporating some driving
factors that could accelerate or determine the nature of change.
Cramer’s V values were the basis of judgment to accept or reject a driver
variable. Although Cramer’s V values do not guarantee the explanatory
444
Ecological Indicators 88 (2018) 439–453
power of the driver variables, a high value (generally greater than 0.1)
is considered as useful. The overall Cramer’s V value using the driving
variables are shown in Table 3.1
3.2.3. Transition probability analysis
The transition probability matrix calculated from the Markov chain
is one of the important inputs to delineate the future LULCC and is
further utilized in ANN model (Fig. S5). The transition probability
matrix of CWS from 2005 to 2010 was used as input to find the land use
change of 2015. Tables 3.2 and 3.3 show the transition probability
matrix for the periods 2005–2010 and 2010–2015 respectively.
3.2.4. Validation of ANN model
The actual land use area of 2015 was compared with the simulated
land use of 2015 based on the ANN model. Table 3.4 shows the chi-
square test for the model validation. The null hypothesis in this case
was the actual and simulated land use area of 2015 were same. How-
ever, this test doesn’t validate the spatial distribution of different land
use categories. The ocular variation checking was done to see the
overall similarity and the more sophisticated Kappa measures of asso-
ciation were finally applied to validate the model.
Since the tabulated chi-square value is greater than calculated one,
the null hypothesis was not rejected (Table 3.4). Thus, ANN model is a
good fit to predict the future LULC of CWS.
There is another two-stage validation. The ROC is an index of dis-
crimination accuracy of any specified quantity of change in land use
class. ROC value of 0.5 represents a random value (model not fit for
explaining real world situation) and any value greater than 0.5 has the
better predicting ability. ROC value of 1 explains all the variations.
The Kno index is the standard Kappa index of agreement which
gives the overall accuracy of the predicted image of 2015. The other
two indices, Klocation and Kquantity, validate the simulation’s ability
to predict location and quantity respectively. Kno, Klocation, and
Kquantity are equal to 1 when the simulation’s success rate is perfect,
and are equal to 0 when the simulation’s success rate is equivalent to
that due to chance (Pontius, 2000).
Quantity agreement and allocation disagreement approach is an-
other important aspect of model validation. Validation results show the
level of agreement of the simulated and actual land use maps regarding
the number of cells in each class and also the level of agreement of pair
of maps regarding the location of the cells in each class (Hyandye and
Martz, 2017).
The result of model validation depicts that overall agreement be-
tween actual and simulated maps of 2015 is 0.974 while the overall
simulation error is 0.026. The simulation error can be partitioned into
0.01 (error due to allocation/disagreement grid-cell) and 0.016 (error
due to quantity/disagreement quantity). Thus, the validation result
from quantity agreement-disagreement between the actual and simu-
lated land use of 2015 shows good predicting ability (Table 3.5).
The kappa indices (k-indices) were also used to assess the accuracy
of the simulation. The k-standard value of simulated and actual land use
map of 2015 is 0.98. The k-no value gives the overall accuracy of the
simulation. The overall accuracy for this case is 0.96 (Table 3.6). Si-
milar results were also found from the study of Daniel et al. (2016),
Hyandye and Martz (2017), Losiri et al. (2016). The ROC value for the
simulation is 0.9 which means that the explanatory power of the model
is not just by chance as ROC value is greater than 0.5. The ROC value is
the correlation between simulated and actual land use changes (Pontius
and Millones, 2011). Berberoğlu et al. (2016) found the ROC value for
the prediction of urban land use as 0.82 using the same method whereas
Lin et al. (2008) found it as 0.9. Thus the ROC value from this present
study was more or less the same. The comparison between the actual
and simulated land use map of CWS of 2015 is shown in Fig. 3.5
K. Islam et al.
3.3. Binary logistic regression for LULCC
The binary logistic regression model is applied to find out the in-
fluence of driving factors on the probability of getting vegetation cover.
The driving forces include elevation, slope, distance to natural water
source, distance to brickfield, distance to locality, distance to highways,
and distance to local road. Elevation, slope, and distance to natural
water source factors are the physical factor while the remaining factors
can be considered socio-economic in a broader sense.
In order to fit the binary logistic regression, it is ideal to check the
data first. One of the important considerations in this model is the effect
of correlation among independent variables. Tolerance and Variance
Inflation Factor (VIF) are two important indices for multi-collinearity. A
tolerance of less than 0.1 and a VIF of greater than 10 of the above-
mentioned indices indicates multi-collinearity problem (Pourghasemi,
2016). Presence of serial/auto-correlation could also be problematic as
the error terms of the explanatory variables are related due to this
445
Ecological Indicators 88 (2018) 439–453
Fig. 3.3. (a) Shift in land use in CWS from 2005 to 2010
time period.
problem. The Durbin-Watson (d.w) test statistic is widely used to detect
such problems. The d.w value of closer to 2 represents the absence of
auto-correlation while a value closer to 0 or 4 denotes the presence of
positive and negative auto-correlation respectively (Canary et al., 2016;
Patriche et al., 2016; Steyerberg et al., 2001). Table 3.7 shows the
tolerance, VIF, and d.w test statistics.
Since the d.w value is closer to 2, it can be inferred that the ex-
planatory variables are free from auto correlation. The VIF value of the
distance from brickfield is greater than 10. For this reason, this variable
is omitted while running the binary logistic regression model.
Out of the seven variables, 3 of them are found to be statistically
significant while predicting the probability of vegetation of the CWS.
Elevation, slope, distance from local road, and distance from highways
contributed positively to the predicted probability of vegetation of the
area. The other three variables contributed negatively. The odds ratio of
the driving factors indicates changes in odds of vegetation as the ex-
planatory factors vary with time and other factors. The odds ratio is
K. Islam et al.
estimated from the co-efficient of the explanatory variables. The value
of odds ratio ranges from 0 to ∞. The value between ‘0’ and ‘1’ indicates
that an increase in the value of explanatory variable leads to decrease in
probability of getting a vegetation pixel. On the other hand, a value
greater than ‘1’ represents that an increase in the value of explanatory
variable leads to the increase of possibility of getting a vegetation pixel.
Thus, increase in elevation and slope gives a higher probability of
getting vegetation pixel. Again, the greater the distance from the road
the greater is the chance of getting vegetation keeping the other vari-
ables as constant.
The interpretation can also be done as a unit increase in the ele-
vation, the log odds of getting vegetation pixel will be increased by
0.0142 units holding the remaining predictor variables fixed. Again,
holding the remaining predictor variables fixed, there is 0.0135% in-
crease in the likelihood of change to vegetative land cover pixel for
slope and 11.2579% increase for the case of road (Table 3.8).
The null hypothesis for this model was that the driver variables do
446
Ecological Indicators 88 (2018) 439–453
Fig. 3.3. (b) Shift in land use in CWS from 2010 to 2015
time period.
not have any impact on the probability of getting vegetation pixel.
However, from the analysis, there is sufficient evidence that elevation,
slope, and distance from road have statistically significant impact on
predicting vegetation cover. Thus, the null hypothesis must be rejected
for these variables only.
3.3.1. Validation of the binary logistic regression model
Validation of the model is an important step to determine the effi-
ciency and further usage of the model. The collected data was divided
into two parts. The first parts contained 80% of the total samples and
was used to train the model. The remaining 20% sample was used for
the validation of the model. A 2∗2 contingency table (Table S4) termed
as confusion matrix is developed using the validation sample data. ROC
and AUC (Area Under Curve) were used to determine the accuracy of
the model.
The AUC is found as 0.61 (Fig. 3.6) which also validates the accu-
racy of the model. Any AUC value greater than 0.5 is interpreted as
K. Islam et al. Ecological Indicators 88 (2018) 439–453

Table 3.5
Validation of ANN model through agreement/disagreement compo-
nents.

Agreement/disagreement Value (%)

Agreement chance 20
Agreement quantity 58.92
Agreement grid-cell 18.48
Disagreement grid-cell 1
Disagreement quantity 1.6

Table 3.6
Accuracy assessment of simulated land use map of
2015 from ANN model.

Types Value

Kno 0.9675
Klocation 0.9833
Fig. 3.4. LULC shift in CWS from 2005 to 2015 time period.
Kstandard 0.9578
ROC 0.91
Table 3.1
Cramer’s V value for driving variables.
better, and if it is 1 then it is absolutely a perfect fit.
Driving Factor Cramer’s V The landscape of the CWS is not homogenous rather it is highly
Slope 0.11 heterogeneous. Again this changes dynamically due to the intervention
Elevation 0.15 of human. Some other studies found an accuracy of greater than 70%
Distance from brickfield 0.12 (Patriche et al., 2016; Pourghasemi, 2016) using binary logistic re-
Distance from highways 0.11 gression model, but this study found it as 61%. Though this accuracy is
Distance from local road 0.07
not the best but still it can be useful considering the nature and extent
Distance from locality 0.08
Distance from natural water source 0.08 of the study area.

3.4. Predicted LULCC of CWS for 2020 and 2025

Table 3.2
Transition probability matrix for LULCC in CWS for 2005–2010 time period. The ANN model as discussed in the earlier section is a good fit for
2010
predicting the future land use change of the CWS. The prediction of
LULCC for the year 2020 and 2025 was the ultimate objective of this
2005 Vegetation Agricultural land Vacant areas Others study. Fig. 3.7a shows the predicted land use of 2020 whereas Fig. 3.7b
shows the land use of 2025.
Vegetation 0.5645 0.0983 0.2489 0.0883
About 50% of the total area still will be covered by vegetation in
Agricultural land 0.6172 0.1638 0.1735 0.0455
Vacant areas 0.4120 0.0737 0.4608 0.0536 2020, but it will be reduced to nearly 47% in 2025. There will be a net
Others 0.5098 0.0953 0.2446 0.1503 increase of 200 ha of barren land from 2015 to 2020 time period. Again,
there will be about 3297 ha of vacant areas in 2025 which will be an
increase of roughly 600 ha from the period 2020 (Table 3.9).
Table 3.3
Transitional probability matrix for LULCC in CWS for 2010–2015 time period.
4. Discussion
2015
From 2005 to 2010, CWS lost more than 2000 ha of vegetated land.
2010 Vegetation Agricultural land Vacant areas Others It may have happened because of the encroachment of about 50,000
Vegetation 0.6421 0.0947 0.2151 0.0480
Rohingya in early 2000 (Nath et al., 2016a). These people live solely on
Agricultural land 0.0491 0.5575 0.2491 0.1443 natural resources. Besides, due to rapid extraction of its resources forest
Vacant areas 0.4724 0.1432 0.3427 0.0417 became fragmented. A similar result was also reflected in the findings of
Others 0.2525 0.1226 0.2880 0.0852 Rahman et al. (2016). Rahman et al. (2016) showed that forest frag-
mentation was less in CWS in the year 2005, but during the 2005–2010
period the vegetation was found highly fragmented. Islam et al. (2016)
Table 3.4
analysed the vegetation change of CWS through vegetation index and
Validation of LULCC prediction of CWS based on ANN model.
they found that the NDVI (Normalized Difference Vegetation Index)
Chi-square test changed from 0.71 to 0.56 during 2005–2010 time period. Local forest
department also identified this problem in early 2000 and eventually, in
Land use class Simulated land use in Actual land use in (O − E)2/E
2003, co-management strategy was taken for the management of CWS
2015 (O) 2015 (E)
under Nishorgo Support Project (NSP), financially supported by the
Vegetation 6604.12 6636.78 0.1607219 United States Agency for International Development (USAID) (Tamima,
Agricultural land 1131.18 1159.65 0.698953 2016). Later on, Integrated Protected Area Co‐Management (IPAC)
Vacant areas 2460.02 2434.95 0.2581182
started their activities in 2010 and then followed by several projects i.e.
Others 627.45 591.39 2.1987582
Management of Natural Resources and Community Forestry in Chunati
Note: χ2 = Σ (O − E)2/E = 3.316; df = 3 and χ20.05 (3) = 7.815 Wildlife Sanctuary (MNRCF-Chunati), Climate Resilient Participatory
Afforestation and Reforestation Project (CRPARP) and Climate Resilient
Ecosystems and Livelihoods (CREL) (GIZ, 2015). The objectives of their

447
K. Islam et al. Ecological Indicators 88 (2018) 439–453

Fig. 3.5. Actual and simulated land use map of CWS-2015 based on ANN model.

Table 3.7
Multi-collinearity and auto-correlation problem diagnostics.

Multi-collinearity diagnostics Auto-correlation

diagnostics
Variables Tolerance VIF Durbin-Watson (d.w)
value

Elevation 0.576 1.736 1.962

Slope 0.784 1.275
Distance from locality 0.154 6.513
Distance from road 0.647 1.546
Distance from 0.379 2.642
highways
Distance from water 0.748 1.337
Distance from 0.094 10.627
brickfield

forest management were to involve people in forest management and to
reduce policeman-ship of forest officials. More specifically, their ob-
jective was to enhance environmental preservation and conservation, to
introduce rational forestland use, to increase public participation and
benefit from the forest, to create forests on marginal and private lands,
Fig. 3.6. AUC and ROC curve for the binary logistic regression model.
to setup institutional strengthening, to improve management practices,
to improve efficient resource utilization etc. Several plantation

Table 3.8
Results of binary logistic regression model.

Estimate Std. Error z value Pr(> |z|) eβ p(x)

Intercept −0.1820 0.0321 −5.675 0.0000*** 0.83364 0.45464

Elevation 0.0142 0.0008 16.736 0.0000*** 1.01429 0.50355
Slope 0.0135 0.0015 8.894 0.0000*** 1.01358 0.50337
Distance from locality −2.6417 2.6463 −0.998 0.3182 0.07124 0.0665
Distance from road 11.2579 4.0435 2.784 0.0054** 77488.7 0.99999
Distance from highways 2.4447 1.8186 1.344 0.1789 11.527 0.92017
Distance from water −1.4554 1.1430 −1.273 0.2029 0.23331 0.18918

Signif. Codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1.

448
K. Islam et al.
activities and alternative income generation initiatives gave back some
of the lost vegetation (around 600 ha) and occupied some barren land
with vegetation (around 300 ha). People living in and around CWS are
very poor, so it is unlikely that they will stop using the forest resources
that they need for their livelihoods until they have some viable alter-
natives. Now, after almost one-decade local people started to realise
that there were not enough resources for them inside the sanctuary
unless they use those resources more sustainably. From 2010 to 2015,
good fruit of their understanding increased vegetation percentage from
55% to 61% (Fig. 3.1).
The transition areas (Table S5 and Table S6) and probability matrix
generated from CA-Markov model also illustrates that the LULCC for the
time period 2010–2015 shows a better positive trend compared to the
previous 2005–2010 period. Nearly 4765 ha were to be persistent as
vegetation with a probability value of 0.64 at that time. However, it
was also found that the probability of entering into the vacant areas
from the vegetation category was 0.21. It is undoubtedly the most
449
Ecological Indicators 88 (2018) 439–453
Fig. 3.7. (a) Predicted LULC map of CWS-2020 based on
ANN model.
serious issue regarding the long-term sustenance of the ecosystem
health of CWS. Again, there was also transition from the vacant areas to
vegetation which was due to the plantation activities of forest depart-
ment through conservation projects and plantation from the revenue
budget of the Bangladeshi government (Figs. 3.3 and 3.4).
In order to predict future LULCC, CA-Markov model and ANN model
are attempted but CA-Markov model is found unfit to predict future
land use of CWS whereas ANN can successfully depict the future sce-
narios. From the prediction, it has been calculated that around 900 ha
vegetation will be lost in next five years and around 1500 ha may dis-
appear in next ten years. There might be many underlying causes of
further degradation. In this study, binary logistic regression is also used
as a useful tool to assess the actual impact of drivers. Elevation, slope
and distance from road are found statistically significant. In other
words, these variables have strong association with deforestation. The
probability of getting vegetation increases with elevation, slope per-
centage and distance from road. Road networks are significant
K. Islam et al. Ecological Indicators 88 (2018) 439–453

Fig. 3.7. (b) Predicted LULC map of CWS-2025 based on

ANN model.

Table 3.9 forest land use in Bangladesh. Our study shows that the distance from
Predicted area statistics of LULC of CWS for the period 2020 and 2025. the local road inside and adjacent to the sanctuary is a dominant factor
describing the probability of getting a site either regenerating as, or
Category Area in Area in Area in Area in
2020 (ha) percentage in 2025 (ha) percentage in
succeeding to forest. Some recent studies discussed the impact of urban
2020 2025 locality and water network upon multi-dimensional land use change
(Aburas et al., 2016; Hughes et al., 2016; Nourqolipour et al., 2014;
Vegetation 5732.12 53 5128.48 47 Peng et al., 2017). Elevation and slope are widely used physical factors
Agricultural land 1082.61 10 1096.65 10
Vacant areas 2601.61 24 3296.87 31
of LULCC. These two variables can significantly describe the nature of
Others 1406.43 13 1300.77 12 land use change of an area. This study agrees with Araya and Cabral
(2010) considering the elevation and slope as predictors of land cover
Total 10822.77 100 10822.77 100 condition and type.
Apart from these geo-spatial variables, socio-economic variables
may play a major role in this context. Population living inside the area
predictors of land use conversion as major human induced activities are
is increasing, and thus it became difficult to manage and fullfil every-
done using it (Green and Ahearn, 2016). Ku (2016) showed that the
body’s demand. Tamima (2016) revealed that local people did not
extended road networks can cause a rapid change in urban land use.
participate in the decision-making process of forest management, and
However, there is hardly any literature that describes the impact of the
they perceived co-management as a threat to their livelihoods.
road network (whether primary or secondary) upon the conversion of

450
K. Islam et al.
Moreover, mistrust and misunderstanding among different stake-
holders, political and ethical conflicts, lack of property rights of tribal
communities, and lack of accountability and transparency were iden-
tified as socio-economic drivers of LULCC of CWS.
With several initiatives conversion rate of forest to other land uses
were throttled but it could never be stopped. So, keeping the reality in
mind, Forest Department (FD) should continue plantation activity in
vacant lands. According to simulation, by 2020 the percentage of va-
cant area will be 24%, and by 2025 it will further increase and occupy
nearly 30% of total area (Table 3.9). These lands can also be brought
under participatory management system through Sustainable Use
Zoning. On the other hand, the cultivable land will get shrunk. Though
change may not be radical, it will still be insufficient compared to the
expected population in 2025. Now through co-management initiatives,
alternative livelihoods/income are being generated but not in a broad
scale. So, in the near future, keeping the prediction in mind, it is high
time to prioritise alternative income generation activities among the
local people. Another important aspect is to reduce the human-elephant
conflict. According to a more recent study, at least 173 people died, and
52 people were injured from 1989 to 2012. However, recently the ca-
sualty increased dramatically. There is no specific data record for CWS,
but in the whole country from 2005 to 2009 and from 2009 to 2012, 57
and 61 people were killed by elephants respectively. Among them,
children and old people comprised nearly 50% (Hossen, 2013). The
local FD office estimated crop damage by elephants, and it was around
5,000,000 BDT (about USD 59,000) per year in and around CWS. So, it
can be easily imagined in the future with less availability of food and
habitat of elephant, conflict between human and elephant will increase.
This study will help FD and other NGOs to make proper zoning i.e.
Ecosystem Management Zone, Habitat Management Zone, Sustainable
Management Zone, Elephant Movement Corridors described in NSP
(2005). According to proposed management zoning of CWS, around
5000 ha (NSP, 2005) will be managed as Habitat Management Zone for
wildlife species (elephant, serow, capped langur) and 3000 ha will be
managed as Elephant Movement Corridors by FD. This study will help
FD to make account of future threats while making plans for afore-
mentioned zonal management.
5. Limitations of the study and future research
Socio-economic drivers such as population growth, economic
growth etc. are some of the important factors associated with land use
change. However, due to the unavailability of such database, it is quite
difficult to incorporate such variables into land use modeling. The
driver variables used in this study are mostly prepared manually as
there is no valid database available. The elevation and slope are the
only two variables which are taken directly from the database with a
slight processing. But the remaining variables are developed from di-
gitization using the Google Earth platform. Thus there might be some
personal errors. The distance to the road variable is a dynamic one as
with the passage of time, new road network forms. To avoid the com-
plexity, this variable is considered as static.
Though this study successfully models the future land use change of
the CWS, it is best to consider a variety of landscape matrices like edge
density, total patch area, contagion, and patch density. Since the study
area is an important habitat for wildlife population, enabling better
assessment of trends for these landscape indices will help in better
understanding the impact of LULCC on the habitat.
6. Conclusion
The CWS has been undergoing a rapid degradation phase from the
beginning of 21st century. Due to manifold pressures, this area is ra-
pidly losing its environmental and biotic integrity. Conversion of ve-
getation to agriculture is a common phenomenon in this region due to
human intervention. However, the most alarming situation is the
451
Ecological Indicators 88 (2018) 439–453
increase of vacant areas inside the sanctuary. This study shows the
trend of land use change from 2005 to 2015. The conversion of one land
use to another indicates that vegetation cover is on the declining trend
whereas vacant land is on the rise. These opposing trends will un-
doubtedly affect the whole ecosystem of this important habitat. About
8258 ha of the total land area was covered by vegetation during 2005
which was about 76% of the total land. Due to rapid conversion to other
land uses, the vegetation cover in 2015 was only 6637 ha. Thus, the
area has already lost almost 15% of its vegetation cover within this
10 years’ period.
This study aims to understand the future land use scenarios of the
area through incorporating Markov model, CA-Markov, and ANN
model. The critical analysis shows that only ANN model is a good fit to
describe the land use of the area for 2020 and 2025. The overall ac-
curacy of prediction using ANN is over 90% which is of course sa-
tisfactory. The validation of the model shows that the disagreement
between the predicted and actual land use of 2015 was about 2%. The
ROC value of 0.91 indirectly illustrates the strength of the model to
predict the future.
The incorporation of the driver variables helps to identify potential
factors of land use conversion in the sanctuary. The analysis result of
binary logistic regression shows that elevation, slope, and distance to
road significantly affect the probability of getting the vegetation cover.
The accuracy of the binary logistic regression model was about 61%
which is quite satisfactory considering the dynamic and heterogeneous
nature of the landscape.
The simulation result shows that the vegetation coverage of the area
will be about 5732 ha in 2020 and 5128 ha in 2025. The percentage
distribution of land depicts that about 53% of the land will be under
vegetation during 2020 whereas this figure will be only 47% during
2025 if this trend continues further. The possession of vacant land ca-
tegory will be 24% in 2020, and it will increase further to reach 31% in
2025. Thus the simulation result figures out the future trend of LULCC.
The findings of this research will help policy makers, management
professionals, and government authorities managing this sanctuary to
take necessary action to stop the future adverse land use conversion.
Conflict of interest(s)
The methods and findings discussed in this research is adapted from
the outcome of the Master of Science thesis of the first author. This
work was conducted at Institute of Forestry and Environmental
Sciences, University of Chittagong during the year 2016–2017. The
authors declare that there is no conflict of interest regarding the pub-
lication of this article.
Funding information
No funding is received for this research.
Acknowledgements
The quality of this article is substantially improved from the com-
ments and suggestions of two anonymous reviewers. The authors are
grateful to them. It is impossible to conduct this research without the
freely available satellite imageries provided by NASA-GLCF and USGS;
ancillary data support provided by Google Earth platform; and eleva-
tion data as provided by SRTM. The authors thank the authority of
these organizations for that. Special thanks goes to Mr. Suvo Moy
Biswas (College of Environmental Science and Forestry, State
University of New York) for the proof reading of the first draft. We are
also grateful to Angela Martin for the time she has spent for the proof
reading and language editing of the final draft of this article.
K. Islam et al.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in the
online version, at http://dx.doi.org/10.1016/j.ecolind.2018.01.047.
References
Abuelaish, B., Olmedo, M.T.C., 2016. Scenario of land use and land cover change in the
Gaza Strip using remote sensing and GIS models. Arabian J. Geosci. 9 (4). http://dx.
doi.org/10.1007/s12517-015-2292-7.
Aburas, M.M., Ho, Y.M., Ramli, M.F., Ash’aari, Z.H., 2016. The simulation and prediction
of spatio-temporal urban growth trends using cellular automata models: a review. Int.
J. Appl. Earth Obs. Geoinf. 52, 380–389. http://dx.doi.org/10.1016/j.jag.2016.07.
007.
Ahmed, B., Ahmed, R., 2012. Modeling Urban land cover growth dynamics using multi-
temporal satellite images: a case study of Dhaka, Bangladesh. ISPRS Int. J. Geo-Inf. 1,
3–31. http://dx.doi.org/10.3390/ijgi1010003.
Al-sharif, A.A.A., Pradhan, B., 2013. Monitoring and predicting land use change in Tripoli
Metropolitan City using an integrated Markov chain and cellular automata models in
GIS. Arabian J. Geosci. 2009. http://dx.doi.org/10.1007/s12517-013-1119-7.
Araya, Y.H., Cabral, P., 2010. Analysis and modeling of urban land cover change in
Setúbal and Sesimbra Portugal. Remote Sens. 2 (6), 1549–1563. http://dx.doi.org/
10.3390/rs2061549.
Aspinall, R., 2004. Modeling land use change with generalized linear models – a multi-
model analysis of change between 1860 and 2000 in Gallatin Valley, Montana. J.
Environ. Manage. 72, 91–103. http://dx.doi.org/10.1016/j.jenvman.2004.02.009.
Azari, M., Tayyebi, A., Helbich, M., Reveshty, M.A., 2016. Integrating cellular automata,
artificial neural network, and fuzzy set theory to simulate threatened orchards: ap-
plication to Maragheh Iran. GISci. Remote Sens. 53 (2), 183–205. http://dx.doi.org/
10.1080/15481603.2015.1137111.
Beaumont, L.J., Duursma, D., 2012. Global projections of 21st century land-use changes
in regions adjacent to protected areas. PLoS ONE 7 (8), 1–8. http://dx.doi.org/10.
1371/journal.pone.0043714.
Berberoğlu, S., Akin, A., Clarke, K.C., 2016. Cellular automata modeling approaches to
forecast urban growth for adana, Turkey: a comparative approach. Landscape Urban
Plann. 153, 11–27. http://dx.doi.org/10.1016/j.landurbplan.2016.04.017.
Beuchle, R., Grecchi, R.C., Shimabukuro, Y.E., Seliger, R., Eva, H.D., Sano, E., Achard, F.,
2015. Land cover changes in the Brazilian Cerrado and Caatinga biomes from 1990 to
2010 based on a systematic remote sensing sampling approach. Appl. Geogr. 58,
116–127. http://dx.doi.org/10.1016/j.apgeog.2015.01.017.
Canary, J., Blizzard, L., Barry, R., 2016. A comparison of the Hosmer–Lemeshow,
Pigeon–Heyse, and Tsiatis goodness-of-fit tests for binary logistic regression under
two grouping methods. Communications In, 0918(March 2016). https://doi.org/10.
1080/03610918.2015.1017583.
Dahdouh-guebas, F., 2002. The use of remote sensing and Gis in the sustainable man-
agemnt of tropical coastal ecosystem. Environ. Dev. Sustainability 4, 93–112.
Dale, V.H., Lannom, K.O., Tharp, M.L., Hodges, D.G., Fogel, J., 2009. Effects of climate
change, land-use change, and invasive species on the ecology of the Cumberland
forests. Can. J. Forest Res. 39, 467–480. http://dx.doi.org/10.1139/X08-172.
Daniel, C.J., Frid, L., Sleeter, B.M., Fortin, M.J., Kriticos, D., 2016. State-and-transition
simulation models: a framework for forecasting landscape change. Methods Ecol.
Evol. 7 (11), 1413–1423. http://dx.doi.org/10.1111/2041-210X.12597.
Daye, D.D., Healey, J.R., 2015. Impacts of land-use change on sacred forests at the
landscape scale. Global Ecol. Conserv. 3, 349–358. http://dx.doi.org/10.1016/j.
gecco.2014.12.009.
Fang, S., Gertner, G., Wang, G., Anderson, A., 2006. The impact of misclassification in
land use maps in the prediction of landscape dynamics. Landscape Ecol. 21 (2),
233–242. http://dx.doi.org/10.1007/s10980-005-1051-7.
Gao, P., Niu, X., Wang, B., Zheng, Y., 2015. Land use changes and its driving forces in
hilly ecological restoration area based on gis and rs of northern china. Sci. Rep. 5
(October 2014), 1–11. http://dx.doi.org/10.1038/srep11038.
Ghebrezgabher, M.G., Yang, T., Yang, X., Wang, X., Khan, M., 2016. Extracting and
analyzing forest and woodland cover change in Eritrea based on landsat data using
supervised classification. Egypt. J. Remote Sens. Space Sci. 19 (1), 37–47. http://dx.
doi.org/10.1016/j.ejrs.2015.09.002.
GIZ. (2015). Final report of the final survey of indicators of “management of natural
resources and community forestry project (MNRCF-Chunati)”. In: Chunati Wildlife
Sanctuary (CWS), Chittagong, Bangladesh. Dhaka.
Goldewijk, K.K., 2001. Estimating global land use change over the past 300 years: the
HYDE database. Global Biogeochem. Cycles 15 (2), 417. http://dx.doi.org/10.1029/
1999GB001232.
Green, G.M., Ahearn, S.C., 2016. Modeling forest canopy trends with on-demand spatial
simulation. Int. J. Geogr. Inf. Sci. 30 (1), 61–73. http://dx.doi.org/10.1080/
13658816.2015.1066791.
Han, H., Yang, C., Song, J., 2015. Scenario simulation and the prediction of land use and
land cover change in Beijing China. Sustainability 7 (4), 4260–4279. http://dx.doi.
org/10.3390/su7044260.
Herold, M., Scepan, J., Clarke, K.C., 2002. The use of remote sensing and landscape
metrics to describe structures and changes in urban land uses. Environ. Plann. A 34
(8), 1443–1458. http://dx.doi.org/10.1068/a3496.
Hossen, A., 2013. Human-elephant Conflict in Bangladesh; Causes and Intensity of
Fatalities.
Hughes, S.J., Cabral, J.A., Bastos, R., Cortes, R., Vicente, J., Eitelberg, D., Santos, M.,
2016. A stochastic dynamic model to assess land use change scenarios on the
452
Ecological Indicators 88 (2018) 439–453
ecological status of fluvial water bodies under the water framework directive. Sci.
Total Environ. 565, 427–439. http://dx.doi.org/10.1016/j.scitotenv.2016.04.153.
Hyandye, C., Martz, L.W., 2017. A Markovian and cellular automata land-use change
predictive model of the Usangu Catchment. Int. J. Remote Sens. 38 (1), 64–81.
http://dx.doi.org/10.1080/01431161.2016.1259675.
Islam, K., 2015. Land Use/Land Cover Change Assessment of Chunati Wildlife Sanctuary
(CWS) Using Satellite Imagery. Institute of Forestry and Environmental Sciences,
University of Chittagong.
Islam, K., Jashimuddin, M., Nath, B., Nath, T.K., 2016. Quantitative assessment of land
cover change using landsat time series data: case of Chunati Wildlife Sanctuary
(CWS), Bangladesh. Int. J. Environ. Geoinf. 3 (2), 45–55.
Islam, K., Jashimuddin, M., Nath, B., Nath, T.K., 2017. Land use classification and change
detection by using multi-temporal remotely sensed imagery: the case of Chunati
Wildlife Sanctuary, Bangladesh. Egypt. J. Remote Sens. Space Sci. http://dx.doi.org/
10.1016/j.ejrs.2016.12.005.
Kantakumar, L.N., Neelamsetti, P., 2015. Multi-temporal land use classification using
hybrid approach. Egypt. J. Remote Sens. Space Sci. 18 (2), 289–295. http://dx.doi.
org/10.1016/j.ejrs.2015.09.003.
Kim, I., Le, Q.B., Park, S.J., Tenhunen, J., Koellner, T., 2014. Driving forces in arche-
typical land-use changes in a Mountainous Watershed in East Asia. Land 3 (3),
957–980. http://dx.doi.org/10.3390/land3030957.
Ku, C.-A., 2016. Incorporating spatial regression model into cellular automata for simu-
lating land use change. Appl. Geogr. 69, 1–9. http://dx.doi.org/10.1016/j.apgeog.
2016.02.005.
Lambin, E.F., Turner, B.L., Geist, H.J., Agbola, S.B., Angelsen, A., Folke, C., Veldkamp,
T.A., 2001. The causes of land-use and land-cover change: moving beyond the myths.
Global Environ. Change 11, 261–269.
Le Hégarat-Mascle, S., Ottlé, C., Guérin, C., 2005. Land cover change detection at coarse
spatial scales based on iterative estimation and previous state information. Remote
Sens. Environ. 95, 464–479. http://dx.doi.org/10.1016/j.rse.2005.01.011.
Lin, C., Wu, C.-C., Tsogt, K., Ouyang, Y.-C., Chang, C.-I., 2015. Effects of atmospheric
correction and pansharpening on LULC classification accuracy using WorldView-2
imagery. Inf. Process. Agric. 2 (1), 25–36. http://dx.doi.org/10.1016/j.inpa.2015.01.
003.
Lin, Y.P., Lin, Y.B., Wang, Y.T., Hong, N.M., 2008. Monitoring and predicting land-use
changes and the Hydrology of the Urbanized Paochiao Watershed in Taiwan using
remote sensing data, Urban Growth models and a hydrological model. Sensors 8 (2),
658–680. http://dx.doi.org/10.3390/s8020658.
López, E., Bocco, G., Mendoza, M., Duhau, E., 2001. Predicting land-cover and land-use
change in the urban fringe. Landscape Urban Plann. 55 (4), 271–285. http://dx.doi.
org/10.1016/S0169-2046(01)00160-8.
Losiri, C., Nagai, M., Ninsawat, S., Shrestha, R.P., 2016. Modeling urban expansion in
Bangkok Metropolitan region using demographic-economic data through cellular
Automata-Markov Chain and Multi-Layer Perceptron-Markov Chain models.
Sustainability (Switzerland) 8 (7). http://dx.doi.org/10.3390/su8070686.
Mallick, J., Kant, Y., Bharath, B.D., 2008. Estimation of land surface temperature over
Delhi using Landsat-7 ETM +. J. Ind. Geophys. Union 12 (3), 131–140.
Megahed, Y., Cabral, P., Silva, J., Caetano, M., 2015. Land cover mapping analysis and
Urban growth modeling using remote sensing techniques in greater Cairo
Region—Egypt. ISPRS Int. J. Geo-Inf. 4 (3), 1750–1769. http://dx.doi.org/10.3390/
ijgi4031750.
Miah, R., 2009. Non-Timber Forest Products and Co-Management : A Case Study of
Chunati Wildlife Sanctuary. In: Making Conservation Work: Linking Rural
Livelihoods and Protected Areas in Bangladesh (pp. 50–65).
Millington, J.D.A., Perry, G.L.W., Romero-Calcerrada, R., 2007. Regression techniques for
examining land use/cover change: a case study of a mediterranean landscape.
Ecosystems 10 (4), 562–578. http://dx.doi.org/10.1007/s10021-007-9020-4.
Nath, T.K., Jashimuddin, M., Inoue, M., 2016a. In: Community-Based Forest Management
(CBFM) in Bangladesh(Vol. 22). Springer International Publishing. http://dx.doi.org/
10.1007/978-3-319-42387-6.
Nath, T.K., Jashimuddin, M., Kamruzzaman, M., Mazumder, V., Hasan, M.K., Das, S.,
Dhali, P.K., 2016b. Phytosociological characteristics and diversity of trees in a co-
managed protected area of Bangladesh: implications for conservation. J. Sustainable
For. 9811 (September). http://dx.doi.org/10.1080/10549811.2016.1231615.
Nguyen, D.T., Iskandar, I., Ho, S., 2016. Land cover change and the CO2 stock in the
Palembang City, Indonesia: a study using remote sensing, GIS technique and
LUMENs. Egypt. J. Remote Sens. Space Sci. http://dx.doi.org/10.1016/j.ejrs.2016.
08.004.
Nourqolipour, R., Shariff, A.R.B.M., Balasundram, S.K., Ahmad, N.B., Sood, A.M., Buyong,
T., 2014. Predicting the effects of urban development on land transition and spatial
patterns of land use in Western Peninsular Malaysia. Appl. Spatial Anal. Policy 1–19.
http://dx.doi.org/10.1007/s12061-014-9128-9.
NSP, 2003. Secondary Data Collection for Pilot Protected Area: Chunati Wildlife
Sanctuary. International Resources Group (IRG).
NSP, 2004. Site-Level Field Appraisal for Protected Area Co-Management: Chunati
Wildlife Sanctuary. International Resources Group (IRG).
NSP, 2005. In: Assessment of Conservation Values of Chunati Wildlife Sanctuary and
Identification of Critical Conservation Areas for Inclusion in Nishorgo Support
Project. International Resources Group (IRG), pp. 1–36.
Patriche, C.V., Pirnau, R., Grozavu, A., Rosca, B., 2016. A comparative analysis of binary
logistic regression and analytical hierarchy process for landslide susceptibility as-
sessment in the Dobrov River Basin Romania. Pedosphere 26 (3), 335–350. http://dx.
doi.org/10.1016/S1002-0160(15)60047-9.
Peng, J., Zhao, M., Guo, X., Pan, Y., Liu, Y., 2017. Spatial-temporal dynamics and asso-
ciated driving forces of urban ecological land: a case study in Shenzhen City China.
Habitat Int. 60 (February), 81–90. http://dx.doi.org/10.1016/j.habitatint.2016.12.
K. Islam et al.
005.
Pontius, R.G., 2000. Quantification error versus location error in comparison of catego-
rical maps. Photogramm. Eng. Remote Sens.
Pontius, R.G., Millones, M., 2011. Death to Kappa: birth of quantity disagreement and
allocation disagreement for accuracy assessment. Int. J. Remote Sens. 32 (15),
4407–4429. http://dx.doi.org/10.1080/01431161.2011.552923.
Pontius, R.G., Parmentier, B., 2014. Recommendations for using the relative operating
characteristic (ROC). Landscape Ecol. 29 (3), 367–382. http://dx.doi.org/10.1007/
s10980-013-9984-8.
Pourghasemi, H.R., 2016. GIS-based forest fire susceptibility mapping in Iran: a com-
parison between evidential belief function and binary logistic regression models.
Scand. J. For. Res. 31 (1), 80–98. http://dx.doi.org/10.1080/02827581.2015.
1052750.
Raghuvanshi, T.K., Negassa, L., Kala, P.M., 2015. GIS based Grid overlay method versus
modeling approach – a comparative study for landslide hazard zonation (LHZ) in
Meta Robi District of West Showa Zone in Ethiopia. Egypt. J. Remote Sens. Space Sci.
18 (2), 235–250. http://dx.doi.org/10.1016/j.ejrs.2015.08.001.
Rahman, M.F., Jashimuddin, M., Islam, K., Nath, T.K., 2016. Land use change and forest
fragmentation analysis : a geoinformatics approach on Chunati Wildlife Sanctuary,
Bangladesh. J. Civ. Eng. Environ. Sci. 2, 20–29. https://doi.org/10.17352/2455-
488X.000010.
Ramachandra, T.V., Setturu, B., Rajan, K.S., Subash Chandran, M.D., 2016. Stimulus of
developmental projects to landscape dynamics in Uttara Kannada, Central Western
Ghats. Egypt. J. Remote Sens. Space Sci. 19 (2), 175–193. http://dx.doi.org/10.
1016/j.ejrs.2016.09.001.
Rizk Hegazy, I., Kaloop, M.R., 2015. Monitoring urban growth and land use change de-
tection with GIS and remote sensing techniques in Daqahlia governorate Egypt. Int. J.
Sustainable Built Environ. 4, 117–124. http://dx.doi.org/10.1016/j.ijsbe.2015.02.
005.
Schneider, L.C., Pontius, R.G., 2001. Modeling land-use change in the Ipswich watershed,
Massachusetts, USA. Agric., Ecosyst. Environ. 85 (1–3), 83–94. http://dx.doi.org/10.
453
Ecological Indicators 88 (2018) 439–453
1016/S0167-8809(01)00189-X.
Sinha, S., Sharma, L.K., Nathawat, M.S., 2015. Improved Land-use/Land-cover classifi-
cation of semi-arid deciduous forest landscape using thermal remote sensing. Egypt.
J. Remote Sens. Space Sci. 18 (2), 217–233. http://dx.doi.org/10.1016/j.ejrs.2015.
09.005.
Sivakumar, V., (2014). Urban Mapping and Growth Prediction using Remote Sensing and
GIS Techniques, Pune, India. In: ISPRS – International Archives of the
Photogrammetry, Remote Sensing and Spatial Information Sciences, XL-
8(December), 967–970. https://doi.org/10.5194/isprsarchives-XL-8-967-2014.
Song, C., Woodcock, C.E., Seto, K.C., Lenney, M.P., Macomber, S.A., 2001. Classification
and change detection using landsat TM data : when and how to correct atmospheric
effects ? Remote Sens. 4257 (00).
Steyerberg, E.W., Harrell, F.E., Borsboom, G.J.J.M., Eijkemans, M.J.C., Vergouwe, Y.,
Habbema, J.D.F., 2001. Internal validation of predictive models: efficiency of some
procedures for logistic regression analysis. J. Clin. Epidemiol. 54 (8), 774–781.
http://dx.doi.org/10.1016/S0895-4356(01)00341-9.
Subedi, P., Subedi, K., Thapa, B., 2013. Application of a Hybrid Cellular Automaton –
Markov (CA-Markov) model in land-use change prediction: a case study of saddle
creek drainage basin Florida. Appl. Ecol. Environ. Sci. 1 (6), 126–132. https://doi.
org/10.12691/aees-1-6-5.
Tamima, U., 2016. Performance evaluation of forest co-management : a case study of
Chunati Wildlife Sanctuary Bangladesh. J. For. Res. 27 (4), 853–861. http://dx.doi.
org/10.1007/s11676-015-0198-6.
Zeng, Y.N., WU, G.P., ZHAN, F.B., ZHANG, H.H., 2008. Modeling spatial land use pattern
using autologistic regression. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, XXXVII, 115–118.
Zhan, X., Sohlberg, R.A., Townshend, J.R.G., D, C., Carroll, M.L., Eastman, J.C., Hansen,
M.C., 2002. Detection of land cover changes using MODIS 250 m data – Google
zoeken. Remote Sens. Environ. 83, 336–350. http://dx.doi.org/10.1016/S0034-
4257(02)00081-0.