Environmental Earth Sciences (2019) 78:546

https://doi.org/10.1007/s12665-019-8557-9

ORIGINAL ARTICLE

Change detection and prediction of urban land use changes

by CA–Markov model (case study: Talesh County)
Hamide Aliani1 · Maryam Malmir1 · Mona Sourodi1 · Sasan Babaie Kafaky2

Received: 5 October 2018 / Accepted: 17 August 2019 / Published online: 26 August 2019
© Springer-Verlag GmbH Germany, part of Springer Nature 2019

Abstract
Earth’s surface has continued to change due to human activities and natural reasons. Land use and land cover (LULC) change
is one of the significant issues which has considerable impacts on environment and its processes. Access to precise and
up-to-date data of LULC through satellite images provides a great opportunity to detect, monitor and model a prediction of
the future changes. The purpose of this research is to monitor and study land use changes, especially in urban land, during
the past years and the possibility of predicting future changes by CA–Markov in Talesh County. In this research, satellite
imagery of ETM 2000, LISS III 2007 and OLI-TIRS 2014 have been used. Supervised classification of images is done
using the maximum likelihood method. Then the accuracy of the generated land use maps was evaluated using the overall
accuracy and kappa coefficients. The results of the evaluation showed that land use maps from 2000, 2007 and 2014 had
kappa coefficients equal to 0.86, 0.85 and 0.89, respectively, and an overall accuracy of 91%, 90%, and 93%. The land use
map for 2028 has been predicted by the CA–Markov model. The results of the model forecast indicate a significant increase
in the size of finished and urban areas by 29/83% and a reduction of the area of agricultural land, forests, and wastelands,
respectively, to the 3/12, 0.59, and 0.48% over the next 14 years in the area under study. The model also showed that the
future development of the city would occur linearly and mainly around the city of Hashtpar, especially on the western and
eastern borders of the city.

Keywords Change detection · Land use · Remote sensing · Urban development · CA–Markov model · Talesh

Introduction the urban ecosystems (Bounoua et al. 2018; Pickard et al.

2017; lu et al. 2019; Abdullahi and Pradhan 2018a, b). This
Researches indicate that man-made changes in land use and issue is globally considerable due to sustainable develop-
land cover (LULC) such as deforestation, cultivated lands ment and it can cause wildlife inhabitant devastation, acid
devastation and urbanization have irreversible impacts on precipitation, desertification, climate change, lower diversity
and many other long-term impacts on the different ecosys-
tems (Rindfuss et al. 2004). Detection of these changes is
* Hamide Aliani critical to monitor the future dynamics. The combination
[email protected]
of Markov chain and Cellular Automata (CA) provides the
Maryam Malmir greatest opportunity to predict and project the changes.
[email protected]
The Markov chain model is widely known for its ability
Mona Sourodi to predict anticipated changes in LULC, urban growth and
[email protected]
urbanization, and weather forecasting (Muller and Middle-
Sasan Babaie Kafaky ton 1994; Zubair 2006; Chang and Chang 2006; Kamusoko
[email protected]
et al. 2009; Subedi et al. 2013; Fathizad et al. 2015; Wang
1
Department of Environmental Science, Faculty of Natural et al. 2014; Roose and Hietal 2018). Markov chain is only
Resources and Environment, Science and Research Branch, able to detect the transformation of LULC classes, but it
Islamic Azad University, Tehran, Iran cannot show the quantity of the changes (Sang et al. 2011).
2
Department of Forestry, Faculty of Natural Resources CA–Markov model is a technique which is raster-based.
and Environment, Science and Research Branch, Islamic Pixel condition shows the LULC class that pixel belongs
Azad University, Tehran, Iran

13
Vol.:(0123456789)
546 Page 2 of 12 Environmental Earth Sciences (2019) 78:546

to and changes in each class depending on the current con- Tajbakhsh et al. (2016) did a study for the analysis and simu-
dition of the pixel and its neighbors (Jabbari and Ahmadi lation of land use changes in Mashhad by Markov–Cellular
2014). The integrated approach in using CA–Markov and Automata model. A 20-year period was considered for the
Markov chain can add spatial neighborhood features to ran- predicted changes. The results show that, during the coming
dom process of Markov chain and predict the future LULC years, urban development on the northern side of the city
changes (Zareh et al. 2012). By predicting land use changes, will be more intense, especially in the north-west region.
we can determine the extent of development and degrada- In another case study, LULC changes in the Usangu Catch-
tion of resources and lead them in the appropriate directions ment (Tanzania) predicted employing a CA–Markov. The
(Brown et al. 2000; Jenerette and Wu 2001; Hathout 2002). simulation of the changes was based on the dynamics of
There are many studies and researches indicating that the the years 2000, 2006 and 2013, respectively. The validation
CA–Markov model, with efficient coordination with GIS procedure followed using simulated and observed LULC of
and remote sensing, can create an appropriate approach in the year 2013. Furthermore, the results revealed 0.6776 and
dynamic modeling of spatial and temporal changes in LULC 0.9125 for κ-standard and overall agreements of the model,
(Li and Reynolds 1997; Myint and Wang 2006; Guan et al. respectively (Hyandye and Martz 2017). Hamad et al. (2018)
2011; Riccioli et al. 2013; Roose and Hietal 2018). In this studied the LULC changes in Halgurd-Sakran Core Zone
model, the Markov chain process controls time changes (Iraq) using Landsat images dated 1993, 1998, 2003, 2008
between classes of land use, based on the possibility of and 2017. These images were classified into bare surface,
conversion (López et al. 2001; Guan et al. 2011), while the pasture, cultivated area and forest applying Random Forest
spatial changes are controlled by the local law determined (RF) method. Moreover, Iraq under siege and Iraq after siege
by the CA spatial filter or suitability maps (Wu et al. 2002; were considered as two main scenarios to project future
He et al. 2006). In the meantime, GIS is used to define the changes to the year 2023 conducting (CA)–Markov chain
initial conditions, determine model parameters and calculate model. Results proved the reliability of these methods for
probabilities and determine neighboring rules (Batty et al. change detection analysis.
1999; Guan et al. 2011). The aim of this study is to evaluate the changes in land
By examining the available resources in this area, it is use of the Talesh County with the help of satellite imagery
observed that the CA–Markov chain model, in combination for the years 2000 to 2014, and modeling and predicting
with remote-sensing images, is used in different parts of the the growth of urban land with the help of remote-sensing
world to successfully model use changes and land cover in technology up to 2028.
urban and rural environments on large-scales (Halmy et al.
2015; Alhamdan et al. 2017; Chen et al. 2013a, b; Osgouei
and Kaya 2017; McCarthy et al. 2018). Modeling land use Materials and methods
changes of Saga, in Japan, for the years 2015–2042 was done
using the CA Markov chain models. The results of this study Study area
indicated an increase in the area of urban land and loss of
natural and agricultural land (Guan et al. 2011). Land use Talesh County, with an area of 2155.91 km 2, is located
changes in the Changping District, in Beijing, have been in the north-west of the Gilan Province (Fig. 1). It
predicted by Landsat satellite images for 2000 (Wang et al. extends between longitudes 48°35′–49°3′E and latitudes
2012). In a study on the effects of land use, changes in the 37°32′–48°16′N. The county has both lowlands and moun-
past and those likely in future have been investigated in a tainous geographical areas. In the flat area, a narrow strip
protected area in the south of Spain by the CA Markov chain of coastal area and in mountain regions mountain ranges,
model. The results of the study have shown agricultural forests and rangelands are seen in different heights. The
development and land use in urban areas and release of the minimum height of the city is 26 m and maximum height
past land uses (Piquer-Rodríguez et al. 2012). Tudun-Wada is 3200 m.a.s.l. The slope of the land varies between zero
et al. (2014) analyzed changes in the forest cover in Namibia, and 70%.
in Nigeria, during 1986–2010 and gave a projection for the
next 21 years by GIS and RS models and the Markov chain
model. The results show that the forest area in the region Materials
is declining due to human activities such as illegal logging
and agriculture. Megahed et al. (2015) have investigated In this study, satellite images of Landsat ETM and TIRS-
the urban growth in the Greater Cairo region (GCR) by the OLI related to 2000 and 2014, respectively, and the LISS
land change modeler. After validation, the model results III images of IRS satellite related to 2007 of Talesh County
show that 14% of the vegetation and 4% of the deserts exist- are used to classify and study land use changes. Specifi-
ing in 2014 will be transformed into urban areas in 2025. cations of satellite images used are given in Table 1. In

13
Environmental Earth Sciences (2019) 78:546 Page 3 of 12 546

Fig. 1  The situation of the area in the province and country

Table 1  Specifications of satellite images of Talesh County from Methodology

2000 to 2014
Type of satellite Sensor Date of data Spatial resolution (m) Pre‑processing of satellite images
Landsat 7 ETM 2000/7/23 Band = 30, pan band = 15
After the preparation of the satellite images, radiometric and
IRS LISS III 2007/8/20 Band = 23.5, pan
band = 6
geometric corrections were done to prepare data for process-
Landsat 8 OLI-TIRS 2014/8/9 Band = 30, pan band = 15
ing and extraction of useful information. In this research,
involving 47 ground control points with proper distribution,
topographic maps at 1: 25,000 and GPS, digital elevation
models (DEM) and sensor orbital data, geometric correc-
addition, Erdas Imagine 14 and Arc GIS 10.3 software are tion of images was performed with ERDAS Software. The
used for the initial preparation, preprocessing and clas- root mean squares error (RMSE) obtained for the sensors
sification of satellite images and IDRISI Selva 17.0 for ETM, LISS III and OLI-TIRS were estimated as 0.51, 0.52,
determining the land use changes and the prediction of and 0.55 pixels, respectively, which is an acceptable error.
changes in the upcoming years. It should be mentioned that first the images of OLI-TIRS

13
546 Page 4 of 12 Environmental Earth Sciences (2019) 78:546

and then the ETM and LISS III images were corrected using and band 3 (red) based on Eq. (1). In the IRS images (LISS
the OLI-TIRS image. In addition, geometric correction was III 2007), the index was calculated through band 3 as infra-
performed to the level of orthorectification. red and band 2 as red, respectively.
For radiometric correction, data window was selected in
dimensions of 100 rows in 300 columns in areas with very
NDVI = (NIR − RED)∕(NIR + RED) (1)
low values and zero and mean spectral values of each row NIR shows a spectral reflectance in the near-infrared and
were calculated. Considering that the numerical values of red bands. Then, the classification was made of each of the
the rows’ mean was less than one unit, it may account for images with a combination of the most appropriate set of
the radiometric errors and shrinkage in images. Moreover, to bands.
increase the accuracy of the extracted data and better detec-
tion of various phenomena, the main bands of the images
Processing of satellite images
and new bands of the generated images were also analyzed
by spectral transformations (PCA, band ratios) and the Nor-
To classify land cover, five land use classes including for-
malized Difference Vegetation Index (NDVI), and the best
est, built-up area, land with poor coverage (rangeland and
bands were selected for the best results using the Optimum
bare land), agricultural land, and water areas are taken into
Index Factor (OIF).
account.
For ETM and OLI-TIRS images, bands 3 and 4, original
Then, with field sampling and the use of GPS, the train-
image PCA1, and a spectral ratio of 4:3, spectral bands were
ing samples for every five classes were acquired. In all, 600
selected, and for the LISS III image, bands 2 and 3, original
samples were collected from the study area so that at least 10
band PCA1, and spectral ratio of band 3–2 were considered.
polygons with an area of 5 ha for each class could be deter-
The NDVI index for Landsat images of ETM 2000 and
mined. To classify images using the taken 70% of training
OLI-TIRS 2014 was calculated using the band 4 (infrared)

Fig. 2  Land use map of Talesh in 2000

13
Environmental Earth Sciences (2019) 78:546 Page 5 of 12 546

samples, supervised classification was performed using the Prediction of land use changes
maximum likelihood and land use map extracted for each of
the images. The maps emerging from this classification are In this research, to predict and simulate in order to improve
shown in Fig. 2. urban development in the Talesh County, the Markov chain
model and CA filter, which is called CA–Markov, were
Accuracy of classification used. Based on the conditional probability and the predic-
tion formula for land use change in the Markov model is
The Kappa coefficient, which shows the accuracy of a clas- calculated using Eq. (2) (Sang et al. 2011):
sification compared to a random classification, is a value
between zero and one with zero indicating total acciden-
S(t + 1) = pij × s(t). (2)
tal classification and one indicating absolutely accurate In this regard S (t) and S (t + 1) are system states at time
classification (Lillesand et al. 2014; Shawul and Chakma t and t +1 and pij is the probability transition matrix in a
2019). It is capable of eliminating random adaptation state that is calculated by Eq. 3:
which is possible in overall accuracy. This coefficient is
⎡ p11 p12 … p1n ⎤
based on observed adaptation (P0) and expected adapta- ⎢ p p … p2n ⎥
tion (Pc). The following formula indicates K index: Pij = ⎢ 21 22 ⎥, (3)
⎢ ………… ⎥
P0 − Pc ⎣ pn1 pn2 … pnn ⎦
K= .
1 − Pc
∑N
In order to assess the accuracy of LULC maps driven (0 ≤ Pij < 1 and Pij = 1, (i, j = 1, 2, … , n).
from ETM 2000, LISS III 2007 and OLI 2014, ground j=1

truth map generated from 70 ortho aerial photos in scale of In the Cellular Automata (CA), model space is defined
1:20,000 as well as 30% ground control points were used. as a network in which each home is called a cell. The
Accuracy assessment is a crucial process in remote amount of each cell is determined by the values of the
sensing and GIS to evaluate the results and includes over- neighboring cells and the cell itself (Aslami et al. 2015).
all accuracy, producer’s accuracy, user’s accuracy and The model of automated cells can be obtained from
Kappa coefficient. In this study, the following formula Eq. 4 (Sang et al. 2011):
was used to calculate overall accuracy for generated maps
(Halimi et al. 2018): S(t, t + 1) = f (S(t), N). (4)
In this regard, S is a finite and distinct set of cellular
Overall accuracy = total number of correct samples
states, N cell fields, t and t + 1 represent different times
× 100% total number of samples. and f the rules of transferring cell states in the local space
The overall classification accuracy indicates the amount (Abdullahi and Pradhan 2018a, b).
of credit of classification, which should be more than 85% The basis of this model is the use and evaluation of land
in land use maps derived from satellite imagery. Follow- use layers of previous years, and then the prediction of the
ing the completion of the classification, the evaluation of spatial distribution of applications for the future by using
classification accuracy was dome. the IDRISI Selva software. This model consists of two
Afterwards, Producer’s accuracy was carried out to main steps: (1) calculation of the conversion probabili-
understand how well the classification procedure has been ties (conversion probability matrix, transformation matrix
done for each and every class. The Producer’s accuracy is and conditional probability layers) using the Markov chain
based on the number of correct pixels in each class and analysis. (2) Allocation of the simulated urban area cov-
the total number of pixels according to Ref. data. This erage (built-up land) based on the CA site operator and
accuracy was calculated regarding the following formula: multi-criteria evaluation (MCE).
In the first stage, land use maps of 2000, 2007, and
Commission error =
off diagonal row elements
, 2014 were extracted from satellite imagery using RS tech-
total of row nology; and the layers for the years 2000 and 2014 were
selected to import into the model and the calculate trans-
Comission error =
off diagonal column elements
, formation matrices and conversion probabilities by using
total of column the Markov chain analysis. In the next step, the conversion
probability matrix created a set of conditional probabil-
Producer’s accuracy(% ) = 100% − error of comission (% ), ity data for the five main user inputs (constructed land,
agriculture, wasteland, jungle, and aquatic areas) for the
User accuracy (%) = 100% − error of comission (%).

13
546 Page 6 of 12 Environmental Earth Sciences (2019) 78:546

period 2000–2014. This probable data set is, in fact, the 1. How similar are the model outcome and truth LULC
result of a predicted calculation of the two old and new map regarding the number of cells in each class?
land use layers (2000 and 2014). 2. How similar are the model outcome and truth LULC
In the last step, in order to simulate the urban land use map regarding the distribution of cells in each class?
map of 2028, the land use map of 2000 as the base map, the
2000–2014 probability matrix of conversion, and the condi- Moreover, validation methods follow two steps. First of
tional probability data were combined by the use of the CA all, accuracy assessment of general features of the system
locator function in the IDRISI software based on the Markov and LULC patterns; and then, mining more spatial details
chain analysis and multi-criteria evaluation. Owing to the and similarity assessment of the role model in a pixel scale
time interval between 2014 and 2028, 14 times repeated (Barredo and Engelen 2010) in this study, to assess the reli-
assignments were made to the CA model. Finally, at the end ability of the CA–Markov outcomes.
of the repetition, the new land use map was produced by First the actual land use maps extracted from the sat-
overlapping all the results obtained from the previous stages. ellite imagery of 2000 and 2007 were introduced into the
CA model; and, then, based on these data, the process of
Validation of the model simulating land use changes was done for 2014. Then, in
order to validate the CA–Markov model, the land use map
The aim of this procedure is to check the reliability of the of 2014, obtained from the Markov model, was compared
model outcomes. Generally, model validation is a statistical to the actual land use map for this year, based on the Kappa
analysis which covers the responses to the following ques- coefficient.
tions (Barredo et al. 2005):

Fig. 3  Land use map of Talesh in 2007

13
Environmental Earth Sciences (2019) 78:546 Page 7 of 12 546

Fig. 4  Land use map of Talesh in 2014

Table 2  The area and coverage Land use class 2000 2007 2014
of different classes of land use
in 2000–2014 Area (ha) Percentage Area (ha) Percentage Area (ha) Percentage

Forest 134932.21 62.59 134672.5 62.47 134086.37 62.19

Built-up land 4154.62 1.93 4876.42 2.26 6022.28 2.79
Agriculture 25389.18 11.77 24956.31 11.58 24638.19 11.43
Water 188.65 0.087 193.18 0.089 198.41 0.092
Other land 50926.34 23.62 50892.59 23.6 50645.75 23.49

Results and discussion

The results of this study indicate that there are wide varia-
tions in the built-up and urban areas of the studied region
with the highest amount of land degradation and transfor-
mation occurring in the years 2007 to 2014. Figures 2, 3 4
show the land use maps for the three periods 2000, 2007, and
2014. The area and coverage of different classes of land use
are presented in Table 2.
The results of the study indicate that the area of forests
Fig. 5  Coverage percentage of each land class in the years 2000–2014 decreased from 134,932.21 hectares in 2000 to 134672.5

13
546 Page 8 of 12 Environmental Earth Sciences (2019) 78:546

Fig. 6  Actual and simulated land use maps for 2014

hectares in 2007, or 259.71 hectares (0.19%). In addition, In order to validate the land use map of 2014 obtained
the area of agricultural land decreased about 432.87 hectares from the Markov model, a comparison was made with
(1.7%) during this time. the actual land use map for this year, based on the Kappa
Results indicate that the area of the built land has grown coefficient.
steadily over time, which means that, from about 4154.62 The overall simulation success rate of the CA–Markov
hectares, in 2000, it increased to about 4876.42 hectares in model for 2014 is estimated as 82.2%. Actual and simulated
2007, showing an increase of 721.8 hectares (17.37 percent). land use maps for 2014 are shown in Fig. 6 (Table 3).
The highest increase occurred in the area of land built dur- The area of each of the actual and simulated land use
ing the second period between 2007 and 2014, registering classes in 2014 is presented in Table 4. The table shows that
an increase from about 4876.42 hectares in 2007 to about each land use site in the region has an error of less than 5%.
6022.28 hectares by 2014, or 1145.86 hectares (23.5%). This figure demonstrates the usefulness and applicability of
However, the level of forest land decreased about 586.13 the Markov model in predicting use variations. In this study,
hectares from 2007 to 2014 (0.43%). The area under agri- the best simulations have been made about the forest use
culture, too, decreased by about 318.12 hectares (27.1%) class. Therefore, the real forest area is 134086.37 hectares
during 2007 to 2014. The coverage percentage of each land and its simulated class is 134160.45 hectares; and its error
use class in the years 2000–2014 is presented in Fig. 5. rate is very small and estimated at 0.05%. This is while the
Statistic parameters of accuracy assessment such as agricultural class simulation is relatively poor and the error
Kappa coefficient and overall accuracy for each LULC maps rate is 4%.
driven from ETM 2000, LISS III 2007 and OLI 2014 were Based on the success of the Markov model in the simu-
calculated. lation for 2014 and the use of the 2000 User Base Map,
the Probability Matrix 2000–2014 and the simulated map

13
Environmental Earth Sciences (2019) 78:546 Page 9 of 12 546

Table 3  Accuracy assessment of the driven LULC maps The ability to use the CA–Markov model and obtain
Class User Producer Kappa coef- Overall high-accuracy outputs has been proven in numerous studies.
name accu- accu- ficient % accuracy % One of these studies is a research by Al-sharif and Pradhan
racy % racy % (2014) using the Markov chain and Cellular Automata in
order to simulate urban land use changes in the Tripoli met-
2000 Forest 84 81.5 86 91
ropolitan area of Libya between 1984 and 2010 to predict
Built-up 88 85.4
land their spatial patterns for the years 2020 and 2025. The results
Agricul- 86 83.4 indicate good performance and approval of the model and
ture that the studied area has been growing very rapidly in the
Water 87 84.4 last decade. This rapid urban development will significantly
Other land 83 80.5 reduce agricultural land. Again, another study (AL-Bakri
2007 Forest 82 79.5 85 90 et al. 2013) modeled land use changes in Oman for 2043 by
Built-up 86 83.4 Landsat satellite imagery and the Markov model. The results
land of this study with high accuracy indicate the transformation
Agricul- 85 82.5 of natural lands to residential plots. This study emphasized
ture
the modeling and evaluation of land use change as an impor-
Water 84 81.5
tant tool for management plans and land use policies. Yang
Other land 86 83.4
et al. (2014) used a combination of optimization of the ant
2014 Forest 87 84.4 89 93
colony, the Markov chain and the automatic cells in mod-
Built-up 89 86.3
eling the time and place of the changes in the city of Bei-
land
jing and evaluated it as a desirable model for computing the
Agricul- 87 84.4
ture quantity and distribution of use changes. Araya and Cabral
Water 92 89.2 (2010) analyzed and modeled urban land use changes using
Other land 91 88.3 the CA–Markov model for the period 1990–2006 in Setúbal
2028 Forest 81 78.6 82 87 and Sesimbra, in Portugal. The predictive power of the
Built-up 82 79.5 model was successfully validated using Kappa changes. In
land these studies and other similar studies, in addition to deter-
Agricul- 79 76.6 mining the status of land use changes at the desired time
ture using the Markov model, the efficiency of this model has
Water 84 81.5 been confirmed (Sohl and Claggett 2013; Mas et al. 2014).
Other land 85 82.5

Conclusion
Table 4  The area of actual and simulated land use in 2014
Land use class Actual 2014 Simulated 2014 Variation
In order to reduce the adverse effects of urban growth on the
percentage environment and maintain optimal ecosystem performance,
time and space patterns of land cover, land use change, and
Forest 134086.37 134160.45 0.05
the factors influencing these changes are significant in the
Built-up land 6022.28 6119.6 2
relative development of economic, social, and environmental
Agriculture 24638.19 25636.43 2.3
policies (Serra et al. 2008; Dewan and Yamaguchi 2009).
Water 198.41 201.07 4
The results of this research show that between 2000 and
Other land 50645.75 49473.45 1.3
2007 forest land area decreased by 0.19%, agricultural land
by 1.7%, and wasteland by 0.07%, while the area of the built
land increased by 17.37%. The highest increase occurred in
of 2014, the simulation of land use was performed for 2028, the area of built land during the second half of the period
which is shown in Fig. 7. The area and percentage of land between 2007 and 2014, or (23.5%) by 2014. However, the
use in Talesh County in 2028 are shown in Table 5. level of forest land (0.44%), agricultural land (1.27%), and
According to the model forecast, the forest area will other lands (0.49%) decreased during the years 2007–2014.
decrease by about 0.59% in 2028 and agricultural land will Then, using urban land use maps obtained from satellite
fall by 13.3% by that year. The size of wasteland will also images (2000, 2007 and 2014), the simulation of urban
decrease by 0.48% in 2028, while the size of the built land development in Talesh County was successfully done
will increase significantly by about 29.83% in the study area. using a combination of the CA–Markov and GIS model.
This model has been validated using real 2014 data and

13
546 Page 10 of 12 Environmental Earth Sciences (2019) 78:546

Fig. 7  Predicted land use map of Talesh in 2028

Table 5  Area and percentage of land use cover of Talesh in 2028 land degradation in the future. The results of the model’s
prediction also showed that the city’s future development
Land use Area (ha) Percentage
would be mainly around the city of Hashtpar, especially on
Forest 133295.36 61.83 the western and eastern borders of the city. The location
Built-up land 7818.9 3.62 of the city of Talesh in the transit axis of Ardebil–Tehran
Agriculture 23868.02 11.07 also led to dwelling on the main axis and linear develop-
Water 207.9 0.1 ment of the city.
Other land 50400.82 23.38 The use of the Markov model in predicting changes and
preparing a map for forecasting changes in various applica-
tions is considered to be important aspects of this study.
its reliability has been satisfactory; therefore, the overall Considering the predictions to improve the strengths of this
simulation success rate of this model is estimated to be prediction and mitigate the negative effects of these changes
82.2%. Based on this validation, the CA–Markov model will lead to sustainable development and conservation of
has been used to simulate land use change by 2028. The natural areas, while paying attention to the needs of the resi-
results of the changes in the region indicate an increase dents of the study area.
in the size of built-up, urban, and aquatic areas and a Also, the results prove that the integrated approach of
reduction in forests, wasteland, and agriculture land over remote sensing and GIS to run models to assess spatial–tem-
the next 14 years in the study area. The process of these poral changes is greatly beneficial to detect the pattern and
changes in the region shows a continuous decline in for- percentages.
est land and an increase in urban areas, leading to severe

13
Environmental Earth Sciences (2019) 78:546 Page 11 of 12 546

References CA–Markov algorithms to monitor and predict future change.

Environ Earth Sci. https​://doi.org/10.1007/s1266​5-018-7392-8
Fathizad H, Rostami N, Faramarzi M (2015) Detection and prediction
Abdullahi S, Pradhan B (2018a) Land use change modeling and the
of land cover changes using Markov chain model in semi-arid
effect of compact city paradigms: integration of GIS-based cel-
rangeland in western Iran. Environ Monit Assess 187:629. https​
lular automata and weights-of-evidence techniques. Environmen-
://doi.org/10.1007/s1066​1-015-4805-y
tal Earth Sciences 77(251):1–15. https​://doi.org/10.1007/s1266​
Guan D, Li H, Su W, Nagaie T, Hokao K (2011) Modeling urban land
5-018-7429-z
use change by the integration of cellular automaton and Markov
Abdullahi S, Pradhan B (2018b) Land use change modeling and the
model. Ecol Model 222(20):3761–3772
effect of compact city paradigms: integration of GIS-based cel-
Halimi M, Sedighifar Z, Mohammadi Ch (2018) Analysing spatiotem-
lular automata and weights-of-evidence techniques. Environ Earth
poral landuse/cover dynamic using remote sensing imagery and
Sci 77(251):1–15. https​://doi.org/10.1007/s1266​5-018-7429-z
GIS techniques case: Kan basin of Iran. Geojournal 83:1067–1077
Al-Bakri JT, Duqqah M, Brewer T (2013) Application of remote
Halmy MWA, Gessler PE, Hicke JA, Salem BB (2015) Land use/land
sensing and GIS for modeling and assessment of land use/cover
cover change detection and prediction in the north-western coastal
change in Amman/Jordan. Journal of Geographic Information
desert of Egypt using Markov-CA. Appl Geogr 63:101–112
System 5(5):509–519
Hamad R, Balzter H, Kolo K (2018) Predicting land use/land cover
Alhamdan MZ, Oduor P, Flores AI, Kotikot SM, Mugo R, Ababu J
changes using a CA–Markov model under two different scenarios.
(2017) Evaluating land cover changes in Eastern and Southern
Sustainability 10:123. https​://doi.org/10.3390/su101​03421​
Africa from 2000 to 2010 using validated Landsat and MODIS
Hathout S (2002) The use of GIS for monitoring and predicting urban
data. Int J Appl Earth Obs Geoinf 62:8–26
growth in East and West St Paul, Winnipeg, Manitoba, Canada. J
Al-sharif AA, Pradhan B (2014) Monitoring and predicting land
Environ Manag 66(3):229–238
use change in Tripoli Metropolitan City using an integrated
He C, Okada N, Zhang Q, Shi P, Zhang J (2006) Modeling urban
Markov chain and cellular automata models in GIS. Arab J Geo-
expansion scenarios by coupling cellular automata model and sys-
sci 7(10):4291–4301
tem dynamic model in Beijing, China. Appl Geogr 26(3):323–345
Al-sharif AA, Pradhan B (2015) A novel approach for predicting
Hyandye C, Martz LW (2017) A Markovian and cellular automata
the spatial patterns of urban expansion by combining the Chi
land use change predictive model of the Usangu catchment. Int J
squared automatic integration detection decision tree, Markov
Remote Sens 38:64–81
chain and cellular automata models in GIS. Geocarto Int
Jabbari MK, Ahmadi S (2014) Modelling urban development using
30(8):858–881
GIS and CA-Markov, pp 200–202 (Farsi language)
Araya YH, Cabral P (2010) Analysis and modeling of urban land
Jenerette GD, Wu J (2001) Analysis and simulation of land-use
cover change in Setúbal and Sesimbra. Portugal. Remote Sens-
change in the central Arizona-Phoenix region, USA. Landsc Ecol
ing 2(6):1549–1563
16(7):611–626
Aslami M, Akimov PA, Kaytukov TB (2015) About verification
Kamusoko C, Aniya M, Adi B, Manjoro M (2009) Rural sustainability
of multilevel wavelet-based numerical method of local struc-
under threat in Zimbabwe–simulation of future land use/cover
tural analysis for two-dimensional problems. Procedia Eng
changes in the Bindura district based on the Markov-cellular
111:57–64
automata model. Appl Geogr 29(3):435–447
Barredo JI, Engelen G (2010) Land use scenario modeling for flood
Li H, Reynolds JF (1997) Modeling effects of spatial pattern, drought,
risk migration. Sustainability 2:1327–1344
and grazing on rates of rangeland degradation: a combined
Barredo JI, Lavalle JI, Kasanko M (2005) Urban scenario modeling
Markov and cellular automaton approach. In: Quattrochi DA,
and forecast for sustainable urban and regional. In: Campagna
Goodchild MF (eds) Scale in remote sensing and GIS. Lewis
M (ed) GIS for sustainable development. CRC Press, Boca
Publishers, New York, pp. 211–230
Raton, pp 245–329
Lillesand T, Kiefer RW, Chipman J (2014) Remote sensing and image
Batty M, Xie Y, Sun Z (1999) Modeling urban dynamics through
interpretation. Wiley, Hoboken, pp 98–100
GIS-based cellular automata. Comput Environ Urban Syst
López E, Bocco G, Mendoza M, Duhau E (2001) Predicting land-cover
23(3):205–233
and land-use change in the urban fringe: a case in Morelia city,
Bounoua L, Nigro J, Zhang P, Thome K, Lachir A (2018) Map-
Mexico. Landsc Urban Plan 55(4):271–285
ping urbanization in the United States from 2001 to 2011. Appl
Lu Y, Wu P, Ma X, Li X (2019) Detection and prediction of land use/
Geogr 90:123–133
land cover change using spatiotemporal data fusion and the cellu-
Brown DG, Pijanowski BC, Duh JD (2000) Modeling the relation-
lar Automata–Markov model. Environ Monit Assess 191:2. https​
ships between land use and land cover on private lands in the
://doi.org/10.1007/s1066​1-019-7200-2
Upper Midwest, USA. J Environ Manage 59(4):247–263
Mas J-F, Kolb M, Paegelow M, Olmedo MTC, Houet T (2014) Induc-
Chang CL, Chang JC (2006) Markov model and cellular automata
tive pattern-based land use/cover change models: A comparison of
for vegetation. J Geogr Res 45:45–57
four software packages. Environ Model Softw 51:94–111
Chen L, Sun R, Liu H (2013a) Research progress of ecological envi-
McCarthy MJ, Radabaugh K, Moyer RP, Muller-Karger FE (2018)
ronment effect in the evolution of urban landscape pattern. Acta
Enabling efficient, large-scale high-spatial resolution wetland
Ecol Sin 33(4):1042–1050
mapping using satellites. Remote Sens Environ 208:189–201
Chen Y, Li X, Liu X, Ai B (2013b) Analyzing land-cover change and
Megahed Y, Cabral P, Silva J, Caetano M (2015) Land cover mapping
corresponding impacts on carbon budget in a fast developing
analysis and urban growth modelling using remote sensing tech-
sub-tropical region by integrating MODIS and Landsat TM/
niques in Greater Cairo Region—Egypt. ISPRS Int J Geo-Inform
ETM + images. Appl Geogr 45(45):10–21
4(3):1750–1769
Dewan AM, Yamaguchi Y (2009) Land use and land cover change in
Muller MR, Middleton J (1994) A Markov model of land-use change
Greater Dhaka, Bangladesh: using remote sensing to promote
dynamics in the Niagara Region, Ontario, Canada. Landsc Ecol
sustainable urbanization. Appl Geogr 29(3):390–401
9(2):151–157
Etemadi H, Smoak JM, Karami J (2018) Land use change assessment
Myint SW, Wang L (2006) Multicriteria decision approach for land
in coastal mangrove forests of Iran utilizing satellite imagery and
use land cover change using Markov chain analysis and a cellular
automata approach. Can J Remote Sens 32(6):390–404

13
546 Page 12 of 12 Environmental Earth Sciences (2019) 78:546

Osgouei PE, Kaya S (2017) Analysis of land cover/use changes using Tajbakhsh M, Memarian H, Shahrokhi Y (2016) Analyzing and mod-
Landsat 5TMdata and indices. Environ Monit Assess 189(4):136 eling urban sprawl and land use changes in a developing city
Palmate SS, Pandey A, Mishra SK (2017) Modelling spatiotempo- using a CA-Markovian approach. Glob J Environ Sci Manag
ral land dynamics for a trans-boundary river basin using inte- 2(4):397–410
grated cellular automata and Markov chain approach. Appl Geogr Tudun-Wada MI, Tukur YM, YaU HMZS, Musa I, Lekwot VE (2014)
82:11–23 Analysis of forest cover changes in Nimbia forest reserve, Kaduna
Pickard BR, Berkel DV, Petrasova A (2017) Predicts of urbanization State, Nigeria using geographic information system and remote
scenarios reveal trade-offs between landscape change and ecosys- sensing techniques. Analysis 4(21):73–83
tem services. Landsc Ecol 32(3):617–634 Wang SQ, Zheng XQ, Zang XB (2012) Accuracy assessments of land
Piquer-Rodríguez M, Kuemmerle T, Alcaraz-Segura D, Zurita-Milla use change simulation based on Markov-cellular automata model.
R, Cabello J (2012) Future land use effects on the connectivity Procedia Environ Sci 13:1238–1245
of protected area networks in southeastern Spain. J Nat Conserv Wang S, Zhang Z, Wang X (2014) Land use change and prediction in
20(6):326–336 the Baimahe Basin using GIS and CA Markov model. In: IOP
Riccioli F, El Asmar T, El Asmar JP, Fratini R (2013) Use of cellular Conference Series: Earth and Environmental Science, IOP Pub-
automata in the study of variables involved in land use changes. lishing 17(1), pp 012074
Environ Monit Assess 185:5361–5374. https​://doi.org/10.1007/ Wu J, Shen W, Sun W, Tueller PT (2002) Empirical patterns of the
s1066​1-012-2951-z effects of changing scale on landscape metrics. Landsc Ecol
Rindfuss RR, Walsh SJ, Turner BL, Fox J, Mishra V (2004) Developing 17(8):761–782
a science of land change: challenges and methodological issues. Wu L, Liu X, Ma X (2018) Prediction of land-use change and its driv-
Proc Nat Acad Sci 101(39):13976–13981 ing forces in an ecological restoration watershed of the Loess
Roose M, Hietal R (2018) A methodological Markov-CA projection hilly region. Environ Earth Sci. https​://doi.org/10.1007/s1266​
of the greening agricultural landscape a case study from 2005 to 5-018-7413-7
2017 in southwestern Finland. Environ Monit Assess 190(7):411. Yang X, Zheng XQ, Lv LN (2012) A spatiotemporal model of land
https​://doi.org/10.1007/s1066​1-018-6796-y use change based on ant colony optimization, Markov chain and
Sang L, Zhang C, Yang J, Zhu D, Yun W (2011) Simulation of land use cellular automata. Ecol Model 233:11–19
spatial pattern of towns and villages based on CA–Markov model. Yang X, Zheng X-Q, Chen R (2014) A land use change model: Inte-
Math Comput Model 54(3):938–943 grating landscape pattern indexes and Markov-CA. Ecol Model
Serra P, Pons X, Sauri D (2008) Land-cover and land-use change 283:1–7
in a Mediterranean landscape: a spatial analysis of driving Zare Garizi A, Sheikh V, Saddodin A, Mahini S (2012) Simulating the
forces integrating biophysical and human factors. Appl Geogr spatiotemporal changes of forest extent for the Chehelchay water-
28(3):189–209 shed (Golestan province), using integrated CA-Markov model.
Shawul AA, Chakma S (2019) Spatiotemporal detection of land use/ Iran J For Poplar Res 20(2):273–285
land cover change in the large basin using integrated approaches Zubair AO (2006) Change detection in land use and land cover using
of remote sensing and GIS in the Upper Awash basin, Ethiopia. remote sensing data and GIS: a case study of Ilorin and its envi-
Environ Earth Sci. https​://doi.org/10.1007/s1266​5-019-8154-y rons in Kwara State. MSc dissertation. University of Ibadan,
Sohl TL, Claggett PR (2013) Clarity versus complexity: Land-use mod- Ibadan, Nigeria, pp 22–35
eling as a practical tool for decision-makers. J Environ Manage
129:235–243 Publisher’s Note Springer Nature remains neutral with regard to
Subedi P, Subedi K, Thapa B (2013) Application of a hybrid cellular jurisdictional claims in published maps and institutional affiliations.
automaton–Markov (CA-Markov) model in land use change pre-
diction: a case study of Saddle Creek Drainage Basin, Florida.
Appl Ecol Environ Sci 1:126–132

13