Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

https://doi.org/10.1007/s12524-019-01003-8(0123456789().,-volV)(0123456789().
,- volV)

RESEARCH ARTICLE

Analysis and Prediction of Urban Growth Using Neural-Network-

Coupled Agent-Based Cellular Automata Model for Chennai
Metropolitan Area, Tamil Nadu, India
Aarthi Aishwarya Devendran1 • Gnanappazham Lakshmanan1

Received: 5 April 2019 / Accepted: 8 May 2019 / Published online: 13 May 2019
 Indian Society of Remote Sensing 2019

Abstract
Chennai is one of the most densely populated cities in India facing challenges in shifting the city to metropolitan or mega
city in the last two decades with continuing agglomeration. To model the growth of Chennai city, we have used cellular
automata-based urban growth models based on the historical datasets. In the present study, urban growth of Chennai
Metropolitan Area (CMA) was predicted for the year 2017 based on 2010 and 2013 dataset and Chennai city master plan
using neural-network-coupled agent-based cellular automata (NNACA) model. Eight different agents of urbanization
including transportation, hotspots, and industries were used in the prediction modeling. On validating the 2017 predicted
outputs, NNACA model with hotspots proved to be better (hits: 498.52 km2) than that of without hotspots (hits:
488.31 km2). Out of the total eight agents of urbanization, the most influencing agent of urbanization of 2017 was
identified to be the neighborhood of ‘Existing built-up of 2013’ using ‘sensitivity analysis’. Further, the urban sprawl of
CMA for 2010, 2013 and 2017 was measured through Shannon’s entropy. The study area was divided into five directional
and distance-based zones with the State Secretariat as the center. Entropy values suggest the need for more careful planning
for further development in the southern region of CMA which has undergone congested urban growth while urbanization is
dispersed in the northern part of the study region which can be thought for future urban developments.

Keywords Urban growth modeling  Chennai Metropolitan Area  Neural-network-coupled agent-based cellular automata
model  Sensitivity analysis  Shannon’s entropy

Introduction data modeling tools, the urban growth prediction models

have become more realistic.
Urbanization is a result of increase in population, migration Cellular automata (CA)-based UGMs are widely used in
of people from rural places to cities. Urban growth moni- the prediction of urbanization (Batty et al. 1999; Batty
toring becomes essential for the sustainable management of 2005). A typical CA consists of five elements: cell space,
natural resources and for the betterment of humans. Thus, cell state, cell neighborhood, transition rules and time (Liu
urban growth models (UGMs) enable policy makers to plan 2009). UGMs based on CA are found to be most efficient
the future urbanization of their cities (Hill and Lindner than any other mathematical models (Batty and Xie 1994;
2010). Nowadays, with the availability of high resolution Triantakonstantis and Mountrakis 2012; Mubea 2014).
temporal satellite datasets along with GIS-based spatial Santé et al. (2010) had given a detailed review of CA-based
urban prediction models. SLEUTH, based on CA model, is
also widely used to predict urbanization and is imple-
& Gnanappazham Lakshmanan mented to predict the urban growth of Xiamen city of
[email protected] China (Hua et al. 2014) and in Matara city, Sri Lanka
Aarthi Aishwarya Devendran (Serasinghe Pathiranage et al. 2018). To make the predic-
[email protected] tion models more realistic, apart from the historical land
1 use data, various agents of urbanization should also be
Department of Earth and Space Sciences, Indian Institute of
Space Science and Technology, Trivandrum, India considered. CA models with agents including road and rail

123
1516 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

networks were used to predict the urban growth of Tehran Chennai district has a coastal length of 19 km, and
city, Iran (Jokar Arsanjani et al. 2013; Aqbelaghi et al. CMA has a total coastal length of 46 km. Famous beaches
2018). Recently, neural networks are coupled with GIS and in Chennai are Marina, Elliot’s or Besant Nagar, Thiru-
CA models for the prediction of urban landscape. Neural- vanmiyur beaches of which Marina beach is the world’s
network-coupled agent-based cellular automata (NNACA) second longest beach. and it attracts number of tourists
model overcomes the uncertainty associated with the every year (Fig. 1). Chennai is well connected to other
determination of transition rules in CA model. Also, cities through its developed transportation networks.
NNACA has the potential to handle the complex, nonlin- Chennai central is the main terminus railway station and is
ear, spatiotemporal urban datasets with less uncertainty adjacent to Chennai Egmore, another major railway junc-
(Basse et al. 2014). NNACA-based model had been used to tion. One of the fastest growing railway hubs is Tambaram
predict the urban growth for the city of Dongguan, China railway station and is located in the southern region of the
(Zhang 2016) and for Changping district of China (Yang study area (Fig. 1). In addition, the city has also recently
et al. 2016). commenced its metro rail service, a rapid transit system, in
In the present study, the urban growth of Chennai 2015. The city has an international airport located at
Metropolitan Area (CMA), Tamil Nadu, India, was pre- Meenambakkam about 14 km from the Chennai city cen-
dicted using NNACA model without and with hotspots, ter. Three major rivers namely Kosasthalaiyar, Cooum and
i.e., the potential location identified for development in the Adyar pass through CMA, and major lakes including
City Master plan, and the more efficient model was used to Sholavaram Lake, Red Hills Lake and Chembarambakkam
identify the most influencing agents of urbanization in Lake are situated within the study region which serve as the
2017 through sensitivity analysis. The distribution of urban major source of drinking water. Due to the boom in
sprawl for 2010, 2013 and 2017 of CMA was also mea- industrial, automobile, electronic and entertainment sec-
sured through Shannon’s Entropy. tors, the study region has experienced rapid urbanization
over the past two decades (Second Master Plan for Chennai
Metropolitan Area 2008a).
Study Area

Chennai, the capital state of Tamil Nadu, is India’s fourth Data and Methods
largest metropolitan cities after Mumbai, New Delhi and
Kolkata. In the present study, the prediction of urban For the prediction of urban growth of CMA, urban maps
growth was implemented for CMA (Fig. 1) using NNACA along with eight agents of urbanization were considered.
model. CMA falls under three districts including Chennai
Urban maps of 2010, 2013 and 2017 were derived using
and parts of Thiruvallur and Kancheepuram districts and is
15 meters resolution (Multispectral PAN merged) satel-
located on the Coromandel Coast. As per Chennai
lite images of Landsat 7 ETM of 2nd June 2010, Landsat
Metropolitan Development Authority (CMDA), 176 km2
8 of 17th May 2013 and 25th March 2017, respectively
of Chennai district, 637 km2 of Thiruvallur district
(https://earthexplorer.usgs.gov/).
including Ambattur, Thiruvallur, Ponneri and Poonamallee
Location of industries within the study area was obtained
taluks and 376 km2 of Kancheepuram district comprising
from Government of India MEPZ (Madras Export Pro-
Tambaram, Sriperumbudur and Chengalpattu taluks con-
cessing Zone) Special Economic Zone (http://www.mepz.
stitute CMA (http://www.cmdachennai.gov.in/index.html).
gov.in/Directory_SEZ.php?page=2&District=Chennai&
It extends over an area of 1189 km2. Chennai is the most
Entity=SEZ%20Unit&SEZ=&Unit=).
densely populated city in Tamil Nadu, with a density of
Places of public interests, public utility centers, road and
26,553 people per km2. Chennai is the largest commercial
railway networks and road junctions derived from
and industrial centers of South India as well as a cultural,
(http://www.openstreetmap.org/) were updated using
economic, educational and information technology (IT)
Google Earth.
centers which attract a large number of migrants to the city.
Population data for each of the revenue villages within
After 2010, due to rapid growth of industrialization,
CMA were obtained from Directorate of Census Oper-
especially textiles and tanneries industries, the rate of
ations, Government of Tamil Nadu (http://www.census.
migration toward Chennai city had increased to multiple
tn.nic.in/).
folds. Majority of Chennai’s economy is based on the
Land values were obtained from Guideline Value and
automobile sector, software services, hardware manufac-
Property Valuation of the Registration Department,
turing, healthcare and financial services. It is also known as
Government of Tamil Nadu (http://www.tnreginet.net/
‘Detroit of India’ with thriving automotive industries.
guideline_value.asp).

123
Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526 1517

Fig. 1 Study area map

123
1518 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

Apart from the field knowledge about the study area, density map of all the revenue villages within the study
Google Earth was used for the validation of urban maps area was prepared based on the census data. (4) Commu-
of CMA during the study periods. tation included all the major road junctions, bus stands and
railway junctions (Fig. 1). In road networks, trunk, primary
and residential roads were categorized as high preference
Urban Cover
roads. Medium preference roads include secondary and
tertiary roads. Outer ring road and expressway were cate-
Landsat data of the years 2010, 2013 and 2017 (as men-
gorized as least preference roads in the study. A 3 9 3 (5)
tioned in ‘‘Data and Methods’’ section) were used in the
urban neighborhood was used in the prediction technique.
preparation of land cover maps of the study region using
For all these agents, along with (6) existing built-up, (7)
support vector machine (SVM) technique. SVM of super-
places of public interest and (8) public utility centers,
vised classification (Sukawattanavijit et al. 2017) was
proximity maps were obtained and fed into NNACA model
adopted to classify the land cover features in the study
as inputs. As per (Second Master Plan for Chennai
region such as built-up, vegetation, waterbody and open-
Metropolitan Area 2008b), development prohibited areas
land. As the present study aims at predicting only the urban
had been identified and included in the NNACA model as
growth of CMA, the land cover maps were converted to
constraints where urban growth was prohibited during the
binary maps with built-up and non-built-up categories by
modeling process. Coastal regulation zone (CRZ)—I,
combining vegetation, waterbody and openland categories
major water bodies, areas around airport at Meenam-
into one. Thus, urban maps with only two categories
bakkam, areas of 100 m around the boundary of Indian Air
namely ‘Built-Up’ and ‘Non-Built-Up’ were obtained for
Force station near Tambaram, Pallikaranai Swamp area and
the years 2010, 2013 and 2017.
green belt areas of 15 m along Poonamallee and Red Hills
bypass roads had been identified as constraints for the
Identification of Hotspots
current study.
In a CA model, the state of any cell at a given time is based
Urban Growth Modeling Using NNACA
on its state at the previous time step and the states of its
neighboring cells (Wolfram 1984). In the current study,
The urban growth prediction was implemented using
based on the Government policy, areas of 500 meters
NNACA model (Fig. 2) with the proximity maps of all the
around OMR (Old Mahabalipuram Road) are found to be
agents along with the constraints (bottom right box of
the urban hotspots owing to the spurt of IT centers (Second
Fig. 2). Multilayer perceptron (MLP) neural network of
Master Plan for Chennai Metropolitan Area 2008b). Hence,
LCM (Land Change Modeler) (https://clarklabs.org/terrset/
a buffer zone of 500 meters around OMR is included in the
land-change-modeler/), TerrSet was used for the prediction
urban map of 2013 as hotspots in the prediction modeling.
of urban growth of CMA.
Thus, for the prediction of urban growth in 2017 based on
Let K be the input neurons to the NNACA model. K: [k1,
NNACA model, two scenarios were considered. One sce-
k2, …, k18], where k1 to k18 represent the proximity maps
nario included hotspot location (IHS Model) in the 2013
of all the agents and the urban neighborhood map. 2n
urban map, and the other excluded the hotspot location 3

(EHS Model). number of hidden neurons are appropriate in modeling

complex scenarios (Yeh and Li 2002), where n is the
Agents of Urbanization number of input neurons (Here, n = 18). In the current
study, 12 hidden neurons were used to simulate the urban
For the prediction of urban growth using NNACA model, growth. To each of the 12 hidden neurons, input (HI) was
eight different agents of urbanization (top right box of assigned
Fig. 2) for the study region were selected based on the X
18

experts’ opinions from CMDA, Urban Planning depart- HI ¼ ðwk  Ik Þ ð1Þ

k¼1
ments and academicians. Of the selected agents, (1)
industries were categorized into four based on their Pol- where Ik is the input neurons representing the proximity
lution Index Score (PIS) (C 60: Red, 41–59: Orange, maps of all the agents (Here, k = 18); w is the weights
21–40: Green and B 20: White) (Ministry of Environment, assigned randomly by the neural network ranging between
Forest and Climate Change 2016). (2) Land values were - 1 and ? 1. The output from the hidden neurons is given
categorized into three based on their prices ([ 1500 INR: by
High land prices, [ 500 AND B 1500 INR: Medium land
prices, B 500 INR: Low land prices). (3) Population

123
Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526 1519

Fig. 2 Methodology adopted in the study for the prediction of urban growth in CMA using NNACA model

1 with every iteration, the weights in the network gets

HO ¼ ð2Þ updated such that the model outputs come closer to the
1 þ eHI
target outputs, thereby minimizing the RMS error for each
where HO is the outputs from the hidden neurons. Sig-
of the output neurons and thereby for the entire network,
moidal functions are often used in artificial neural networks
and the entire process is continued till the RMS error
as an activation function to introduce nonlinearity in the
reaches the minimum. An accuracy rate of 80% is con-
model (Bishop 1995). The outputs from the hidden neurons
sidered to be acceptable (Eastman 2012), and it can be
were fed into the output neurons as input ðOI Þ
considered that the model has learnt well based on the input
X
12 datasets. In the current study, an accuracy rate of 92% was
OI ¼ ðwh  HO Þ ð3Þ reported.
h¼1

where h is the number of hidden neurons. Outputs from the Influence of Agents of Urbanization
output neurons ðOO Þ are given by on the Prediction Output
1
OO ¼ ð4Þ It is established that NNACA is the most efficient model in
1 þ eOI
predicting the urban growth when compared to other tra-
Each of the two output neurons in the model gets a value ditional CA models and also the inclusion of hotspots, if
ranging between 0 and 1. These ‘model outputs’ obtained any, improves the prediction output (Aarthi and Gnanap-
from the NNACA model along with the ‘true outputs’ also pazham 2018; He et al. 2018). But the choice of agents of
known as ‘targets’, obtained from the historical datasets, urbanization is subjective and is based on the experts’
were used to assess the accuracy of the prediction modeling opinions and introduces certain amount of uncertainty into
through root-mean-square (RMS) error calculation. Thus, the model. Hence, a sensitivity analysis is performed to
for each of the pixel in the study area, the NNACA esti- identify the most influencing agents of urbanization in the
mates the RMS error based on Eq. (5) study region. Initially the prediction model was run with all
X1 the eight agents together and then individually. Based on
ETotal ¼ ðtarget  model outputÞ2 ð5Þ their prediction accuracies, the most influencing agent of
2
urbanization for the study area was identified (Al-Ahmadi
where ETotal = EO1 þ EO2 ; EO1 and EO2 : errors from output
et al. 2013).
neurons 1 and 2, respectively. MLP with back propagation
algorithm is widely used in urban prediction modeling, as

123
1520 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

Validation that the urban cover in 2017 had increased more than
double the area in 2010 which is evident from the popu-
Each of the predicted outputs is validated against the lation data. CMA had a population of 7.04 million in 2001
observed data through which the efficiency of the predic- which increased to 8.87 million in 2011 and it is estimated
tion model is calculated. Error matrix or contingency to have a population of 11.19 million in 2021 (Second
table is the most widely used accuracy assessment tech- Master Plan for Chennai Metropolitan Area 2008c). This is
nique which calculates the overall accuracy (OA) and due to the fact that many IT centers, industries, educational
kappa coefficient (k) values (Ahmed et al. 2013). institutions, banking sectors and so on had emerged in the
study region which attracted many rural people to migrate
Assessing the Type of Urban Growth Based to CMA and the number of industries within the study
on Entropy region had increased five times from 7782 in 2001 to
42,000 in 2017 (Chennai District Statistical Hand Book
To assess the direction and distribution of urban sprawl, 2017). Hotspot location was included into the urban map of
normalized Shannon’s entropy was calculated (Mosam- 2013 in a linear pattern of 500 m around OMR for the
mam et al. 2017). In this study, urban maps of the years prediction of urban cover in 2017 to capture the urban-
2010, 2013 and 2017 were used. The study area was ization especially due to the emergence of IT centers in this
divided into eight directional zones [North–North East (N– region.
NE), North West–North (NW–N), West–North West (W–
NW), South West–West (SW–W), South–South West (S– Shannon’s Index for Assessing the Type of Urban
SW), South East–South (SE–S), East–South East (E–SE) Sprawl
and North East–East (NE–E)] with the State Secretariat as
the center. Since the study area lies on the east coast, the From 2010 to 2017, the urban cover in CMA was
directional zones with insignificant areas were merged with increasing (Fig. 4). The study area was divided into five
the neighboring zones, i.e., NE–E is merged with N–NE major directional zones with the State Secretariat as the
zone and E–SE and SE–S were merged with S–SW zone center (Fig. 5a). It is observed that the maximum urban-
resulting in five zones (N–NE, NW–N, W–NW, SW–W ization (192.03 km2) occurs in SW–W zone of CMA which
and S–SW). Further, the study area is divided into five includes T. Nagar, Thirumazhisai. Location of commercial
buffer zones of each 7 km based on the distance from the industries, amenities including hospitals, educational
State Secretariat. Based on the urbanization in each of the institutions, and religious centers are higher, and also the
zones, the normalized entropy ðHn Þ was calculated accessibility to transportation network is much higher in
  this zone. Hence, this zone experiences highest urban
1X n
1
Hn ¼ pðxi Þ log ð6Þ growth in CMA. The S–SW zone is the second most
n i¼1 pð x i Þ
urbanized (139.79 km2), and the major locations within
where n is the number of zones (n = 5); p(x) probability of this zone include Adyar, Perungudi. Location of IT centers
built-up in each zone; values range from 0 to loge(n). and industries mainly causes the urbanization in this zone
Entropy values closer to 0 indicate a compact or closely (https://www.sipcot.com/industrial_complex.html). The
packed type of distribution, whereas a dispersed pattern of third most urbanized in the study area is the W–NW zone
urban growth is observed with the values closer to loge(n). (134.87 km2) including Anna Nagar, Avadi and location of
SIPCOT (State Industries Promotion Corporation of Tamil
Nadu) Industrial Estates drive the urbanization in this zone.
Results and Discussion Vyasarpadi and Minjur are located within the NW–N zone,
the fourth urbanized in CMA (83.16 km2), in which auto-
Land Cover Maps mobile and petro-chemical-based industries are situated.
The least urbanization is found in the N–NE zone
The land cover maps of the study region produced an OA (22.88 km2) which comprises Tondiarpet, Ennore. Urban-
of 92.23%, 87.87% and 87.57% and k values of 0.89, ization occurs in this zone mainly due to the industries
0.8437 and 0.8412 for the year 2010, 2013 and 2017, related to coastal activities.
respectively. The land cover maps (Fig. 3) show a The entropy values of urbanization of 2010, 2013 and
remarkable increase in built-up from 2010 to 2017. CMA 2017 were found to be 0.981, 1.114, and 1.0845, respec-
comprises of total area of 1189 km2 of which area of urban tively. The maximum entropy value with five zones is
cover in 2010, 2013 and 2017 is mapped as 237.41 km2, 1.6094 (loge5). It could be seen that the entropy values kept
400.57 km2 and 572.11 km2, respectively. It could be seen increasing from 2010 to 2013 and it decreased after 2013.
This shows that the urban sprawl was distributed till 2013

123
Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526 1521

Fig. 3 Land cover maps of CMA during the study periods. a June 2010; b May 2013; c March 2017

W–NW zones experienced distributed growth in 2017. The

entropy analysis of the study region clearly indicates that
compact type of urban growth was found toward southern
region, whereas distributed growth was found toward
northern region of the study area.
The rate of urban growth is higher within the 7 km from
the State Secretariat followed by the 14 km, 21 km, 28 km
and 35 km zones, respectively (Fig. 5b). It is important to
note that the entropy values of urbanization of the 7 and
14 km zones kept decreasing from 2010 to 2017. These
two zones fall within the Chennai Corporation limits. Thus,
it is implied that the Chennai Corporation had begun to
experience congested type of urban growth since 2010
itself. Teynampet, Vyasarpadi, Guindy, Madhavaram are
some of the locations within these zones where the
employment opportunities, accessibility to transportation
and availability of socio-economic factors including hos-
pitals, schools, recreation center, etc., are higher (Sekar and
Kanchanamala 2011). The third highest urbanized zone
(21 km zone) had started to experience congested urban
growth after 2013. Major locations within this zone are
Meenambakkam, Red Hills. Since the Chennai Corporation
had already become saturated with urbanization, further
urban growth had started to occur in the fringes of the
Fig. 4 Urban sprawl of the observed urban of the study region in Corporation boundary. The 28 km and 35 km zones
2010, 2013 and 2017 including Thiruninravur, Thirumazhisai, Sithalapakkam,
Perungalathur experience distributed type of urban sprawl
and after the year 2013 congested or concentrated type of and show an increase in urbanization in 2017. It can also be
urban growth was observed in the study region (Fig. 5a). In noted that the southern part of CMA experiences higher
2017, the concentrated type of urban growth was observed growth than the northern region, maybe because the
in S–SW, SW–W directions. Because of the availability of southern region falls within the State boundary. Thus,
employment opportunities and socio-economic amenities, urbanization is lesser in the northern zones (N–NE, NW–
southern CMA is highly urbanized than the north and N) as they are at the cross-state boundary. CMDA (Hous-
settlements occur mainly in the southern region and thus ing And Urban Development Department, Chennai
urbanization in south gets congested. N–NE, NW–N and Metropolitan Area 2018) had announced the expansion of

123
1522 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

Fig. 5 Entropy values of the study region. a Entropy values of CMA for the study periods for five directional zones with the State Secretariat as
the center; b entropy values of CMA for the study periods for five distance-based buffer zones from the State Secretariat

CMA from 1189 km2 to an area of 8878 km2 including the (73.59 km2) than in the EHS model (83.8 km2). However,
entire Chennai, Kancheepuram, Thiruvallur districts and the false alarm rate was found to be higher in IHS model
Arakkonam and Nemili taluks of Vellore district. The need (106.57 km2) than in EHS model (88.57 km2). Hence,
for expansion was mainly due to the rapid urbanization in sensitivity analysis was performed to improve the effi-
Sriperumbudur, Thiruvallur, Ambattur, Sholinganallur. ciency of IHS model to minimize the areas of misses,
Thus, this analysis clearly highlights the importance of the thereby identifying the most influencing agent of urban-
Government’s decision in the expansion of CMA for ization of CMA in 2017.
effective urban planning policies.
Identifying Most Influencing Agents
Predicted Urban Growth of Urbanization

EHS and IHS models with all the inputs (Fig. 6) predicted IHS and EHS models predicted 498.52 km2 and
576.92 km2 and 606.35 km2 of the study region as built-up 488.31 km2, respectively, as built-up in 2017 (as discussed
areas, respectively, whereas the observed area of built-up in ‘‘Predicted Urban Growth’’ section). In sensitivity
in 2017 was found to be 572.11 km2 in 2017 (Fig. 7). The analysis, each of the agents was used in predicting the
accuracy of the prediction output of IHS (O.A: 85.51%; k: urban growth of CMA individually (Table 1). It could be
0.71) is found to be higher than the accuracy of EHS model seen that the maximum area of hits (502.42 km2) was
(O.A: 84.82%; k: 0.6966). Upon validating the prediction achieved by using only ‘Existing Built-Up’ of 2013 as the
outputs with the observed urban map of 2017, the areas of agent of urbanization in the IHS model. Also, the area of
hits (correctly predicted built-up) are also found to be misses was reduced (69.7 km2) when only the existing
higher in IHS model (498.52 km2) than EHS model built-up was used as the agent of urbanization when
(488.31 km2). Misses correspond to under-prediction and compared to area of misses (73.59 km2) of IHS model with
false alarms quantify the over-prediction in the prediction all the eight agents. This could be attributed to the fact that
model. For an efficient model, apart from the higher OA the study region experiences congested type of urban
(higher areas of hits), it is also important to have lesser growth as it is evident from the entropy values (as dis-
areas of false alarms and misses, as given by k values cussed in ‘‘Shannon’s Index for Assessing the Type of
(Shafizadeh-Moghadam et al. 2017). In this study, higher Urban Sprawl’’ section). In a congested urban growth
area of hits is found associated with IHS model. Also, the pattern, settlements occur in already urbanized regions.
areas of misses were found to be lesser in IHS model Thus, existing built-up itself was enough to model the

123
Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526 1523

Fig. 6 Inputs of NNACA model for the urban growth prediction for industries; l places of public interests; m public utility centers; n high
CMA. Proximity maps of a hotspots; b existing built-up; c commu- land prices; o medium land prices; p low land prices; q population
tation; d railways; e high preference roads; f medium preference density map; r urban neighborhood map; s constraints (color
roads; g least preference roads; h red category industries; i orange figure online)
category industries; j green category industries; k white category

urban growth of the study region (Hamad et al. 2018) while to 752.56 km2 in which observed area of built-up in this
in our earlier study, for the urban growth prediction of region in 2017 was 271.7 km2. NNACA model was
Sriperumbudur Taluk located in the southwest direction of implemented with all the agents of entire CMA which
CMA, inclusion of the agent ‘Industries’ improved the resulted in 206.73 km2 area of built-up (Fig. 8). However,
model accuracy. when the agents of CMA excluding Chennai Corporation
Further to assess the influence of agents of urbanization were used, the EHS model predicted 210.94 km2 area of
of Corporation boundary over the entire study region, built-up in 2017 with improvement in the accuracy from
Chennai Corporation was excluded from CMA and urban 0.63 to 0.65 and decrease in areas of misses (Table 2).
prediction was implemented. Since hotspot does not fall This further emphasizes the fact that the effects of
into this excluded boundary, urban growth prediction was agents of urbanization vary from region to region and in the
carried out without hotspot as an agent of urbanization. The current study; the agents of urbanization of Chennai Cor-
area excluding Chennai Corporation from CMA extends up poration could have comparatively lesser impact in the

123
1524 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

Fig. 7 Observed and validated urban cover of CMA. Observed urban cover in a 2010; b 2013; c 2017. Validated urban cover of 2017 using
d EHS model with all the agents; e IHS model with all the agents; f IHS model with existing built-up of 2013

Table 1 Sensitivity analysis of agents of urbanization using IHS region away from the corporation boundary. Also, urban
model for the prediction of urban cover in 2017 growth for this region was predicted for the year 2020
Agents of urbanization Area of validation outputs (km2) using the urban maps of 2013 and 2017 and agents of CMA
excluding Chennai Corporation of 2017. The NNACA
Urban hits Misses False alarms
model predicted 343.51 km2 to be urbanized within a span
Existing built-up 502.42 69.7 102.69 of 3 years.
Hotspots 489.27 82.84 115.84
Commutation 481.64 90.48 123.43
High preference roads 495.18 76.93 109.91 Conclusion
Medium preference roads 491.28 80.83 113.79
Least preference roads 486.89 85.23 118.18 We conducted the study to model the urban growth of
Railways 486.52 85.59 118.51 CMA using NNACA model without and with hotspots, i.e.,
Red category industries 480.58 91.53 124.41 the proposed region for development based on the Master
Orange category industries 483.52 88.59 121.64 Plan of CMA of 2008 with eight agents of urbanization.
Green category industries 491.51 80.6 113.54 Inclusion of hotspot improved the accuracy of the urban
White category industries 489.26 82.86 115.84 growth significantly. To understand the direction of urban
High land prices 490.65 81.46 114.46
growth, entropy analysis was carried out. Results suggest
Low land prices 484.62 87.49 120.44
that urbanization of CMA is of congested type and thus the
Medium land prices 484.67 87.44 120.43
agent, existing built-up, makes the other agents insignifi-
cant. Sensitivity analysis was performed to find the most
Places of public interests 491.37 80.74 113.74
appropriate agents of urbanization as the selection of
Public utility centers 491.59 80.52 113.52
agents remains always uncertain. In our study, existing
Population 491.56 80.55 113.53
built-up proved to be the most influencing agent of
urbanization for CMA in 2017. Results of entropy analysis

123
Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526 1525

Fig. 8 Validation outputs of NNACA model for the region excluding of entire CMA; c Urban Prediction for 2020 with agents of CMA
Chennai Corporation from CMA. a NNACA model with agents of excluding Chennai Corporation (which gave more accurate results for
CMA excluding Chennai Corporation; b NNACA model with agents 2017, i.e., based on a)

Table 2 Accuracy and validation of predicted urban in 2017 in the region excluding Chennai Corporation from CMA using EHS model
Overall accuracy Area (km2) of validation outputs
O.A. (%) k Urban hits Misses False alarms

With agents of CMA excluding Chennai Corporation 84.01 0.6533 210.94 60.74 59.52
With agents of entire CMA 83.8 0.6268 206.73 64.97 56.87

show that this is the crucial moment to develop the city Al-Ahmadi, K., See, L., & Heppenstall, A. (2013). Validating spatial
plan with due care for further development activities in the patterns of urban growth from a cellular automata model. In A.
Salcido (Ed.), Emerging applications of cellular automata (p.
southern and southwest region of CMA which has under- 26). London: InTechOpen.
gone congested urban growth, whereas urbanization is Aqbelaghi, A. S., Ghorbani, M., Farhadi, E., & Shafiee, H. (2018).
dispersed in the northern part of the study region which can Environmental approach in modelling of urban growth: Tehran
be thought for future urban developments. Further, the City, Iran. Asian Journal of Water, Environment and Pollution,
15(2), 47–56.
urban growth for the year 2020 was also predicted for the Basse, R. M., Omrani, H., Charif, O., Gerber, P., & Bódis, K. (2014).
region excluding Chennai Corporation from CMA using Land use changes modelling using advanced methods: Cellular
NNACA model which would enable the urban policy automata and artificial neural networks. The spatial and explicit
makers to take appropriate development actions for the representation of land cover dynamics at the cross-border region
scale. Applied Geography, 53, 160–171.
sustainable allocation and urbanization of the study area. Batty, M. (2005). Cities and complexity: Understanding cities with
Urban growth models would yield more realistic prediction cellular automata, agent-based models, and fractals. Cambridge,
with improvements in the algorithms which would enable MA: MIT Press.
the users to determine the most appropriate CA elements Batty, M., & Xie, Y. (1994). From cells to cities. Environment and
Planning B: Urban Analytics and City Science, 21(7), S31–S48.
including neighborhood configuration based on their input Batty, M., Xie, Y., & Sun, Z. (1999). Modeling urban dynamics
datasets. through GIS-based cellular automata. Computers, Environment
and Urban Systems, 23(3), 205–233.
Bishop, C. M. (1995). Neural networks for pattern recognition. New
York: Oxford University Press.
Chennai District Statistical Hand Book. (2017). Department of
References Economics and Statistics, Government of Tamil Nadu. https://
cdn.s3waas.gov.in/s313f3cf8c531952d72e5847c4183e6910/
Aarthi, A. D., & Gnanappazham, L. (2018). Urban growth prediction uploads/2018/06/2018062923.pdf. Accessed 31 March 2019.
using neural network coupled agents-based cellular automata Eastman, J. R. (2012). IDRISI selva tutorial, manual version 17.
model for Sriperumbudur Taluk, Tamil Nadu, India. The http://uhulag.mendelu.cz/files/pagesdata/eng/gis/idrisi_selva_
Egyptian Journal of Remote Sensing and Space Science, 21(3), tutorial.pdf. Accessed 31 March 2019.
353–362. Hamad, R., Balzter, H., & Kolo, K. (2018). Predicting land use/land
Ahmed, B., Ahmed, R., & Zhu, X. (2013). Evaluation of model cover changes using a CA-Markov model under two different
validation techniques in land cover dynamics. ISPRS Interna- scenarios. Sustainability, 10(10), 3421.
tional Journal of Geo-Information, 2(3), 577–597.

123
1526 Journal of the Indian Society of Remote Sensing (September 2019) 47(9):1515–1526

He, J., Li, X., Yao, Y., Hong, Y., & Jinbao, Z. (2018). Mining Second Master Plan for Chennai Metropolitan Area. (2008c). http://
transition rules of cellular automata for simulating urban www.cmdachennai.gov.in/Volume3_English_PDF/Vol3_Chap
expansion by using the deep learning techniques. International ter03_Demography.pdf. Accessed 31 March 2019.
Journal of Geographical Information Science, 32(10), Sekar, S. P., & Kanchanamala, S. (2011). An analysis of growth
2076–2097. dynamics in Chennai Metropolitan Area. Institute of Town
Hill, A., & Lindner, C. (2010) Modelling informal urban growth Planners, 8(4), 31–57.
under rapid urbanisation—A CA-based land-use simulation Serasinghe Pathiranage, I. S., Kantakumar, L. N., & Sundaramoorthy,
model for the city of Dar es Salaam, Tanzania. Ph.D. thesis (p. S. (2018). Remote sensing data and SLEUTH urban growth
46). Dortmund: Technical University of Dortmund. model: As decision support tools for urban planning. Chinese
Housing and Urban Development Department, Chennai Metropolitan Geographical Science, 28(2), 274–286.
Area. (2018). http://www.cmdachennai.gov.in/pdfs/go/2018/ Shafizadeh-Moghadam, H., Asghari, A., Tayyebi, A., & Taleai, M.
go13.pdf. Accessed 31 March 2019. (2017). Coupling machine learning, tree-based and statistical
Hua, L., Tang, L., Cui, S., & Yin, K. (2014). Simulating urban growth models with cellular automata to simulate urban growth.
using the SLEUTH model in a coastal peri-urban district in Computers, Environment and Urban Systems, 64, 297–308.
China. Sustainability, 6(6), 3899–3914. Sukawattanavijit, C., Chen, J., & Zhang, H. (2017). GA-SVM
Jokar Arsanjani, J., Helbich, M., & de Noronha Vaz, E. (2013). algorithm for improving land-cover classification using SAR and
Spatiotemporal simulation of urban growth patterns using agent- optical remote sensing data. IEEE Geoscience and Remote
based modeling: The case of Tehran. Cities, 32, 33–42. Sensing Letters, 14(3), 284–288.
Liu, Y. (2009). Modelling urban development with geographical Triantakonstantis, D., & Mountrakis, G. (2012). Urban growth
information systems and cellular automata. Boca Raton, FL: prediction: A review of computational models and human
CRC Press. perceptions. Journal of Geographic Information System, 4,
Ministry of Environment, Forest and Climate Change. (2016). 555–587.
Government of India. http://pib.nic.in/newsite/PrintRelease. Wolfram, S. (1984). Cellular automata as models of complexity.
aspx?relid=137373. Accessed 31 March 2019. Nature, 311(5985), 419–424.
Mosammam, H. M., Nia, J. T., Khani, H., Teymouri, A., & Kazemi, Yang, X., Zhao, Y., Chen, R., & Zheng, X. (2016). Simulating land
M. (2017). Monitoring land use change and measuring urban use change by integrating landscape metrics into ANN-CA in a
sprawl based on its spatial forms: The case of Qom city. The new way. Frontiers of Earth Science, 10(2), 245–252.
Egyptian Journal of Remote Sensing and Space Science, 20(1), Yeh, A. G.-O., & Li, X. (2002). Urban simulation using neural
103–116. networks and cellular automata for land use planning. In D.
Mubea, K. W. (2014). Scenarios of urban growth in Kenya using E. Richardson & P. van Oosterom (Eds.), Advances in spatial
regionalized cellular automata based on multi temporal landsat data handling (pp. 451–464). Berlin: Springer.
satellite data. Ph.D. thesis (p. 3). Bonn: University of Bonn. Zhang, X. (2016). Urban growth modeling using neural network
Santé, I., Garcı́a, A. M., Miranda, D., & Crecente, R. (2010). Cellular simulation: A case study of Dongguan City, China. Journal of
automata models for the simulation of real-world urban Geographic Information System, 8(3), 317–328.
processes: A review and analysis. Landscape and Urban
Planning, 96(2), 108–122.
Second Master Plan for Chennai Metropolitan Area. (2008a). http:// Publisher’s Note Springer Nature remains neutral with regard to
www.cmdachennai.gov.in/Volume1_English_PDF/Vol1_Chap jurisdictional claims in published maps and institutional affiliations.
ter03_Economy.pdf. Accessed 31 March 2019.
Second Master Plan for Chennai Metropolitan Area. (2008b). http://
www.cmdachennai.gov.in/Volume2_English_PDF/DR-English.
pdf. Accessed 31 March 2019.

123