Urban Growth Modelling Based on CA-Markov Approach on
Bengaluru India
Jyothi Gupta
1 a
and Raghunandan Kumar
2
1
School of Architecture, CHRIST (Deemed to be University), Bengaluru, 560074, Karnataka, India
2
Department of Civil Engineering, CHRIST (Deemed to be University), Bengaluru, India
Keywords: Geospatial Modelling, Urban Growth Predictions, Cellular Automata-Markov Chain Model, Land Use / Land
Cover Change (LULC), Remote Sensing.
Abstract: The theme of this research is to create spatial patterns for Bengaluru city in India to understand the urban
growth over the past 40 years. The problem of our re-search addresses developing an integrated Geospatial
Modelling Approach to as-sess Urban Growth patterns in Bengaluru Metropolitan Region. This study uses
the various logical methods to create the Land Use/Land Cover (LULC) Map, all the datasets in google earth
engine are categorised in the Supervised Classification. Machine Learning Processes such as Random Forest
(RF), Classification and Regression Tree (CRT), and Support Vector Machine (SVM) classifiers are
considered for this Classification. The Classifier’s performance is evaluated using statistical measures like
overall Accuracy and kappa statistics. Classes with multiple parameters are carried out with the Hybrid
Cellular Automata- Markov (CA-Markov) method, which is capable of duplicating changes through one
grouping to another. This hybrid model supports model both spatial 3D and temporal time-based changes.
The main product after modelling predicts LULC for 2041 and 2051. The argument is that CA-Markov,
Shannon entropy will allow us to define how much area of all classes will be changed in 2041 and 2051.
1 INTRODUCTION
Land use is a phrase that is referred to how humans
use the land and its resources, as well as the purposes
for why they do so. The environment or vegetation
type present, like forests or farmland, is referred to as
land cover. Artificial changes in the earth's crust are
referred to as land use/land cover (LULC), often
called land change (Bhat et. al, 2015). Landcover use
has been identified as a fundamental cause of climatic
change on geographical and time dimensions,
appearing as a critical environmental concern and one
of the major research initiatives on global change
research on a local scale (Baqa et al, 2021).
2 AIM AND OBJECTIVES
AIM: To develop a cohesive CA-Markov Model
Approach to assess Urban Growth patterns in
Bengaluru Metropolitan Region.
a
https://orcid.org/0000-0003-0612-0188
Objective of the Study: Visualize and analyse the
Spaciotemporal transformation in Land use /Land
cover (LULC) from -1991,2001,2011,2021(40
years). Simulate the past LULC and forecast the
future development of the Bengaluru Metropolitan
Region using CA Markov Model 2031, 2041.
Identifying Specific Regions where intense Urban
Growth can occur in Bengaluru Metropolitan Region.
Table 1: Methodology Flow using Google earth engine.
Google
earth engine
(GEE)
Export
Landsat
images in
tiff format
Creating
train/test
data for
class
Use of
Classifier
Train the
classifier
using train
data
Classify as
image or
feature
selection
Data
catalog
DEM
OSM
Data
(Road
layer)
Landsat 5,7
and
Sentinel
-2
Data from
Slope,
Aspect
Euclidean
distances
On roads
and Urban
Image pre
-
processing
LULC
change
analysis
LULC
change
matrix
Supervised
LULC
Transition
Potential
Transition
Potential
Map
Reference
Maps
Gupta, J. and Kumar, R.
Urban Growth Modelling Based on CA-Markov Approach on Bengaluru India.
DOI: 10.5220/0012876600003739
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 1st International Conference on Artificial Intelligence for Internet of Things: Accelerating Innovation in Industry and Consumer Electronics (AI4IoT 2023), pages 385-391
ISBN: 978-989-758-661-3
Proceedings Copyright © 2024 by SCITEPRESS Science and Technology Publications, Lda.
385
classificatio
n
Modelling
(MLP)
Accuracy
Assessment
91,2001,21,
21
CA Markov
Simulation
Model
Validation
for 21
Calibrated
CA
-
Markov
Model 31-
41
Predicted
LULC maps
The table 1 shows the flow in methodology, which
is used in this study.
3 DATA COLLECTION
The Remote Sensing Data of LANDSAT
Multispectral, TM, ETM+, and OLI/TIRS &
Sentinel-2 MSI: Multispectral Instrument, Level-2A
data is used for Supervised Classification such as
Random Forest (RF), Classification and Regression
Tree (CRT), and Support Vector Machine (SVM) in
Google Earth Engine (GEE). Using these datasets,
land use land cover (LULC) for
1991,2001,2011,2020 are Simulated in Google Earth
Engine. Digital Elevation Model (DEM) from Earth
Data by NASA is Downloaded in table 1-2. The
Digital Elevation Model (DEM) Slope is Generated
in ArcGIS Pro. The Slope gives the identification of
terrain, whether the Terrain is Steep or flat (Mishra et
al, 2014). The low slope value will have flat terrain,
and a high slope value will have steep terrain. From
the Slope, the aspect is generated. The aspect
identifies the downscale direction of a high-value
change rate through one cell towards its neighbours.
Road Layer has been obtained from the Open Street
map. Euclidean Distance to Roads and Railways are
Considered. Euclidean Distance to Built-up is also
considered. Table 2 is created to show the link
attachment.
Table 2: Data collection from Online link.
Year
Data
GEE Spatial resolution / Link
1991
Landsat 5 Series
TM AND ETM -
GEE
30 m
2001-2011
Landsat 7 Series
TM AND ETM -
GEE
30 m
2021
Sentinel 2 GEE
datasets
10m to 60 m
Administrative and
city boundary
Shapefile-https://www.diva-
gis.org/
DEM (Digital
Elevation/Terrain
model) for Slope
and Aspect
Earth Data -
https://earthdata.nasa.gov/
Census data for 4
decades
Census of India
https://censusindia.gov.in/
Road network and
Railway
Open Street Map
https://www.openstreetmap.org/
4 CA-MARKOV MODEL
The CA-Markov model a hybrid model which
develops the traditional Markov model with the
Cellular Automata model (CA). The CA methods are
utilized to regulate the spatial dynamics of the GIS
platform (Jain et al, 2016). The spatiotemporal raster
Based da-ta modelling is employed to show what has
changed for constant data over time across Land
use/land cover categories using transition probabilities.
When it comes to land-use change projections, the
Markov model concentrates upon quantity (Jadawala
et al, 2021). The spatial parameters of this model are
inadequate and don't account for the different forms of
land use types of variations in the spatial magnitudes.
The CA model prepares a robust area conception; this
means it can handle complex space systems in terms of
space-time dynamic evolution (Yadav et al, 2021). The
CAMarkov model, which combines Markov and CA
theories, is concerned with time series and space for
prediction purposes. This could effectively simulate
changes in quantity and space of land use patterns
throughout time and space. The LULC maps were
created using the Google Earth engine, then exported
as Geo Tiff files and divided into four categories.
Water is in class 1, Vegetation is in class 2, Barren is
in class 3, and Built-up is in class 4. These LULC maps
were converted into rst format from Geo Tiff Format
in Clark Labs TerrSet IDRISI software. The land
change modeller(figure1-3) helps to make the forecast
LULC diagram centred on equally the previous LULC
map and future LULC plan.
This panel in figure 1 creates several Transitional
maps. Changes, persistence, gains, and losses can be
mapped by land use/cover class, as well as transitions
and transfers by class. Change patterns in
environments influenced by hu-man intervention can
be complex and challenging to recognize. A
geographic trend analysis tool was developed to aid
understanding in such circumstances. This is the
polynomial trending surface that best fits the
changing pattern. A call to a TREND module
analyzes this choice as show in figure 1a.
To Predict 2031 & 2041 Land Use Land Cover,
Clark Labs TerrSet IDRISI software was employed in
2 ways to build transition areas & transition area
probability matrices.
For LULC change analysis, the Land use/Land
cover Change module software Land Change Modeler
(LCM) is employed. The Change Module investigates
the difference among two LULC photos, namely the
previous and latter land cover photographs as show in
in figure 1b. All the Parameters should be converted to
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
386
Figure 1: (a) Changes in map and Spatial trends and (b) Change analysis’s and LCM parameters.
rst for-mat from Geo-tiff Format in Land Change
modeler.
Clark Labs TerrSet IDRISI software will estimate
LULC parameters based on previous and current
Land use/land maps to generate Change possibility
matrix reports that reflect the chance of both LULC
class transitioning to alternative session.
Secondly, a CA-Markov model has helped to
forecast the Transition in the LULC categories for
2031 and 2041. Additionally, with the use of two
LULC maps created from satellite photographs. The
model is used to determine the set of a random
process, X (t), at every point in time-period, t1, t2,
tn,tn + 1; consequently, the unplanned processes will
explain in equation 1.
(()) = , ( 1) =  1, (1) =
(()) =  (1)
tn shows the present time and tn+1 denotes time in
future; t1, t2, t3, t4……, tn 1 implies continuous
time frame moments in the previous time. According
to current realities, the future remains independent of
the previous time. Hence, the future random process
is not affected by someplace it occurs. It is not where
it used to be or where it is today. If M[k] is the
Markov chain, and xn is a group of N states (x1, x2,
x3…... xn), The chance of Transition between
condition i to condition j for single time instant is
given by Equation 2.
.
(
[
+ 1
]
=
|
[
]
=
)
(2)
The Land Change Modeler module provides three
techniques for constructing transition potential maps
associated with sub- models and independent
Parameters: a multi-layer perceptron (MLP) neuronic
network link, logistic regression, and a machine
learning tool like similarity -weighted instance (Sim-
Weight). The MLP correctly forecasts the plot that will
transfer since the picture of a subsequent stage to the
indicated simulated period, depending on the
projections. MLP surpasses alternative strategies in
estimating the correlation among nonlinear land-use /
land cover LULC changes and explanatory variables in
equation 3-4. When several transition types are
modeled, it is more versatile and dynamic than the
others.
( + 1) =

× () (3)
where 0 ≤ Pij < 1 and n ∑ j=1 Pij = 1, (i, j = 1, 2, . .
., n).
The following formula is used to definite the CA
cellular automata model:
(, + 1) = [(), ] (4)
where S(t) and S(t+1) are the organization rank at
periods t and t + 1, correspondingly. N represents the
cellular field, t, t + 1 represent distinct intervals, f
symbolizes the transforming rule of cellular conditions
in a particular region, S represents the group of
restricted and distinct cellular conditions, and Pij
represents the evolution probabilities in a phase.
5 SHANNON ENTROPY (HN)
Shannon Entropy is a commonly used metric of
spatial dispersion or concentration that is widely used
in the research of the urban sprawl phenomenon. The
Hn measurement depends on the entropy concept,
which was first designed to quantify information. It is
a valuable and dependable metric for deciding the
level of compactness & dispersion of urban
expansion.
Urban Growth Modelling Based on CA-Markov Approach on Bengaluru India
387
Figure 2: Shannon Entropy Equation & Obtained results.
Figure 3: Built-up changes and Line graph for Shannon Entropy.
Figure 4: Spatial Urban pattern LULC showing growth (a) 1991, (b) 2001,2022,2021.
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
388
Figure 5: Spatial Urban pattern LULC showing growth (a) 2021, (b)2022.
Figure 6: Model Accuracy.
Urban Growth Modelling Based on CA-Markov Approach on Bengaluru India
389
Figure 7: Bangalore Builtup area (a) 2001 (b) 1991.
Where Pi is the fraction of a geophysical
parameter in i-th zone and n symbolises the overall
sum of zones as seen in figure 2. The entropy value
can range amid 0 and log (n). A number around 0
implies a relatively tight circulation, while a value
close to log(n) show a scattered distribution. The mid-
level of log(n) is regarded as the threshold level;
hence, a city with just an entropy value higher than
the threshold value is referred to be a spreading city.
From 1991 to 2021, the most significant shift has
occurred in the Southeast direction, where the Built-
up area has urban growth as shown in figure 3-4.
We can see the Spatial Change patterns of
Bengaluru’s significant barren land from 1991 and
how it changed to a built- up area in 2021 due to urban
sprawl. We can see the patterns of change analysis of
how vegetation increased from 1991 to 2021 and how
it progressively reduced and then rose. Gains and
losses, change transitions, and change analysis are
patterns we've seen where substantial barren land has
been turned into Urban land.CA Markov is the most
common and effective modeling method for many
researchers who often use for modeling urban growth.
We forecasted forthcoming LULC land use/land
cover for 2031 and 2041 using the CA-Markov
model, and we have calculated the future Area in sq-
km of all classes. Which will aid in identifying where
and in which direction built up would increase and
which city planners can use to prepare for future
expansion.
The dynamic learning process begins with a
strong learning rate but decreases gradually over
repetitions till the last knowledge proportion is
stretched at what time the highest sum of repetitions
is extended. If a huge fluctuation in the RMS
inaccuracy is found during the first number of
reiterations, the learning rates (begin and finish) are
condensed by part, and the technique is repeated in
figure 6. LCM keeps the MLP's (Multi-layer
Perception) other variables at their default settings.
On the other hand, LCM does not make any specific
changes to outputs. Because changes are being
simulated, LCM filters out any circumstances that do
not meet the context of any given transition from the
transitional potentials. In figure 3 if the change is
from Barren to Vegetation, values will only occur in
a pixel before Barren, then Transition potential maps
are generated. The dynamic learning process begins
with a strong learning rate but decreases gradually
over repetitions till the absolute learning rate is
extended once the extreme sum of iterations is gotten.
If a enormous fluctuation in the RMS mistake is
found during the first number of iterations, the
absorbing charges (begin and finish) are abridged by
quasi, and the procedure is repeated. LCM keeps the
MLP's (Multi-layer Perception) other variables at
their default settings. On the other hand, LCM does
not make any specific changes to outputs. Because
changes are being simulated, LCM filters out any
circumstances that do not meet the context of any
given transition from the transitional potentials. In
AI4IoT 2023 - First International Conference on Artificial Intelligence for Internet of things (AI4IOT): Accelerating Innovation in Industry
and Consumer Electronics
390
figure 7 if the change is from Barren to Vegetation,
values will only occur in a pixel before Barren, then
Transition potential maps are generated.
In 1991, Barren Land accounted for around 75 %
of the overall Bengaluru District Boundary, while
Vegetation accounted for 9.2 %, and built-up area
accounted for 15 % of the total Bengaluru District
Boundary. Then, in 2001, we can see that bare land
decreased to 30.7 %, while built-up has expanded 9.2
% since 1991 and Vegetation rose exponentially to
29.2 %. Between 2001 and 2011, barren land was
reduced by 8%, and Vegetation was decreased by 4%,
resulting in an 8.2% increase in an urban area in 2011
as barren plot was transformed into the urbanized
area. From 2011 to 2021, barren land was reduced by
14.81%, and Vegetation has been increased by 2%,
resulting in a 13.9 % increase in an urban area in 2021
(figure 5-7).
Because the Accuracy of the Land Change
Modeler is 79.02 percent, we can claim it will predict
about 80 percent of the time. As shown, Bengaluru’s
future urban growth and the direction in which the
city is expanding are visible.
Between 2021 and 2031, bare land will be reduced
by 5.78%, and Vegetation will be reduced by 4%,
resulting in a 7.88% increase in urban areas in 2031.
The total built-up area will increase by 54.6 %.
Between 2031 and 2041, bare land will be reduced by
4 %, and Vegetation will be reduced by 2%, resulting
in a 6.4 % increase in urban areas in 2041. The total
built-up area will increase by 61 %.
6 CONCLUSIONS
Using Shannon Entropy, we can see that the most
substantial change from 1991 to 2021 happened in the
Southeast direction, where the Built-up region has
increased. This study concludes the challenges and
issues of urbanization in Bengaluru (Gupta J, 2022).
The solutions to these concerns are GIS data and
raster data are employed. Raster data are collected
from the google earth engine & GIS Data are gathered
from different web portals & studied various research
literature in the journal about the problem. To bring
this study to a close, qualitative, and quantitative tools
were examined. This paper explains the logical
method, which must be associated with the CA
Markov Model and the Shannon Entropy Study.
This report requires Future research of
Bengaluru’s changing spatial patterns of urban
growth. It is challenging to identify significant
differences between agriculture and parks because of
the low spatial resolution of Landsat 5 & 7 (Gupta et
al, 2015). The CA Markov model has a drawback in
that it cannot be employed for short time intervals.
While calibration is the most crucial procedure for
determining which parameters are appropriate for the
model, this model has been run more than 15 times.
Each time the parameters change, the results vary.
REFERENCES
Bhat, V., Aithal, B. H., & Ramachandra, T. V. (2015).
Spatial patterns of urban growth with globalisation in
India’s Silicon Valley. Organized By Department of
Civil Engineering, Indian Institute of Technology
(Banaras Hindu University), Varanasi-221005 Uttar
Pradesh, India, 98.
Baqa, M. F., Chen, F., Lu, L., Qureshi, S., Tariq, A., Wang,
S., ... & Li, Q. (2021). Monitoring and modeling the
patterns and trends of urban growth using urban sprawl
matrix and CA-Markov model: A case study of
Karachi, Pakistan. Land, 10(7), 700.
Mishra, V. N., Rai, P. K., & Mohan, K. (2014). Prediction
of land use changes based on land change modeler
(LCM) using remote sensing: A case study of
Muzaffarpur (Bihar), India. Journal of the
Geographical Institute" Jovan Cvijic", SASA, 64(1),
111-127.
Jain, S., Siddiqui, A., Tiwari, P. S., & Shashi, M. (2016).
Urban growth assessment using CA Markov model: A
case study of Dehradun City. 9th International
Geographic Union.
Jadawala, S., Shukla, S. H., & Tiwari, P. S. (2021). Cellular
automata and markov chain based urban growth
prediction. International Journal of Environment and
Geoinformatics, 8(3), 337-343.
Yadav, V., & Ghosh, S. K. (2021). Assessment and
prediction of urban growth for a mega-city using CA-
Markov model. Geocarto International, 36(17), 1960-
1992.
Gupta, J. (2022, November). Statistical Assessment of
Spatial Autocorrelation on Air Quality in Bengaluru,
India. In International Conference on Intelligent Vision
and Computing (pp. 254-265). Cham: Springer Nature
Switzerland.
Gupta, A. J., & Subrahmanian, R. R. Design Challenges of
Project Deliverables in Construction Industry in
Bangalore.
Urban Growth Modelling Based on CA-Markov Approach on Bengaluru India
391