This article is an openaccess article distributed under the terms and conditions of the Creative Commons Attribution license (
The primary objective of this research is to predict and analyze the future urban growth of Dhaka City using the Landsat satellite images of 1989, 1999 and 2009. Dhaka City Corporation (DCC) and its surrounding impact areas have been selected as the study area. At the beginning, a fisher supervised classification method has been applied to prepare the base maps with five land cover classes. In the next stage, three different models have been implemented to simulate the land cover map of Dhaka city of 2009. These have been named as “Stochastic Markov (St_Markov)” Model, “Cellular Automata Markov (CA_Markov)” Model and “Multi Layer Perceptron Markov (MLP_Markov)” Model. Then the bestfitted model has been selected by implementing a method to compare land cover categories in three maps: a reference map of time 1, a reference map of time 2 and a simulation map of time 2. This is how the “Multi Layer Perceptron Markov (MLP_Markov)” Model has been qualified as the most appropriate model for this research. Later, using the MLP_Markov model, the land cover map of 2019 has been predicted. The MLP_Markov model extrapolates that builtup area increases from 46% to 58% of the total study area during 2009–2019.
Like many other cities in the world, Dhaka, the Capital of Bangladesh is also the outcome of spontaneous rapid growth. As the growth of population in Dhaka is taking place at an exceptionally rapid rate, it has become one of the most populous Mega Cities in the world.
Dhaka City has undergone radical changes in its physical form, not only in its vast territorial expansion, but also through internal physical transformations over the last decades. These have created entirely new kinds of urban fabric. In the process of urbanization, the physical characteristics of Dhaka City are gradually changing as plots and open spaces have been transformed into building areas, open squares into car parks, low land and water bodies into reclaimed builtup lands,
Dhaka is now attracting a huge amount of ruralurban migrants from all over the country due to wellpaid job opportunities, better educational, health and other daily life facilities. This kind of increasing and over population pressure is putting adverse impacts on Dhaka City like unplanned urbanization, extensive urban poverty, water logging, growth of urban slums and squatters, traffic jams, environmental pollution and other socioeconomic problems [
If this situation continues then Dhaka would soon become an urban slum with the least livable situation for the city dwellers. Therefore, the primary objective of this paper is to forecast the future urban land cover changes of the selected study area within the greater Dhaka City.
With the advancement of technology, reduction in data cost, availability of historic spatiotemporal data and high resolution satellite images, Remote Sensing (RS) and Geographic Information System (GIS) techniques are now very useful for conducting researches like land cover change detection analysis and predicting the future scenario [
Many researchers have conducted number of researches to detect the land use/land cover change pattern over time and predict the future growth of urban areas. They have introduced and applied different techniques to achieve the research objectives. Among them, Griffiths
Dewan and Yamaguchi have tried to evaluate land cover changes and urban expansion in greater Dhaka, between 1975 and 2003 using satellite images and socioeconomic data. A supervised classification algorithm and the postclassification change detection technique in GIS have been implemented by them. They have found the accuracy of the Landsatderived land cover maps ranged from 85% to 90% [
Emch and Peterson have quantified mangrove forest cover change in the Sundarbans of southwest Bangladesh from 1989 to 2000 using Landsat Thematic Mapper (TM) satellite imagery. They have used three image processing techniques: Normalized Differential Vegetation Index (NDVI), maximum likelihood classification and subpixel classification [
Kashem has implemented SLEUTH urban growth model to simulate the historical growth pattern of Dhaka Metropolitan Area. SLEUTH model incorporates Slope, Landuse, Exclusion layer (where growth cannot occur), Urban, Transportation and Hillshade data layers. SLEUTH uses a modified Cellular Automata (CA) to model the spread of urbanization [
Wang and Mountrakis have developed a GISbased modeling framework titled MultiNetwork Urbanization (MuNU) model, which integrates multiple neural networks, to predict the urban growth of Denver Metropolitan Area, CO, USA [
Dhaka is located in central Bangladesh at 23°43′0″N, 90°24′0″E, on the eastern banks of the Buriganga River (
To prepare the base maps for analysis purpose and applying the different methods to achieve the research objectives, the Landsat satellite images (1989, 1999 and 2009) have been collected from the official website of US Geological Survey (USGS). Landsat Path 137 Row 44 covers the whole study area. Map Projection of the collected satellite images is Universal Transverse Mercator (UTM) within Zone 46 N–Datum World Geodetic System (WGS) 84 and the pixel size is 30 m. Five land cover types have been identified for this research (
(
Details of the land cover types.
Land Cover Type  Description 


All residential, commercial and industrial areas, villages, settlements and transportation infrastructure. 

River, permanent open water, lakes, ponds, canals and reservoirs. 

Trees, shrub lands and semi natural vegetation: deciduous, coniferous, and mixed forest, palms, orchard, herbs, climbers, gardens, innercity recreational areas, parks and playgrounds, grassland and vegetable lands. 

Permanent and seasonal wetlands, lowlying areas, marshy land, rills and gully, swamps, mudflats, all cultivated areas including urban agriculture; crop fields and ricepaddies. 

Fallow land, earth and sand land infillings, construction sites, developed land, excavation sites, solid waste landfills, open space, bare and exposed soils. 
The Landsat satellite images used for analysis are of different qualities and dates (
Details of the Landsat satellite images. (Source: US Geological Survey, 2011).
Respective Year  Date Acquired (Day/Month/Year)  Sensor 

1989  13/02/1989  Landsat 4–5 Thematic Mapper (TM) 
1999  24/11/1999  Landsat 7 Enhanced Thematic Mapper Plus (ETM+) 
2009  26/10/2009  Landsat 4–5 Thematic Mapper (TM) 
Supervised classification relies on the
Training sites are the areas defined for each land cover type within the image. The chosen color composite is used for digitizing polygons around each training site for similar land cover. Then a unique identifier is assigned to each known land cover type [
This is the stage of creating the spectral signature for each type of land cover. This is done by analyzing the pixels of the training sites. When the digitization of training sites is finished, the statistical characterizations of each land cover class are needed. These are called signatures [
After developing signature files for all the land cover types the images have been classified using a hard classifier called “Fisher Classifier”. Fisher classifier uses the concept of the linear discrimination analysis [
After image classification, sometimes many isolated pixels may be found [
Land cover maps of the study area.
The next stage of image classification process is accuracy assessment. It is not typical to ground truth each and every pixel of the classified image. Therefore some reference pixels are generated [
User’s accuracy for category K is the percent of category K in the reference information, given that the map shows category K. Producer’s accuracy for category K is the percent of category K in the map, given that the reference information shows category K [
Accuracy totals (1989).
Class Name  Reference Totals  Classified Totals  Number Correct  Producer’s Accuracy  User’s Accuracy 


18  21  18  100.00%  85.71% 

46  38  35  76.09%  92.11% 

66  61  54  81.82%  88.52% 

43  35  31  72.09%  88.57% 

77  95  75  97.40%  78.95% 




Accuracy totals (1999).
Class Name  Reference Totals  Classified Totals  Number Correct  Producer’s Accuracy  User’s Accuracy 


62  72  60  96.77%  83.33% 

28  24  20  71.43%  83.33% 

56  53  47  83.93%  88.68% 

33  30  29  87.88%  96.67% 

250  250  217  85.92%  85.92% 




Accuracy totals (2009).
Class Name  Reference Totals  Classified Totals  Number Correct  Producer’s Accuracy  User’s Accuracy 


112  115  108  96.43%  93.91% 

22  19  17  77.27%  89.47% 

43  45  41  95.35%  91.11% 

28  28  24  85.71%  85.71% 

45  43  39  86.67%  90.70% 




The overall accuracy represents the percentage of correctly classified pixels [
There is 14.8%, 13.2% and 8.4% reported map error in the maps of 1989, 1999 and 2009 respectively. These errors in the maps are clearly not negligible. It is a very common problem in land change science that the amount of error in the maps is nearly as large as the amount of change on the ground. Thus it is necessary to compare the error in the maps with how much difference there is during 1989–1999 and 1999–2009.
The amount of land change ranges from approximately 2% to 20% for 1989–1999, while the figure is approximately 1% to 17% for 1999–2009 (
Amount of differences (1989–1999).
Land Cover Type  1989  1999  Change in Area (%)(1989–1999)  

Area (km^{2})  %  Area (km^{2})  %  

37.6569  8.447  129.2292  28.991 


67.3218  15.102  42.5034  9.535 


109.5714  24.581  94.3929  21.175 


62.0640  13.923  53.7264  12.052 


169.1559  37.947  125.9181  28.247 







Amount of differences (1999–2009).
Land Cover Type  1999  2009  Change in Area (%)(1999–2009)  

Area (km^{2})  %  Area (km^{2})  %  

129.2292  28.991  204.4008  45.853 


42.5034  9.535  33.6645  7.552 


94.3929  21.175  79.9578  17.937 


53.7264  12.052  49.9914  11.215 


125.9181  28.247  77.7555  17.443 







Finally it can be stated that few misclassifications have been observed in the classified land cover maps of Dhaka city. The reasons may be as follows:
The same spectral characteristics of some land cover types. For example, in case of 1989 base map, certain builtup areas were misclassified as fallow land. Again, in most cases it was really difficult to separate water bodies and low/cultivable lands categories. The reasons may be the seasonal variations of the satellite images for different years and the similar spectral properties of land covers in some cases. The images collected for 1999 (November) and 2009 (October) are from the same winter season. But the image of 1989 (February) is from another season, summer. This kind of variation creates problems while preparing base maps for analysis.
Moreover, less image spectral resolution has directed to spectral mixing of different land cover types. This has caused spectral confusions among the cover types. It is also important to mention that the images of 1999 and 2009 represent winter season while the image of 1989 represents spring season. Therefore other seasonal images can be important evaluating the land cover change pattern of this kind of highly dynamic urban environment like Dhaka.
Again the spatial resolution of the images is important. For this research purpose, Landsat satellite images have been chosen that are only commercially available but can be found in free publicdomain. The main problem of working with Landsat images is low resolution. The spatial resolution of Landsat Image is 30 m. IKONOS, QuickBird or other satellite images with higher resolution can be better option.
The next limitation regarding this research is the collection of reference data or maps. The reference data are necessary for ground truthing purpose of the base maps (1989, 1999 and 2009) that have been prepared from the Landsat satellite images. But reference maps of the respective years (1989, 1999 and 2009) are not available. Therefore the base maps of Dhaka city of the years 1987, 1995 and 2001, collected from Survey of Bangladesh (SoB), have been used for referencing purpose. Google Earth images (2010) are used as reference data for ground truthing the base map of 2009.
This kind of research needs extensive field visit for image classification and assessing the accuracies of the satellite images. Another point is the verification of the older images. For older images it is not possible to visit the field to find out the actual land cover types. These things can be improved by recent field visit to collect Global Positioning System (GPS) data for land cover verification/ground truthing purpose. To solve the problems with older satellite images, the historical base maps of similar years should be collected.
The overall accuracies of the base maps can be achieved better if the above mentioned limitations can be reduced.
In remote sensing, “Change Detection” is defined as the process of determining and monitoring the changes in the land cover types in different time periods. It provides the quantitative analysis of the spatial distribution in the area of interest [
NP equals the number of patches of the corresponding patch type (class). Number of patches of a particular class or land cover type is a simple measure of the extent of subdivision or fragmentation of the class [
Number of patches over the years.
In case of builtup area, the core southern part of Dhaka city has remained the same. While the northeast and southwest parts have converted to builtup areas. The northern part of Dhaka city has gained water body followed by a massive decrease in the southeast and southwest parts (
Gains and losses in land cover types (1989–2009).
No particular pattern on gains or losses is found for vegetation. In cases of low land the changes are evident in eastern and western parts. Fallow land has decreased markedly and the losses are clear in northwestern and mid parts (
The first model that has been implemented is given the name as “Stochastic Markov Model (St_Markov)”, because this model combines both the stochastic processes as well Markov chain analysis techniques [
A Markov chain is a stochastic process (based on probabilities) with discrete state space and discrete or continuous parameter space [
In a Markov chain the probability of the next state is only dependent upon the current state. This is called Markov property and stated as [
The probability of a Markov chain ξ_{1}, ξ_{2},........can be calculated as [
The conditional probabilities:
These are called the “Transition Probabilities” of the Markov chain [
Let’s consider a Markov chain with n states s_{1}, s_{2},.......,s_{n}. Let p_{ij} denote the transition probability from state s_{i} to state s_{j},
The transition matrix (n × n) of this Markov process is then defined as [
Predictions of the future state probabilities can be calculated by solving the matrix equation [
p(t) = p(t − 1)∙P (6)
With increasing time steps, a Markov chain may approach to a constant state probability vector, which is called limiting distribution [
p(∞) =
At the beginning, Markov chain produces a transition matrix (
Markov probability of changing among land cover types (1989–1999).
Builtup Area  Water Body  Vegetation  Low Land  Fallow Land  


0.6649  0.0268  0.0533  0.0298  0.2252 

0.2125  0.1074  0.1030  0.1969  0.3802 

0.1675  0.0853  0.3304  0.1173  0.2995 

0.0766  0.4006  0.0514  0.3446  0.1267 

0.4126  0.0144  0.2603  0.0199  0.2928 
Cells expected to transition to different classes (1989–1999).
Builtup Area  Water Body  Vegetation  Low Land  Fallow Land  


95,476  3,845  7,658  4,278  32,332 

10,034  5,074  4,865  9,300  17,953 

17,569  8,945  34,655  12,302  31,409 

4,574  23,914  3,070  20,572  7,566 

57,732  2,009  36,415  2,789  40,964 
The matrix of transition probabilities (
After analyzing
Markovian conditional probability images.
The next step is to make one single land cover map for future prediction aggregating all the Markovian conditional probability images. This prediction is performed by a stochastic choice decision model. Stochastic choice creates a stochastic land cover map by evaluating and aggregating the conditional probabilities in which each land cover can exist at each pixel location against a rectilinear random distribution of probabilities [
Stochastic Choice creates a stochastic land cover map by evaluating and aggregating the conditional probabilities in which each land cover can exist at each pixel location against a rectilinear random distribution of probabilities [
The second model that has been implemented is named as “Cellular Automata Markov Model (CA_Markov). CA_Markov combines the concepts of Markov Chain, Cellular Automata (CA) [
Stephen Wolfram defined CA as follows: “
Final St_Markov predicted land cover map (2009).
The components that comprise an elementary cellular automaton are as follows [
(a) The physical environment or the space represented by an array of cells, on which the automaton exists (its lattice).
(b) The cell in which the automaton resides that contains its state(s).
(c) The neighborhood around the automaton.
(d) Transition rules that describe the behavior of the automaton.
(e) The temporal space in which the automaton exists.
A CA model represents discrete dynamic system consisting of four elements [
where,
It shows that the state of the
Here,
The local transition rule is given by a rule table where given the sizes of ∑ and
where each of the
Now applying the local transition rule to all the cells in the CA’s lattice, the next configuration of the CA can be computed by its induced global map [
In brief, the standard CA can be generalized as follows [
where,
CA_Markov is useful for modeling the state of several categories of a cell based on a matrix of Markov transition areas; transitional suitability images and a user defined contiguity filter. A Markov model applies contiguity rule like a pixel near to an urban area is most likely to be changed into urban area [
The 3 × 3 mean contiguity filter for CA_Markov modelling.
The suitability maps determine which pixels will change as per the highest suitability of each land cover type. The higher the suitability of a pixel, the possibility of the neighboring pixels to change into that particular class is higher.
Preparing a suitability map for each land cover type is difficult in terms of data and information availability. It is not possible to incorporate all types of factors or constraints that exist within the study area. Therefore a simple assumption has been assumed for fuzzy factor standardization.
The basic assumption for preparing suitability images is the pixel closer to an existing land cover type has the higher suitability. It means a pixel that is completely within vegetation has the highest suitability value (255) and pixels far from existing vegetation pixels will have less suitability values. The farthest pixels from vegetation will show the lowest suitability values. Here the suitability decreases with distance. Though this idea is not perfect always, like in case of water body and vegetation the scenario can be different. In reality, a pixel near to forest is more likely to convert into builtup area rather in forest. But this can be considered to be the research limitation.
Therefore a simple linear distance decay function is appropriate for this basic assumption. It serves the basic idea of contiguity. The land cover maps have been standardized (
Suitability images of each land cover type (1999).
At the end, the Markov transition area matrix (
CA_Markov projected land cover map of Dhaka city (2009).
The term “Artificial Neural Network (ANN)” has been inspired by human biological nervous system [
MLP neural network uses the back propagation (BP) algorithm. The calculation is based on information from training sites [
where, w_{ij} = the weights between node i and node j; O_{i} = the output from the node i
The output from a given node j is computed as [
f = a nonlinear sigmoid function that is applied to the weighted sum of inputs before the signal passes to the next layer
This is known as “Forward Propagation”. Once it is finished, the activities of the output nodes are compared with their expected activities. In normal circumstances, the network output differs from the desired output (a set of training data, e.g., known classes). The difference is termed as the error in the network [
η = the learning rate parameter; δ_{j} = an index of the rate of change of the error; α = the momentum parameter.
The process of the forward and backward propagation is repeated iteratively, until the errors of the network minimized or reaches an acceptable magnitude [
In general, the larger the number of nodes in the hidden layer, the better the neural network represents the training data [
where, N_{h} = the number of hidden nodes; N_{i} = the number of input nodes; N_{o} = the number of output nodes
The number of training sample also affects the training accuracy. Too few samples may not represent the pattern of each category while too many samples may cause overlap. Again too many iterations can cause over training that may cause poor generalization of the network [
where, N = the number of elements; i = the index for elements; e_{i} = the error of the ith element; t_{i} = the target value (measured) for i^{th} element; a_{i} = the calculated value for the i^{th} element.
The basic concept of modeling with MLP neural network adopted in this research is to consider the change in builtup area over the years. In general, it means other land cover types are primarily contributing to increase the builtup area. At this stage, the issue of which variables affect the change to builtup area (1989–1999) has been considered. Therefore, only the transitions from “water body to builtup area”, “vegetation to builtup area”, “low land to builtup area” and “fallow land to builtup area” have been considered for model simulation. These four transitions have been termed as “All” here.
Transition from all to builtup area (1989–1999).
It is logical that new areas will be converted to builtup area where there are existing builtup areas. Therefore six driver variables have been selected for MLP_Markov modeling. These are (1989–1999): Distance from all to builtup area, distance from water body, distance from vegetation, distance from low land, distance from fallow land and empirical likelihood image).
The empirical likelihood transformation is an effective means of incorporating categorical variables into the analysis (
Now it is important to test the potential explanatory power of each variable. The quantitative measures of the variables have been tested through Cramer’s V [
After getting satisfactory Cramer’s V values for all the driving variables, now the turn is to run MLP neural network model. For this purpose, 10,000 iterations have been chosen. The minimum number of cells that transitioned from 1989 to 1999 is 4,794. Therefore, the maximum sample size has been chosen as 4,794. For each principal transition particular weights have to be obtained. The RMS error curve has been found smooth and descent after running MLP neural network. After all these combinations, the MLP running statistics gives a very high accuracy rate of 91.36% (this accuracy is a measure of calibration, not validation). Based on these running statistics the transition potential maps have been produced (
Empirical likelihood image of changing into builtup areas (1989–1999).
Transition potential maps from all to builtup area (1989 to 1999).
Using this kind of MLP neural network analysis it is possible to determine the weights of the transitions that will be included in the matrix of probabilities of Markov Chain for future prediction. The transition probabilities are shown in
MLP_Markov projected land cover map of Dhaka City (2009).
Transition probabilities grid for Markov chain (1989 to 1999) in MLP modeling.
Builtup Area  Water Body  Vegetation  Low Land  Fallow Land  


0.7823  0.0174  0.0347  0.0194  0.1463 

0.2079  0.1264  0.1008  0.1927  0.3721 

0.1529  0.0779  0.3887  0.1071  0.2734 

0.0695  0.3634  0.0467  0.4054  0.1150 

0.3825  0.0133  0.2413  0.0185  0.3445 
Now the task is to select the most appropriate model. In general sense, model validation refers to comparing the simulated and reference maps. But the traditional way of validating a model or comparing two maps, using Kappa statistics, is now outofdate [
Therefore, a method of comparing three maps (a reference map of time 1, a reference map of time 2 and a simulation map of time 2) has been implemented for model validation [
The three maps comparison method consists of two components of agreement and three components of disagreement. The components of agreement are persistence simulated correctly and change simulated correctly [
Among the three models, the percentages of disagreement components are lowest (28.066%) while the percentages of agreement components (71.934%) are found highest for MLP_Markov (
Components of agreement and disagreement for model validation.
Name of Component  St_Markov (%)  CA_Markov(%)  MLP_Markov (%) 

Persistence Simulated Correctly  44.424  48.459  50.578 
Change Simulated Correctly  18.003  19.196  21.356 




Change Simulated As Persistence  11.369  10.39  9.788 
Persistence Simulated As Change  16.736  13.54  10.751 
Change Simulated As Change to Wrong Category  9.468  8.415  7.527 




This is how, implementing the three maps comparison method, it is found that the simulated map of “MLP_Markov 2009” is showing the best results in terms of the percentages of disagreement and agreement components. This is why; MLP_Markov is giving the best outputs among the three modeling techniques.
Moreover, if we look at
Therefore, at the end, the MLP_Markov model has been selected for predicting the land cover map of Dhaka City for the year of 2019.
Comparison of the three models based on basic modeling elements.
Name of Component  St_Markov  CA_Markov  MLP_Markov 

Temporal Dependency  √  √  √ 
Spatial Dependency  ×  √  √ 
Spatial Distribution  ×  √  √ 
Spatial Proximity  ×  √  √ 
Suitability Analysis  ×  √  √ 
Transfer Function  ×  ×  √ 
Weights of the Connections  ×  ×  √ 
Training/ Testing Accuracy Assessment  ×  ×  √ 
Transition Potential Maps  ×  ×  √ 
Maps of the components of agreement and disagreement.
The base maps of 1999 and 2009 have been used to predict the land cover map of 2019 (
MLP_Markov projected land cover map of Dhaka City (2019).
The predicted map of 2019 reveals that 58% of the total area will be occupied by the “builtup area” cover type (
Percentages of presence of land cover types over the years (1989–2019).
This research has revealed that the increase in builtup area is prominent in Dhaka City over the years (1989–2009). Later this paper has presented three different methods to simulate the land cover map of 2009 being persistent with the inherent changing characteristics. The methods have been named as “Stochastic Markov (St_Markov)”, “Cellular Automata Markov (CA_Markov)” and “Multi Layer Perceptron Markov (MLP_Markov)” model. Then the MLP_Markov model has been found most appropriate, based on the three map comparison method, for future prediction (the land cover map of 2019).
Our hope might be realized if the error in the base maps is reduced to the point where the error becomes smaller than apparent change in land. We hope the interpretation of depicting the future scenario in quantitative accounts, as demonstrated in this research, will be of great value to the urban planners and decision makers, for the future planning of a modern and habitable Dhaka City. Moreover, it is our belief that this kind of research has a high potential to contribute towards the sustainable urban growth both at any local and regional level in the world.
This research work is a part of the M.Sc. thesis conducted by the corresponding author which has been supported by the European Commission, Erasmus Mundus Programme, M.Sc. in Geospatial Technologies, Framework Partnership Agreement 20070064/001 FRAME MUNB123. The author would like to thank Pedro Latorre Carmona, Mário Caetano, Edzer Pebesma, Nilanchal Patel and Filiberto Pla Bañón for their thoughtful suggestions and enduring guidance at different stages of this research.
The author would also like to express gratitude to the European Commission and Erasmus Mundus Consortium (Universitat Jaume I, Castellón, Spain; Westfälische WilhelmsUniversität, Münster, Germany and Universidade Nova de Lisboa, Portugal) for providing the fund and other research opportunities.
Finally we thank the anonymous reviewers and the editors for their constructive comments that improved the quality of this paper.