Introduction
Background of the research
Like many other cities in the world Dhaka, the capital of Bangladesh, is also the outcome of spontaneous rapid growth without any prior or systematic planning. 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 urbanisation, 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 built-up lands etc.
This new urban fabric is to be analysed to understand the changes that have led to its creation. Therefore, it is necessary to track the changes of modern Dhaka City which mainly includes the changes in the physical form of the city.
Problem identification
Dhaka is now attracting a huge amount of rural-urban migrants from all over the country due to well-paid job opportunities, better educational, health and other daily life facilities (Dewan et al. 2007). This kind of increasing and over population pressure is putting adverse impacts on Dhaka city like converting wetlands/natural vegetation/open space/bare soil to urban built-up areas (Khan, 2000). All these are creating numerous problems like unplanned urbanisation, extensive urban poverty, water logging, growth of urban slums and squatters, traffic jam, environmental pollution and other socio-economic problems (Dewan et al. 2007).
Literature review
Remote Sensing (RS) and GIS techniques are being widely used to assess natural resources and monitor environmental changes. It is possible to analyse land use/ land cover change dynamics using remotely sensed data and linking it with socio-economic or bio-physical data using GIS. The incorporation of RS can help analysing this kind of research in variety of ways like land cover mapping, detecting and monitoring land cover change over time, identifying land use attributes and land cover change hot spots etc (Lambin, 2001).
With the advancement of technology, reduction in data cost, availability of historic spatio-temporal data and high resolution satellite images, RS techniques are now very useful for conducting researches like land cover change detection analysis (Das, 2009).
Objective of the research
The general objective of this research is to map, detect, quantify and analyse the land cover changes of Dhaka City over time. This research has been conducted to achieve the following broad objective:
To quantify and investigate the characteristics of urban land cover changes (1989-2009) within the study area using satellite images.
Materials
Study area profile
The proposed study area for this research is Dhaka City Corporation (DCC) and its surrounding impact areas (Figure 1). This selected study area almost covers the biggest urban agglomeration and is the central part of Bangladesh in terms of social and economic aspects. Therefore, this area has huge potentiality to face massive urban growth in near future based on the current trend of rapid urbanization.
Figure 1: Location of the study area within greater Dhaka city, Source of left map: Banglapedia, National Encyclopedia of Bangladesh, 2011.
Satellite images
This research is dependent on secondary data. To prepare the base maps for analysis purpose and applying the different methods to achieve the research objectives, Landsat satellite images (1989, 1999 and 2009) have been collected from the official website of U.S. Geological Survey (USGS). Table 1 shows the details of the Landsat satellite images used for analysis.
Table 1: Details of Landsat satellite images
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 meters. Figure 2 illustrates the location of DCC on the Landsat satellite images for different years. The surroundings of DCC have also been included within the study area for detecting land cover changes.
The Band Combination used, for the base Landsat satellite images (Figure 2), is 432 Red-Green-Blue (RGB). Map Projection used for DCC Boundary is Bangladesh Transverse Mercator (BTM) and datum is D_Everest_1830.
Figure 2: Dhaka City Corporation and its surroundings [Map Prepared by the Researcher; Source: USGS, 2010 and DCC, 2008]
Reference data
For the purpose of ground-truthing/ referencing, several base maps of Dhaka City (for the year of 1987, 1995 and 2001) have been collected from the Survey of Bangladesh (SoB). Again, for comparing the images some other reference satellite images (IRS image of 1996 and Landsat satellite image of 2003) have been collected from the Department of Urban and Regional Planning, Bangladesh University of Engineering and Technology (BUET), Dhaka, Bangladesh.
The land use maps of all the wards of DCC (2008) have been collected from DCC and Rajdhani Unnayan Kartripakkha (RAJUK) or Capital City Planning and Development Authority. Google Earth is another option to get some ideas about the recent land cover pattern of Dhaka city. These reference data have been used for training site selection and preparing land cover maps.
Base map preparation
The collected Landsat satellite images (1989, 1999 and 2009) have been used for preparing the base maps for land cover change detection analysis.
Image enhancement
Image enhancement is a kind of image modification that enables the capabilities of human vision to identify and select regions of interests (Billah et al. 2004). Composite generation technique has been performed for this particular research.
Composite generation
Landsat TM records 7 spectral bands. For visual purpose any 3 bands are combined that are acting a False Color Composite (FCC). Figure 3 shows several composites using different band combinations from the same TM images (Eastman, 2009).
Figure 3: Composites using different band combinations
Using the basic colours red, green and blue (RGB) it is possible to prepare different FCC images (Eastman, 2009). These FCC images are useful to distinguish between different cover types or ground objects like buildings, roads, and vegetation.
The FCC of RGB= bands 4, 3 and 2 has been chosen for this research. This combination normally makes urban areas appear blue, vegetation red, water bodies from dark blue to black, soils with no vegetation from white to brown (Eastman, 2009). Figure 4 shows the FCC (RGB= bands 4, 3 and 2) image of the study area for different time periods.
The study area is selected by choosing the same geographic position of all 7 bands for the same time period. This helps to maintain the same number of rows and column.
Figure 4: False color composite (RGB=4, 3 and 2) maps of the study area
Image classification
Image classification refers to grouping image pixels into categories or classes to produce a thematic representation (Canada Centre for Remote Sensing, 2010). Image classification comprehends various operations that can be applied to photographic or image data. There are two basic methods of image classification: supervised and unsupervised (Eastman, 2009). Supervised classification relies on the priori knowledge of the study area (Canada Centre for Remote Sensing, 2010). Therefore, for this research, a supervised classification method has been used.
In case of supervised classification, the user develops statistical description for various known land cover types that is called signature development. Then a procedure is used to identify the similar pixels/signature for different land cover types for the whole image.
Training site development
Training sites are the areas defined for each land cover type within the image. The chosen colour composite is used for digitising polygons around each training site for similar land cover. Then a unique identifier is assigned to each known land cover type. More than one training site for each type has been identified for making the land cover images. Five land cover types have been identified for this research (Table 2). The training sites developed for this research are based on the reference data and ancillary information collected from various sources as mentioned earlier. For developing training sites properly, all the satellite images have been analyzed with respect to spectral and spatial profiles. This is performed to make sure that the digital numbers (DNs) of different land cover types are acceptable prior to final classification (Eastman, 2009).
Table 2: Details of the land cover types
Signature development
This is the stage of creating the spectral signature for each type of land cover. This is done by analysing the pixels of the training sites. When the digitisation of training sites is finished, the statistical characterisations of each land cover class are needed. These are called signatures. Signatures are developed incorporating the vector files containing training sites and the bands used for analysis. These signature files contain statistical information about the reflectance values of the pixels of the training sites for each land cover type (Eastman, 2009).
Classification After developing signature files for all land cover classes the next step is to classify the images based on these signature files. This can be done by: hard or soft classifiers. These are kind of statistical techniques to classify the whole image pixel by pixel based on each known type particular signature (Eastman, 2009).
For this research, a hard classifier called ‘Fisher Classifier’ has been chosen. Fisher classifier uses the concept of the linear discrimination analysis (Eastman, 2009). Fisher Classifier performs well when there are very few areas of unknown classes and when the training sites are representative of their informational classes (Eastman, 2009). This is why fisher classifier is appropriate for this particular research, because most areas for the classes are known.
Generalisation After image classification, sometimes many isolated pixels may be found (Eastman, 2009). These isolated pixels may belong to one or more classes that differ from surrounding pixels. Therefore, it is necessary to generalise the image and remove the isolated pixels.
Filtering is the solution for this type of problem. In RS analysis, filtering refers to the removal of certain spectral or spatial frequencies to highlight features in the remaining image. Mode filters are good for filling gaps between polygons after a vector-to-raster conversion (Eastman, 2009).
Therefore, a 3×3 mode filter has been applied to generalize the fisher classified land cover images. This post-processing operation replaces the isolated pixels to the most common neighbouring class. Finally the generalized image is reclassified to produce the final version of land cover maps for different years (Figure 5).
Figure 5: Land cover maps of Dhaka city
Accuracy assessment
The final stage of image classification process is accuracy assessment. Accuracy Assessment is a kind of process to compare the classification with ground truth or other data. It allows evaluating a classified image file (Eastman, 2009).
It is not typical to ground truth each and every pixel of the classified image. Therefore some reference pixels are generated. These are points on the classified image. Each point of reference pixels represents specific geographic coordinate of the image. These reference pixels are randomly selected (Congalton, 1991).
The randomly selected points within the classified image list two sets of class. The first set of class values represents the actual land cover type. The second set of class values are known as reference values. The ground truthing of reference values are possible through field visit, comparing base maps, aerial photos, previously tested maps or other data (ERDAS IMAGINE, 2006). The number of reference pixels is an important issue for accuracy assessment. It has been proved that more than 250 reference pixels (± 5%) are needed to estimate the mean accuracy of classification (Congalton, 1991). Therefore 250 reference pixels have been generated; using stratified random distribution process, for each classification image to perform accuracy assessment.
Assessment procedure
The collected base maps have been used to find the land cover types of the reference points. Figure 6 shows the randomly selected 250 points for each base year. From the accuracy assessment cell array, three types of reports are generated (ERDAS IMAGINE, 2006). These are:
- Error Matrix: it compares the reference points to the classified points in a c×c matrix, where c is the number of land cover classes.
- Accuracy Totals: it calculates statistics of the percentages of accuracy that is based on the error matrix and
- Kappa Coefficient.
Figure 6: Selected stratified random points for ground truthing
Accuracy results
User’s accuracy measures the proportion of each land cover class which is correct. On the other hand, producer’s accuracy measures the proportion of the land base which is correctly classified (ERDAS IMAGINE, 2006). Producer’s and User’s accuracy are also found consistently high ranging from 71%-100% (Table 3, 4 and 5) for all the years. Therefore, higher values of producer’s and user’s accuracy indicate that the prepared base maps are quite good enough for further analysis.
The overall accuracies for 1989, 1999 and 2009 are found 85.20%, 86.80% and 91.60% respectively, with Kappa statistics of 0.8054, 0.8294 and 0.8592.
It means few misclassifications have been observed in the classified land cover maps of Dhaka city. This may be because of the same spectral characteristics of some land cover types. For example, in case of 1989 base map, certain built-up 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. 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.
It is typical that most land cover classification images are 85% accurate (Eastman, 2009). Therefore, it can be stated that the classification accuracy achieved for this research is above satisfactory level.
Table 3: Accuracy totals (1989)
Table 4: Accuracy totals (1999)
Table 5: Accuracy totals (2009)
Change detection analysis
Change detection is important because it helps the researcher to understand and monitor the land cover change pattern (e.g. urbanisation, deforestation, agricultural land management) within the study area.
Change in area
It is clear that over the years (1989 to 2009) built-up area has increased in huge percentage (from 8.4% to 46%). It is also noteworthy that fallow has decreased in good rate (from 38% to 17%). Other land cover types have decreased in a very small amount (Figure 7).
Figure 7: Land cover change in area (%)
Gains and losses by category
Figure 8 illustrates that built-up area has increased over the years while there is slight loss in this category. It means some parts of the previously existed built-up areas have converted to some other land cover classes, while vast new area has transformed into built-up area from other classes.
Gains in built-up area are evident in all three combinations. Again fallow land cover type is decreasing in large percentage in all the years. The changes (in terms of gains and losses) in other land cover types are almost the same or not influencing. Therefore an abrupt increase in built-up area and decrease in fallow land cover type is quite clear from this kind of analysis (Figure 8).
Figure 8: Gains and losses of land covers by category (unit: % of area)
Contributors to net change experienced by built-up area
Figure 9 illustrates which land cover type is contributing more to net change in built-up area. It is found that fallow land is contributing most converting towards built-up area followed by water body and vegetation.
Figure 9: Contributors to net change experienced by builtup area (Unit: % of area)
Gains and losses in land cover types
In case of built-up area, the core southern part of Dhaka city has remained the same. While the north-east and south-west parts have converted to built-up areas. The northern part of Dhaka city has gained water body followed by a massive decrease in the south-east and south-west parts (Figure 10).
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 north-western and mid parts (Figure 10).
Figure 10: Gains and losses in land cover types (1989-2009)
Summary of land cover change detection analysis
In summary, it can be stated that in any combinations of time periods (1989-1999 or 1999-2009 or 1989-2009) some similarities have been found. Those are as follows:
- The increase in built-up area is prominent.
- The net contribution to built-up area is mainly from fallow land followed by water body.
- The expansion of urban area is following the northern and western portions of Dhaka city.
- The growth of urbanisation is haphazard indicating the absence of proper planning.
Moreover, fallow land and water body types are basically converting into built-up areas over the years. This is the general trend of land cover change pattern for Dhaka city from 1989-2009.
Conclusion
Dhaka, as the capital city of Bangladesh, encounters a host of problems perpetuating for decades. With the growing importance of the city and mounting up pressure of population, the problems are getting more and more precarious every year. Amid all the problems, uncontrolled urban growth/ development is perhaps the most irritating one particularly for the planning and development control authorities like DCC and RAJUK.
The unwanted urban growth originates not only due to ineffective development control but also due to failure of the equilibrium between demand and supply of buildable and liveable urban land. Therefore, regular monitoring of the plan implementation is necessary together with the urban development trend in new areas. An early measure in tackling problems can not only save huge public money but also reduce the miseries of the city dwellers. In this regard, performing this kind of change detection analysis, using Remote Sensing Techniques, to monitor the urban growth is highly recommended.
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 the modern and much liveable Dhaka City.
Acknowledgement
This research article is a part of the M.Sc. thesis conducted by the author. The author would like to thank Dr. Raquib Ahmed, Professor and Chairman, Department of Geography & Environmental Studies, University of Rajshahi, Bangladesh for his thoughtful suggestions and enduring guidance at different stages of this research. The author would also like to express gratitude to Dr. Pedro Latorre Carmona, Dr. Mário Caetano, Dr. Edzer Pebesma and Dr. Nilanchal Patel for their support and comments regarding necessary corrections.
References
- Billah, M. and Rahman, G. A 2004, Land Cover Mapping of Khulna City Applying Remote Sensing Technique. In: Proceedings of the 12th International Conference on Geoinformatics (Sweden: Geospatial Information Research: Bridging the Pacific and Atlantic, University of Gävle), pp. 707-714.
- Canada Centre for Remote Sensing, Natural Resources Canada, Ottawa, Ontario, K1A 0Y7, , explored on 15-11-2010.
- Congalton, R. 1991, A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sensing of Environment, 37, 35-46.
- Das, T. 2009, Land Use / Land Cover Change Detection: an Object Oriented Approach, Münster, Germany. Thesis: Master of Science in Geospatial Technologies, Institute for Geoinformatics, University of Münster, Germany.
- Dewan, A. M. and Yamaguchi, Y. 2009, Landuse and Land Cover Change in Greater Dhaka, Bangladesh: Using Remote Sensing to Promote Sustainable Urbanization. Journal of Applied Geography, 29, 390–401.
- Eastman, J. R. 2009, IDRISI Taiga Guide to GIS and Image Processing (Manual Version 16.02) [Software] (Massachusetts, USA: Clark Labs, Clark University).
- Khan, N. S. 2000, Temporal Mapping and Spatial Analysis of Land Transformation Due to Urbanization and its Impact on Surface Water System: A Case from Dhaka Metropolitan Area, Bangladesh. International Archives of Photogrammetry and Remote Sensing, Vol. XXXIII, Part B7, Amsterdam.
- Lambin, E.F. 2001, Remote Sensing and GIS Analysis. In: International Encyclopedia of the Social and Behavioral Sciences, edited by N.J. Smelser and P.B. Baltes (Oxford: Pergamon), pp. 13150-13155.
- The ERDAS IMAGINE 9.1 On-Line Help System, 2006, Leica Geosystems Geospatial Imaging, LLC [Software] (Worldwide Headquarters, 5051 Peachtree Corners Circle, Norcross, GA 30092-2500, USA).
