Figure 1: Study area - Kolkata administrative boundary with 10 km buffer.

Kolkata is a highly populated city in India that has a population of 14.11 million as per provisional census of 2011. Figure 2 describes the temporal increase in population during the last 100 years.

Figure 2:  Population growth since 1901 (Source: Census of India).

Table 1:  Population growth in Kolkata during 1971- 2011

Data: Urban dynamics was analysed using temporal remote sensing data for the period between 1973 and 2010. The time series spatial data acquired from Landsat Series Multispectral sensor (57.5m) and thematic mapper (28.5m) sensors for the period between 1979 and 2010 were downloaded from a public domain (http://glcf.umiacs.umd.edu/data) as indicated in Table 2. Base layers and training data (for classifying remote sensing data) were derived from the Survey of India (SOI) topographic maps of scales 1:250000 and 1:50000. Features such as extent of water bodies, certain roads etc., were derived from 1:50000 map, while 1:250000 was used mainly for training data. Further, these layers were resampled to 30 m.

Table 2:  Data used in the Analysis

Data Analyses: Figure 3 outlines the process of data analyses, which includes preprocessing, analysis of vegetation cover and land use, and lastly, the gradient wise zonal analysis of Kolkata using landscape metrics.

Figure 3: Procedure adopted for classifying landscape and computation of metrics.

• Preprocessing: Remote sensing data (Landsat series) for Kolkata of different time periods were downloaded from Global Land Cover Facility (http://www.glcf.umd.edu/index.shtml and http://www.landcover.org/), United States Geological Survey (USGS) Earth Explorer (http://edcsns17.cr.usgs.gov/NewEarthExplorer/) and Glovis (http://www.glovis.usgs.gov). The remote sensing data obtained were geo-referenced, geo-corrected, rectified and cropped with respect to the study area. Geo-registration of remote sensing data (Landsat data) was done using ground control points collected from the field using pre-calibrated GPS receiver (Global Positioning System) and also from known points (such as road intersections, etc.) collected from geo-referenced topographic maps published by the Survey of India. The Landsat satellite data of 1979 (with spatial resolution of 57.5 m x 57.5 m (nominal resolution)) and that of 1989 - 2010 (with spatial resolution of 28.5 m x 28.5 m (nominal resolution)) were resampled to 30m each in order to maintain uniformity in spatial resolution across different time periods. The study area included the Kolkata administrative area with 10 km buffer.

Land use analysis was carried out using supervised pattern classifier - Gaussian maximum likelihood algorithm. Remote sensing data was classified using signatures from training sites that include all the land use types detailed in table 3. Mean and covariance matrix were computed using estimates of maximum likelihood estimator. Maximum Likelihood classifier is used to classify the data using these signatures generated. This technique is proven to be a superior classifier as it uses various classification decisions using probability and cost functions (Duda et al., 2000, Ramachandra et al., 2012). Mean and covariance matrix are computed using estimates of maximum likelihood estimator. Land uses during different time periods were estimated using the temporal data through open source program GRASS - Geographic Resource Analysis Support System (http://ces.iisc.ernet.in/grass). Signatures were collected from field visits and with the help of Google Earth. 60% of the total generated signatures were used in classification, and 40% signatures were used in validation and accuracy assessment.  Classes of the resulting image were reclassified and recoded to form four land-use classes. The noise in the classified image of 1973 was removed by smoothing filter (3 X 3 median filters).

Table 3: Land use classes

• Accuracy assessment: These methods evaluate the performance of classifiers (Mitrakis et al., 2008).  This is done through comparison of kappa coefficients (Congalton et al., 1983), which are common measurements used to demonstrate the effectiveness of the classifications (Congalton, 1991; Lillesand & Kiefer, 2002). Recent remote sensing data (2010) were classified using the collected training samples. Statistical assessment of classifier performance, based on the performance of spectral classification considering reference pixels, was done which included computation of kappa (κ) statistics and overall (producer's and user's) accuracies. For earlier time data, training polygon along with attribute details were compiled from earlier published topographic maps, vegetation maps, and revenue maps.

Division of zones into concentric circles (Gradient Analysis): Each zone was divided into a concentric circle of incrementing radius 1 km from the center of the city. The gradient analysis helped in visualising the changes at local levels with the type and role of agents. This helped in identifying the causal factors and locations experiencing various levels (sprawl and compact growth) of urbanization in response to the economic, social and political forces. This approach (zones, concentric circles) also helped in visualizing the forms of urban sprawl (low density, ribbon, leaf-frog development). The built up density in each circle is monitored overtime using time series analysis. This helps the city administration in understanding the urbanization dynamics to provide appropriate infrastructure and basic amenities.

  …...............(1)

Where,  Pi is the proportion of built-up in the ith concentric circle. As per Shannon’s Entropy, if the distribution is maximally concentrated in one circle, the lowest value zero will be obtained.

Table 4: Landscape metrics calculated in the analysis