METHOD

The process of urbanisation and sprawl in Bangalore (study area) have been assessed as outlined in Fig. 2, which includes (i) Land use analysis, (ii) Modeling and prediction, (iii) Urban Sprawl Analysis

Land use analysis:

The land use analysis was carried out for the 3 temporal data using the Gaussian maximum likelihood classifier.  The data were classified under 4 different classes as shown in Table I.Satellite data classification involved (i) Preprocessing (ii) Classification (iii) Accuracy assessment.

Preprocessing: The raw satellite images geo-corrected, followed by radiometric correction and resampled to 30 m resolution to maintain uniformity for multi temporal data comparisons and for modeling.

Table I. Land use categories

Land use classification and accuracy assessment: The method involves i) generation of False Color Composite (FCC) of remote sensing data (bands – green, red and NIR). This helped in locating heterogeneous patches in the landscape ii) selection of training polygons covering 15% of the study area and uniformly distributed over the entire study area, iii) loading these training polygons co-ordinates into pre-calibrated GPS, vi) collection of the corresponding attribute data (land use types) for these polygons from the field, iv) Supplementing this information with Google Earth/Bhuvan.Land use classification was done using supervised pattern classifier - Gaussian maximum likelihood algorithm based on various classification decisions using probability and cost functions [40].

Fig .2. Procedure followed to model, understand the landscape changes

Land usesduring the different period were computed using the temporal remote sensing data through open source GIS: GRASS- Geographic Resource Analysis Support System (http://ces.iisc.ernet.in/grass). Four major types of land use classes considered were built-up, vegetation, cultivation area (since major portion is under cultivation), and water body. 60% of the derived signatures (training polygons) were used for classification and the rest for validation. Statistical assessment of classifier performance based on the performance of spectral classification considering reference pixels is done which include computation of kappa (κ) statistics.

Modeling and Prediction:The land use pattern is evolving dynamically and follows the Markovain random process properties with various constrains that include average  transfer  state  of  land  use  structure  stable and different land use classes may transform to other land use class given certain condition (Such as non-transition of urban class to water or vice versa). Thus Markov was used for deriving the land use change probability map for the study region and was applied using Markov module. The probability distribution map was developed through Markov process. A  first-order  Markov  model  based on probability  distribution  over  next  state of the current cell that is  assumed  to  only  depend  on  current  state [41]. CA was used to obtain a spatial context and distribution map. CA’s  transition  rules  use  its current neighborhood  of pixels to  judge  land  use type  in  the future.  State  of  each  cell  is  affected  by  the  states  of  its neighboring  cells  in  the  filter considered.  Besides using CA transition rule and land use transition is governed by maximum probability transition and will follow the constraint of cell transition that happens only once to a particular land use, which will never be changed further during simulation. CA coupled with Markov chain was then used to predict urban land use state in 2020

Urban sprawl analysis:Post prediction and valdation with spatial metrics were employed to quantify the urban growth pattern. Spatial metrics aid in quantifying  the land use pattern at a particular time [1][42]. Spatial metrics such as Number of urban patches (NP), Normalised landscape shape index (NLSI), edge density (ED), Clumpy, Pladj were used in the analysis.These selected metrics signify patch, contagion, edge, shape and adjacency.