METHOD

The process of urbanisation and sprawl in Bangalore (study area) have been assessed which includes (i) Land use analysis, (ii) Modeling and prediction.Land use data was used from the previous analysis (Bharath S et al., 2012). This data was reclassified into Urban and non-urban for Geomod analysis. But other modelling techniques such as CA Markov and LCM same data with 4 classes as described in table 1 was considered.

Modeling and Prediction: CA MARKOV: 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 of IDRISI. 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 . 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

Table 1. Land use categories

Land use Change Modeller (LCM): an ecological modeller module in IDRISI Taiga was used for modelling the land use scenario based on the data of 2008, 2010 and 2012. LCM module provides quantitative assessment of category-wise land use changes in terms of gains and losses with respect to each land use class. This can also be observed and analysed by net change module in LCM (IDRISI manual). The Change analysis was performed between the images of 2008 and 2010, 2010 and 2012, to understand the transitions of land use classes during the years. Threshold of greater than 0.1 ha. Were considered for transitions.  CROSSTAB module of IDRISI was used between two images to generate a cross tabulation table in order to see the consistency of images and distribution of image cells between the land use categories. Multi-Layer perceptron neural network was used to calibrate the module and relate the effects of agents considered and obtain transition potential sub models. Further markov module was used to generate transition probabilities, which were used as input in cellular automata for prediction of future transitions. This has been analysed with an inbuilt module of LCM or using the CA_Markov in IDRISI.

GEOMOD: GEOMOD was used for modeling the spatial patterns of urbanisation and predict likely land use changes. GEOMOD simulates the spatial pattern of land use changes [56], or change between two land categories (Binary images of urban and non-urban). GEOMOD selects the location of the grid cells based on the following decision rules:

1. Persistence: simulates one way change.
2. Regional stratification: simulate land use changes within a series of regions called strata.
3. Neighborhood constraint: It is based on a nearest neighbor principle, whereby restricting land change within any one time step to cells that are on the edge between landscape A and landscape B
4. Suitability map considering drivers.

If there is a net increase in the Class A category as the simulation proceeds from a beginning time to the ending time, then GEOMOD will search among the Class B grid cells in order to select the cells that are most likely to be converted to become Class A during the time interval and vice versa [56][57][58].

Suitability map is created using GEOMOD Module in IDRISI TAIGA (http://clarklabs.org) considering drivers. Each driver is considered as real number (%), obtained by comparing the driver map to the beginning time land-cover map. Site suitability of each cell is calculated using the equation (1) below,based on each reclassified attribute,

.................(1)

Where: R(i) = suitability value in cell(i), a = particular driver map, A = the number of driver maps,

Wa= the weight of driver map a, and Pa(i) = percent-developed in category ak of attribute map a, where cell(i) is a member of category ak. Predictions were done considering three population growth rates of 5% (current average population growth of Karnataka)