Method

Analysis can be delineated into five major steps
(1) data acquisition from public repositories such as USGS, land use classification and accuracy assessment using open source software GRASS.
(2) Land use analysis and baseline validation.
(3) Developing data for three scenarios based on urban growth.
(4) Calibrating the model to find out the best weights based on the influence on the neighbourhood pixels.
(5) Development of model and validation of the model with predicting of future scenario based on validated data and calculation of spatial metric for analysis of landscape configuration in various scenarios

Landscape configuration in various scenarios

1. Image pre-processing Remote sensing data were acquired for specific years from USGS earth explorer (http://www.usgs.gov/). Remotely sensing data obtained were geo-referenced, rectified and cropped pertaining to the study area.

2. Land use analysis Land use analysis was performed using supervised pattern classifier—Gaussian maximum likelihood classifier (GMLC). GMLC has been already proved as one of the superior classification techniques due to use of cost function and probability determination techniques employed (Duda et al. 2000; Bharath et al. 2018). Land use classification was performed and the classified data is categorised into four major classes built-up area, vegetation, open area, and water body as described in Table 2.

Essentially land use analysis is performed in four broad steps as follows
a. Stacking as image composite-false colour composite to identify various classes of patches (bands–near infrared, green and red green).
b. Collection of training data as training polygons using pre-calibrated GPS AND and using google earth (validated and corrected with a shift in position). Training data was collected in order to classify and also to validate the results of the classification. 70% of the training data collected were used to derive the user-classified map. 30% of exact ground truth was used to validate it.
c. Land use classification using GMLC using GRASS GIS (Geographical Analysis Support System) an open source software has been used for the analysis, which has the robust support for processing both vector and raster files.
d. Validation of land use by performing accuracy assessment and kappa statistics: accuracy assessments helps the data generators to determine the quality of the information. The test samples from user classified map and validation ground truth maps generated is use to generate well-known methods in validation of land use using error matrix and calculate kappa (κ) statistics.

3. Population growth rate has been a major influencing factor of urban growth in Bangalore. Hence using this as a factor three scenario was designed as per the suitability map. The first scenario considered current growth population rate of Bangalore (approach. 5% per annum) as business as usual scenario, further with the national growth rate at 2–3%, were considered as deviations for designing the next two scenarios. The second scenario was designed based on a decreased growth rate of 3% matching the national average and the third scenario was based on increased growth rate of 7%.

4. Modelling land use scenarios GeoMod was used for modelling the land use pattern and to predict the future scenarios. GeoMod is built on a grid-based model that persists the maps as grids of data to simulate the urban pattern of land use change and has a capability to predict both as time forward or time backwards (Pontius and Schneider 2001; Pontius et al. 2001; Dushku and Brown 2003). GeoMod works on a binary map and simulates the land use change based on this binary map as two categories (binary images are first created as urban and non-urban). Map considered as a base to predict is both as time forward or time backwards (Pontius and Schneider 2001; Pontius et al. 2001; Dushku and Brown 2003). GeoMod works on a binary map and simulates the land use change based on this binary map as two categories (binary images are first created as urban and non-urban). Map considered as a base to predict is supplied by the user and map to be used as validation is also used (in case of validation) else only the grid cells are mentioned, along with the land use change driver’s associated with weights. This grid are assigned as one of the two categories for the ending time based on the various decision rules.
   a. Map-persistence GeoMod can simulate two change but in different transitions. In a single transition, it simulates only a single way change.
   b. Grid based differentiation or stratification It would simulate land use change within a strata or specific region.
   c. Neighborhood constraint It is based on a nearest neighbor principle and treats cells restrictions for one time change with the edge between two portions A&B in the land use
   d. Site suitability This is generated empirically using several maps and the land-use transition map from the beginning time
   Site suitability of each cell is calculated using the equation below
   

   

   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.

5. Developed suitability map along in conjunction with land use maps of the year 2008 and 2010 was considered as base data to simulate urban growth in the year 2012. The model performance was assessed by comparing the simulated year 2012 urban map to the actual year 2012 classified map. Accuracy is calculated for each analysis. This process is reiterated until the simulation reaches the threshold accuracy by changing the driver behaviour and influence characteristics. Once the model is trained to land use of 2010 and 2012 is used as base data to predict urban pattern growth for the year 2014 to the year 2020 in a time step of 2 years and land use is quantified. Spatial metrics using open source software fragstat was then computed. This was performed for all scenarios to observe the change in urban land use by 2020 depending land use configuration. As suggested by Ramachandra and Bharath (2012) urban land use dynamics can be characterized by few spatial metrics as tabulated in Table 3 based on shape, edge, complexity criteria.

Fig. 2 Procedure for modelling

Table 2 Land use categories

	Indicators	Formula
1	Number of urban patches (NUP)	NPU=n, NP equals the number of patches in the landscape
2	Normalized landscape shape index (NLSI)	∑ii==N1 pi NLSI = si ,N where si and pi are the area and perimeter of patch i, and N is the total number of patches
3	Total edge (TE)	TE=∑mk=1eik, where, eik = total length (m) of edge in landscape involving patch type (class) i; includes landscape boundary and background segments involving patch type i
4	Clumpiness index (Clumpy)

Table 3: Landscape Metrics analysed