Introduction

Urbanization is an irreversible physical process involving largescale structural changes in the landscape with an increase in builtup and population densities, thereby affecting the region's ecology and environment. This process has gained momentum during the last two decades due to globalization with opening up markets leading to the accelerated economic activities (Bharath and Ramachandra, 2016; Ramachandra et al., 2013). Cities in India are urbanising at an unprecedented and irreversible rate, as the global proportion of urban population has increased from 28.3% in 1950 to 50% in 2010 (Gerland et al., 2014). As population and its activities increase in a region, the boundary of the city expands to accommodate growth along the urban fringes, leading to urban sprawwith the fragmented urban morphology, thereby impacting local ecology at peri-urban areas and city outskirts (Ramachandra et al., 2012a,b). Landscapes covered with tree cover provide diverse services ranging from micro climate moderation, sequestering carbon (emitted in urban environment), groundwater recharge, etc. while maintaining the natural balance (Parker, 1995). Unplanned urbanization in recent times has fueled the sprawl with lack of adequate infrastructure and basic amenities, necessitating holistic integrated governance approaches for sustainable management of natural resources (Ramachandra et al., 2015). Bangalore the Silicon Valley of India, has witnessed 1005% increase in paved surfaces with 88% decline in green spaces and 79% of water spread regions (Ramachandra and Bharath, 2016). Bangalore known as garden city in seventies would have about 6 percent of green spaces by 2020 (Bharath et al., 2014). Inventorying, mapping and monitoring of trees would help in green initiatives to meet the basic human needs.
Availability of multi resolution spatial data acquired synoptically through space borne sensors help in inventorying, mapping and monitoring of natural resources cost effective way (Gougeon and Leckie, 2006; Ward and Johnson, 2007; Konijnendijk, 2003; McHale et al., 2009; Hirschmugl et al., 2007). The tempora spatial data help in the understanding of urban growth pattern, urbanization rate, and the underlying problems of urban sprawl, etc. These, ultimately, aid in better administration through the provision of basic amenities. Urban sprawl has been characterized considering indicators such as growth, social conditions, aesthetics, decentralization factor, accessibility conditions, density, open space availabilities, dynamics, costs, and social benefits (Ramachandra et al., 2012a,b; Vishwanath et al., 2015). Advances in geoinformatics has aided in optimal exploitation of remote sensing data for inventorying and mapping of natural resources and also extraction of features such as trees, buildings, etc. (Preto, 1992; Ramachandra et al., 2012a,b, 2015; Bharath and Ramachandra, 2016).
Improved spatial, spectral and radiometric resolutions of spatial data with collateral data (collected from field) aid in generating a range of spatial statistics of feature attributes such as structure, crown, etc., which will help in the location specific interventions. Most of the optical data obtained through space borne sensors have either higher spatial resolution or spectral resolution but not both due to various limiting capabilities (Kumar et al., 2012) including cost. These can be overcome by using image fusion or pan sharpening (Ramachandra et al., 2011; Choi, 2006; Alparone et al., 2007; Chen et al., 2011) by integrating better spatial information from panchromatic and spectral information from multi-spectral (MS) data (Kumar et al., 2009). Techniques such as intensity hue saturation (Choi, 2006), Principle component analysis (Gonz Audícana et al., 2004), ICA-independent component analysis (Petrovic and Xydeas, 2004; Chen et al., 2011; Dong et al., 2013) have helped in obtaining optimal information through fusion of multi resolution spatial data. PCA was useful in retaining spectral information, while ICA was able to retain high spatial resolution. Recent developments such as Hyper spectra color space resolution merge, MIHS, Wavelet transformations help in attaining detailed accurate information (Nikolakopoulos, 2008).

1.1. Hyperspectra color space resolution merge

Considers an image with ‘n’ input bands and a single pan band fuse in ‘n-1’ angles of Hypersphere (Padwick et al., 2010). square of mean and standard deviation of multispectral intensity and panchromatic intensity is calculated as shown in (1)

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

Then, the forward color transform is done from the native color space to hypersphere color space using mean and standard deviation as per Padwick et al. (2010). Later, intensity match of transformed P² and I² is done. Further sharpening is done by considering the square root of P² to obtain I_adj. The intensity component is reverse transformed from HCS to native space using hyperspectral color transform (Padwick et al., 2010).

1.2. High pass filter fusion

Pixel sizes (cell sizes) are extracted as input information. The ratio between the cell sizes of the multispectral data and high spatial resolution data is computed and a high-resolution data is filtered through a high pass convolution filter of size relative to input pixel sizes. The low spatial resolution data is resampled comparable to the high-resolution data based on four nearest neighbors. The HPF image is weighted relative to the global standard deviation of the multispectral bands as per equation (2).

..............(2)

Where:
W = weighting multiplier for HPF image value, SD(MS)=standard deviation (SD) of the MS band to which the HPF image is being added, SD(HPF)= standard deviation (SD) of the HPF image, M =modulating factor to determine the crispness of the output image.
Further each pixel value is calculated using equation (3).

..............(3)

Non-linear absolute histogram equalization is performed match the data type and range of input data.

1.3. Modified intensity hue saturation fusion

The modified IHS method helps in fusing spatial data differing in spectral responses by assessing the spectral overlap between each Multi Spectral band and the high-resolution PAN band and weighting the merge based on these relative wavelengths (Siddiqui, 2003; Yakhdani and Azizi, 2010).

1.4. Wavelet fusion

Wavelet method based on breaking down of wavelengths of the spectrum into wavelets or packets of discrete reflectance was used to merge the high-resolution image into the intensity values. The wavelet transform provides a framework to decompose images into a number of new images, each one of them with a different degree of resolution (Nikolakopoulos, 2008). Scaling functions were derived from Bi-orthogonal spline column wavelets. The fourth order shift invariant decomposition transform and fourth order reconstruction spline are as per King and Wang (2001). Panchromatic image is decomposed into four images. The first image was replaced with intensity image and an inverse discrete wave transform was performed considering other three decomposed images.This provides a sharper image which is then transformed back to RGB.
Extraction of tree cover: Tree cover extraction algorithms were developed considering that vegetation center is radiometrically brighter than the edge. Methods of optimal match of predefined shapes with local radiometric values were proposed by Larsen (1998). Brandtberg (1999) used edge segments and neighbors with region growing for extraction. Further, Walsworth and King (1999), developed and compared two vegetation delineation techniques involving radiometric surface aspect and other with high pass filter for temporal analysis. This work considers minima and maxima of local radiometric value based on ground measurements of maximum and minimum tree crown size, girth with predefined edges. Based on the training data collected from the field trees were extracted using matching radiometric quality and range.