Evaluation of Different Techniques to Detect Land Use / Land Cover Change Over an Area

A through observation and mapping of land-use (LU) and land-cover (LC) is essential these days for proper use of limited natural resources and hence accelerating sustainable planning and evaluating development. LULC plotting helps in categorising different variety of attributes resting in this planar earth like forest, vegetation, built-up or urban infrastructure, water, bare soil etc. A timely and precise change detection of these surface geographies will provide the footing for a better support of relationships and interfaces between human and natural phenomena. With the advancement of technology and resources it is the responsibility of a developing nation to gather adequate information regarding some basic unavoidable and frequently changing features like land use/land cover (LULC) in various growing cities to keep away the, uncontrolled development. In these recent years the use of satellite images from various sources are found the most productive way of mapping land cover. These satellite images were organised into several LULC classes considering diverse procedures. Based on these principles, one can expect a variation in the quantification of LU/LC. Hence a strong need is felt to evaluate different techniques to detect the LU/LC change over an area. To that end this research evaluates the results of various classification algorithms for land cover mapping using LISS III data of Bhubaneswar for the year 2013. These algorithms work on their respective principles and hence generates several productive plots. This study assesses the results for different classification methodologies for LULC mapping by considering LISS III data of Bhubaneswar for the year 2013. NDVI (Normalized Difference Vegetation Index) is applied to classify the greenery and the non-greenery part in my research work. Principal Component Analysis (PCA) is applied to minimise the attribute dimensionality and noise from the attribute and hence enhance the spectral values. The classification methodology applied in this study includes: Iso -Clustering, Maximum Likelihood classification (MXL), Spectral Angle Mapper (SAM), and Decision Tree Classifier (DTC) to distribute the raster data into five classes i.e. Forest, Agriculture, Barren lands, Built up (Urban / Rural) and Aquatic sources. The precision of the output image of LULC of Bhubaneswar is done generating an error matrix which is generated at the margin line for which ground zero value of the land was assembled from google earth. The area (%) for each class of different classification methodology is correspondingly compared. Results of this study show that Decision Tree Classifier (DTC) with kappa value 0.85 and overall accuracy 88.40% are the best technique for LU/LC change detection