Modeling Land Cover Dynamics through Multi-Resolution Remote Sensing Data

Student Name: Uttam Kumar

Land cover (LC) refers to the physical state of the earth’s surface and immediate surface in terms of the natural environment (such as vegetation, soils and ground water) and the man-made structures (e.g. buildings). Changes in LC induced by human and natural processes play a major role in global as well as regional scale patterns of the climate and biogeochemistry of the earth system. Variations in LC influence surface-atmosphere fluxes of sensible heat, latent heat, and momentum, which in turn influence weather and climate. Hence understanding LC dynamics at the local/regional as well as at global levels is essential to evolve appropriate management strategies to mitigate the impacts of LC change.

LC dynamics can be captured through the remote sensing data of various spatial, spectral and temporal resolutions. These information constitute a vital input for monitoring and modeling LC dynamics. However the data of multi-resolution sensors need to be integrated to get optimal information through image fusion techniques. Fusion of data from multiple sensors aids in delineating objects with comprehensive information due to the integration of spatial and spectral information. In this context, numerous image fusion techniques are adapted and assessed for their applicability in landscapes of three different altitudinal ranges (low altitude range - Deccan Plateau, moderate altitude - Western Ghats and high altitude - Western Himalayas) for different resolution ratios such as IKONOS / IRS LISS-III (1:4) Panchromatic (PAN) and Multispectral (MS), Landsat ETM+ MS and PAN (1:2), IRS LISS-III MS and MODIS 250 m (1:10), IRS LISS-III PAN and MODIS 250 m (1:50) and IKONOS PAN and MODIS (1: 250).

Hard and soft classifications are the two ways of estimating LC composition in a landscape. Various supervised and unsupervised classification techniques have been tested for hard classification. In case of coarse resolution data, soft classification technique such as Linear un-mixing using Ordinary Least Squares and Orthogonal Sub-Space Projection are adopted. The end members are extracted using NFIND-R, scatter plot, pixel purity index and 3-dimenisonal visualization. Nonlinear unmixing of pixels is addressed using Neural Network based multi-layer perceptron. An algorithm is proposed to indicate the spatial distribution of class abundance within a coarse resolution pixel by pixel swapping technique which maximizes the autocorrelation between the pixels, assuming similar LC. By hybridizing the abundance estimates from soft classification and posterior probabilities from the ground data, a new method of estimating prior probabilities for image classification.

Temporal phenomenon in a landscape is captured by spatio-temporal data analysis. Various change detection techniques such as Principal component analysis, Correspondence analysis, NDVI differencing are used for detecting changes in LC. Case studies of Greater Bangalore, Himalayan region and Uttara Kannada district are demonstrated using urban fragmentation and forest fragmentation on temporal scale. The temporal dynamics of the urban area is modeled using time series data using cellular automata and statistical models. Several spatial metrics are used to quantify the measures for spatial pattern. Finally the models are calibrated and tested for validation using ground data.