|
LAND SURFACE TEMPERATURE RESPONSES TO THE LAND COVER DYNAMICS IN WESTERN GHATS
(PDF)
T V Ramachandra, Srijith A H and Bharath S
Energy and Wetlands Research Group,
Centre for Ecological Sciences,
Indian Institute of Science - 560012
5. Method
There have been extensive studies to understand the land use of Western Ghats by different researchers through different data, methods at different time periods. They have used field mapping, aerial photographs, satellite images etc. for inferring status of forests. The present study has been carried out by using satellite images to understand LU dynamics and the role of LU categories influence on LST. The two major approaches used for getting LU information from satellite imagery are visual interpretation and digital image processing. Visual interpretation uses various scene elements like size, shape, tone, texture, association etc. to identify and delineate objects. Digital image classification is the process of assigning pixels to respective classes on the basis of spectral properties (Lillisand et al., 2004).
5.1. Land Use Analysis:
The steps involved in preparation of LU maps are listed below and shown in Fig. 5.1.
(i) Check for quality parameters: The satellite images of the study area are checked for pixel quality and cloud cover using data quality and science quality flags.
(ii) Image pre-processing and Geo-referencing: The remote sensing data (satellite images) satisfying the quality parameters have been corrected for radiometric errors like faulty data lines, repeating lines etc. The image then has been corrected for geometric errors by nearest neighbourhood method using suitable ground control points obtained through field survey and from Google Earth. The geo-referenced images have been then projected to Geographic Coordinates System (EPSG: 4326).
(iii) Image segmentation - Agro-Climatic Zones: The satellite images have been cropped to different zones ie. coast, hills and plains using suitable masks based on the agro-climatic zone of the study area. These are further cropped based on state boundaries to study the change in LU classes over a period. These processes are carried out using different GIS software as discussed in the Materials section.
(iv) LU classification: The LU analysis has been carried out using supervised ISODATA (Iterative Self Organizing Data Analysis Technique) Classification technique. In ISODATA clustering, the image data are first classified by aggregating them into natural spectral groupings or clusters present in scene. The basic premise is that values within a given cover type shall be close together in measurement space; whereas data in different classes shall be comparatively well separated. Iterative Self Organizing Data Analysis Technique (ISODATA) is a widely used variant of K-means technique used for pattern recognition (Tou et al., 1974). The world ‘self-organizing’ refers to the way clusters are located which are inherent in data. It uses the minimum spectral distance formula to form clusters (Chen et al., 2000). In this number of clusters are set initially and clustering has been carried out on basis of deletion, splitting or merging during the process (Fig 5.2). Once an iteration is completed with the assignment of pixels to the cluster, the statistics describing each cluster are evaluated. If the distance between the mean of two clusters are less than pre-defined distance the two clusters are merged. On the other hand, if a single cluster has a standard deviation that is greater than a predefined maximum value, the cluster is split in two. Clusters with fewer than the specified minimum number of pixels are deleted. The clustering is carried out until (i) a maximum number of iterations are performed, or (ii) a maximum percentage of unchanged pixels has been reached between two iterations.
The satellite image was first classified into 16 clusters over 10 iterations. During every iteration an arbitrary mean is assigned to each cluster and the pixels are allocated in the clusters closest to the mean. New means are calculated during each iteration for each cluster based on the pixels present. The process is repeated until each pixel is each pixel is assigned to the closest mean. Each of the clusters is separated out and individually compared with field data. If the clusters are found to be confusing between two classes, these are noted and those showing confusion are grouped and further classified by same algorithm (ISODATA). Confusions are found to exist during classification of LU between plantation and forest, built up and open spaces, crop land and scrub vegetation etc. The method of allocating the cluster to each land use class is based on the visual observation. The land use maps are prepared and categorised into these classes described in Table 5.1.
(v) Validation of LU Maps: The land use maps have been validated for classification error by computing Error matrix. The reference image has been prepared by classification of training sites taken from ground data. The Kappa coefficient analysis is a discrete multivariate technique used in accuracy assessment for determining statistically if one error matrix is significantly different from another (Bishop et al., 1975)
5.2. LST MAPS:
The steps involved in preparation of LST maps are listed below and show in Fig. 5.3.
(i) Check for quality parameters: The satellite images are checked for pixel quality and cloud cover using data quality and science quality flags.
(ii) Image pre-processing and Geo-referencing: The satellite images satisfying the quality parameters have been corrected for radiometric errors like faulty data lines, repeating lines etc. The image then has been corrected for geometric errors by nearest neighbourhood method using suitable ground control points obtained through field survey and from Google Earth. The geo-referenced images have been then projected to Geographic Coordinates System (EPSG: 4326).
(iii) Image segmentation - Agro-Climatic Zones: The satellite images have been cropped to different zones i.e. coast, hills and plains using suitable masks based on the agro-climatic zones of the study area. These are further cropped onto by state boundaries to understand the change in land use classes over the study period.
(iv) Land Surface Temperature Maps: The LST based on agro-climatic zone has been evaluated. The maximum temperature Tsmax, minimum temperature Tsmin and mean temperature Tsmean have been evaluated for different zones of the study area.
(v) Validation of Land Surface Temperature Maps: The LST maps has been validated on the basis of air temperature (at 2m) taken at 13 different locations all across the study area. The ground data has been taken from NASA Climatology Resource – Global Coverage. The reference data set provides air temperature at a height of 2m from ground level. The ground maximum temperature Tgmax, minimum temperature Tgmin, and mean temperature Tgmean is compared with land surface temperatures (Tsmax, Tsmin and Tsmean) for the same locations. A mean land surface temperature of about 2-3 °C higher than air temperature is deemed to be acceptable (Cheng et al., 2005).
6.3. Rate of Land Use Change:
The rate of change of each land use category is shown by two indices i.e. total changed area A, and the rate of change ρ. These are calculated as follows:
|
