STUDY AND APPLICATION OF SOFT COMPUTING TECHNIQUES FOR HOTSPOT MONITORING USING SATELLITE DATA

Abstract

The aim of the dissertation is to elaborate the remote sensing methods for monitoring subsurface lire (hotspots) in Jharia Region, (Jharkhand) India; as the Jharia coal field contains almost half of subsurface mine fires within the Indian coal fields [1]. Thus, detecting and monitoring such hotspots are mandatory. Since ground based monitoring are quite expensive and difficult task, exploiting the potential of satellite images have been tried as an alternative solution. For this purpose, freely available satellite images (e.g., MODIS, NOAA/AVHRR, and LANDSAT) are being used for our study. This study involves the application of most renowned soft computing techniques such as: supervised classification (parallelepiped, minimum distance) and unsupervised classification (ISODATA, K-means) over optical data: MODIS, NOAA/AVI-IRR, - and LAN DSA'I'. NE)VI plays an important role for the detection of hotspot due to the fact that hotspot region usually has bare ground such that neither bushes nor grasses grows over hotspot region. Thus, NDVI classified image into hotspot and non-hotspot regions is used. The accuracy of the classified image is assessed using the metrics: hotspot detection accuracy (HDA) and false alarm rate (FAR). The assessed value indicates that there is room for improvement. Thus, an attempt based on heuristic method- genetic algorithm (GA) have been carried out, since it has higher chances to result in an optimal classification of hotspot and non-hotspot pixels due to its ability to search for the optimal hypothesis over a larger search space. Therefore, the attempt of GA based KMI (K-Means Index) indicates that the detection of hotspot with an accuracy of 81%-I 11)A and 11%-FAR over MODIS dataset. Such high HDA and low FAR over detection of hotspot and an attainment of good temporal resolution recommends use of MODIS dataset for area estimation over hotspot coverage in Jharia region. But due to fragment size of hotspot in comparison to spatial resolution of MODIS, major amount of hotspot are present partially within a pixel (i.e., mixed pixel issues). In order to perform hotspot area estimation over such coarse resolution image, subpixel analysis is performed; by refining the per-pixel spectral-based detected hotspot from MODIS image by proposing a method that uses a subpixel spectral detection method called GEM (a target - constraint approach). Constrained energy minimization (CEM) is very efficient in the detection of small hotspots very effectively as well as it requires only a prior knowledge of target spectral signature. Due to the requirement of hotspot pure spectral signature, we have used LANDSAT- 5TM image for the endmember selection using PH. With such refined detected hotspots, the estimated area coverage of hotspot were found to be of 11.09 Km2 (on I 4-Mar-20 15) and when validated with week and yearly variation; it is observed that hotspot of 0.165 Km2 of variation been observed within two weeks interval and 2.647 Km2 of increased I-Iotspot coverage is observed over a period of two years.