Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/14502
Authors: Singh, Krishna Kant
Keywords: Natural Disasters;Property;Earthquakes;Rapid and Inevitable
Issue Date: Oct-2014
Publisher: Dept. of Earthquake Engineering iit Roorkee
Abstract: Natural Disasters are one of the major threats to life and property and they impose serious threat to the society. Amongst the various natural disasters, earthquakes are one of the most destructive events. Earthquakes are sudden, rapid and inevitable in nature thus it is not possible to avoid them. Therefore, it becomes necessary to provide quick mitigation methods in order to reduce the effect of such catastrophic events. The effect of earthquakes is widespread and covers large geographical area therefore quick response is required to carry out fast recovery and rehabilitation operations. In the early days, when remote sensing data was rarely available, the rescue teams where dependent on the field surveys which took longer period of time to provide accurate information about the extent and location of damage. But in the recent years, remote sensing data has proved to be a useful tool for getting information about the disaster affected areas. Initially, remote sensing images were visually analyzed to extract information but later researchers worked to develop automated methods for extracting correct and accurate information about the affected area. This thesis aims at developing semi automated methods for the analysis and classification of satellite images to detect earthquake induced changes including tsunami inundated areas and landslides. In this work, three novel methods for identification of different earthquake induced changes using medium resolution satellite images are proposed. The first method proposed is Fuzzy Kohonen Local Information C-Means (FKLICM) for land cover classification of remote sensing images. The FKLICM algorithms had certain limitations and thus to overcome these limitation another classification algorithm, Generalized Improved Fuzzy Kohonen Clustering Network (GIFKCN) is proposed. This algorithm is used to develop a method for identification of earthquake induced tsunami inundated areas. Finally, the GIFKCN classifier is used in combination with ASTER Global Digital Elevation Model (GDEM) to identify earthquake induced landslides. The first method developed is used to classify remote sensing images into different land cover types. It involves fusion of Multispectral (MS) bands and pan band using Brovey transform to obtain a higher resolution image the fused image is dimensionally reduced using Principal Component Analysis. The first component of Principal component analysis known as PC-1 is classified using the ii proposed novel neuro fuzzy clustering algorithm named Fuzzy Kohonen Local Information C-Means (FKLICM).The proposed method is applied on Landsat 7 ETM+ image with eight bands over the Ishinomaki city, Japan area acquired on 18 February 2003 to classify it into three classes vegetation, water, land & urban area. The classification result obtained from FKLICM is compared with other state of the art methods and it is observed that the result form FKLICM is better than the other methods. The second method in this thesis is for identification of tsunami inundated areas using the proposed novel neuro-fuzzy classifier that hybridizes Generalized Improved fuzzy partitions FCM (GIFP-FCM) with Kohonen Clustering Network (KCN) known as Generalized Improved Fuzzy Kohonen Clustering Network (GIFKCN). The novel classifier is trained using the spectral indices like Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Difference Built-up Index (NDBI) for the identification of different land cover classes. The inundation that occurred in Ishinomaki city located in Miyagi prefecture Japan due to the 11 March 2011 Tohoku tsunami is identified using GIFKCN clustering algorithm. Pre Tsunami satellite image acquired on February 18, 2003 and Post Tsunami satellite image acquired on 20 March 2011 from Landsat 7 ETM+ and Landsat 5 Thematic Mapper (TM) respectively are classified using the novel GIFKCN clustering algorithm. Finally, the change in water class is identified to find inundated areas. It is observed that much of the inundation occurred along the Kitakami River. The third method is developed for the identification of earthquake induced Landslides from bi temporal satellite images using GIFKCN classifier and ASTER GDEM is proposed. The pre and post earthquake images are preprocessed, the preprocessed images are classified into four classes using the proposed GIFKCN classifier. The classifier is trained using the spectral indices like NDVI, NDWI, NDBI. The pre and post classified images are used to detect the changes in the vegetation class. Since landslides often result in loss of vegetation thus the points at which vegetation class is changed to barren land are identified as candidate landslides. A rule set is created using DEM derived slope and aspect values. The rule set is applied on the candidate landslides to remove false positives. The method is applied on Sikkim area to detect the 18th September 2011 earthquake induced landslides using pre and post earthquake Landsat 5 and EO-1 ALI images acquired on27th August 2011 and 19th October 2011 respectively. The digital elevation data is obtained from ASTER Global iii Digital Elevation Model (GDEM).The accuracy assessment of the method is done and the number of Landslides detected is also computed. The three methods proposed in this thesis yield good results and can be used for effective disaster mitigation processes. These semi- automatic methods provide quick and accurate damage response and can be used by mitigation teams and decision managers to carry out fast recovery and rehabilitation process.
metadata.dc.type: Thesis
Appears in Collections:DOCTORAL THESES (Earthquake Engg)

Files in This Item:
File Description SizeFormat 
G24361-K.K.S.-T.pdf5.27 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.