Determination of building damage after Van-Ercis earthquake by using very high resolution satellite/aircraft platforms and GIS technology

Disasters, which may be natural or technological phenomena defined as an occurrence of widespread severe damage, injury, or loss of life or property that a society cannot cope and undergoes severe disruption. Among the natural hazards, earthquake is the most destructive disaster and causes huge loss of lives, heavily infrastructure damages and great economical loses. High numbers of deaths accompanying earthquakes almost result from structural collapse of poor, low-cost houses. According to the statistics about natural hazards, more than a million earthquakes occur which is equal to two earthquakes per minute in the world. During the time period of last decade, for instance between 2001–2011, natural disasters have brought about 780.000 deaths and approximately % 60 of this loss is reported due to the earthquakes. Most of the earthquakes originate along the boundaries between plates in the world. Turkey is one of the most seismically active countries in the world due to the junction of Arabian, African and Eurasian plates. According to the statistical data, every 2-3 years, destructive earthquakes occur in this country. Turkey is located on the Alpine-Himalayan mountain belt, one of the world's most active regions in terms of earthquakes, and has an active structure in terms of seismicity due to its location. The collision of African and Arabian plates with the Eurasian plate has an important role in the formation of tectonics of Turkey and its surroundings. As a result of this collision, the North Anatolian Fault Zone (NAFZ) on the Northern part of the mainland and the Eastern Anatolian Fault Zone on the Southern part of Turkey are formed. The NAFZ is a right lateral strike slip fault system, starting from Karlıova in the East and continuing to the Saroz Gulf in the length of 1200 km. The EAFZ is a left lateral strike-slip fault in NE-SW direction. Starting from Karlıova, this fault system runs for 600 km and passes through Kahramanmaraş province and joins the Dead Sea Fault Zone. A great earthquake (Mw=7.1) struck Van city center and its vicinity towns in the Eastern part of Turkey on October 23th 2011 at 13:41 local time (10:41 GMT). The epicentre coordinates of Van-Ercis earthquake was at 38° 37′ 40.8″ N, 43° 29′ 9.6″ E near Tabanlı village, 16 km far away from the Northern part of Van city centre. The main shock was occurred at 26 km depth and its duration was 50 seconds. After the main earthquake, 11 aftershocks above M>5 were occurred. According to the information provided by Prime Ministry Disaster and Emergency Management Presidency, it was recorded that, Van-Ercis earthquake caused 604 loss of life and 2608 wounded. In Van province and its town of Ercis, about 4000 buildings heavily damaged or/and totally collapsed due to the great and destructive earthquakes. Besides, a majority of the casualities and building damages occurred due to the irregular urbanization in Eastern part of Turkey.The main purpose of this doctoral dissertation is to determine and extract the buildings which have been demolished or severely damaged after earthquake by using different image dataset (very high spatial resolution satellite images and orthophotos) with image classification and change detection methods and also to establish a 3D interactive GIS based disaster database to visualise and to build queries for the application. To assess how well remote sensing technologies can perform in detecting structural damages suffered after earthquake, conventional pixel based and object based image classification procedures were applied to all dataset for five case study areas selected. Among the five case study areas, two of the them had homogeneous urban patterns and the other three areas had heterogeneous urban patterns. The two homogeneous case study areas include mainly farm lands, vegetation areas and sparse buildings whereas heterogeneous case study areas had more dense urbanization. However the first heterogeneous case study area (named as HET 1) has patterns like a mixture of homogeneous and heterogeneous case study areas. This site was especially selected due to having the city stadium, which is used as a tentcity after the earthquake. Since rapid post-earthquake damage assessments are being efficiently made in more detail and accurately using very high spatial resolution (VHR) satellites due to ease of the interpretation of damage status of each building block or even each individual building extracted from the images by using advanced image processing techniques; in this thesis, 3 different image datasets were used ranging from VHR satellite image to orthophotos acquired pre- and post- earthquake together with the Street wiev images. As a orthophoto images, two different orthophoto datasets produced by different directorates were used. First orthophoto dataset was taken before the earthquake and produced by Directorate General of Land Registry and Cadastre. Post-earthquake orthophotos were obtained the next day after the earthquake and produced by General Command of Mapping (GCM) which is the national mapping agency of Turkey. Both orthophoto datasets have same spectral resolution (i.e. 3 spectral bands - RGB), however their spatial resolutions (i.e. pre-earthquake orthophoto dataset has 30 centimeter whereas post-earthquake orthophoto dataset has 45 centimeter spatial resolution) are different. As a VHR satellite image dataset, the Quickbird- 2 (QB2) and Worldview- 2 (WV2) images were used. Quickbird-2 is a very popular high-resolution commercial satellite with 0.6 m panchromatic and 2.4 m multispectral mode spatial resolution and has 4 spectral bands. Worldview- 2 is the first high-resolution commercial satellite with 0.5 m panchromatic and 2.0 m multispectral mode spatial resolution and has 8 spectral bands. The third dataset used was Street view images of the disaster areas at the city centre and obtained by the Earthmine Inc. on November 9th, 2011. In remote sensing, image classification is the most commonly used image processing method in order to get information about the Earth features and to categorize all pixels into one of several land cover/land use (LULC) classes. In general, most LULC classifications have been done by using a pixel based classification methods which can be either a supervised classification, unsupervised classification or hybrid clssification. These pixel based classifiers analyze the spectral properties (i.e. not spatial or contextual information) of every pixel within the area of interest. However, after the launch of VHR satellites, the efficiency of the traditional pixel-based image processing is uncertain due to higher internal variability and noise within each land-cover class in the VHR multispectral imagery. As an alternative to pixel based image analysis, the Object Based Image Analysis (OBIA), which takes not only the spectral information but also the shape, contextual and semantic information, is being used although it has also some limitations. In this thesis, first conventional pixel based classification approach was performed but due to limitations and mixed pixel problems encountered, the classification results were not found efficient. Hence, the object based image analysis was done in the next phase of the study. In the first phase of OBIA application, multi-resolution segmentation was applied and the most convenient parameters (such as scale, color, shape, smoothness and compactness) were chosen to obtain the objects in each case study areas. Among these parameters, scale parameter was considered the most significant one that determines the size of image objects. After the segmentation, the number of the classes that will be taken into consideration in the classification step were specified. For this purpose, the selected main land cover classes for pre-disaster image classification were determined as 7 classes in total; i.e. agricultural land, vegetation (forest, bushes, lawn, gardens), buildings (residential and public), open land, mixed areas, shadow and road (asphalted, main networks, turnouts, unpaved roads). For the post-disaster image classification, the selected classes were 11 in total and named as agricultural land, vegetation (forest, bushes, lawn, gardens), buildings (residential and public), tent city area, collapsed buildings, debris areas, open land, playground, mixed areas, shadow and road (asphalted, main networks, turnouts, unpaved roads). As a final step, supervised and unsupervised object based image classification methods were applied to five different case study areas using 2 different image datasets (i.e. VHR images and orthophotos). For the accuracy assessment, classification accuracy analyses was performed in all case study areas using with conventional error matrices after the unsupervised and supervised classification phases, it was observed that the supervised classification (%57-98) was more successful than the unsupervised classification (%47-96). When the accuracy analysis after the supervised classification is examined, it is seen that the orthophoto obtained before the earthquake and the high spatial resolution (WV2) data set provide higher accuracy (91%) in the homogeneous regions. In addition, the accuracy in heterogeneous regions is close to (70%) in both sets of data. Accuracy analyzes of the data set after the earthquake show that the QB2 satellite image and the orthophoto image (HGK) are close to each other (65%) in heterogeneous areas. However, the accuracy of the QB2 satellite data (70% and 77%) in both homogeneous regions was lower than that of orthophoto images (92% and 96%) in the same regions. This difference is due to the seasonal difference of the QB2 data. In this study, the damage, which is usually quantified in terms of the extent or density of collapsed structures, was assessed using object-based change detection approach. There were 494 buildings in 5 case study areas before the earthquake. After the earthquake, 39 buildings were classified as collapsed. The total areal extent of the changes calculated from the orthophoto dataset was 11140 m2 for the heterogenous case study areas while for the homogenous case study areas the change was determined as 2011 m2. For the VHR satellite dataset, the total areal extent of the changes for heterogenous case study areas were found as 11039 m2 while for homogenous areas was calculated 2158 m2. The differences in the areal extents were caused by the misclassification. After the processing of VHR satellite dataset and orthophotos dataset with OBIA and change detection analysis, the final step was to visualize the results in a 3D GIS based established. For this purpose, in the first step QGIS is used in order to digitize all thematic mapping results. Then geological and up to date fault maps were digitized and added as layers in Arc Catalog for the earthquake database management system. As a final phase, the ArcScene software was used to establish a 3D queryable interactive database of the study area. As shown in this PhD thesis, remote sensing and GIS play crucial role for evaluating and examining pre- and post- disaster status effectively. This integrated tool is very important for all countries, which are prone to all different types of disaster (natural hazards or through human factors) every year and should be used in the disaster management systems. As demonstrated in this study, the VHR satellite images offer significant advantages over ground-based survey for assessing building damages especially where the affected area is extensive and access limited. However, some limitations and challenges (cloud coverage, image resolutions, data acquisition in a timely manner, indistinguishable moderate/minor damage states from non-damage, seasonal effects, mixed pixels etc.) should also be considered and managed carefully. The damages, especially in the case of major disasters, a comprehensive disaster and emergency management system is needed in our country that uses advanced space and geoinformation technologies to be prepared for disasters and reduce potential damages. In this context, the priority needs to be given to education, increase of national capacity, resource and human power in the field of modern disaster management and advanced geoinformation technologies.