Monitoring and Estimating Flood Damages by Object-Oriented Change Detection of Optical and Radar Earth Observations

The Earth’s land-covers are exposed to several types of environmental changes, issued by either human activities or natural disasters. On 15 April 2017, severe precipitations in the west and southwest regions of Iran caused flooding in the rivers of Ilam, Lorestan and Khuzestan provinces. The peak of these rainfall was in the Karoon Basin and the Dez Dam, causing an unprecedented flood in recent years with an intensity of eight thousand cubic meters per second. The occurrence of this flood has led to damages to these villages and agricultural plains. Due to the occurrence of natural disasters across the world, there is a strong need to develop an automated algorithm for fast and accurate extraction of changed landscapes within the affected areas. Such techniques can accelerate the process of strategic planning and primary services for people to move into shelters, damage assessment, as well as risk management during a crisis. Therefore, a variety of CD techniques has been previously developed, based on various requirements and conditions. However, the selection of the most suitable method for change detection is not easy in practice. To our best of knowledge, there is no existing CD approach that is both optimal and applicable in the cases of (a) using a variety of high spatial resolution optics and radar remote sensing images, (b) lack of labelled samples, (c) noisy multi-temporal images (d) non-linearly separable change and no-change classes. In addition, a low degree of automation is not optimal for real-time CD applications. Hence this paper aims to resolve the above-mentioned problems.
Recently, the Object-Based Image Analysis (OBIA) are applied considered for change detection, namely, Object-Based Change Detection (OBCD) techniques. The OBCD method uses image objects for analysing change information by comparing segmentation and classification results. Otherwise, it can extract change information by comprehensively analysing the objects of multi-temporal images using spectral, texture and structure information. Consequently, this method has normally high detection accuracy and robustness. However, since both pixel and object based methods partition the observation space linearly or rely on a linear combination of multi-temporal data, suffer from high false alarm detection rates. As a result, they can be inefficient for images corrupted by either noise or radiometric differences, which cannot be normalized. This is particularly true for high spatial resolution images, where the class distributions are strongly overlapped.
Within the last two decades, kernel-based methods have demonstrated and reliable results success in many remote sensing applications, in particular for classification and change detection problems. The main idea of kernel methods is based on the fact that the nonlinear decision function can be obtained by running a linear algorithm in a higher dimensional feature space.
In order to resolve these problems, an integrated object-level CD method based on an object-based classifier and Support Vector Data Description (SVDD) method is proposed. In order to using the information contents of radar and optical data simultaneously for enhancing the accuracy of change map, the kernel based fusion method was proposed. In addition, parameter determination of the proposed method is addressed automatically by using an inter-cluster distance based approach. In order to evaluate the efficiency of the proposed method and extract the flooded areas, optical and radar remote sensing images from before and after of Shoosh 2017’s flood, acquired by sentinel-1 and 2, were used. The accuracy analysis of results showed a great flexibility for change detection of by finding nonlinear object-level solutions to the problem. Furthermore, the comparative analysis of proposed object-level CD method in the case of using fused radar and optical images (overall accuracy (O.A.): 94.24, area under ROC (AUC): 0.98) respect to Iteratively Reweighted-Multivariate Alteration Detection (IR-MAD) (O.A.: 85.10, AUC: 0.87), Principal Component Analysis (PCA) (O.A.: 77.89, AUC: 0.78) and Spectral Angle Mapper (SAM) CD methods (O.A.: 80.67, AUC: 0.82) showed that the accuracy of the change maps is relatively improved.
The proposed CD approach leads to an acceptable level of accuracy for both optical and radar imagery. The results confirmed the fundamental role and potential of using both optical and radar data for natural hazard damage detection applications. The microwave signals have high sensitivity to water content of wetland and flooded areas which increase the intensity of the backscatter signal. Consequently, radar sensors have high potential in detecting environmental changes during natural disasters with adverse weather conditions.