Urban Area Monitoring and Intraurban Change Detection using SAR Data

Most human activities are concentrated in urban areas, which cover a small portion of the Earth's surface, but cause planetary-scale issues, such as air and water pollution, land degradation, and heat island phenomena. The increasing availability of fine temporal and spatial resolution SAR (Synthetic Aperture Radar) sensors on board of Earth Observation (EO) satellites provides a new perspective in investigation and monitoring of global and regional human activities. Thanks to the principle of radar imaging and its inherent characteristics, e.g., double-bounce scattering, SAR directly enables to retrieve the urban structure in the horizontal and vertical dimensions. Additionally, the utilization of SAR polarimetry makes it possible to separate mixed backscattering echoes. Finally, exploiting Interferometric SAR (InSAR) technology, more 3D information can be extracted and considered. In this scenario, although many approaches have been developed to detect urban changes in single-polarization SAR data, few researches have focused on urban change detection using multi-polarization SAR sdata. Moreover, many of the approaches in technical literature map urban extent expansion or intra-urban change locations, but without exploring, in the latter case, detailed change characteristics in terms of their spatial and/or temporal patterns. In this thesis, new approaches are developed to automatically map and explore intra-urban changes using fully polarimetric SAR (Quad-PolSAR), Dual-polarimetric SAR (Dual-PolSAR), as well as temporal sequences of SAR data, even in conjunction with multispectral data. More specifically, this thesis reports the methods listed as follows. 1, A superpixel-based multi-pattern change detection technique that uses SAR polarimetry. Quad-PolSAR (Radarsat-2) and Dual-PolSAR (Sentinel-1) data are used to map 2D/3D changes inside urban areas. The results confirm the possibility of an effective 2D/3D change detection with fully polarimetric SAR, but show poor performances using dual-polarization data. 2, A temporal change pattern extraction approach that uses coherence in multi-temporal Dual-PolSAR series. Sentinel-1 SAR complex data are used to detect changes in the temporal domain. Due to the weak contribution of polarimetry to change mapping, coherence temporal series are considered to investigate temporal multi-pattern intra-urban changes. 3, An intra-urban change vector analysis exploiting SAR and multispectral data. The use of SAR in conjunction with other optical sensors is a promising yet challenging technique, because of the different spatial resolutions and data modalities. A change vector analysis and visualization approach is proposed to automatically extract changes and enable an easier understanding of the intra-urban changes that are revealed. 4, A hierarchical bi-clustering algorithm for multiple-pattern change investigation. To address temporal pattern analysis in wide urban areas (e.g., megacities or urban clusters) a hierarchical bi-clustering method is introduced to address the unpredicted number of clusters and better discriminate among urban change patterns referring to new constructions, demolitions, and renovations. All the approaches proposed in this thesis have been applied to monitor urbanization in both developing and developed countries, and have been validated by an extensive comparison with ground truth data in many different areas of the world. Specific focus has been given to the P.R. China and South East Asia, where urban areas are quickly evolving and intra-urban changes can be more easily spotted and recognized.