SYNTHETIC APERTURE RADAR REMOTE SENSING BASED TECHNIQUES FOR EARTH OBSERVATION

Abstract

Microwave remote sensing using synthetic aperture radar (SAR) has brought significant advancements to the field of remote sensing. SAR, being an active sensor, enables observation of the Earth's surface regardless of day or night conditions. Unlike optical imaging systems, which rely on visible light, SAR operates in the microwave frequency range, allowing operation under adverse weather conditions such as cloud cover, fog, rain, and haze. SAR sensors emit electromagnetic waves capable of sub-surface penetration and are sensitive to the dielectric properties of the target medium. SAR remote sensing can be categorized into two major domains: SAR Polarimetry (PolSAR) and SAR Interferometry (InSAR). PolSAR utilizes polarization diversity, employing multiple polarimetric channels (co-polarized and cross-polarized), such as HH, HV, VH, and VV, to identify targets and retrieve parameters based on their orientation, shape, alignment, and permittivity. On the other hand, InSAR utilizes the phase information of complex SAR data to detect topographical variations and vertical structures of scattering features. Techniques such as Differential SAR Interferometry (DInSAR) and Persistent SAR Interferometry (PsInSAR) are employed for time-series monitoring of land movement. This thesis focuses on the development and utilization of SAR remote sensing techniques for various earth observation applications. The thesis presents the utilization of a time-series SAR Interferometry based approach for Land deformation monitoring. Initially, the study provides a methodology for estimating multi-dimensional land subsidence velocity using time-series Sentinel-1 SAR datasets through Persistent SAR Interferometry (PsInSAR). While traditional InSAR techniques offer high accuracy in estimating land subsidence at anmm scale, they are limited to measuring deformation velocities solely along the radar line of sight (LOS). Consequently, the complete deformation velocity, including horizontal and vertical components, remains uncertain. This research addresses the limitations by extracting PSI-based multi-dimensional land deformation velocity in horizontal and vertical directions within an urban area, specifically focusing on Lucknow city in Uttar Pradesh, India. By leveraging ascending and descending pass Sentinel-1 datasets, the study derives LOS velocity vectors and employs them to infer multidimensional subsidence velocity. Further, the assessment of land deformation and its drivers in a rapidly changing Himalayan foothill region is done. Over recent decades, land deformation has become a significant concern in Himalayan foothill regions due to a combination of natural processes and human activities, including seismic events, landslides, heavy rainfall leading to floods, and groundwater depletion. These factors pose severe risks to land surfaces, infrastructure, and underground structures such as tunnels and pipelines. The study aims to establish a framework for quantifying land deformation and analyzing its underlying causes through the application of a time-series InSAR approach using Sentinel-1 satellite data spanning from 2015 to 2021. The estimated deformation velocities were ranging from -4 mm to +2 mm per year. Additionally, changes in land use and land cover (LULC) are examined as potential drivers of land deformation. Landsat-8 imagery from 2015 and 2021 is used to generate LULC maps, which are compared with deformation analysis results. The study reveals an expansion of urban and agricultural areas at the expense of vegetated lands, potentially contributing to land subsidence due to excessive groundwater extraction. Moreover, areas with rugged terrain and steep slopes experience significant subsidence, likely linked to increased landslide occurrences triggered by heavy precipitation events. High precipitation is identified as a key catalyst for landslides and soil erosion, particularly in steep terrain. Furthermore, the study area, located in a seismically active zone with unstable slopes, is highly susceptible to land deformation induced by seismic activity. The developed methodology provides a valuable tool for ongoing monitoring and assessment of land deformation and its causes, with potential applicability to regions facing similar challenges. Later, the analysis of land subsidence activity in the Himalayan town of Joshimath is done. The region is prone to various natural disasters, including earthquakes, landslides, and human-induced activities such as construction projects. Recently, the geologically critical region of Joshimath has witnessed a rise in land subsidence and sinkage, posing risks to inhabitants. This study utilizes remote sensing methods, specifically time-series SAR Interferometry-based Persistent Scatterer Interferometric SAR (PsInSAR), to monitor land deformation measurements in mm/year using multi-temporal Sentinel-1 datasets. Additionally, feature tracking analysis is done using high-resolution Planet datasets, corroborating the PsInSAR results. The study estimates land deformation for different time periods using interferometric SAR datasets, with the highest subsidence observed during 2022-2023. The maximum deformation velocity is concentrated in the northwestern region of the town, particularly around Singhdwar, with velocities ranging from +103.22 mm/year to +187 mm/year. Other areas experience varying degrees of subsidence, with the south-east region showing rapid land subsidence. The research also investigates the causative factors of land deformation and assesses ground conditions using UAV datasets in critically affected areas. The work also provides recommendations for future development policies in Himalayan regions susceptible to land deformation. Further, the modeling of urban groundwater variation using land deformation is implemented, utilizing time-series Sentinel-1 datasets. Urban intensification has significantly impacted groundwater reserves, a crucial freshwater source. Groundwater exploitation has surged, particularly in urban areas, to meet population growth and development demands. This emphasizes the importance of proper monitoring of groundwater variations, which is a complex process for not being directly accessible for physical measurements. Therefore, developing an advanced indirect method is crucial for high-resolution, long-term monitoring of groundwater reserves in heavily exploited zones like cities, enabling the study of local variations and their impacts. This study focuses on monitoring groundwater variations using time-series Sentinel-1 Interferometric SAR (InSAR) data, specifically employing the PsInSAR technique to detect land deformation based on phase information from permanent scattering targets. The study area, Lucknow City (India), was assessed using 58 ascending and 60 descending pass images between 9/10/2014 and 2/7/2020. Groundwater level measurements were obtained from the Central Groundwater Board, Government of India (CGWB), for various seasons. Besides, Landsat 5 and 8 datasets were utilized to analyze the pattern of urban growth for a 30-year period and predict the near future scenario. An in-depth analysis of all the components revealed a direct relationship between land deformation, groundwater variations, and urban expansion, with a high correlation coefficient (0.886) observed between groundwater level changes and deformation measured at groundwater wells within deformation zones. The research work is also done towards developing a novel algorithm for estimating the snow wetness parameter using a hybrid polarimetric SAR technique. Snow wetness is a key input for mapping avalanche hazard zones and predicting wet snow avalanches, which have a high risk of affecting roads, railway networks, and human settlements. Existing methods for quantitatively analyzing snow wetness mostly rely on fully polarimetric SAR data, with very few utilizing hybrid polarimetric SAR (PolSAR) data. This study introduces a novel methodology for estimating snow wetness using the C-band hybrid polarimetric RISAT-1 SAR dataset. Using radar remote sensing to analyze the behaviour of such a snowpack requires information on the surface and volume scattering characteristics. The study utilizes modeled generalized surface and volume scattering parameters based on X-Bragg's reflection coefficients and Fresnel transmission coefficients for the inversion of surface and volume snow permittivity, respectively. The investigations were carried out in February 2014 for a study area of the Manali region in Himachal Pradesh, India. The retrieved snow estimates exhibit a coefficient of determination of 0.86 and a root mean square error of 0.667 when compared to in-situ measurements. Additionally, it is observed that the snow wetness estimates obtained from the proposed method using the RISAT-1 dataset outperform those derived from the fully polarimetric RADARSAT-2 dataset using the conventional Shi and Dozier methodThe research also focuses on the development of a deep-learning model aimed at detecting floods. SAR remote sensing has gained significant acclaim due to its ability to operate effectively regardless of weather conditions, offering continuous monitoring day and night. As a result, it has become a widely embraced technology for flood mapping and monitoring applications. In this study, Sentinel-1 satellite datasets sourced from the Sen1Floods11 benchmark flood label datasets were utilized to map floods across 11 distinct flood events. The aim was to differentiate between permanent water bodies and flooded areas using SAR imagery, with a focus on exploiting the polarimetric properties inherent in SAR datasets for flood mapping purposes. The study employed this dataset to improve flood detection by implementing the developed DB-SEN1FloodNet, also known as DeeplabV3PlusMX. The proposed model exhibited excellent performance during testing, achieving impressive metrics such as an overall test accuracy of 96%, an average precision of 98%, a mean recall of 94%, an F1 score of 96%, and a Mean IOU (Intersection over Union) of 0.65%. The research presented in this thesis significantly contributes to various techniques within Synthetic Aperture Radar (SAR) remote sensing, including time-series SAR interferometry, SAR backscatter modeling, and Deep Learning applications in SAR. These techniques are particularly valuable for assessing and monitoring land deformation, estimating snow geophysical parameters, and determining flood areas using SAR data. The thesis concludes with suggestions for future research directions to guide potential researchers in furthering advancements in SAR remote sensing methodologies.