Advanced methods for simulation-based performance assessment and analysis of radar sounder data

Radar Sounders (RSs) are active sensors that transmit in the nadir electromagnetic (EM) waves with a low frequency in the range of High-Frequency and Very-High-Frequency and relatively wide bandwidth. Such a signal penetrates the surface and propagates in the subsurface, interacting with dielectric interfaces. This interaction yields to backscattered echoes detectable by the antenna that are coherently summed and stored in radargrams. RSs are used for planetary exploration and Earth observation for their value in investigating subsurface geological structures and processes, which reveal the past geomorphological history and possible future evolution. RS instruments have several parameter configurations that have to be designed to achieve the mission science goals. On Mars, radargram visual analyses revealed the icy layered deposits and liquid water evidence in the poles. On the Earth, RSs showed relevant structures and processes in the cryosphere and the arid areas that help to monitor the subsurface geological evolution, which is critical for climate change. Despite the valuable results, visual analysis is subjective and not feasible for processing a large amount of data. Therefore, a need emerges for automatic methods extracting fast and reliable information from radargrams. The thesis addresses two main open issues of the radar-sounding literature: i) assessing target detectability in simulated orbiting radargrams to guide the design of RS instruments, and ii) designing automatic methods for information extraction from RS data. The RS design is based on assessing the performance of a given instrument parameter configuration in achieving the mission science goals and detecting critical targets. The assessment guides the parameter selection by determining the appropriate trade-off between the achievable performance and technical limitations. We propose assessing the detectability of subsurface targets (e.g., englacial layering and basal interface) from satellite radar sounders with novel performance metrics. This performance assessment strategy can be applied to guide the design of the SNR budget at the surface, which can further support the selection of the main EORS instrument parameters. The second contribution is designing automatic methods for analyzing radargrams based on fuzzy logic and deep learning. The first method aims at identifying buried cavities, such as lava tubes, exploiting their geometric and EM models. A fuzzy system is built on the model that detects candidate reflections from the surface and the lava tube boundary. The second and third proposed methods are based on deep learning, as they showed groundbreaking results in several applications. We contributed with an automatic technique for analyzing radargram acquired in icy areas to investigate the basal layer. To this end, radargrams are segmented with a deep learning network into literature classes, including englacial layers, bedrock, and echo-free zone (EFZ) and thermal noise, as well as new classes of basal ice and signal perturbation. The third method proposes an unsupervised segmentation of radargrams with deep learning for detecting subsurface features. Qualitative and quantitative experimental results obtained on planetary and terrestrial radargrams confirm the effectiveness of the proposed methods, which investigate new subsurface targets and allow an improvement in terms of accuracy when compared to other state-of-the-art methods.