https://hdl.handle.net/10214/28941

Wet terrestrial ecosystems play a vital role in the Arctic, linking various hydro-ecological and permafrost processes. However, climate warming is altering these ecosystems, impacting their functional properties and spatial extents. Despite their importance, current maps are often limited by thematic detail, spatial resolution, or coverage. To better assess ecosystem status, vulnerability, and changes, more comprehensive datasets are needed. This Ph.D. addressed this research gap by studying several emerging Earth observation (EO), remote sensing (RS), and machine learning (ML) techniques for enhanced wet terrestrial ecosystem mapping and monitoring across the Arctic at various spatial scales, from local to regional. Each chapter examined new spaceborne synthetic aperture radar (SAR) datasets, given SAR's sensitivity to moisture and inundation. The initial chapters focused on high and very-high resolution (VHR) SAR imagery. First, the recently launched Radarsat Constellation Mission (RCM) was assessed for permafrost landscape mapping, with an emphasis on polygonal ice-wedge wetlands. An accurate Convolutional Neural Network (CNN) algorithm was developed for this task, suggesting strong potential for applications beyond local scales. Then, VHR Capella Space imagery was assessed for fine-scale classification of Arctic surface water features, by developing a deep learning (DL) U-Net model. Evaluation of the proposed U-Net showed favorable performance compared to other state-of-the-art ML models, while visual analysis underscored the precise detail captured by Capella imagery. This holds important implications for low latency, detailed surface water mapping, such as in flood-prone areas where conventional products are limited. Chapter 4 extended the applications of novel SAR techniques to better characterize Arctic hydro-ecological conditions at regional scales. By applying interferometric SAR (InSAR) analysis and ML on a time-series of Sentinel-1 imagery, an effective methodology was developed for monitoring seasonal ecosystem wetness conditions. The last chapter prioritized the operationalization of big EO data, including multi-wavelength SAR, for large-scale Arctic wetland mapping. By leveraging cloud-computing processing, a scalable and efficient approach for classifying Canada’s wetlands was established. These results will support future efforts around baseline circumpolar wetland mapping and reporting. Overall, this Ph.D. developed a suite of enhanced methodologies for characterizing wet Arctic ecosystems, by establishing the contributions of several rapidly developing geotechnologies.

https://hdl.handle.net/10214/28941