Development of methods to map and monitor peatland ecosystems and hydrologic conditions using Radarsat-2 Synthetic Aperture Radar

Peatland ecosystems exhibit a wide range of biophysical conditions and Synthetic Aperture Radar (SAR) remote sensing provides a method to collect information about these conditions over large areas.The ability to extract useful hydrologic and vegetation information across peatlands is currently limited due to complex interactions of spatially and temporally-variable conditions on the SAR response.The overarching purpose of this thesis was to advance our understanding of SAR backscatter response to peatland hydrology and vegetation and to develop new approaches for remote mapping and monitoring of peatland environments with SAR.Specifically, this thesis aimed to 1) improve methods for peatland ecosystem mapping and classification accuracy assessment with a Random Forest classifier;and 2) develop methods for surface soil moisture and water table depth retrieval in peatlands using SAR remote sensing data.At Alfred Bog, a peatland in eastern Ontario, Canada, active remote sensing data (SAR and Light Detection and Ranging) were used for these purposes.A Random Forest classification workflow was developed, resulting in an improved peatland ecosystem mapping technique.Recommendations for appropriate training data sample selection with this classifier were also developed.This workflow enabled the creation of a sitewide peatland ecosystem map, which was used to better understand the SAR response to hydrological and vegetation conditions within the different peatland ecosystem classes.For the retrieval of surface hydrologic information, groups of highly correlated variables were identified from a large number of SAR parameters iii (including SAR intensity, polarimetric decomposition and discriminator variables) and a subset of these were compared with trends in soil moisture, water table and vegetation spatial variability and change over time.The Freeman-Durden Power due to Rough Surface parameter was found to be positively correlated with soil moisture, while the Touzi AlphaS1 parameter was found to be negatively correlated with water table depth from the surface.Various polarimetric parameters were used to build statistical models of soil moisture and, in some cases, CART-models resulted in high explained variance but independent validation indicated that these models were over-fit.These results are important, as many examples were found in the literature where, through statistical models, SAR was reported to be a strong predictor of soil moisture but models were not properly validated.To determine if models could predict soil moisture from SAR at times when no field measured data existed, linear mixed effects models were built that accounted for the temporal autocorrelation due to the repeated measures design of field data.While some models resulted in high explained variance, most of the explained variance was attributed to the variability between peatland classes and/or the specific date that the image was acquired, rather than the SAR data itself.These models also presented challenges in independent validation.Overall, this thesis points to some fundamental limitations on our ability to accurately monitor peatland hydrology with SAR due to the complexity of the scattering response where complex surface conditions exist.It highlights a need for extensive field monitoring campaigns and testing to further refine approaches for remote hydrologic monitoring in natural environments.iv Acknowledgements First, I would like to thank my supervisor, Murray Richardson, for his support and guidance over the past few years.Thank you for taking me on as a student, encouraging me to gain experience a wide variety of environments and technologies, passing on so many important skills and for being available to talk issues and questions through throughout this entire process.I would also like to thank my committee members (Doug King and Scott Mitchell) for providing valuable comments and guidance throughout the various stages of this research.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.