Remote Sensing of Soil-Snow-Vegetation Continuum Over Boreal Forests and Permafrost Landscape Using L-Band Radiometry
Kaydedildi:
| Yayımlandı: | ProQuest Dissertations and Theses (2025) |
|---|---|
| Yazar: | |
| Baskı/Yayın Bilgisi: |
ProQuest Dissertations & Theses
|
| Konular: | |
| Online Erişim: | Citation/Abstract Full Text - PDF |
| Etiketler: |
Etiket eklenmemiş, İlk siz ekleyin!
|
| Özet: | The soil–snow–vegetation continuum plays a fundamental role in regulating global water, energy, and carbon cycles, yet it remains one of the least understood and observed components of the Earth system due to the scarcity of in situ observations in snow-covered regions and the limitations of existing satellite retrievals.This thesis advances the use of L-band passive microwave radiometry for characterizing the coupled soil–snow–vegetation continuum in boreal and Arctic landscapes. A new emission model, TO-snow, was developed by extending the existing tau–omega framework through a closed-form solution of Maxwell’s equations, treating the intervening dry snow layer as a low-loss medium. Constrained inversions over CONUS with SMAP brightness temperatures show that explicitly accounting for snowpack effects significantly reduces retrieval biases, yielding soil moisture and vegetation optical depth (VOD) estimates with root-mean-squared errors of 0.060 m3 m−3 and 0.124, respectively, when validated against International Soil Moisture Network data and MODIS-derived indices. Building on this foundation, the model was further used to retrieve VOD and ground permittivity during snow-covered months across the Arctic–Boreal Zone, generating new global datasets causally validated against biomass, tree height, ground temperature, and carbon flux, providing critical indicators of ecosystem functioning in snow-dominated Northern Hemisphere landscapes.Leveraging the insights from physical modeling, this work introduces deep learning frameworks to capture the temporal complexity of ground freeze–thaw (FT) processes. A deep convolutional autoencoder reframes FT detection as a probabilistic time-series anomaly detection problem, retrieving FT states from SMAP observations with an 11% accuracy improvement over conventional normalized polarization ratio thresholding and reducing uncertainties across heterogeneous snow- and vegetation-covered terrains. Finally, this thesis provides the first demonstration of a direct link between peak snow water equivalent (SWE) and temporal variability in L-band surface emission. Using a transformer-based architecture, SWEFormer, the study shows that long-range dependencies in L-band time series can be exploited to estimate peak SWE. To address the challenge of limited in-situ data for model training, a transfer learning approach is employed, transferring knowledge from reanalysis datasets to sparse in-situ training samples. SWEFormer outperforms operational global products (ERA5, GlobSnow, AMSR-E/2), particularly in boreal watersheds where canopy attenuation limits high-frequency microwave retrievals.Taken together, these contributions underscore the unique advantages of L-band radiometry for snow-covered landscapes, bridging observational gaps that have long hindered monitoring of cold-region hydrology and carbon fluxes. By combining physics-based radiative transfer modeling with modern deep learning approaches, this thesis establishes new theoretical and practical frameworks for soil moisture, vegetation, FT dynamics, and peak SWE retrievals, with direct implications for land data assimilation, Earth system modeling, and the prediction of Arctic–boreal feedbacks under climate change. |
|---|---|
| ISBN: | 9798263320287 |
| Kaynak: | ProQuest Dissertations & Theses Global |