PhD Final Defense – Siddharth Singh
- Event Type
- Seminar/Symposium
- Sponsor
- Civil and Environmental Engineering
- Location
- 3350, Yeh Center
- Date
- Dec 9, 2025 10:00 am
- Originating Calendar
- CEE Seminars and Conferences
Integrated Multi-Model, Multi-Frequency Remote-Sensing Framework for Characterization of Seasonal Snowpack Dynamics
Advisor: Ana P Barros
This dissertation develops and validates a multi-model, multi-frequency framework for retrieving seasonal snowpack properties by integrating physically based modeling, remote sensing, and statistical inference. It addresses a critical challenge in remote sensing snowpacks: the accurate estimation of snow surface and volume characteristics such as snow grain size, snow water equivalent, and snow depth under forest dominated heterogeneous land cover and topographic conditions using multisource satellite and airborne observations.
The first component of the study focuses on the spatial scaling behavior of surface reflectance and derived snow properties across multiple resolutions. A ten-year time series of reflectance from Landsat, MODIS, and VIIRS sensors over the Western Canadian domain of NASA’s Arctic and Boreal Vulnerability Experiment is analyzed to quantify and correct subgrid biases introduced by mixed pixels, particularly in forested terrain. Subgrid forest fraction, spatial heterogeneity, and spectral behavior are used as explanatory variables in a random forest model trained on high-resolution Landsat data to predict mean and standard deviation of reflectance at the VIIRS scale. The model achieves high predictive skill for mean reflectance and spatial variability and enables accurate correction of snow grain size, snow indices, and vegetation metrics, reducing biases in coarse-resolution retrievals. These results demonstrate that learning-based retrieval frameworks must explicitly incorporate spatial variability and land cover complexity to produce robust and physically consistent estimates.
The second component presents a Bayesian physical-statistical framework for estimating snow water equivalent and snow depth using dual-frequency airborne SAR observations collected during the SnowEx 2017 campaign in Grand Mesa, Colorado. A multilayer snow hydrology model forced by downscaled numerical weather prediction forecasts is used to generate prior distributions of snow properties. These are coupled with a forward radar scattering model and inverted through Bayesian inference. SnowSAR observations are aggregated to moderate spatial resolution and represented as equivalent single- or two-layer snowpacks. The retrievals achieve high accuracy with average residuals of five to seven percent compared to snowpit measurements and recover snow depth distributions that agree closely with airborne LiDAR, meeting key performance targets for satellite mission readiness.
The third component extends the retrieval framework to forested environments by incorporating vegetation effects using a coupled system of models that simulate snowpack evolution, vegetation scattering, and microwave interactions. Forest and background parameters are estimated using Ku-band SAR data and interpolated across space using kriging. The retrievals are validated against collocated LiDAR and snowpit measurements, showing strong agreement in both mean values and spatial distributions. The analysis also identifies key sources of uncertainty, including land cover misclassification, LiDAR underestimation under canopy, coarse atmospheric forcing, and empirical canopy parameterization.
These studies show that accurate snow retrievals over complex terrain require both physically grounded modeling and careful representation of spatial heterogeneity. The framework developed in this work integrates reflectance decomposition, multilayer snow simulation, and dual-frequency radar inversion to deliver consistent and transferable retrievals across resolutions and land cover types. The approach is suitable for natural, non-urbanized landscapes and supports future satellite missions aiming to monitor global snow water resources.