Siddharth Singh

This study presents a physical-statistical Snow Water Equivalent (SWE) retrieval framework in forested areas using dual-frequency X- and Ku-band SAR airborne measurements. The methodology builds on previous work coupling snow hydrology and microwave propagation and backscatter models and introduces a parameterization of microwave propagation and scattering within the forest canopy based on the Water Cloud Model (WCM) modified to account for canopy closure effects. The retrieval framework was applied to SnowSAR measurements from four flights over Grand Mesa, Colorado and its performance was evaluated against snow pit observations and LiDAR snow depth measurements. Prior distributions of snowpack properties were generated using a multilayer snow hydrology model (MSHM) forced with Numerical Weather Prediction (NWP) analysis data. SAR measurements were spatially averaged to 90 m and 30 m resolution for retrieval. Prior distributions of vegetation and ground parameters were initialized using Ku-HH measurements, with effective soil and vegetation parameters estimated for frozen conditions. Soil parameters were estimated in open areas and spatially interpolated to nearby forested areas using ordinary kriging. SWE and snow depth retrievals for forested pixels at 90 m resolution were considered successful by accepting a relative residual backscatter (RRB) tolerance in the Bayesian optimization up to 30 % for individual pixels with incidence angles between 30°–50° along SnowSAR flight paths. Successful retrievals capture both the mean and spatial variance of snowpack properties across the Grand Mesa plateau consistent with the LiDAR survey with RRB generally below 5 %. Validation against collocated LiDAR snow depth and snow pit SWE measurements from the SnowEx ’17 campaign show a root mean square error (RMSE) of 0.033 m (< 8 % of maximum SWE for pits) and improved spatial patterns compared to snow hydrology predictions driven by NWP alone. The errors are larger in pixels with mixed land-cover (e.g., forest-grassland boundaries, land-margins of frozen ponds and lakes, and mixed forest) due to increased uncertainty in the estimation of vegetation parameters. Absolute relative differences (ARD) between LiDAR snow depth and SAR snow depth retrievals are below 10 % for 62 % of the forested pixels at 90 m resolution, and the fraction of successful retrievals increases to 82 % for ARD < 20 %. Retrievals at 30 m resolution achieve dramatic improvement with 78 % of the retrievals for ARD < 10 % due to reduced mixed land-cover uncertainty. These results demonstrate the feasibility of dual-frequency Bayesian SWE retrieval in forested landscapes by combining physical modeling with remote sensing at high spatial resolution enabled by SAR.

This study presents a physical-statistical Snow Water Equivalent (SWE) retrieval framework in forested areas using dual-frequency X- and Ku-band SAR airborne measurements. The methodology builds on previous work coupling snow hydrology and microwave propagation and backscatter models and introduces a parameterization of microwave propagation and scattering within the forest canopy based on the Water Cloud Model (WCM) modified to account for canopy closure effects. The retrieval framework was applied to SnowSAR measurements from four flights over Grand Mesa, Colorado and its performance was evaluated against snow pit observations and LiDAR snow depth measurements. Prior distributions of snowpack properties were generated using a multilayer snow hydrology model (MSHM) forced with Numerical Weather Prediction (NWP) analysis data. SAR measurements were spatially averaged to 90 m and 30 m resolution for retrieval. Prior distributions of vegetation and ground parameters were initialized using Ku-HH measurements, with effective soil and vegetation parameters estimated for frozen conditions. Soil parameters were estimated in open areas and spatially interpolated to nearby forested areas using ordinary kriging. SWE and snow depth retrievals for forested pixels at 90 m resolution were considered successful by accepting a relative residual backscatter (RRB) tolerance in the Bayesian optimization up to 30 % for individual pixels with incidence angles between 30°–50° along SnowSAR flight paths. Successful retrievals capture both the mean and spatial variance of snowpack properties across the Grand Mesa plateau consistent with the LiDAR survey with RRB generally below 5 %. Validation against collocated LiDAR snow depth and snow pit SWE measurements from the SnowEx ’17 campaign show a root mean square error (RMSE) of 0.033 m (< 8 % of maximum SWE for pits) and improved spatial patterns compared to snow hydrology predictions driven by NWP alone. The errors are larger in pixels with mixed land-cover (e.g., forest-grassland boundaries, land-margins of frozen ponds and lakes, and mixed forest) due to increased uncertainty in the estimation of vegetation parameters. Absolute relative differences (ARD) between LiDAR snow depth and SAR snow depth retrievals are below 10 % for 62 % of the forested pixels at 90 m resolution, and the fraction of successful retrievals increases to 82 % for ARD < 20 %. Retrievals at 30 m resolution achieve dramatic improvement with 78 % of the retrievals for ARD < 10 % due to reduced mixed land-cover uncertainty. These results demonstrate the feasibility of dual-frequency Bayesian SWE retrieval in forested landscapes by combining physical modeling with remote sensing at high spatial resolution enabled by SAR.

This study presents a physical-statistical Snow Water Equivalent (SWE) retrieval framework in forested areas using dual-frequency X- and Ku-band SAR airborne measurements. The methodology builds on previous work coupling snow hydrology and microwave propagation and backscatter models and introduces a parameterization of microwave propagation and scattering within the forest canopy based on the Water Cloud Model (WCM) modified to account for canopy closure effects. The retrieval framework was applied to SnowSAR measurements from four flights over Grand Mesa, Colorado and its performance was evaluated against snow pit observations and LiDAR snow depth measurements. Prior distributions of snowpack properties were generated using a multilayer snow hydrology model (MSHM) forced with Numerical Weather Prediction (NWP) analysis data. SAR measurements were spatially averaged to 90 m and 30 m resolution for retrieval. Prior distributions of vegetation and ground parameters were initialized using Ku-HH measurements, with effective soil and vegetation parameters estimated for frozen conditions. Soil parameters were estimated in open areas and spatially interpolated to nearby forested areas using ordinary kriging. SWE and snow depth retrievals for forested pixels at 90 m resolution were considered successful by accepting a relative residual backscatter (RRB) tolerance in the Bayesian optimization up to 30 % for individual pixels with incidence angles between 30°–50° along SnowSAR flight paths. Successful retrievals capture both the mean and spatial variance of snowpack properties across the Grand Mesa plateau consistent with the LiDAR survey with RRB generally below 5 %. Validation against collocated LiDAR snow depth and snow pit SWE measurements from the SnowEx ’17 campaign show a root mean square error (RMSE) of 0.033 m (< 8 % of maximum SWE for pits) and improved spatial patterns compared to snow hydrology predictions driven by NWP alone. The errors are larger in pixels with mixed land-cover (e.g., forest-grassland boundaries, land-margins of frozen ponds and lakes, and mixed forest) due to increased uncertainty in the estimation of vegetation parameters. Absolute relative differences (ARD) between LiDAR snow depth and SAR snow depth retrievals are below 10 % for 62 % of the forested pixels at 90 m resolution, and the fraction of successful retrievals increases to 82 % for ARD < 20 %. Retrievals at 30 m resolution achieve dramatic improvement with 78 % of the retrievals for ARD < 10 % due to reduced mixed land-cover uncertainty. These results demonstrate the feasibility of dual-frequency Bayesian SWE retrieval in forested landscapes by combining physical modeling with remote sensing at high spatial resolution enabled by SAR.

Time-series of surface reflectance and derived land cover and snowpack properties, including snow grain size (SGS), Normalized Difference Snow Index (NDSI), and Soil-Adjusted Vegetation Index were analyzed across multiple spatial resolutions using Landsat (30 m), MODIS (500 m), and VIIRS (1 km) observations over a 10-year period in the Western Canadian domain of NASA's Arctic and Boreal Vulnerability Experiment (ABoVE). The objective is to demonstrate a scalable framework for decomposing VIIRS reflectance signals and associated surface property estimates by accounting for the subgrid-scale heterogeneity introduced by mixed landcover types such as forests, wetlands, and open snow fields. The approach is to separate snow and vegetation reflectance contributions within coarse-resolution pixels and leverage this information toward improving retrieval of SGS, NDSI, and SAVI. The analysis showed that differences between VIIRS 1 km coarse scale reflectance and Landsat averaged reflectance at the same resolution are closely linked to subgrid-scale variability in forest fraction, spatial organization of forests (quantified by pixel-based chi square heterogeneity), and the spectral behavior of the subgrid 500 m MODIS and VIIRS bands. To address these discrepancies across scales, a random forest model was trained against high-resolution Landsat data to predict separately the mean and standard deviation of subgrid reflectance within each VIIRS pixel for forested and non-forested areas using VIIRS observations and ancillary data (topography and land-cover). Model results show strong agreement with Landsat for mean NIR reflectance (RMSE < 0.07, R² > 0.90) and good skill in capturing spatial variability (R² ∼ 0.5-0.7), particularly in heterogeneous forested regions. Predicted reflectance distributions were used to estimate NDSI, SAVI, and SGS, demonstrating improved agreement with Landsat-derived values and reducing known biases in VIIRS-based retrievals.

Time-series of surface reflectance and derived land cover and snowpack properties, including snow grain size (SGS), Normalized Difference Snow Index (NDSI), and Soil-Adjusted Vegetation Index were analyzed across multiple spatial resolutions using Landsat (30 m), MODIS (500 m), and VIIRS (1 km) observations over a 10-year period in the Western Canadian domain of NASA's Arctic and Boreal Vulnerability Experiment (ABoVE). The objective is to demonstrate a scalable framework for decomposing VIIRS reflectance signals and associated surface property estimates by accounting for the subgrid-scale heterogeneity introduced by mixed landcover types such as forests, wetlands, and open snow fields. The approach is to separate snow and vegetation reflectance contributions within coarse-resolution pixels and combine this information to produce a joint (mixed) distribution toward improving retrieval SGS, NDSI, and SAVI. The analysis showed that differences between VIIRS 1 Km coarse scale reflectance and Landsat averaged reflectance are closely linked to subgrid-scale variability in forest fraction, spatial organization of forests (quantified by pixel-based chi square heterogeneity) and the spectral behavior of the subgrid 500 m MODIS and VIIRS bands. To address these discrepancies, a random forest model was trained using high-resolution Landsat data to predict separately the mean and standard deviation of subgrid reflectance within each VIIRS pixel for forested and non-forested areas. Model results strong agreement for mean NIR reflectance (RMSE < 0.07, R² > 0.90) with Landsat and good skill in capturing spatial variability (R² ~ 0.5-0.7), particularly in heterogeneous forested regions. Predicted reflectance distributions were used to estimate NDSI, SAVI, and SGS, demonstrating improved agreement with Landsatderived values and reducing known biases in VIIRS-based retrievals. This work provides an unambiguous pathway to integrate VIIRS daily observations with machine learning based corrections for recovering the statistics of subgrid reflectance necessary to produce physically meaningful retrievals of snow cover properties in the absence of high-resolution data.

A physical–statistical framework to estimate snow water equivalent (SWE) and snow depth from synthetic aperture radar (SAR) measurements is presented and applied to four SnowSAR flight-line data sets collected during the SnowEx'2017 field campaign in Grand Mesa, Colorado, USA. The physical (radar) model is used to describe the relationship between snowpack conditions and volume backscatter. The statistical model is a Bayesian inference model that seeks to estimate the joint probability distribution of volume backscatter measurements, snow density and snow depth, and physical model parameters. Prior distributions are derived from multilayer snow hydrology predictions driven by downscaled numerical weather prediction (NWP) forecasts. To reduce the signal-to-noise ratio, SnowSAR measurements at 1 m resolution were upscaled by simple averaging to 30 and 90 m resolution. To reduce the number of physical parameters, the multilayer snowpack is transformed for Bayesian inference into an equivalent one- or two-layer snowpack with the same snow mass and volume backscatter. Successful retrievals meeting NASEM (2018) science requirements are defined by absolute convergence backscatter errors ≤1.2 dB and local SnowSAR incidence angles between 30 and 45^∘ for X- and Ku-band VV-pol backscatter measurements and were achieved for 75 % to 87 % of all grassland pixels with SWE up to 0.7 m and snow depth up to 2 m. SWE retrievals compare well with snow pit observations, showing strong skill in deep snow with average absolute SWE residuals of 5 %–7 % (15 %–18 %) for the two-layer (one-layer) retrieval algorithm. Furthermore, the spatial distributions of snow depth retrievals vis-à-vis lidar estimates have Bhattacharya coefficients above 94 % (90 %) for homogeneous grassland pixels at 30 m (90 m resolution), and values up to 76 % in mixed forest and grassland areas, indicating that the retrievals closely capture snowpack spatial variability. Because NWP forecasts are available everywhere, the proposed approach could be applied to SWE and snow depth retrievals from a dedicated global snow mission.

A physical-statistical framework to estimate Snow Water Equivalent (SWE) and snow depth from SAR measurements is presented and applied to four SnowSAR flight-line data sets collected during the SnowEx’2017 field campaign in Grand Mesa, Colorado, USA. The physical (radar) model is used to describe the relationship between snowpack conditions and volume backscatter. The statistical model is a Bayesian inference model that seeks to estimate the joint probability distribution of volume backscatter measurements, snow density and snow depth, and physical model parameters. Prior distributions are derived from multilayer snow hydrology predictions driven by downscaled numerical weather prediction (NWP) forecasts. To reduce noise to signal ratio, SnowSAR measurements at 1 m resolution were upscaled by simple averaging to 30 and 90 m resolution. To reduce the number of physical parameters, the multilayer snowpack is transformed for Bayesian inference into an equivalent single- or two-layer snowpack with the same snow mass and volume backscatter. Successful retrievals, defined by absolute convergence backscatter errors ≤ 1.2 dB and local SnowSAR incidence angles between 30° and 45° for X- and Ku-band VV-pol backscatter measurements, were achieved for 75 % to 87 % for all grassland pixels with SWE up to 0.7 m and snow depth up to 2 m. SWE retrievals compare well with snow pit observations showing strong skill in deep snow with average absolute SWE residuals of 5–7 % (15–18 %) for the two-layer (single-layer) retrieval algorithm. Furthermore, the spatial distributions of snow depth retrievals vis-à-vis LIDAR estimates have Bhattacharya Coefficients above 94 % (90 %) for grassland pixels at 30 m (90 m resolution), and values up to 76 % in mixed forest and grassland areas indicating that the retrievals closely capture snowpack spatial variability. Because NWP forecasts are available everywhere, the proposed approach could be applied to SWE and snow depth retrievals from a dedicated global snow mission.