University of Twente Student Theses
Snow Depth Estimation Using Different SAR Image Modalities
Duggal, Tishya (2024) Snow Depth Estimation Using Different SAR Image Modalities.
|
PDF
4MB |
| Abstract: | Estimating snow depth is vital for various purposes like hydrology, climate modelling, avalanche risk assessment, and winter sports. While remote sensing techniques, particularly Synthetic Aperture Radar (SAR), have made significant advancements in estimating snow depth, there remains a lack of comparative assessments of various SAR modalities. This includes employing different polarization modes, frequencies, and sensor platforms for accurate snow depth estimation over complex terrains. This study thus employs SAR data from various platforms, including Sentinel-1, SAOCOM, and UAVSAR, to investigate the impacts of different polarisation configurations (dual polarimetric (DP) and quad polarimetric (QP)), different frequencies (C- and L- bands), and different SAR sensor platforms (airborne and spaceborne) on snow depth estimation. The study focuses on five sites situated in the mountainous terrains of the western USA. By utilizing NASA’s SnowEx snow depth data as a reference and employing a Random Forest regression model, this study reports a comparative assessment of SAR-based snow depth estimation performance across polarization, frequency, and sensor platforms. The results indicated varying degrees of accuracy among the different SAR datasets, with UAVSAR operating as an airborne system in quad-polarised mode and its L-band frequency consistently achieving the best performance across most study areas, exhibiting the lowest RMSE and MAE values. Following it, spaceborne SAOCOM’s quad-polarimetric L-band data performed notably well but still with higher RMSE and MAE values than UAVSAR. Next in line, dual-polarimetric spaceborne data from both Sentinel-1 (C-band, 10 m spatial resolution) and SAOCOM (L-band, 50 m spatial resolution) often showed comparable error levels, with Sentinel-1 achieving slightly lower RMSE and MAE values amongst the two datasets due to an advantage of higher spatial resolution. Overall, the SAOCOM dual-polarimetric dataset yielded the least accurate results across most study areas. Visual inspection of the maps revealed that the UAVSAR predictions had the closest resemblance to the reference snow depth maps among all other datasets. The models struggled to predict the full range of snow depth values accurately thus leading to a compressed range of values in the predicted maps. The difference maps indicated that large errors were mainly concentrated in regions with extreme snow depth values, where the models tended to underestimate the higher values and overestimate the lower values. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 38 earth sciences, 74 (human) geography, cartography, town and country planning, demography |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/101408 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page