University of Twente Student Theses
Navigating the Snowscape : Utilizing SAR Interferometric Coherence and ICESat-2 Data for Machine Learning-Based Snow Depth Estimation
Kabadwal, Manas (2024) Navigating the Snowscape : Utilizing SAR Interferometric Coherence and ICESat-2 Data for Machine Learning-Based Snow Depth Estimation.
|
PDF
18MB |
| Abstract: | Accurate snow depth estimation is crucial for understanding hydrological cycles, predicting water resources, and assessing the impacts of climate change on snow-covered regions. Snow depth data provides vital insights for climate modeling, water management, and disaster risk reduction, especially in regions dependent on snowmelt for freshwater resources. Enhancing the precision of snow depth measurement techniques is, therefore, essential for both scientific research and practical applications. This thesis delves into the utilization of ICESat-2 LIDAR data as a reference for training machine learning models to predict snow depth using SAR data from SAOCOM 1B and Sentinel-1. A meticulously crafted methodology was employed, encompassing advanced preprocessing, coregistration techniques, absolute coherence calculation, and sophisticated machine learning-based prediction models. The study’s comprehensive approach revealed that SAOCOM 1B’s L-band SAR data significantly surpasses Sentinel-1’s C-band SAR data in terms of snow depth prediction accuracy, primarily due to its superior penetration depth and more dependable coherence measurements. The Random Forest model was identified as the most effective machine learning algorithm, achieving an exceptional R-squared value of 0.8212 with the gt1r beam from ICESat-2, thereby underscoring the potential use of LIDAR-SAR data for snow-monitoring model generation. The feature importance analysis underscored the pivotal role of multi-polarization data (HH, HV, VH, VV) in augmenting the prediction accuracy, emphasizing the necessity of utilizing diverse polarization channels to capture the comprehensive range of snowpack characteristics. Despite the comparatively lower performance of Sentinel-1 data, this study highlighted potential avenues for enhancement in C-band SAR technology and data processing techniques. The findings elucidate the critical importance of selecting appropriate SAR datasets and leveraging robust machine learning models for accurate snow depth estimation. This research provides invaluable insights into the comparative strengths and limitations of different SAR datasets and different beams of ICESat-2, significantly contributing to the advancement of snow monitoring practices and a deeper understanding of snowpack dynamics within the broader context of climate change. |
| Item Type: | Essay (Master) |
| Faculty: | ITC: Faculty of Geo-information Science and Earth Observation |
| Subject: | 33 physics, 38 earth sciences, 43 environmental science, 54 computer science |
| Programme: | Geoinformation Science and Earth Observation MSc (75014) |
| Link to this item: | https://purl.utwente.nl/essays/102807 |
| 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