Machine Learning-Based Mapping of Lake Ice Cover from SWOT KaRIn Backscatter: Preliminary Results
Bibliographic record
Abstract
Lakes are key components of the global freshwater system, playing a crucial role in climate regulation, hydrological cycling, and maintaining ecological balance. As highly sensitive indicators of climate change, lakes are recognized by the Global Climate Observing System (GCOS) as an essential climate variable (ECV), with lake ice cover (LIC) and lake ice thickness (LIT) identified as two of its thematic products. In the Northern Hemisphere, many lakes develop seasonal ice cover, which significantly influences local energy balance, ecosystem function, and socio-economic activities such as transportation, fishing, recreation, and tourism. Understanding the spatial distribution and temporal dynamics of the lake surface conditions is essential for numerous applications. For instance, accurate mapping of ice cover dynamics in lakes is crucial for predicting lake ice phenology, estimating ice thickness, and assessing the impacts of climate change on lake ecosystems. Due to the steady decline in long-term in situ observations of lake ice and overlying snow properties in recent decades, there is an increasing reliance on satellite remote sensing. These spaceborne approaches provide an effective alternative for investigating lakes at regional and global scales, providing a comprehensive and cost-effective means of monitoring these dynamic water bodies. The Surface Water and Ocean Topography (SWOT) mission, launched in December 2022, introduces the Ka-band Radar Interferometer (KaRIn), which offers high-resolution measurements of both surface water elevation and backscatter. While primarily designed for hydrological and oceanographic studies, SWOT's KaRIn backscatter data have shown sensitivity to surface conditions, suggesting potential applications in cryospheric monitoring. Our analyses demonstrate that KaRIn backscatter is sensitive to lake surface conditions and exhibits a clear distinction between ice-covered and open-water areas, providing a basis for classification efforts. Building on these initial findings, we develop a Random Forest classification model to distinguish between lake ice and open water using SWOT KaRIn backscatter data. Random Forest is a robust machine learning algorithm known for its effectiveness in handling complex, nonlinear relationships and has been widely used in remote sensing classification studies. The study focuses on five lakes in the Northern Hemisphere: Lake Teshekpuk in Alaska, as well as Kluane Lake, Great Slave Lake, Lake Athabasca, and Great Bear Lake, all located in Canada. Lake Teshekpuk and Kluane Lake are consistently observed during both the Cal/Val and Science phases of the SWOT mission, while Great Slave Lake, Lake Athabasca, and Great Bear Lake were partially covered during the Cal/Val phase, with full spatial coverage during the Science phase. The time period considered for performing the classification spans from March 30 to July 10, 2023 (Cal/Val phase), and from September 1, 2023, to July 15, 2025 (Science phase). To support classification and provide accurate labelling, we use supplementary satellite datasets including Sentinel-1 SAR, Sentinel-2 MSI, MODIS Aqua/Terra, and Landsat 8/9. These datasets provide additional information on surface conditions to ensure accurate reference labelling. By leveraging SWOT's high-resolution data and the capabilities of machine learning, this study aims to enhance the monitoring of lake ice phenology. It offers valuable insights into the value of Ka-band for mapping lake ice.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".