Ice detection for small lakes with satellite imagery and machine learning
Bibliographic record
Abstract
The role of satellite imagery in the continuous surface classification of the Earth has grown rapidly through the start of the 21st century. Advancements in machine learning and satellite imaging technology have facilitated the realisation of automated ice cover detection. Whilst sea ice detection has garnered great interest from researchers, the detection of lake ice has amassed considerably less focus. Specifically, few studies have evaluated ice detection methods for small lakes, which, albeit less prominent than larger water bodies, possess environmental and economic significance. This bachelor's thesis compares combinations of prevalent satellite imagery technologies and machine learning methods to find the optimal combination for small-lake ice detection. To discern the prevalent machine learning methods and imaging technologies, the trends and results of previous ice detection studies were explored. Synthetic aperture radar (SAR) and multispectral imaging (MSI) were identified as the two major imaging technologies applied to ice detection. For the machine learning methods, support vector machines (SVMs) and convolutional neural networks (CNNs) were found to be widely employed for ice detection research. The open availability and resolution of the SAR and MSI data provided by the Sentinel-1 and Sentinel-2 pairs of satellites led to their selection as the data source for an ice detection experiment detailed in this thesis. The experiment was performed by constructing several machine learning models for each combination of the chosen satellite imagery technologies and machine learning methods. The models were trained and initially assessed on data from two Canadian lakes. The global applicability of the models was then evaluated with a test set of images from lakes, which were not sources for the training set. The results of the experiment showed a significant difference in water-ice classification performance between the SAR and MSI models. Perfect validation accuracies were achieved with the MSI models, while the highest validation accuracy reached with a SAR model was 89.29%. For the test set, a CNN trained with MSI data attained the highest accuracy of 98.57%. The most accurate classification performance for SAR models was acquired with an SVM model, which managed a 73.20% accuracy. While the models trained on SAR data were only qualified for local indicative ice predictions, the test results suggest the CNN trained on MSI data to be globally applicable for small-lake ice detection.
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 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".