A Machine Learning Framework to Automate the Classification of Surge‐Type Glaciers in Svalbard

Abstract

Surge‐type glaciers are present in many cold environments in the world. These glaciers experience a dramatic increase in velocity over short time periods, the surge, followed by an extended period of slow movement, the quiescence. This study aims at understanding why only few glaciers exhibit a transient behavior. We develop a machine learning framework to classify surge‐type glaciers, based on their location, exposure, geometry, climatic mass balance and runoff. We apply this approach to the Svalbard archipelago, a region with a relatively homogeneous climate. We compare the performance of logistic regression, random forest, and extreme gradient boosting (XGBoost) machine learning models that we apply to a newly combined database of glaciers in Svalbard. Based on the most accurate model, XGBoost, we compute surge probabilities along glacier centerlines and quantify the relative importance of several controlling features. Results show that the surface and bed slopes, ice thickness, glacier width, climatic mass balance, and runoff along glacier centerlines are the most significant features explaining surge probability for glaciers in Svalbard. A thicker and wider glacier with a low surface slope has a higher probability to be classified as surge‐type, which is in good agreement with the existing theories of surging. Finally, we build a probability map of surge‐type glaciers in Svalbard. The framework shows robustness on classifying surge‐type glaciers that were not previously classified as such in existing inventories but have been observed surging. Our methodology could be extended to classify surge‐type glaciers in other areas of the world.

Global clustering of recent glacier surges from radar backscatter data, 2017–2022

“…The approach also provides surge timing. Bouchayer and others (2022) use statistical approaches (machine learning) to classify surge-type glaciers on Svalbard. From the 25 glaciers surging during 2017–22 according to our data, three to four mostly small and narrow ones have a surge probability of smaller 50% in the automated classification.…”

Automated Detection of Glacier Surges from Sentinel-1 Surface Velocity Time Series—An Example from Svalbard

“…However, the time component of glacier surges are often times not available in large scale studies, but crucial to many modern-day approaches in resolving unanswered questions about surge probability, distribution, and behaviour. Further measurements and detection allowing large-scale comparison could fill gaps in knowledge existent up to this day [32]. However, with extending both spatial and temporal coverage, automatization and computational inexpensive data handling become a factor.…”

Overcoming Overfitting Challenges with HOG Feature Extraction and XGBoost-Based Classification for Concrete Crack Monitoring

“…(4) improved accuracy XGBoost has several adjustable hyperparameters, such as learning_rate, subsample, and colsample_bytree, to optimize the model's accuracy. By optimizing these hyperparameters, XGBoost is able to produce more accurate models than other machine learning techniques [32]- [36].…”