Lake ice monitoring in the European Alps

Abstract

Lake ice cover provides a variety of ecological and cultural ecosystem services. Therefore, the loss of lake ice cover can lead to a range of ecological consequences. With increasing pressure on account of climate-change the timing of seasonal lake ice freeze-up and break-up, also referred to as lake ice phenology (LIP), is subject to change. Recent research shows that lakes around the Northern Hemisphere have progressively earlier ice‐off dates, later ice‐on dates and a resulting shorter ice duration, up to the point where lakes fail to freeze. Due to its critical linkage to the characterization of the Earth's climate, lake ice phenology has been introduced by the Global Climate Observing System (GCOS) as part of the 54 Essential Climate Variables. We present a proof of concept for an extensive monitoring system for the European Alps based on a multi-sensor remote sensing approach. As part of this thesis, we created a complete processing chain implemented in the Google Earth Engine for the extraction of lake ice coverage and lake ice phenology down to lake sizes of 0.1 km² using random forest classifiers. Our accuracy assessment with webcam-based ground observations shows that we can extract phenological key dates with a high accuracy (R² of 0.98) and a mean difference of -0.17 ±4.3 days. Our analysis using leave-one-lake-out cross-validation showed that the performance for unseen lakes drops significantly. Thus, model generalization over the entire European Alps has to be proven in future studies with more extensive validation data before establishing an operational monitoring system.