Radar altimetry for classifying surface conditions of subarctic lakes during freezing and thawing periods

Ice cover on subarctic lakes is an important indicator of climate change at local- or regional scales. This study proposes a new approach for classifying altimetry data from Jason-2 and SARAL/Altika satellite missions to characterize surface states of subarctic lakes. It focuses on Great Slave Lake (Canada) during freeze-up and thaw periods. For the first time, parameters from altimetry waveforms were used in an unsupervised clustering to establish distinct clusters from waveforms observed from Jason-2 and SARAL/Altika during the freeze-up and the thaw period. Clusters are assigned to the different surface states (open water, pure ice and leads) based on a priori altimetry and radiometric information. The statistics of these clusters were then used to construct two trained models of supervised classification based upon KNN (K-nearest neighbour) and SVM (support vector machine). The SVM-based model yielded the best results (accuracy of 92% with Jason-2, and 98% with SARAL/Altika). It was used to classify all waveforms considered in the study from the nominal orbits of Jason-2 (2008-2016) and SARAL/Altika (2013-2016). Results were superimposed onto Moderate Resolution Imaging Spectroradiometer (MODIS) products for qualitative visual and semi-quantitative assessments.

Automated summary

This study proposes a new approach for classifying surface states of subarctic lakes using satellite data. Two supervised classification models based on SVM were constructed, achieving high accuracy in identifying different surface states. The models were applied to classify multi-year satellite data and compared with other products for evaluation.