Home – Home – Jornals – Earth’s Cryosphere 2026 number 3

The article considers the application of machine learning methods for predicting the retreat rates of the coasts composed of frozen unlithified sediments. To test machine learning technologies, three well-studied coastal sites in the southwestern part of the Kara Sea were selected. These sites have been monitored by the Laboratory of Geoecology of the North at Lomonosov Moscow State University since the 1980s. Data have been collected at 10-meter intervals for each site, including categorical characteristics of the coastal zone (geomorphological level, lithology, dominant permafrost process) and quantitative indicators of coastal retreat rate. A sequential application of correlation analysis using the Random Forest algorithm and the dispersion analysis (Welch’s ANOVA) made it possible to identify features that most significantly affect the retreat rate of the bluff, as well as to determine zones with the highest risk of intensive changes and average values of the coastal retreat rate with a certain combination of categorical features. The study proposes an approach that enables the segmentation of the sea coastline based on its resilience to key natural factors affecting coastal dynamics in the permafrost zone using machine learning technologies.