A comparative analysis of interannual changes in the ice coverage of the Pechora and the Kara Seas has been conducted based on the information of open sources (satellite observations) gathered by the Arctic and Antarctic Research Institute and the open dataset with mapped ice conditions ArcticDEM. The following problems of the study have been solved: the analysis of the most common approaches to building a forecast based on the analysis of ice conditions has been conducted; a deep learning algorithm to solve the problem of analysis of ice fields has been developed and tested; a medium-term forecast of the condition of ice fields in the area of the Pechora and the Kara Seas has been prepared based on the approbation model example; the obtained results have been analyzed and processed. In the course of study, an algorithm of automated mapping of ice conditions based on the satellite imagery data has been developed; the possibilities of its application to forecast ice conditions automatically, i.e., without using a manmade mapping, have been described. It has been demonstrated that the use of images based on multiple spectral bands and the combination of data from different satellites can minimize the influence of clouds and weather conditions. The model gave the real forecasting accuracy for a period of 10–14 weeks equal to 49–53 % for the Barents (Pechora) and the Kara Seas of 2018–2021. In case the model was fine-tuned and included data from the Barents (Pechora) and Kara Seas before 2018 in the training sample, the accuracy on the data after 2018 increased up to 52–56 %. The feature of the model is its ability to operate with different classes of ice and precisely determine its amount. The developed method increases the efficiency of operational control over the ice conditions in the Arctic seas. Safe navigation routes for shipping transport in the Pechora Sea have been modeled.