Assessing Changes in Boreal Vegetation of Kola Peninsula via Large-Scale Land Cover Classification between 1985 and 2021

• Main Authors: Sklyar, E, Rees, G
• Format: Article in Journal/Newspaper
• Language: English
• Published: MDPI AG 2022
• Subjects: boreal forest; northern vegetation; human impact; climate change; land cover; monitoring; satellite imagery; Landsat; Russia; Arctic; Kola Peninsula; Climate change; kola peninsula; Scott Polar Research Institute; Tundra
• Online Access: https://www.repository.cam.ac.uk/handle/1810/344085; https://doi.org/10.17863/CAM.91509

Peer reviewed: True Funder: Cambridge Overseas Trust Funder: Trinity College Funder: Scott Polar Research Institute in Cambridge, UK <jats:p>The effective monitoring of boreal and tundra vegetation at different scales and environmental management at latitudes above 50 degrees North relies heavily on remote sensing. The vastness, remoteness and, in the case of Russia, the difficulty of access to boreal–tundra vegetation make it an ideal technique for vegetation monitoring in the Kola peninsula, located predominantly beyond the Arctic circle in the European part of Russia. Since the 1930s, this area has been highly urbanised and exposed to strong influence by a number of different types of human impact, such as toxic pollutions, fires, mineral excavation, grazing, logging, etc. Extensive open archives of remote sensing imagery as well as recent advances in machine learning further enable the efficient use of remote sensing methods for assessing land cover changes. Here, we present the results of mapping northern vegetation land cover and changes in it over a large territory, in time and under human impact based on remote imagery from Landsat TM, ETM+ and OLI. We study the area of about 37,000 km2 located in the central part of the Kola peninsula in the boreal, pre-tundra and tundra between 1985 and 2021 with a time interval of approximately 5 years and confirm the correlations between the human pressure and the level of vegetation changes. We put those into the perspective of year-on-year changes in the temperature and precipitation regimes and describe the recovery of the damaged original boreal vegetation (dominated by spruce) through pine and deciduous vegetation. As a by-product of this study, we develop and test an approach for the semi-automated processing and classification of Landsat images using the novel TensorFlow machine learning technique (widely spread across other disciplines) that enables high-throughput classification, even on conventional hardware.</jats:p>