Abstract
<jats:p>The objective of this study was to identify spatial and temporal relationships between changes in climatic parameters, such as temperature and precipitation, and land use and land cover dynamics on the Kerch Peninsula for the period 1990–2020. The classification of Landsat multispectral images (TM, ETM+, OLI) utilizing a convolutional neural network resulted in the identification of eight LULC classes. ERA5 climate data were corrected using U-Net neural trained on data from four weather stations in the study area, thereby increasing the final accuracy. Spearman's pixel-wise correlation analysis revealed negative correlations of temperature with LULC in areas of artificial landscaping (forested areas, farmland), where vegetation reduces localized warming through evapotranspiration; positive correlations in the Southwest, where aridization and temperature increase are accompanied by a transition to low-productive classes (bare soils, barren lands); support of grass cover in the Northwest and increased degradation in urbanized and lakeside zones in the Southeast.</jats:p>