In this study, we modeled and analyzed the four-decade dynamics of five key forest attributes — dominant tree species, forest site productivity class, forest type based on ground vegetation, growing stock, and stand age — across the Khanty-Mansi Autonomous Okrug–Yugra (Western Siberia) for the period from 1984 to 2024. We used Landsat image time series processed with the LandTrendr algorithm, NDVI and NBR vegetation indices, topographic variables derived from the FABDEM digital elevation model (elevation, slope, aspect, curvature), high-resolution CHELSA climate data, and ESA WorldCover land cover data as predictors and forest inventory data as target variables. Forest attributes were mapped using SegFormer deep learning models. Classification models achieved top-1 accuracies of 74.2% for dominant tree species, 71.0% for site productivity class, and 69.6% for forest type, with corresponding top-2 accuracies of 88.9%, 88.2%, and 85.2%. Regression models yielded a mean absolute error of 25.2m3/ha for growing stock (R2=0.70) and 28.5 years for stand age (R2=0.69). The replacement of primary pine forests by secondary broadleaved forests emerged as the most significant trend in the dynamics of tree species composition. Over the 41-year period, the area of pine-dominated forests decreased from 34.7% to 25.8% of the total area of the region, while the area of broadleaved stands increased from 3.9% to 10.3%, indicating post-disturbance replacement of conifers by birch- and aspen-dominated forests. Concurrently we observed an increase in areas occupied by species typical of late successional stages — Siberian pine and larch — from 3.0% to 7.66% and from 0.03% to 1.24%, respectively. An assessment of forest site productivity dynamics indicated a general improvement in site quality, as evidenced by a reduction in the proportion of low-productivity (class V) forests from 39.3% to 24.3%, and an increase in the share of classes IV and III from 4.6% to 15.1% and from 2.0% to 5.6%, respectively. Mean growing stock increased from 90.7 to 104.9m3/ha, while mean stand age remained relatively stable (120–130 years). The most pronounced forest type change was the decline in lichen-dominated highland forest types in the northern taiga from 11.6% to 2.5%, driven by increased fire frequency and slow lichen recovery. These findings provide the first annual maps of boreal forest dynamics in Western Siberia characterizing the combined effects of climate and disturbance on forest structure and productivity and supporting regional forest and carbon management. Keywords: Boreal forests; Western Siberia; Landsat time series; Deep learning; Tree species composition; Forest site productivity; Growing stock; Geospatial modeling.