<p>The non-Lambertian characteristics of mountain shadows resulting from variations in relief significantly affect the effectiveness of tree species mapping. This study aimed to explore the impact of topography-induced mountain shadows on forest canopy spectral characteristics and forest mapping across multiple seasons, while quantitatively assessing their response to topographic correction models. Using multi-seasonal Sentinel-2 data, we extracted tree species distribution in a temperate forest landscape in northeast China, applying the random forest (RF) algorithm with various sampling modes (s/i = 0/100, 25/75, 50/50, 75/25, and 100/0) for shaded (s) and illuminated (i) areas. The effectiveness of four commonly used topographic correction models (Teillet, Cosine-C, SCS, and SCS + C) in mitigating mountain shadows was evaluated. The results indicated that mountain shadows affected the consistency of forest canopy spectra, particularly in autumn, when the coefficient of variation (CV) differed by 13.95–31.05%. In the s/i = 50/50 mode, mapping performance in both shaded and whole areas was higher than that in the other sampling modes, improving the mapping accuracy by 1.20–8.84% and 0.04–8.08%. In the models tested, the shadow mitigation effects were better in spring and autumn than in summer, with impacts reduced by 0.02–7.38%. In autumn, combining s/i = 50/50 and the Teillet model achieved the best mapping accuracy (86.72%) for the whole area, improving by 2.10–12.63%. This study suggests that mountain forest mapping should fully consider the effects of mountain shadows and carefully select the appropriate image seasons, sampling modes, and topographic correction models.</p>