Coal-fired power stations are significant sources of soil contamination with heavy metals and a source of hazard to human health. The soil samples (n = 25) selected in the area around Novocherkassk Power Station (Rostov Region, Russia) within a radius of up to 20 km revealed the enrichment with Pb, Cu and Zn. The heavy metals (HM) content in soil is reduced in the following sequence: Mn > Cr > Zn > Ni > Cu > Pb > Co. The correlation diagrams of the HM total content in soils revealed a significant association between the following HM pairs: Cu-Pb, Ni-Cu, Cd-Ni, Cd-Cu (r ≥ 0.7, p < 0.001). The concentration coefficient (Kc) and the total pollution coefficient (Zc) were used to estimate anthropogenic pollution. The use of generalized additive model (GAM) to detect the dependence of HM distribution on factors revealed the significance of the source distance. The influence of wind rhumb on HM distribution has a complex nonlinear nature. A GAM shows a good performance for all data sets: R2 = 0.71, 81% deviance explained for Zn, R2 = 0.85, 91% deviance explained for Cd, R2 = 0.63, 70% deviance explained for Ni. Thus, GAM model reveals significant factors (Dist_km, rhumb) in forming pollution by heavy metals in studied impact zone and proved a valuable approach to assess the degree and sources of pollution in soils on a large scale.