Soil sampling and mapping of trace elements are challenging in Arctic regions. However, understanding the spatial distribution of them is crucial for effective environmental management and conservation efforts. Validating such digital maps is also a key consideration, especially in data-constrained conditions. In this study, we first evaluated the feasibility of digital soil mapping methods for spatial modeling of trace and heavy metals in the Russian Arctic with limited soil data. The random forest method was used to predict topsoil concentrations of As, Cd, Cr, Hg, and Ni, in combination with environmental covariates, including climate, relief, vegetation, geology, geographic position, anthropogenic impact, soil properties, and soil class maps. We used a clustered legacy dataset and tested map accuracy with spatial and nonspatial cross-validation (CV) techniques. We also estimated the uncertainty of the predictions and the area of applicability (AOA). We found that spatial and nonspatial CV resulted in different model performances, where conventional CV methods showed better performance. The mountainous area was beyond the AOA of all predictive models. Soil and climate variables were key predictors in the RF models. Despite the challenges of modeling trace elements in this study with sparse data, future efforts should prioritize expanded sampling in underrepresented environments, especially with distinct soil-forming factors, and multistage validation strategies tailored for clustered datasets. The presented mapping can be a starting point for regional and national initiatives to map trace elements and metals in northern regions.