The main objective of this paper is to predict soil attributes in unsampled areas using geostatistical models, By improving the prediction parameters of selected data, using environmental covariates characteristic of Antarctic ice free areas. In this study, 58 soil samples from a grid were collected at 0-10 cm depth in Keller Peninsula, King George Island, Antarctica. The soil chemical analysis was performed, and the values of potassium, calcium and magnesium were determined for each soil sampled. Simple kriging (SK) and Random Forest interpolator were used in this work to predict the values of the studied soil attributes in non-sampled areas. We used a Terrestrial Laser Scanner (TLS) to generate a cloud of points, to obtain digital elevation models (DEMs) of 1, 5, 10, 20 and 30 meters cell size. The use of covariates did not improve the parameters of soil bases prediction in the studied area. The final maps did not show great differences based on RMSEs, mainly related to the great spatial variability of soil attributes in this region, despite soil thematic maps evidencing visual difference.