LAUSR

Abstract Acid sulfate soils are one of the most environmentally harmful soils existing in nature. This is because they produce sulfuric acid and release metals, which may cause several ecological damages. In Finland, the occurrence of this type of soil in the coastal areas constitutes one of the major environmental problems of the country. To address this problem, it is essential to precisely locate acid sulfate soils. Thus, the creation of occurrence maps for these soils is required. Nowadays, different machine learning methods can be used following the digital soil mapping approach. The main goal of this study is the evaluation of different supervised machine learning techniques for acid sulfate soil mapping. The methods analyzed are Random Forest, Gradient Boosting and Support Vector Machine. We show that Gradient Boosting and Random Forest are suitable methods for the classification of acid sulfate soils, the resulting probability maps have high precision. However, the accuracy of the probability map created with Support Vector Machine is lower because this method overestimates the non-AS soils occurrences. We also compare these modeled probability maps with the conventionally produced occurrence map. In general, the modeled maps are more objective and accurate than the conventional maps. Moreover, the mapping process using machine learning techniques is faster and less expensive.