Mapping soil drainage classes: Comparing expert knowledge and machine learning strategies

Soil drainage is an essential factor that influences plant growth and various biophysical processes, such as nutrient cycling and greenhouse gas fluxes. Therefore, soil drainage maps are fundamental tools for managing crops, forests, and the environment. This study compared two approaches to mapping soil drainage classes in the state of São Paulo, Brazil, using geographic information systems (GIS). The first approach employed expert knowledge (EK) to develop a simple model based on soil color and texture, while the second used machine learning (ML) with an extensive set of covariates and a decision tree algorithm. To evaluate the full, operational implementation of soil mapping, this study assessed the two approaches in terms of accuracy, labor efficiency, transferability, interpretability, and agreement/disagreement statistical methods. In terms of accuracy, the ML-based strategy showed greater agreement with the reference map (53 %) compared to the EK approach (50 %). However, the EK strategy was more time- and resource-efficient, as well as being more transferable and interpretable due to the simplicity of its rules based on soil properties. Given its higher interpretability and ease of application, the EK approach was recommended as the most suitable for operational soil drainage mapping in tropical environments. • Use of remote sensing and predictive models to map soil drainage classes. • Qualitative and statistical criteria for the assessment of prediction performance. • Two strategies for mapping soil drainage classes in Brazil were tested and rated. • The ML-based strategy had the highest soil drainage map accuracy. • In the selection criteria, the EK-based strategy received the best score.