Spatial modeling of soil textural classification in a fragment of the Atlantic Forest biome using the random forest algorithm
Description
This study aimed to model the spatial distribution of soil textural classes in the Itupararanga Environmental Protection Area (EPA), Brazil, using the Random Forest (RF) algorithm. A total of 138 soil samples were collected considering variations in topography and land use. Sand, silt, and clay fractions were interpolated using the Inverse Distance Weighted (IDW) method and used as predictor variables. The dataset was divided into 70% for training and 30% for testing, and model performance was assessed using validation metrics. The RF model (ntree = 60; mtry = 3) showed excellent predictive performance, achieving 97% overall accuracy, a Kappa index of 0.95, and an Out-of-Bag error of 2.5%. Clayey and sandy clay soils predominated, representing 82.4% of the study area. Higher clay and silt contents were associated with steeper slopes, indicating greater susceptibility to erosion. The results demonstrate the effectiveness of integrating geoprocessing and machine learning for soil texture mapping and environmental planning.
Files
Institutions
- Universidade Estadual Paulista (Unesp)São Paulo, São Paulo