Investigating soil properties influencing the depth and degree of lessivage in Florida soils using a random forest modeling approach

• Published in: Geoderma Regional Vol. 40; p. e00916
• Main Authors: Colopietro, Daniel J., Pachon, Julio, Bacon, Allan
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2025
• Subjects: acidity; algorithms; Argillic horizon; argillic horizons; clay; Clay translocation; coagulation; Eluvial zone; energy; Florida; forestry equipment; Illuvial zone; Lessivage; Machine learning; mineralogy; particle size; Random forest modeling; sand; surface area; U.S. Soil Taxonomy; Florida; Argillic horizon; MPE; WRB; Lessivage; RMSPE; NRCS; BS; SD; OC; UI; ANOVA; Random forest modeling; EA; PDP; Db; OM; CEC; OOB; RSF; Illuvial zone; RSG; ICE; RMSE; TEB; NCSS; VI; Clay translocation; Machine learning; Eluvial zone; LD; USDA
• ISSN: 2352-0094; 2352-0094
• DOI: 10.1016/j.geodrs.2025.e00916

The translocation of clay sized particles is an extensive soil process that occurs in over half of the total surface area covered by soil globally. However, investigations into lessivage are usually limited in extent and scale, and results from experimental data do not reflect natural conditions since control is exerted on various soil properties, such as particle size distributions, and clay mineralogy. Using the NCSS Soil Characterization Database and the USDA Soil Taxonomy definition for an argillic horizon, a random forest machine learning approach was used to investigate lessivage in Florida soils. Within the database, 395 soil profiles observed and described in Florida were identified to have an argillic horizon and lack the presence of a lithologic discontinuity. The coefficient of determination (R2), and the root mean square error (RMSE) were used to estimate the performance of the random forest model. The validation results showed that the model for the depth of lessivage achieved an R2 of 0.59 ± 0.07 and an RMSE of 27.86 ± 2.62 cm and the model for the degree of lessivage achieved an R2 of 0.68 ± 0.14 and an RMSE of 5.91 ± 3.25. A variable importance analysis indicated that the concentration of OC, the CEC, and the total extractable bases in the eluvial zone were the dominant soil properties in predicting the depth of the argillic horizon. The response in the predicted depth to the argillic horizon was similar with each of the aforementioned correlated variables and indicative of the flocculating mechanism, whereby polyvalent cations anchor OM to particle surfaces which then enables particle bridging. Whereas, extractable acidity in the eluvial zone and sand concentration in the illuvial zone were the dominant soil properties in predicting the degree of lessivage. As extractable acidity in the eluvial zone increased, the degree of lessivage decreased which could be indicative of coagulation due to an increase in the activity of Al3+ and H+ ions. Another mechanism identified in clay immobilization in this study is the cessation of the wetting front as a result of the loss of energy. As the depth to gleying increased, the predicted depth to the argillic horizon increased and the degree of lessivage increased. However, based on the variable importance analysis, this mechanism was not determined to be dominant in Florida soils.