Soil texture mapping: a novel approach combining interpolation techniques and decision tree classifiers

Boudibi, Samir; Benguega, Zineeddine; Fadlaoui, Haroun; Khomri, Zine-eddine; Aissaoui, Azeddine; Sakaa, Bachir; Otmane, Tarik; Bouzidi, Narimen

doi:10.22098/mmws.2026.19187.1767

Soil texture mapping: a novel approach combining interpolation techniques and decision tree classifiers

مقالات آماده انتشار

نوع مقاله : پژوهشی

نویسندگان

¹ Centre de Recherche Scientifique et Technique sur les Régions Arides, CRSTRA, Biskra, Algeria

² Centre de Recherche Scientifique et Technique sur les Régions Arides, CRSTRA, Biskra, Algeria, and Department of Agronomic Sciences, Laboratory of Saharan Bioresources: Preservation and Valorization, Kasdi Merbah University, Ouargla, Algeria

10.22098/mmws.2026.19187.1767

چکیده

This study proposes a reproducible and GIS-based methodology for digital soil texture mapping by integrating geostatistical interpolation with deterministic decision tree classifiers (DTCs) derived from the United States Department of Agriculture (USDA) soil texture classification system. A total of 68 topsoil samples (0–20 cm) were collected across the irrigated area of northern Biskra province (southeastern Algeria) and analyzed for sand, silt, and clay contents. Among the most commonly applied interpolation techniques, ordinary kriging (OK), simple kriging (SK), and inverse distance weighting (IDW) were tested to generate continuous spatial distribution maps of soil particle fractions. Since the objective of this research was methodological demonstration rather than comprehensive benchmarking of interpolation algorithms, the method showing slightly better cross-validation (LOOCV) performance was selected. OK produced marginally lower RMSE values (15.93% for sand and 13.11% for silt) and satisfactory coefficients of determination (R²=0.758 for sand and 0.687 for silt) and was therefore adopted. To preserve the compositional constraint (sand + silt + clay=100%), clay content was derived from interpolated sand and silt maps. Four deterministic DTCs were implemented within the GIS environment to convert particle fraction rasters into continuous USDA texture classes. The final texture map demonstrated an almost perfect agreement with observed classifications (Kappa coefficient=0.898). The proposed framework emphasizes methodological simplicity, transparency, and applicability under moderate sampling density without reliance on auxiliary environmental covariates or complex machine learning models. Although interpolation uncertainty may influence classification near texture boundaries, the approach provides a practical and scientifically robust solution for soil texture mapping in data-limited regions.

کلیدواژه‌ها