Multitemporal and Multivariate Pedological Pattern Analysis of Machinery-Based Tillage Systems (No-Till and Chisel) Integrating Machine Learning Frameworks

Long-term tillage management fundamentally reshapes soil’s physical and chemical environment, yet an integrated, predictive characterization of the distinct chemical signatures induced by no-tillage (NT) versus chisel tillage (CT) remains limited. We analyzed an eight-year dataset (2010–2017) from a long-term experiment in Iowa, USA, focusing on pH, available phosphorus (Bray1-P), and macro- and micronutrients (K, Ca, Mg, Cu, Fe, Zn) at two depths (0–5 and 5–15 cm). A convergent multi-method framework combined robust univariate statistics, multivariate ordination (PCA, PERMANOVA), linear mixed-effects models, and machine learning (Random Forest and Firth-penalized logistic regression). Results reveal a clear stratification–homogenization pattern. NT is associated with surface accumulation of Zn (+14%), Fe (+16%), and Cu (+5%), with mild acidification (−0.4 pH units) and high temporal stability. CT favored vertical nutrient redistribution, marked by subsurface K enrichment (up to 6% higher than NT), progressive alkalinization, and greater temporal variability. Predictive modeling highlighted subsurface K and surface Zn/Fe as key discriminators, with Firth regression confirming their complementary effects. These findings indicate that long-term NT and CT are associated with distinct, depth-specific chemical configurations—integrated systems defined by concentration gradients, temporal stability, and element covariation—rather than isolated element changes. This work provides a robust, quantitative framework for diagnosing soil management history and characterizing the pedochemical imprint of tillage.