Soil Quality Indicators and Machine Learning to Assess Iron Mining Waste Piles Revegetation in the Eastern Amazon

Abstract

In the eastern Brazilian Amazon, iron mining significantly impacts the environment. This study evaluated soil recovery in revegetated mining waste piles in the Carajás Mineral Province, focusing on chemical, physical, and biological indicators. Revegetation stages (0–2, 3–5, 6–8, and 9–11 years) were evaluated and compared to native tropical forest reference soils over 4 years, during the dry season. After 9–11 years, soil organic carbon (SOC), microbial biomass carbon (MBC), and enzymatic activity in revegetated soils reached levels similar to reference conditions. In early stages (0–2 years), microbial carbon was a key component, but from year 3 onward, plant‐derived organic material likely contributed to increased SOC. Distance‐based redundancy analysis showed significant temporal differences in microbial variables (p < 0.001), with SOC, Al3+, Mn, Ca2+, pH (CaCl2), and soil texture driving these changes. Random Forest modeling proved effective in identifying key soil indicators of recovery stages, with model performances of Overall Accuracy (OA) = 0.80 and Cohen's Kappa coefficient (CKC) = 0.74, achieving high predictive accuracy. Key predictor variables included β‐1,4‐glucosidase, MBC, clay content, Fe, and SOC. While findings demonstrate that revegetation improves soil quality and carbon dynamics, limitations include sampling restricted to the dry season and potential variability in mining waste materials and climate conditions. Still, this study highlights the importance of combining soil quality indicators with machine learning to support sustainable land management in tropical environments, as shown in the revegetation of mining waste piles in the Eastern Amazon.