Perspectives on machine learning: Predicting the combined effects and strategies of water management and biochar treatment on soil Cd activity and Cd accumulation in rice

Abstract

Currently, for cadmium (Cd) pollution in farmland, both the control of Cd contamination and the requirement of water conservation are urgent. Biochar passivation is a commonly used method for remediating Cd contamination in soils; however, under different water management conditions, the appropriate Cd concentration range for biochar remediation in Cd-contaminated soil remains to be investigated. In this study, machine learning was used to investigate the combined effects of biochar and water management on active Cd (DTPA-Cd, available Cd, and exchangeable Cd) in soils and Cd accumulation in brown rice. Moreover, the appropriate Cd concentration range for biochar remediation in Cd-contaminated soil under water flooding conditions and alternating wet-dry conditions was predicted. The results showed that active soil Cd and brown rice Cd were best fitted by the optimized GBDT model. The importance of factors affecting Cd uptake in rice grains was ranked as follows: soil properties (43.7 %) > biochar properties (30.3 %) > experimental conditions (24.6 %) > rice properties (1.4 %). Machine learning predictions highlighted that, for soils with pH values of 5.0, 5.5, 6.0, 6.5, and 7.0, when the Cd concentration in acidic soil exceeds 0.94, 1.25, 1.68, 2.12, and 2.42 mg·kg^-1, respectively, the use of water management alone should be avoided. Moreover, when the Cd concentration in acidic soil exceeds 1.25, 1.60, 1.95, 2.30, and 2.85 mg·kg^-1, respectively, even under flooded conditions, the application of biochar alone is not recommended, and other restoration techniques (e.g., silicon fertilizer, and green manure) should be coapplied.