Quantitatively modeling soil–water distribution coefficients of three antibiotics using soil physicochemical properties

► Fourteen soil physicochemical properties were applied as the descriptors. ► The models were developed for soil–water partition coefficients of three antibiotics. ► Three models have high robustness and precise predictability. ► The models could predict soil–water partition coefficients of the antibiotics. Using 14 parameters featuring soil physicochemical properties and the partial least squares (PLSs) regression method, three quantitative models were respectively developed for the soil–water distribution coefficients (logKd) of oxytetracycline (OTC), sulfamethazine (SMZ) and norfloxacin (NOR) in 23 Chinese natural soil samples from cultivated lands in 19 provinces of China. The cross-validated correlation coefficients (Qcum2) of three models are 0.866, 0.765 and 0.868, and the standard deviations (SDs) are 0.16, 0.21 and 0.15 respectively. The high Qcum2 and low SD values indicate that three models have high robustness and precise predictability. Six parameters including pH, clay content, free Fe oxides (DCB-Fe), free Al oxides (DCB-Al), Ca content and Al content are greatly significant in the OTC model, three ones including pH, clay content and DCB-Fe are greatly significant in the SMZ model, and five ones including pH, clay content, DCB-Fe, Ca content and organic matter (OM) are greatly significant in the NOR model. The high VIP values of pH (1.17–1.24), clay content (0.81–1.10) and DCB-Fe (0.90–0.99) show that the three sorts of soil physicochemical properties play dominant roles in governing the partition balance between soil and water of three antibiotics.