Sorption of phenylurea herbicides by soils: Modelling uncertainty and ruling factors with an artificial neural-based fuzzy inference system

TítuloSorption of phenylurea herbicides by soils: Modelling uncertainty and ruling factors with an artificial neural-based fuzzy inference system
AutoresB. K. Agbaogun, J. M. Alonso, K. Fischer
TipoComunicación para congreso
Fonte the Society of Environmental Toxicology and Chemistry (SETAC) North America 39 th Annual Meeting, Sacramento (USA), Society of Environmental Toxicology and Chemistry (SETAC) North America, 2018.
ISSN1087-8939
AbstractRetention of pesticides by soils is both spatially variable and also one of the most sensitive factors determining losses to surface and groundwater. To date, only a few work has been done to explain this process in tropical soils especially, and generally to uncover the factors that govern the process in both temperate and tropical soils. The purpose of this study was therefore to evaluate the influence of various and interrelated specific soil properties and the pesticide specific molecular descriptors on sorption of pesticides in tropical soils, and to develop and test a simple explainable sorption model. The sorption behavior of five representatives of the phenylurea family (PUH) were studied in twelve soils of contrasting characteristics, stemming from the Southwestern Nigeria. Sorption isotherms and coefficients were obtained by equilibrating the soil samples with 0.01M CaCl 2 solutions spiked with increasing concentrations of the target PUHs. HPLC-DAD was used to quantify the target PUHs in solutions at equilibrium. The isotherm data were fitted to Freundlich and linear distribution equations (R2 ≥ 0.96 and R2 ≥0.76, respectively) to obtain the isotherm and distribution coefficients (Kf, and Kd). Spearman rank correlation was used to determine the specific soil and PUHs properties that have significantly high correlations with Kf or Kd. Significant correlations were established between Kd or Kf and the following factors: soil properties (pH, cation exchange capacity, organic carbon, extractable Fe and Mn content, and mass proportions of clay and silt), and the pesticide molecular descriptors [molecular weight (M w ) and hydrophobicity index (logK ow )]. These were used alongside with PUHs concentrations as potential descriptors. However, due to the problem of multicollinearity between these predictors, multiple linear regression and other similar statistical approach were inadequate to elaborate a model, hence the use of an artificial neural-based fuzzy inference system (ANFIS). Several models of different combinations of the factors were then elaborated under 10-fold cross-validation. Furthermore we carried out a sensitivity analysis of the factors to establish the most important predictor. In conclusion, we successfully used ANFIS to construct a holistic 4 factor system that could model the sorption of PUHs in tropical soils. The model should assist in understanding the fate of pesticides, and for their risk assessment in tropical soils.