Evaluation of Two Statistical Tools (Least Squares Regression and Artificial Neural Network) in the Multivariate Optimization of Solid-Phase Extraction for Cadmium Determination in Leachate Samples

This work proposes the use of multivariate optimization as a procedure for cadmium determination in leachate samples via flame atomic absorption spectrometry after solid phase extraction using a minicolumn packed with Amberlite XAD-4 modified with 3,4-dihydroxybenzoic acid. The variables related with the preconcentration (pH, sampling flow rate and buffer concentration) were optimized using Doehlert design. Two statistical modeling tools (least squares regression and artificial neural networks) have been applied to the data and their performances were compared. Digestion procedures of the leachate by heating in acid medium and ultraviolet radiation were evaluated being the latter more appropriate to prevent loss of Cd by volatilization. The developed procedure has promoted an enrichment factor of 9, with detection and quantification limits (3s_b) of 0.72 and 2.4 µg L^-1, respectively, and precision - expressed as relative standard deviation percentage - of 4.0 and 6.4% (RSD%, n = 4 for 5.0 and 20.0 µg L^-1, respectively). Addition/recovery tests for Cd were carried out and values between 97 and 112% were obtained. The procedure was applied for cadmium determination in leachate samples collected at the sanitary landfill of Jaguaquara-BA, Brazil.

cadmium; landfill leachate; solid-phase extraction; Doehlert design; least squares regression; artificial neural network