GEMAS: PREDICTION OF SOLID-SOLUTION PHASE PARTITIONING COEFFICIENTS (K d ) FOR OXOANIONS AND BORIC ACID IN SOILS USING MID-INFRARED DIFFUSE REFLECTANCE SPECTROSCOPY LESLIE J. JANIK,*y SEAN T. FORRESTER,y JOS  E M. SORIANO-DISLA,yz JASON K. KIRBY,y MICHAEL J. MCLAUGHLIN,yx CLEMENS REIMANN,k and THE GEMAS PROJECT TEAM** yContaminant Chemistry and Ecotoxicology Program, Sustainable Agriculture Flagship Program, Waite Campus, CSIRO Land and Water, South Australia, Australia zDepartment of Agrochemistry and Environment, University Miguel Hernández of Elche, Alicante, Spain xSchool of Agriculture, Food and Wine, University of Adelaide, Glen Osmond, South Australia, Australia kGeological Survey of Norway, Trondheim, Norway (Submitted 17 April 2014; Returned for Revision 21 August 2014; Accepted 24 November 2014) Abstract: The authors’ aim was to develop rapid and inexpensive regression models for the prediction of partitioning coefficients (K d ), defined as the ratio of the total or surface-bound metal/metalloid concentration of the solid phase to the total concentration in the solution phase. Values of K d were measured for boric acid (B[OH] 3 0 ) and selected added soluble oxoanions: molybdate (MoO 4 2– ), antimonate (Sb[OH] 6– ), selenate (SeO 4 2– ), tellurate (TeO 4 2– ) and vanadate (VO 4 3– ). Models were developed using approximately 500 spectrally representative soils of the Geochemical Mapping of Agricultural Soils of Europe (GEMAS) program. These calibration soils represented the major properties of the entire 4813 soils of the GEMAS project. Multiple linear regression (MLR) from soil properties, partial least- squares regression (PLSR) using mid-infrared diffuse reflectance Fourier-transformed (DRIFT) spectra, and models using DRIFT spectra plus analytical pH values (DRIFT þ pH), were compared with predicted log K d þ 1 values. Apart from selenate (R 2 ¼ 0.43), the DRIFT þ pH calibrations resulted in marginally better models to predict log K d þ 1 values (R 2 ¼ 0.62–0.79), compared with those from PSLR-DRIFT (R 2 ¼ 0.61–0.72) and MLR (R 2 ¼ 0.54–0.79). The DRIFT þ pH calibrations were applied to the prediction of log K d þ 1 values in the remaining 4313 soils. An example map of predicted log K d þ 1 values for added soluble MoO 4 2– in soils across Europe is presented. The DRIFT þ pH PLSR models provided a rapid and inexpensive tool to assess the risk of mobility and potential availability of boric acid and selected oxoanions in European soils. For these models to be used in the prediction of log K d þ 1 values in soils globally, additional research will be needed to determine if soil variability is accounted on the calibration. Environ Toxicol Chem 2015;34:235–246. # 2014 SETAC Keywords: Anions Metal/metalloid Solid-solution partitioning coefficients (K d ) Mid-infrared spectroscopy Partial least squares regression Soil INTRODUCTION Solid-solution partitioning, which can be quantified using a partitioning coefficient (K d value), provides a simple measure of the distribution and potential mobility of metal/metalloid concentrations in soils [1,2]. The K d value is commonly expressed as the ratio of the surface-bound or total element concentration of the solid phase to the total element concentra- tion in the solution phase of soils [1–3]. The K d value, together with total element concentrations, can be used to provide information on the potential risk of metal/metalloids in soils through an understanding of the total concentration that is in a mobile or potentially bioavailable form in soils [1,3]. In spite of the usefulness of metal/metalloid partitioning data in our understanding of bioavailability and mobility, the determination of K d is time consuming and expensive for large-scale applications [3]. Regression models that can directly relate soil properties to K d values may provide a more rapid and cost-efficient approach to determining the partitioning of metals in a large number of soils with varying physical and chemical properties. However, regression modeling techniques themselves may be unsuitable for the direct prediction of soil solution concentrations [1]. Although direct bivariate regression models can be used to determine the relationship between metal/metalloid K d values and soil parameters such as pH, multivariate methods that utilize rapid spectroscopy techniques (e.g., near-infrared [NIR] and mid-infrared [MIR]) to determine important soil components may be better suited to predicting K d values [3]. Partial least-squares regression (PLSR) has been used recently to predict log K d þ 1 values for cationic metals (e.g., copper, zinc, and lead) from MIR diffuse reflectance Fourier-transformed (DRIFT) spectra of soils [3]. The use of DRIFT with PLSR for soil characterization was first reported by Janik and Skjemstad [4] and later extended to a wider range of soil properties [5–8]. The basis for this method relies on the correlation between the soil properties and infrared spectral * Address correspondence to Les.Janik@csiro.au ** S. Albanese, M. Andersson, R. Baritz, M.J. Batista, A. Bel-lan, M. Birke, D. Cicchella, A. Demetriades, B. De Vivo,W. De Vos, E. Dinelli, M. _Duriš, A. Dusza-Dobek, O.A. Eggen, M. Eklund, V. Ernstsen, P. Filzmoser, D.M.A. Flight, M. Fuchs, U. Fügedi, A. Gilucis, M. Gosar, V. Gregorauskiene, W. De Groot, A. Gulan, J. Halami c, E. Haslinger, P. Hayoz, R. Hoffmann, J. Hoogewerff, H. Hrvatovic, S. Husnjak, G. Jordan, M. Kaminari, V. Klos, F. Krone, P. Kwe cko, L. Kuti, A. Ladenberger, A. Lima, J. Locutura, P. Lucivjansky, A. Mann, D. Mackovych, B.I. Malyuk, R. Maquil, R.G. Meuli, G. Mol, P. N egrel, P. O’Connor, K. Oorts, R.T. Ottesen, A. Pasieczna, W. Petersell, S. Pfleiderer, M. Po nnavi c, C. Prazeres, U. Rauch, S. Radusinovi c, M. Sadeghi, I. Salpeteur, R. Scanlon, A. Schedl, A.J. Scheib, I. Schoeters, E. Sellersj€ o, I. Slaninka, A. Šorša, R. Srvkota, T. Stafilov, T. Tarvainen, V. Trendavilov, P. Valera, V. Verougstraete, D. Vidojevi c, Z. Zomeni. Published online 5 December 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/etc.2821 Environmental Toxicology and Chemistry, Vol. 34, No. 2, pp. 235–246, 2015 # 2014 SETAC Printed in the USA 235