Environ. Sci. Technol. 2009, 43, 4049–4055
Direct Comparison between Visible Near- and Mid-Infrared Spectroscopy for Describing Diuron Sorption in Soils M O H S E N F O R O U Z A N G O H A R , * ,†,§ DANIEL COZZOLINO,‡ RAI S. KOOKANA,§ RONALD J. SMERNIK,† SEAN T. FORRESTER,§ AND DAVID J. CHITTLEBOROUGH† School of Earth and Environmental Sciences, University of Adelaide, PMB 1, Glen Osmond 5064, Australia, The Australian Wine Research Institute (AWRI), Waite Road, Glen Osmond 5064, Australia, and Commonwealth Scientific and Industrial Research Organisation (CSIRO) Land and Water, PMB 2, Glen Osmond 5064, Australia
Received October 24, 2008. Revised manuscript received April 14, 2009. Accepted April 14, 2009.
Both visible near-infrared (VNIR) and mid-infrared (MIR) spectroscopy have been claimed to better predict pesticide sorption in soils than other methods. We compared the performances of VNIR and MIR spectroscopy for predicting both organic carbon content (fOC) and the sorption affinity (Kd) of diuron in 112 surface soils from South Australia. Separate calibration models were developed between VNIR and MIR spectra, and fOC and Kd using partial least-squares (PLS) regression. MIR clearly outperformed VNIR for predictions of both fOC and Kd in soils. Correlation (R2) and accuracy (RPD) indices were 0.4 and 1.3 for the VNIR-PLS model versus 0.8 and 2.3 for the MIR-PLS model, respectively, for Kd prediction. PLS loadings for sorption prediction were compared in terms of the soil information they contained. While VNIR loading did not include any direct spectral information regarding soil minerals, MIR loading included peaks associated with sand, clays, and carbonates. Perhaps by better predicting fOC and integrating the effects of OC as well as minerals, the MIRPLS model provided a better prediction for diuron Kd values in our calibration set.
Introduction Conventional laboratory methods of soil analysis are usually time-consuming, expensive, and sometimes use toxic materials. When it comes to the analysis of a large number of soils, in particular, they may not be even feasible. However, there is an increasing need for soil characteristic data for agricultural and environmental purposes. Therefore, alternative analytical techniques are being sought to provide accurate yet inexpensive and rapid quantitative soil data. During the last two decades, infrared (IR) spectroscopy has increasingly been used for quantitative analysis of soil * Corresponding author phone: +61 (8) 8303-3831; fax: +61 (8) 8303-6717; e-mail:
[email protected]. † University of Adelaide. ‡ AWRI. § CSIRO. 10.1021/es8029945 CCC: $40.75
Published on Web 05/04/2009
2009 American Chemical Society
parameters. In this context, multivariate calibrations are developed between IR spectra and the reference analytical data of the soil property of interest (1). Since 1988, when Haaland and Thomas (2) first utilized partial least-squares (PLS) regression techniques for spectral analysis, attempts have been made to quantitatively analyze soils by IR spectroscopy. Janik et al. (3) reviewed several studies in which the IR spectroscopic methods in combination with chemometrics (e.g., PLS regression) were used to predict soil properties. More recently, Sorensen and Dalsgaard (4) showed that the performance of the calibration models by near-infrared (NIR) spectroscopy and PLS analysis was excellent for the determination of clay content but not for total carbon. In situ estimation of soil clay content has also been achieved by visible near-infrared (VNIR) spectroscopy (5). Unlike total carbon, soil organic carbon content was shown to be highly predictable using the VNIR-PLS modeling approach (6). Cozzolino and Moron (7) and Zimmermann et al. (8) went one step further and showed that NIR and mid-infrared (MIR) spectroscopy are able to successfully quantify the OC content in different soil fractions. Furthermore, NIR spectroscopy has been shown to be capable of predicting the chemical structure of soil OC, as well as its content (9). Since the chemical and physical properties of minerals and organic matter in soils can be determined by IR techniques, hypothetically, any other soil characteristics that are related to or dependent on those properties could also be predicted by these techniques. One such property is the sorption of nonionic organic compounds (e.g., pesticides) in soils. Previously, the sorption of nonionic pesticides was reported to be solely related to soil OC content (10). In current literature, however, the extent of sorption of nonionic pesticides onto the soil matrix is reported to be influenced by the effects of variability of organic matter chemistry among soils (11-13), interactions of organic matter with minerals that may block the sorption sites on the organic matter (14), and direct sorption by minerals (15). In our previous studies, MIR-PLS calibration models provided a good prediction of soil sorption coefficients (Kd) of nonionic test compounds atrazine and diuron because the MIR spectra contain information on the amount and nature of soil mineral and organic components (16, 17). The MIR-PLS model can integrate this information for an improved prediction compared to the KOC (OC-normalized sorption coefficient) modeling approach for the prediction of Kd, which is based only on the soil OC content (10). The variance explained in calibration (R2) was improved from 0.42 to 0.69 and the standard error (SE) was reduced from 7.26 to 5.57 (16). In another study, Bengtsson et al. (18) found that the NIR-PLS technique predicted the sorption coefficients of two nonionic pesticides (lindane and linuron) in some Swedish soils with high R2 of around 0.85 and very low root-mean-square error (RMSE) of 0.14. They did not report the extent of improvement in the prediction by the NIR-PLS method as compared to the KOC model. A number of studies directly comparing the efficacy of NIR versus MIR spectroscopy for predicting soil characteristics (e.g., soil OC content) have found much better performance by MIR spectroscopy (19-21). This is probably because the spectral information in the MIR region for soil OC is of higher quality than that in the NIR region (20). The peaks in the MIR region result from fundamental molecular vibrations as opposed to the NIR region where peaks result from overtones and combinations of fundamental vibrations. VOL. 43, NO. 11, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4049
No direct comparison between NIR and MIR for predicting the sorption of pesticides in soils has been published yet. Sorption data are essential input data required in environmental models. While IR techniques are increasingly being used for obtaining soil data, a direct comparison between NIR and MIR spectroscopy methods could assist researchers and environmental modelers in choosing the right IR region of the electromagnetic spectrum when it comes to the prediction of sorption of pesticides in soils. The emphasis in this comparison is on the quality of the information available in each IR region (VNIR versus MIR) for predicting diuron (a nonionic pesticide) sorption in soils.
Materials and Methods Test Compound. Diuron [3-(3,4-dichlorophenyl)-1,1-dimethylurea] was chosen as a test compound, representing nonionic pesticides (see Supporting Information for a chemical structure diagram as well as some physiochemical properties of diuron). It was purchased as white crystalline powder (>99% purity) from Sigma-Aldrich (Sydney, Australia). Diuron is a substituted phenylurea-based pre-emergence herbicide for the control of many agricultural and nonagricultural weeds, which is applied to soil. It is absorbed by roots and translocated to shoots where it inhibits photosynthesis (22, 23). As with other nonionic pesticide molecules, soil organic matter plays the predominant role for its sorption onto the soil matrix. It is chemically stable and persistent to hydrolysis within the normal pH range found in the environment (pH 4-9) (23). Soils. A set of 112 soils was used in this study, including 101 soils from our previous study (16), along with 11 additional calcareous soils. In our previous study (16), a couple of calcareous soils in that set were identified as spectral outliers in Principal Component Analysis (PCA) and this caused problems with the subsequent PLS analysis for these soils. By increasing the number of calcareous soils from 3 to 14 in this study, we avoided this problem (see below). All soil samples were from the surface A horizon of soil profiles available in a South Australian Soils Database (24). The soils cover a wide range of OC content and diuron sorption affinities. Soil samples were stored air-dried in the archive and had already been sieved to
