Estimating particle-size fractions of soil dominated by silicate minerals from geochemistry B. G. RAWLINS a , R. WEBSTER b , A. M. T YE a , R. L AWLEY a & S. L. O’HARA c a British Geological Survey, Keyworth, Nottingham, NG12 5GG U.K., b Rothamsted Research, Harpenden, Hertfordshire, AL5 2JQ, U.K., and c Department of Geography, University of Nottingham, University Park, Nottingham, NG7 2RD U.K. Summary Measuring particle-size distribution (psd) is so time-consuming and expensive that it is rarely affordable for systematic survey. In young soil dominated by silicate minerals the size fractions have characteristic elemental compositions, and so they might be predicted from geochemical analysis. We tested the feasi- bility of predicting the psd from concentrations of elements in samples of topsoil taken in surveys of two large regions in eastern England and on which the psd had been determined. Of the 35 elements measured we chose eight (a) from our general knowledge of mineral composition, (b) after a principal component analysis to avoid redundancy and (c) experience elsewhere. Of the eight elements, five were used (Al, Fe, Ni, Ti and Zr) to build multiple linear regression models for the predictions. The equations were assessed by their coefficients of determination, R 2 , and the effectiveness of their predictions, expressed as root- mean-square errors (RMSEs) on validation sets of data of known psd. The models accounted on aver- age for 89% of the variance in the clay size-fraction and 82% for sand. The corresponding median RMSEs were 4.9% and 8.8% on medians of 17% and 58%, respectively. The silt size-fraction was less well predicted; R 2 was only 0.58, and the median RMSE was 10.6% on a median of 22%. We judge our approach, in which the regression models may be regarded as pedotransfer functions, to have been moderately successful and to merit attention in similar circumstances in other regions. Introduction One of the most important attributes of the soil, if not the most important, at any place is the particle-size distribution (psd). It largely determines the soil’s hydraulic behaviour and water stor- age, its handling characteristics under tillage and its susceptibil- ity to erosion. The clay component is the principal source of potassium for plant growth, and most micronutrients are sorbed to it. Hand texturing is much used to estimate psd; with experience and calibration against known standards it can be used reliably to estimate particle-size fractions (Hodgson et al., 1976). For many years psd has been measured in the laboratory by com- binations of sieving and sedimentation in water, or by laser granulometry, which detects the scattering of light, after the removal of organic matter. These methods are both laborious and time-consuming, however, and pedologists have sought cheaper, more rapid substitutes. Estimates of psd which are based on the attenuation of gamma rays (Elias et al. 1999) have also been made, although this approach has not been widely adopted for large-scale soil survey. A promising, alternative technique is visible and near infra- red diffuse reflectance spectroscopy (VNIR-DRS) which detects differences in the spectra reflected from soil and resulting from bending and stretching of atomic bonds. The spectra are there- fore related to mineral composition, which can be used to esti- mate particle-size fractions. An advantage of this particular technique is that it does not require the prior removal of organic matter. The approach is based on statistical models between soil reflectance at specific wavelengths, commonly their first deriva- tives, in a series of samples and measurements of their particle size. Published studies suggest that when applied to independent samples, predictions from these models have modest error and bias (Brown et al., 2005; Waiser et al., 2007). There are limitations to the widespread application of VNIR- DRS, however. First, there are subtle differences between re- flectance spectra recorded by different individual spectrometers; so statistical models developed for one instrument cannot be transferred automatically to spectra from another. Second, coarse-textured soils dominated by quartz have few spectral features, and this makes it difficult to establish reliable statistical models for estimation of their particle-size fractions. Finally, estimates cannot be made for soil samples with unusual spectra (spectral outliers; Shepherd & Walsh, 2002) because such Correspondence: B. G. Rawlins. E-mail: bgr@bgs.ac.uk Received 23 July 2008; revised version accepted 7 October 2008 116 # 2009 NERC Journal compilation # 2009 British Society of Soil Science European Journal of Soil Science, February 2009, 60, 116–126 doi: 10.1111/j.1365-2389.2008.01112.x European Journal of Soil Science