How qualitative spectral information can improve soil profile classification?

Abstract

Soil classification is important to organize the knowledge of soil characteristics. Spectroscopy has increased in the last years as a technique for descriptive and quantitative evaluation of soils. Thus, our objective was to assess qualitative and quantitative methods on soil classification, based on model profiles. Soils in different environments in the Roraima state, Brazil, were evaluated and represented by 16 profiles, providing 109 soil samples, which were analyzed for particle size distribution, chemical attributes and spectral measurement. Visible-near infrared spectra (350–2500 nm) of soil samples were interpreted in terms of intensity, shape and features. The soil color obtained using a spectroradiometer and a colorimeter, and by a soil expert was compared. Descriptive and qualitative analyses were performed for all spectra of the soil profile samples. The descriptive evaluations of the spectral curves from all horizons of the same profile were used to identify the diagnostic attributes and assign a profile to a taxonomic class. This was possible because spectra of samples had specific shapes, features and intensities that combined to present a specific signature. The Outil Statistique d’Aide à la Cartogénèse Automatique and cluster quantitative analyses could not correctly group similar soil classes and they still need to be improved in order to extract all the variability of the spectral data to discriminate soil classes. Soil color quantification by the Munsell system using both equipments showed greater R2 and lower error than that achieved by a soil expert, due to influences of subjectivity inherent in human assessments. Based on this specific case, it was clear that the automatic system may be more consistent than the pedologist’s visual method. Future studies should focus on the development of an online tool that integrates a descriptive approach and spectral information of a given soil profile to determine its probable taxonomic class.

Soil Color and Mineralogy Mapping Using Proximal and Remote Sensing in Midwest Brazil

“…), value (lightness), and chroma (purity, similar to saturation), which can be calculated from spectra using mathematical formulas. We used spectral reflectance data to calculate the Munsell soil color at three depth intervals (0-20 cm, 20-60, and 60-100 cm), according to Marques et al [19] and Rizzo et al [20]. The method used as input only the reflectance values between 380 and 780 nm (visible spectral range), and followed the steps: (1) Spectra were integrated using color-matching functions (x, y, z) to the XYZ color system for illuminant D 65 (daylight) and 2nd standard observer [50], (2) XYZ tristimulus values were converted to the CIELAB color system (L*a*b*), (3) coordinates a* and b* were used to calculate hue angles and chroma, while value was estimated by L*, and (4) hue angle was converted to Munsell notation using a color conversion table [51].…”

“…We did not verify the accuracy of the color estimations at each site because: (1) We lacked colorimeter records in our dataset; (2) spectral data were acquired under the same conditions as in reference works [19,20]; and (3) the mathematical procedures of reference, implemented in this section, provided similar color estimations to the colorimeter measurements, with R 2 ranging from 0.68 to 0.96 and RMSE between 0.19 and 0.57 [19,20].…”

“…As soil color and mineralogy are important proxies used to distinguish different soil types or to infer related soil attributes [17], they play an important role in soil cartography. Some studies have used reflectance spectroscopy (350-2500 nm) as input data to estimate the color and/or its mineralogy [18][19][20][21][22][23][24], but only a small number of works mapped their spatial distribution. At the moment, Viscarra Rossel et al [25] performed one of the few studies on soil color mapping, where they accurately mapped (R 2 0.67) iron oxides and the color of Australian soil using reflectance spectra (350-2500 nm) and geostatistics.…”

“…Reflectance spectroscopy data (350-2500 nm) was successfully used in pedometry as predictor of the soil color and mineralogy [18][19][20][21][22][23][24], nevertheless, only a small set of research mapped their spatial patterns. At the moment of this work, Viscarra Rossel et al [25] performed one of the few studies on soil color mapping, where the authors also mapped iron oxides of Australian soil using reflectance spectra (350-2500 nm) and geostatistics.…”

“…Besides that, XRD measurements are qualitative and not conducive to quantitative analyses [34]. Soil color in national datasets [33] also lacks or does not contain spatial referencing or was visually approximated, which has been proven to be less consistent than modern colorimeter measurements [19].…”