Reflectance spectroscopy vis-NIR and mid-IR applied for soil studies

• Autor(a) principal: Araujo, Suzana Romeiro
• Data de Publicação: 2013
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: http://www.teses.usp.br/teses/disponiveis/11/11140/tde-02042013-133023/
• topic: Absorção; Absorption; Análise do solo; Carbon; Carbono; Data modelizing; Espectroscopia; Heavy metal; Metal pesado do solo; Modelagem de dados; Poluição do solo; Soil analysis; Soil physical chemical properties; Soil pollution; Solos - Propriedades físico-químicas; Spectroscopy

Effective agricultural planning and environmental monitoring requires basic soil information. However, analyzing soil properties by conventional methods is often expensive and time consuming. In addition, these analyses result in chemical residues, which may be environmentally hazardous. In recent decades near-infrared diffuse reflectance spectroscopy (400-2500 nm) has been shown to be a viable alternative for rapidly analyzing soil properties. Information needs to be mathematically extracted from the spectra in order to correlate them with soil properties, and multivariate statistics are often used to calibrate soil prediction models.However, soils evaluated by the mid-IR region (4000 to 400 cm-1) warrants new studies. The primary aim of this study was to investigate the feasibility to use soil spectral data and chemometrics methods to predict soil properties, in order to reduce the number of conventional soil analyses. The understanding of the relationships between spectral characteristics and the physic-chemical properties of soils were evaluated in three different studies with soils of: (i) spectral library (Chapter 1), (ii) amazonian region (Chapter 2), (iii) soils contaminated with heavy metals and tannery sludge (Chapter 3).It was possible to identify regions of the vis-NIR and mid-IR spectra that showed absorption features due to water, iron oxides, and clay minerals. In Chapter 1 the predicted models for clay and soil organic matter showed high accuracy. It reflects the influence of the direct spectral responses of these properties in the NIR. The division of the large library into smaller subsets based on variation in the spectra characteristics was the best alternative to quantify soil attributes in tropical soils by Partial Least Square regressions. Another alternative would be to use Boosted regression trees for the whole library. In Chapter 2, the mid-IR predicted models outperformed the vis-NIR. Comparison of the interpolation results revealed that the predictions of the PLS regression (mid-IR and vis-NIR) adequately reproduced the spatial pattern of the properties evaluated, especially soil organic carbon and cation exchange capacity and, had the ability to predict the soil properties of unknown samples from a different geographical location. In Chapter 3, the metals adsorption to soil constituents caused expressive changes in soil spectral curves, showing spectral differentiation between highly contaminated soil and soils that are relatively contaminant-free. The results indicate that the Cr pseudo-total content can be predicted by spectroscopy reflectance with both sensors data. Fe and Mn also can be predicted accuratley by vis-NIR. The vis-NIR models outperformed the mid-IR. Besides these results, the vis-NIR instrument has less complicated sample and can be used directly in the field using portable spectrorradiometers.