Brightness Values-Based Discriminant Functions for Classification of Degrees of Organic Matter Decomposition in Soil Thin Sections

Abstract

Read online

The decomposition of organic matter represents a fundamental pedogenetic process, since it impacts the carbon cycle and the release of nutrients to the soil. However, quantitative research aimed at micro-scale in situ analysis is scarce, despite its relevance in the decomposition process. Therefore, the objectives of this research were to generate discriminating functions of the degrees of organic matter decomposition, based on the brightness values associated with each morphological stage, and from this step, to generate thematic maps. Soil thin sections of forest and compost soils were selected, and petrographic microscope images with three light sources were taken: plane polarized light (PPL), crossed-polarized light (XPL), and crossed polarizers and a retardation plate (gypsum compensator) inserted (XPLλ). Subsequently, the RGB (red, green, blue) image was broken down into three bands, resulting in nine bands for each image. Two thousand sampling points were generated for each band, obtaining brightness values for each decomposed organic matter stage. The points were classified into four categories based on their degree of decomposition: no (A), light (B), moderate (C), and strong (D), in addition to porosity (P). Linear discriminant analysis was performed to obtain classification models for each level of decomposition. The results show that each degree of organic matter decomposition can be highlighted through specific light sources and a set of bands, with an overall accuracy of >94% and kappa coefficients of >0.75 for all classes. In addition, the resulting functions were validated in training images and high-resolution mosaics to create final thematic maps. The use of linear models automated the production and quality of thematic maps at the microscopic level, which can be useful in monitoring the organic matter decomposition process.

Keywords

• linear discriminant analysis
• micromorphometry
• thematic maps
• supervised training
• variable selection