Detection of soil organic matter using hyperspectral imaging sensor combined with multivariate regression modeling procedures

Abstract

Soil organic matter (SOM) plays an important role in the soil-plant system, increasing the search for faster and non-destructive methods for detecting this attribute. For this purpose, the use of hyperspectral imaging sensors is promising. This study aimed to evaluate the possibility of predicting organic matter through hyperspectral images in the laboratory associated with multivariate regression modeling procedures. There were 384 soil samples collected at eight depths in an Oxisol, in the State of Paraná, Brazil. Principal component and linear discriminant analysis were applied to the spectrum to group the sets according to depth. The hyperspectral imaging technique combined with the model's regression coefficient was used to classify the images relative to the content of organic matter. The partial least squares regression (PLSR) model was developed to correlate spectral data from the sensor with SOM contents acquired by the conventional method. The results obtained in the prediction were R2 = 0.75, r = 0.87 and RPD = 2.1. These results suggest that the model generated by the image sensor was able to discriminate the organic matter contents of the soil at different depths by capturing spectral variations.