Espectroscopia de reflectância vis-NIR-SWIR-MIR na caracterização de perfis e avaliação de atributos de solos típicos do Nordeste do Brasil

Abstract

Soil is essential for life on Earth and for the performance of ecosystem services, making efficient monitoring of its attributes essential. Traditionally, soil is evaluated through wet chemistry laboratory analyses, but this technique has proven to be costly and time consuming. As a promising alternative, remote sensing using reflectance spectroscopy stands out for allowing rapid and low cost pedological analyses, which is especially advantageous for the study of heterogeneous regions susceptible to degradation, such as Northeast Brazil. Given this context, the objectives of this study were: to explore the potential of visible to mid-infrared spectra to characterize the spectral behavior of soils in the Northeast region; to quantify the physical and chemical attributes of these soils, including salinity and sodicity, using spectroscopy in the 350 - 15000 nm range and multivariate statistical algorithms. A total of 114 soil samples were evaluated, covering nine orders, collected from 24 soil profiles distributed across 13 cities in Ceará. Through wet chemistry analyses, the following attributes were determined: organic carbon, total nitrogen, available phosphorus, exchangeable potassium, calcium, magnesium, sodium, and aluminum, electrical conductivity, pH, base sum, cation exchange capacity, base saturation, and sodium saturation percentage. Spectral analysis evaluated soil reflectance in the visible, near infrared, short wave infrared (vis-Nir-SWIR), and mid infrared (MIR) ranges, with data preprocessed using Savitzky Golay smoothing and conversion to absorbance. The results of the conventional analysis were evaluated using descriptive statistics. The smoothed data were subjected to principal component analysis (PCA), and the PCA scores were used for cluster analysis with the unsupervised Fuzzy K means classification algorithm. To quantify the attributes, predictive models were developed with raw and preprocessed spectra using Partial Least Squares Regression, Support Vector Machine with linear and radial kernel functions, and Cubist Algorithm algorithms. The performance of the predictive models was evaluated by the metrics R², RMSE, RPD, and RPIQ. The results of the qualitative analysis rev ealed spectral behaviors with distinct patterns between soil orders. Unsupervised classification grouped samples based on horizon characteristics, with the MIR region showing greater sensitivity to identify more subtle variations between horizons. In the quantitative analysis, all attributes were predicted with at least satisfactory performance, except for calcium, which showed unsatisfactory performance with R² below 0.50. The models developed with MIR data consistently outperformed those in the vis NIR SW IR range for most attributes. The predictions of electrical conductivity and sodium saturation percentage showed reasonable performance, highlighting the potential of the technique for diagnosing soil salinity and sodicity. These highlighting the potential of the technique for diagnosing soil salinity and sodicity. These results confirm the potential of reflectance spectroscopy as an efficient and alternative tool for characterizing and predicting attributes in heterogeneous soils.