Espectroscopia de reflectância na quantificação de atributos em solos do semiárido

Abstract

The use of reflectance spectroscopy in soil characterization has become an efficient alternative in identifying physical, chemical and biological attributes, with the main objective of this research being to quantify and characterize soil attributes in the Brazilian Semiarid region, using the statistical method of Regression. by Main Components (PCR). Therefore, the physical attributes were analyzed: sand, silt and clay, the chemical attributes: calcium, magnesium, potassium, assimilable phosphorus, sodium, aluminum, pH (H2O), potential acidity, CTC calculation and base saturation, as well as the variability of soil attributes, based on spectral information in the range of 350 to 2500 nm from 164 soil samples from Quixadá, Pentecoste, Fortaleza (CE), Goiana (PE), Rio Largo (AL), Neópolis (SE), Cruz das Almas (BA) and Brejo (MA). The chemical and physical results and spectral analyzes were obtained with the FieldSpec Pro FR 3 equipment, subjected to pre-processing techniques, continuous spectrum removal methods and conversion of values to absorbance. For the quantitative analysis and prediction of soil attributes, the spectra of the samples were used, as well as the pre-processed data. To build models for quantifying soil attributes, the PCR algorithm was used. For calibration, 70% of the soil curves were used and 30% for validation. To validate the models, the leave-one-out cross-validation method was used, evaluated following the metrics of the Coefficient of Determination (R2), Root Mean Square Error (RMSE) and the Deviation Performance Ratio (RPD). In relation to the R2 results, for the physical attributes the use of pre-processed data presented the best model when compared to continuum removal and absorbance, for the chemical attributes Mg and Ca they presented the best models in all the processing carried out. For RMSE, sand using raw data, silt removal from the continuum and clay using absorbance showed better precision, while for chemical attributes, K was the attribute that presented the best precision in the three processes, when compared to the other components. For the RPD, the sand presented an adequate model with the raw spectra, but requires adjustments, and for the chemical attributes of Ca and Mg, the raw spectra presented adequate models, but also in need of adjustment. The application of other processing techniques and regression methods can yield more significant results.