PLS Regression Coupled with FTIR Analysis as a Fast Tool to Assess Properties and Nutrient Pools of Biochar-Based Fertilizers

Abstract

ABSTRACT Biochar-based fertilizers (BBFs) agronomic performance depends on its nutrient pools, electrical conductivity (EC) and pH, conventionally assessed by laboratory classical wet methods that are non-environmentally friendly, time-consuming, and costly. However, the nutrient pools and properties of BBFs can be predicted by partial least square regression (PLS) based on Fourier Transform Infrared Spectroscopy (FTIR) analysis. Thus, we aimed to use PLS regression based on FTIR analysis to predict the pH, EC and nitrogen (N), phosphorus (P), and potassium (K) pools in BBFs produced by different routes. Sixteen BBFs were produced by mixing, previously or after the pyrolysis process of low-grade phosphate acidulated with different inorganic acids (sulfuric, hydrochloric and nitric acids), addition or no different sources of Mg (magnesium chloride and serpentinite). PLS models based on the normalized FTIR dataset in BBFs are considered good when R2 during calibration and cross-validation phases are higher than 0.81, values founded in our study to prediction of pH, EC, total, and mineral N, total P, and P soluble in water, citric acid, formic acid, and neutral ammonium citrate, and total K of BBFs. Besides, the difference between R2 of calibration and cross-validation lower than 0.2 founded indicated no overfitting. Furthermore, the PLS models had high prediction capacity (R2 of prediction higher than 0.55) and high accuracy (R2 to y-randomization <0.50). Thus, PLS regression models based on the normalized FTIR dataset in BBFs were models characterized by the good performance, robustness, and nonrandom able to predict the nutrient pools, pH, and EC in BBFs.