Calibration and validation statistics of partial least-squares regression models.

Calibrations were evaluated as follows (Saeys et al (2005): excellent (R2/r2 ≥ 0.9, RPD ≥ 3.0), reliable quantitative predictions (R2/r2 ≥ 0.75 and <0.9, RPD ≥ 2.0 and <3.0), differentiation between high and low values (R2/r2 ≥ 0.65 and <0.75, RPD ≥ 1.5 and <2.0, unsuccessful (R2/r2 <0.65, RPD <1.5). Transf: transformations for regression analyses. Abs: Log1/R (R = reflectance). 1D: first derivative. R2: coefficient of multiple determination (calibration). SEC: standard error of calibration. SECV: standard error of cross validation. r2: regression coefficient NIRS predicted vs. observed values. SEP: standard error of prediction (validation). RPD: ratio of SD of reference values (validation) to SEP. RPIQ: ratio of the interquartile distance IQ ( = Q3–Q1) of reference values (validation) to SEP. Corr. OM: Pearson correlation with organic matter content.