AIC formulae in soil spectroscopy

Partial least squares regression (PLSR) is widely used in soil spectroscopy; however, the methodology used for selection of the optimal number of components remains inconsistent. In this study, we review and compare commonly used formulations of the Akaike Information Criterion (AIC) for selecting PLSR components. Three different AIC formulations reported in the literature were evaluated. To this end, we used visible–near-infrared (vis–NIR) spectra to predict soil organic carbon (SOC; n = 310) and soil texture ( n = 50) data. Despite substantial mathematical differences among the AIC formulations and resulting differences in selected component numbers, model performances—assessed using root mean square error (RMSE), ratio of performance to interquartile distance (RPIQ), and Nash-Sutcliffe model efficiency (NSE) for cross-validation and external validation—were largely similar. Our findings highlight the need for greater awareness regarding AIC formulation and interpretation in soil spectroscopy applications.