Recent advances in multivariate analysis coupled with chemical analysis for soil surveys: a review

Purpose Soil is a complex open system covering various physical and chemical attributes. In soil testing, multivariate analysis (MVA) has an important application because it allows the interpretation of a large amount of data for the design of relevant environmental scenarios. The purpose of this research is to summarize recent applications of MVA for identifying soil types or characteristics and for predicting soil attributes with a critical evaluation. Methods Based on a comprehensive search of the available database, in this review, we have provided updated information on the most representative classification and regression MVA applied in the past decade in soil surveys. Regression MVA were compared in terms of applicability, efficiency, and predictive power of different soil attributes. Results Principal component analysis (PCA) allows the grouping of soils into independent clusters according to their differences in texture or physicochemical composition, which may mirror local or regional environmental signatures. PCA is also used to reduce the dimensionality of spectral data before their application in regression MVA. Partial least square regression (PLSR) is the most commonly applied regression MVA for predicting soil attributes after the correlation of spectra (e.g., Vis–NIR) vs. conventional analysis results. The resulting PLSR models, evaluated by correlation coefficient ( R 2 ) and root mean squared error (RMSE), can be valid for the estimation of several soil attributes (e.g., organic carbon, clay). Conclusions Application of regression MVA may have limitations in predicting some soil attributes. Objective interpretation of the dynamic nature of soils requires the selection of representative samples as well as appropriate MVA, which can have significant potential in an effective soil survey.