Estimation of wet aggregation indices using soil properties and diffuse reflectance near infrared spectroscopy: An application of classification and regression tree analysis

Soil aggregation is critical for assessing soil health; however, conventional aggregation measurement is laborious and expensive. The performance of near infrared diffuse reflectance spectroscopy (NIR) and basic soil properties for estimation of wet aggregation indices was investigated. Two samples sets representing different soils from across Lake Victoria Basin in Kenya were used for the study. A model calibration set (n = 136) was obtained following a conditioned Latin hypercube sampling, and validation set (n = 120) using a spatially stratified random sampling strategy. Spectral measurements were obtained for air-dried (<2 mm) soil using a Fourier-transform NIR spectrometer. Soil laboratory reference data were also obtained for wet aggregation indices (WSA): macro, micro and unstable fractions using two different wet-sieving pretreatments. Soil properties were screened as candidate predictors of WSA using Classification and Regression Tree (CART regression) analysis. WSA were calibrated to soil predictors and to smoothed first derivative NIR spectra using partial least squares (PLS) regression. Key soil predictors were: soil organic carbon and pH water (macro), water dispersible clay (WDC) (micro) and exchangeable sodium (unstable). Full cross validation of NIR PLS prediction of stable macro, micro, unstable aggregates, and for WDC gave RPD (ratio of prediction deviation) of 1.4-2.0. Independent testing of NIR PLS gave RPD = 1.4 for macro and RPD = 1.2-1.0 for unstable and soil predictors. NIR could estimate macro and unstable fractions with moderate reliability, and; NIR was superior over soil properties for stability pedotransfer purposes. Further efforts should widely test performance for a wider range of soil types and calibration strategies for improved geographic transferability of models.