Predicting clay content on field-moist intact tropical soils using a dried, ground VisNIR library with external parameter orthogonalization

- We collected simulated in-situ VisNIR spectra from field-moist, tropical soils.
- We used orthogonal projection to remove effects of water from field-moist spectra.
- Partial least squares models for clay content were calibrated using air-dry soils.
- Projection allowed air-dry models to predict clay content for field-moist spectra.
- Parameterization of the projection is robust to calibration dataset variability.

The effect of variable soil moisture, which is found in natural field conditions, is the single most limiting aspect that limits proximal implementation of VisNIR spectroscopy for predicting soil properties using dried-ground spectral libraries. Though the external parameter orthogonalization algorithm (EPO) has shown promise in removing the effect of soil moisture on soil spectra of intact-field moist soils without having to know the soil moisture, EPO has not been widely tested and has not been tested on soils with highly weathered mineralogy (oxides and kaolinite). Thus, the objective of this work was to test the effectiveness of EPO on intact field moist soil spectra and a dried-ground spectral library from highly weathered soils located in Brazil and to use this diverse dataset to assess the sensitivity of the EPO-PLS parameterization and performance to changes in the structure of the calibration spectral dataset. A dried-ground spectral library of 1515 soils collected from Piracicaba and Sao Paulo State, Brazil was transformed using the EPO P-matrix from 80 surface and subsurface soils collected independently of the library and scanned at field-moist intact and at dried-ground condition. Results show that EPO can remove the effect of soil water from field-moist spectra for tropical soils with kaolinitic and ferritic mineralogies. Predicted clay content improved from 320 to 120 g kg− 1 for spectra before and after EPO, respectively. Bootstrapping analysis was performed to assess the sensitivity of the EPO-PLS procedure to changes in the structure of the calibration spectral dataset. All EPO-PLS parameterizations were constrained to a small set of values and small changes to EPO-PLS parameterization had little observed effect on model performance. Large spectral libraries, those developed at the national or continental level, will contain soils of varying mineralogy. While research has shown that EPO is effective on smectitic soils as well as on kaolinitic soils, it is still unclear to what extent mineralogy controls EPO effectiveness.