Penetrometer-Mounted VisNIR Spectroscopy: Implementation and Algorithm Development for In-Situ Soil Property Predictions

Abstract

Many applications in agriculture and environmental sciences rely on high-quality spatially explicit soils data. Due to the costs of sample collection, preparation, and laboratory analysis, traditional techniques for collection of new soils data are often expensive. In this study, we developed the framework for a novel soil measurement technique; a penetrometer-mounted visible near infrared (VisNIR) spectrometer. The penetrometer-mounted VisNIR probe is capable of measuring soil properties of in situ soils at high-depth resolutions (i.e. 5-cm vertical spacing). A fully functional in situ VisNIR probe could reduce the cost of soil measurement by supplementing or replacing traditional soil measurement techniques. For in situ VisNIR to be a viable tool, in situ VisNIR needs to be compatible with existing spectral modeling techniques designed for spectra collected from air-dried and ground soils in the laboratory. One issue with in situ VisNIR is that, unlike spectra collected under laboratory conditions, in situ spectra are altered by in situ effects (e.g soil moisture, structure, field temperatures, etc.) and therefore are incompatible with existing laboratory approaches. Using soils from central Texas, we tested two methods for mitigating in situ effects; direct standardization (DS) and external parameter orthogonalization (EPO). Our tests indicate that EPO was more effective than DS. We further tested EPO on tropical soils from Brazil. The EPO performed well on these soils demonstrating that EPO can be applied to a wide variety of soil types. Finally, we tested the EPO on in situ spectra collected using the penetrometer-mounted VisNIR probe and again, the EPO performed satisfactorily. By iii coupling the EPO with a penetrometer-mounted VisNIR probe we have demonstrated the viability of in stiu VisNIR. The penetrometer-mounted system can utilize existing laboratory-based spectral modeling tools for prediction of soil properties at high-depth-resolutions and is a viable tool for rapid, cost effective soil measurement.