Improving In-Situ Estimation of Soil Profile Properties Using a Multi-Sensor Probe

Area of Science:

• Soil Science
• Remote Sensing
• Agricultural Engineering

Background:

• Optical diffuse reflectance spectroscopy (DRS) is a laboratory technique for soil property estimation.
• In-situ DRS offers potential for rapid, non-destructive, and cost-effective field measurements.
• Accurate in-situ soil property assessment is crucial for precision agriculture and environmental monitoring.

Purpose of the Study:

• To evaluate the in-situ estimation of various soil physical and chemical properties using visible and near-infrared (VNIR) spectra.
• To assess the added value of combining VNIR spectra with depth, apparent electrical conductivity (ECa), and cone index (CI) data (DECS).
• To identify optimal preprocessing and modeling techniques for accurate in-situ soil property estimation.

Main Methods:

• A Veris P4000 instrument collected VNIR spectra, ECa, CI, and depth data to 1 m depth in two Missouri fields.
• Soil core samples were analyzed in the laboratory for soil organic carbon (SOC), total nitrogen (TN), moisture, texture, cation exchange capacity (CEC), Ca, Mg, K, and pH.
• Visible and near-infrared (VNIR) spectra were analyzed alone and in combination with DECS using partial least squares regression (PLSR), neural networks, regression trees, and random forests.
• Gaussian smoothing filter preprocessing was evaluated alongside various calibration methods.

Main Results:

• The combination of Gaussian smoothing filter preprocessing and PLSR yielded the best model performance for most soil properties.
• The DECS approach improved the estimation of silt, sand, CEC, Ca, and Mg compared to VNIR spectra alone, with >5% improvement for Ca.
• Field-specific differences in estimation accuracy were observed, indicating potential site-specific calibration needs.

Conclusions:

• In-situ estimation of profile soil properties is feasible using a multi-sensor approach combining VNIR spectroscopy with ECa, CI, and depth data.
• Partial least squares regression (PLSR) with Gaussian smoothing preprocessing is recommended for optimal results.
• Further research may be needed to address field-specific variations in estimation accuracy for robust in-situ soil monitoring.