Multi-Sensor Soil Probe and Machine Learning Modeling for Predicting Soil Properties

Area of Science:

• Agricultural Engineering
• Soil Science
• Digital Agriculture

Background:

• Traditional soil sampling is labor-intensive, costly, and provides limited spatial resolution.
• Developing accurate digital soil maps requires integrating diverse data sources with varying resolutions.
• Precision agriculture demands high-resolution soil data for optimized crop management.

Purpose of the Study:

• To develop and evaluate a data-driven, in situ proximal multi-sensor approach for digital soil mapping.
• To investigate the impact of scale mismatch between high-resolution sensor data and coarser laboratory measurements.
• To compare four data integration methods for soil prediction modeling in almond orchards.

Main Methods:

• Engineered a novel Digital Soil Core (DSC) Probe with seven distinct sensors (friction, force, dielectric, resistivity, imagery, acoustics, spectroscopy).
• Collected in situ DSC sensor data and spatially co-located soil cores for laboratory analysis.
• Applied Partial Least Squares Regression (PLSR) to compare four data integration methods (A, B, C, D) for soil prediction models.

Main Results:

• Method C, allocating laboratory data to all sensor data within a layer, outperformed Methods A and B.
• Method D, directly linking high-density sensor data to crop responses, bypassed laboratory costs and proved most cost-effective.
• The DSC System provides mm-scale vertical soil data up to 120 cm depth in approximately 60 seconds.

Conclusions:

• The DSC System enables high-density, in situ soil characterization for creating agricultural digital twins.
• Directly linking multi-sensor soil data to crop performance (Method D) offers a cost-effective solution for precision agriculture.
• This approach facilitates industrial-scale precision agriculture by providing actionable soil-crop relationship insights.