Feasibility assessment on use of proximal geophysical sensors to support precision management

A study was conducted at three sites in North Dakota to strengthen understanding of the usefulness of different proximal geophysical data types in agricultural contexts of varying pedology. This study hypothesizes that electromagnetic induction (EMI), gamma‐ray sensor (GRS), cosmic‐ray neutron sensor (CRNS), and elevation data layers are all useful in multiple linear regression (MLR) predictions of soil properties that meet expert criteria at three agricultural sites. In addition to geophysical data collection with vehicle‐mounted sensors, 15 soil samples were collected at each site and analyzed for nine soil properties of interest. A set of model training data was compiled by pairing the sampled soil property measurements with the nearest geophysical data. Eleven models passed expert‐defined uncertainty criteria at Site 1, 16 passed at Site 2, and 14 passed at Site 3. Electrical conductivity (EC), organic matter (OM), available water holding capacity, silt, and clay were predicted at Site 1 with an R‐squared of prediction (Rpred2)$(R_{pred}^2)$ > .50 and acceptable root mean square error of prediction (RMSEP). Bulk density (BD), OM, available water capacity, silt, and clay were predicted with Rpred2$R_{pred}^2$ > .50 and acceptable RMSEP at Site 2. At Site 3, no soil properties were predicted with acceptable RMSEP and an Rpred2$R_{pred}^2$ > .50. These results confirm feasibility of our method, and the authors recommend the prioritization of EMI data collection if geophysical data collection is limited to a single mapping effort and calibration soil samples are few. Core Ideas Framework given to assess relationships between proximal sensing and soil properties. Correlations between sensing data and soil properties varied among three study sites. Electromagnetic induction was most consistently useful in soil property prediction.