Managing Global Resources for a Secure Future

2017 Annual Meeting | Oct. 22-25 | Tampa, FL

401-4 Measuring Multiple Soil Properties Simultaneously Using Proximal Soil Sensor Data Fusion.

See more from this Division: SSSA Division: Soil Physics and Hydrology
See more from this Session: Proximal and Remote Sensing Techniques in Soil Physics and Hydrology

Wednesday, October 25, 2017: 2:20 PM
Marriott Tampa Waterside, Grand Ballroom I and J

Wenjun Ji, Bioresource Engineering, McGill university, Ste-anne-de-Bellevue, QC, Canada, Asim Biswas, 50 Stone Road East, University of Guelph, Guelph, ON, CANADA and Viacheslav Adamchuk, Department of Bioresource Engineering, McGill University, Montreal, QC, Canada
Abstract:
Sensor data fusion, a multifaceted process integrating automatically detected and geospatially correlated data or information from multiple proximal soil sensors (sources), may provide a realistic alternative to characterize the complexity of soils in a more comprehensive and robust but less risky way in precision agriculture applications. In this study, we integrated or fused the data from a gamma-ray sensor, an apparent electrical conductivity (ECa) sensor, a commercial ruggedized multi-sensor platform carrying a visible and near-infrared (vis-NIR) optical sensor and a soil ECa sensor, and topography data from a real time kinetic GPS sensor. The information was used to predict 8 soil properties including soil organic matter (SOM), pH, buffer pH (BpH), phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), and aluminum (Al). Partial least square regressions (PLSRs) were used to predict soil properties from each individual sensor and different sensor combinations (sensor data fusion). By fusing the data from all the proximal soil sensors, SOM, pH, BpH, Ca, Mg, and Al were predicted simultaneously (R2>0.5, and RPD>1.50). Improved predictions were observed for most soil properties based on sensor data fusion than those based on individual sensors. After choosing the optimal sensor combination for each soil property, the predictive capability was compared using different data mining algorithms, including support vector machines (SVM), random forest (RF), multivariate adaptive regression splines (MARS), and regression trees (CART). Improved and robust predictions for SOM, Ca, Mg, and Al were observed using SVM over PLSR. The predictive capability was followed by RF and MARS, with CART offering the worst result. Robust predictions of pH and BpH were only obtained by MARS and PLSR, respectively. Thus, we conclude that proximal soil sensor data fusion paired with data mining algorithms is a promising way to offer the essential soil information needed for precision agriculture applications.

See more from this Division: SSSA Division: Soil Physics and Hydrology
See more from this Session: Proximal and Remote Sensing Techniques in Soil Physics and Hydrology