Soil Prediction using High Density Data for Understanding Field Variability and Crop Management

Authors: Md Saifuzzaman*, Bioresource Engineering Department, McGill University, Canada, Viacheslav Adamchuk, Bioresource Engineering Department, McGill University, Canada, Asim Biswas, School of Environmental Sciences, University of Guelph, Canada, Pierre Dutilleul, Plant Science Department, McGill University, Canada
Topics: Spatial Analysis & Modeling, Soils, Canada
Keywords: Proximal soil sensing, topography, prediction method, soil properties, field variability
Session Type: Paper
Day: 4/6/2019
Start / End Time: 8:00 AM / 9:40 AM
Room: 8217, Park Tower Suites, Marriott, Lobby Level
Presentation File: No File Uploaded


The availability of high-density multivariate sensor data, along with recent advances in analysis methods, have served to: (i) predict soil nutrients in agricultural fields, (ii) optimize model predictions and (iii) determine their spatial variability. Proximal soil sensor (PSS) measurements and their ability to support the prediction of soil attributes of eleven agricultural fields in Ontario, Canada, were investigated. High-accuracy topography and apparent soil electrical conductivity (ECa), mapped using DUALEM-21S and RTK GNSS sensors, respectively, served in the characterization of field-scale soil variability. Targeted sampling strategies were adopted, and lab analysis of six properties (soil pH, BpH, Soil Organic Matter, Phosphorus, Potassium, and Cation Exchange Capacity) were processed in an effort to understand soil variability across the fields. DUALEM sensor variables were found to be colinear with one another for three fields. Topographic variables, slope, and topographic wetness indices were moderately correlated with the remainder of the sensor variables. Upon calculation and comparison of Pearson’s correlation coefficients (r) between sensor variables and field-measured soil properties, the topographic parameters and ECa (PRP1: 0 – 0.5 m) sensor variable were correlated (r≥60) and effectively predicted SOM, P, and CEC at different locations. Ordinary least square served to generate a best-fit line. Model error (Standard Error of estimate and Standard Deviation) along with the coefficient of determination (adjusted R2) were evaluated as tools for assessing the accuracy of prediction of targeted soil properties. Among the 11 agricultural fields, the VN and R50 fields represented the best-structured data resulting in the least prediction error.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login