Improvement of Soil Property Mapping in Northern Ontario's Great Clay Belt Using Multi-Source Remotely Sensed Data

The prediction for the soil properties of texture, calcareous substrate reaction to acid, and ELC (Ecological Land Classification) moisture regime from environmental covariates derived from multi-source remotely sensed data was conducted for study areas located in the District of Cochrane in Ontario (49 50 N, 81 84 W). Random forest (RF) and support vector machine (SVM) approaches were applied to model soil property classifications for 3 adjacent regions: Hearst, Gordon Cosens Forest (GCF) and Abitibi River Forest (ARF). LiDAR (light detection and ranging) data was exploited to derive the detailed vegetation properties of canopy height model (CHM) and gap fraction, as well as a digital elevation model (DEM) for generating topographic covariates. The results indicate that vegetation covariates, particularly LiDAR-derived vegetation properties, were important in the prediction of the soil properties of interest. The most accurate models had accuracy scores greater than 0.7 and Cohens kappa greater than 0.5.