Spatially locating soil classes within complex soil polygons – Mapping soil capability for agriculture in Saskatchewan Canada

This paper proposes a simplified approach to mapping soil capability, as defined by the Canada Land Inventory (CLI), based on the hypothesis that the primary determinants of soil capability may be surrogated by Normalized Difference Vegetation Index (NDVI) derived from Earth Observation (EO) data integrated with other biophysical information. A case study in which a Decision Tree classification method with a boosting algorithm was used in spatially locating individual soil capability classes as estimated in the complex symbol of the CLI database was conducted in Saskatchewan Canada. The input metrics used for the classification include the first four principal components of the original NDVI images, phenological parameters, topographic factors, land cover and spatial dependence images. Validation showed high Kappa coefficients for the mapped soil capability classes within homogeneous soil polygons and high R-squares between the mapped soil area and CLI-estimated area within heterogeneous polygons. Results confirm the hypothesis that integrating parameters derived from the Moderate Resolution Imaging Spectro-radiometer (MODIS) 250 m time-series Normalized Difference Vegetation Index (NDVI) with ancillary data may serve as a comprehensive tool for classification of soil capability.

► The use of complex soil classes is problematic in environmental modeling.
► We used MODIS satellite imagery to identify soil capability classes in Saskatchewan Canada.
► Inputs also included phenological and topographic factors, land cover and spatial dependence.
► Validation showed high Kappa coefficients and high R-square values.
► Results confirm that integrating imagery with ancillary data can be used to map soil capability.