Article ID Journal Published Year Pages File Type
4572865 Geoderma 2016 11 Pages PDF
Abstract

•Soil thickness successfully mapped, with increased accuracy for the 0 to 5 m range.•Generalized Linear Model Residual Kriging was best method for shallow soils.•Novel data inputs (exposed bedrock, well data) increased accuracy for modeled results.•Exposed bedrock data helped constrain shallow soil predictions.•Slope and convexity not found to be important factors for predicting soil thickness.

The Critical Zone (CZ) is defined as the outer layer of the solid Earth, extending from the vegetation canopy through the pedosphere and down to the bottom of the weathered bedrock zone. Most biological, chemical and physical interactions take place in the CZ, and it is here that most terrestrial life is found. Being able to predict the thickness of the pedosphere (i.e. soil thickness, from the mineral soil surface down to bedrock) in the CZ can lead to a better understanding of rates of physical/chemical change, such as carbon sequestration, soil erosion, and water storage. The objective of this study was to accurately map the thickness of the pedosphere on a landscape scale for a ca. 3400 km2 landscape in Southern British Columbia (BC). The data inputs used were exposed bedrock (EB) points identified from orthophotos, well record (WR) data, and in-situ observations of soil thickness, for which conditioned Latin hypercube sampling (cLHS) was used to define a set of locations where soil thickness was determined. Four methods were then used to model soil thickness as a function of environmental data layers derived from a digital elevation model and satellite imagery: Generalized Linear Model (GLM), Random Forest (RF), GLM Residual Kriging (GLMRK) and RF Residual Kriging (RFRK). An equal weighted random sampling scheme of 100 EB, WR, and in-situ soil thickness points was used with each model. A second sampling scheme was used with the same WR and in-situ soil thickness points and an additional 5000 randomly sampled EB points, used to improve prediction accuracy with limited data sources. Of the modelling methods used, GLMRK proved to be the best method for shallow thicknesses (0 to 2 m) as assessed by Root Mean Square Error (RMSE) values (1.87 m) with the equal weighted sample scheme. The addition of the 5000 EB points substantially improved predictions for shallow soil thickness (RMSE 0.9 m) as well as soil thickness in the 2–5 m range, while having negligible impact on predictions of thicker soils. This demonstrates that an exposed bedrock layer can help constrain shallow soil thickness predictions when used in conjunction with geostatistical approaches such as GLMRK and RFRK for mapping soil thickness.

Related Topics
Physical Sciences and Engineering Earth and Planetary Sciences Earth-Surface Processes
Authors
, , , ,