2D versus 3D predictive soil mapping

The raster layers in this dataset are the model predictions of various soil properties being created as prototype maps for more quantitative natural resource management. They are the model outputs for the paper currently in preparation for the journal Geoderma entiteled: "Cautionary notes on uncertainty and accuracy patterns for predictive soil property mapping by depth".

The files are named in groups of 2-4 with the .tif file being the base raster grid and the other fields being supporting files used by various GIS software. The naming convention starts with the soil property abbreviation as follows and association with 2D or 3D approaches as follows:

SOC_00cm_logK = 2D % wt soil organic carbon

oc_0_cm = 3D % wt soil organic carbon 

sand_f_vf_psa_00cm = 2D % fine + very fine sands

sand_f_vf_psa_0_cm = 3D % fine + very fine sands

ph_h2o_0_cm = 3D 1:1 soil pH in water

ph_h2o_2D_00cm = 2D 1:1 soil pH in water

EC_2D_00cm = 2D 1:2 water soil electrical conductivity

ec_12pre_0_cm = 3D 1:2 water soil electrical conductivity

The next abbreviation is QRF indicating that the model was a quantile regression forest. The end of the filename can have several different formats:

_QRF: This is the soil property prediction raster.

_QRF_95PI_h: This is the upper 95% Prediction Interval raster.

_QRF_95PI_l: This is the lower 95% Prediction Interval raster.

QRF_95PI_relwidth: This is the 95% relative prediction interval raster that normalizes the width of the prediction interval to the original observation set used to train the QRF model.

_bt after any of the file names simply means that the layer was back-transformed from a layer in transformed units (e.g. log).