Continuous-Soil-Properties project
Prediction of continuous soil properties at 30m resolution for the Continental USA
Properties to be predicted
Property |
Units |
Notes |
Clay |
g/kg or % |
|
Sand |
g/kg or % |
|
Silt |
g/kg or % |
|
Depth |
cm |
|
pH |
unitless |
|
rock fragments |
(m3 m-3) |
|
Available water capacity |
(mm mm-1 ?) |
probably need to run Rosetta or Saxton and Rawls |
Potential properties to be predicted
Property |
Units |
Notes |
ECEC |
(cmolc/kg) |
|
N |
(%) |
|
C |
(%) |
|
electrical conductivity |
(dS/m) |
|
Semi-detailed Structure of Project.
A. Pedon dataset cleaning and prep [per property?]
A.1 Forest Service NRM data [Colby]
(0-5, 5-15, 15-30, 30-60, 60-100, 100-200 cm) [per property]
C.1 Extract DEM derivatives
C.2 Extract spectral derivatives
C.3 Extract other geospatial covariates
D. Build Models [per property]
D.1 Build list to hold each regional model per property [big point of discussion here]
D.2 Set up training routine (cross validation repeated 10 times or something).
D.3 Training/evaluation data split
D.4 Build continental model
D.5 Build regional models
E.1 Evalute continental model
E.2 Evaulate each regional model
F. Spatial Predictions [how to use regional models?]
H. Locations where the addition of legacy or new sample data would prove most beneficial to model improvement and/or uncertainty reduction.