Predicting Soil Components and Value-Added Soil Information in Canyonlands National Park, Utah
What – We modeled Biological soil crust (BSC) class (classes 0-6), map unit component (see legend), ecological site, soil series, soil depth and surface texture class for Canyonlands National Park.
Why – BSC in arid ecosystems influence infiltration and erosion rates and are important for soil resource management. Hopefully these maps will be useful for Park Service Planning and be an additional value added product derived from NRCS mapping efforts.
We used Random Forests with 1000 trees and the default number of random variables.
Pedon information used as training data was obtained from the NRCS and cleaned (checked for correct spelling, etc.).
Biological soil crust classes were rounded to the nearest integer.
Distance between pedons was calculated and Pedons <90 m apart were manually selected and removed from the training dataset.
Several concerns about this dataset exist: many pedons are photo interpreted and many do not have complete data entered.
NASIS data was obtained from the National Park Service but it came in a format for AnalysisPedon which was prohibitively difficult to use.
OOB errors for each predicted attribute are listed in the PDF Out-of-bag_Errors. While overall OOB error is informative it is often more useful to look at the individual class OOB errors. OOB error is highly dependent upon class size.
Maps of BSC class and associated probability are in Biol_soil_Crust_maps.pdf. Other predicted soil attributes are in OtherMaps.pdf. Variable importances for each prediction are in Variable Importance.pdf.
* Please contact Colby Brungard (email@example.com) if you are interested in viewing the pdf files.
Assessing Soil Variability, Wind Erosion Potential, and Relict Wetlands in Snake Valley, Utah
|Snake Valley, Utah|
Spatial Prediction of Soil Carbon Using WorldView-2 Imagery and Elevation Data for a Desert Wetland Ecosystem, Utah