Geospatial Data for Modeling Soil Carbon Stocks across Pacific Northwest Watersheds

Rainier), and the Colville Watershed (CVW) located in northeastern Washington. The Wetland Intrinsic Potential (WIP) tool was implemented for each study watershed to model the gridded land surface as a continuous probability of wetland presence, with each grid pixel containing a value from 0-100%. Geospatial datasets related to vegetation, climate, lithology and geology, and topography were gathered to determine predictors for SOC stocks.

Google Earth Engine was used to obtain satellite imagery for calculation of vegetation spectral indices from the five year median of Sentinel-2 reflectance. Two model types were used in the research to model SOC stock and SOC percent: a linear mixed effects model (LMM) and a quantile random forest (RFM). The LMM was used to test specific hypotheses about important predictors and examine predictor coefficients.

The RFM was used to accommodate potential non-linear relationships in the data as well as incorporate a more flexible approach to predictor selection. Mapped predictions of SOC stock across all four study watersheds were generated using the model with the best fit to the test dataset. Additional geospatial masks were used to remove surface water areas and urban zones.

The data are provided in comma separated values (CSV), GeoTIFF, and GeoPackage formats.