Quantifying catchment scale regolith physicochemical variation has the potential to yield important insights to the evolution and function of the subsurface critical zone. The objective of this study was to quantify and predict soil properties within a forested catchment in southern AZ, USA. The catchment, Marshall Gulch (MG), is part of the Jemez River Basin-Santa Catalina Mountains Critical Zone Observatory (CZO). MG is 5-hectare, mixed-conifer forested catchment situated on granitic parent material, with a mean elevation of 2400m, mean annual temperature of 10°C and mean annual precipitation of 75 cm. The study was designed using a combination of principle component analysis (PCA) of available geospatial data and a conditioned Latin Hypercube Sampling (cLHS) scheme. Geospatial data determined by the PCA to account for 95% of landscape variance included modeled soil depth, slope, soil wetness index, normalized difference vegetation index (NDVI), and National Agriculture Imagery Program (NAIP) bands 3/2. The cLHS scheme was performed using these geospatial data and determined 20 optimal sample locations for collecting soil material down dug to the depth of refusal, and sampled according to genetic horizon. Soils were characterized using methods of X-ray Fluorescence (XRF), X-ray Diffraction (XRD), particle size, color, pH, electrical conductivity (EC), C/N isotopes, and loss on ignition (LOI). Regression models predicting soil depth (cm), carbon (kg m-2), clay (%), Na flux (kg m-2), pH, and volumetric strain were developed using the 20 sample locations, the six geospatial data layers determined from the PCA, and reverse step-wise multiple linear regression. Results indicated strong correlations of soil properties with the drainage systems in the MG catchment. Deeper soils, higher clay content, higher carbon content, and greater Na loss were observed and modeled within the drainages of the catchment relative to adjacent slopes and ridgelines. Regression R2 values ranged from 0.35 (soil depth, cm) to 0.50 (carbon, kg m-2) and all models exhibited p-values <0.05. Regression model residuals were extrapolated spatially using an optimized Inverse Distance Weighting (IDWopt), followed by cross validation and determination of spatial autocorrelation using Moran’s I. Cross validation mean errors approached zero and residual errors were randomly distributed throughout the catchment as determined by Moran’s I, indicating unbiased models with no spatial autocorrelation in predicted residuals. Spatial maps of prediction residuals were added to soil properties predicted with the regression models to yield final interpolated maps of profile data at the scale of the entire catchment. Overall we found soil properties varied predictably with topography and landscape position throughout the catchment, yielding strong correlations to the convergent drainages within the study site. These data provide a clear basis for modeling water, carbon, and sediment transport within the catchment, in addition to laying the foundation for quantifying catchment scale soil-water and carbon storage capacities.
Rasmussen C., Holleran M. (2013): Using statistical methods to quantify and predict catchment scale soil variability. Abstract presented at AGU Chapman Conference on Soil-mediated Drivers of Coupled Biogeochemical and Hydrological Processes Across Scales, Biosphere 2, Tucson, Arizona, 21-24 October 2013 (Talk).