Geostatistical Methods for Predicting Soil Moisture Continuously in a Subalpine Basin

Williams, Katherine

This soil moisture spatial survey (SMSS) study demonstrates the power of geostatistical methods for quantifying physical variables spatially. The goals of this study were to explore and compare geostatistical predictive models and interpolation methods and to produce a continuous soil moisture surface throughout the sub-alpine forested basin of a watershed in Rocky Mountain National Park, Colorado. Through a combination of regression, prediction, variable derivation, and interpolation, a high resolution, continuous surface was produced. The final Geographically Weighted Regression (GWR) model used to predict volumetric water content based on derived Slope, Vegetation Type, and Percent Cover had an Adjusted R2 value of 0.71 and showed no significant spatial autocorrelation of the residuals. The Vegetation Type and Percent Cover surface were classified and geoprocessed from IKONOS-2 Near Infrared-Visible Imagery and 1 meter LiDAR digital elevation model data. Both field measurements and predicted measurements from the GWR model were used to create interpolation surfaces using Ordinary Kriging (OK) and Inverse Distance Weighting (IDW) methods. The IDW surface outperformed the OK surface by showing lower error and modeling the desired microtopographical features in greater detail. Using field measurements only, the soil moisture surface captured the average soil moisture trend along transects in an existing Hillslope Study, but did not portray the variability along the slope. Using the GWR model to predict soil moisture at points in a 5 meter grid improved the texture of the surface and captured more of the local variability on the slope. This study demonstrates a unique combination of data and methods for quantifying the spatial distribution of an environmental variable.