Author: Ryan Matthew Will

Publisher:

Published: 2017

Total Pages: 119

ISBN-13:

"Quantifying soil organic carbon (SOC) in complex terrain is challenging due to its high spatial variability. Generally, limited discrete observations of SOC data are used to develop spatially distributed maps of SOC by developing quantitative relationships between SOC and available spatially distributed variables. In many ecosystems, remotely sensed information on aboveground vegetation can be used to predict belowground carbon stocks. In this research, we developed maps of SOC across a semi-arid watershed based on discrete field observations and modeling using a suite of variables inclusive of hyperspectral and lidar datasets; these observations provide insights into the controls on soil carbon in this environment. The Reynolds Creek Experimental Watershed (RCEW), in SW Idaho, has a strong elevation gradient that controls precipitation and vegetation. Soil samples were collected to 30 cm depth using a nested sampling approach, across the watershed (samples, 279 data points, in 28 plots, discretized with depth, total n=1344) and analyzed for SOC content. Point SOC data was combined with a suite of predictor variables from traditional, lidar and hyperspectral datasets to calibrate Random Forest and Stepwise Multiple Linear Regression models that predict SOC distribution across RCEW. In this study, SOC generally increased along the precipitation-elevation gradient corresponding with an increase in the diversity and abundance of vegetation. We found that variable soil bulk densities and areas of high rock content strongly influenced mass/unit area SOC values. Interestingly, rock content was also negatively correlated with percent SOC. Local variability of SOC in this study was high with the variability at the plot scale about 1/3 of that observed at the watershed scale. Our research suggests that vegetation indices calculated from spectral data are the best predictors of SOC storage in this system. Roughly 60% of the variance in SOC data is explained using Normalized Difference Vegetation Index while two hyperspectral vegetation indices, Modified Red Edge Simple Ratio and Modified Red Edge Normalized Difference Vegetation Index explain over 70%. The addition of Lidar variables modestly improved SOC prediction, explaining 75% of variability in SOC."--Boise State University ScholarWorks.