Theses and Dissertations Collection

Remote sensing has many applications and benefits in the field of earth science. Hyperspectral sensing can be used in remote sensing to collect visible (350-750 nm) and near-infrared (750-2500 nm) wavelengths, and can produce environmental data more efficiently than traditional laboratory methods. In this thesis, the ability of hyperspectral sensing to predict elements of interest in soil and wildfire ash were tested. Spectral data from forest soils and forest fire ash were collected using an Analytical Spectral Devices (ASD) FieldSpec 4 hyperspectrometer. Ex-situ reflectance of 282 soil samples from the Lake Tahoe basin (California, USA) were collected and the soil was evaluated for Total P, Bray-1 P, water extractable P, and Mehlich-3 P using conventional laboratory methods. Prediction models were built using Random Forest (RF) and Partial Least Squares Regression (PLSR). PLSR models had the highest R2 of validation values for Bray-1 P (0.800) and Total P (0.796), while RF observed the highest prediction accuracy for Mehlich-3 P (0.739) and water extractable P (0.712). For PLSR models, the R2 of calibration values were often lower than those of validation, which suggests potential overfitting in the models. RF did not seem to experience overfitting.Forty wildfire ash samples collected across the western United States were analyzed for reflectance and Cd, Cu, Pb, and Zn content. The sample size was prohibitively small for PLSR modeling, which produced no R2 of calibration values above zero. The R2 of validation for PLSR prediction of Cu was observed as high as 0.760, but that may be inflated due to model overfitting. RF models were much more consistent and yielded R2 of validation up to Cu = 0.771, Zn = 0.661, Cd 0.651, and Pb = 0.357. The model results support the use of vis-NIR hyperspectral systems to measure environmental data in controlled laboratory settings. Future work should be done with larger sample sizes. Additionally, prediction models could be used in the future with field-based remote sensing systems to assess forest watershed P loading potential and the risk of heavy metal contamination in wildfire ash.