Examining the Relationships Between Chemical Properties and VNIR Reflectance Properties of Kentucky Soils using In-situ Data

Abstract:

The main focus of this study is to investigate and develop relationships between soil chemical properties and their spectral properties. In-situ hyperspectral data of 19 sites representing the major terrestrial ecosystem types in Kentucky terrestrial ecosystem were used to investigate the relationship between soil spectral Visible/Near-Infrared (VNIR) reflectance and soil chemical properties. Soil samples were obtained from 0 – 60 cm depth with sampling every 5 cm. VNIR reflectance of the soil samples were measured for each of the samples along with the corresponding 15 soil chemical properties including soil pH, total nitrogen, total carbon (C), total extractable Fe, Mg, Ca, and K. A strong relationship (Pearson r > 0.6) between total carbon and soil reflectance between 1700-1800 cm^-1 and 2800-2900 cm^-1 were detected for the top 10 cm soil samples. Whereas, soil extractable Fe showed a negative relationship with soil spectra (800-900 cm^-1 and 3400-3500 cm^-1). Partial least square analysis was used to develop predictive models from a set of random selected soil samples. In addition, classification and regression tree results were developed to identify the spectra that were significant predictors of in-situ soil properties. Results showed that two organic matter features (1551 and 1607 cm^-1) and one clay feature (3676 cm^-1) were the significant predictors of total C for all soil samples. Our results can aid in developing models to map soil properties using remotely sensed data that can be helpful to further our understanding of soil-vegetation interactions.