Soil reflectance spectroscopy is a rapid and relatively cheap alternative to standard laboratory analysis that enables the simultaneous estimation of multiple properties from a single measurement. It is, therefore, highly suitable for the high-throughput analysis of large soil datasets. 

A reflectance spectrum is a signal that contains information about multiple soil properties, such as soil organic carbon, soil texture, moisture, and mineral composition. However, most existing models treat soil spectroscopy as a black-box regression problem, learning a direct mapping from spectra to soil properties. Because these models lack a physical understanding of how each soil property affects the measured signal (reflectance spectrum), they often generalize poorly across different instruments, moisture levels, and field conditions.

To address this, this project aims to move beyond purely statistical mapping by isolating the spectral signatures of specific soil properties. The student will design and conduct laboratory experiments using artificially mixed samples to study the effect of each soil property on the soil spectra (e.g., varying clay or organic matter content while keeping other factors constant). The results of this study will enable us to understand how much each soil property contributes to soil spectra, identify the most important regions of the soil spectrum for each property, and eventually pave the way toward developing models that generalize better than existing black-box models.