From Spectra to Mineralogy: a Remote Sensing Approach to Earth's Arid Dust Source Regions

Author: Keebler, Abigail May

Year: 2026

Degree: Dissertation (Ph.D.)

Advisor: Ehlmann, Bethany L.

Committee Members: de Kleer, Katherine R.; Ehlmann, Bethany L.; Frankenberg, Christian; Thompson, David R.

Option: Planetary Sciences

Abstract

Mineral dust plays a critical role in Earth’s climate and biogeochemical systems, influencing radiative forcing, cloud microphysics, and nutrient fertilization of terrestrial and marine ecosystems. These impacts are strongly dependent on dust mineralogy, particularly the abundance and speciation of iron-bearing phases, clays, and carbonates, which control both the optical properties and chemical reactivity of dust aerosols. Despite this importance, mineralogical properties of dust source regions remain poorly constrained at regional to global scales, limiting the representation of dust processes in Earth system models. Hyperspectral visible to shortwave infrared (VSWIR) remote sensing offers a promising pathway for addressing this gap by enabling spectrally resolved characterization of surface mineralogy globally. However, translating reflectance spectra into quantitative mineral abundances remains challenging due to the nonlinear nature of VSWIR spectra of intimate mineral mixtures. This dissertation combines hyperspectral remote sensing, field spectroscopy, and laboratory analyses to improve quantitative interpretation of mineral dust source regions and to evaluate the extent to which mineralogical properties can be retrieved from VSWIR observations. The work is structured around three complementary studies that collectively link global satellite observations, field measurements, and empirical modeling approaches. First, a global analysis of arid dust source regions is conducted using hyperspectral observations from the Earth Surface Mineral Dust Source Investigation (EMIT) mission. A systematic sampling and filtering framework is developed to extract a representative dataset of bare soil reflectance spectra from more than one billion observations. The resulting dataset characterizes global variability in surface albedo and mineralogical absorption features across major dust source regions, revealing distinct regional spectral endmembers associated with differences in mineralogy, including bright iron oxide- and kaolinite-rich Saharan surfaces and darker clay- and carbonate-dominated Asian surfaces. These results demonstrate substantial compositional diversity in dust source regions and provide constraints on surface radiative properties relevant to Earth system modeling. Second, we investigate the physical and compositional controls on spectral variability using a novel dataset of co-located in situ VNIR reflectance spectra and laboratory measurements of mineralogy, grain size, and iron speciation. We compare the spectral variability captured in this dataset to that of the EMIT global dust source dataset to evaluate the range of global variability represented. We assess compositional controls on spectral variation, including absorption feature presence, position, and strength, as well as overall reflectance and continuum shape. Clay and carbonate absorption features show systematic but non-linear relationships with mineral presence and abundance, reflecting overlapping absorptions and mixed-phase effects. In contrast, iron oxide abundance exhibits strong, approximately linear relationships with diagnostic absorption features in fine-grained clay-rich sediments. We also identify a distinct population of hematite-bearing sands that display strong absorption features despite low hematite abundance. Radiative transfer modeling shows that grain size and bright mineral matrices significantly modulate iron oxide spectral expression. Overall, these results highlight that composition and sediment physical context in control spectral variability. Third, the dissertation evaluates empirical approaches for predicting quantitative mineral abundances from VSWIR spectra. Using the coupled spectral-mineralogical dataset, a partial least squares regression model is trained to estimate mineral and iron species abundances. Results show that mineral components which control continuum shape and albedo over the full spectral range, including quartz, feldspar, and iron oxides, can be predicted accurately, while phases that express diagnostic feature in a narrow spectral range, such as clays and carbonates, are less-well predicted. Application of the calibrated models to EMIT reflectance data demonstrates that PLSR models successfully identify the major minerals present in the ground-truth data. Compared with EMIT mineralogy products, the empirical approach provides several advantages. In particular, the models improve quantitative retrievals of carbonate abundance, and more frequently resolve complex multi-mineral assemblages containing combinations of clays, carbonates, and evaporite minerals within individual spectra. The models produce reasonable values when applied to spectra from sediment types outside the training dataset, suggesting promising transferability across heterogeneous dust source environments. Together, these findings demonstrate both the potential and current limitations of empirical inversion approaches for hyperspectral mineral retrieval and highlight their utility as a complement to existing feature-based remote sensing frameworks. Together, these results provide a framework for improving quantitative interpretation of hyperspectral observations of Earth’s bare sediment surface. By linking global-scale satellite data with field-based measurements and empirical modeling, this work advances the ability to retrieve physically meaningful mineralogical information from VSWIR remote sensing. These improvements are essential for better constraining the radiative and biogeochemical impacts of mineral dust in Earth system models and for extending hyperspectral approaches to future Earth and planetary remote sensing missions.