Quantitative bedrock characterization using multispectral TIR data: pushing the limits of TIMS and ASTER in a desertic environment

Thermal Infrared (TIR) spectroscopy constitutes a powerful diagnostic tool for compositional analysis of geological materials. Numerous studies have made use of that tool to investigate geologic problems. However quantitative interpretation is complicated by the trade-off that exists in remote sensing data between spectral and spatial resolutions. The pixel size of hyperspectral datasets usually prevents detailed geologic interpretation while instruments with higher spatial resolution offer limited spectral resolution. We collected natural samples of two types of igneous lithologies (basalt and rhyolite) in the Gila Bend region in an attempt to compare TIR spectra at 3 different levels of resolution: a) laboratory spectra taken on individual samples with hyperspectral resolution (~2 wavenumbers); b) 6-point spectra from Thermal Infrared Multispectral Scanner (TIMS) imagery taken at meter-scale spatial resolution; and c) 5-point spectra from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) imagery taken with a 90x90 m²/pixel spatial resolution. Identification of each lithology is straightforward in image datasets, using qualitative processing (such as decorrelation stretches). Quantification of the proportion of caliche however - a relatively simple process on hyperspectral laboratory data - proved to be much more challenging, and yielded contrasting results with changing spatial and spectral resolutions. Moreover we observed a mismatch between remotely sensed spectra and the expected spectra derived from laboratory spectra of natural samples. This can only partially be attributed to atmospheric effects, and other factors need to be considered when comparing the datasets. The laboratory spectrometer offers the most ideal conditions, sampling individual and uniform rock surfaces and minimizing the effects of atmospheric CO₂ and H₂O. However it fails to incorporate the diversity that can be observed in a natural outcrop. TIMS, with a pixel size of ~2m, samples radiated energy from the soil and various types of rock fragments, possibly with different levels of weathering. One level of spatial resolution down, ASTER, with a pixel size of 90 m, likely samples radiated energy from in place outcrops as well as rock fragments. In an effort to accurately recreate the TIR signature of individual pixels in the remote sensing data we adopted a systematic approach, sampling various outcrops within each unit, and sampling all the different types of rock fragments present on each outcrop. We then used laboratory spectra of natural samples degraded to TIMS and ASTER resolutions and combined in various ways, as end-members for the deconvolution of the scenes. We also adopted the opposite approach, using Principal Component Analysis (PCA) to isolate the major spectral end-members present in each scene, and compare them to the natural samples we collected. Our results illustrate the challenges inherent to using multispectral remote sensing data in a quantitative way and emphasize the importance of collecting natural samples to guide the image analysis process.