Comparison of Satellite Hyperspectral Data in Geological Mapping—Data Section

Next-generation satellite hyperspectral (HS) sensors offer significant potential for large-scale regional mineral mapping. However, like all satellite sensors, their results rely on complex correction processes that require removing the effects of atmosphere, topography, and geometry before accurately acquiring reflectance spectra. These corrections are often performed manually, and different methods can yield different results. Rupsa et al. compared hyperspectral satellite data from PRISMA, EnMAP, and EMIT with airborne imagery data acquired by HyMap sensors to investigate the consistency between these datasets and their applicability in geological mapping. Based on geological importance, relatively good exposure, arid climate, and data availability, Rupsa et al. selected the Marinkas-Quellen and Epembe carbonate complexes in Namibia for comparison. Qualitative and three different quantitative comparisons were performed on the hyperspectral data of these locations, including: (1) a correlation comparison of reflectance values ​​in the visible-near-infrared (VNIR) to short-wave infrared (SWIR) spectral range; (2) a comparison of spectral indices for specific minerals in each scene; and (3) a comparison of spectral abundance estimated using linear unmixing. The results show significant differences in consistency among different sensors across the visible to short-wave infrared (VNIR to SWIR) spectral range, with better consistency in the VNIR range and poorer consistency in the SWIR range. The study found that the SWIR spectral data from the EnMAP and EMIT sensors were the most reasonable (showing the most pronounced absorption characteristics), while the endmember spectral abundances of the HyMap and PRISMA sensors were consistent with geological variations. This study highlights the necessity of accurate radiometric and topographic correction in the SWIR range for geological mapping.

Rupsa et al. used high-quality (cloud-free, smoke-free, and dust-free) hyperspectral data from three satellites: EnMAP, EMIT, and PRISMA. However, the data were not acquired at the same time, and there is a time interval between different datasets. All hyperspectral data, including those from satellites and airborne platforms, required calibration and atmospheric correction to convert the measured digital signals into the true surface reflectance on the sensors.

Figure 1. Coverage area of ​​each sensor

Table 1. Detailed information on different hyperspectral sensors

Study area

Marinkas-Quellen, located in southern Namibia near the South African border, is a carbonate complex containing silicate and carbonate rocks, including calcareous, magnesian, and ferruginous carbonate variants. This carbonate complex dates from the Cambrian to the early Cambrian period and its formation is associated with inland rift activity.

The Epembe carbonate complex, located in northern Namibia, is primarily composed of calcareous carbonates (Sovites) accompanied by very narrow regions of magnesian carbonates. Dating back to the early Mesozoic era, this complex lies within a sub-northwest trending disjoint zone running north to south.

The selection of these two research areas was based on the following considerations:

Geological significance: Namibia’s carbonate complexes are associated with rare earth elements that are important for the green energy transition.

Good exposure: The geological structure of the study area is well exposed, which is conducive to remote sensing research.

Arid climate: Arid climates reduce vegetation cover, making rock features easier to detect by satellite sensors.

Data availability: Satellite multispectral data and ground rock sample data are available for the study area.

Figure 2. Geographical location and geological features of the study area

Research Methods

1. Data Preprocessing

Satellite hyperspectral data providers typically offer different levels of preprocessed data, including:

• Geometric and sensor calibration data (Level L0): Raw data, without any calibration processing.
• Radiation-corrected top-level atmospheric radiation (L1 level): These data have been radiation-corrected to provide improved radiation information, but still include atmospheric and other influencing factors.
• Atmospherically corrected surface reflectance estimates (L2 level): Atmospherically corrected data provide estimates that are closest to the true surface reflectance.

L2 data was used in this study (see Table 2). Sensor performance and inconsistencies between sensors caused by inaccuracies in their respective preprocessing workflows have been taken into account. These different preprocessing processes are as follows:

• Geometric correction: Ensures spatial accuracy of images and corrects any distortion caused by sensor movement or terrain changes.
• Radiation correction: Converts the recorded digital signal into a real radiation value that can be used for analysis, eliminating the influence of sensor characteristics.
• Atmospheric correction: Removes atmospheric scattering and absorption effects to obtain reflectivity close to that of the Earth's surface.

Table 2 Information on each dataset in the study

PRISMA

The PRISMA L1 radiation data generation process includes flat-field thresholding correction and dark pixel methods to generate and update radiation, spectral, and geometric parameters. Subsequently, the L2 processor uses L1 top-level atmospheric radiation data and panchromatic bands as input to generate the final atmospherically corrected surface reflectance data.

EnMAP

The L2 data used in this study by EnMAP was derived using the Python-based Climate Correction Module (PACO) developed by the German Aerospace Center (DLR). While PACO is primarily based on the ATCOR-IDL code, it is known for its unique and rigorous calibration and validation procedures. The model automatically determines scene-specific biome density, ozone levels, and season using sensor configuration parameters such as solar zenith angle and observation off-axis angle. These parameters are then used for ozone, cirrus, and haze correction. The outputs generated by the process include image masks, aerosol optical thickness, wavelength maps, and low-level atmospheric reflectance images. The final L2A reflectance product is a combination of these outputs. EnMAP also provides a range of user-selectable acquisition parameters, such as observation angle, which is particularly important in areas with significant topographic variations.

EMIT

EMIT represents a potentially significant advance in hyperspectral imaging, employing an innovative detector to capture the entire VNIR-SWIR spectrum on a single sensor. This technology opens new opportunities for NASA's upcoming SBG (Surface Biology and Geology) missions, helping to mitigate the discrepancies between the VNIR and SWIR ranges. EMIT uses radiative transfer modeling tools to generate the final reflectance product and performs radiometric calibration to ensure consistency of assumptions and reduce model-induced errors. It also leverages bright cloud features in the scene to monitor short-wavelength (VNIR) calibration. Furthermore, EMIT uses cross-calibration with other on-orbit instruments to evaluate the absolute accuracy of pre-defined calibration protocols. EMIT's novel preprocessing workflow ensures consistent product quality throughout its lifecycle. Its 60-meter spatial resolution effectively meets the needs of its large-scale scientific missions, such as mapping minerals and dust flows. EMIT's low Earth orbit deployment on the International Space Station also provides opportunities for high-frequency data acquisition.

HyMap

The airborne hyperspectral dataset was acquired using the HyMap sensor at an altitude of 2000 meters, with 5-meter spatial sampling. Geometric and radiometric corrections were performed by HyVista, the company that acquired the data. Atmospheric corrections used a continental aerosol model and a mid-latitude summer atmospheric model, assuming an ozone concentration of 340 ppm and a visibility of 75 km, to estimate surface reflectance.

In the HyMap scene received from Eppembe, vegetation patches have been masked. However, this masking was not performed in other scenes because at resolutions of 30 meters and 60 meters, most pixels contain some vegetation.

 引用   :   Chakraborty, R.; Rachdi, I.; Thiele, S.; Booysen, R.; Kirsch, M.; Lorenz, S.; Gloaguen, R.; Sebari, I. A Spectral and Spatial Comparison of Satellite-Based Hyperspectral Data for Geological Mapping. Remote Sens. 2024, 16, 2089. https://doi.org/ 10.3390/rs16122089