In-Scene Atmospheric Correction
Testing In-Scene Atmospheric Corrections of Hyperspectral Thermal Data from Nadir- and Oblique-Viewing Geometries Matt R. Smith, Alan R. Gillespie, Hugau Mizzon. University of Washington, Seattle, WA, USA Lee K. Balick. Los Alamos National Laboratory, Los Alamos, NM, USA Juan Carlos Jiménez-Muñoz, Jose A. Sobrino. University of Valencia, Valencia, Spain In-Scene Atmospheric Correction (ISAC) is applied to nadir-, horizontal- and slant-looking hyperspectral thermal scenes. Nadir and horizontal views are well-corrected by this method but slant geometries are not, requiring more complicated corrections.
METHODS In-Scene Atmospheric Correction (ISAC) ISAC is a method developed by Young et al. (2002) to determine relative atmospheric parameters (transmissivity, path radiance) using only data from within the scene. It is predicated on two main assumptions: (1) the atmosphere is homogeneous and (2) there is something approximating a blackbody in the scene.
ABSTRACT In-Scene Atmospheric Correction (ISAC) is an algorithm that exclusively uses data from within the scene to extract transmissivity and path radiance terms in the radiative transfer equation for the purpose of extracting atmospheric information and determining surface mineralogy. Here we test its usefulness for a variety of hyperspectral viewing geometries: nadir-, horizontal- and slanted-views. We find that it works well for nadir-looking and horizontal-looking geometries, but not for slanted views. For slanted views, with highly variable path lengths within a single scene, we evaluate the complicating factors which may result in the complicated and noisy final emissivity spectra.
Young et al. (2002)
Figure 2. Fitting a line to the scatterplot of measured to equivalent blackbody radiance.
The In-Scene Atmospheric Calibration (ISAC) [Young et al., 2002] algorithm uses the following steps: 1. Convert the raw at-sensor measured radiance to temperature in each band for every pixel (Figure 1) 2. Pick the band the shows the highest temperature for each pixel; compare all pixels to find the most popular highest-temperature band
INTRODUCTION
3. Using only those pixels that share the most-popular highesttemperature band: make a scatterplot plotting the measured atsensor radiance vs. the equivalent blackbody radiance for that pixel s maximum temperature. (Figure 2)
One of the foremost problems with hyperspectral thermal imagery is atmospheric correction. Many techniques have been devised to solve this problem: empirical-line calibration [e.g. Qaid et al., 2009], atmospheric modeling using MODTRAN [e.g. Richter and Schlapfer, 2002], using known atmospheric water-vapor spectral features to reconstruct atmospheric spectra [Tonooka, 2001], and in-scene atmospheric corrections (ISAC) which assume that there is a population of materials in the scene that approximates a blackbody [Young et al., 2002]. Most of these techniques have been devised to correct for airborne/satellite nadir viewing geometries. When correcting for side-looking or oblique-viewing geometries, previous corrections have mainly relied on MODTRAN atmospheric modeling, with the variable path lengths calculated using a digital terrain model (e.g. Richter and Schlapfer, 2002). Of these correction algorithms, ISAC is the most easily applicable to a variety of scenes, since it requires the least a-priori knowledge of the scene. However, ISAC makes an assumption that there is a substantial population of blackbody-like materials in the scene, an assumption that is not always possible, especially in arid landscapes. Therefore, we test the applicability of ISAC to arid scenes from nadir-, horizontal-, and slant-viewing geometries and assess our ability to extract useful mineralogic data in each of these cases.
4. According to the radiative transfer equation:
Young et al. (2002)
Ls = ε B(λ,T) τ + Lu Figure 1. Converting measured radiance (upper plot) to equivalent brightness temperature (lower plot) using the Planck equation
DATA AND RESULTS For this study, we employed two hyperspectral thermal datasets: The Aerospace Corporation s Spatially Enhanced Broadband Array Spectrograph System (SEBASS) and the TELOPS Hyper-cam. SEBASS is a pushbroom scanner that collects in 128 bands between 7.57 and 13.51 µm (NEΔT≤0.05 K). Hypercam is a Fourier-transform Spectrometer, and can collect up to 150 bands between 7.7 – 11.5 µm (NEΔT=0.03 K). Hyper-cam has successfully been employed previously to retrieve surface emissivities by Balick et al. [2009].
When the blackbody radiance, B(λ,T), is plotted versus measured radiance, Ls, the slope of that line will be the surface emissivity in that band, ε, times the transmissivity of the intervening atmosphere, τ, and the intercept will be the atmospheric path radiance, Lu. The surface emissivity is always between 0-1, and therefore, a line fit to the top of the data points will fit that data which is closest to a blackbody, eliminating the emissivity term, and leaving only transmissivity. The path radiance, transmissivity and emissivity terms (Fig. 3) solved in this way yield relative values, and may be converted to absolute values with additional scaling using water vapor features.
All data for this study were taken within Death Valley, CA. Nadir views were taken of Badwater Basin by SEBASS, horizontal views were taken of Hell s Gate by SEBASS and Hyper-cam, and slant-views were taken of Badwater Basin from Dante s View by Hyper-cam. Figure 3. Emissivity spectra from a SEBASS image before and after ISAC correction
- NADIR VIEWS -
- SLANT VIEWS -
Panoramic SEBASS image
ISAC was performed on this airborne SEBASS image and emissivity spectra were extracted. SEBASS
Here ISAC provided useful emissivity spectra and allowed us to easily identify surface mineralogy: sulfates, clays, and quartz. Retrieved emissivity spectra from the scene may contain some residual atmospheric terms, causing slight deviations from pristine lab spectra.
We applied ISAC to two blocks of different path lengths, one each in the near- and far-ground of the image. The recovered atmospheric parameters (transmissivity [plotted at left] and path radiance) varied strongly between the two and surface mineralogy absorption features are apparent in our atmospheric parameters, indicating a flawed correction. However, corrected emissivity spectra (plot at right) are still able to show the surface mineralogical signatures. This shows that finding a blackbody in the scene may be difficult in arid environments even when obvious blackbody analogs (bushes) are present.
SULFATES
TELOPS HYPERCAM
0.1 km
300 m Danilina et al. (2009)
QUARTZ
Figure 4. False-color image using SEBASS of Badwater Basin, Death Valley, CA [R: 8.28 µm, G: 9.55 µm, B: 8.59 µm]
PHYLLOSILICATES (CLAYS)
CARBONATES
- HORIZONTAL VIEWS -
Figure 7. ISAC-corrected emissivity spectra from slant view of Badwater Basin floor taken with Hyper-cam from a nearby Dante s View peak..
Figure 8. Photograph taken from Dante s View of Trail Canyon Fan, with two fan units (#7, #8) marked
ISAC, when applied to TELOPS images, leaves a noisy emissivity spectrum, when compared with nadir- and horizontal-viewing geometries (Fig. 7). We hypothesize that this is due to one, or a combination, of the following factors: 1. Strong variations in path length between the near- and far-field violate the assumption – necessary for ISAC – that the atmosphere is homogeneous throughout the image
QUARTZ
Similar to nadir views, we are also able to extract useful emissivity information with ISAC on horizontal views of Hell s Gate using SEBASS data; quartz and dolomite emissivity spectra were easily identified.
TELOPS data view of Hell s Gate
2. Strongly variable and non-unity emissivities in the scene can effect the measured radiance
We do not see a strong path-length dependence in the ratio of 10 to 8 µm (Figure 11), as would be expected for a scene affected by large path length effects, these effects must not be the primary factor controlling the applicability of ISAC here. This is a surprising finding, since the atmospheric path length varies so strongly between the top and the bottom of the image..
Figure 5. False color image using SEBASS of Hell s Gate, Death Valley, CA [R: 8.3 µm, G: 9.18 µm, B: 11.15 µm]
Figure 9. Hyper-cam image of the Trail Canyon Fan and plots of uncorrected measured radiances
ε = 0.9
ε = 0.8
Figure 10. Ratio of 10 to 8 µm in uncorrected radiance data
3. Instrumental artifacts may exhibit control on scaling of measured radiance across the image This is expressed as a sub-circular patterns within the 10 to 8 µm ratio (marked as a white circle in Figure 10). However, since the observed pattern is not strictly circular, other factors (e.g. path length, surface characteristics) also exert an influence.
ε=1 Figure 11. Variable emissivities can affect the relative path length effects. For a low-emissivity surface, the contrast between a near and far target diminishes and may affect ISAC. In Figure 10, foreground salts display in the influence of emissivity on measured at-sensor radiance.
CONCLUSIONS
Figure 6. False color image of a TELOPS scene. [R: 8.3, G: 9.17, B: 11.15]
As we narrow the field of view and reduce the signal-to-noise ratio in the TELOPS scene, fewer blackbody-like materials are present in the scene, resulting in more noisy extracted emissivity profiles. However, strong absorptions are still visible, such as those for quartz (plot above at right)
Here, we find that ISAC is capable of correcting hyperspectral thermal imagery for atmospheric effects from nadir- and horizontal-viewing geometries. After employing ISAC, we may retrieve useful emissivity spectra from these images. However, for slant-view geometries, ISAC fails to yield useful emissivity spectra. Slant-views are complicated by the possibility of a varying atmosphere, surface emissivities or instrumental artifacts that result in vaguely radially symmetrical patterns in the image. Adaptions involving image segmentation according to path length may improve our ability to apply ISAC to oblique views. REFERENCES Balick, L., et al.., IEEE Geoscience and Remote Sensing Letters 6(1), 52-56 (2009). Danilina, I., et al., IEEE 1st Workshop Hyperspectral Imaging & Signal Process. Evol. Rem. Sens., Grenoble (2009). Qaid, A.M. et al. Geo-Spatial Information Science, 12, 3, 197-201 (2009). Richter, R. & Schlapfer, D. International Journal of Remote Sensing, 23, 2631-2649 (2002). Tonooka, H. IEEE Transactions on Geoscience & Remote Sensing, 39, 3, 682-692 (2001). Young, S.J. et al. J. Geophys. Res. 107, p.4774 (2002).
ACKNOWLEDGMENTS The authors would like to acknowledge support by the Department of Energy Grant DE-FG52-08NA28772
METHODS In-Scene Atmospheric Correction (ISAC) ISAC is a method developed by Young et al. (2002) to determine relative atmospheric parameters (transmissivity, path radiance) using only data from within the scene. It is predicated on two main assumptions: (1) the atmosphere is homogeneous and (2) there is something approximating a blackbody in the scene.
ABSTRACT In-Scene Atmospheric Correction (ISAC) is an algorithm that exclusively uses data from within the scene to extract transmissivity and path radiance terms in the radiative transfer equation for the purpose of extracting atmospheric information and determining surface mineralogy. Here we test its usefulness for a variety of hyperspectral viewing geometries: nadir-, horizontal- and slanted-views. We find that it works well for nadir-looking and horizontal-looking geometries, but not for slanted views. For slanted views, with highly variable path lengths within a single scene, we evaluate the complicating factors which may result in the complicated and noisy final emissivity spectra.
Young et al. (2002)
Figure 2. Fitting a line to the scatterplot of measured to equivalent blackbody radiance.
The In-Scene Atmospheric Calibration (ISAC) [Young et al., 2002] algorithm uses the following steps: 1. Convert the raw at-sensor measured radiance to temperature in each band for every pixel (Figure 1) 2. Pick the band the shows the highest temperature for each pixel; compare all pixels to find the most popular highest-temperature band
INTRODUCTION
3. Using only those pixels that share the most-popular highesttemperature band: make a scatterplot plotting the measured atsensor radiance vs. the equivalent blackbody radiance for that pixel s maximum temperature. (Figure 2)
One of the foremost problems with hyperspectral thermal imagery is atmospheric correction. Many techniques have been devised to solve this problem: empirical-line calibration [e.g. Qaid et al., 2009], atmospheric modeling using MODTRAN [e.g. Richter and Schlapfer, 2002], using known atmospheric water-vapor spectral features to reconstruct atmospheric spectra [Tonooka, 2001], and in-scene atmospheric corrections (ISAC) which assume that there is a population of materials in the scene that approximates a blackbody [Young et al., 2002]. Most of these techniques have been devised to correct for airborne/satellite nadir viewing geometries. When correcting for side-looking or oblique-viewing geometries, previous corrections have mainly relied on MODTRAN atmospheric modeling, with the variable path lengths calculated using a digital terrain model (e.g. Richter and Schlapfer, 2002). Of these correction algorithms, ISAC is the most easily applicable to a variety of scenes, since it requires the least a-priori knowledge of the scene. However, ISAC makes an assumption that there is a substantial population of blackbody-like materials in the scene, an assumption that is not always possible, especially in arid landscapes. Therefore, we test the applicability of ISAC to arid scenes from nadir-, horizontal-, and slant-viewing geometries and assess our ability to extract useful mineralogic data in each of these cases.
4. According to the radiative transfer equation:
Young et al. (2002)
Ls = ε B(λ,T) τ + Lu Figure 1. Converting measured radiance (upper plot) to equivalent brightness temperature (lower plot) using the Planck equation
DATA AND RESULTS For this study, we employed two hyperspectral thermal datasets: The Aerospace Corporation s Spatially Enhanced Broadband Array Spectrograph System (SEBASS) and the TELOPS Hyper-cam. SEBASS is a pushbroom scanner that collects in 128 bands between 7.57 and 13.51 µm (NEΔT≤0.05 K). Hypercam is a Fourier-transform Spectrometer, and can collect up to 150 bands between 7.7 – 11.5 µm (NEΔT=0.03 K). Hyper-cam has successfully been employed previously to retrieve surface emissivities by Balick et al. [2009].
When the blackbody radiance, B(λ,T), is plotted versus measured radiance, Ls, the slope of that line will be the surface emissivity in that band, ε, times the transmissivity of the intervening atmosphere, τ, and the intercept will be the atmospheric path radiance, Lu. The surface emissivity is always between 0-1, and therefore, a line fit to the top of the data points will fit that data which is closest to a blackbody, eliminating the emissivity term, and leaving only transmissivity. The path radiance, transmissivity and emissivity terms (Fig. 3) solved in this way yield relative values, and may be converted to absolute values with additional scaling using water vapor features.
All data for this study were taken within Death Valley, CA. Nadir views were taken of Badwater Basin by SEBASS, horizontal views were taken of Hell s Gate by SEBASS and Hyper-cam, and slant-views were taken of Badwater Basin from Dante s View by Hyper-cam. Figure 3. Emissivity spectra from a SEBASS image before and after ISAC correction
- NADIR VIEWS -
- SLANT VIEWS -
Panoramic SEBASS image
ISAC was performed on this airborne SEBASS image and emissivity spectra were extracted. SEBASS
Here ISAC provided useful emissivity spectra and allowed us to easily identify surface mineralogy: sulfates, clays, and quartz. Retrieved emissivity spectra from the scene may contain some residual atmospheric terms, causing slight deviations from pristine lab spectra.
We applied ISAC to two blocks of different path lengths, one each in the near- and far-ground of the image. The recovered atmospheric parameters (transmissivity [plotted at left] and path radiance) varied strongly between the two and surface mineralogy absorption features are apparent in our atmospheric parameters, indicating a flawed correction. However, corrected emissivity spectra (plot at right) are still able to show the surface mineralogical signatures. This shows that finding a blackbody in the scene may be difficult in arid environments even when obvious blackbody analogs (bushes) are present.
SULFATES
TELOPS HYPERCAM
0.1 km
300 m Danilina et al. (2009)
QUARTZ
Figure 4. False-color image using SEBASS of Badwater Basin, Death Valley, CA [R: 8.28 µm, G: 9.55 µm, B: 8.59 µm]
PHYLLOSILICATES (CLAYS)
CARBONATES
- HORIZONTAL VIEWS -
Figure 7. ISAC-corrected emissivity spectra from slant view of Badwater Basin floor taken with Hyper-cam from a nearby Dante s View peak..
Figure 8. Photograph taken from Dante s View of Trail Canyon Fan, with two fan units (#7, #8) marked
ISAC, when applied to TELOPS images, leaves a noisy emissivity spectrum, when compared with nadir- and horizontal-viewing geometries (Fig. 7). We hypothesize that this is due to one, or a combination, of the following factors: 1. Strong variations in path length between the near- and far-field violate the assumption – necessary for ISAC – that the atmosphere is homogeneous throughout the image
QUARTZ
Similar to nadir views, we are also able to extract useful emissivity information with ISAC on horizontal views of Hell s Gate using SEBASS data; quartz and dolomite emissivity spectra were easily identified.
TELOPS data view of Hell s Gate
2. Strongly variable and non-unity emissivities in the scene can effect the measured radiance
We do not see a strong path-length dependence in the ratio of 10 to 8 µm (Figure 11), as would be expected for a scene affected by large path length effects, these effects must not be the primary factor controlling the applicability of ISAC here. This is a surprising finding, since the atmospheric path length varies so strongly between the top and the bottom of the image..
Figure 5. False color image using SEBASS of Hell s Gate, Death Valley, CA [R: 8.3 µm, G: 9.18 µm, B: 11.15 µm]
Figure 9. Hyper-cam image of the Trail Canyon Fan and plots of uncorrected measured radiances
ε = 0.9
ε = 0.8
Figure 10. Ratio of 10 to 8 µm in uncorrected radiance data
3. Instrumental artifacts may exhibit control on scaling of measured radiance across the image This is expressed as a sub-circular patterns within the 10 to 8 µm ratio (marked as a white circle in Figure 10). However, since the observed pattern is not strictly circular, other factors (e.g. path length, surface characteristics) also exert an influence.
ε=1 Figure 11. Variable emissivities can affect the relative path length effects. For a low-emissivity surface, the contrast between a near and far target diminishes and may affect ISAC. In Figure 10, foreground salts display in the influence of emissivity on measured at-sensor radiance.
CONCLUSIONS
Figure 6. False color image of a TELOPS scene. [R: 8.3, G: 9.17, B: 11.15]
As we narrow the field of view and reduce the signal-to-noise ratio in the TELOPS scene, fewer blackbody-like materials are present in the scene, resulting in more noisy extracted emissivity profiles. However, strong absorptions are still visible, such as those for quartz (plot above at right)
Here, we find that ISAC is capable of correcting hyperspectral thermal imagery for atmospheric effects from nadir- and horizontal-viewing geometries. After employing ISAC, we may retrieve useful emissivity spectra from these images. However, for slant-view geometries, ISAC fails to yield useful emissivity spectra. Slant-views are complicated by the possibility of a varying atmosphere, surface emissivities or instrumental artifacts that result in vaguely radially symmetrical patterns in the image. Adaptions involving image segmentation according to path length may improve our ability to apply ISAC to oblique views. REFERENCES Balick, L., et al.., IEEE Geoscience and Remote Sensing Letters 6(1), 52-56 (2009). Danilina, I., et al., IEEE 1st Workshop Hyperspectral Imaging & Signal Process. Evol. Rem. Sens., Grenoble (2009). Qaid, A.M. et al. Geo-Spatial Information Science, 12, 3, 197-201 (2009). Richter, R. & Schlapfer, D. International Journal of Remote Sensing, 23, 2631-2649 (2002). Tonooka, H. IEEE Transactions on Geoscience & Remote Sensing, 39, 3, 682-692 (2001). Young, S.J. et al. J. Geophys. Res. 107, p.4774 (2002).
ACKNOWLEDGMENTS The authors would like to acknowledge support by the Department of Energy Grant DE-FG52-08NA28772
