See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/40125430

Evaluation of aerosol mapping methods using AVIRIS imagery

Article · January 2004

Source: OAI

CITATIONS READS
7 242

5 authors, including:

Christoph Popp Daniel Schläpfer

ReSe Applications LLC ReSe Applications LLC
45 PUBLICATIONS 1,153 CITATIONS 188 PUBLICATIONS 4,546 CITATIONS

SEE PROFILE SEE PROFILE

Stephan Bojinski M.E. Schaepman

World Meteorological Organization University of Zurich
40 PUBLICATIONS 1,555 CITATIONS 262 PUBLICATIONS 10,033 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Christoph Popp on 23 May 2014.

The user has requested enhancement of the downloaded file.

 EVALUATION OF AEROSOL MAPPING METHODS USING
AVIRIS IMAGERY

Christoph Popp 1 , Daniel Schläpfer1, Stephan Bojinski2, Michael Schaepman 3, and Klaus I. Itten 1
University of Zurich, Switzerland

1 INTRODUCTION

Natural and anthropogenic aerosols affect solar electromagnetic radiation on its propagation through
the Earth’s atmosphere. Their influence is twofold, directly by reflecting, scattering or absorbing
processes and indirectly serving as cloud condensation nuclei (King et al, 1999). These processes in-
fluence the radiation budgets significantly from local to global scales thereby forming a major un-
certainty in climate models.

Remote sensing has in recent years become an important tool to determine aerosol properties
thus leading to a better understanding of the role of aerosols in the atmosphere. Due to the possibility
of providing spatially homogeneous data, airborne or spaceborne sensor measurements constitute a
main achievement to other approaches (e.g., sun photometry). Further, quantification of atmospheric
aerosol characteristics is also needed as input parameter for the atmospheric correction of imagery
data since aerosols modulate the optical signal reflected from a surface target into the sensor’s field of
view. For those reasons, different methods for the image-based retrieval and spatial mapping of aero-
sol parameters have been developed. Their principles are strongly connected to number, position and
width of spectral bands, ground sampling distance, repetition rate, and view angle of the particular
sensor type (King et al, 1999, gives an overview).

The objective of this paper is the application of standard aerosol mapping methods using year
2000 AVIRIS data measured over the greater Los Angeles (USA) area. Moreover, a new technique
especially developed for aerosol retrieval by means of imaging spectroscopy data named Aerosol Re-
trieval by Interrelated Abundances (ARIA) (Bojinski, 2003) is introduced as an alternative. Above
mentioned methods are all based on the occurrence of dark surface types in the investigated area. A
rural region in the Santa Monica Mountains as well as a flat urban area in the vicinity of downtown
Los Angeles have been chosen as test sites. Both scenes contain various surface types. Columnar aero-
sol optical depth (τa) at 550 nm for different aerosol regimes is calculated using a look-up table ap-
proach. The retrieved results are validated using the MODIS level2 aerosol product and in relation to
the terrain surface reflectance and structure.

1
Department of Geography, RSL, University of Zürich, Winterthurerstr. 190, CH-8057 Zürich, Switzerland,
Phone: +41 1 635 52 50, Fax: +41 1 635 68 46, E-mail: chpopp@geo.unizh.ch
2
Global Climate Observing System (GCOS) Secretariat, c/o WMO, Geneva, Switzerland
3
Wageningen University, Centre for Geo-Information, Wageningen, The Netherlands

- 1 -
2 METHODOLOGY

Aerosols attenuate the radiation due to their diverse particle size distribution and chemical composi-
tion in a wide wavelength range, mainly in the ultraviolet and visible part, except for large dust parti-
cles whose signatures are diminishing with longer wavelengths. Scattering at aerosol particles can be
described by the Mie scattering approximation. The aerosol optical depth (τa) describes the scattering
and absorption of aerosols integrated over an atmospheric column.

The principle behind remote sensing of aerosol properties might be described using radiomet-
ric transfer equations. The contribution of radiance reflected by atmospheric aerosols into the sensors
field of view over dark surfaces basically can be expressed as:

ρa (λ) = ρ a p p (λ) - ρground (λ), (1)

where ρa p p is the apparent reflectance measured by the sensor and ρground denotes the reflectance at a
surface target. Equation 1 illuminates the importance of an appropriate presumption of ρground(λ) in
case of deriving aerosol parameters. Dark surface types are chosen as ideal targets for the inversion of
aerosol optical depth mainly because of three reasons: firstly, the lower ρground(λ) the better ρa p p(λ)
can be separated into a contribution of ground reflectance and a contribution of aerosol scattering;
secondly, ρground(λ) might be estimated more precisely than for brighter targets with regard to abso-
lute error; and third, absorption and backscattering of ground reflected radiance on aerosols can be
neglected. Aerosol optical depth then is derived by upward-calculation of ρ ground(λ) for aerosol pa-
rameter variations carried out numerically by pre-calculated look-up tables.

1.1 Dark Target

The dark target approach requires the occurrence of dense dark vegetation in the investigated area
(Kaufman and Sendra, 1988). For instance, trees or bushes are commonly dark in the visible part of
the electromagnetic spectrum. In atmospherically uncorrected imagery, dark vegetation pixels are
identified by a threshold of a vegetation index. The surface reflectance of these pixels can subse-
quently be approximated by 0.02 +/- 0.01 reflectance in the red VIS channel of a particular sensor.

1.2 MODIS Band Ratio

MODIS’s band ratio method (Kaufman et al., 1997) is based on an empirical relationship between
SWIR reflectance at λ=2.1 µm and visible reflectance at λ=0.66 µm and λ=0.49 µm for a large variety
of natural surfaces. Compared to the VIS wavelength range where most aerosol interaction with elec-
tromagnetic radiance occurs, a weak aerosol effect in the SWIR region can be assumed. Candidate
Pixels are masked in the SWIR by thresholding and assumed ground reflectance in the VIS are calcu-
lated using the empirical ratios.

1.3 ARIA

Dissimilar to the above mentioned methods, the Aerosol Retrieval by Interrelated Abundances (ARIA,
Bojinski, 2003) now makes explicitly use of the high number of spectral channels provided by
AVIRIS and other imaging spectrometry instruments. ARIA is based on unmixing of the spectrally
contiguous sensor signal in the SWIR range using a priori defined land surface reflectance endmem-
bers. The derived abundances are interrelated between VIS and SWIR spectral ranges, if spectral auto-
correlation is given. In this case, the abundances are derived from unmixing in the SWIR and then are
used with the same endmembers to re-mix an assumption on surface reflectance in the VIS.

- 2 -
1.4 Further Methods and Trends

The combination of high spectral resolution with high signal to noise ratio bears the potential for a
variety of other promising approaches. A method comparable to the ARIA method has been devel-
oped for SeaWIFS data by Von Hoyningen-Huene et al. (2003). It iteratively mixes the surface re-
flectance from reference spectra in the VNIR spectral range on the basis of vegetation abundance,
derived at the same spectral range. Good results have been reported for this method on SeaWIFS, but
its implementation on AVIRIS data has not yet been performed and tested. Another option is the use
of the oxygen absorption at 760nm to retrieve the total optical thickness as it is done with MOS-B on
IRS-1 (Pflug et al., 1999; Zimmermann 1991). The method requires spectral resolution below 5 nm
and therefore is not suited to be used with AVIRIS data. A further method has been developed for
GLI on ADEOS-1 (Höller et al., 2003). It detects the strong aerosol signature in the UV bands be-
tween 380 and 420 nm to estimate the aerosol contents and aerosol single scattering albedo. Also that
approach is not well suited for AVIRIS data as noise levels are high in these critical wavelength re-
gions.

3 DATA BASIS

The NASA Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) onboard a U-2 aircraft flew
over the Californian coast line from downtown Los Angeles westward to Point Mugu near Oxnard on
16 September 2000 at 11.00 local time and on a nominal altitude of 20 km above sea level. The
complete flight strip measures 95 km in length and 12.3 km in width covering a wide range of differ-
ent surface types. The area is sun-illuminated from the south-east and is cloud-free.

For comparison of aerosol retrieval methods, an urban and a mainly rural region have been
chosen as test areas, in the following referred to as ‘City’ and ‘Topanga’ scene. Further, a USGS
digital elevation model (DEM) with a spatial resolution of 10 m is used for orthorectification and at-
mospheric processing of the imagery.

Figure 1 illustrates the described AVIRIS image, the positioning of the two selected scenes as
well as the corresponding digital elevation model in the lowest panel. The ‘Topanga’ scene includes
mountainous areas of the Santa Monica Mountains to the north and west reaching terrain heights
around 700 meters above sea level. These hilly areas are mostly covered by chaparral and dry bush
lands, some small patches of bare soil also show up, most probably the result of periodic bush fires in
this region. To the southeast, urban areas sprawl into the Santa Monica Mountains. To the south and
west, the Pacific Ocean clearly is divided from the continent through a bright strip of sandy beaches.

The ‘City’ scene contains almost exclusively plane urban area with bigger patches of dark green
vegetation which are mostly referred to as city parks. Here, brightness is proportional to building den-
sity. The Inglewood oil field at the southwestern boarder of the scene consists of bare soil with lowest
average reflectance in the VIS.

- 3 -
Figure 1: Location of the AVIRIS flight strip and the investigated scenes. In the lowest panel, the corresponding
DEMs are shown.

- 4 -
4 APPLICATION USING AVIRIS IMAGERY

The imagery has been preprocessed in two steps. First an orthorectification is carried out with a para-
metric approach using the PARGE software (Schläpfer and Richter, 1998). Second, atmospheric proc-
essing of the imagery data is done using ATCOR-4 (Richter and Schläpfer, 2002). The atmospheric
processing is done for two reasons: atmospheric correction with varying visibility and subsequent
comparison to specific suitable and well-known endmember spectra (e.g. ocean water) can provide a
first approximation of mean τ a across a scene (τ a and visibility are directly related). Aerosol optical
depths in between 0.22 and 0.35 in combination with a maritime or rural aerosol type have been
found by this method to be typical conditions. Secondly, the atmospheric preprocessing is done to
remove Rayleigh scattering as height dependent disturbing factor. However, topographic correction
has intentionally not been applied in order to keep the nature of shadowed areas as dark reference
targets.

The calculation of look-up tables is performed using MODTRAN 4 (Berk et al., 1989) for a
maritime, a rural as well as an urban aerosol model after Shettle and Fenn (1979) which are assumed
to be the most dominant and plausible aerosol models in the investigated area. In this study, US stan-
dard atmosphere 1976 is used as input parameter in MODTRAN 4 and ATCOR-4.

To complement the later achieved aerosol mapping over land for all three methods, aerosol op-
tical depth should also be derived over water surfaces for a spatially homogenous final aerosol map. A
water mask is obtained setting thresholds of ρ app (0.665) <= 0.02 and ρ app (0.865) <= 0.02, respec-
tively (Carrère et al., 1997). Water reflectance at λ=1.00 µm then is assumed to be 0.00 afterward be-
ing inverted using the pre-calculated look-up tables (Gao et al, 2000).

1.1 Dark Target

Identification of dense, dark vegetation pixels is achieved with the atmospherically resistant vegetation
index (ARVI) (Kaufman and Tanré, 1992). The ARVI threshold is flexibly set such that 5% of the
image pixels are classified. The darker half of these pixels in the NIR band finally is chosen as candi-
date pixels and their ρ ground (0.665µm ) is assumed to be 0.02 +/- 0.01.

1.2 Band Ratio

Masking of candidate pixels is done by setting two thresholds in the SWIR channel: a minimum of
0.005 reflectance avoiding the identification of water pixels for which this method does not work. A
maximum of 0.05 reflectance is set such that at least 5% of the image pixels are selected. Otherwise
this threshold is increased to 0.01. The overall brighter city scene requires the 0.01 threshold. To fin-
ish, an assumed ground reflectance in the red VIS is determined for every masked pixel applying the
standard band ratio factor of 0.5 between SWIR and VIS band.

1.3 ARIA

Spectral autocorrelation has been investigated for spectral bands between 0.4 and 0.7 mm in the VIS,
and the SWIR1 and SWIR2 region, respectively (Bojinski, 2003). Reflectance spectra for soil (895
spectra from a spectrum database), vegetation (1239), and rocks/minerals (917) surfaces have been
analyzed. Averaged band correlations have been calculated. It has been revealed that for soil and
vegetation, the SWIR1 and SWIR2 spectral ranges are uniformly correlated to the VIS part which has
not been the case for rocks and minerals. Hence, rocks/mineral spectra are excluded from the analysis.
Determining suitable candidate pixels for the inversion is done using the spectral angle mapper algo-

- 5 -
rithm (Richards and Jia, 1999). Spectral angles calculated between the atmospherically uncorrected
image and one of the used endmember spectra equal or less than 0.15 are considered fitting. Next, the
assumed VIS reflectance of the masked pixels is re-mixed using the abundances derived in the SWIR
spectral region.

A spatially contiguous product finally is achieved by completing the land τ a values with those
derived over water followed by an interpolation of τ a values to fill the gaps between pixels carrying
aerosol information.

2 RESULTS, COMPARISON, AND DISCUSSION

2.1 Quantitative Validation

Aerosol optical depth is calculated for maritime, rural, and urban aerosol models for the dark target,
band ratio, and ARIA method. Mean values of τa, their standard deviations, and the expectable errors
for each result are illustrated in Table 1 for the Topanga scene and Table 2 for the City scene. For
rural and maritime models, small differences can be observed in all categories. Use of the urban
model yields values that stand noticeably apart in both scenes. In the dark target method, no average
results could be determined as the fit of simulated to real data during the inversion process was inap-
propriate for the urban aerosol model. Maritime and rural results for mean τ a in the dark target and
the band ratio method are higher than for the ARIA method. In the Topanga scene, in-scene varia-
tions are approximately equally high for the dark target and ARIA methods, although at a higher
overall τa level for the latter, and slightly higher for the band ratio method. In the city case, ARIA in-
scene variation is lowest.

An urban aerosol regime does not seem to fit the atmospheric conditions the time the data has
been collected. Considering maritime and rural results only, in absolute terms, the ARIA methods av-
erage error is smaller than in the dark target method and higher than in the band ratio method. Toler-
able accuracies of Δτa +/- 0.1 are exclusively attained by the band ratio retrieval in the Topanga scene
for assumed rural and maritime aerosol regimes.

The given results seem to corroborate a similar performance of ARIA and band ratio method.
It has to be noted, that all suited pixels have been included for the statistics. The ARIA method allows
for aerosol detection on almost double the amount of pixels than the band ratio. If only the inter-
secting mask is evaluated, the performance of ARIA is clearly superior by a factor of 2 (i.e. standard
deviations at 0.1-0.3 compared to 0.5-0.6 in Δτ a).

Method Aerosol mean τa stddev τa Δτ a

ARIA maritime 0.21 0.04 0.13
ARIA rural 0.18 0.03 0.11
ARIA urban 0.44 0.12 0.40
Band Ratio maritime 0.24 0.06 0.10
Band Ratio rural 0.23 0.06 0.10
Band Ratio urban 0.56 0.22 0.32
Dark Target maritime 0.30 0.04 0.18
Dark Target rural 0.31 0.04 0.18
Dark Target urban - - -

Tab.1:Topanga scene results obtained with ARIA, band ratio, and dark target methods for different aerosol models.

- 6 -
 Method Aerosol mean τa stddev τa Δτ a
ARIA maritime 0.28 0.09 0.15
ARIA rural 0.25 0.08 0.14
ARIA urban 0.86 0.51 1.49
Band Ratio maritime 0.34 0.10 0.11
Band Ratio rural 0.33 0.10 0.12
Band Ratio urban 0.82 0.31 0.51
Dark Target maritime 0.44 0.10 0.16
Dark Target rural 0.47 0.10 0.17
Dark Target urban - - -

Tab.2: City scene results obtained with ARIA, band ratio, and dark target methods for different aerosol models.

2.2 Qualitative Validation

Figure 2 shows the interpolated and smoothed τ a mapping results using a rural model for the three
methods evaluated in this study for the Topanga scene and in the City scene, respectively. Dark water
retrieval results are added to the raw aerosol map before interpolation. White areas denote unrealistic
high τa values appearing where very few candidate pixels could be found. In the Topanga scene, the
previously made conclusions are confirmed. The highest τ a values can be found in urban or semi-
urban areas in the eastern part of the image, the lowest are found in vegetated mountainous regions in
the western part for all methods. All derived results show a relative strong correlation with geographi-
cally varying surface covers. This is not a very realistic representation of the spatial distribution of
aerosols in the Earth’s atmosphere since this rather depends on other factors (e.g., wind direction,
aerosol size distribution). The highest overall τa values are retrieved using the dark target method, the
lowest applying ARIA. Band ratio and ARIA maps show a similar structure, with stronger variability
for the primary. Furthermore, red and blue areas are stronger speckled in the band ratio outcome.
The dark target result seems smoother than the band ratio but includes significant higher τa values.

The aerosol inversion over water provides realistic uniform τ a at reasonable absolute values.
Artificially high τ a values only show up along the coastline, most likely due to suspended matter or
shallow water depths which undermine the dark water assumption. This might also explain the errone-
ously classified water pixels in the band ratio method. Assumed that aerosol concentrations usually do
not change much on short distances, the contrast between over land and over ocean retrieved τ a values
can be considered as a supplementary criterion. Here, the dark target result shows the best fit whereas
there are higher contrasts between ocean and vegetated areas in the other two methods.

In the City scene, applying the band ratio and the dark target methods, the highest τ a values
are found in areas including high building-density. The lowest τa values show up in the oil field with
its low VIS reflectance in the southwest of the scene. The speckled and fragmentary resulting map
from the band ratio and the dark target retrieval do not look very promising. The results appear most
reasonable for the ARIA method, with relatively low spatial variation of τa. Compared to the Topanga
scene, τ a values in the City scene are significantly higher for the dark target and the band ratio
method, except around some obviously low reflectance regions. This discrepancy cannot simply be
assigned to higher air pollution in industrialized urban areas. It is rather a consequence of lacking
appropriate dark pixels in the image. Because there are at least 5% of the image pixels chosen, these
pixels are generally brighter than those chosen in the Topanga scene. For this reason, their real
ground reflectance is strongly underestimated which finally leads to higher τa values.

A last criterion is the relation between aerosol mapping results and terrain shape. Lower aero-
sol load is expected at higher altitudes (compare Figure 1 for the digital terrain model). Here, only the
Topanga scene can be evaluated. A certain relation between hill crests and aerosol contents can be

- 7 -
observed for both, the ARIA and the band ratio method. However, the interpolation routine applied to
fill the gaps of non-evaluated pixels apparently reduces the terrain effect significantly. Much more
obvious effects would be expected for a method working in a pixel-wise manner.

Figure 2: Final τa mapping results for ARIA, band ratio (BR) and dark target (DT) methods for an assumed ru-
ral aerosol model in the Topanga scene (left column) and the City scene (right column). White areas denote pixels
with values that are considered unreasonably high. Aerosol optical depth values are given at 0.55 µm.

2.3 Validation with MODIS Data

Result qualities are assessed through indirect validation with the MODIS level 2 aerosol product due to
lack of direct validation data such as sun photometry or other ground-based radiometric measure-

- 8 -
ments. The mean τa value obtained by the MODIS aerosol product for the Topanga scene is 0.08 +/-
0.07 and for the City scene 0.19 +/-0.09. The MODIS aerosol type is given as smoke for all pixels
derived from geographically and seasonally defined climatology after d’Almeida et al. (1991).
MODIS τa values are linearly interpolated for the centers of the Topanga and City scene on account
of the large difference in scale – the ground resolution of the MODIS aerosol product is 25 km. Con-
siderable differences can be observed between the results obtained in this study and the MODIS aero-
sol product. Generally, this may be due to the difference in scale. The ARIA results come closest to
the reference product τ a values. For the other two methods, not even the error bars overlap to those
provided by the MODIS data. This cautiously verifies a slightly better quality of the ARIA method.

3 CONCLUSIONS

Aerosol optical depth has been derived for the same data using three different retrieval methods. Two
traditional methods which make use of only a few spectral bands are compared to a recently devel-
oped method labeled ARIA which includes channels of a wider spectral range in order to determine
atmospheric aerosol properties. An assumed urban aerosol model obviously does not represent at-
mospheric conditions during the data acquisition in none of the evaluated scenes. Results for maritime
and rural aerosol regimes are at a comparable level for every inversion. Application of the traditional
methods on areas with a lack of sufficient low reflectance surfaces in the VIS leads to unsatisfactorily
results. In vegetated regions, their derived τa values are in expectable ranges taking visibility from the
most conceivable a priori atmospheric correction as reference. The prospects of the dark target and
the band ratio methods might rather be in fast image-based assessment of visibility values as input pa-
rameters for the atmospheric correction of remote sensing data, than in spatially homogenous aerosol
mapping. Further, their applicability remains limited to areas with large parts of appropriate dark sur-
face types.

In-scene smoothness is more reasonable for ARIA. In lack of suitable ground truth data, the
MODIS level 2 aerosol product has been used for validation, whereas ARIA mean τa values showed
best correspondences. Nevertheless, the calculated errors of every inversion are at high levels. Both of
the methods provided adequate independency between surface cover reflectance and retrieved τa,
which is relevant in interpolated gaps.

Particular methods and algorithms for remote sensing of aerosol parameters have to be devel-
oped in imaging spectroscopy with regard to more accurate and reliable spatial mapping of aerosols.
The advancement of imaging spectrometer technology will lead to sensors allowing to make use of
techniques whose potential is known but which cannot be applied to a sensor of AVIRIS type per-
formance. The APEX system (e.g.; Schaepman et al., 2003) will offer full spectral coverage down to
380 nm at spectral resolutions below 5 nm, which allows for application of methods known from
SeaWIFS, MOS-B, or GLI to full resolution imaging spectrometry data. A promising approach to be
followed with APEX-type instruments is the evaluation of the oxygen feature in combination with the
strong aerosol signatures at and below 400nm.

4 ACKNOWLEDGEMENTS

We thank Dar Roberts and Phil Dennison (both University of California) for providing digital terrain
data. Part of this work has been supported by ESA/ESTEC contract 14906/00/NL/DC. NASA/JPL and
the AVIRIS team are thanked for providing the AVIRIS data.

- 9 -
 5 REFERENCES

Berk, A., Bernstein, L., and Robertson, D., 1989: MODTRAN: a Moderate resolution Model for Lowtran7. Techni-
cal Report, Spectral Sciences Inc., Burlington, Massachusetts, pp. 38.
Bojinski, S., 2003: Imaging Spectroscopy of Aerosols using a Reference Spectrum Database. PhD Thesis. Remote
Sensing Series, Vol. 40. University of Zurich, Switzerland, pp.87.
Carrère, V., Dubuisson, P., Roger, J., and Santer, R., 1997: MERIS Level 2 Atmospheric Correction over Land :
Pixel Identification, Data Simulation and Algorithm. Physical Measurements and Signatures in Remote Sens-
ing V1. Ed. by G. Guyot & T. Phulpin, A.A. Balkema, Rotterdam, The Netherlands, pp. 27-37.
D’Almeida, G., Koepke, P., and Shettle, E., 1991: Atmospheric Aerosols: Global Climatology and Radiative Char-
acteristics. Deepak, Hampton, Virginia, pp. 561
Gao, B., Montes, M., Ahmad, Z., and Davis, C., 2000: Atmospheric Correction Algorithm for Hyperspectral Re-
mote Sensing of Ocean Color from Space. Applied Optics, 39(6):887-896.
Höller R., Higurashi A., Teruyuki N. and Nakajiama T.Y., 2003: Retrieval of aerosol optical thickness and single-
scattering albedo over land from GLI, 35th conference of the remote sensing society of Japan. RSSJ, HIVE
Nagaoka, Japan, pp. 275-276.
Kaufman, Y. and Sendra, C., 1988: Algorithm for Automatic Atmospheric Corrections to Visible and Near-IR Satel-
lite Imagery. International Journal of Remote Sensing, 9(8):1357-1381.
Kaufman, Y. and Tanré, D., 1992: Atmospherically Resistant Vegetation Index (ARVI) for EOS-MODIS. IEEE
Transactions on Geoscience and Remote Sensing, 30(2):261-270.
Kaufman, Y., Wald, A., Remer, L., Gao, B., Li, R., and Flynn, L., 1997: The MODIS 2.1-mm Channel-
Correlation with Visible Reflectance for Use in Remote Sensing of Aerosol. IEEE Transactions on Geo-
science and Remote Sensing, 35(5):1286-1297.
King, M., Kaufman, Y., Tanré, D., and Nakajima, T., 1999: Remote Sensing of Tropospheric Aerosols from Space:
Past, Present, and Future. Bulletin of the American Meteorological Society, 80(11):2229-2259.
Pflug B., Krawczyk H. and Posse P., 1999: Atmospheric Ground Truth Measurements for Validation of MOS-Data
and Angstrom Inversion Algorithm, 3rd International Workshop on MOS-IRS and Ocean Colour. Institute of
Space Sensor Technology and Planetary Exploration, DLR, Berlin, pp. 121-129.
Richards, J. and Jia, X., 1999: Remote Sensing Digital Image Analysis. Springer-Verlag, Berlin, Germany, 3rd Edi-
tion, pp. 363.
Richter R. and Schläpfer D., 2002: Geo-atmospheric processing of airborne imaging spectrometry data. Part 2: At-
mospheric/Topographic Correction. International Journal of Remote Sensing, 23(13): 2631-2649.
Schaepman M.E., Itten K.I., Schläpfer D., Kaiser J.W., Brazile J., Debruyn W., Neukom A., Feusi H., Adolph P.,
Moser R., Schilliger T., De Vos L., Brandt G., Kohler P., Meng M., Piesbergen J., Strobl P., Gavira J., Ul-
brich G. and Meynart R., 2003: APEX: Current Status of the Airborne Dispersive Pushbroom Imaging Spec-
trometer. R. Meynart (Editor), International Symposium on Remote Sensing. Sensors, Systems, and Next
Generation Satellites VII. SPIE, Barcelona, Vol. 5234, pp. 202-210.
Schläpfer D. and Richter R., 2002: Geo-atmospheric processing of airborne imaging spectrometry data. Part 1: Para-
metric Ortho-Rectification Process. International Journal of Remote Sensing, 23(13): 2609-2630.
Shettle, E. and Fenn, R., 1979: Models for the Aerosols of the Lower Atmosphere and the Effect of Humidity Varia-
tions on their Optical Properties. Technical Report AFGL-TR-79-0214, Air Force Geophysics Laboratory,
Hanscom AFB, Massachusetts, pp. 94.
Von Hoyningen-Huene, W., M. Freitag, and J.P. Burrows, 2003: Retrieval of Aerosol Optical Thickness over Land
Surfaces from Top-of-Atmosphere Radiance, Journal of Geophysical Research-Atmospheres, 108:4260.
Zimmermann G., 1991: Fernerkundung des Ozeans. Akademie Verlag, Berlin, pp. 419.

- 10 -

View publication stats