tanbul 2004
CORRECTION OF ATMOSPHERIC AND BIDIRECTIONAL EFFECTS IN
n MULTISPECTRAL ADS40 IMAGES FOR MAPPING PURPOSES
1278
U. Beisl* *, N. Woodhouse^
1283
* Leica Geosystems GIS & Mapping GmbH, Heinrich-Wild-Strasse, CH-9435 Heerbrugg, Switzerland
1289 Ulrich.Beisl 9 gis.leica-geosystems.com
? Leica Geosystems GIS & Mapping LLC, 10840 Thornmint Road, Suite 100, San Diego, CA 92127, USA
1295 Neil. Woodhouse @gis.leica-geosystems.com
1301 Commission VII, WG VII/1
506 KEY WORDS: Atmosphere, Mapping, Correction, Algorithms, Automation, Model, Mosaic, Sensor
1314
ABSTRACT:
1321
The radiometry of photogrammetric images is influenced by various effects from outside the camera. One prominent effect is the
additional path radiance from atmospherically scattered sun light. This occurs especially at short wavelengths and long atmospherical
1329 p p y E E p
path lengths, which gives rise to an increasing blueshift towards the borders of the images. Another effect is the bidirectional
reflectance from the ground surface (BRDF). This effect depends on the illumination and the viewing geometry as well as on the
wavelength and is caused by a varying amount of subpixel shadows on the ground. At high solar elevation, frame sensors encounter
a bright area within the image, the so-called hot spot; line scanners like the ADS40 show an across-track brightness gradient. This
prevents precise intra- and intercomparison of images, affects spectral ratios and is adverse to proper mosaicking. In order to correct
the blueshift in the images a dark pixel subtraction algorithm is applied to the data, which accounts for the largest effects of the
atmosphere. The algorithm takes into account the view angle dependence of the path radiance by calculating column statistics. For
the bidirectional effect an automated semi-empirical algorithm is presented to correct a set of line scanner images simultaneously to a
defined viewing and solar geometry. Statistics of the image brightness are calculated and the model fitted to the averages. These
statistics can be calculated either from all pixels or can be class specific. The method is applied to ADS40 data after system and
atmospheric correction in order to produce a well-defined input for the orthorectification and mosaicking. As an example, images
from a flight campaign are processed to orthophotos and mosaicked without further dodging or feathering.
1. INTRODUCTION hue which appears towards the borders of the images (i.e. for
large view zenith angles. For a definition of the angles see
“In most cases, the atmosphere is perceived as a hostile entity Figure 1).
whose adverse impacts must be neutralized or eliminated before
remotely sensed data can be properly analyzed.” (Schott, 1997)
Despite the fact that atmospheric effects reveal physical
properties of the atmosphere, this is usually not the aim of
taking images.
Almost the same is true for bidirectional effects.
"The non-Lambertian nature of the terrestrial surface is a major
source of unexplained variability in wide-swath satellite sensor
data acquired in the solar reflective wavelength, hindering
quantitative analysis in the spectral, temporal, and locational
domains." (Chopping, 2000)
The bidirectional effects contain information about the
geometric and biophysical structure of the object, which is a
fascinating field of research, albeit progress is slow in
developing universal models and model inversion techniques.
Applications of this research area range from new classification
methods to crop yield predictions.
But again, this is usually not the intention of mapping imagery.
Therefore, fast empirical models which remove those effects are
sufficient.
2. ATMOSPHERIC EFFECTS
The most prominent atmospheric effects in aerial images even
under clear sky conditions are a general brightening and a blue
* Corresponding author.
Figure 1. Coordinate system for reflection (0; incident solar
zenith angle, 0, reflected view zenith angle, ¢ view
azimuth angle).
For operational purposes (no pressure cabin in plane) flight
elevations below 4000 m (12.000 ft) are preferred. This reduces
the amount of atmosphere between sensor and ground compared
to satellite systems. However, in order to cover a reasonable
