Transcript Slide 1

Geog 577 Paper Discussion
Dahl Winters
December 6, 2006
Pat S. Chavez, Jr. 1996. Image-Based Atmospheric Corrections –
Revisited and Improved. Photogrammetric Engineering and
Remote Sensing 62(9): 1025-1036.
Background
Radiometric Correction Models
The objective of a radiometric atmospheric correction procedure is to devise a
model to convert a satellite’s digital photon counts (DCs) to ground
reflectances. Derivation of the different model parameters depends on the
information available (in-situ ground data or entirely image-based data).
1. DCs converted to at-satellite radiances by removing the gain and offset
effects introduced by the imaging system.
2. At-satellite radiances converted to surface reflectances by correcting for both
solar and atmospheric effects.
Objective
The objective is to develop a purely image-based atmospheric correction method for
satellite images. This would enable easier use of historical satellite images, or images
taken in inaccessible areas where field measurements cannot be done.
The DOS (dark-object subtraction) method is strictly image-based, but has unacceptable
accuracy because it corrects only for the additive scattering effect of the atmosphere,
and not for its multiplicative transmittance effect.
In this paper, the DOS model is expanded upon by including a simple correction for the
multiplicative transmittance effect, which is caused by both scattering and absorption.
Methods
Data: Spectral data from Moran et al (1992) suitable for testing multiple atmospheric
correction methods under a variety of conditions.
Two ways of deriving the required multiplicative transmittance-correction coefficient are
presented: the COST and default TAUz methods.
COST: uses the cosine of the solar zenith angle, which is a good first-order approximation
of the atmospheric transmittance for the study sites and dates.
TAUz: uses the average of the transmittance values computed using in-situ atmospheric
field measurements made during 7 different satellite overflights.
Results
Two entirely image-based radiometric correction models are presented, generating results
with comparable accuracy to those developed from models using in-situ field
measurements. These are variations of the DOS (dark-object subtraction) model with
the addition of a multiplicative transmittance correction.
The corrections generated by the entirely image-based COST model are as accurate as
those generated by models using in-situ atmospheric field measurements.
This means, at least for the atmospheric conditions existing at the study sites and times,
the COST model can be used for atmospheric correction without the need of doing
field measurements.
Figure 1
Figures 2 and 3
Table 4
Moran et al. 1992, Uncorrected/Apparent Reflectance
Table 5
Moran et al. 1992, DOS 1-percent reflectance
Table 6
Moran et al. 1992, SSD Reflectance
Table 7
Moran et al. 1992, HBC Reflectance
Table 8
COST Reflectance, New Image-Based Model
Table 9
Default TAUz Reflectance, New Image-Based Model
Discussion
The COST model works well for images taken of a semi-arid to arid environment. Both
the COST and TAUz models approximate the transmittance values for a non-arid
environment. Further testing is needed for non-arid environments, different
atmospheric conditions, and images with > 55 degree solar zenith angles.
Overcorrections:
• The COST model uses a cosine-function correction for multiplicative transmittance
that may overcorrect at higher zenith angles, making the TAUz model more
appropriate for such images.
• Tables and scatter plots show that all models overcorrect for very low reflectances,
with the DOS model overcorrecting the least.
The additive scattering correction is more important for darker reflectances, while the
multiplicative transmittance correction is more important for brighter reflectances.
Future Research: The data used (TM bands 1-4) did not cover the full range of
reflectances; future studies should use targets including more of the dynamic range of
reflectances in all TM bands. Accuracy differences between the visible and IR bands
or soils vs. vegetation may actually be a situation of bright vs. dark.