Adams & Gillespie: Remote Sensing of Landscapes with Spectral Images
Remote Sensing of Landscapes with Spectral Images A Physical Modeling Approach John B. Adams & Alan R. Gillespie
with Spectral Images
A Physical Modeling Approach
John B. Adams & Alan R. Gillespie
Wheat Fields, Eastern
The authors invite readers to ask questions about the contents of the book. Questions that are of general
interest will be posted, along with replies by the authors.
We also encourage readers to send us comments about the book.
In what ways have you found the book useful?
Can you suggest improvements?
Please alert us to errors that we have missed.
John Adams and Alan Gillespie are on the faculty of the Department of Earth and Space Sciences
in the W.M. Keck Remote
Sensing and Planetary Sciences Lab at the
To send messages e-mail email@example.com
Mail correspondence to:
Remote Sensing Laboratory
Department of Earth and Space Sciences
University of Washington
Question #101. Posted 15 August 06. From a student: "Why did you use so many Landsat images to illustrate the
book instead of new images like AVIRIS or Hyperion? Aren't multispectral images out of date?"
Reply to #101. Most of the principles that apply to analyses of multispectral data apply also to hyperspectral data.
Hyperspectral images excel because they capture fine spectral detail, but we didn't emphasize "spectral mining" or
mineral identification in the book. The Landsat images in the book were used by the authors for field studies. The
images date back to the times of the field projects. They have the advantage that we knew a lot about the conditions
on the ground when the images were acquired. Some of the studies are "old" because we are, too. We also wanted to
emphasize the importance of being able to extract information from the rich archive of Landsat images that dates back
to the 1970s. Current studies of environmental change at landscape scales depend on being able to compare present and
Question #102. Posted 15 August 06. From a student: "How come you hardly mention AVHRR, MODIS and many
other types of environmental monitoring images?"
Reply to #102. AVHRR and MODIS images are great tools for working at regional to global scales, but they are not
the first choices to take into the field to solve problems on the ground. In our experience, as the size of the pixel
footprint gets larger, it usually becomes more difficult to interpret the spectral information in terms of what is
on the ground. There are lots of spectral images out there, but we did not attempt to organize the book around
imaging systems. We avoided images that were difficult for most field investigators to acquire because of cost or
Question #103. Posted 15 August 06. From J.W: "In the discussion of spectral mixture analysis on page 139,
you say that you can remove shade by normalizing the fractions of the other endmembers. But when I do that I
sometimes get strange results. What is going on?"
Reply to #103. The example on page 139 uses the synthetic image: all endmember fractions are between 0 and 1.
In real images it is common to have some pixels that have "overflow" fractions (<0 or >1). If overflow fractions
are included when normalizing, the results of the arithmetic can indeed be strange. However, we can work around
this problem (page 172) by truncating the overflow fractions. Fractions <0 are set to 0, and fractions >1 are set to 1.
To understand the rationale for doing this, it may help to review why overflow fractions occur in the first place.
For example see pages 132 and 274. It is helpful to note that the truncation is most effective if only a small fraction
of the data are affected.
Question #104. Posted 15 August 06. From a colleague: "How can you say that making accurate measurements
of the proportions of constituents is NOT the main goal of mixture analysis? I thought that was the whole point."
Reply to #104. Your view is supported by a lot of the literature on mixture analysis. And the search is still
on in the remote-sensing community for deeper mathematical insight, better analytical methods and better
image-processing algorithms. However, as judged by the published literature, remarkably few of the studies
of spectral mixing in images actually demonstrate a need for high accuracy of endmember fractions.
Furthermore, fractions rarely are normalized for shade. For years we worked hard to improve fraction accuracy,
also thinking that that was the main objective. What we learned from many days in the field was that each time we
got more accurate results for one component of a scene (endmember) we lost coherence with some other part.
Of course, if we used the right endmembers and had a simple scene we got high accuracy. For most images of
complex natural surfaces, though, we had to settle for uncertainty in the fractions. We came to realize
more and more that the patterns of endmember fractions in the images were of great value for interpreting materials
and processes on the ground, providing that close attention was paid to the image context. This is why we caution
about getting obsessive about fraction accuracy, and instead emphasize how SMA can be a powerful tool for
dissecting the spectral information in an image.
Comment #101. Posted 15 August 06. "The cover is beautiful and the layout is great."
Comment #102. Posted 15 August 06. "It is very dense."
We are not aware of any published reviews of the book at this time. If you find one, please let us know!
We will post reviews as they become available.
These are currently being added.
Download Synthetic Image as text file.
The image dimension is 20x20x4.
Download the Synthetic Image with G endmember as text file.
The image dimension is 20x20x5.
Page 311: reference should be 1997, not 1994
Page 355: correct reference is:
Strahler, A.H. (1997). Vegetation canopy reflectance modeling: recent developments and remote sensing perspectives.
Remote Sensing Reviews 15, 179-194.