Development of Software to Locate Arrays of Spectral Signatures on the Land
Dave Weidling
Project Background and Objectives
   I propose to continue development of a software program that displays the population of spectral signatures found in a selected section of a Landsat Thematic Mapper image.  The software which I designed and call “Spectral Explorer”  is at an early “proof of concept” phase of development and is written in the Visual Basic language.  At this point it is capable of importing modified grid files of band information from ArcView, displaying them in a static and animated fashion, subselecting pixels whose 5 band spectral profile lies within an area of interest in “spectral space” and then writing out a mask file which can then be imported to Arcview where only the selected pixels will appear. 

  The program arose out of simple curiosity to see what a population of landscape spectral signatures might look like.  I wrote the software so that a single signature is a line on a chart, connecting six points in spectral space.  The program then steps through a selected region of a TM image displaying each signature in turn as a colored multi-jointed line and then the whole population as displayed as a polygon with approximately twelve nodes in the chart of spectral space (Plate 1).

   The original question I asked in writing the code was, “what places in our region are like the core of the Arcata City Forest, from the point of view of the Thematic Mapper?”The answer that the software generated appears to indicate that very few places exhibit the same range of spectral signatures as shown in Plate 2.  The interesting part is that by adjusting the boundary of the selected zone of spectral signatures just slightly, large, formerly non-selected areas of probable old-growth redwood in the southern Humboldt redwood parks appear.  Questions arise from this as to whether there is a real difference between the two populations of redwoods.

   I propose to rewrite the software in Java for this project, make it more graphical and easy to use and add an important new capability.  It is possible that in the future, people and machines are going to an increasing extent to be perceiving reality as multispectral because more information is derived from the same instance of viewing one’s surroundings that way.  As ccd devices with sensitivity outside of the normal human visible range become cheaper it becomes apparent that there is nearly free information being offered by nature at all times that we are not making full use of.  This was impressed upon me when in another project I was delighted to find that a NIR filter over a standard digital camera produced acceptable imagery. 

   If one allows this speculation as to the utility of a multispectral viewpoint, then arrays of multispectral signatures will surely tell even more than an individual multispectral signature.  By use of arrays I mean deriving increased understanding about the significance of a given spectral signature in geographic space from the presence and location of other similar or differing spectral signatures near it.  Here we are trying to develop a topology of spectral signatures.

 

   With this in mind I Propose to modify my software to allow a simple graphical querying of a TM image in the form of describing three spectral signatures and requesting, “tell me where riparian coastal vegetation is surrounded by pasture with both intersected by a road.”   The purpose of such an example query could be to determine likely spots for deer to be crossing a road while following a stream course from upland areas to the Arcata Bay.

   Java has been selected as the development language for this project because of its object orientation.  Analysis and visualization tools are becoming based on the object oriented model (Pundt and Brinkkotter-Runde, 1998) and data itself is being seen as a collection of objects (Camara et al, 1996). Java provides these approaches to be fully realized as it is widely known to be completely compliant to the object model of reality.  The results of this project can set the stage for development of other more powerful visualization tools due to the reusability of classes developed during this project, with future projects (Silver, 1995).  The available Java integrated development environments, one of which I intend to employ, are becoming quite rich in tools if not any easier to use and will provide my medium to work in for this project.

    This project deals with the Integration of multispectral image data with geographic information systems.  This teaming can have important uses in promoting human and ecological welfare. Researchers in the Mediterranean are using multispectral data to understand the variability in durum wheat yield and suggest the most appropriate places and methods of cultivation for this important food crop (Nieves et al, 2000).  A more dire example of the use of multispectral imagery impinging upon human life is the recent research into finding buried land mines at the Naval Surface Warfare Center in Florida.  Their results are considered “quite good and encouraging” with “probabilties of detection close to 1” (Clark et al, 2000).  After numerous field trips with my wife, a Forest Service botanist I am amazed how much remains unknown regarding plant populations.  with possible climate changes and ongoing invasions of exotic plants I hope that the exploration of plant and feature identification by using multispectral data might make some contribution to our ability to make informed decisions regarding the landscape.

Proposed Data Sources

  1. 1:24,000 USGS topographic map of the Arcata area, to be digitized.
  2. Landsat Thematic Mapper image of Northwest California
  3. Ground truth data from test sites in the immediate area verified by a Trimble GPS unit
  4. Target locations derived from a Java-based version of Spectral Explorer software

Timeline

The proposed timeline for this project is as follows:

Anticipated Products and Outcomes

The primary product will be a Java application that will be used as a tool to explore multispectral space.  The most exciting outcome will be the ability to specify simple spatial arrays of spectral signatures and find them in the multispectral image.  Location data pertaining to those pixels of interest will either be computed in the developed software or back in ArcView.  At that point field verification of the results will be attempted.  The desired outcome will be to go to a location and see if the query has produced a result that makes sense on the ground.  Along with this field checking will come a statistical analysis of the success of the exercise. 

References

Bhanu, B.,  P. Symosek, and D. Subhodev, 1995. Analysis of terrain using multispectral images, Pattern Recognition, 30(2):197-215.

Camara, G., R. Souza, U. Freitas and J. Garrido, 1996.  Spring: integrating remote sensing and GIS object-oriented data modelling, Computers and Graphics, 20(3):395-403.

Clark, G., S. Sengupta, W. Aimonetti, F. Roeske and J. Donetti, 2000.  Multispectral image feature selection for land mine detection, IEEE Transactions on Geoscience and Remote Sensing, 38(1):304-311.

Nieves, A., D. Villegas, J. Casadesus, J. Araus and C Royo, 2000. Spectral vegetation indces as nondestructive tools for determining durum wheat yield, Agronomy Journal, 92(1):83-91

Pundt H. and K. Brinkkotter-Runde, 1998. Visualization of spatial data for field based GIS, Computers and Geosciences 26(2000):51-56.

Silver, D., 1995. Object-oriented visualization, IEEE Computer Graphics and Applications 22(3):54-62.

 

Plates


 Plate 1.  Spectral Explorer (version 25)  The red area is a display of one of the six bandsimported on the upper left.  The white area on the left would display the raw integer values of a given band if the file was was smaller than the 309 x 279 values indicated at upper left. The chart at the lower right is a representation of the population of pixels imported and the yellow line is the spectral signature of a single pixel.  The twelve sliders are used to delineate a zone of interest within the total population of signatures.  Note that this is experimental software, not a finished product.

 

Plate 2.  Results of use of Spectral Explorer to search for exact matches to the spectral signature of what I arbitrarily determined to be the “core” of the Arcata City Forest.  The large darker green area is the mask created by Spectral Explorer with its successful matches highlighted in bright green.  A large green clump at upper center is the forest itself.  The bright reddish area at the borders is the underlying Landsat TM image.

 

Plate 3.  Applying the technique used in plate two led to few results outside the immediate Arcata area, but minor adjustment of signature matching parameters revealed what is probably the old growth redwoods found in the series of state parks adjacent to highway 101 in southern Humboldt County.  These are highlighted in yellow.  Interestingly, these are a tiny subset of all the forest vegetation in the area.