GEO/EVS 425/525 Exercise 14:Dark Object Subtracting and Band Ratioing

This exercise deals with three issues that have achieved prominence in remote sensing: dark-objectsubtraction, simple band ratios, and indices formed from ratios involving more than one band.

Dark-Object Subtraction

The dark object in a satellite image represents the area of zero reflectance below the pixel with thelowest reflectance values in the image. It represents a number of things, including shadow, scattering,and electrical gain in the sensor. Often, especially when we are planning to use band ratios, it isimportant to eliminate the dark object from our image. In this laboratory, you will use the spatial modelerto develop a general program to [1] subtract the dark object, [2] take band ratios of the more significantbands in your image, and [3] compare an empirical dark-object-subtracted band ratio to empirical reality.

You should carry out this exercise using your Thematic Mapper quadrangle image – and you should planto concentrate especially on areas of water! Bring your image up on your viewer. Using ImageInfo, lookat the histograms of the 6 or 7 bands comprising it. Verify that the dark object is greatest in band 1 andless in the other reflected bands, and that it may be lacking altogether in bands 5 and 7. In the imageyou have chosen, are there pixels with zero value for band 1?

Write a program using the spatial modeler that subtracts the dark object. The program will include inputand output rasters connected by a function of the form {image} - {Global Minimum(Image)} (if there areno pixels of zero value) or {image} - {Global Minimum(Image) Ignore (0)} (if there are background pixelsassigned zero value; note that this will work even if there are no pixels of zero value). Compare theinitial image with your dark-object-subtracted image as it appears in the viewer. Do you see adifference? Why is this? Now compare the histograms of the two images using ImageInfo. Do you seea difference?

Simple Band Ratios

The logic of band ratios is based on the fact that all materials reflect much more intensely in some bandsthan in others. For example, chlorophyll absorbs visible light rather broadly, but the greatest absorptionis in the red and blue bands, so that relatively more green is reflected – That’s why leaves are generallygreen in color. However, vegetation reflects even more highly in the near infrared spectral range, whereit doesn’t absorb at all. Consider which TM bands reflect most highly in areas of healthy vegetation andwhich reflect least. A ratio of the former to the latter would be very high in areas of healthy vegetationand low in barren areas.

Write a program using spatial modeler that makes a ratio of the two bands. Again, the program will haveinput and output rasters connected by a function of the form {image band for high reflectance} / {imageband for low reflectance}. When you’ve run the model, look critically at your results. Do they makesense? Does it make a difference whether you use the dark-object-subtracted image (you should usethat one!) or the raw image?

It is also possible to distinguish various minerals by their relative reflection in different TM bands. Forexample, iron oxide reflects much more strongly in band 3 than in band 1, so that a Band-3/Band-1 ratioprovides a mechanism for searching out iron oxides. Other ferrous minerals can be found using a Band-5/Band-4 ratio. Clay minerals reflect more strongly in band 5 than in band 7, so that a Band-5/Band-7ratio is a means of finding them. These indices can be found in ERDAS Imagine by clicking on ImageInterpreter -> Spectral Enhancement -> Indices. Try running all of them (note, you can do this in onepass using the “Mineral Composite” index. Do the results make sense? You should include printoutsof each of the band-ratio images produced in this section in your portfolio.

More Complex Ratios

You may have noticed that the ratios you evaluated in the previous section were unbounded – that isthey had no set minimum or maximum, and it can be difficult to interpret the output. One strategy that isoften used – especially with vegetation analysis is to use a slightly more complex index which isbounded and which can be more easily interpreted.

Vegetation analysis is seldom based on the simple Band-4/Band-3 ratio you (should have) used above.The Normalized Difference Vegetation Index (NDVI) is much more commonly used – indeed it’s one ofthe most common analytical tools in the remote-sensing toolbox. It is simply [Band-4 - Band-3]/[Band-4+ Band-3]. To see the utility of the index, consider the situation in which vegetation is very healthy andshows minimal reflection in band 3. The index will approach 1. Conversely, if near-infrared absorptionis high, so that band-4 reflection is minimized, the index will approach -1. Write a program using spatialmodeler to calculate NDVI, and run it for your dark-object-subtracted image. Remember that you willhave to define (as 0) the case in which both band 3 and band 4 have the value of 0 (so that you don’tdivide by 0) using an “Either” conditional statement. Does the result make sense? You should includeyour NDVI image in your portfolio.

Questions to Consider

1. How would you evaluate the relative utility of classification and band ratios as tools to extractinformation from satellite imagery?

2. Is a more complex ratio such as NDVI always more useful than a simple ratio?

3. Why are the spatial-model forms of the indices included in ERDAS Imagine more complex thanthe form of the models you wrote in this exercise?

Portfolio

1. Image of your quadrangle using simple vegetation band ratio

2. Image of your quadrangle using simple iron oxide band ratio

3. Image of your quadrangle using simple ferrous mineral band ratio

4. Image of your quadrangle using simple clay mineral band ratio

5. Image of your quadrangle using NDVI.