| | |  | An example of subtracting one of the filtered “dry” images from the filtered “flooded” image | |
Exercise 2: Change detection – band math
In this exercise, as well as in exercise 3, you will be using change detection techniques. These techniques allow you to obtain more information on the changes between the different images by analysing them together.
Using the available filtered images, you will first implement mathematical expressions. You will subtract the initial state (pre-flood) image from the final state (post-flood) image (i.e. final-initial). This technique will produce new images, characterising the differences between every pair of pre- and post-flood images.
Open 05102004_f.tif, 25102005_f.tif and 10102006_f.tif as a single dataset. Implement the following mathematical expressions:
- 10102006_f.tif – 05102004_f.tif. The resulting image represents the changes between 05102004 (pre-flood) and 10102006 (post-flood)
- 10102006_f.tif – 25102005_f.tif. The resulting image represents the changes between 25102005 (pre-flood) and 10102006 (post-flood)
| | | Final result of the subtraction of the average of dry images from the post-flood image |
In the resulting images, no or little changes are displayed in grey (medium backscatter values). Strong changes are either displayed as dark or bright areas, depending on whether the change has been from bright to dark or from dark to bright respectively. The bigger the backscattering differences between the corresponding pixels, the brighter or darker the resulting image will be.
This means that all pixels that became considerably darker in the post-flood image will be dark in the image resulting from subtraction, whereas pixels that became much brighter in the post-flood image will appear bright. Therefore the presence of new water-covered areas in the post-flood image (i.e. flooded areas), should produce dark pixels in the “subtraction image”.
Despite the fact that both pre-flood images are of the same season as the post-flood image, the presence of differences is inevitable. For that reason, the pre-flood images can be combined, in order to produce a new image, which will better represent a mean “dry” situation. This action will help minimise any differences between the pre-flood images (the more “dry” images available, the better the result) and subsequently render the subtraction result more reliable. Therefore, to implement an average dry image, apply the expression: (05102004_f.tif + 25102005_f.tif)/2 and save the resulting image as pre-flood_mean.tif.
Close everything you have opened. Now open 10102006_f.tif and pre-flood_mean.tif in one dataset and implement another subtraction, by using the post-flood image and the arithmetic mean of the pre-flood images: 10102006_f.tif – pre-flood_mean.tif. Save the resulting image as subtraction.tif
Alternatively you can produce the subtraction.tif image directly, without producing the pre-flood_mean.tif separately, by using the following expression: 10102006_f.tif – (05102004_f.tif + 25102005_f.tif)/2. The result will be the same. However, the pre-flood_mean.tif is necessary because you will need it for the next exercise.
1. Observe the subtraction.tif image and find the changes (very dark and very bright areas).
2. What do the dark areas represent?
3. Based on what you have learned so far about the interpretation of radar images and the dielectric properties of the soil, can you tell what the bright pixels around the flooded areas are? Bear in mind that the post-flood image was acquired two days after the inundation, which means that a considerable amount of water was probably already absorbed by the soil or discharged through the drainage network before the satellite overpass.
| |
