Lab 7 Julia Janicki. Introduction and methods

Transcription

1 Lab 7 Julia Janicki Introduction and methods The purpose of the lab is to map flood extent after a flooding event that occurred in Houston, Texas. Two Sentinel-1 images with C-band wavelength were used for this lab, one was taken on July 24-31, 2017 and represents normal conditions, the other was taken after the flooding event on August 29-30, The raw SAR intensity data were processed using SNAP to create ready-to-use products of a subset of the scene by applying radiometric correction, Lee Sigma filter to reduce speckling, and terrain correction. The two processed and subsetted images (Figures 1 and 2) were then imported into ENVI in order to conduct a change detection. The two images were subtracted (Figure 3) by using the band math tool and the difference image was then density sliced with a threshold at -0.2 in order to produce the final image that displays the flooded areas, with the areas with the value of less than -0.2 indicating flooded areas and greater than -0.2 indicating normal areas. Results The result displaying flooded areas in an area of Houston can be seen in Figures 5 and 6, where the blue represents flooded areas. For this particular subsetted area, through visual interpretation it seems that at least ⅓ of the whole area is flooded. In particular, the areas along the two rivers are especially flooded, especially the area in between the two rivers. The reservoir near the bottom of the images was already filled with water in the normal image so it was shown in the final image as non-flooded. Discussion The result that displays flooded areas seems to be generally successful. Radar is good at providing information on physical structures and water content, which is good for the purpose of this lab. Water appears dark in radar images due to specular reflection since smooth surfaces such as water surfaces reflects energy away. Because of this property water is very easily identified in radar images and visual interpretation of flooded areas before and after a flooding event is not difficult. The final image showing the flooded areas seem to be mostly correct through visual interpretation of the before and after processed Sentinel-1 images, as well as comparing it to the image visualized using RGB multi-channel image display (Figure 4). The only area that is ambiguous is the reservoir, since it was already filled with water in the normal image, the density slicing change detection technique did not display that area as flooded, and it would be difficult to tell if it is in fact flooded or not using this this technique. Moreover, the final results seem to be relatively successful after comparing it with other sources such as the Hurricane Harvey flood map produced by the UC Davis Natural Hazards and Mitigation Group. In this map (Figure 7), similarly along both rivers and in particular between

2 the two rivers most areas are flooded. Also similarly the area to the west of the river that is towards the left end of the image are not flooded. However, in this map more areas seem to be flooded compared to the map produced for this lab, as much as about ½ of the image seem to be flooded. A different threshold could be tested for the density slicing method to see if the results improve, for example trying to move the threshold to a value closer to 0 such as 0.1. Alternatively, a different change detection technique could be applied instead, such as Support Vector Machine or other supervised classification algorithms, since these technique tend to be more accurate compared to density slicing.

3 Figures Figure 1. Processed Sentinel-1 image of the subsetted area in Houston, Texas before the flood, taken on July 24-31, 2017 at the C band ( cm), processed with the VV polarization, with 20 m (range) x 22 m (azimuth) resolution and 10 m pixel spacing.

4 Figure 2. Processed Sentinel-1 image of the subsetted area in Houston, Texas after the flood, taken on August 29-30, 2017 at the C band ( cm), processed with the VV polarization, with 20 m (range) x 22 m (azimuth) resolution and 10 m pixel spacing.

5 Figure 3. The difference image of the two processed Sentinel-1 images of the subsetted area in Houston, Texas taken on July 24-31, 2017 and August 29-30, The very dark and very bright areas in the difference image are the areas that have the most change between the two images, with the dark areas being flooded areas.

6 Figure 4. An image that visualizes the flooded area using RGB multi-channel image display. The red are from the normal conditions image while green and blue are from the post-flooding event image. Blue areas indicate flooding, and stable water bodies such as reservoirs are indicated by dark red and black.

7 Figure 5. The thematic map displaying the flooded areas in blue and normal areas in beige by thresholding the difference image near -0.2, with to -0.2 representing flooded areas and -0.2 to 27.7 representing dry areas.

8 Figure 6. The vectorized flooded areas created from the thematic map overlaid on top of the difference image.

9 Figure 7. The screenshot of the same general area of the final result in the Hurricane Harvey Water Extent map produced by the UC Davis Natural Hazards and Mitigation Group.

10 Figure 8. Composite image of the (from left to right, top to bottom): (1) processed normal image (2) processed post-flood image (3) difference image (4) thematic map displaying the flooded areas (5) flooded areas overlaid on top of the difference image (6) image that visualizes the flooded area using RGB multi-channel image display.