Utility of Spatial Filtering Techniques in the Remote Sensing of Soil Erosion in the Sefid-Rud Reservoir Catchment in Iran

Disfani, Mohammad Najafi (1989) Utility of Spatial Filtering Techniques in the Remote Sensing of Soil Erosion in the Sefid-Rud Reservoir Catchment in Iran. PhD thesis, University of Glasgow.

Full text available as:
Download (16MB) | Preview


The objective of this study is to investigate the applicability of Landsat Thematic Mapper digital images assisted by computer analysis to the study of soil erosion. The study aims to identify the sources of sediment and areas of dissected land in the catchment basin of the Sefid Rud reservoir in northern Iran. First, histogram equalization is deliberately applied to the original band 3 to reduce the noise and unv/anted edges and lines in the dark tail of the histogram, mainly vegetation, and the light tail, the non-eroded areas, and also to improve the visual appearance of edges and lines on the processed image. The next step is high pass filtering, unlike the conventional edge detection technique in which the first step is low pass filtering. In this instance, the result of low pass filtering was that faint edges, evidence of the gullies, were removed and highly eroded areas appeared as non eroded areas. Therefore low pass filtering was replaced with high pass filtering, which highlighted faint edges and lines. The next step is detecting the edges and lines. When using the edge and line detecting technique for detecting dissected lands one needs to take into account that a gully might appear as two or three edges if its width is more than one pixel or as one line if it is just one pixel or less than one pixel in width on the Thematic Mapper image. Therefore an algorithm should be chosen which has the ability to detect both edges and lines. The existina edge and line detecting filters such as the Sobel , the Robert, compass, the Laplacian convolution masks and the directional line detecting technique were evaluated. The Sobel and the Robert operators were found to be powerful edge detecting techniques, but the Laplacian convolution mask was found to be the best for detecting the badland and gullied areas because it has the ability to detect faint edges as well as coarse edges. Not only does it detect both edges and lines, but it also gives stronger weight to the lines than the edges. Only edges and lines in gullied areas were of interest for detecting the dissected lands, but all other artificial and natural lines and edges were also detected. The result of applying the Laplacian function appears on the screen as black, white and gray pixels. The black pixels are non-eroded land, white pixels are eroded and gray pixels are transitional between eroded and non eroded. To change the transitional pixels to either eroded or non eroded and also for printing the image as hardcopy the thresholding function of IAX was applied to the edge detected image. In order to mask out the noise within the vegetated areas caused by edges of plots of different crops the vegetation index (VI) was added to the detected image. In the derived image black pixels are evidence of gullies and white pixels are non dissected lands. In this image it is possible to find out the relative proportion of dissected and non dissected land globally and / or within the regions of interest. Although it is possible to measure the proportions of dissected and non dissected land and they are also visually distinguishable, they have not been categorised so far. To provide a map with categories of dissection, the first step is to smooth the image. To obtain the smooth image a low pass filter was used. Two ways were tested for producing the map of dissected lands from the smoothed image. In the first method one of the strongest edge detecting techniques, the Sobel operator was used on the smoothed image of dissected lands. In the result boundaries were detected and eroded and non eroded areas outlined. In the second method for categorising the smoothed image, the density slicing function of IAX was used to split the dissected land into different levels of severity. We concluded that the second method gives a better result. It was found in previous work that among erosion features gullies are recognizable on Thematic Mapper data. Detection of gullies and gullied areas by means of classification, whether supervised or unsupervised, was not successful in this study area. We came to the conclusion that the application of a Laplacian mask on the enhanced band 3 image could detect dissected lands. When aerial photographs and Thematic Mapper data are compared, the advantage of aerial photographs was that gullies actively cutting headwards were detectable, but on the Thematic Mapper data distiguishing between active and non active gullies was impossible. Aerial photographs are a very good tool to detect all kinds of erosion features (sheet, rill, and gully), but in my study area applying this new method (DLDT) on Thematic Mapper data can provide as much detail of soil erosion as is included in previous soil erosion maps made from aerial photographs. The Sobel and the Robert operators were found to be very strong edge detectors, but the ability of the Laplacian convolution mask for detecting gullies was greater. (Abstract shortened by ProQuest.).

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Keywords: Remote sensing, Geographic information science and geodesy, Physical geography
Date of Award: 1989
Depositing User: Enlighten Team
Unique ID: glathesis:1989-77976
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 30 Jan 2020 15:45
Last Modified: 30 Jan 2020 15:45
URI: http://theses.gla.ac.uk/id/eprint/77976

Actions (login required)

View Item View Item


Downloads per month over past year