Share Email Print
cover

Proceedings Paper

Sub-pixel mapping of water boundaries using pixel swapping algorithm (case study: Tagliamento River, Italy)
Author(s): Milad Niroumand-Jadidi; Alfonso Vitti
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Taking the advantages of remotely sensed data for mapping and monitoring of water boundaries is of particular importance in many different management and conservation activities. Imagery data are classified using automatic techniques to produce maps entering the water bodies’ analysis chain in several and different points. Very commonly, medium or coarse spatial resolution imagery is used in studies of large water bodies. Data of this kind is affected by the presence of mixed pixels leading to very outstanding problems, in particular when dealing with boundary pixels. A considerable amount of uncertainty inescapably occurs when conventional hard classifiers (e.g., maximum likelihood) are applied on mixed pixels.

In this study, Linear Spectral Mixture Model (LSMM) is used to estimate the proportion of water in boundary pixels. Firstly by applying an unsupervised clustering, the water body is identified approximately and a buffer area considered ensuring the selection of entire boundary pixels. Then LSMM is applied on this buffer region to estimate the fractional maps. However, resultant output of LSMM does not provide a sub-pixel map corresponding to water abundances. To tackle with this problem, Pixel Swapping (PS) algorithm is used to allocate sub-pixels within mixed pixels in such a way to maximize the spatial proximity of sub-pixels and pixels in the neighborhood.

The water area of two segments of Tagliamento River (Italy) are mapped in sub-pixel resolution (10m) using a 30m Landsat image. To evaluate the proficiency of the proposed approach for sub-pixel boundary mapping, the image is also classified using a conventional hard classifier. A high resolution image of the same area is also classified and used as a reference for accuracy assessment. According to the results, sub-pixel map shows in average about 8 percent higher overall accuracy than hard classification and fits very well in the boundaries with the reference map.

Paper Details

Date Published: 14 October 2015
PDF: 7 pages
Proc. SPIE 9638, Remote Sensing of the Ocean, Sea Ice, Coastal Waters, and Large Water Regions 2015, 96380G (14 October 2015); doi: 10.1117/12.2194579
Show Author Affiliations
Milad Niroumand-Jadidi, Univ. degli Studi di Trento (Italy)
Alfonso Vitti, Univ. degli Studi di Trento (Italy)


Published in SPIE Proceedings Vol. 9638:
Remote Sensing of the Ocean, Sea Ice, Coastal Waters, and Large Water Regions 2015
Charles R. Bostater; Stelios P. Mertikas; Xavier Neyt, Editor(s)

© SPIE. Terms of Use
Back to Top