You are in:Home/Publications/Determination of Shallow Water Depths using Inverse Probability Weighted Interpolation: a hybrid system-based method

Dr. mahmoudgoma :: Publications:

Determination of Shallow Water Depths using Inverse Probability Weighted Interpolation: a hybrid system-based method
Authors: Mahmoud Salah
Year: 2016
Keywords: Naïve Bayesian, Multilayer Perceptron, Fuzzy Majority Voting, bathymetry, Landsat-8, reflectance, echo-sounding.
Journal: International Journal of Geoinformatics
Volume: 12
Issue: 1
Pages: 45-55
Publisher: Not Available
Local/International: International
Paper Link: Not Available
Full paper Not Available
Supplementary materials Not Available

A method for the determination of bathymetry from the new and freely available 11-band Landsat-8 multispectral satellite imagery and sample depth measurements has been proposed. The method starts with a few reference points, where both reflectance and water depths are known. Several preprocessing operations were done before depth estimation. First, the image was geometrically and atmospherically corrected and then segmented into land and water. Second, Lansat-8 blue band was converted to reflectance. Third, two different algorithms were used to assign each pixel to each of the reference points. The algorithms used include: Naïve Bayesian (NB) as a statistical model; and Multilayer Perceptron (MLP) as a neural network model, which offer complementary information. Outputs are probability images corresponding to each known depth. In order to achieve a robust decision about the obtained probabilities, the Fuzzy Majority Voting (VMV) algorithm was then applied for combining measures of probability from the two algorithms. Finally, the water depths were derived from the combined probabilities based on an inverse probability weighted interpolation technique (IPWI). The proposed method enabled the retrieval of water depths of less than 5 m at a relatively high level of accuracy, 0.19 m. However, accuracy further decreases for the water depths of more than 10 m.

Google ScholarAcdemia.eduResearch GateLinkedinFacebookTwitterGoogle PlusYoutubeWordpressInstagramMendeleyZoteroEvernoteORCIDScopus