Independent Component Analysis (ICA) performance to bathymetric estimation using high resolution satellite data in an estuarine environment

• Type: Article in International Conference Proceedings Book
• Date: 2014

Authors

Teodoro, AC (Author) FCUP; Rute Almeida (Author) FCUP; Goncalves, M (Author) Other

Conference proceedings International

Title: REMOTE SENSING FOR AGRICULTURE, ECOSYSTEMS, AND HYDROLOGY XVI

Conference on Remote Sensing for Agriculture, Ecosystems, and Hydrology XVI was part of the International Symposium on Remote Sensing

Amsterdam, NETHERLANDS, SEP 22-25, 2014

Indexing

ISI Proceedings

ISI Web of Knowledge - 0 Citations

Scopus - 0 Citations

Scientific classification

FOS: Engineering and technology > Environmental engineering

CORDIS: Technological sciences > Technology

Other information

Authenticus ID: P-00A-5ST

DOI: 10.1117/12.2067196

Abstract (EN): The use of satellite remote sensing data is a valid alternative to the classical survey bathymetric methods for bathymetric estimation in shallow waters. Multispectral satellite data has been used to produce bathymetric maps by considering the pixel reflectance as a depth indicator. Teodoro et al., (2010) already proposes a model for the estimation of depth based on Principal Component Analysis (PCA) of an IKONOS-2 image, for the Douro River estuary (Porto, Portugal). In this work, alternative univariate and bivariate models are proposed for the same IKONOS-2 image based on PCA and Independent Component Analysis (ICA). The PCA is the standard method for separating mixed signals. Such analysis provides signals that are linearly uncorrelated. Although the separated signals are uncorrelated they could still be depended, i.e., nonlinear correlation remains. The ICA was developed to investigate such data. Fast ICA algorithm was used in Matlab (R). The results obtained were compared with the bathymetric estimation trough PCA. Best univariate ICA based model allowed to estimate depth with a mean error of 0.00m [with 1.15 of standard deviation], outperforming the best PCA based univariate model results of 0.39[1.34], even with the first PCA component explains 80% of data variance. With bivariate models is possible to reduce the standard deviation of the error to 1.01m.

Language: English

Type (Professor's evaluation): Scientific