Derivative-Neural Spectroscopy for Hyperspectral Bathymetric Inversion*

*This article was part of the 2005 Nystrom Competition. A portion of this research was supported by a NASA Graduate Student Researchers Program (GSRP) fellowship, with John R. Jensen, University of South Carolina, and Richard Miller, NASA John C. Stennis Space Center. Curtiss Davis, Robert Leathers, and T. Valerie Downes performed Hydrolight simulations. K. Carder, C. Mazel, R. Zimmerman, H. Dierssen, E. Louchard, and R. Pamela Reid supplied field data. Zimmerman provided the HyperTSRB-derived data, with preprocessing by Dierssen. All data were made available under the auspices of the Office of Naval Research–funded Coastal Benthic Optical Properties (CoBOP) program. The author thanks the anonymous reviewers for their constructive remarks, which improved the quality and clarity of this article.

Abstract

Bathymetry is an important variable in scientific and operational applications. The research objectives in this study were to estimate bathymetry based on derivative reflectance spectra used as input to a multilayer perceptron artificial neural network (ANN) and to evaluate the efficacy of field and simulated training/testing data. ANNs were used to invert reflectance field data acquired in optically shallow coastal waters. Results indicate that for the simulation-based models, nonderivative spectra yielded more accurate bathymetry retrievals than the derivative spectra used as ANN input. However, for the empirical field-based models, derivative spectra were superior to nonderivative spectra as ANN input. This study identifies circumstances under which derivative spectra are useful in bathymetry estimation, and thus increases the likelihood of obtaining accurate inversions.

Key Words:

• artificial neural network
• bathymetry
• derivative
• hyperspectral
• remote sensing

Notes

Notes: ANN=artificial neural network; RMS=root-mean-square; 1stDERIV=first derivative, 2ndDERIV=second derivative, 3rdDERIV=third derivative, 1ON=one output neuron, and 6ON=six output neurons.

Notes: ANN=artificial neural network; RMS=root-mean-square; 1stDERIV=first derivative, 2ndDERIV=second derivative, 3rdDERIV=third derivative, and 1ON=one output neuron.

Notes: ANN=artificial neural network; RMSE=root-mean-square error; MAE=mean absolute error; 1stDERIV=first derivative, 2ndDERIV=second derivative, 3rdDERIV=third derivative, 1ON=one output neuron, and 6ON=six output neurons. Values in parentheses indicate results from the n=50 validation; ANN-HL values not in parentheses are for the n=738 validation.

Notes: ANN-HL denotes models developed using Hydrolight pseudodata; ANN-HTSRB denotes models developed using HyperTSRB data; 1ON=one output neuron, and 6ON=six output neurons. For ANN-HL trials, values in parentheses indicate results from the n=50 validation; ANN-HL values not in parentheses are for the n=738 validation. ANN=artificial neural network; RMSE=root-mean-square error; MAE=mean absolute error.

Notes: ANN=artificial neural network; PRESS=predicted residual error sum of squares; 1stDERIV=first derivative, 2ndDERIV=second derivative, 3rdDERIV=third derivative, 1ON=one output neuron, and 6ON=six output neurons.

*This article was part of the 2005 Nystrom Competition. A portion of this research was supported by a NASA Graduate Student Researchers Program (GSRP) fellowship, with John R. Jensen, University of South Carolina, and Richard Miller, NASA John C. Stennis Space Center. Curtiss Davis, Robert Leathers, and T. Valerie Downes performed Hydrolight simulations. K. Carder, C. Mazel, R. Zimmerman, H. Dierssen, E. Louchard, and R. Pamela Reid supplied field data. Zimmerman provided the HyperTSRB-derived data, with preprocessing by Dierssen. All data were made available under the auspices of the Office of Naval Research–funded Coastal Benthic Optical Properties (CoBOP) program. The author thanks the anonymous reviewers for their constructive remarks, which improved the quality and clarity of this article.