Artificial Neural Network (ANN) Based Inversion of Benthic Substrate Bottom Type and Bathymetry in Optically Shallow Waters - Initial Model Results
| 详细信息 | 推荐本文 | |
- 作者：P. V. Nagamani (1) pvnagamani@gmail.com
Prakash Chauhan (2)
Nivedita Sanwlani (2)
M. M. Ali (1) - 关键词：Benthic substrate – ; Neural networks – ; Inversion – ; Sea grass – ; Coral sand
- 刊名：Journal of the Indian Society of Remote Sensing
- 出版时间：March 2012
- 出版年：2012
- 期刊代码：74_0974-3006
- 类别：ear
- 卷：40
- 期：1
- 页码：137-143
- 数据来源：sp
摘要
Ocean-colour remote sensing in optically shallow waters is influenced by contribution from the water column depth as well as by the substrate type. Therefore, it is required to include the contribution from the water column and substrate bottom type for bathymetry estimation. In this report we demonstrate the use of Artificial Neural Network (ANN) based approach to spectrally distinguish various benthic bottom types and estimate depth of substrate bottom simultaneously in optically shallow waters. We have used in-water radiative transfer simulation modeling to generate simulated top-of-the-water column reflectance the four major benthic bottom types viz. sea grass, coral sand, green algae and red algae using Hydrolight simulation model. The simulated remote sensing reflectance, for the four benthic bottom types having benthic bottom depth up to 30 m were generated for moderately clear waters. A multi-layer perceptron (MLP) type neural network was trained using the simulated data. ANN based approach was used for classification of the benthic bottom type and simultaneous inversion of bathymetry. Simulated data was inverted to yield benthic bottom type classification with an accuracy of ~98%for the four benthic substrate types and the substrate depth were estimated with an error of 0%for sea grass, 1%for coral sand and 1–3%for green and red algae up to 25 m, whereas for substrate bottom deeper than 25 m depth the classification errors increased by 2–5%for three substrate bottom types except sea grass bottom type. The initial results are promising which needs validation using the in-situ measured remote sensing reflectance spectra for implementing further on satellite data.