Seawater-type based neural networks for Ocean Color data inversion

Ari Saptawijaya, Davide D'Alimonte, Tamito Kajiyama

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The retrieval of Ocean Color (OC) data products is here investigated by using Multi Layer Perceptron (MLP) neural nets. Synthetic data have been generated for this scope with a forward OC model. These samples have then been used to train and assess the MLP performance considering different seawater types (WTs) with optical properties driven by: chlorophyll (Chl-a), colored dissolved organic matter (CDOM), and non-pigmented particulate matter (NPPM), as well as a mixture of Chl-a, CDOM and NPPM (denoted MIXI). Acknowledging that MLP classification results represent WT posterior probabilities, an integrated machine learning approach is set up by joining MLPs for data regression and classification in a composite scheme. Results indicate that this approach is valuable to support the use of regional ocean color inversion schemes by decomposing the overall challenge in sub-components, optimally addressing each of them, and combining the individual solutions in a principled framework.

Original languageEnglish
Title of host publication34th Asian Conference on Remote Sensing 2013, ACRS 2013
PublisherAsian Association on Remote Sensing
Pages1897-1904
Number of pages8
ISBN (Print)9781629939100
Publication statusPublished - 1 Jan 2013
Event34th Asian Conference on Remote Sensing 2013, ACRS 2013 - Bali, Indonesia
Duration: 20 Oct 201324 Oct 2013

Publication series

Name34th Asian Conference on Remote Sensing 2013, ACRS 2013
Volume2

Conference

Conference34th Asian Conference on Remote Sensing 2013, ACRS 2013
Country/TerritoryIndonesia
CityBali
Period20/10/1324/10/13

Keywords

  • Neural network
  • Ocean color
  • Regional algorithms
  • Remote sensing

Fingerprint

Dive into the research topics of 'Seawater-type based neural networks for Ocean Color data inversion'. Together they form a unique fingerprint.

Cite this