Open AccessRetrieval of the spectral diffuse attenuation coefficient K d ( λ ) in open and coastal ocean waters using a neural network inversion
Author(s) -
Jamet C.,
Loisel H.,
Dessailly D.
Publication year - 2012
Publication title -
journal of geophysical research: oceans
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.67
H-Index - 298
eISSN - 2156-2202
pISSN - 0148-0227
DOI - 10.1029/2012jc008076
Subject(s) - seawifs , remote sensing , attenuation , atmospheric correction , inversion (geology) , ocean color , artificial neural network , attenuation coefficient , wavelength , satellite , radiometry , environmental science , geology , computer science , physics , optics , artificial intelligence , phytoplankton , chemistry , paleontology , organic chemistry , structural basin , astronomy , nutrient
The diffuse attenuation coefficient, K d ( λ ) is a fundamental radiometric parameter that is used to assess the light availability in the water column. A neural network approach is developed to assess K d ( λ ) at any visible wavelengths from the remote sensing reflectances as measured by the SeaWiFS satellite sensor. The neural network (NN) inversion is trained using a combination of simulated and in‐situ data sets covering a broad range of K d ( λ ), between 0.0073 m −1 at 412 nm and 12.41 m −1 at 510 nm. The performance of the retrieval is evaluated against two data sets, one consisting of mainly synthetic data while the other one contains in‐situ data only and is compared to those obtained with previous published empirical (NASA, Morel and Maritorena (2001) and Zhang and Fell (2007)) and semi‐analytical (Lee et al., 2005b) algorithms. On the in‐situ data set from the COASTLOOC campaign, the retrieval accuracy of the present algorithm is quite similar to published algorithms for oligotrophic and mesotrophic ocean waters. But for Kd(490) > 0.25 m −1 , the NN approach allows to retrieve Kd(490) with a much better accuracy than the four other methods. The results are consistent when compared with other SeaWiFS wavelengths. This new inversion is as suitable in the open ocean waters as in the turbid waters. The work here is straightforwardly applicable to the MERIS sensor and with few changes to the MODIS‐AQUA sensor. The algorithm in matlab and C code is provided as auxiliary material.