Yongzhen Fan is the author of the following paper:
https://www.sciencedirect.com/science/article/pii/S0034425717303310
Atmospheric correction over coastal waters using multilayer neural networks
He's also contributing to the IOCCG report coming the spring of 2019
Q (30. August 2018)
Dear Yongzhen Fan,
I have currently come across your very interesting paper:
Atmospheric correction over coastal waters using multilayer neural networks
It is of great interest to me as I am currently working on the processing chain of a hyperspectral ocean observing small satellite project:
https://www.ntnu.edu/ie/smallsat
https://www.ntnu.edu/ie/smallsat
I want to see if the approach proposed in your paper can be suitable for our application.
My first and foremost request is regarding the training data used.
Is this data made available at any point?
Not only the simulated data, but also other types of data.
Not only the simulated data, but also other types of data.
Furthermore,
is any kind of documentation is available?
The code generated?
The SeaDAS NIR alogrithms (old and new?)
The code generated?
The SeaDAS NIR alogrithms (old and new?)
I know that these questions are not very specific,
but any kind of help will be greatly appreciated!
best regards,
Sivert Bakken
Sivert Bakken
A(31. August 2018)
Dear Sivert Bakken,
Thank you for interested in our algorithm. The small satellite project looks very interesting and with a hyper-spectrometer, the system should be capable of providing not only HAB warnings but also what types of algae is blooming.
Our AC algorithm has been applied to many ocean color sensors, such as SeaWiFS, MODIS, VIIRS, GOCI, Sentinel-3 OLCI, and we are working on other sensors as well. I believe our algorithm is suitable for your hyperspectral instrument. Do yo have more detailed description or specification of the HSI instrument? I could not find those information online. Once we have the detailed information, we can make more careful assessment.
The training dataset we use were created by the coupled radiative transfer model that we developed over the years. Since each instruments has its unique band configuration, bandwidth, relative response functions, signal to noise ratio, etc., the training dataset for each sensor is specifically tailored for that sensor. There is no general dataset that is suitable for different sensor, so we haven’t made these data available yet. However, we have created validation dataset for the new IOCCG report on several ocean color sensors, these data should be made available when the report is released next spring.
I am not sure what kind of document you are looking for, but I think my paper describes our algorithm quite well, including what kind of atmospheric model, aerosol models and ocean IOP models we used in our radiative transfer simulation; how we create the training dataset and how we select the input parameters for our simulation. For the training algorithm, there are numerus existing packages available in Python, Matlab, Tensor Flow, etc.
We used SeaDAS for pre-processing of the satellite data. But now we are integrating our algorithm to the EAS SNAP platform, we are making our own plugins, so that our algorithm will be running independently. We already have a prototype for the Sentinel-3 OLCI sensor, other plugins are under development.
Hope this answers your questions and please feel free to contact me if you have other questions.
Best,
Yongzhen Fan