about input data for IOP model(QAA and GIOP)

[jiwoni-ind]

Hi, Smith

Thank you for sharing the source code for IOP algorithms(QAA and GIOP).
This is very useful for using sensor data not provided by OCSSW.

I have one question for using this.
First of all, I want to use this calculate IOP using remote sensing reflectance from GOCI-II.
However, I keep getting various errors.

So I'd like to know if you have any data you used when testing this code.
Or could I know information such as the format of the input data?
For reference, I applied IOP using the get_tile_data function of your utils library.

I hope you will be able to guide me through.

Best regards
Ji-Won

[lifeodyssey]

Hi,

I have used him code. So maybe I can help you this question.
For example, in the QAA.py you can saw the input of QAA function.
You can add the band of GOCI-II and use it as a input.

By the way, did the GOCI-II data open to public now？

[BrandonSmithJ]

@jiwoni-ind I'm assuming this is referring to the MDN repository, and not this one on band-adjustment. In which case, both QAA and GIOP can be used by passing in the Rrs (as a numpy array shaped [number of samples, number of bands]), and the wavelengths (as a numpy array shaped [number of bands]).

GIOP additionally needs the sensor name passed in, e.g. OLI for Landsat-8. The problem in your case is, GIOP needs an estimate of chlorophyll to work, which is calculated via OC3. There is no GOCI-II definition for OC3, and so you will not be able to use GIOP for that sensor unless the OC3 definition is included.

To use get_tile_data with these functions, you'll need to reshape the input correctly first - get_tile_data returns Rrs in the shape of [height, width, number of bands], whereas you need [number of samples, number of bands]. Once Rrs is reshaped, you should be able to pass it into the QAA function (along with the bands) without issue.

Let me know if this answers your question.

[jiwoni-ind]

Hi,

I have used him code. So maybe I can help you this question. For example, in the QAA.py you can saw the input of QAA function. You can add the band of GOCI-II and use it as a input.

By the way, did the GOCI-II data open to public now？

@lifeodyssey

Thank you for your willingness to help.

Currently, GOCI-II data are provided by the Korea National Oceanic and Atmospheric Research Institute.
Here is the link: (I don't know if access is possible.)
http://www.khoa.go.kr/nosc/satellite/searchL2.do

If access is not possible, please help with GOCI data.
GOCI data is provided by Ocean Color Data web. (see link below)
https://oceandata.sci.gsfc.nasa.gov/directaccess/GOCI/

First, added information from meta.py GOCI-II.

• SENSOR_LABEL = 'GC2' : 'GOCI-2'
• SENSOR_BANDS = 'GC2' : [380, 412, 443, 490, 510, 555, 620, 660, 680, 709, 745, 865]

When using QAA.py, the code is executed as follows.

import utils2, QAA
import numpy as np

fname = "D:/data/GK2_GC2/LA_L2/AC/20210809/GK2B_GOCI2_L2_20210809_031530_LA_S007_AC.nc"
#fname = "D:/data/COMS/L2/G2021054001644.L2_COMS_OC.nc"
data = utils2.get_tile_data(filenames=fname, sensor='GOCI-2', allow_neg=True, rhos=False, anc=False)
wavelengths = data[0]
Rrs = data[1]
qaa = QAA.model(Rrs, wavelengths)

Also, when using GIOP.py, the code is executed as follows.

import utils2, GIOP
import numpy as np

fname = "D:/data/GK2_GC2/LA_L2/AC/20210809/GK2B_GOCI2_L2_20210809_031530_LA_S007_AC.nc"
data = utils2.get_tile_data(filenames=fname, sensor='GOCI-2', allow_neg=True, rhos=False, anc=False)
wavelengths = data[0]
Rrs = data[1]
sensor = 'GOCI-2'

giop = GIOP.model(Rrs, wavelengths, sensor)

[jiwoni-ind]

@jiwoni-ind I'm assuming this is referring to the MDN repository, and not this one on band-adjustment. In which case, both QAA and GIOP can be used by passing in the Rrs (as a numpy array shaped [number of samples, number of bands]), and the wavelengths (as a numpy array shaped [number of bands]).

GIOP additionally needs the sensor name passed in, e.g. OLI for Landsat-8. The problem in your case is, GIOP needs an estimate of chlorophyll to work, which is calculated via OC3. There is no GOCI-II definition for OC3, and so you will not be able to use GIOP for that sensor unless the OC3 definition is included.

To use get_tile_data with these functions, you'll need to reshape the input correctly first - get_tile_data returns Rrs in the shape of [height, width, number of bands], whereas you need [number of samples, number of bands]. Once Rrs is reshaped, you should be able to pass it into the QAA function (along with the bands) without issue.

Let me know if this answers your question.

@BrandonSmithJ

Thank you. I noticed the problem with my source code in your reply.

The shape of the data read using get_tile_data is [2780, 2780, 12] as [number of row, number of col, number of bands].
I think the problem was that I used it as an input value for QAA and GIOP.

When get_tile_data is used, the type of the return value is a (2,) tuple.
The first index of the tuple contains bands.
In the case of GOCI-II, [380, 412, 443, 490, 510, 555, 620, 660, 680, 709, 745, 865] is returned.
The second index contains the value of Rrs for each band, and is a three-dimensional numpy array. (As mentioned above, it is in the form of (2780,2780,12).)
I used these two return values as input values (wavelengths, Rrs) for QAA and GIOP.

But I have one question. The QAA function requires Rrs, wavelengths.
As you said, Rrs needs to be reshape such as [number of samples, number of bands]. What does sample count mean here?

[BrandonSmithJ]

Sample count is just the number of (in this case) pixels you want estimates for. So ignoring any masking applied to the pixels, you would have height * width = 2780 * 2780 = 7,728,400 samples. Reshaping the Rrs array via e.g. Rrs.reshape((-1, Rrs.shape[-1])) and feeding that into QAA should get you the results you expect. The results can then be reshaped back into the image shape if that's the format you need.