Dear Bilal,

In your data (MODIS) spatial resolution and coordinate system are same, so problem in GCP's, location accuracy problem. I think if you change GCP's in 2nd image and do georegistration again based on 1st image (take GCP's from this image), it will match.  Hope it will be helpful to you. Thanks.

Dear Mohammad,

If there is discrepancy in paths or rows or angle is a little different or atmospheric conditions are not same, such a problem arises, therefore, geometric correction is considered suitable enough to overcome this problem.

You can pick a few stable features as control points and apply an affine transformation to refine the relative registration of both images. You will need at least 3 non-collinear points well distributed throughout the image. 

Hope this helps.

Dear Dr Hernán De Angelis,

Both have the correct geolocation information because I have hundreds of images with same problem. 

Dear Dr Firoz Khan

Both the images have same dimension (columns and rows) and I used MODIS Conversion Toolkit (MCTK) to perform geometric correction. 

Dear Dr Sayjro,

Thanks for your suggestion - as my data has resolution of 500 m and also have hundreds of images so I think it will be difficult task to perform affine transformation in batch processing. what do you think?

Dear Brother Alam,

Thanks for your input but these links look not suitable for mentioned problem.

Dear Muhammed,

as Sayjro said you can use ground control points for image to image registration. ENVi provides for your problem an automatic registration function - which works quite well. If you write a short IDL skript you can also do hundreds of registrations automatically. Hope this helps you!!!

Greetings

Dear Mohammad

The affine transformation is the geometric transformation. Using certain plugins and software you can do it in bulk and quickly.

http://fiji.sc/Interactive_Affine

http://www.imagescience.org/meijering/software/transformj/

Dear Ronny Richter,

I have wrote an IDL code based on "ENVI_register_doit" but  I have some problem as I am not getting output file. If possible then I can send you to have a look on it to find my mistake.

Thank you.

Dear Dr Firoz Khan

Thanks for providing these useful links - I will try to work on them.

Thank you.

Dear Brother, these images are of the same day and time, the only difference is that these two are different data products of MODIS.

If both images use same spatial resolution and coordinate system, check the coordinates of upper-left corner of both images,  divide coordinate differences by spatial resolution, should can solve the problem. 

If you didn't keep coordinates of upper-left corner  in images, check the map projection parameters you used.

Dear Xianjun Hao,

Yes, both the images do not have same UL and LR coordinates but now I used CONGRID function in IDL to get same UL and LR coordinates. Now both images have the same coordinates but the shift is still persistent. 

That's strange.  Could you tell me which two products you used?  If you see  significant shift, should be some issues in map projection.

MYD02HKM and MYD04 cloud mask at 500 m resolution!

Yes, there is a significant shift.  I found few IDL codes to align pixels of two images but I don't know how to use them as I am not expert in IDL.

I see.  MYD02HKM and MYD04 are not gridded data. The spatial resolution of radiance  is 500m (nadir), but geolocation is 1km in the  MYD02HKM  file. For MYD04_L2 file, you used mask at 500m resolution, but the spatial resolution of geolocation data in MYD04_L2 data file is 10km.   If you use tools to process the two file separately, may cause spatial differences because of interpolation at different resolutions. 

1. If you just need the data, don't need geolocation to map the data, it is simple, reflectance data and the mask data are exactly the same size and resolution (4060*2708),  can show them by row vs col.

2. If you need to use geolocation to map the data at 500m resolution,  can try the following steps:

1)  Get 500m geolocation data through spatial interpolation of 1km geolocation data swath by swath ( per 20 rows of 500m data).

2) Define a 500m grid,  do spatial resampling with  500m geolocation data for both reflectance data and mask data (nearest neighbor) on the grid.

Dear Xianjun Hao, 

Thanks for understanding about the issue and providing the solutions - I have the same issue with MYD35 cloud mask as well. I use MCTK to process these datasets. Actually I want to mask clouds present in the radiance data and also want to map clouds as a separate image- could you please guide me in detail how to process these datasets to mask cloud cover pixels using cloud mask from either MYD04 or MYD35. I normally use ENIV/IDL for processing. If possible then please help me to solve this issue by sending exact steps at my email: [email protected]

Thank you.

I usually use Matlab, don't have IDL/ENVI now. If you need to grid the datasets, you can use MODIS MS2GT tool from http://nsidc.org/data/modis/ms2gt/

This maybe the most convenient approach, convert the two data sets  to the same grid, can visualize it conveniently with grid and projection information.

Without this tool,  you need to write some code. Since I am not familiar with IDL, can't help much.  I will try to find a document and send to you by email.

Para alinear de manera correcta pixeles de imágenes yo hago el siguiente proceso:

1.- Descargo las imágenes, y reviso si están corregidas geométrica, ortogonal y radiometricamente

2.- posteriormente con un software como erdas corto las imágenes por coordenadas

3.- reproyecto las imágenes y les fijo a ambas las mismas coordenadas de esquinas, y numero de filas - columnas...

saludos. 

Dear Xianjun Hao,

I prefer to solve this problem for swath data as I have a lot of processed data based on MYD02HKM swath product. Is there any other way to solve this problem without converting swath data into grid data?

If you prefer swath data,  it is simple. 

1.  Calculate geolocation data at 500m resolution

  I am not sure if you need geolocation data at 500m resolution.  If you need, can interpolate the 1km geolocation data to 500m. Please see attached file (page 11) for details.  Take a look of the attached document, you can get IDL code online for MODIS geolocation interpolation.

2.  At swath level, you use two data sets at the same resolution.

    Dimension of Radiance in MYD02HKM:   5*4060*2708

    Aerosol/cloud mask in MYD04:   4060*2708

   So, for a selected band, the radiance data and mask data are already aligned spatially.

  R(4060, 2708) mask(4060, 2708)

  You can do any calculations conveniently.  If you want to map them, just need to use 500m geolocation data from step 1.

Hope these information helpful.

Dear Dr Xianjun,

Many thanks for your understanding the problem and provide me the solution. I understand it and will work on it. I will let you know my progress. I really appreciated your contribution.

Regards

Bilal

Dear Bilal,

In your data (MODIS) spatial resolution and coordinate system are same, so problem in GCP's, location accuracy problem. I think if you change GCP's in 2nd image and do georegistration again based on 1st image (take GCP's from this image), it will match.  Hope it will be helpful to you. Thanks.

Dear Dr. Rahman,

Thanks for your input. At this time, I am working on the solution as suggested by Dr. Xianjun. I also would like to work on your suggestion but if could you please tell me in detail or steps then it will be great and easy for me to work on it. 

Thank you!

Dear Dr Xianjun Hao,

My problem to align the swath data is solved now.

Many thanks for your kind help.

Regards,

Bilal 

Great. 

you can use image registration in ENVI software. resampling is useful also in raster size but image registration, image-image will give you different options and in the registration workflow you can remove tie points with high error

Execute these steps

1.- reference the images between them .... you take one as a reference pattern and the other as a reference image

2.- cut the scene of the image so that the two cover similar geographical domains

3.- reproject the images assigning similar dimensions of rows and columns, corner coordinates (they are assumed to be the same as in the previous step), huso and datum .....

regards....

Dear just do co-registration between your two images

good luck