Hi Gang,

I'm running into some possible artifacts derived from projecting SeaWiFS L2 chlor_a images (from HRPT files) onto a standard (cylindrical) projection I'm using. I'm using the seadisp projection function to do this, then saving the files as flat FLOAT binaries.

What I'm seeing is some very low values, mostly at the edge of clouds. There are enough of these so that the histogram of the projected image (compared to the original) has an uptick in the 'gram as it approaches zero.

It's been suggested that the projection algorithm is milling some zeros (or bad data points) into the projected values. I was using nearest neighbor to avoid that exact problem. Do you have any suggestions as to why this may be happening? There's nothing in SeaDAS or IDL documentation which explains the details of the nearest neighbor search / selection process (for example, what happens if there are two equidistant LAC pixels from the geographic target? are they averaged?).

Since statistical analysis of the data is the goal here, this may be an important detail.

Thanks,
Norm

Norm,

I'm sure others will want to chime in on the subject, but I'll get things started ...

As you alluded to, any time a L2 file is reprojected, the values (and, thus, statistics) will change.  Particularly when the output image is smaller than the original, as multiple pixels will be dumped into the same geographic bin.  We tend to recommend not reprojecting L2 files, if possible, when statistics are involved, but acknowledge that it is necessary sometimes. 

When projecting the data, what output size are you using?  Also, are you designating a missing value?  The latter may one way to prevent zeros from being included.  We'd like to recreate the problem. 

SeaDAS makes use of many standard IDL routines to reproject the imagery. 
Agreed, the documentation on this subject is limited.

Best,

     Jeremy