[Gdal-dev] Re: AVHRR NOAA level 1b

[Charles Wivell]

Hi all;

To create a truly accurate geocoded product from AVHRR, I recommend using a full satellite/sensor/earth model. The AVHRR sensor is very non-linear in pixel resolution. At nadir it might be 1.1 km, but at end of scan it is over 6 km (in the scan direction). The resolution varies approximately linearly from 1.1 km at zero degrees to about 2 km at +/- 45 degrees and then the pixels bloom rapidly as the scan drops off the edge of the earth. Also at those larger scan angles the parallax correction becomes important, thus a true ortho-correction should be used.

This should also be a warning to users of AVHRR data. A lot of the time producers of AVHRR derived products tell you their product is at a resolution of 1.1 km. The raw AVHRR data might have been resampled to an output pixel size of 1.1 km, but the true resolution of the product is not 1.1 km (except for the nadir pixels).

The MODIS sensor, in the way it scans, is a lot like the AVHRR sensor, so the resolution of the MODIS pixels vary along scan like AVHRR. The MODIS orbit is a little lower, so the blooming isn't quite so bad (about 1.1 to about 5km). But the MODIS sensors scans many lines at a time where as AVHRR scans a single line, so MODIS also has what is called the "bow tie" effect per scan. At nadir, MODIS has a slight scan gap but at end of scan, the scans have complete overlap ( a point on the ground is imaged by at least 2 scans, sometimes three) .

I guess after saying all that, my point is that sometimes (depending on the complexity of the sensor system) a simple warp is not the way to go to get accurate products.


Chuck Wivell

MN DNR Grand Rapids MN.

>>> Juan Castaneda <juan.castaneda at nuigalway.ie> 5/19/2006 5:59 AM >>>
We have a receiver and download images continously. The software we have
produces level 1b images and can export to .img. Those img are correctly
geocoded and they work fine. But I cannot set that software up to create
.img images for each l1b produced. So I thaught GDAL would do the work.

I have tryed using gdalwrap -tps but it doesn't seem to work. It
produces errors of 600Km... it's way too much. I'm more like a GIS guy
so I took the GCPs from gdalinfo and using ArcGIS I attempted to
geocode. It produced pretty much the same results as gdal. I'm wondering 
if I am missing something! should I have to apply some other correction? 
Maybe I am taking the same assumptions as gdalwrap and they are wrong. 
Maybe the GCPs read from gdal are not correct.

I appreciate any thoughts

Juan C

Markus Neteler wrote:

> On Fri, May 19, 2006 at 09:58:50AM +0100, Juan Carlos Castaneda
> wrote:
> 
>> Dear list,
>> 
>> I'm trying to use GDAL to geocode AVHRR level 1b images. It's not
>> giving me good results. I went to the archives and Google to find
>> what's happening and I read that the algorithm to geocode is not
>> the best for these type of image. From the documentation: "that
>> GDAL simple affine georeference model completely unsuitable for the
>> NOAA data. So you should not rely on it. It is recommended to use 
>> the thin plate spline warper (tps). Automatic image rectification
>> can be done with ground control points (GCPs) from the input file.
>> NOAA stores 51 GCPs per scanline both in the LAC and GAC datasets."
>> 
>> 
>> Any suggestions how to do the "thin plate spline warper (tps)" for
>> these images? Can it be done with GDAL? I've used gdalwrap -tps but
>> it's not working neither. I used the GCPs and they don't seem to
>> give accurate positions on the image. Can anybody give me some
>> advice?
> 
> 
> Please use gdalwarp with -tps option:
> 
> http://www.gdal.org/gdalwarp.html "-tps Enable use of thin plate
> spline transformer based on available GCPs. Use this instead of the
> -order switch. "
> 
> Please let us know how it performs (I am also interested in the AVHRR
> geocoding business).
> 
> Markus
_______________________________________________
Gdal-dev mailing list
Gdal-dev at lists.maptools.org 
http://lists.maptools.org/mailman/listinfo/gdal-dev