[gdal-dev] reproject python numpy binary swath/lat/lon

[David Hoese]

Rutger,

Thanks for the reply.  I tried the s_srs tag and it still didn't help.  
I also tried using the -geoloc flag which I had tried earlier too and 
still get the 441 out of 441 failed to transform error.  I have tried 
PyTroll, but there KDTree algorithm seems to be too slow for the amount 
of data I'm using (not sure why), I have contacted them in the past.

I'm attempting to project satellite data to an AWIPS grid.  Assuming I 
have the grid defined correctly with proj4 parameters, I've been using 
"+proj=latlong +datum=wgs84" as my s_srs parameters.  Does that sound 
correct for satellite data?  So my command line call (still for testing) 
becomes:

gdalwarp -s_srs "+proj=latlong +datum=wgs84" \
-te 59.844 -123.044 14.335 -65.091 \
-ts 5120 5120 \
-t_srs "+proj=lcc +lon_1=-95.000  +lon_0=-113.133 +lat_2=25.0001 
+lat_1=24.9999 +lat_0=16.369" \
-geoloc -of GTIFF \
image.vrt out.tif

The output grid is 5120x5120 and I wasn't sure if -te was lat/lon 
points, but that's what I used.  -te has lat/lon of upper left pixel of 
the grid and lat/lon of the lower right pixel of the grid.  Does any of 
this look correct?  Thanks for the help.

-Dave

On 6/12/12 7:04 AM, gdal-dev-request at lists.osgeo.org wrote:
> Message: 2
> Date: Tue, 12 Jun 2012 02:08:21 -0700 (PDT)
> From: Rutger<kassies at gmail.com>
> To:gdal-dev at lists.osgeo.org
> Subject: Re: [gdal-dev] reproject python numpy binary swath/lat/lon
> Message-ID:<1339492101512-4980686.post at n6.nabble.com>
> Content-Type: text/plain; charset=us-ascii
>
> Hello David,
>
> A while back i did some tests with GDAL geolocation arrays using MODIS swath
> data. I found that specifying the "s_srs" tag when running gdalwarp gets rid
> of the 'too many points' error, even if you already have a SRS in your VRT.
> If you use gdalwarp, your target grid can be completely described by the
> -t_srs -tr and -t switches. The Python equivalent of these are the
> dataset.SetProjection() and dataset.SetGeoTransform(). Getting the input SRS
> from only the coordinate values seems impossible to me, the SRS should be
> either read from some metadata or specified by the user.
>
> For me the geolocation arrays worked technically well, and very convenient.
> However. i found the resulting image quality very disappointing (for my
> specific MODIS case) compared to MRTswath and Pyresample. This might be due
> to the irregular coordinate grids as a result of the swiping image sensor or
> something like that. Maybe in different cases it would give good quality
> results.
>
> I ended up using the Pyresample module which is part from the Pytroll
> project. See:
> http://pytroll.org/index.html  http://pytroll.org/index.html
> This worked great for my case.
>
>
> Regards,
> Rutger