[GRASS-user] Once more: raster projection transformation: LAEA -> UTM

[Paul Kelly]

On Tue, 22 Jul 2008, Jana K. wrote:

>
> I've already posted this a couple of days ago, but I am afraid somehow it got
> lost in other questions.
> I believe someone must know an answer, so please, HELP me :-(:
>
> I need to transform a map from LAEA coordinates into UTM. I created a new
> location with UTM projection. I used r.proj from within the new location,
> but  running it,  I received the following error message:
>
> Input Projection Parameters: +proj=laea +lat_0=52.0000000000
> +lon_0=10.0000000000 +x_0=4321000.0000000000 +y_0=3210000.0000000000
> +a=6378137 +rf=298.257222101 +no_defs +towgs84=0.000,0.000,0.000
> Input Unit Factor: 1
>
> Output Projection Parameters: +proj=utm +zone=34 +a=6378137
> +rf=298.257223563 +no_defs +towgs84=0.000,0.000,0.000
> Output Unit Factor: 1
>
> Can't work with xy data
>
> I don't understand it as the data are clearly georeferenced.

Well it looks like your current (output) location is broken in some way. 
How was it originally created? Does
g.region -p
show an XY Location? Or a UTM location? Maybe you could try recreating the 
projection information by running g.setproj, or g.proj -c with an 
appropriate input source.

> Now question 2:
>
> My raster data are categorical (landuse), and during the projection
> transformation I want to go from high resolution (100m) to rather crude
> resolution (2500m). How GRASS handles such operation? I would expect that It
> assigns the highest frequency value to the new cell (I mean if one new cell
> contains 625 old pixels, it will be assigned the value which occurs most
> frequently). Am I right?

If you read the r.proj manpage, it says:
"The projected data is resampled with one of three different methods: 
nearest neighbor, bilinear and cubic convolution.

The nearest option, which performs a nearest neighbor assignment is the 
fastest of the three resampling methods. It is primarily used for 
categorical data such as a land use classification, since it will not 
change the values of the data cells. The bilinear option determines the 
new value of the cell based on a weighted distance average of the 4 
surrounding cells in the input map. The cubic option determines the new 
value of the cell based on a weighted distance average of the 16 
surrounding cells in the input map.

The bilinear and cubic interpolation methods are most appropriate for 
continuous data and cause some smoothing. Both options should not be used 
with categorical data, since the cell values will be altered. If nearest 
neighbor assignment is used, the output map has the same raster format as 
the input map. If any of the both interpolations is used, the output map 
is written as floating point. "

So I think this clearly suggests that with categorical data your only 
option is nearest neighbor, which doesn't do what you want. Perhaps you 
could resample with r.resamp.stats in the original location before 
reprojecting.

Paul