[GRASS-user] Multiple labels and r.series

[John C. Tull]

On Nov 11, 2008, at 9:04 PM, Glynn Clements wrote:

>
> John C. Tull wrote:
>
>>>> I am creating a raster map that essentially summarizes the
>>>> intersection of a series of labeled rasters that each have
>>>> values of either 1 or null (presence/absence data). I am
>>>> using r.series like this to perform the summation across the
>>>> rasters that are all named with a prefix of presence_:
>>>>
>>>> r.series input="`g.mlist rast
>>>> pattern='presence_*' sep=,`"
>>>> output=presence_sum method=sum --o
>>>>
>>>> Can anyone think of a way to achieve this summation of
>>>> intersections in a way that will also allow for each cell to
>>>> have a label that concatenates all labels for the respective
>>>> raster maps that are present at that cell? E.g., let's
>>>> say three of the rasters are present at a particular cell,
>>>> and the labels for those rasters are 'raster1',
>>>> 'raster4', and 'raster7'. I would like the
>>>> final map to have a label of 'raster1, raster4,
>>>> raster7'.
>>>>
>>>> Any and all suggestions are greatly appreciated.
>>>
>>>
>>> the trick here is to r.reclass 0/1 absence/presence maps to be 0/2^n
>>> before running the sum.  So you then get a sum like 13, which would
>>> mean
>>> the 8+2+1 maps were all present, or if the sum was 5 you'd know the
>>> 1+4
>>> maps were present for that cell, and not others.
>>>
>>>
>>> see also r.bitpattern and r.mapcalc bitwise operators.  (sorry I  
>>> don't
>>> know the proper name/wikipedia link for this technique, I'm not a
>>> computer
>>> scientist)
>>
>> I follow your logic here. I am still hopeful that someone will have
>> some sort of script or r.mapcalc magic to generate concatenated  
>> labels.
>>
>> As I'm not a computer scientist either, I fear I would spend more  
>> time
>> trying to decipher the bitwise values and meanings than would be
>> practical. That is a good suggestion, nonetheless. I should have  
>> added
>> that I am using about 96 raster layers.
>
> The main problem here is that labels are associated with categories,
> not with individual cells.
>
> r.cross will take a set of input maps, and generate a unique category
> (and concatenated label) for each unique combination of categories.
> Howver, it won't handle 96 maps (it is currently limited to 10 maps in
> 6.3.0, increased to 30 maps in 6.4 and 7.0).
>
> My first suggestion is to try using r.cross on batches of 10 maps at a
> time, then use r.cross again on the 10 intermediate results. This may
> fail if the total number of combinations is large, or if any cell has
> too many labels (there may be a limit on the length of a label), and
> will probably be quite slow (possibly unacceptably so).
>
> Ultimately, it may not even be possible to generate such a map, given
> the 32-bit limit for category numbers. You could theoretically have up
> to to 2^96 distinct categories, although in practice you would be
> limited by the size of the map. But you could end up with each cell
> having a distinct combination of labels.

I will explore this option. I'm using 6.4 svn, so I can push the  
envelope a bit.

Thanks,
John