[postgis-devel] Status of raster mapalgebra?

[Pierre Racine]

Hi Nicklas,

>What is the status of map algebra functionality?

There is a one raster version on the way (Jorge what is the status?) and a two raster version planned. You can see slow plpgsql prototypes of those two functions in raster/script/plpgsql

>Is there any code that could be tested?

Just the plpgsql versions. But they might be sufficient to make a demo or proof of concept.

>In beginning of Mars I will give a talk about PostGIS i Stockholm. I
>will try to tell about the news in 2.0.

MapAlgebra will be a very progressive work:

1) One raster version - Pixel by pixel
2) One raster version - Accepting neighbour pixels in expression
3) One raster version - Accepting neighbour pixels from the whole coverage (hence getting raster values using world coordinates over the whole table instead of just raster coordinates)
4) Two raster version
5) Two raster version working with non-aligned rasters (forcing resampling of one to be aligned on the other one)

1) might do it in 2.0.
2) maybe.
The rest probably/certainly not. It will depend on the date of release.
5) need GDAL based ST_Resampling first (this is still far away).

>Most of my raster experience is about finding clear cuttings by
>comparing satellite imagery from different years. It would have been
>very nice to be able to show that in PostGIS. The software that were
>used for that purpose was something called Enforma that as far as I
>understand just compared band for band what had happened since last
>year. The different bands give different result depending on what
>wave-lengths the photosynthesis uses compared to what is absorbed to
>naked ground and so on.

For this I guess you would need the two raster version computing difference between two bands years after years.

>Will there be anything to try during February?

I suggest you use the plpgsql versions as a proof of concept and comment on their usage. They might however be broken right now because of some change in the API but they should be easy to repair.

>Am I right that map algebra will be the tool to use?

Yes. Something like:

SELECT ST_MapAlgebra(rast1, 1, rast1, 2, "rast2 - rast1") FROM yourrastertable WHERE ST_Intersect(rast1, rast2)

Later MapAlgebra should be encapsulated by a number of specific functions. In your case probably something like ST_Difference(raster1, band1, raster2, band2)

>I have no idea how much work there is to do so I don't know what I am
>asking for.

The two raster version is very complicated (you must deal with nodata values, non-aligned rasters). I'm working on an optimized version (setting large zones or raster with memcpy when possible) which still a lot of work.

Pierre