[postgis-users] Is this a perverse use-case for raster types?

[James Hiebert]

Hi All,

I'm considering using PostGIS rasters for storage of raster data at my organization and I'm looking for some advice (or perhaps a reality check).  I work for a region climate services provider and the vast majority of our data (by volume, not necessarily complexity) are output from climate models.  These are generally a n-by-m raster with one band for each timestep.  There could be upwards of 36k to 72k timesteps for a typical model run.  We have hundreds of model runs.

So my question is, is it insane to be thinking of storing that many bands in a PostGIS raster?  Or more specifically, is this _not_ a use case for which PostGIS rasters were designed?  I notice that most of the examples in the docs and in "PostGIS In Action" focus only on images and I can imagine that handling multispectral satellite images as being more of the intended use case.

I did a little benchmarking of a typical use case of ours ("What's the average temperature inside a some polygon, e.g. a river basin?").  I noticed that the run time for doing a ST_Clip(raster, band, geometry) and ST_Intersects(raster, band, geometry) appears to be super-linear even when doing it on just a single band.  I ran the following query:
SELECT rid, st_height(st_clip(rast, 1, the_geom)), st_width(st_clip(rast, the_geom)) FROM basins INNER JOIN bcsd ON ST_Intersects(rast, 1, the_geom) WHERE rid = <rid> (where basins is table of river basins with one single polygon and bcsd is a table with a raster column "rast").
for a set of rasters with increasing number of bands, and the time to run the query is shown in the attached plot.  Since the raster properties are presumably shared across all the bands, it seems odd to me that run time would increase.  I would expect it to be _contant_ (with constant number of pixels), but I suppose that that's my own ignorance as to how the PG type extensions work?

Comments or explanations are welcome.

~James

-- 
James Hiebert
Lead, Computational Support
Pacific Climate Impacts Consortium
http://www.pacificclimate.org
Room 112, University House 1, University of Victoria
PO Box 1700 Sta CSC, Victoria, BC V8V 2Y2
E-mail: hiebert at uvic.ca
Tel: (250) 472-4521
Fax: (250) 472-4830
-------------- next part --------------
A non-text attachment was scrubbed...
Name: clip_time.png
Type: image/png
Size: 20010 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20121029/ba081e96/attachment.png>