[postgis-users] Raster Points in Polygon (was Re: PostGIS Raster for Met Data)

[Bborie Park]

Hey Michael,

All the code I'm currently writing is for inclusion in the PostGIS 
Raster project.

Specialized C programs running against flat files will almost always be 
faster than PostGIS Raster.  But, if PostGIS Raster can return an answer 
in a reasonable time frame (<= 10 seconds ideally), my users will be happy.

Currently, my testing of a polygon filter against a daily coverage 
limited by a half month takes ~30 seconds.  But, this is because of the 
inefficiencies in ST_Intersection.

If you have any glaring performance inefficiencies, please do let us 
know so that we see if some work could be done to speed things up.

-bborie

On 08/25/2011 04:46 PM, Michael Akinde wrote:
> Hi,
>
> Sounds interesting. Is your code online anywhere?
>
> All of our source code is GPL and available at https://github.com/wdb. WDB is the backend database server for yr.no, Norway's largest weather site with ~3M visitors a week. It's heavily optimized for point retrieval from grids - which it does very effectively (an average yr.no weather page retrieves approximately 3000 points in<0.2s). We also recently implemented a NetCDF-Java interface to the database, to facilitate easy link up with WMS services (at least in theory).
>
> Our plan now is to extend and optimize the functionality for polygon retrieval and point data, since the system is getting leveraged for a variety of different types of projects.
>
> I don't think we'll be moving our data into PostGIS rasters (our experiments so far haven't shown any encouraging results performance-wise - perhaps not surprising, given that our current algorithms is essentially C code working on flat-files); but I hope that we can leverage the algorithms discussed in the thread to do the geo-spatial calculations for the more nasty queries using PostGIS. Pierre was kind enough to demonstrate that it can be done fairly easily - now I just have to figure out how to do it efficiently enough that our users will be happy.
>
> st_intersect(raster, raster) sounds very interesting, but what would be the result from the function?
>
> Regards,
>
> Michael A.
>

-- 
Bborie Park
Programmer
Center for Vectorborne Diseases
UC Davis
530-752-8380
bkpark at ucdavis.edu