[postgis-users] polygons crossing 0/360 longitude

[Pritchard, MJ (Matt)]

Hi Shane (& Carlos),

Thanks for your posts about this: this is really helpful as I have been
grappling with this issue, too. However I'm still stuck...
Do you have an example of a spatial search query that works, Shane,
using geometries loaded using your lon_wrap() function? 

The problem I have had is false hits caused by postgis incorrectly
assigning a bounding box to a geometry spanning the dateline, which then
spans the *rest* of the longitude range. E.g. if your geometry's
longitude range is 170 to -170, its resulting BB will be -170 to 170
(implying it crosses the Greenwich meridian, not the dateline), hence
the false positive if you search over Greenwich.

E.g. If you take a geometry like

'POLYGON(((-170 20,170 20,170 5,-170 5,-170 20)),-1'    
(which spans the dateline)

And split it as per your method into

'MULTIPOLYGON(((360 20,360 5,190 5,190 20,360 20)),((0 5,0 20,170 20,170
5,0 5))),-1'

Then apply a search like

select * from matt_test where geom1 && geometryfromtext('POLYGON(( -10
10, 10 10, 10 5, -10 5, -10 10))',-1);

then I still get a false positive. AFAICT, the problem is present
whether your geometries are all in SRID=4326 or all in SRID=-1

Have you got some other method of searching that avoids this problem?

Many thanks,

Matt

------------------------------------------- 
NERC Earth Observation Data Centre 
Space Science & Technology Dept. 
CCLRC Rutherford Appleton Laboratory 
Chilton, Didcot OX11 0QX 
01235 445645 (tel) / 445848 (fax) 
http://www.neodc.rl.ac.uk/