Re: Geometry vs Geography (what to use) - Mailing list pgsql-sql
| From | Lee Hachadoorian |
|---|---|
| Subject | Re: Geometry vs Geography (what to use) |
| Date | |
| Msg-id | CANnCtnJAF2Ux3esB=w=3D-Lbg8xfbs=Viby5Uy84YY8mQZ9RRA@mail.gmail.com Whole thread Raw |
| In response to | Re: Geometry vs Geography (what to use) (Michael Moore <michaeljmoore@gmail.com>) |
| Responses |
Re: Geometry vs Geography (what to use)
(Michael Moore <michaeljmoore@gmail.com>)
|
| List | pgsql-sql |
Mike,
ST_DWithin takes a parameter as distance in the units of the SRID. Since 4326 is decimal degrees (latitude and longitude), you are manually converting from degrees to miles. This conversion will only come close to working if you are covering a pretty small area and if you look up the conversion based on the latitude, and will *always* be hobbled that that a degree longitude is not the same length as a degree latitude (although they are close at the equator).
The standard way to accomplish this is to either project the coordinates to an appropriate planar coordinate reference system, or to use the geography data type. If you use geography, the units are in meters.
You could ask your DBAs to change the SRID of the table, but you can also transform them as part of your query.
Option 1 (using SRID for UTM 10N, see below)
SELECT pc1.city AS city,
pc1.postalcode AS postalcode
FROM tpostalcoordinate pc1
WHERE ST_DWithin(ST_Transform(geo_position, 32610),
(SELECT ST_Transform(pc2.geo_position, 32610)
FROM tpostalcoordinate pc2
WHERE pc2.postalcode = '95050'
AND pc2.countrycode2tcountry = 'US'),
80467.2)
I've selected SRID 32610 as the appropriate UTM zone based on the searched ZIP code 95050. If you need to do these calculations over a different area, you would have to use a different UTM zone, or pick a wide-area CRS (something that would apply to the continental US), with some loss of accuracy.
Option 2 (using geography)
SELECT pc1.city AS city,
pc1.postalcode AS postalcode
FROM tpostalcoordinate pc1
WHERE ST_DWithin(geo_position::geography),
(SELECT pc2.geo_position::geography
FROM tpostalcoordinate pc2
WHERE pc2.postalcode = '95050'
AND pc2.countrycode2tcountry = 'US'),
80467.2)
Neither query has been tested.
Best,
--Lee
On Tue, Apr 5, 2016 at 11:28 AM, Michael Moore <michaeljmoore@gmail.com> wrote:
Steve,On Mon, Apr 4, 2016 at 6:09 PM, Steve Midgley <science@misuse.org> wrote:On Mon, Apr 4, 2016 at 5:20 PM, Michael Moore <michaeljmoore@gmail.com> wrote:I am converting from Oracle to postgres. We have an application that takes a value for MILES as an input and returns a list of SCHOOLS that are within range. This all works fine in Oracle, no problem. In Oracle the datatype of the table field is SDO_GEOMETRY. Our DBAs have set up the same table in Postgres with datatype geometry(Point,4326). Here is where the problem comes in. I am trying to use the ST_DWithin function and it kind of works except that one of the input parameters to this function is expressed in RADIANS.
SELECT pc1.city AS city,pc1.postalcode AS postalcodeFROM tpostalcoordinate pc1WHERE ST_DWithin(geo_position,(SELECT pc2.geo_positionFROM tpostalcoordinate pc2WHERE pc2.postalcode = '95050'AND pc2.countrycode2tcountry = 'US'),50 * .01539)So, in the above query I am selecting all the cities with a 50 mile radius. It WORKS ... kind of. The thing I don't like is my screwy way of converting radians to miles. ( miles * .01539 ). It's just a best guess that gives me results 'almost' the same as Oracle's. The second thing I don't like is the 'almost' the same. I'm am guessing the difference in the result set is due to the planar calculation vs the spheroid calculations. So my questions are
- Should I be using GEOGRAPHY(POINT,4326) instead of GEOMETRY(POINT,4326)I assume this would allow me to express the distance meters and it would do spheroid calculations which should give me results more consistent with Oracle's?
- Any advice. Is there something else I should be doing? What did I miss?
Just reading this article[1] makes me think that you might get better spherical calculations for lat/long using the Geography data type.Regarding ST_DWithin itself, I'm confused about your convertion to radians. The docs I'm looking at say that the third parameter to that function is meters, not radians. (And radians aren't a measure of distance anyway, as I'm sure you know, but which is also confusing me in your message). So are you just converting 50 miles to meters?Also, maybe your conversion factor is accurate, and it is possible that the *Oracle side* is using a linear projection (treating lat/long like a cartesian grid), and you are doing proper spherical math on the Pg side, and that's why your results are close but not quite right?I hope this helps your work.. Best,SteveI think I read somewhere in the documents that Radians was the parameter for Geometry and Meters is the parameter for Geography. Even with Radians, if the radius of the circle is known then so too is it for 1 radian. I'm guessing there is a default circle size for a geometry(Point,4326) datatype. I'll keep digging and look at the docs you suggested.
Regards,Mike