[postgis-users] Distances off in the Southern US

[Dan Putler]

Hi Mike and Nicolas,

The fact that Mike's two calculations resulted in the same value to 12 
decimal places is more that a little fishy. I seem to remember a similar 
issue coming up on this list sometime ago (roughly a year ago is my, 
faulty, memory). The difference between southern and northern cities in 
the US just shouldn't be an issue if the data was read in correctly.

Dan

On 08/15/2011 06:18 AM, Nicolas Ribot wrote:
> On 15 August 2011 12:50, Mike Hostetler<mike at squarepegsystems.com>  wrote:
>> Hello,
>> I'm somewhat new to GIS and I have a problem that I thought appeared to be
>> simply using a wrong projection or datum, but it seems to be a bit more
>> subtle than that.
>> I have a table of cities in the US and I'm trying to find distances between
>> them. When I use a city that is in the northern US, it works fine.  When I
>> try to find the distance between two cities in the Southern US, the distance
>> becomes way off.
>> I setup a Geometry in my cities table and populated it like the following:
>> select AddGeometryColumn('cities','geom',32661,'POINT',2);
>> UPDATE cities SET  geom=transform(setsrid(makepoint(longitude,
>> latitude),4269), 32661)
>> (I find the latitude and longitude from the Yahoo Geocode service)
>> A distance calc from McHenry, IL to Dallas, TX is calculated as:
>> select distance( (select geom from cities where id=26251), (select geom from
>> cities where id=67) )*0.000621371192 as miles;
>>       miles
>> ------------------
>>   996.717850542391
>> (Google Maps reads as 972, off by 25 miles or off around 4%)
>>
>> But Birmingham, AL, to Miami, FL is calculated as:
>> leader=# select distance( (select geom from cities where id=26251), (select
>> geom from cities where id=67) )*0.000621371192 as miles;
>>        miles
>> ------------------
>>   996.717850542391
>> (Google Maps reads as 767, off by 120 files, or 13%).
>> I can handle a little error, as long as it's somewhat small (<5%).  But this
>> is way off.
>> Again, it smells to be to be a datum or projection issue to me, but I'm not
>> sure how to find the sweet spot to be accurate everywhere.
>> Your input is appreciated.
>>
> Hi Mike,
>
> Some remarks:
> • Are you sure Yahoo! Geocoder Service returns degree expressed on
> 4269 coordinate system ? I read it is 4326.
> • How did you get the distances with Google Maps ? Did you use travel
> directions ? If so, the 972 miles is by driving on roads, not by
> flying the shortest distance (great circle).
>
> The following service:
> http://www.geobytes.com/CityDistanceTool.htm?loadpage
> gave me a direct distance of 808 miles or 1300km for McHenry - Dallas
> and a distance of 658 miles or 1059 km for Birmingham - Miami.
>
> Then, by running these queries the distance between cities seems to be good:
>
> select foo.city, bar.city, st_distance(foo.geom,
> bar.geom)*0.000621371192 as miles
> from
> (select 'McHenry' as city, 'IL' as state, 'POINT(-88.267314
> 42.326215)'::geography as geom) as foo,
> (select 'Dallas' as city, 'TX' as state, 'POINT(-96.795404
> 32.778155)'::geography as geom) as bar;
>
> city   |  city  |      miles
> ---------+--------+------------------
>   McHenry | Dallas | 807.029700174124
>
> select foo.city, bar.city, st_distance(foo.geom,
> bar.geom)*0.000621371192 as miles
> from
> (select 'Birmingham' as city, 'AL' as state, 'POINT(-86.811504
> 33.520295)'::geography as geom) as foo,
> (select 'Miami' as city, 'FL' as state, 'POINT(-80.237419
> 25.728985)'::geography as geom) as bar
>
> city    | city  |      miles
> ------------+-------+------------------
>   Birmingham | Miami | 666.318599210394
>
> The slight differences between CityDistanceTool and Yahoo! services
> come from the cities coordinates: the two services do not position
> cities at the same coordinates.
>
> Also please note I'm using the new GEOGRAPHY Postgis type that allows
> direct distance computation: no need to transform data back to a
> planar coordinate system.
> If you use geometry type with long/lat coordinates, you may have a
> look at st_distanceSphere and st_distanceSpheroid.
>
> Nicolas
> _______________________________________________
> postgis-users mailing list
> postgis-users at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-users