GeoDjango: distance calculation

[kmishler]

We are starting to use geodjango (revision 7003) and are having
problems using the api to calculate distance.

class Airport(models.Model):
icao = models.CharField('ICAO', max_length=4)
location = models.PointField()

The location field will eventually be populated using lat/longs from a
text file.

For now, I populated 3 records in the table:

Airport.objects.create(icao="KBWI",
location=fromstr('POINT(-76.668333333333337 39.175361111111108)',
srid=4326))
Airport.objects.create(icao="KDEN",
location=fromstr('POINT(-104.67317777777778 39.861655555555558)',
srid=4326))
Airport.objects.create(icao="EGLL",
location=fromstr('POINT(-0.4613888888888889 51.477499999999999)',
srid=4326))

When I try to use the api to get distance, I do not get the correct
results:

point = Airport.objects.get(icao="KBWI").location
for airport in Airport.objects.distance(point):
print airport.icao + ": " + str(airport.distance) + ", " +
str(Distance(degree=airport.distance).nm)

EGLL: 77.1935295397, 0.000727473678024
KBWI: 0.0, 0.0
KDEN: 28.0132524428, 0.000263997564426

select icao, ST_distance(ST_GeomFromText('POINT(-76.668333333333337
39.175361111111108)',4326), location) from common_airport;
icao | st_distance
------+------------------
KBWI | 0
KDEN | 28.0132524427637
EGLL | 77.1935295397234


select icao,
distance_sphere(ST_GeomFromText('POINT(-76.668333333333337
39.175361111111108)',4326), location) from common_airport;
icao | distance_sphere converting manually to nm
------+----------------------
KDEN | 2393555.0105449 1292.41631
KBWI | 2.81660258041303e-10 0
EGLL | 5833168.46744864 3149.65875


Using distance_sphere, I get the correct results. Is there any way to
use the django api to get distance_sphere?

I also tried using the SRID 32140:

class Airport(models.Model):
icao = models.CharField('ICAO', max_length=4)
location = models.PointField(32140)

I used the same code to populate the records - it's nice that
geodjango can do that.

When I try to use the api to get distance, I get closer results, but
they're still not correct:

point = Airport.objects.get(icao="KBWI").location
for airport in Airport.objects.distance(point):
print airport.icao + ": " + str(airport.distance) + ", " +
str(Distance(meter=airport.distance).nm)

EGLL: 6258767.81814, 3379.46426466
KBWI: 0.0, 0.0
KDEN: 2442966.58517, 1319.09642828

Is there any way to use the django api to get the distance?


Versions:
WhiteBox Enterprise Linux 3.0 (clone of Red Hat enterprise 3.0)
gcc 3.2.3 / glibc 3.2.3
Python 2.4.2
ctypes 1.0.2
PostgreSQL 8.2.4
postgis 1.3.2
geos 3.0.0
proj4 4.6.0
curl 7.17.1 (needed for gdal)
expat 2.0.1 (needed for gdal)
gdal 1.5.0

[Justin Bronn]

> We are starting to usegeodjango(revision 7003) and are having

> problems using the api to calculate distance.
>
> class Airport(models.Model):
> icao = models.CharField('ICAO', max_length=4)
> location = models.PointField()
>

First, a reminder to add the `objects.GeoManager()` to your models --
it appears that you just omitted it in this post (or else you would
have received errors).

> Using distance_sphere, I get the correct results. Is there any way to
> use the django api to get distance_sphere?

I just clarified this in the documentation - the current distance API
currently works only with projected coordinate systems. Geodetic
systems, such as WGS84 (the default), will just use the degree
conversion factor, thus producing inaccurate results. I know of
`distance_sphere`, however, it only supports calculation between
points in PostGIS. Thus, geodetic support was purposely delayed until
I could come up with a better solution.

> I also tried using the SRID 32140 ...

> When I try to use the api to get distance, I get closer results, but
> they're still not correct:

SRID 32140 is "NAD83 / Texas South Central", thus I would not be
surprised if you obtained invalid results when using it for geographic
data that is outside it's bounds (e.g., KDEN, or Denver International
Airport, is a long way from South Texas). See http://www.spatialreference.org/ref/epsg/32140/.

One option is to use custom SQL to perform the calculations using
`distance_sphere` -- a patch adding such support would be even better
(I created ticket #6414 for this purpose). An interim measure would
be to just support such calculations for PointFields, which would
requiring some hacking on the `PostGISDistance` SQL construction
object in django/contrib/gis/db/backends/postgis/query.py

Best Regards,
-Justin

[kmishler]

Justin: Thank you very much for fixing ticket #6414. We are starting
to use the API and it works.

Would it be possible to change the distance calculations to use
distance_sphere instead of distance_spheroid? Or have an option to be
able to use either one? For some of our calculations, it is much too
slow to use distance_spheroid - the slight loss in accuracy is worth
the enormous gain in speed.

Karen

[Justin Bronn]

> For some of our calculations, it is much too slow to use distance_spheroid -
> the slight loss in accuracy is worth the enormous gain in speed.

I did some performance tests and confirmed your findings. I've
created a ticket that includes a patch that changes the default
geodetic distance calculation to use `ST_distance_sphere` instead:

http://code.djangoproject.com/ticket/6715

My informal tests indicate that calculations are 30-40% faster, with a
maximum of 0.25% loss in accuracy. I haven't committed yet because I
wanted to solicit comments on the API that enables spheroid
calculations.

-Justin