Bryon Jacob created JENA-1915:
---------------------------------
Summary: spatial:greatCircle appears to be returning wrong answers
Key: JENA-1915
URL: https://issues.apache.org/jira/browse/JENA-1915
Project: Apache Jena
Issue Type: Bug
Components: Spatial
Affects Versions: Jena 3.13.1
Reporter: Bryon Jacob
the `spatial:greatCircle` function appears broken - sometimes returning
negative numbers. returning negative numbers can never be correct from a
distance function - but I've tried to produce something more useful than that
as a defect report and repro...
I was trying to "port" some SPARQL queries where I'd written the great circle
function (Haversine formula) to use the spatial:greatCircle function - which
means I had a pretty good test case to see where the issue arises. I
simplified my query down to a repeatable test that demonstrates the issue:
note that in my original query, I was using a custom jena function to compute
`radians` - since that function won't be available to you in stock jena to try
this out, I've precomputed the radians values and put them in the VALUES block
next to lat/lon pairs. This should be runnable on any stock Jena instance with
the jena-geosparql functions loaded. Note that for most of the distances
computed, the two distances agree quite closely - but for the last two, the
jena function returns a negative number, where the hand-computed value is a
correct distance between those points
{code:java}
PREFIX m: <http://www.w3.org/2005/xpath-functions/math#>
PREFIX spatial: <http://jena.apache.org/function/spatial#>
# this is a namespace of custom functions, which won't be available in stock
Jena - to reproduce
# this, I've pre-computed the radian values and included them in a VALUES block
below...
# PREFIX f: <http://data.world/function/functions#>
SELECT ?lat1 ?lon1 ?lat2 ?lon2 ?φ1 ?φ2 ?Δφ ?Δλ ?d1 ?d2
WHERE {
VALUES (?lat1 ?lon1 ?lat2 ?lon2 ?φ1 ?φ2 ?Δφ ?Δλ) {
# in my original query, I was computing radians using a custom radians
function, which won't be available
# in stock Jena, so to make this reproducible I precomputed the
radians values needed for the Haversine
(41.2572 -95.9656 41.2592 -95.9339 0.7201 0.7201
0.0000 0.0006)
(41.2572 -95.9656 41.2482 -96.072 0.7201 0.7199
-0.0002 -0.0019)
(41.2572 -95.9656 41.5871 -93.626 0.7201 0.7258
0.0058 0.0408)
(41.2572 -95.9656 51.0472 -113.9998 0.7201 0.8909
0.1709 -0.3148)
(41.2572 -95.9656 40.7528 -73.9876 0.7201 0.7113
-0.0088 0.3836)
(41.2572 -95.9656 49.7237 13.3422 0.7201 0.8678
0.1478 1.9078)
(41.2572 -95.9656 -33.9065 18.4175 0.7201
-0.5918 -1.3119 1.9964)
(41.2572 -95.9656 -33.8646 151.2099 0.7201
-0.5910 -1.3111 4.3140)
}
# calculate the "great circle" distance between the two (lat φ, long λ)
points, in kilometers.
# these are the function calls in my original query, commented out and
replaced with VALUES above
# BIND (f:radians(?lat1) AS ?φ1)
# BIND (f:radians(?lat2) AS ?φ2)
# BIND (f:radians(?lat2 - ?lat1) AS ?Δφ)
# BIND (f:radians(?lon2 - ?lon1) AS ?Δλ)
BIND (m:sin(?Δφ / 2) * m:sin(?Δφ / 2) + m:cos(?φ1) * m:cos(?φ2) * m:sin(?Δλ
/ 2) * m:sin(?Δλ / 2) AS ?a)
BIND (2 * m:atan2(m:sqrt(?a), m:sqrt(1 - ?a)) AS ?c)
BIND (6371 AS ?RadiusOfEarthInKm)
BIND (?RadiusOfEarthInKm * ?c AS ?d1)
# call the Jena function for comparison
BIND(spatial:greatCircle(?lat1, ?lon1, ?lat2, ?lon2,
<http://www.opengis.net/def/uom/OGC/1.0/kilometre>) AS ?d2)
}
{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
