QGIS 3.4 distance matrix returns the wrong point

Question

I'm using Distance Matrix to find the closest point between one set (4 points) and another set (4 points), detailed below. All results look okay except one, where the returned point is not the closest. Any ideas what is it that I'm doing wrong?

Set 1 (input)

Set 2 (target)

The output layer shows the following. However, TargetID for 1 should be 3, and not 2. By the way, I did another distance matrix layer where the "Use only the nearest (k) target point" was 0, giving an output of all distances between all points, and indeed the distance between 1 (bilbao) and 3 (oviedo) was smaller than the distance between 1 (bilbao) and 2 (alcala de henares), which suggests that qgis knows the correct distance between points but returned a wrong one when I'm looking for the nearest point only.

Distance Matrix layer

Answer 1

It looks like a bug and/or improper projection.

Using distances in long-lat degrees (which doesn't mean much), the result is the proper one. Using true ground distances (in meters), the reported distances are the proper ones.

with src as (select 1 as id, st_makepoint(-2.92391,43.25722) as geom),
dest (id, geom) as (values (2, st_makePoint(-3.36708, 40.48227)),(3,st_makePoint(-5.84372, 43.36232)))
SELECT src.id, dest.id, 
       ST_Distance(src.geom,dest.geom) d_degrees, 
       ST_Distance(src.geom::geography,dest.geom::geography) d_meters
from src, dest;

 id | id |    d_degrees     |    d_meters
----+----+------------------+-----------------
  1 |  2 | 2.81011514913535 | 310402.55232612
  1 |  3 | 2.92170095083326 | 237160.04795382

Looking at the code, we see that two different functions are used to get the nearest neighbor (using the spatial index) and to compute the distance. My guess is that the function computing the nearest neighbor does not consider the unit nor cast to geography.

I would recommend projecting your data to a CRS that use meters, like UTM.

You might also want to fill a bug report.