[Taxacom] Inappropriate accuracy of locality data
taxacom3 at achapman.org
Thu Dec 2 14:15:41 CST 2010
Indeed Bob - all of these and more are forms of precision, and I will
make a note for any updating of the Best Practices Guide.
I have only briefly touched on this topic here - one could write a book
and probably should!
Doug - Datums are still important - the maximum difference between
datums found so far is 3.552 km (reference in the Best Practices
Guide). It is important to use the appropriate datum or WGS84 for the area.
Bob - you suggested rounding of the lat/long - i.e. reducing the
precision of the recording, but this can also lead to problems
(depending on the reduction in precision) as it moves the point as
lat/long is only a point representation of the left hand bottom of grid
square. One is better using an appropriate precision and the second
field of 'uncertainty'. In this way, no information is lost.
Locations within caves can be difficult, and I often see the lat/long of
the entrance given and then directions from the cave entrance. This is
probably the simplest way of recording the location. As to people
deliberately falsifying locations this can also be a problem - it
happens with respect to people attempting to hide the locations of
sensitive taxa, for example. In these cases we recommend decreasing the
precision and recording that information and *not* moving (randomly or
otherwise) the lat/long as this is misleading and can lead to false
analyses - not good for our science. It is thus important that
documentation of what has been done accompany the geolocation
information. Another report on recommendations for "Dealing with
Sensitive Data" was commissioned by GBIF and can be found on their web
On 2/12/2010 6:21 PM, Bob Mesibov wrote:
> Hi, Arthur.
> Since you've opened this can of worms, allow me to extract one of the little annelids and wiggle it:
> I know (at least) 5 sorts of 'precision'. The first is the one you call instrument precision. The second is multi-reading precision: the spread of location values seen when you take readings from the same spot at different times (ideally, over several days) and under different weather conditions. Multi-reading precision is larger than instrument precision. I suspect it's also anisotropic.
> As you point out, precision isn't accuracy, and a GPS simply cannot tell you how far your GPS location is from your actual location. Garmin reckons 15 m RMS from true for its eTrex under favourable conditions. Taking an average from a multi-reading set of values and calling that the best estimate of real location is probably safe (but I wonder whether there are systematic biases in some spots, e.g. urban and rockbound places).
> The third sort of 'precision' is the *implied* reliability of a published, GPS-based locality with no accompanying uncertainty figure. Doug Yanega's 41 52 38.84 N implies plus/minus 0.005s of latitude, which is wildly inappropriate for both an eTrex and a typical biological collection. I advise rounding that off to 41 52 39 N, which has an implied reliability of 1s, or about 30 m. A reader can reasonably infer from that latitude figure that the locality was closer to 39s than to 38s or 40s. (But it would be better if the author had included an uncertainty as well.)
> The fourth sort of 'precision' applies to transects. Best practice is obviously to specify the track, or the start and end points for a straight-line transect. I know that some field workers, me included, locate the transect collections at the midpoint of the X-m-long transect if the transect is short enough - in our opinion.
> The fifth sort of 'precision' applies in the special case of cave locations. I don't know what the situation is outside Australia, but here caves exist in a very fuzzy space - and I'm not referring to the problem of a cave collecting locality 200 m from the cave entrance. Caving clubs keep cave locations secret, and government conservation bureaucrats go along with this secrecy to keep the cavers onside. I have been allowed to publish cave millipede localities from Tasmania generalised to 1 minute of lat/long, but even that upsets some of the cavers, who would prefer a cave name only.
> In any case, I agree 100% with Doug that all this is pretty academic. In my multipedes collecting-event database, the uncertainty field contains categorical plus/minus values: 25 m, 50 m, 100 m, 200 m, 500 m, 1 km, 2 km, 5 km, 10 km and GOK (God only knows). These are point-radius distances. Like Doug, I struggle getting historical records down the category scale.
More information about the Taxacom