re Geocoding Localities

Don Kirkup d.kirkup at LION.RBGKEW.ORG.UK
Tue May 21 13:48:44 CDT 1996

> >From: Chrissy van Hilst
> >Subject: Geocoding Localities
> >
> >I am working on a geocoding project at USNM. We are in the process of
> >working up protocols for enhancing specimen locality data by attaching
> >latitude and longitude. The goal is to be able to use these data in GIS
> >applications to look at everything from change over time to species
> >distribution. I'm looking for feedback from other collections that may
> >have done this or thought about it and other members of the community
> >that may have thoughts and ideas. One of my concerns is augmenting
> >imprecise localities with very precise lat/longs. We want this data to
> >be meaningful and mappable but we do not want to imply more precision
> >than we really have. Please email me with your comments at
> >cvanhils at

We have c. 15,000 collections records in the Brunei Checklist database
Most of the records are georeferenced and each set of coordinates is
assigned an error-code (in meters) depending on how they were derived.

Our scale of errors varies from c. +/-100m for GPS derived coordinates,
to locations given to the nearest minute, to gazetteer derived localities
for which we assign an arbritary error of +/-2,500m.
We also have some collections for which the only geographical information
is 'Brunei', but thankfully not too many!

For a pictorial representation of species ranges a map with different
symbols for the different error codes is probably enough to convey the
For analytical purposes however we felt that the georeference could not
satisfactorally be treated as a point, but should be treated as a

The first step of our approach was to simply draw a circle with a radius
for each collection point equivalent to that of the error-code value in
the database.

The precision of a locality was then further refined by finding the
intersection of the error-circle with graphical coverages related to
observations in our database. For example, collection observations in the
database includes forest and soil types. We also have graphical coverages
for both of these. The refined locality is calculated by overlaying the
error-circle with the forest type and soils coverages and finding the
intersection between these three graphical layers and the tabulated
values in the database.

For collections along 'linear' features such as roads and rivers rather
than draw an error-circle, we create a 50m buffer around the feature then
overlay and link to the database in the same way..

This involves a fair amount of work even though we have managed to
semi-automate the process (using PC and Workstation ArcInfo).

We felt that this was the best we could achieve and this was necessary if
we were to incorporate as many collection observations as possible for
the interpretation of our remote-sensed data.

Don Kirkup
Brunei Checklist Project
RBG Kew.

More information about the Taxacom mailing list