the dangers of using Google Earth/Maps to georeference


Bombus vagans, beautifully captured by J. C. Lucier.

My colleague ran into an interesting problem while georeferencing our bumble bees yesterday. It was a nice lesson for our lab group, so I thought I’d post the story here for everyone …

We’re in the process of affixing unique identifiers to all of our bee specimens, and we’re capturing the information from their labels: where and when they were collected and by whom. These metadata are typed into a spreadsheet that we eventually import into our database. The verbatim data are then queued to be georeferenced (i.e., figuring out the latitude and longitude for each locality), either manually or by using some informatics trickery to be described later. My colleague needs the Bombus data ASAP for an interesting and timely niche modeling study, so she’s carefully poring over several georeferencing resources, like Google Earth, gazetteers, etc. to predict lat and longs for these specimens. She ran into a Bombus vagans individual with this collecting event label:

USA: North Carolina
Judson
19.VII.1923
JC Crawford

Easy. We’ll just go to Google Maps, search for Judson, NC, grab the lat and long (35.002335, -78.80888), and be done with it. Next specimen … right? Well, Bombus vagans, also known as the wandering bumble bee, is largely a boreal species (see also the specimens in GBIF). Finding a specimen in the coastal plain habitat near Fayetteville, NC seemed unlikely—or maybe this bee was living up to its common name! A bit of detective work revealed that J. C. Crawford collected several other bees in the mountains of western North Carolina five days later (24.VII.1923) and that this collector’s typical hunting grounds in 1922 and 1923 were exclusively in the mountains near Bryson City. Hmmm… Diving deeper into the mystery my colleague found that there was a second Judson, NC in the 1920s, located just downriver from Bryson City on the Little Tennessee. It disappeared in the 1940s after being flooded by Fontana Lake (which resulted from the Fontana Dam). Ah ha!

Next step: look up some old maps of Swain County and estimate the lat and long for the old town of Judson: 35.40355, -83.561347 (with a generous error of 1000 meters). If we really wanted to be precise we’d try to find better maps from the 1920s-1940s, but this first pass will get my colleague going on her niche modeling analysis.

So, what did my lab group learn? 1) it pays to know the natural history of the species you’re georeferencing, 2) precise collection dates and collectors’ names are useful, 3) we can’t employ a single tool to predict latitude and longitude, no matter how user-friendly Google Earth is(!), 4) we’re better off georeferencing our specimens at the time of collection, rather than after-the-fact, 5) we need to document how we determined the lat / long and make sure this information is added to the database, especially when detective work was required, 6) natural history collections are vital sources of irreplaceable information. What if we wanted to sample bees in Judson, NC today? Perhaps there are some aquatic species we could sample using fish nets and SCUBA gear?

Note: a best practices guide for georeferencing is available from GBIF.


6 Responses to “the dangers of using Google Earth/Maps to georeference”

  1. ihateaphids says:

    Luckily, rare oddballs have little effect on resulting models for species with tremendous numbers of records like bumbles…as long as the full geog. range of the species is represented in the occurrence data, and there are no systematic biases in the data (like consistent misID, or inconsistent lumping/splitting of subspecies). Of course, it depends on scale of the study, little effect for global range predictions; predictions of habitat suitability at a high resolution of course need highly accurate locality data (1000m won’t cut it, maybe even 10 won’t cut it for certain types of studies and predictor variables!).

  2. Alex Wild says:

    Good post. Reminds me of a “Trinidad” in California that has caused some confusion about Neotropical ant distributions in the literature.

  3. GailK says:

    #5 We always need to document how we determined the lat / long. We’ve been doing this now for >15 years for the therevid PEET project, starting when there were few electronic gazetteers, and I now wish we’d been even more explicit about the growing number of sources during certain periods of our history. Different gazetteers/maps will give different results. How many times have you looked up an address that you know well in Google maps/earth and found they’re off by a house? The historical angle provides the chance for more detective work, however, and it takes special workers to pursue some of these mysteries. Good work in tracking this one down and sharing lessons learned!

    P.S. Should also add keeping track of the geodetic datum for GPS measurements.

  4. Old museum specimens often throw up interesting label data posing interesting locality labels. I recall a specimen of an hemipteran that had the location Queenscliff as the only locality for a specimen of collected in 1880s in the South Australian Museum. The only extant Queenscliff was way out of the known range of the bug, but searching through historical maps, we found that the locality is today known as Kingscote on Kangaroo Island. Makes sense when you consider the name changed after the death of Queen Victoria and cliff and cote mean the same thing!

    So, always check historical maps and not just current gazetteers.

  5. Lee Belbin says:

    Most don’t seem to equate taxonomy with gazetteers, but there are parallels. Most gazetteers don’t have temporal information (or old feature names or synonomy etc) – and should: a ‘gazeteer concept’? :)

    PS: I assume you are aware of http://www.museum.tulane.edu/geolocate/, not that it solves all problems.

  6. [...] errors in assigning localities, which can be tricky when two areas have similar names (see “the danger of using Google Earth/Maps to georeference“). Overall, I have come to appreciate the value of geographic data checking for georeferenced [...]

Leave a Reply