I am currently collating a list of specimens from databases of various natural history museums for further research. However, a well known issue associated with majority of historical data is the lack of appropriate latitude and longitude which prevents one from using that data.

There have been ways to overcome that data - such as drawing a buffer around a region and providing a range of uncertainty associated with that location.

For instance, the function - biogeomancer from the package 'spatial' in R, automates the process of georeferencing, provided there are a few textual descriptions such as "2 miles west of XYZ". See documentation here.

However, my main concern is in using such a protocol for regions as big as 200 square km. Is there a way one can overcome that issue? I would love to use this rich trove of museum data, provided I can handle the uncertainty associated with its location.

An example of some specimens in my dataset is shown below. Please note that many of them come with mentions of elevation, but most of the records are very vague.

enter image description here


In the comments section, one of you mentioned the purpose of this question and what I tend to achieve from the same.
1. I am interested in how once can reduce the radius of uncertainty from a really wide polygonal region to a smaller radius of uncertainty (if possible).
2. This information will help me carry out future spatial analysis such as species distribution modeling / occupancy modeling for instance.

  • Have you heard of GBIF ? gbif.org It may already have what you're looking for.
    – GISHuman
    Commented Apr 3, 2018 at 16:14
  • @GISKid Yup, this is the data from GBIF. Unfortunately, majority of that data lacks good georeferncing. Commented Apr 3, 2018 at 16:15
  • Interesting question! However, it's a little unclear what you're hoping to do - the items are already georeferenced in that they have location information, albeit for polygonal regions. Could you edit and expand on how you hope to 'deal' with the issue of large polygons? Is it to help with a spatial analysis?
    – Simbamangu
    Commented Apr 4, 2018 at 4:21
  • 1
    What method are you using for your SDM? And what size study area? Depending on those two - I would think that a large area of uncertainty would reduce the usefulness of a SDM, personally. I would instead eliminate the data that has a large polygon of uncertainty and stick to using occurrences that are 'more precise'. Especially if # of occurrences isn't an issue
    – GISHuman
    Commented Apr 4, 2018 at 13:40
  • 2
    Seems that you might be able to narrow, on a case by case basis, some locality data - e.g. using raster elevation data of 3500' ± 250' to mask within the Santhapara polygon. Without using 'detective skills' and additional data, you are stuck with taking the centroid of the polygon - and (speaking from experience) this is dangerous! Why? Now you have what looks like accurate point data, but it is not, and this can get lost during sharing or steps of analysis.
    – Simbamangu
    Commented Apr 4, 2018 at 13:56

1 Answer 1


Consider the dates of the occurrences, and try to to get (build, georeference) a map of the roads, railway lines, bridges and towns (villages, train stations) known or available at that time for the region, as naturalist usually departed from a known village and at least partially use an existing road or railroad to get to the areas where they got the specimens. Sometimes it really reduces the probable area of collection/ocurrence. If there some more ecological info on the species, you can rule out areas, eg. open vs forest lands, wetlands vs drylands, also with some ancillary info on the historic distribution of these ecosystems.

Although I would definitively not use these locations for training the sdm model, you could use the model results to reduce the location uncertainty of those badly georeferenced occurrence in combination with the data mentioned above.

Some recent papers address the bias effect of these uncertain locations and if using summarized environmental data can be used to compensate for this fuzzy location:

Where is positional uncertainty a problem for species distribution modelling? https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1600-0587.2013.00205.x

Quantifying the degree of bias from using county‐scale data in species distribution modeling: Can increasing sample size or using county‐averaged environmental data reduce distributional overprediction? https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5551104/

  • Thank you for your answer Priscilla. I am currently trying to obtain historical land cover maps and data that provides information on roads and townships. The hope is to cross-reference this information with tags from historical museum specimens to better geo-reference them. While the SDM aspect is not a major issue, I would really want to georeference them accurately as I want to sample these specimens for genetic analysis. Any thoughts on the latter? Commented Apr 10, 2018 at 18:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.