Greetings! I am a biologist working with iNaturalist data to plan future field surveys. Any ideas on how to replicate the geoprivacy polygons to accurately display uncertainty in locations? For example, the accuracy on a certain observation may be listed as 27 km. However, when I download data from GBIF and then plot a 27 km buffer around that point, the circle is much larger than the polygon displayed in iNat.
So I guess a second question is: what does the accuracy “tag” actually mean? When I measure it out the polygons visualized on an individual observation, it seems like a ~15km x 20 km rectangle, so perhaps that accuracy is the diagonal of the polygon (25 km). If this is indeed the case, how to best recreate/visualize that information?
Perhaps a circular buffer with a 15km radius will do a decent approximation… still, that’s not quite the same as the random point is not the center of that 15kmx20km polygon… Open to any and all ideas!
Welcome to the Forum @wild_ecology!
Obscuration on iNaturalist uses the 0.2 x 0.2 degree grid “rectangle” that contains the true location of the observation. The east-west dimension of this grid of course varies with latitude. So coming up with a single radial distance to use may not get you what you are looking for.
I think the accuracy distance of obscured observations is based on the diagonal of the grid rectangle that was applied, this being the farthest possible distance from the displayed (obscured) point location to the actual point location for that observation. That distance will vary from about 22 km at the poles to about 31 km at the equator.
Probably the best and most accurate way to visualize the information outside of iNaturalist is - you guessed it - replicating the 0.2 x 0.2 degree grid “rectangle” for each observation.
EDIT: also, if the observer supplied an accuracy distance themselves, that may get added to the grid diagonal to get the total accuracy distance - but I would have to do some exploration to confirm that. So one might need to buffer the grid polygon by that additional distance for visualization purposes. I don’t think iNaturalist does that for display purposes - they just use the 0.2 x 0.2 degree polygon.
Excellent point, @jdmore! I haven’t looked into GBIF’s possibilities to represent polygons rather than points (plus accuracy). If that possibility is there, I’d favour exactly what you’re proposing: sharing iNat’s obscured locations as rectangles of 0.2 x 0.2 degree. Centroids of these rectangles with large accuracies are not a good representation outside iNat, and certainly lead to a lot of confusion.
Just to clarify, I believe what gets shared outside of iNaturalist is a random point location somewhere inside that rectangle (the same point that displays, along with the rectangle, within iNaturalist for obscured locations), not necessarily the rectangle centroid. Since that point could be near one of the rectangle corners, it requires twice as large of an accuracy distance for correct representation. With centroids one could use half of the diagonal for the accuracy distance, instead of the full diagonal.
Thank you for confirming my suspicion. I very much understand why locations are obscured, but sharing the random point to GBIF (and even visualizing it as a point and not an error polygon on iNat itself) strikes me as a sure fire way to add confusion, error, and wasted resources to many studies, as this is not readily apparent on GBIF or those not super familiar with iNaturalist. Obviously due diligence is the responsibility of those using open data, but I wish the error polygon was uploaded/displayed and not a random point. This random point implies a false level of accuracy, particularly when it falls in particular jurisdictions.
I haven’t looked into it myself, but don’t the iNat data on GBIF also include the accuracy radius? If not, they certainly should, as all point data (not just those from obscured observations) need to be evaluated in light of that radius. And the radius of obscured points will be appropriately huge.