Hello forum.
I am currently writing my master thesis in biology about how effective iNaturalist is in it’s recognition of coleoptera. The project is still in it’s early phase, and eventhough my project is mainly concerned with the ‘biology-part’ I still have some questions regarding the app-data collection.
I can read that observations are given a georeference (or geotag), so that it is possible (if not obscured) to view the location of an observation. However, I am interested in if the geotag also affects the ‘species-suggestions’ that arrive, when logging an observation. Will the suggestions be adjusted based on what else has been found in the area or what is possible when concerning a general species distribution based on temperature ect.?
If yes, will it then effect the ‘pilot’-study I am doing on a museum-collection containing danish beetle-species (since they are logged as being found at the same location, but originally will be found across the country)? Is it possible to then de-able georeferencing?
I am completely new to the iNat forum, so I apologize if this question has been posed in a completely wrong place and I wish you all a great day.
Hi Mia, welcome to the iNaturalist forum! Do you have an account on iNaturalist.org yet? If you are adding photos of specimens, you might find it easier to do this using the website interface, because you will be able to batch edit the locations to reflect where they were collected. Using the app it will select the location you are standing (unlikely to be where the collection was made), and the website will make it a lot faster to manually select the correct locations. Quick tutorial on using the web uploader here.
That said, iNaturalist observations should be collections that you collected yourself, rather than digitizing other people’s collections. You can read more about that here. Depending on what your project is, I wonder if you could use the function of the computer vision feature, but not actually upload any observations? (Computer vision also might not be very good at IDing pinned specimens since it’s primarily trained on organisms that were found in situ, usually outdoors.)
As far as how location affects the suggestions by computer vision - the model looks at species globally that appear visually similar to the image, and then commonly observed related species that have been seen nearby are often inserted into the results. Some species suggestions will be both “visually similar” and “seen nearby” at that time of the year. It’s not as robust as looking at temperature or using species range maps, aside from locations of observations of that species that have been submitted to iNaturalist. (There are suggestions for refining that process here, for people interested in that discussion.) One of the iNaturalist staff did a presentation about the how the model works in October 2020. If that didn’t answer your question, please feel free to follow up. : )
Thank you so much for your answer bouteloua ! that is really helpfull.
For my pilot study i will not upload any observations, I am simply curious about how well and which species suggestions will pop up in the app under “what did you see” (and how correct they are, if the right species will be in the top-10-suggestion). However, I will take into account that pinned species is not optimal for several reasons.
Later in spring and summer, I will do a field study regarding some selected species of coleoptera, a mixture of what i expect the app will have no problems recognizing and taxa that could be harder to recognize (maybe they are rare or have many similair-looking species in the genus, or small, dull-colored ect.). So it is just to have a general overview, and something to do, while out-door conditions are not optimal. :) To make better hypotheses I will need to know a bit more about how the model works, so I will definately read the links you have posted!
I also do have an account in iNaturalist with some observations allready.
One other thing you may wish to check with rare species is if those taxa are even included in the computer vision model or not. I’m not an expert on this, but I believe the “cut-off” is “a taxon must have at least 100 verifiable observations and at least 50 with a community ID to be included in training” based on https://www.inaturalist.org/blog/31806-a-new-vision-model
Fyi if you are using observations which are not yours, and as noted above this means you should not be uploading them as observations, there is a page where you can simply access the computer vision system by providing a photo.
It will run the computer vision tool against the photo without creating an observation.
If you are a chrome user, there is also an extension you can add to your browser that via colour coding tells you how confident the system is in the recommendation (it will always recommend something, even if it has low confidence in its ‘best’ match).
The extension is called inaturalist enhancement suite. I’m writing off my mobile site don’t have the exact link, but you can search for the name.
And the retraining is only done perhaps twice a year, so the key cutoff is were there that number uploaded when the system was last retrained, not today.