Accuracy circle delivers important context to geocoordinates

General
1

Have been reading the thread linked below, which is already closed, so I can’t comment there.

https://forum.inaturalist.org/t/project-is-not-reporting-my-observations/82194/3

Seems OP @tim122 had some observations that were not included in a specific project. That problem was “solved” by manipulating accuracy.

I would think that on a data sharing platform like iNat, It might be important to mention: Location data including accuracy, isn’t something that should be manipulated or changed ad libitum. The scientific value of irreproducible or forged data is still zero, even when it comes with pretty pictures.

2

Looking at the original thread, it seems the user was just frustrated by a platform limitation where a large accuracy circle caused an observation to spill outside a local bioblitz boundary, excluding it completely. This makes sense as iNaturalist collection projects filter observations strictly based on their geographic boundaries. If an observation’s accuracy circle is too large and spills outside the project’s bounding box, the platform automatically excludes it to protect data integrity. While their fix of artificially shrinking the circle solved their immediate project issue, you are right that doing so could compromise the reproducibility of the data. In the case of the situation you linked I think it was reasonably adjusted though. I’m interested to hear what others think. Are you suggesting the real solution here is a feature request for how iNaturalist collection projects handle boundary overlaps, rather than users manually overriding their true GPS telemetry? Or something along those lines?

3

I am not really interested in a technical solution. I just think it’s important to discuss data quality and data manipulation openly.

I started out using iNat via its website uploader at the end of 2018, early 2019. Means i took fotos in the field with a simple camera and later tried to locate the whereabouts of the observation on the satellite image during upload on the website. I have to admit, this procedure resulted in very poor location data: Two years later, i had long switched to using the app, i tried to refind some of my plant observations from 2019. While i refound some, i also failed spectacularly to reproduce others of my own data points.

Reproducibility is an important character of science and scientific data. And for me, the reproducibility of location data became much better after switching to the app. I think one of the best advices one can give to anybody using iNat is probably: Use the app for recording and uploading and never ever touch location or accuracy.

4

Observation locations often have no accuracy data, just a latitude and longitude. When there is an accuracy radius, this is most often automatically generated (along with lat and long) by the GPS in the observer’s phone and included in the image EXIF data. If the GPS fix is imprecise the automatically generated accuracy radius may be quite large.

Of course, pretty much every photo is taken by an actual human who has their own recollection of when and where they saw this organism. That means that the observer often has the knowledge that would be needed to refine the accuracy radius—“Actually, I know I saw these mites somewhere in this small area.” So long as the observer is confident that the accuracy circle still includes the actual location, editing the location accuracy radius does not imply irreproducible or forged data. There’s no particular region to prefer inaccurate automatic location data over accurate manually corrected data.

5

I really don’t think that’s true. Mostly I leave the location unchanged on my own observations. But there are a myriad factors that cause incorrect locations, and fixing those observations is fine, too.

6

In my experience, e.g. while making a series of iNat observations along a road. There is still a low percentage of iNat observations with automatic geocoordinates that are off by a few 100m. This is easily visible as single dots on that days map seem to have jumped randomly. Sure put them manually back in line, if you want. Or if you have backup GPS data from another device, put these in.

How do you improve location data? And how do you correct automatic GPS data @rupertclayton?

7

I had a hard time refining the locations when uploading my older observations, and I know some of them are still way off… but I did my best.

My solution nowadays for improving location accuracy is to use the iNaturalist app in the field, whether I’m photographing with my phone or with a camera (even when I don’t have an internet connection).

When I take photos with my phone, I add them to the iNaturalist app right away and simply refine the location. When I use my camera, I create an observation without media, refine the location, and leave a broad ID or add a note so I know which photos it refers to. Later, when I’m back home, I upload the media to those observations and edit them as needed.

I actually tap on the location and accuracy fields, and this usually improves the precision. Sometimes the location generated by the phone’s GPS can be surprisingly inaccurate.

8

Then you are lucky in where you’re observing! I have a few areas that I regularly observe in where something (the angle of the hills?) consistently throws off the location fixes and/or accuracy circle. Sometimes the radius of the circle is 1km or more. And these are not large or remote areas: one is the campus nature preserve of the university where I work and another is a county park. So being off by that much (as well as being incorrect) often puts the observation in a totally different habitat. Unless I go in and estimate the locations and accuracy circles based on trails, satellite-map visible features, etc., the data will actually be much less accurate than they should be.

Similarly, if the project the person was trying to make sure their observations were included in was for a specific place, they presumably knew their observations were made inside that area and roughly where. As long as the adjusted circle includes all the possible actual locations, it’s fine to shrink it down from something that includes a bunch of locations you know are incorrect

9

The only bioblitz or the like that I’ve participated in where regions matter is City Nature Challenge, and none of my obs were close to the boundary of the two regions, but I have made some obs where I later changed the coordinates.

This oak I originally placed at the traverse point in the street, but later in the survey got approximate coordinates of the hole in the ground left by Helene toppling the tree.

This ivy and several other observations are at the back of a lot I’m currently surveying. The current locations are calculated from concrete monuments whose coordinates turned out to be both about two meters off. I’m planning to revise the coordinates and circles when I shoot the back irons with the total station.

And here’s an observation on a road: a tick. I know I was on the shoulder on the south side of the road, but where along the road I was I know only that I was between two exits which are several kilometers apart.

10

I’ve definitely had that experience (just a few observations that seem “off-track”). And I’ve had more dramatic displacements for all of the photos over a couple hours. And photos that entirely failed to be geolocated. For a while, I would keep Google Maps open on my phone and make a point of using that app to get a GPS fix before taking photos, which is a slow-but-fine workflow for plant or fungus photography but useless for animals. There are quite a few forum threads discussing GPS limitations of mobile phones.

Nothing more complicated than remembering what part of a trail I was on when I took a particular photo and doing my best to reposition the pin and accuracy circle to reasonably match my confidence in the location. Using the satellite imagery can be helpful, but it’s worth considering that the satellite photos are typically not precisely aligned either. If only a few observations are misplaced, then comparing locations and timestamps with others from the same day can really improve accuracy.

Great to have input from an actual surveyor on this!

11

I think this is the sort of thing being suggested. I’m glad you have a “low percentage” like this; I routinely have 20+ observations for the day that show up way out a km or more away from where I was. I go back in at the end of the day after I’ve uploaded everything and move these points back to exactly where I was. These observations often have very large accuracy bubbles, and once I move the point to the correct location, that huge accuracy bubble is entirely meaningless. Sometimes the automated location is correct but the automated accuracy bubble is way bigger than it needs to be (includes plenty of space where the observation most certainly wasn’t). In both those situations, it would be reasonable to shrink the bubble down. Similarly, if I know I observed something in a park during a bioblitz, but the automated accuracy bubble includes lots of land outside the park, it’s reasonable to shrink it down to be within the park. Obviously don’t shrink down an accuracy bubble if you’re unsure that it still includes the place you made the observation. But think it’s safe to say most users’ accuracy bubbles are generated automatically by the app. And if they’re bigger than they need to be, shrinking them down as small as they can be while still including the location where you were is not a form of fudging data.

12

I agree with this (and do this myself when needed), but I’ve also definitely seen examples where users just shrink their accuracy circles to get within a project place without other considerations, which I think is a problem and the one the OP was specifically talking about.

One reason for that is probably, if the user knew how the accuracy circle worked to begin with, they likely would have corrected an error and set it smaller or such that it was inside the location they were looking to be in prior to asking about it!

I also see that many users don’t have a good idea of where they are on the landscape apart from the GPS coordinates recorded by their phones (in fact, use of GPS-based mapping tools is associated with reduction in some navigational skills: https://www.sciencedirect.com/science/article/pii/S0272494424001907). I would guess that iNat Forum users are a highly biased subset of people who are much more proficient than average at knowing where they are and remembering it hours to days later (not bragging, but I definitely meet this description). I’d trust a Forum user much more than an average user if they told me that they were in the boundaries of a place. Many people might be totally wrong in this belief. I’d personally give the benefit of the doubt to the GPS (which is also generally conservative in my experience, erring on the side of larger accuracy values) over the average user remembering where they were and with what certainty some substantial period of time after the observation was made.

So in short, I do think that the OP identifies a real issue, though one that I don’t think is particularly common (since changing the accuracy circle is a bit of an obscure function). I’ve approached the issue by also noting that users shouldn’t shrink the accuracy value of an observation solely to get within a place boundary, but because they are certain that the true location lies within the circle. More abstractly, its similar to educating users how/when to use the agree button - not to achieve a specific end (like RG) but to accurately represent their own knowledge/data about the observation.

13

I completely agree. The location circles are totally arbitrary. The just go off of one point - whichever point is registered in the database’s record of a named location, and then use that as the center of a radius. This often results in asinine mixtures of habitat, such as the circle including a large part of an airport runway along with a richly diverse riparian habitat that is not even on the airport property. I have an intimate knowledge of every place I ever record an observation at, and I will almost always adjust the circle to encompass the area of habitat where the observation was made.

For instance, if I am at Tyler State Park, and see a butterfly in a wildflower meadow, then I will want the location circle to include the areas of the park that have large wildflower meadows, but I will NOT want that location circle to include part of the housing development that is adjacent to the park. Location circles are far too arbitrary to just let them be. It is better to either manipulate the point/circle that automatically pops up when you select a named location, or to select the specific location yourself using a satellite map, and then set the size of the circle based on factors such as adjoining habitat type, vulnerability of the specimen observed, etc.

14

No, that is really not good advice. I do not have an iPhone, so I cannot give you any real data on its accuracy. The Android iNat app, Lightroom app, camera app (Samsung, Itel and Redmii) and HD Camera app have all been used by me over a period of more than two years with well over 60 000 photos to check real data.

The results are simple: location and accuracy vary wildly and is often not recorded at all or sometimes so far off you might as well say the photo was taken on the moon.

The Android iNat app records an accuracy sometimes, not always (I had an instance for 4 days where obs recorded in South Africa ended up in Bangladesh with an accuracy of 10m). Samsung’s camera app adds location data (might be accurate) if the camera is continuous use for at least 30 minutes. Redmii I did not use long enough for proper data, but it never added location data. Itel’s camera app adds data after ± 15 minutes of continuous use. All of these apps stop adding the location data after 5-10 minutes of not using it and then using it again, none of them add accuracy.

Lightroom’s camera app does add the location data after ±10 minutes of being open. Anything up to the first 30 - 60 minutes of use is completely wrong (up to 60 kilometers/37 miles out in some areas I’ve used it). It does not record location accuracy. HD Camera suffers from similar issues.

I’ve used GAIA GPS, CalTopo and some others to try and record hikes on another phone as I am walking, with various waypoints recorded (especially if it is something rare or not seen before) and this has had varying results, mostly not worth it so I’ve stopped doing that as well, it just adds an additional thing to remember for no good result. I do these hikes for pleasure as well…

My opinion is that, with all of these issues, if you are going to leave the location and accuracy exactly as it is, you might as well not bother having it. A location being off by 60 km for one of these https://www.inaturalist.org/taxa/527591-Morella-cordifolia puts it in the sea or in the mountains, where it does not grow naturally at all… or these https://www.inaturalist.org/observations/317592392 and https://www.inaturalist.org/observations/210838176 add difficulty that is not necessary, especially if an identifier comes along two years after the fact - how will the user remember then?

The best thing to do is to walk and every 30 minutes or so or at a beacon that can easily be identified, stop and take a landscape image. Then it is only a window of 30 minutes that have to be checked and probably less than 2 kilometers in distance, which is better use of accuracy and location than 60km differences.

If you are on a well marked trail, take images of the trail beacons as you go (some trails have a marker every km, this is very helpful). Off trail is a bit more difficult, but you will be amazed how good a tree, a farm dam, a rock formation or a cross-road can stand out on a satellite image. Use google earth for better resolution. And sometimes (my experience near Kutaisi) there are pretty good recordings of cattle tracks on GAIA GPS and google earth, which are good beacons as well, bearing in mind that off trail hiking covers less distance than a clean trail in the same time. City or town walking is easy, just make a note of the address or city block every 10 minutes or so. But try your level best to add decent location data and some sort of accuracy circle below (my arbitrary figure) 50m or 150ft manually and always check the location before adding obs.

15

This is an incorrect description of how accuracy circles are generated in most cases. This description seems to only apply to circles generated when users enter a text ID for a location and let the system return a value for accuracy (which is generally algorithmically generated and may or may not make much sense).

Most accuracy values are generated by mobile devices around the GPS coordinates that they generate (the ways that iOS and Android apps generate these have been covered in other forum posts). A smaller number of accuracy values are generated by direct human input.

You talk about the relevancy of

for accuracy circles, but whether it contains all the same type of habitat is not one of the criteria for whether or not a circle would be considered to be “correct”. The only criteria for evaluating a circle’s “correctness” for iNat’s purposes is whether it contains the true location of the observation. For instance, I would not make a circle larger to try to accurately contain a patch of habitat. I suppose that a user could do this for themselves. But the data will be more useful to IDers and data end users if the circle is as small as it can be (while still being correct) regardless of the habitat types present within it. If users are interested in habitat data, there are other better options for capturing that (notes, observation fields, etc.).

16

Observe in airplane mode with an eBird list running to maintain precise and frequent GPS fixes.

While not fully reliable, for me this results in most observations having accurate fixes and precision values of <10 m, even in relatively remote and rugged areas.

17

But the automatically generated (a.k.a. “asinine”) location circle most often does not even include the exact point where the observation took place, so one MUST make the circle larger, or move the center of the circle, in order for it to be accurate.

I am speaking as an iNaturalist user who only ever uses a desktop computer with the full iNat website for entering observations. I have zero interest in using a cell phone to enter observations “on the fly”, as that does not allow me to go through dozens (or sometimes even hundreds) of photos of the thing I observed, and select the very best ones, edit them precisely, etc.

If a cell phone has a way of knowing where one was when they took a photo, and then that info is entered into an observation by default, then yes that seems to be the way to go for those who prefer to just use the iNaturalist app on their phone instead of using a proper computer for data entry.

18

That is a good way to look at it, as most casual users only encounter this situation as a result of place boundaries. Being aware of where you are in relation to a project’s boundary while you are in the field would seemingly help this problem significantly.

19

I think there are two different “automatic accuracy circles” being conflated here.

This form of automatic circle is only an issue for desktop users who enter their location by typing the name of the location. Since most iNat users submit from mobile devices, this isn’t the type of automated circles most users have to deal with.

This refers to the automatic circles generated by the mobile device when you take a picture, which generally do include the correct location, but it’s best practice to check them because sometimes the GPS glitches and puts the point in the wrong spot or adds a huge accuracy circle.

This may be true of the automated circles generated by typing a location in the browser, but not of the automated circles generated by mobile devices. They usually do include the location (but not always, so best to check them). In those cases “move the center of the circle” is preferred over “one MUST make the circle larger”, assuming you know where you were on the map.

I’d venture that this is by far the most common way location data is entered on iNat, given how much of the user base is on the mobile apps.

20

I don’t use automatically generated location data. I use my camera, which has no location. I upload on desktop. Everything is manual for me. Sometimes I do adjust the location data after I make it though most of the time not. Rarely I upload phone pics but location is turned off. (I think only once or twice ever did I take pics via the inat app to have it create location for me)

It is frustrating to me now to learn that a project will exclude certain things due to part of the circle being outside their boundary. I’m not sure on a fix for that. I just know my mindset in setting location:

The center of the circle is my best guess on where I was. The uncertainty radius is set large enough to encompass all possible locations I could’ve been. This means the circle will be large enough to contain many locations I know I wasn’t. If I was on a nature trail, the circle might include things far off trail, other trails, things outside a park boundary

I could understand by all this being frustrated that you know you were within boundaries of a project but due to circle radius you’re excluded. However. I do agree that scientifically it’s best to not change it. Because most observations the project might include will not be specially changed like that. Instead I would hope that the projects could change how they handle location data (not sure if this is something done by whatever project this is or how inat’s set up)