At a popular state park there are 3919 observations by 85 different observers, all at the same spot–the Headquarter’s Building. Apparently, people go observing at the park, but rather than geotag, they put all the observations at the headquarter’s landmark with an accuracy circle in most cases not including the site where the organism was photographed. Technically, these should all be casual, since the locations are not accurate. Or do people think this is ok? I wonder how pervasive this issue is elsewhere.
The observers are likely just entering the name of the park into the search bar, not placing the points manually. For locations entered like this, iNat pulls a location and accuracy circle from Google (I think). If you look at almost any popular location, you will see a collection of points like this. Often the accuracy circle is fairly large and includes most/all/way more than the focal location. But there are some places with weird geometries (or where the place isn’t defined well in google) where the accuracy circle isn’t great. If you’re sure the true location isn’t in the accuracy circle, you can downvote “Location is Accurate” in the DQA.
16 Likes
this happens everywhere. even in the state park that i assume you noted, there are two different locations that have high concentrations of observations, based on where probably G Maps was defaulting its geocoding at the time:
two spots:
- https://www.inaturalist.org/observations?nelat=29.370656765555115&nelng=-95.62714576721191&subview=map&swlat=29.370581967077573&swlng=-95.62723159790039
- https://www.inaturalist.org/observations?nelat=29.371030757118465&nelng=-95.63186645507812&order=asc&page=6&subview=map&swlat=29.370955958915708&swlng=-95.6319522857666
temporal distribution of the observations:
- https://jumear.github.io/stirfry/iNatAPIv1_observation_histogram?interval=month&d1=2020-01-01&nelat=29.370656765555115&nelng=-95.62714576721191&subview=table&swlat=29.370581967077573&swlng=-95.62723159790039
- https://jumear.github.io/stirfry/iNatAPIv1_observation_histogram?interval=month&d1=2020-01-01&nelat=29.371030757118465&nelng=-95.63186645507812&order=asc&page=6&subview=table&swlat=29.370955958915708&swlng=-95.6319522857666
i don’t think the issue is that places aren’t defined well exactly. i think this usually has more to do with the fact that Google Maps needs precise locations when providing directions. so their goal when they define their places is not necessarily to encompass the entire boundaries of the place.
8 Likes
I’ve seen some instances where Google doesn’t include all parts of a park or other location - most commonly on less “popular” or well-known natural areas (like land trust properties or state water conservation areas vs big national parks). Sometimes ones with outholdings or weird geometries (a bit sticking off in one direction). The circle for those locations usually does a good job capturing the part of the location that Google does include or the bulk of the place, but not the weirder bits. On the other hand, the center points for locations in Google are generally quite good/precise and make sense (like being on Park HQ) in my experience.
4 Likes
This is a very common practice. This is an expected outcome when people (like myself) upload images from cameras without GPS capabilities and must rely initially on a location provided by Google Maps. It does take an extra step to relocate a location pin to a precise site after getting that general location, but in many instances, the OP may have visited many locations within a park, refuge, or forest and is unable or unwilling to individually reset the location of every point. You are correct that this is “inaccurate” in a literal sense but it is a level of fuzziness I think we have to live with on iNat. Otherwise, vast thousands of useful observations would be needlessly consigned to Casual.
In my own experience, I see this happening at my former place of work, “Balcones Canyonlands National Wildlife Refuge”, a 45,000-acre region with a patchwork of Federally and privately owned tracts. The Google Maps point for that placename (title) is actually on a private tract that is not part of the Refuge but there is nothing we can presently do about it. I just have to understand and accept that level of generality. It’s easy to recognize such points because the pin will be located in the middle of the words “Balcones Canyonlands National Wildlife Refuge” on the map. In cases I’ve tracked down, the observation is actually somewhere within a one to five mile radius of that map title.
24 Likes
It’s imperfect but I personally wouldn’t mark them as being inaccurate in DQA, especially if there’s that many observations with the same issue. If anything, I’d comment and tell the user it should be corrected. And hopefully they correct it.
12 Likes
Have you tried geotagging the photos? Whenever I use my DLSR camera that doesn’t have GPS, I use the app Geotag Photos Pro on my iPhone to record the track and when I’m back home, I attach the locations to all the photos. Works very easily. I wonder if many people don’t know how simple this process is.
6 Likes
I think the better way to think about this is that the posted location of every observation is wrong, even if only by a few centimetres, and so the question is what level of inaccuracy is tolerable? Thankfully we have the possibility to provide an accuracy circle on each observation, meaning that each end user of the data can decide their own threshold of inaccuracy based on their own purposes.
I try to give a precise location within a few metres for all my observations, but it is not always possible. I have photos that I took on analogue cameras years before iNat was invented, and which I know were taken in a particular reserve or town, but am unable to be any more specific. These observations are still valuable (arguably very valuable, given they are historic by comparison to the vast amount of data on iNat) and I think, so long as I set my accuracy circle appropriately in each case, I don’t believe there should be any real limit to how wide that circle of inaccuracy can go.
13 Likes
This doesn’t mean that the information is useless. If you are looking for information in a larger area like a county, state or continent the information is relevant. In the end, the researcher decides whether the observations are relevant to their study.
14 Likes
I manually add location on my phone for some pictures but it actually often has the same issue as described in the post here. Even if I add it as coordinates, a lot of observations end up having the exact same location when I upload them, when they really should have a difference of, say, 20-100 feet. Sometimes I adjust the exact pinpoint when uploading but a lot of the time I’m like “well, close enough, anyways”.
3 Likes
This is a common problem, and I agree it is alarming. I made a feature request some time ago to make it easier to see how locations were created, but it didn’t get much traction.
You’re describing the difference between accuracy and precision. In the original example given, the locations are neither accurate nor precise. Locations that are accurate but not precise are an improvement on this. I would be comfortable with having high confidence (e.g., 90%, 95%, 99%) that the true location is within the “circle of accuracy” but even the “accuracy” [precision] values of some GPS enabled devices have a disappointingly lax definition, because they offer standard deviation instead of standard error:
For iOS no definition is provided, although it must exist, despite the impossibility of a perfect statistical definition.
I’d only note that this issue is much much worse on many other citizen science sites, where the workflow to upload observations with an accurate and precise set of point coordinates is less accessible. One of the things I really appreciate about iNaturalist is that it is easy to generate high-quality data with precise coordinates.
9 Likes
I think it’s also relevant to point out that many GPS coordinates for museum specimens on sites like GBIF and their uncertainty values are generated in a very analogous way. For most historical records created before the advent of GPS tech in the field, the description is textual. Things like:
2 mi NNE of Town X
YYYY Park
ZZZZ County
There are various routines that will generate a coordinate and uncertainty value for locations like these, and it is pretty much the same idea as what iNat does via Google. I’ve spent a decent chunk of time geolocating old locations from collections like these, and I think it’s quite likely that the locations from iNat/Google are on average equal or better in quality than those for many historic observations (which generally means through the 80s or even 90s). The location coverage of Google etc. now is much better than the geolocation software was even 5-10 years ago when many collections were/had digitized a lot of their collections.
On the whole, I would guess that, given the recency of the data, overall location accuracy on iNat is much higher than professional collections from the distant to recent past (when the bulk of collections were made), and probably a little worse than contemporary professional collections. In my mind, this isn’t a major worry.
And, as always, it is on the end user of the data to decide what they can/cannot profitably use. I’d much rather have the data and the decision to exclude it if I choose rather than not have the data at all due to someone else’s fear it might be not be accurate enough.
21 Likes
Dan, I use small Canon point-and-hope cameras for portions of my iNat efforts. There are any number of phone apps which might allow me to keep track of a GPS-ed path, and if I were energetic enough, I could even take an exact GPS point for every observation I make with those cameras…but I’ve found none of the track apps to be friendly to use for post-processing. I’ll look into the one you mention. Also, since I use my iPhone heavily (in tandem with the Canons) for capturing many observations in the field (with the native camera and/or iNat app), the battery use for GPS track apps on top of that might be excessive, especially if I expect to be away from a recharge for most of a day in the field.
In practice, for the vast majority of photos I take with those small Canons, I can manually place a pin (in batches) at upload within a few hundred yards and then add an appropriate circle of uncertainty for them. I’m comfortable with this level of location accuracy. When most natural history data are rolled up at the county, watershed, or ecoregion level for scientific analyses, the locations I’m providing are more than sufficient, even if they are slightly inexact. I take many/most plant pics with my GPS-enabled iPhone 14, so there’s no problem with those.
4 Likes
When I clicked through the obs for the two locations that pisum linked, the accuracy circle that is large enough that there’s no way to say for certain that many didn’t occur within that circle. It’s not like the circle only encompasses the parking lot or the building itself – it encompasses paths, waterways, wooded areas, etc. Unless one is absolutely certain the observation didn’t occur within that circle, such as recognizing a background feature, they should not be marked casual, in my opinion.
My DSLR does not have GPS and therefore will not record the location. I have it linked to a phone app that will provide location data – but it tends to do so at only certain points (when the connection is good?), so that the images will all be listed as having come from one or two points during my trek, rather than exactly where each was taken. No more accurate than what is happening with people tagging the HQ (less so, since it defaults to a miniscule accuracy circle).
As gcwarbler mentioned, that GPS link app is a major battery drainer, and even though I carry a huge backup battery pack, it is a problem and my phone has been drained multiple times before I was done birding. I also am using my phone to run eBird (which gives me a track of where I visited, at least) and the camera on the phone for close-up obs.
I do my best to move the pins and provide an appropriately sized accuracy circle, but often when I was out for several hours, my memory of where each picture was taken is going to be less than perfect, so sometimes I have to generalize to a park with a large accuracy circle.
In addition, when I’m taking pics of birds, especially on lakes and swamps, I’m using a super telephoto lens, and the GPS is going to be where I was standing, not where the bird (or dragonfly or whatever) I took a picture of was. Those are always a guess as I try to remember the angle and distance… often I have to just make the point encompass half the lake or something. If I stood in one place for a bunch of them, they’re going to be just like these.
7 Likes
I intentionally turn off all location data for my photos (unless they’re taken in the app directly) and manually place each photo. Most of mine are from my cameras rather than my phone anyway.
It’s a bit silly I suppose, but I don’t need even more things tracking my every move, and turning off all that location data also makes your batteries last longer.
3 Likes
there are some observations where it’s obvious that the organism can’t be found in the circle. for example, https://www.inaturalist.org/observations/152316244 shows a duck in water, and the circle does not incorporate any bodies of water.
however, it is possible to surmise that all the observations with the exact same coordinates and accuracy were recorded using the same method, and i think the proper interpretation here is just that the method used to record the location in these observations is a different method than one that tries to encapsulate the location of the organism within an accuracy circle.
i’ve ranted in other threads that there’s absolutely no consistency in how locations are determined across all observations in the system, and this is just another example of that fact. that said, i would not consider the location of the example duck observation incorrect even if i can say for sure that the duck was nowhere within the accuracy circle. instead, i would say that the location of the duck is correct if you interpret it as falling within the State Park, as defined by the geocoded location of the Park by G Maps. so if i came across these as a researcher, rather than throwing them out entirely, i would just adjust the locations of these observations as needed to make them conform to whatever rules i’m using to determine locations.
4 Likes
We’ve had a spate of forum posts deploring the accuracy/precision of locations or timing for our observations. (The photos in one observation might have been taken 20 minutes apart!) I find myself annoyed.
Possible conflict of interest: My camera doesn’t do the GPS thing. I map observations by hand, sometimes to a few meters of accuracy (as far as I can tell from the map). Sometimes, I photograph at a park and report all observations from a central point in the park with an accuracy circle large enough to include the whole park, or at least all the parts I visit. I think that for many purposes that’s good enough.
People should do a better job of putting accuracy circles around their locations, but this is not obvious to most users. It’s something we have to work with, watch out for, and where possible teach people about. But observations with less than perfect accuracy or precision are not necessarily useless. They’re not even misleading for people who are aware of what citizen science projects can accomplish.
How much precision you need depends on what you’re using the data for. If you want to relocate an uncommon plant, you need great precision (not to mention accuracy). If you’re mapping at the level of U.S. counties or states, an accuracy circle of a few hundred meters is just fine. In fact, a it’s usually OK if the plant was outside the accuracy circle reported (but somewhere nearby). Some iNaturalist data isn’t good enough for your purposes, but that doesn’t mean it’s useless for everyone.
If the plants are all reported to be at the the headquarters of the park, the probability that they’re all in the park is very, very high. Probably few are actually at the HQ, but for many purposes that doesn’t matter. (Example: making a species list for the park). Terrestrial plants marked as being in the middle of a lake are very probably along the shore of that lake, whatever the size of the accuracy circle.
Don’t throw out data that many people will consider good just because it doesn’t meet the highest possible standards.
13 Likes
With Geotag Photos Pro, the process is very simple:
- I open the app on my iPhone and there’s a time/date synchronization feature where it just shows the current date and time down to the nearest second. I adjust my camera’s time to match this.
- I start recording a track with the app.
- I go about the day, taking photos.
- When I’m back home I stop recording the track (or do it earlier), then upload it.
- I copy my photos to a folder on my PC.
- I run the Geotag Photos Pro program on my PC, selecting both the track and folder containing the photos. It then adds GPS coordinates to all the photos in one button click. Then you’re done.
The app continuously records the track, so the photos will all have accuracies as good as what your phone sees in the field.
1 Like
Canon’s Camera Connect app itself should be able to – among other things – record locations that you can then save to your photos while they’re still in your camera (before you download them to your computer for post-processing). (every major camera maker has their own apps that handle this sort of thing, as far as i know.)
some relevant links:
In Sugarloaf Ridge State Park (in California) where I used to work, iNat has 1202 observations placed at the same spot, in the location on the map where Google Maps shows the name of the park.
Most of these have location accuracy of 1.19km. This is the default location and accuracy when one types “Sugarloaf Ridge State Park” into the upload page. I’m curious where that accuracy number comes from, as it is different for different locations.
5 Likes