Obscuring rectangle on obs with large accuracy circles may not encompass true location

#1

Previously discussed here (and maybe other places too).

“Because this observation is obscured, a 0.2 x 0.2 degree cell encompassing the hidden true location is shown in place of the normal marker.”

By definition, an obscuring rectangle should contain the true location of an observation. However, if an observer has used a large accuracy circle, the obscuring rectangle can and often does fail to include the true location. As a hypothetical example, I observed a squirrel in the bay area. I entered it with a large accuracy circle that does include the true (terrestrial) location. I then obscured the location and got an obscuring rectangle that pretty clearly does not encompass the true location.

unobscured

obscured1

One solution already suggested was to not use an obscuring rectangle if the area given by the accuracy circle is larger than the area an obscuring rectangle would have since the user has essentially already obscured the location.

4 Likes

Off-trail observations
Geoprivacy, Obscuring, and Auto Obscure Discussion
#2

That might not always be the case. In some cases the “center” of the uncertainty circle may indeed be the true location, and either there was a problem with the accuracy data, or the user thought they were obscuring the location by adding a large accuracy buffer.

Obscuring might be more reliable if the opposite was done – hide the accuracy circle if an obscuring rectangle is already in place. That way the “dot” is more certainly randomized, and the (in-)accuracy distance is still available in the observation data if needed.

1 Like

#3

Sorry, I’m not sure what you mean. The current behavior already is to hide the circle if there is an obscuring rectangle.

1 Like

#4

Ah, ok, I didn’t realize that, now I’m seeing the issue. It still concerns me that using the observer’s coordinates with the large uncertainty circle, in place of the obscuring rectangle, could unintentionally reveal true coordinates in some cases.

So maybe use an obscuring rectangle with different symbology, dashed instead of solid outline or something, to indicate that the true location might lie beyond the rectangle outline?

0 Likes

#5

If the size of the obscuration (is that a word?) is to be 10km then place the pin at a random point within 10km of the centre of the circle, and make the rectange size equal to the diameter of the accuracy circle plus 10km. That box includes all of the accuracy circle and the centre of it is at least as random as one obscuring position data with a 0km diameter circle. Might be a big box but that’s what the accuracy circle is telling us.

(ie there is no reason to position the box within 10km of a random spot within the accuracy circle, just base it on the centre)

(replace 10km with 0.2 degrees if you like)

–Tony

1 Like

#6

Another potential solution!

For accuracy circles already larger than a standard obscuration rectangle, though, I don’t think the obscuration rectangle would need to be any larger than the diameter of the accuracy circle – as long as the pin position is randomized within it. Just match the rectangle to the extent of the circle, and that should still capture the true location, if accuracy was specified correctly.

Thinking about this raises a related question, though – when creating the random point for an obscuration rectangle, is there a procedural “geofence” around the true coordinates, to prevent the random point from randomly ending up too close for comfort to the real coordinates? @kueda?

0 Likes

#7

There is not. If there was you’d have situations where a threatened organism that occurs in one small spot in an obscuration cell would appear everywhere in the obscuration cell except the true coordinates…thereby revealing the true coordinates the more people observe it.

1 Like

#8

Thanks, that makes sense. Guess I was thinking a pretty small fence, maybe 200 meter radius, compared with the 0.2 degree rectangle, which would take a huge number of observations to find statistically, and then so many people would already know where it is anyway… But maybe that wouldn’t be protective enough to be worthwhile.

BTW, does the 0.2 degree size apply even at high latitudes, where it would get pretty narrow? Observation space may be creeping north/south-ward over the next century…

1 Like

#9

i know someone is going to disagree with me here but if the uncertainty circle is so huge why display it on the map at all? The point isn’t meaning fun when you don’t know where you were within 1000 km (and you should probably gain some situational awareness so you don’t get lost in the woods).

Yes i know there are fringe cases but… do the fringe cases need to be on the range maps? Just put a huge square over the map on the observation page and leave the observation at that.

1 Like

#10

Personally, I’m less often looking at taxon range maps and more often looking at the map for an observation I’m trying to ID. A pin with a large accuracy circle is way more useful for IDs than no location information at all. And much less misleading than rectangles that don’t encompass the true location. I think what’s best to display on the observation map and what’s best to display on the taxon range map are different (but related) questions.

I think there are a few reasons people use large circles:

  • They’re trying on purpose to obscure a location. For example, I recently saw an elephant observation in an obscuring rectangle that was not in a place where there are elephants. In comments the user alluded to putting an accuracy circle around the entire country for the purposes of obscuring the location.
  • By accident. The user knows the location but didn’t notice the mistake or doesn’t know how to fix it.
  • They really don’t know the location. For example, a photo taken of a bison in Yellowstone in 1985. A circle is drawn around the entire park. (This is what the circle is intended for.)

Maybe the first time a user enters an observation with a large circle, a prompt could come up. It could suggest marking the observation obscured or private instead of using a large circle if the purpose is not to share the true location. It could also point to help documents on how to narrow the circle using the app or website for those that need a little assistance.
(This isn’t really a solution so much as a way to lower the rate of occurrence for cases that shouldn’t have large circles, but hopefully at least a step in the right direction.)

4 Likes

#11

It does. And yes, this means the obscuration area is smaller and narrower at the poles, but these are extraordinarily remote places. If there’s any evidence organisms there are under serious threat of poaching or collecting, we’d consider changing the obscuration cell size at higher latitudes, but I suspect exchaust from commercial air flight poses a larger threat to these creatures does than poaching.

2 Likes

#12

It sounds like the issue here is how we display this information, and not with the level of obscuration applied, so IMO, we don’t change the way the coordinates are obscured since that would not make the actual location any harder to guess. We want to convey two things:

  1. The amount of uncertainty
  2. The fact that the coordinates are obscured

To achieve both, I propose that for observations where the positional accuracy is larger than 0.1 degrees, we display such a location as a rectangle that contains the positional accuracy circle:

735bedd4e9927f8f876c130e9fd814aa361b4794 (red rectangle)

It would convey a larger amount of uncertainty, but at these scales I’m not sure that really matters, and it would clearly convey the fact that coordinates have been obscured, though not the actual area within which they have been obscured. I dunno, not a perfect solution, but the lack of an obvious perfect solution is why have never addressed this.

3 Likes

#13

@kueda this is what I had in mind for how to derive the rectangle, but without also displaying the circle.

If this is an auto-obscured taxon, I would rather see the rectangle without the accuracy circle, and with the pin location randomized within the rectangle, just like in a standard obscuration display. Only difference is that the rectangle is bigger to match the user’s accuracy level.

The reason being, I am still concerned that the center of that positional accuracy circle could represent the true coordinates in some cases. Those cases being,

  1. The user mistakenly thought that they could obscure their observation by simply specifying a large accuracy/uncertainty distance, instead of by also placing the “center” some distance away from the true coordinates, or

  2. The user made an accurate observation, and wasn’t intending to obscure the observation at all, but accidentally included a large uncertainty radius, perhaps due to erroneous data from their phone or other equipment. If it’s an auto-obscured taxon, we would still want auto-obscuration to work as expected, even if that wasn’t the user’s intent, and not reveal the user’s true coordinates.

Sorry if this is repetitive, just afraid I may have been clear as mud before…

2 Likes

#14

Ignore the blue circle in my illustration. I am proposing replacing it with the red rectangle and a circular map marker.

2 Likes

#15

Ah, got it. And presumably that marker is randomized? (Doesn’t look like it in the diagram)

I think that is sufficient to convey what you listed - amount of uncertainty and fact of obscuration.

3 Likes

#16

Well, the issue there is that I can still figure out the exact pin location by looking at how the square is centred, which will sometimes be the more-or-less exact actual sighting location.

What you could do though is this: image

Where green is the rectangle that will be displayed, and the red lines are equal to the location accuracy.

(The existing blue obscuring rectangle and the red lines wouldn’t need to be displayed)

1 Like

#17

True @reuvenm, I suppose if enough users were aware of the meaning of a larger than usual obscuration rectangle, and of how that rectangle is constructed, that could be an issue.

And I also hadn’t considered before that the position of the standard obscuration rectangle is also randomized with respect to the true coordinates. Meaning, I’m guessing, the true coordinates could even be right on an edge of the rectangle.

So that being the case, I agree that your solution is the only way to guarantee that the true coordinates are contained within the obscuration rectangle (your green).

I was hoping we could avoid applying that solution for observations with accuracy values greatly smaller than 0.2 degrees, which I am guessing are the majority. But even if the accuracy value is 5 meters, for example, one would still want to limit randomization of the rectangle so as to keep the true coordinates at least 5 meters inside the edges of the rectangle. And if we’re doing that, might as well just use your solution for all of them.

0 Likes

#18

You can figure out the true coordinates we have in the database, but in this case that information is not actually useful in finding where the actual coordinates are b/c by declaring a large uncertainty radius, the observer is already saying “these aren’t the exact latitude and longitude; the exact coordinates are somewhere in this giant circle.” We’re talking about situations where the actual, precise coordinates have not been added to iNat.

Regarding your diagram, I think you’re just proposing what I proposed, unless I’m misunderstanding you.

0 Likes

#19

In theory, yes.
In reality, I think accuracy circles are used really poorly a lot of the time. I see a lot of observations where the accuracy circle probably doesn’t encompass the actual observation, or where the accuracy circle is far larger than it needs to be. For this specific case, I don’t know how many observations there are with very large accuracies where the pin is actually pretty accurate as to the location, but I’m sure there are some. I wouldn’t be surprised if there’s quite a lot.

In particular, as mentioned above, I’m sure there are cases where people have been slightly confused and have used large accuracies as an attempt to obscure coordinates, yet have still placed the point accurately.

Your suggestion appears to be a square with side length of twice the observation accuracy, centered on the observation pin. Maybe I’m misunderstanding, but if not, this will allow anyone to locate the observation pin exactly if they want. In other words you haven’t actually obscured the observation at all.

My suggestion is a square with side length of twice the observation accuracy plus 0.2 degrees, centered on the centre of the normal obscuration rectangle. This will only provide you the same information as any other obscured observation, i.e. the 0.2 degree box the observation pin is found in.

3 Likes

#20

This is the best algorithm, and I would actually argue it should be applied to all obscured observations, regardless of their accuracy value.
Calculation for all observations would be:

  1. Disregarding accuracy value, calculate obscuring rectangle via the current method (0.2 x 0.2 degrees).
  2. Extend each side of the obscuring rectangle by the accuracy value (i.e. the radius of the accuracy circle). For observations with accuracy 0 m or undefined, the rectangle is the same as what is given by the current method. For all other observations, the square is enlarged, sometimes a little, sometimes a lot.

This obscures the pin equally for all observations, always includes the full extent of the accuracy circle, and displays as a rectangle rather than a circle to indicate obscuring has happened.

1 Like