New DQA for landscapes (with no obvious subject)

General
1

Since this issue bubbles up on the Forum often, I started a Feature Request. But what I really wanted was to open up discussion on the Forum.

Description of need: We have a DQA for Not A Single Subject. That is a key and lock to one door. In that room are obs with multiple pictures, and each is of something different. My little dog, that bird (in a tree) and oh look a dandelion. There were years of discussion before we got that new DQA. More discussion when we got it. Still discussing today …

February 2019 requested 125 votes accumulated https://forum.inaturalist.org/t/easy-way-to-mark-multiple-species-observations/278

DQA added in February 2024 https://www.inaturalist.org/blog/90263-forum (scroll down to New Data Quality Assessment condition for photos unrelated to a single subject)

https://forum.inaturalist.org/t/easy-way-to-mark-multiple-species-observations/278/174 post in that thread from Carrie Seltzer

https://help.inaturalist.org/en/support/solutions/articles/151000171680-what-do-i-do-if-the-observation-has-multiple-photos-depicting-different-species-

We need a counterpart to the ‘single subject’ I would call it landscape – and include any obs with varied biodiversity and no indication from the observer what the particular intention for an ID is here. One picture with dog, tree, bird and dandelion. Clouds. Mountains. Seashore. Grassland. Forest. iNatters are engaging with nature as requested – but newbies still need to learn the rule that an obs is of one taxon. We also have projects which record wider views of vegetation (not the intended purpose of iNat, but those and other obs are here already)

https://www.inaturalist.org/projects/vegmaphoto-s-afr

Feature request details:

New DQA Landscape. This would also push the obs to Casual, so that other identifiers don’t see it in their Needs ID queue (unless they choose to include Casual). The observer is free to add an ID, notes or comment, and counter the DQA Landscape vote. Or an identifier can ID that mango tree, and counter the vote. Could also be used for beetle or daisy ? – altho there I prefer to choose one of the above, then willing to withdraw if the observer replies.

For these multiple species, with no clearly intended taxon we have a hamster wheel of threads on the forum.

For example https://forum.inaturalist.org/t/appropriate-use-of-dqa-for-broad-images-of-plant-habitats-without-clear-subject/82973/49

Feature Requests down the years https://forum.inaturalist.org/t/top-10-inat-feature-requests-of-all-time/67669/11

2

My concern with this DQA is that it would be an “intent-based” vote. What I mean by that is the other votes are based on some objective quality of the observation- was the photo taken where and when it claims? Is there an individual organism present in all the photos? Is the photo real or AI-generated? This DQA would be asking “did the observer really mean for a particular organism to be the focus?” So the same exact photo with the same exact data might be casual if the observer “means” to observe the river/rainbow/boulder/sunset, yet RG if the user “means” to observe the grass/trees/shrubs in the same image. Of course an observer can already submit the same photo multiple times to focus on multiple subjects. But as identifiers we’re asked to respect the observer’s intent, and if it’s unclear, to assume the user meant well. This DQA seems like it would be an entirely subjective vote for identifiers to use to get observations they don’t like to Casual grade, by assuming the observer intended to submit something that doesn’t fall under the definition of an “observation”. I think if it can’t already be taken to Casual through “no evidence of organism” or “as good as it can be at plant kingdom”, then there must be something in the photo that is an identifiable organism. And the assumption has to be that that is the focus unless the observer explicitly says otherwise.

3

And the first example here ? With the ‘rainbow’ comment. That is the explicit focus, but not iNat’s intention of which taxon. There is nothing we can do with it taxon wise, which accords with ‘respect the observer’s intention’

4

And you already have all the tools needed for that:

  • DQA, no evidence of organism
  • reviewed
  • move on

I fail to see how a novel DQA item makes any difference. Instead I worry that a DQA item for “I can’t tell what the focus is, and I know that nobody else can either” will be used as a silent image quality indicator, ripe for abuse.

5

Such observations may be annoying for identifiers, but they are not violating any guidelines and they are technically perfectly valid observations.

There is no rule that the subject of interest has to have a prominent place in the media evidence. The subject may be two leaves in the corner of the image, or a single bird call at 0:14 in the recording – if that is what the observer chooses to focus on.
There is no rule that observers have to explicitly mark or indicate what their subject of interest is (although in practice they will likely find that they get more engagement and responses from other users if they do this).

6

But I am unwilling to use – no evidence of organism – for a landscape which includes visible (if not identifiable) plants. That is an abuse of that DQA.

I do – Mark as Reviewed, and move on.

The issue comes to the forum from many people. Sometimes daily. Today’s question was not mine.

7

That isn’t very strong evidence for this being a widespread problem. My impression is that the multiple-photos-with-different-species problem is much more common than the landscape-with-no-obvious-subject problem.

Also, is there any real difference between a landscape and, say, a “groundscape” of mostly dead leaves, bark and twigs? I seem to see many more observations like the latter, where perhaps an insect/spider disappeared underneath some detritus before the observer could photograph it properly. Of course, these kinds of “groundscape” photos often include heavily camouflaged subjects that are very easy to miss. So in addition to the potential image quality abuse previously noted in this thread, I would also worry that this DQA could inadvertently remove some perfectly valid observations from the needs-id pool, simply because some identifiers failed to find wally/waldo.

Mere ambiguity doesn’t seem a strong enough reason to demand a special DQA. It’s certainly frustrating to deal with sometimes, but I think it’s just an ineliminable part of the identification process.

8

If there are visible plants, why not identify it as “plant”?

I do agree that there are gaps, but don’t think a landscape DQA is the right solution. Maybe because it isn’t exactly addressing the right problem? I’m thinking about it.

9

I think a canonical case for “no evidence of organism” are observations like an empty field with the ID “Blue Jay” and a caption like “there was a Blue Jay here but it flew away before I could get a photo”. So in this case there is no evidence for the organism which the user stated was the focus of their observation, so it’s fine to click this, even though there may be identifiable e.g. grasses in said picture. I think the “rainbow” example is similar, where the observer made it clear what their focus was, and there’s no evidence of a corresponding organism (the focus isn’t even an organism!), so I think in that particular case it’s not an abuse of the DQA to mark it as such. Though I do think, if doing so, a short comment explaining what and why you are doing is called for.

That’s just that particular case, I do agree marking a landscape photo as “no evidence of organism” where the observer hasn’t added a caption suggesting an invalid subject is an abuse of the DQA. But I don’t think this is best fixed by adding a new DQA item. I know such observations can be frustrating, but I think leaving a comment and picking out an individual organism which can be identified will lead to a much better new user experience than just marking some DQA item and confining it to the Casual abyss.

Hopefully one day we get better onboarding, but I think for now a lot of new users are going to be quite confused by what exactly an iNaturalist observation is supposed to be, and are going to submit things that probably go against best practices, and while I understand this can lead to frustration for identifiers at times, I think it’s better to gently nudge users in the right direction than throw out their observations, especially when it’s quite easy for identifiers who don’t want to deal with this to just mark as reviewed and move on.

10

The only problem with marking such observations reviewed and moving on, is that they accumulate for others. The older of “unknown” observations you go through, the higher proportion of landscapes and random shots of the ground you will find, left there by all the previous identifiers who also didn’t know what to do with them. These take time that could have been better spent on observations of specific organisms.

Something very broad like “vascular plants” for that blurry treeline on the horizon, still takes two people identifying it and the second knowing to mark “as good as can be” before it moves to casual, which seems like a clumsy way of doing things.

I agree we need a DQA for landscapes or ground shots with no clear subject, and in my area I run into a lot more of those observations than ones that need to be split, or have the wrong date or location. Similar to the number of cultivated plants that need to be marked not wild. It’s very frustrating not having a DQA that’s clearly meant for them.

I do share concerns that it could be over-used, as there are many photos that should have been cropped, but where it’s reasonable to assume which organism was the most likely focus - for example a showy flower. I worry some frustrated identifiers could get over zealous.

I’ve seen other suggestions that could address that, such as sending the observer a notification for DQA votes and how to fix them. In this case, “Identifiers were unable to determine which organism in the photo you would like identified. Please crop the image or leave a comment explaining.” The DQA could automatically reset once photos are edited, a comment is left, or a broad ID is made, if any of those things happen.

The current system is more like first several identifiers ignore it. Finally someone asks the observer for clarification. 9 times out of 10 they don’t answer. More identifiers waste time looking at it. Finally someone sees the old question that was never answered, and makes a broad ID like “plants.” It sits at that level for another two years until, if we’re lucky, someone bothers to confirm “plants” and mark “as good as can be.”

Almost more frustrating to me than landscapes where nothing is identifiable are those shots where everything would be if the observer would just pick one. There might be turf grass, clover, plantain, dandelions, and fallen maple leaves scattered evenly throughout. None centered. None blooming. No initial ID. No second photo cropped to a specific organism. No answer to being asked “Which of these plants would you like identified?”

Yes, we can pick one at random to get it out of the way of other identifiers, but if the observer doesn’t care and it’s a common yard weed not very interesting to science, it feels so futile to do so. I would much rather a single click DQA and an automatic prompt to the observer to fix it.

11

I think rather than a new DQA, what is needed here is better guidance on what to do in these situations. As others have indicated, there is technically nothing wrong with posting a broad landscape photos with several different organisms in it, but it gets frustrating having to deal with them. Different identifiers seem to have different strategies - choose one organism to ID, ID at a higher taxon and mark yes to “ID is as good as it can be”, vote “no evidence of organism”, etc. If the observer notes in the description/comments what they want, that is usually easy enough to follow but even that can cause problems - e.g. in the scenario listed above by @yongestation in which the observer indicates that their subject is a blue jay, but the blue jay is not actually present in the photos.

All of these possible strategies are relatively easy for an identifier to go with, but the hesitation and problems mostly just come from the fact that people aren’t sure what to do. If we had better guidelines and it was clear what we should do in situations like this, I think these observations would get dealt with much more quickly. My current strategy is to ID to the highest level that incorporates everything that might reasonably be a focus and leave a comment asking the observer what they would like identified. If they don’t reply within a set amount of time (usually 2-3 months) I will usually then choose one to identify and indicate which I’ve chosen. This of course though takes time and means I have to monitor those observations so it’s obviously not an effective strategy for those who want to deal with these quickly. But I do think that better guidelines would drastically cut down on this issue.

12

My problem when I come across these is that it’s often something that’s fairly clearly a dicot (or at least ‘vascular plants’) but not identifiable past that. Do I agree with ‘plants’ and tick that it’s ‘as good as it can get’ just to get it out of the way, even though it could go further, just not to a useful level? Or do I take it down to dicots and then be forced to wait maybe another two years for a second person to confirm and tick?

A DQA would enable it to be dealt with by one person, and potentially in a way that would make it easy to look through observations thus made casual if anyone wanted to go through and try to rescue things. Having said that, I like the idea of DQAs also triggering notifications to observers with hints on how to improve their observations.

13

That is on me. I find an ID to plant Kingdom not useful. And will not use them myself. Which is why I methodically clear IDs down to Epifamily for the Cape Peninsula. Family I expect to trigger the relevant identifier. I would rather keep going with IDs in a batch where I can trigger the taxon specialist’s filters – and get a finer ID.

14
  1. A DQA to Casual is a courtesy to future identifiers.
  2. A relevant DQA would be searchable and sortable.
  3. It is ONE vote, which is easily overturned if abused. Not a train smash.
  4. And, we each have our own blinkered view of iNat. I choose to pick over Unknowns, Newbies (who will get a kind and thoughtful response), trapped at Disagreement. So I see landscapes every day as I ID. If you are a taxon specialist perhaps you never see the issue, and the strings of comments from identifiers. Just as I less often see tangled discussions that make my head spin. Too many legs !! And 2 families separated by IN!!
    https://www.inaturalist.org/observations/115070117
    https://www.inaturalist.org/observations/11133612

Is there a chunk of iNat text in the guidelines already ? Don’t landscape, choose one of the above.

15

But, if we had a dedicated DQA, taxon specialists could retrieve that batch. A second pair of eyes, programmed to see lizards, will notice the gecko behind the blurry leaves. Or that dicot twig … is a snake. They do bubble up in Kingdom Disagreement to Life.

16

Honestly, at present these often just get shuffled into Plants anyway, so it’s just a question of whether you’d like them buried in the mountain of plant (down to dicot, probably) IDs or removed with a hopefully searchable DQA, which I would imagine would contain fewer observations. Though I wouldn’t use a ‘landscape’ DQA for such groundscapes unless they’re far enough away that nothing is distinguishable anyway.

17

I’m never sure about that either. One problem is that different identifiers are responding differently. The person who called a landscape “plants” can see that there are some dicots, but may also be some monocots, some conifers, and how much of that green in the distance is from moss-covered rocks rather than vascular plants? One view is that “Based on the evidence” (which does not include any indication of focus on a specific organism) it cannot be improved beyond plants. “Respect the observer’s intention” is here being expanded to include respecting the observer’s lack of intention. There is no indication they were photographing any particular order of plants, so some may feel we shouldn’t identify one at random.

If I’m the first person to take it on I’m more inclined to pick one that I can get as specific as possible. But if another identifier had called a treeline “vascular plants” I don’t want to come along and say “I see some dicots in it” creating the need for a third identifier. So I usually just skip those, hoping the next person who believes it should be left at a level that includes everything in the photo will be the one to agree and mark “as good as can be.” I’m sure too many of us take that approach though, which is why they are so slow to clear out.

Edited to add: I’m sure this goes the other way too. If I’m the first to ID it and I pick “dicots” someone who thinks “That should have been left at vascular plants, since the observer didn’t indicate whether they meant most of the treeline, that one possible pine on the left, or the grassy field in front of it.” is not going to bother disagreeing to “vascular plants,” so we sort of need two identifiers who also happen to be using the same standards just to get it to some useless high level ID, so it can be sent to casual.

18

Is it rude to find it wryly amusing that your desire to avoid the need for a third identifier… just requires another identifier anyway?

And I take your point about the difficulty of different people treating them different ways. I guess that’s another thing a DQA would achieve: creating a specific way people can deal with them consistently.

19

Except it’s not clearly a data quality issue. I think we’re all in agreement that ambiguous observations can create problems for identifiers in various ways. The question is whether the DQA is the right mechanism to handle such cases (assuming they’re a big enough problem to demand special handling at all). The current consensus in this thread seems to be the DQA isn’t the right tool for this job.

Which is another good reason for not doing that. Most of these sorts of makeshift remedies are effectively just hackery: they’re using pre-existing mechanisms in ways that they weren’t specifically designed for. Even worse, they often don’t really solve the problem at all, but rather just move it somewhere else.

We already know that the right way to deal with ambguous observations is to ask the observer for clarification. This can and is already done right now using comments - but, ideally, we’d also like some formal way to flag the observation as “needs further info” which can act as a filtering option. That way, most of these observations can just remain where they should be: in the unknowns (since that’s what they truly are).

While we’re waiting (possibly forever) for something like this to be implemented, we could fall back on some more hackery to achieve something similar. I’m not sure if this has been suggested before, but an observation-field could act as a “needs further info” flag, and then the &field / &without_field url parameters could be used for filtering. Establishing a convention like this might be a good way to test the potential of the general idea and also provide a simple way to judge how many observations might be affected.

20

So you’d prefer them to be left buried in a growing pile of ‘unknowns’, which is just as inaccessible? I’m not really convinced that’s the answer either.

Yes, I tend to leave a comment for the observer if it’s recent, though less likely for older ones because of an even lower rate of response, but very few ever respond. Hence, we’re left with what seems to me a growing difficulty of finding ‘useful’ observations because now it’s the potentially identifiable observations that are buried in increasing quantities of ‘no one knows what to do with this’.