When you do annotations, are you annotating what can be seen in the photo, or the organism itself? This may seem like a picky question, but it has big implications sometimes, and what I can find about this issue is sometimes contradictory or at least inconclusive. Here’s how the issue came up:
I recently annotated a photo of a fallen sourwood inflorescence as “No live leaves.” Another user disagreed with this on the basis that we couldn’t tell if the tree had leaves from the photo, so that category should be left blank.
We seem to have two reasonable, but different, interpretations of what annotations are. My assumption was that I was annotating the observation - if someone wanted to find out what sourwood leaves look like, or train CV models to ID sourwood leaves, they won’t find this observation useful. However, if you want to research phenological stages (something I wouldn’t personally do with iNat data quality, but never mind that), then my “no live leaves” would be misleading and likely just plain wrong.
But for the “Flowers and Fruits” category, this brings up another problem. If the annotation is about the organism, then I would rarely feel comfortable marking “no flowers or fruits” for an observation of a tree. Decent-sized trees are almost never photographed in sufficient resolution or from enough vantage points to be sure there is not any flower or fruit on it somewhere out of (or even within) view.
And turning to animals, there is the “Evidence of Presence” category, which is about the observation itself, and the “Alive or Dead” category which is clearly about the organism, not the observation.
So in summary, we have some annotation categories that are clearly about the observation, others that are clearly about the organism, and some that can go either way. What do we do about those ambiguous categories?
I don’t have an “official” answer to your question but here’s what I do in certain instances:
Flowers and Fruits:
If I can see a flower bud, flower, and/or fruit in any photo, I add one or more annotations.
If none of the above are visible in any photo, and there is no evidence of flowering whatsoever, I annotate “No Flowers or Fruits”.
If none of the above are visible in any photo, but there is some evidence of flowering (perhaps a bare flower stalk or an empty seed pod), I leave the annotation blank.
Leaves:
If I can see breaking leaf buds, colored leaves, and/or green leaves in any photo, I add one or more annotations.
If none of the above are visible in any photo, and a photo of the whole plant is included, I annotate “No Live Leaves”.
If none of the above are visible in any photo, but a photo of the whole plant is not included, I leave the annotation blank.
My understanding is that the annotation is referring to the organism, and if it’s unclear which option applies, the annotation shouldn’t be used. It’s worth noting that the CV does not “see” annotations, so they won’t impact CV training one way or the other. As far as someone looking for images of leaves, they’d probably search for ones with a “green leaves” on that annotation. Whether it’s listed as a “no leaves” or left blank, the images would be excluded from such a search either way. The point of the annotations I believe is to have phenology data. Annotating “no leaves” on all observations where leaves are simply not visible wouldn’t really add to the phenology information. Some of the other annotations have a “can’t tell” option, which is what I’d select here if it were an option. Given the options available though, I think skipping the annotation since the relevant parts of the tree aren’t showing is the best option.
Agreed. I would never use any annotation if the image isn’t sufficient to determine which option to apply. I would want to distinguish between “no flowers” and “I can’t tell”. Negative data is still data, and I don’t think “no” and “I can’t tell” should be conflated.
No flowers or fruits: No flowers, fruits, or seeds (in whole or part) are visible
The word “visible” to me implies “visible from the evidence provided”, i.e. it doesn’t matter if there might be flowering parts out of frame, since those are not visible.
No Evidence of Flowering (within Angiospermae): Media provides no evidence of reproductive structures
Admittedly I am biased because a while ago, I made it a goal to annotate every plant observation that I’ve IDed. I go through the observations in the Identify tab with the “Reviewed” box checked, and leaving an annotation blank is problematic because then it fills up the queue. Which is why I made the call to just annotate “No Flowers or Fruits” if I don’t see any in the photos.
There are some edge cases, like if I see the flower stalk but the flowers/fruit/buds are out of frame (example). I used my best judgement and annotated those as “No Flowers or Fruits”. I’m glad they changed this from “No Evidence of Flowering”, which would have ruled out all reproductive structures including presumably flowering stalks!
Flower stalks are evidence of flowering in the past, not at present. Phenology is about what is occurring right now, not what one assumes is occurring or what has occurred in the past.
I think the general way that the annotations are presented and described suggests that they should apply to the observation rather than the organism. In fact even the “Alive or Dead” category to me is clearly about the observation, simply because of the “Cannot Be Determined” option. I don’t think “Cannot Be Determined” ever actually applies to an organism - it’s either alive or dead, and this should be easy to determine unless there are some extreme edge cases that I’m missing. But in terms of an observation, this category makes sense. Maybe you have just found a feather on the ground. Is the bird that it came from alive or dead? It’s impossible to tell, because the bird isn’t there. But that’s got nothing to do with the organism itself, its based entirely on the evidence that you have of it. So I think I would generally say that annotations are for the observation rather than the organism. That said though, I generally don’t annotate things with “no evidence” or “cannot tell” unless it is very clear, and I mostly prefer to leave these blank if I can’t tell.
Since I most often use annotations to find pictures of an organism in a particular state, I always assumed it was from the available evidence. If the plant was flowering but none of those flowers were in any of the pictures, it doesn’t do any good for it to show up when someone has filtered for pictures of the flowers.
The usual cases for cannot be determined here are insect cocoons (is there a live bug in there or not?), and situations where the organism is not the evidence (ie Alive/Dead doesn’t apply to a leaf mine or feather or beaver dam, because it’s impossible to know if the organism they came from is alive or dead when the evidence was observed).
Exactly, in those cases the annotation isn’t about the organism at all but is about the observation and the evidence that we have. Even for the cases of cocoons, the organism itself is very much either alive or dead, it’s just that we don’t have the capacity to tell.
I don’t call us ‘citizen scientists’ If you add text please use [square brackets] that is what they are for.
There is another discusssion where people are, neither scientists, NOR citizens. So that is something else to chew on. Drifts back to defending the ivory tower academics from them out there. But iNat is egalitarian. We each get the same ONE vote.
I see fruit – whether it is green, ripe or shed. I use the taxon photos to help with my IDs. What does the fruit of this plant look like ?
This doesn’t fit my experience as an IDer on iNat. I don’t annotate very much (just “Dead” for obviously dead), but I routinely come across lizards where it can’t be determined if the individual pictured is alive or dead. Sometimes the pic is from far away, but the lizard has a posture/pose that I’ve never seen in life. Sometimes the lizard is clearly wounded (common example is someone taking one from their cat), but not obviously mortally so. Sometimes a lizard is trapped (spiderweb, glue trap) and it isn’t clear if it has died of exposure/envenomation. I usually ask the observer if they were able to assess dead/alive in person in these cases. Sometimes I get a response, sometimes not.
On a general note, I’m not entirely sure what the original question is asking about when it comes to “individual vs. observation”, but more generally, the annotations should be for whatever the target individual organism is in the observation. Each observation is intended to be for a specific focal organism (sometimes the observer indicates one, sometimes not) (see definition of observation). This is why there’s no option to annotate “Both” for “Alive or Dead” (the focal organism can’t be in both states simultaneously) - even if the observation were to show multiple individuals that are both dead and alive. Other annotation fields are also structured this way.
This type of scenario is the topic of this current thread:
This question makes it obvious that it’s important to define observations properly. On iNat an observation is defined as follows:
An observation records an encounter with an individual organism, or recent evidence of an organism, at a particular time and location.
Obviously in an observation of a flower fallen from a tree, or a feather dropped by a bird, or a shed skin left behind by a caterpillar, you are NOT observing an organism, but recent evidence of an organism only. Therefore annotations to observations of recent evidence, relate first and foremost to the recent evidence that was observed, not to the organism that wasn’t observed at all. So a flower fallen from a tree might be annotated: no leaves. A shed skin of a caterpillar should be annotated: larvae/caterpillar.
But there are exceptions: Some annotations make no sense when related to the recent evidence, and always relate to an organism. The classic case of this is the annotation of “alive or dead”. Recent evidence of an organism can rarely ever be considered alive. A shed skin or a feather, a track or a construction might partly consist of living matter, but they can’t live independently so the annotation “alive or dead” makes no sense. In cases like that, the annotation can relate to the organism, even if the organism itself wasn’t observed. A observation of a feather of a mallard can therefore be annotated as “male”, and as alive or dead: “Can not be determined”. But that’s the exceptions to the rule then.
My personal approach as a plant identifier fits exactly with the one detailed by @trscavo: Annotate based on what the evidence shows, and if there is insufficient evidence to demonstrate that the plant has “No flowers or fruits” or “No live leaves” then I leave that blank.
There are several different interlocking tensions embodied in iNat’s annotation system. From @carlsalk’s initial question we can see that Tension #1 is to define the scope to which an annotation applies. Does a negative annotation apply only to what we see (“There are no leaves in these photos”) or to the organism (“There is enough evidence to conclude that the tree has no leaves.”)
Tension #2 is that the annotation system serves multiple purposes.
Annotations associated with observations are a source of phenology data for researchers, and (via the phenology graphs on each taxon page) for anyone who might be curious. To my mind, this purpose has the highest priority. I try to avoid making annotations that would create misleading data.
Annotations also allow iNat users to filter the photos of a particular taxon to show only the photos from observations with a particular phenology annotation. So, I can see all the photos of a plant species from observations annotated as having “Fruits or Seeds”. I do find this useful, but it can be hit or miss because annotations are made at the observation level, so a particular photo may not include the fruits or flowers that prompted the annotation.
Annotations are available as a search filter in the Explore and Identify interfaces—potentially helpful, but not something I’ve really found a need for.
I mentioned another tension in point 2 above: Tension #3 is that annotations are given to observations but may be used in image searches and turn up unexpected results.
Tension #4 is that an observer may rely on information beyond the evidence in the observation photos and audio. As an extreme example, let’s say I add a couple observations with no media just to document my own experiences. I may well want to note that I saw an adult female partridge and a fruiting pear tree. Without any evidence, an identifier can’t really confirm or deny my assertions. Generally, I give observers the benefit of the doubt on this unless the evidence strongly suggests that they’re wrong.
The uncertainty in these cases is all due to the quality and type of evidence presented in the observation though, not some inherent uncertainty within the organism itself. A far away lizard in an odd pose is still either alive or dead, it’s just that we may need to get closer to it to gather enough evidence to determine which it is. Likewise, if it’s not clear from the image(s) whether a trapped lizard is alive or dead, this can usually be very easily verified in life. A wounded animal is a bit more of an edge case I’ll admit, but it could be determined by watching what happens to the animal after the encounter (which is not feasible in most situations!). “Cannot Be Determined” very much seems to be based on a human’s experience with an animal rather than with the animal itself.
I agree that seems like a reasonable thing to do, but I still don’t do it. My main hesitation is: what’s the benefit? If the annotation is meant to paint a picture of phenology, then a “no leaves” datapoint that really means “there weren’t any leaves in this particular photo, but who knows if the plant had them” isn’t useful. If the point is to allow users to search for pictures of certain plant parts, no one is going to search using a double negative. i.e. if you want photos of flowers, you search flowers; if you want leaves, you search leaves. No one will search for “no leaves” in an attempt to find a flower. So it doesn’t help with searches either. So while I don’t see it as a violation of any rules to annotate all the photos zoomed in on just a flower as “no leaves”, I can’t see any benefit to doing so, and it just muddies the phenology data.
I mean, at some level, of course everything we observe and annotate is based on our own experience. I just really feel like “I can’t tell from this picture” shouldn’t ever be conflated with actual negative data. “I looked at the plant and saw it had no leaves growing on it” and “I have no clue if the plant had leaves or not, but they’re not visible in this image” are entirely different claims. Regardless of how you want to frame “about the observation” vs. “the organism being observed”, I don’t see any value in conflating “not visible here” with “not present on the organism”.
