Platform(s), such as mobile, website, API, other: All
URLs (aka web addresses) of any pages, if relevant: Observation submission pages
Description of need:
With the rise in popularity of leafminers, powdery mildews, and other plant-pathogens on iNat, the CV is increasingly learning both plants and the various pathogens that effect the plant. The CV has no way of knowing whether users intend to identify the plant or the thing on the plant, and as a result, thousands of observations are uploaded with IDs for mildews, leafmines, rusts, etc. that are meant to be observations of the plant itself. This is particularly a problem when the pathogen/mine selected is not even present on the plant, which is a very common occurrence, as anyone who identifies leafminers or pathogens can attest.
Feature request details:
I’ve been thinking of ways to better-inform observers as to whether they’re selecting an ID for a plant or a pathogen on the plant, and the best I’ve come up with is using the iconic taxa color coding system to color code the suggestions themselves. I had Gemini make a few mock-ups of what this might look like:
In each case, the CV isn’t sure whether the intent is to observe the host organism or the parasite, but it displays the options in a way that even the newest of users with no knowledge of the species suggested can at least tell whether they’re clicking on a plant, a fungus, or an insect. I think this would go a long way to preventing the perpetual issue of plants misidentified as pathogens that may or may not even be present. It would therefore reduce kingdom-level disagreements that in essence throw observations back into the “Unknowns” pile.
I would like iNat to display the broader taxon. Still confused by why iNat sometimes spells out, this is a member of the Family … and other times doesn’t bother.
This tiny thumbnail is a fish, so it won’t be the insect you intend. (Not Kingdom disagreement, but still flat out wrong)
For some reason iNat seems to top the list with the disease / pest, trailed by the intended plant ‘host’.
A recent obs was of rose hips, which the observer clearly thought was a gall, a disease of some sort. No idea that a rose can make fruit, called rose hips.
Another idea I considered was making a push for the leafminer and fungus pathogen taxon photos to all be close-ups of the pathogen. For example, here are two photos I took of the same leaf miner:
When viewing it as a thumbnail, that first photo screams “leaf”, while the second photo is clearly some funky bug trail. If it were up to me, I’d do away with all the “whole leaf” thumbnail photos for plant parasites and replace them with closeups like the second photo above. But that’s something done on a taxon-by-taxon basis, so I figured some form of labeling would be a good top-down approach with the same goal.
This would also be good for the drop down when typing in the name because there are some animals and plants that have the exact same genus name or very similar common names.
This is a good example showing how the cropping behavior for an entire community can affect how the CV learns. The CV learns by the training data fed to it. If we, or even a handful of people (but enough) feed it images not focused well on the target. This will always be the result. Even if 9/10 people cropped to target. 100 uncropped images that looks more like the whole leaf can still teach the CV “that leaf identifies this moth” rather than “this mine identifies this moth”.
I agree in principle, but ultimately there are many taxa where people have not taken appropriate closeups of the organism in question. I do a substantial amount of curation of taxon photos for plant parasites, and I often run up against a lack of high-quality photo availability. Leafminers are slightly better off than many leaf spots and mildews, but it would be difficult to get everything in perfect order for this proposed solution.
That of course is setting aside the other major issues – the existing unfixed problem of not automatically removing a photo from any observation whose Community Taxon no longer matches the taxon entry for which that photo is a default; the frequent setting of non-Research-Grade (or simply unreviewed) photos as defaults by people who do not have the expertise to confirm the ID in the first place; and so on.
To the actual main topic of the thread: I would support icons (and color-coding those icons) for disambiguation, and in principle it should not be difficult for staff to implement. However, seemingly the vast majority of kingdom-level misidentified observations are from inexperienced users who would not necessarily look closely enough to appropriately interpret the extra information provided by icons and color-coding. (I think those calling for icons and not merely colour-coding have it right, given that I usually can’t remember off the top of my head what the arbitrary iNaturalist “Iconic Taxon” colours even go with – the one given to insects is very difficult to distinguish from the one given to fungi, for example.) Most often, schoolchildren or people who just want “the answer” right now end up just picking the top suggestion regardless of whatever is conveyed by the taxon entry photos or even a common name for the taxon. I’m not saying we should be discouraged from adding more disambiguating points to the suggestions, just that I doubt it would substantially cut down on the existing issues for the most part.
The problem is bad enough for some species that I think any fix is better than no fix. I regularly ID new observations of the leafminer Liriomyza violivora, and a truly ridiculous number of them—over 80%—consist of leaves without any mining on them. Many of these taxa have very few identifiers, so these data quality issues can pile up for years unless someone consistently puts in the effort to correct the record. Even if the majority of inexperienced users would ignore icons, it would still go a long way to building a system where identifier effort is not wasted on kingdom disagreements instead of producing RG-level data.
I’ve found a similar error rate with a few of the moths I ID as well. It probably deserves its own thread, but another issue that probably contributes to this is that a lot of observers will include a photo of the host plant as one of the photos in the miner observation, which means even if their identification is correct, they’re feeding the CV training one photo per observation of the plant rather than the mine. So it’s no wonder it gets confused. I’d be very unpopular if I started DQAing these for violating “single subject”, but that’s honestly what should be done if the mine isn’t even visible in one of the photos.
I agree that anything is better than nothing though, and some sort of “you’re about to click a bug when you mean to click a plant” warning is a good start.
I am blue-green colorblind, and agree that icons would be preferable to colors for this. This is especially true for new users, who might not know what color each taxon is associated with, but can tell a leaf icon from a beetle icon.
I am also colorblind and essentially just ignore (unconsciously at this point) most color info since I know there’s a good chance I am going to get it wrong. The recommended number of distinct colors for charts and graphs is often quotes somewhere at 5-8, and even these can be difficult to match meanings for without a legend. So I think, with the number of colors and lack of legend, the value of the colors alone would be low
I think the icons would help more since they are also a visual match to the general appearance of the organism - essentially, the same kind of data as in the photo. It’s generally easier for the brain to make meaning this way (e.g., bug photo → bug icon vs. bug photo → green color). I have no personal preference on whether the icons are colored or not, but I would note that the colors can mean there are different levels of contrast with text icons, which could matter for some users one some screens under some conditions (like strong daylight, etc.)
Also, I note that the spec for the feature request is for “observation submission pages” but I think for consistency’s sake it should be for “CV identification submission pages”. In other words, anywhere that a user would enter a CV-based suggestion., this would be implemented. This would be more consistent for one thing, but also might cut down on the dreaded “double nonsense CV ID → RG” pipeline that afflicts some of these taxa/observations. If a second IDer (who is otherwise adding an incorrect CV ID that they can’t really support on their own expertise) sees the icon, it might trigger them to do a double take and reconsider.
I support colour codes and symbols both as options, maybe that users can choose between, since not everyone can benefit from colour highlights due to vision.
But I understand iNat staff are trying to move away from any colour designations from major taxon levels, which I hope they’ll reconsider eventually.
Shrug, thats what the guidelines are for and you would be using the DQA correctly. I never include the host image with the subject. I always keep them as seperate observations.
Also there is a new CV engineer staff member, however they don’t seem to be on the forums, or haven’t been given the staff tag. So unsure if the community will continue to have essentially silence regarding CV issues.
I would love icons/color or some sort of UI indication of what the suggestions are. I can’t express how nuts it drives me when the CV will say “we’re pretty sure it’s in this tribe” then the top suggestion due to the geomodel isnt even in that tribe.
Like, imagine someone say, “I’m pretty confident you have a Bumblebee, my top species suggestion because no Bumblebees I know live in your area is the Brown Sweat Bee.”
Pardon, you think what? Really the worst part is you HAVE to be familiar with the taxonomy to even realize this. Say the CV is pretty sure it’s in Simuliidae, and then the top suggestion is genus Chaoborus. How many people do you think will realize that genus is not even in that family?
I believe the suggestion UI needs an overhaul, we should be told if the suggestions are an insect, plant, and if its even in the parent taxon suggestion. Which is the “we’re pretty sure it’s in x”.
Sorry for the small rant, it really does drive me nuts.
Why does iNat provide NO INFORMATION to point out that the chosen sp is a plant, while the obs is for ‘marine life’ ‘in the sea’ ?? Australia vs Texas too.
Why offer a land-based plant for marine life anyway ??
PS we cannot blame iNatters, especially newbies, for not using the info we DO NOT give them !!
PPS here – where no disambiguation is required, iNat provides the info. TWICE.
What if we added a setting that makes you choose the Iconic taxa it’s in before you select the species (with an unknown option being available) that’s on by default for new users, but is also togglable off for people who know how to use iNat
And the homonym for centipedes, does not bother to say I’m a daisy. Now trapped at Life by Kingdom Disagreement and iNat ‘working as intended, but not as expected’
