There’s been a lot of talk lately about where iNaturalist is headed and what the next few years might look like. The site has come a long way since it first started, and it’s amazing to see how much it’s grown and how many people now rely on it for learning, documenting biodiversity, and connecting with nature.
All of that progress naturally raises the question: what should the future of iNat look like? I know I’ve been thinking about it, and I’m sure plenty of others have too. Are there things you’d like to see improved? Features you wish existed? Changes that would make using the site smoother or more enjoyable?
This isn’t meant to be a complaint thread—more of a place to share hopes, ideas, and directions you’d love to see iNat explore. What would make the platform better for you as an observer, an identifier, or just someone who enjoys being part of the community?
You know, I was just thinking the other day that I hope they don’t change too much and ruin the experience. Right now, iNat is pretty darn good in most respects. If it ain’t broke, don’t fix it!
I would love to see a more intentional and robust integration of taxonomy into the computer vision and geomodel. Right now, iNat’s models are only trained on the lowest eligible taxonomic level. This means that if a species qualifies for the CV & geomodel, the models don’t learn anything about the distibution or identification of the genus as a whole (the same is true for higher taxonomic levels such as tribe or family, etc.). This can lead to bad outcomes in various ways:
Species A and B are plants in genus X, and these two species are very difficult to distinguish from photos. In Europe, only Species A occurs, but in North America there are both species. This leads to lots of Research Grade observations of A in Europe, and none of either species in North America. For this reason, the CV will learn that anything that looks like genus X must be species A (which is wrong) and the geomodel will learn that genus X is only expected to be in Europe (which is also wrong). You will never get reliable suggestions in North America, where the proper identification should just be “genus X,” because the genus itself is not in the model.
Species A, B, C, and D are moths in genus Y. Species A has a distinctive greenish coloration which allows it to be identified, but species B, C, and D are similar shades of brown and are difficult or impossible to identify from photos beyond the genus level. Therefore, species A is the only member of genus Y in the CV, and the automatic suggestions will only suggest genus Y for moths with green wings, even though most species of this genus are brown. Species B, C, and D get consistently misidentified as genus Z, a different brown moth, because genus Y is not in the CV model independently from species A.
There is very little identifiers can do to fix this problem; it’s baked into the way the taxon selection is designed. I know at least one identifier who avoids making identifications on some taxa to avoid making the species eligible for the CV and causing it to “forget” the genus. I hate that that’s a problem we need to worry about!
Instead of adding more gimmicks to the app, I would love it if it wouldn’t freeze on my iPhone! I often can’t even use it in the field because the functions freeze. It’s really frustrating when wanting to record a birdsong, but it’s frozen on my screen.Not a complaint, as 99% of my observations are photos which I download onto the website version from my laptop anyway.
As someone who frequently records bird song, I’d love to see standardized spectrogram generation added to audio observations like eBird and Xeno-canto. Sonograms would make the identification of audio obs exponentially easier, and this is especially invaluable in regions where verified sound observations of local fauna are limited, like in Indonesia, and for documentation of species that are better heard than seen. I’ve seen people discuss scale, especially with respect to the utility of multiple taxa; in general I’ve found the <12kHz scale used by ebird to be sufficient for the majority of taxa, while Xeno-canto’s <20kHz is a bit harder on the eyes but can include bats as well. I’m sure having any spectrogram visualization would be better than none.
Semi-related, ebird has also recently begun to automatically normalize audio uploaded to the macaulay library to -3dB, which also goes a long way to helping ID, as the overwhelming majority of audio I encounter on the site requires me to put a stethoscope to my laptop speaker at maximum volume. I don’t blame people for doing this, necessarily, when they don’t know what tools are available for normalization, or best practices for audio media upload. Automatically normalizing audio will also be a huge help to audio ID.
While I’m wary of the extent that we ‘gamify’ the website, I think an improved social aspect to the site could be refreshing. I don’t know exactly how this could manifest, but I always like to see what my friends are seeing and where they are. Maybe when I look in my following tab, I could see a small map that shows where a recent bulk of observations was taken? eg. when I see “X user added 14 observations” I can see a small map of Kelowna, or Oaxaca, or Melbourne, similar to the map we see after uploading our own observations.
Lastly I think that identifiers could benefit from improved notifications. When someone adds an ID that has a comment (separate from just a comment) I can’t see that in my notifications, I can only see it when I open that observation, or if I happen to see it in a mountain of updates on my home page. It would be useful to see this when checking notifs, or to have a separate notification for this, comments, or dissenting IDs, which are usually the biggest reasons to return to an observation.
For dark mode I use the Dark Reader browser plugin, it works pretty well - usually. Occasionally it inverts the colors of images, but I haven’t run into that recently.
Higher resolution images and smarter computer vision would be nice. I would like curators to be able to flag taxa that can’t be reliably identified by sight so the CV knows to suggest the parent taxon rather than suggesting species that are unlikely to be correct. A checkbox on the species page should do it.
Regarding higher resolution images, it would be cool if there was an algorithm that only saved as much resolution as there really is - for example a blurry photo doesn’t need to be 4k, but a really nice camera photo that has a lot of resolution could have more pixels saved.
I’m looking forward to see notifications on subscribed places of interests in android app. Somehow I’m notified on observations only in desktop web version.
I would also like improvements for such use cases as participating in different variations of Big Year projects. It is still impossible to easily find out what nearby species you haven’t observed in current year (“goals” as far as I know show you unobserved species). In order to do it right now i’m using custom made console application.
We use iNaturalist for our river protection citizen science projects and chose it primarily for the “one app for everything” approach. Audio ID would really strengthen that pitch — especially when onboarding beginners. Right now we have to explain why they need a separate app for frog calls or bird songs, which undermines the whole idea. Native audio identification would make iNaturalist a much easier sell.
iNat definitely has many good quality and correctly IDed observations of both frog calls and bird songs. Auto-ID suggestions for pictures are a feature of iNat, but not a be all and end all, and users really shouldn’t be depending on IDs from Merlin or other audio ID apps (or iNat’s photo-based CV, for that matter) anyway. Those suggestions are often wrong or just a starting point for an ID. If users aren’t sure of the species ID for an audio observation, that is fine - they can just ID as “Anurans” or “Birds” and wait for feedback from the community. This is often better than blindly accepting an algorithm-generated ID suggestion if the observer really has no idea of the ID anyways. There are lots of audio observations with initial IDs to these levels.
Fair point on data quality — but for beginners the value is the learning moment in the field, not a definitive ID. I’m speaking from experience leading field excursions: even just getting a genus suggestion is hugely motivating compared to recording audio and then… nothing. Right now iNat doesn’t do anything practical with sound recordings beyond storing them. A rough “probably a Rana sp.” in the field is exactly what turns a hesitant newcomer into someone who starts paying attention and eventually contributes better IDs.
I think the biggest thing I’d like to see would be community wiki-style ID guides built into each taxa’s About page. There’s a ton of knowledge that would be fairly easy to share that way, and a community wiki could bring accountability, transparency, and externally sourced references to the process (as opposed to chatbot summaries, which I think are an unnecessary and suboptimal solution to the problem).
A lot of the other stuff I’d like to see is about data management, like:
Allow tagging of individual photos rather than whole obs for things like type of evidence
Support for uploading or generating spectrograms and reflecting those in the data structure (have an audio/spectrogram type of evidence for example)
More clearly-defined data grade categories (i.e. distinguish between data-deficient “casual” and non-wild “casual”; decouple the captive value from the Needs ID value)
More UX design to both ease and encourage people to use annotations (for example, I find that it’s very hard to search for observations of specifically live animals; people almost never tag things “alive,” and they are also fairly inconsistent about tagging things as “dead”)
Better support for video clips. I don’t think iNaturalist needs to be hosting entire videos with audio or anything, but it already kinda-sorta supports GIFs. It would be great to either formalize that support, or have an initiative to identify and implement support for a different format of animated image that has a relatively low footprint (I’ve seen people say there are more efficient options than GIFs). Supporting direct video uploads and conversions seems like too much to ask, but if a user can clip out and convert a bit of video to a suitable format, I think that would be valuable; sometimes a video is much clearer than the stills that can be extracted from it, and sometimes a bit of movement is enough to clearly identify something. A cap to file size would be a fair thing to implement if necessary.
Also, I’d love it if the taxonomic trees were restructured to reflect modern cladistics.
Everyday about a third of the observations of my taxon of interest are uploaded incorrectly. It’s so exhausting and I’m very close to just giving up even trying to correct them.
More powerful ID models, particularly multi-image reasoning (treating multiple photos of the same observation as a set rather than individually), and better handling of hybrids and subspecies.
Easier filtering of results without having to hand-craft URLs with &tags parameters.
At-scale detection of duplicates, likely misIDs, and observations containing multiple species. And some native equivalent of jeanphilippeb’s ‘Unknown’ projects — a way to flag observations as likely belonging to a specific family or genus so they get routed to the right identifiers rather than sitting in Unknown indefinitely.
can you elaborate? Currently, you (or any user) could add an ID of family or genus and the observation would then be easy for the right identifier to find.