Overthinking plant identification

[yerbasanta]

Great question and discussion!

I’d just quickly add, I have personally started with and still primarily rely on the ‘gestalt’ or ‘feel’ for various family, genera, and species ID of local plants I’m familiar with, but am coming around to using keys and local/county species guides/lists (like CalFlora) to at least rule out or narrow down more tricky IDs to the species level (whether mine or others’).

I think the original question had a really important consideration that I haven’t seen in the discussion yet (might have missed it), which is that there is some risk in not making higher-level IDs–or stretching your comfort zone a bit.

Leaving some observations at order or family level (don’t get me started on the many ‘Plantae’ IDs I’ve gotten) I believe can put an observation into limbo, since there’s 12 million observations for ‘Plantae’ between Jan 1 1920 (earliest date available)–Jan 1 2020 in Plantae, but nearly an order of magnitude (1.6-1.7 million) observations even for the species-rich orders and families Asterales + Asteraceae.

I assume some braver souls than I trawl Plantae for ID purposes, but that seems a monumental task!

On a side note, I’ve only recently gotten into subspecies/variety-level IDs, for which I almost currently always rely on dichotomous keys and species range (location) maps, and habitat/substrate to determine just what I’m looking at–here’s one example where I waffled around on a Cirsium occidentale species before coming to a subspecies ID through use of range and habitat maps with some help from @morganstickrod.

I’ve also started looking through some of my older observations, which can be somewhat embarrassing/humbling to see how little I knew just 2-3 years ago about the observations I was making at a species level!

[teellbee]

somewhat embarrassing/humbling to see how little I knew just 2-3 years ago about the observations I was making at a species level!

Um, I hope you don’t feel that way. iNat’s training philosophy seems to be letting the user discover things for themselves. There isn’t really an indoctrination or training scheme, particularly for identification. Unless one reads the Forum messages, there’s not a lot of “best practices” guidance. Well, there is a training effort afoot currently for some ID training, but that is just coming to be now.

[jasonhernandez74]

@someplant

wildflower guides will often have only the genus.

That is probably true now that they have all gone to photographs; but under the Peterson system, with line drawings allowing eight to twelve species on a page, there are comparisons of species within a genus.

[lowclouds]

My take on this:
I’ve become less afraid of wrong IDs on my own observations, although, as @sedgequeen said, if I’m not sure, I’ll put a comment mentioning it.
However, for IDing other folks’ observations, I hold myself to a higher standard and will only go to the lowest level I’m confident is correct. Sometimes, I’ll just leave a comment if I think the original ID is questionable.
One thing I’ve found useful about going through dichotomous keys (as much as I think they’ve really reached the end of their useful life,) is that they teach you what to look for and to photograph. Sometimes, for example, if you need pappi, or seeds, or fruit bracts to make an ID, you just need to resolve to go back when those characters are there and document the evidence for a reasonable ID. Some plants, like Jepsonia, are easy to ID even though the flowers happen months after all the vegetative parts have disappeared, but others, like Atriplex or Stephanomeria, require that you get the fruits.
Maybe some day, we’ll figure out how to reify the AI knowledge and turn it into useful polyclave/multi-entry keys

[Italopithecus]

@someplant

But these keys often rely on technical features that just aren’t available in most iNat observations. For example, at step 5 the key for Asteraceae of New England (link ) tells you to look at pappi, which means if I am to use this key I’m pretty much stuck. Additionally, identifying down to genus or family using keys can be pretty hard in the first place, again because the relevant information may just be completely missing. Also all of this is assuming I’m using the key correctly and not misinterpreting plant anatomy (such as bracts, sepals, etc.). And then there’s the cases where I just don’t have any available key; this is particularly noticeable for cultivated plants

Through constant practice you have to narrow down to the genus or subgenus level only with the sight of the plant and then, if possible, right to the species level. Of course this is not always possible as some genera of group of species need the observations of various, and often minute, characters. The knowledge of the flora of a specific area and data on the environment are essential in order to exclude some taxa.
Keys are often useful only with a specimen in hand but in few cases are almost unuseful so that istinct is what can lead you to the identification.

[fffffffff]

Also a big thing to think for iders is subspeciess ids, when you add them first time it doesn’t change comunity taxon, and observation stays up at e.g. Plantae, inviting someone who at least can id the species/genus/family is a good practice for such situations (also users are afraid to id higher levels hen ssp is added which often results in a forgotten observation that in fact has a good id).

[murphyslab]

Like you, I’m not an expert in any field of biology, but I do have a good amount of experience just from exploring on my own and being curious about the plants that I saw where I grew up and when I was out and about.

Since I’m from British Columbia, I’ve opted to focus there. It helps catalogue the local biodiversity AND it also helps me to learn the local flora and fauna. It’s one thing to just narrow the location, but it also helps to narrow things to ones I’m largely familiar with and which I know are locally found. Like you, I have some local sources that I regularly consult.

But largely these are species where I know:

1. There’s a narrow set of possible species because of the geographic limits, so it’s usually going to be one of the ones I know.
2. These are species where I have enough familiarity to ID without resorting to a key (or that one essential piece of the key has been internalized)

So to make this all go smoothly, I made my own custom search URL. There’s an excellent tutorial on the iNaturalist custom URL search features.

I use the “Search multiple taxa using a List” method, for which I keep a list of BC species (most are native, but some are imports & invasives) that I feel relatively confident in identifying. That list has a list_id value (e.g. list_id=1139967 … though that isn’t mine). There’s also an associated place_id for British Columbia (place_id=7085).

So if I wanted to do IDs in Canada (place_id=6712) with a particular list (list_id=1139967), the URL would look like this:

https://www.inaturalist.org/observations/identify?place_id=6712&list_id=1139967&quality_grade=needs_id

I keep that bookmarked so I can quickly get back to identifying local & familiar flora.

The downside of that method is that it only catches something if there’s a preliminary ID of a taxon in your list. Two ways to address that shortcoming are:

1. To include higher level taxons in your list.
2. Create a list for the higher level taxons that you want to ID within & repeat with that.

At present I’m not great with mosses, save for a few. So I only include the small handful of exact moss species that I know & don’t have any higher level moss taxons.

It isn’t a perfect method, but it’s darn near close enough. And it does allow me to be more selective and contribute in a more useful fashion. Plus my home province has an unending stream of observations that need IDs. (We seem to have several professional biology folks who upload in bulk!)

[arboretum_amy]

@someplant I just wanted to say that I very much appreciate your identifying, particularly on cultivated plants. They are a largely underserved group and it is nice to have another person outvoting bad computer vision suggestions.