For observations known to be in the genus Ligustrum in Virginia, North Carolina, South Carolina, Georgia, Florida, Tennessee, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, and Texas:
- Let iNaturalist use its artificial intelligence (iNat’s AI) to suggest identifications.
- Choose the first species iNat’s AI suggests.
Why are the suggestions of iNat’s AI for privets so reliable?
Simple: Frustrated with the number of misidentified observations in this genus, a few naturalists took it upon ourselves to get those misidentifications corrected. As we did, we noticed a few interesting results:
- In new observations of privets, the observer was more frequently making a correct identification.
- When people reviewed older “Needs ID” observations, they were also getting better at identifying privets correctly.
- When we were doing our own reviews, if we were patient enough to let iNaturalist make a suggestion, the suggestion was more likely to be correct. (As @sambiology recently told me, agreeing with a suggestion is easier and faster than entering a name—so experts can review observations a lot faster when the first suggestion is correct.)
Who did the work?
So far as I know, @alisonnorthup and I contributed the most work, but there are probably a good many others who played a part. If you also gave time between October 2017 and October 2021 to clean up misidentifications in the genus Ligustrum, let me know and I will edit this to give you credit, too.
Why just these 12 states?
Without getting too mired in the details:
- To some extent, this approach does work everywhere.
- I live in Texas, and Alison was in graduate school here. Trying to make sense of observations in our immediate area, we discovered that only four species of privet are found here. Indeed, that was true throughout the state.
- Most of our review elsewhere focused on these species.
- Privets in the other 11 states turned out to be exclusively or at least predominantly these species. This approach probably works well in other areas that fit that description—for example, California, New Zealand, and Australia.
- North of these states, other privets are also common. Having seen only photos of those species, I feel comfortable identifying only the clearest of cases.
What are these four species of privet?
-
L. lucidum, commonly called glossy privet or tree privet;
-
L. japonicum, sold in nurseries as waxleaf ligustrum and commonly called Japanese privet;
-
L. sinense, commonly called Chinese privet; and
- and L. quihoui, commonly called quihoui privet. This species is found a large swath of Texas and Oklahoma, and smaller populations are found in Arkansas, North Carolina, and Virginia. It occurs in a few isolated locations elsewhere in these 12 states.
Oddly enough, because these four species had often been misidentified as three other privets—L. vulgare (common privet), L. obtusifolium (border privet), and L. ovalifolium (garden privet or California privet)—we seem to have improved the ability of iNat’s AI to identify them, too. I am not sure how great of an improvement we made on those, but it seems to me that they are suggested erroneously less frequently than before.
Wasn’t that a lot of work?
Absolutely yes! As I write this, according to iNaturalist:
- I have reviewed 29,672 observations identified as Ligustrum spp. (That’s 42 percent of the 70,266 observations identified to that genus at this date.)
- Alison has credit for identifying 2,783 observations of Ligustrum spp. If her ratio of observations reviewed to observations identified is the same as mine, she would have reviewed about 4,600 observations to run up that many IDs.
- Both of those numbers are underestimates of the reviews done. Total identifications does not count observations that are Casual grade, observations that were already correctly identified by so many people that we chose not to add ours, nor observations that for whatever reason have not reached Research Grade. Neither number shows how many observations we reviewed that turned out not to be in this genus.
To be clear, I am not complaining about the way identifications and reviews are counted. My point is that looking at these numbers for the genus Ligustrum does not give an accurate estimate of the work done. If you are thinking about doing this in another genus, keep that in mind. If having an accurate estimate of this work is important, perhaps someone can develop queries that give an accurate estimate of each outcome I mentioned above.
How long did it take?
In personal time, it took untold hours of working through and reviewing IDs. I gave up TV and Twitter time. I don’t know how Alison managed it.
In calendar time, the work stretched out over more than two years. These were the big steps:
- October 2017, Alison started her reviews. Working steadily over a period of months, she compared specimens and keys to photos and to her own observations. During this period, I didn't do much more than keep up with current observations and chime in on an identification when Alison mentioned me in it.
- December 2019—26 months later—I started my own systematic review, one species of privet at a time, first in Texas and then in every state along the coast to Virginia, plus Oklahoma, Arkansas, and Tennessee. I had a lot of time on my hands, so I was able to complete this review in about two weeks. Towards the end of this effort, I started noticing that when an observation was a privet, iNat’sAI was much more likely to first a privet first. Furthermore, when that privet was one of these four species, that suggestion was almost always the correct privet.
- March 2020, it dawned on me that whenever the first suggestion for a privet wasn't a privet, iNat’s AI was usually suggesting one of the same dozen or so other species. So how often had other iNaturalists chosen one of those species when the observation had actually been a privet? The reviews we had done so far would have missed those misidentifications. I took advantage of the Covid shutdowns to review observations identified as those species, too. As I weeded out the misidentified privets from each species, that species became much less likely to be the first suggestion for observations that were actually privets. Gradually iNat’s AI reached the point of almost never suggesting a species that wasn't a privet first when the observation was a privet. We had arrived!
What lessons does this hold for other iNaturalists?
The main lesson is that if iNat’s AI is making the wrong suggestions in a taxon you are interested in, you have the power to set it straight. Just make a dedicated effort to cleaning up misidentifications in that taxon. These approaches might increase your chances for success:
- Find an achievable goal. Perhaps I am wrong, but with all the hybrids that form within Section Lobatae (red oaks), I doubt that anyone will be able to clean it up neatly.
- Be sure you are right. The only thing worse than adding to the confusion of a scrambled taxon would be to scramble the order of a taxon you misunderstand.
- Team up with others. Define the problem together and develop a strategy for fixing it. Again, be sure you are right about the problem you think exists.
- Focus your efforts at first: one species, one locale. Then work through all the related species in that locale or all locales for that one species.
- Strike a heavy blow. You’ve got to be able to correct misidentifications faster than they are being made. Organizing a virtual sprint to clean up one taxon in a weekend is one example of a heavy blow.
- Be persistent and consistent.
- Misidentifications go both ways. If you find that a species is frequently misidentified as the taxon you are cleaning up, go check out observations identified as that species for instances that are actually your taxon.
- Keep other iNaturalists informed. When you correct a misidentification, explain why. The explanation doesn't necessarily have to be long. Sometimes my only comment was along the lines of “sessile fruits.” Those comments can attract others who will help with the work.
- Be open minded and humble. Some of what you are sure of could well turn out to be wrong. Even newbies can contribute valuable insights. You will almost certainly learn more than you had thought there was to know.
- Most of all, keep it fun. Don’t let the drudgery of correcting errors interfere with the sense of adventure of exploring the natural world.
由使用者
baldeagle
在
2021年12月27日 16:25
所貼文
評論
新增評論