Tech Tip Tuesday: Navigating Suggested Identifications

I forgot that July is one of my favorite months. I realized it as soon as I saw the doe with her delicate, spindle-legged fawn nibbling the long grass along the forested edge of my backyard. I remember a similar pair appearing around this same time last year, their footsteps careful and eyes wide as they paced through the short grass. I had seen evidence for weeks—cropped jewelweed near the stairs, angled prints in the soft soil along the garden’s perimeter—however, I had not laid eyes on them until recently. They must have heard the door open because by the time I was fully outside, they were nervously stepping into the shaded tangle of vines and limbs.

After a long dry spell, the forest has come back to life. Plants that were lingering, parched, now are vibrant. In the tangle of thorns and leaves underneath the back porch, Black Raspberries are creeping into view, fruits turning from white to rose pink and soon to inky purple. To me, July feels like a full expression of life in the Northeast. If you’re wondering what you can expect to see in July, check out this month’s Field Guide and Bees of July for some ideas.

This Week on Tech Tip Tuesday

Sometimes the ability of iNaturalist software to guess a species’ identification for us seems like magic. How did it know that the animal in the photo was more likely to be a Gray Squirrel than an American Mink? If we journey backwards a couple months through the Tech Tip Tuesday archives, you might remember that I wrote a piece about how iNaturalist uses a type of artificial intelligence (AI) called “computer vision” that is trained to recognize features in a photograph that help distinguish one species from another. Thanks to this technology, you can often get fairly accurate suggested identifications that help you better understand the plant, animal, or fungi that you’re seeing in the field.

The key phrase in that last sentence is “suggested identifications”. While it can be tempting to select the top species, genus, or family recommended, it’s important to think through a recommendation before choosing it. Although the AI often gets identifications correct, especially for frequently submitted species, it isn’t right 100% of the time. In some cases, it may provide incorrect suggestions, either because the photograph wasn’t clear, the species isn’t very common, or can’t be identified from a photo alone. But, don’t despair—think of this as a learning opportunity! By exploring the suggested species, you will learn features that will help you make your own identifications and you will become more familiar with other organisms that you might encounter in the future.

So, what should you look at when deciding whether a certain species, genus, or family might be a good fit?

  1. Think about it—one of the first things you can do to rule out an incorrect suggestion is to take a moment and consider the species on the list. Are any of them species you recognize and, if so, are they wildly out of place? Or, if you don’t recognize them is there anything in their name that might give you pause? Sometimes, the AI will suggest species from a completely different continent. If you’re in Williston and it’s suggesting a Springbok, I would say there’s a 99.5% chance that it’s wrong (about that last 0.5%—hey, you never know).
  2. Compare the photos—sometimes you can quickly tell that a suggestion is wrong by looking at the pictures for the suggested species, genus, or family. This can easily be done when uploading either by clicking on the two arrows next to the suggestion in the app or by clicking on “view” next to the suggestion on the computer. You may not always be able to say that a suggestion is incorrect based on the photos—sometimes species have multiple color morphs, like the Asian Lady Beetle. However, if you aren’t seeing any photos that look similar to your observation, it’s definitely a reason to pause before selecting it.
  3. Look for “visually similar/seen nearby”—when looking over the list of suggested species, it’s important to pay attention to which species are considered “visually similar and seen nearby” (this appears below the species’ name). This can help when deciding between two species who look fairly similar—if one is visually similar and one is both visually similar and seen nearby, chances are good that it’s the second one.
  4. Check the species’ distribution—besides relying on “seen nearby”, you can also find the suggested species on a map and see how your observation compares to the species’ range. If the suggested species doesn’t seem present in your area then chances are good that it isn’t correct. However, in some cases, a species may appear outside of its range, either as an introduced species or a misplaced migrant. If you believe that this is the case, choose the genus or family and add a comment asking for others’ input before selecting the species.
  5. Learn more—if you haven’t been able to rule in or out a suggestion by using the previous steps, take some time to learn more about a suggested species, genus, or family. Develop a better understanding of what conditions it needs to survive in an area, where it’s usually found, and if there are any key features to look for.

Sometimes, you can improve the suggestions by either improving your photo (if your observation is still nearby) or selecting a different photo. Many people don’t realize that, although you can upload multiple photos at once, iNaturalist only looks at the first photo to make its suggestions. So, if your first photo of a tree is the bark and it isn’t yielding any accurate suggestions, try choosing a leaf photo for the first picture. However, it’s still important to include the bark because it will help other users to confirm or contradict the identification.

There are also some species that the AI can’t recognize because there aren’t enough observations available for training. If this happens, other users will likely be able to help provide a more specific identification after you have uploaded the observation. In some instances, a definitive, species-level identification may not be possible. Some species’ identifying features are not visible in an ordinary picture and may require a microscope or a behavioral observation. If you’re unable to select a suggestion, it’s perfectly fine (and encouraged) to pick a broader category, such as “bird” or “fern”, or “animal” or “plant”. Other users will then find these observations and help narrow them down.

TTT Task of the Week

This week, take some time to evaluate suggested identifications before selecting them, if that’s not something you already do. Take some time to look up additional information about the suggested species, paying close attention to their ranges’ and life history traits that may help indicate a correct or incorrect suggestion.

That’s all for this week. Thanks for helping us map Vermont’s biodiversity, stay safe, and happy observing!

P.S. If you want to learn more about how computer vision works, check out these two articles:

https://www.inaturalist.org/blog/31806-a-new-vision-model

https://www.theatlantic.com/science/archive/2017/07/an-app-for-identifying-animals-and-plants/535014/

Posted on 07 de julho de 2020, 06:08 PM by emilyanderson2 emilyanderson2

Comentários

Love your descriptive sharing of the deer observations in your yard. Can I use my devoured hostas as documentation for deer?

To your point in #5: I have learned how to upload multiple photos. However, how do I go about moving a secondary photo to the AI sensitive 1st place? Ruth Stewart

Publicado por ruthstewart mais de 3 anos antes

Adicionar um Comentário

Iniciar Sessão ou Registar-se to add comments