Recommendations for plant identification apps were recently making the rounds on social media. Of course it would be nice to accurately identify an unknown weed in your field by taking a picture instead of sending it to an “expert” or a diagnostic lab. However, many of the top-rated plant identification apps are not particularly accurate at weed identification.
Unfortunately, you’re not likely to know the app is giving you a wrong answer unless you’re the type of person who can already accurately identify weeds.
I recognize that my criticism of plant identification apps might be met with some healthy skepticism. I’ve spent most of my adult life trying to refine my weed identification skills. I probably sound like the blacksmith who tried to convince the locals that the curve of a well-made horseshoe is far superior to any round piece of vulcanized rubber. However, dependable identification apps will require much larger databases of accurately identified photos. They may get there sooner rather than later, but in the meantime, it’s wise to double check an app’s suggested ID instead of blindly accepting it as definitive.
I wanted to compare three plant identification apps and put them to the test, so free versions of “PictureThis,” “PlantNet” and “PlantSnap” were downloaded. All three have options for monthly subscriptions that provide an ad-free interface and other features.
Pictures of 11 different weeds were taken from my own farm, from research trials and from a plant that was sent to me for identification (it was thought to be waterhemp, but an agronomist wanted confirmation). The results for each app are shown in Table 1 (below). It is most important to focus on the success of being able to identify the correct species, rather than correctly identifying the genus of the plant. Correctly identifying the genus would be like someone bringing you a plant from their field, asking you if it was waterhemp and you saying, “I don’t know but it looks like it’s from the pigweed family”… thanks Dr. Linnaeus for that wonderful insight. Can I call you Carl?
What about waterhemp? The suspected waterhemp sample was identified as waterhemp by both Dr. François Tardif and myself. The sample was also sent to a lab (Harvest Genomics) where they did a molecular test for identification and herbicide resistance.
The species was identified as waterhemp and resistant to Group 9 herbicides (e.g. glyphosate), Group 2 herbicides (e.g. Classic) and Group 14 herbicides (e.g. Reflex). It was only the “PictureThis” app that identified this plant as being from the Amaranthus genus, but it guessed that the species was Palmer amaranth, arguably a worse species than waterhemp, but incorrect nonetheless.
Summary: For this particular test, the PictureThis and PlantNet apps accurately identified the correct species 36 per cent of the time. Not overly impressive. However, both apps do have nice and intuitive user interfaces. Given that this technology has progressed rapidly in the last 10 years, it should only improve with time. In the meantime, this old chunk of coal is going to happily use traditional identification guides and keys to identify unknown weedy plants.