ArbNation Join free

Tree ID apps and keys: what the accuracy studies found

One field test put two free photo identification apps at 66.5 and 79.2 per cent top-match accuracy. A second test of six apps found none above about 88 per cent. Neither test centred on trees.

Checked against the sources at the bottom of this page on October 16, 2026. Rules, fees and pay change: the source has the last word.

Photo-recognition plant apps have been tested in two papers. A 2026 AoB Plants paper from the University of Alberta tested two free apps on 5,291 field photos. A 2023 PLOS One paper from the University of Galway and the University of Leeds scored six apps on herbaceous plants in Ireland. Both name limits for trees.

What the studies measured

The two accuracy studies, with the figures as published
ItemAoB Plants, 2026PLOS One, 2023
AuthorsWanigasinghe and colleagues, University of Alberta and Alberta Biodiversity Monitoring InstituteCampbell, Peacock and Bacon
Test set5,291 field images of 119 species from 54 plots of 0.8 square metres, in grassland and forest in Alberta38 herbaceous species photographed in their habitats in Ireland with one phone, six apps
MeasureTop-1 identification accuracy, the share of photos where the first suggestion was rightA repeatable scoring system, with first-attempt species identification also reported
ResultFlora Incognita 79.2 per cent across all photos and iNaturalist 66.5 per cent. Expert plot surveys are estimated at 90 to 95 per cent.None of the six apps passed about 88 per cent. Pl@ntNet identified 88.21 per cent of flower images and 80.36 per cent of leaves to species on the first attempt.

Limits for trees

The Alberta authors report that Flora Incognita reached 83.7 to 86.7 per cent per photo for shrubs, trees and herbs, against the 90 to 95 per cent estimated for experts. They add that the tree sample was too small for statistical analysis, and call for later plot studies with larger plots in forested sites. The paper cites a separate analysis of photo-based apps on 55 tree species but does not repeat its figures.

The Irish paper tested herbaceous plants only. Its authors conclude that apps are an aid to identification and not to be assumed correct, particularly where a plant might be toxic or harmful.

What changed the result

  • Flowers and fruit. In Alberta, photos showing reproductive structures raised accuracy, with gains of 8.9 per cent in Flora Incognita and 17.1 per cent in iNaturalist, particularly for graminoids. In Ireland, apps identified flowers better than leaves.
  • Photo quality. High-quality images raised accuracy for graminoids and herbs in iNaturalist.
  • Several photos. Combining images of the same individual improved iNaturalist by 6 per cent and Flora Incognita by 1.7 per cent. The authors add that photographing each species took two to three times as much field time as an expert's visual survey.
  • Training set. The Alberta authors link the lower iNaturalist score to a model trained on 90,290 taxa, against about 16,000 in Flora Incognita.

Studies that rank apps differently

The Irish authors found Pl@ntNet the most successful of six apps. They point out that two earlier studies ranked it lower, and that scoring systems shape the order. One school-focused study they cite scored Pl@ntNet at 55 per cent correct, while another app in that study reached 92.6 per cent. The authors take this as evidence that rankings move with the scoring method.

Free public-interest tools

The Woodland Trust offers a free Tree ID app for Android and iPhone covering the United Kingdom's native and non-native trees. It identifies by leaves, bark, twigs, buds, flowers or fruit in any season, and the Trust calls it an A-Z tree guide. Pl@ntNet describes itself as a citizen science project available as an app that identifies a plant from a picture. Its home page counters show 84,710 species, 1,475,342,717 images and 77 floras. Guides by place: tree identification by country and winter tree identification.

Sources