What is it about?
Can plant identification apps miss unusual or newly introduced species because those species are not expected to be there? This study explores how location and plant characteristics influence artificial intelligence (AI)-based plant identification. Apps like iNaturalist and PlantNet use computer vision to analyze plant photos and generate species suggestions. Although these tools are designed to help users quickly identify plants, their suggestions are influenced not only by visual similarity but also by geographic information used to prioritize species that are expected to occur near the observation site. While this approach generally improves performance, it may also affect species suggestions. This study examined how several factors influence AI-generated plant identifications. We tested whether increasing location precision changes the likelihood of correctly identifying species that are rare or not established in the region compared with established plants. Identification accuracy was also compared between native and long-established introduced species, as well as between the two applications. In addition, we evaluated whether performance varies among plant families, including the grass family (Poaceae), and whether photographs showing flowers or inflorescences are easier to identify than images showing only leaves. Overall, this study evaluates how location, plant characteristics, and application design influence AI-based plant identification performance.
Featured Image
Photo by Yvette Aubusson on Unsplash
Why is it important?
AI-based plant identification apps can generate large amounts of biodiversity data, and their reliability depends on understanding how they generate species suggestions. This study highlights how location information influences AI-generated suggestions. Species that are outside their expected distribution range may be more difficult to detect, which is particularly relevant for monitoring novel or potentially invasive species where early detection is critical. Our findings also show that plant characteristics and plant family can influence AI performance, revealing potential biases in identification outcomes. This study supports the responsible use of AI tools in ecological monitoring as citizen science platforms and automated identification systems become more widely used in environmental management and conservation decision-making.
Perspectives
Artificial intelligence is here to stay and will continue to play an increasing role in many sectors as funding and programs work to integrate AI into jobs and public services. From my perspective as someone involved in identifying invasive plants that are not yet established or are present only in small populations in Canada, this study addresses an important and timely question. Individuals involved in monitoring programs should verify species occurrences reported by citizens on these applications, but they should also remain aware of how identification accuracy can be affected by different factors. Traditional monitoring approaches, including field observation, sampling, and expert verification, should continue to be used alongside emerging digital tools. Programs conducting surveys for invasive plants outside their known distribution ranges should avoid relying on AI applications as the sole method of species confirmation. Professional expertise remains essential, as AI-based systems can still produce identification errors, particularly for certain groups such as grasses, where our study showed lower accuracy.
Andréanne Charron
Canadian Food Inspection Agency
Read the Original
This page is a summary of: Investigating key drivers influencing AI-based detection and identification of plants, PLOS One, March 2026, PLOS,
DOI: 10.1371/journal.pone.0342712.
You can read the full text:
Contributors
The following have contributed to this page
