A January study used machine learning to map 31 plant traits, including leaf area, wood density and seed mass. It combined species observations, expert vegetation plots and measured traits with satellite and environmental information. The maps estimate community characteristics; a photograph does not directly measure a plant’s entire biology.[1]
Three records, three different jobs
An observation establishes that somebody encountered an organism at a place and time. iNaturalist’s requirements for a verifiable record include a date, location and photographic or audio evidence, together with an assessment that the organism was wild. Community identification provides another check. Such a record is evidence of presence; it is not by itself a census of the surrounding vegetation.[3]
Trait databases contribute different information. TRY retains original measurements while standardising names and units, recording measurement context and flagging possible errors or duplicates. That work makes measurements from separate research projects comparable enough to combine. Species identity supplies a link between the observation and the measurements.[2]
In the study, gradient-boosted trees learned from these combined records. Validation used spatially separated expert plots. At approximately one-square-kilometre resolution, correlations reached 0.63; 15 of the 31 traits exceeded 0.50. Those are correlations, not percentages of correct predictions.[1]
An average depends on what receives the weight
Consider an invented example with two species: one a metre tall, the other ten metres tall. Equal numbers of records for each would give a record-weighted mean height of 5.5 metres. But if the short species occupied 90% of the vegetation and the tall species 10%, a cover-weighted mean would be 1.9 metres: 0.9 × 1 plus 0.1 × 10.
Both calculations are correct. They answer different questions because the weights represent different things. The first describes the balance of the records; the second describes the stated balance of vegetation cover. Adding observations alone does not make those definitions interchangeable.
The paper uses observation frequency as a proxy for abundance where plot measurements are unavailable. Assigning species-average traits also misses variation within a species. The maps therefore extend coverage while retaining those measurement limits.[1]
Sources & further reading
- Crowdsourced biodiversity monitoring fills gaps in global plant trait mapping Nature Communications · 30 Jan 2026 · Accessed 20 Oct 2026
- TRY database: scope and data curation TRY Plant Trait Database · Accessed 20 Oct 2026
- What is the Data Quality Assessment and how do observations qualify to become Research Grade? iNaturalist · Accessed 20 Oct 2026