AI will map bee boxes and flowers to study pollination’s economic value

A beehive box viewed from a satellite can look like a water tank, a haystack or an unremarkable pale rectangle. Yet identifying where managed hives are placed, and how long they stay there, could help answer a surprisingly difficult economic question: what is a field's crop yield worth without the pollination activity of insects that may forage across farm boundaries? Researchers at the University of North Texas and three partner universities are developing AI tools to turn scattered ecological observations into a wider, time-sensitive map.[1]

The project has received a $3.6 million grant from the US National Science Foundation, of which $1.4 million is allocated to North Texas. The team includes ecologists, computer scientists, geographers, entomologists and economists at North Texas, Michigan State University, the University of California, Santa Barbara, and the University of Maryland. The grant is financing research rather than reflecting an achieved financial benefit to agriculture. The university announced the programme on 29 October.[1]

Ecologist Elinor Lichtenberg describes the existing research bottleneck as one of geographic and temporal coverage. Researchers may travel to a farm, count flowers and insects, and assess one crop at a particular moment. Pollination, however, changes across a season. Blooming fields, neighbouring habitat and managed hive placement change from week to week. Observations made at a small number of sites can leave economists with an incomplete picture of the service wild and managed bees provide across a larger agricultural landscape.[1]

Small boxes, large areas

Computer scientist Xiaohui Yuan and colleagues are using sequences of high-resolution satellite images to recognise bee boxes and other small objects. This is a demanding computer-vision task because objects occupy few image pixels, may be partially hidden and can resemble structures unrelated to beekeeping. The team has had to work with ecological experts to distinguish the boxes from water tanks and haystacks and to refine the labelled examples used for training. That collaboration matters: a model that counts the wrong objects could make the resulting economic estimates look more precise while actually introducing systematic errors.[1]

The researchers plan two complementary tools. One will estimate floral availability at a location and time, a possible indicator of the resources available to pollinating insects. The other will locate very small objects in remotely sensed imagery, beginning with managed honeybee boxes. Combining these observations with crop development and yield data could allow economists to investigate whether proximity to hives, their duration at a site and the surrounding habitat correlate with changes in agricultural outcomes.[1]

A detected box does not reveal the exact number of living bees inside it. Its presence may provide a rough indication of managed-hive activity, but colony strength, bee health, flight behaviour and the flowers actually visited are not captured directly by satellite pixels. Likewise a landscape with abundant flowers need not supply the right pollen or nectar to every crop at the relevant time. The project includes field verification by Lichtenberg's laboratory precisely because remote observations require biological checks.[1]

The small-object detector began development under a university seed grant in 2024, and the team says initial findings appeared in Expert Systems With Applications in 2026. The newly announced grant expands the work into a broader research programme. It would therefore be wrong to treat the entire $3.6 million project as a laboratory result already achieved, or to claim a measured increase in crop yields from the new tools. Their promised contribution is a better way to collect data on the conditions under which pollination helps farms.[1]

Pollination is not a free input

Farmers can pay for managed bee services by renting hives, or support wild pollinators by preserving land and habitat. Both options have costs. Reserving field edges or nearby parcels for flowering vegetation may reduce the area planted with crops; renting managed hives imposes a direct fee and depends on availability. Previous research by some members of the team examined when each approach might be advantageous. The new project will ask how coordinated decisions across nearby farms change that comparison.[1]

The collective question is important because bees do not respect property boundaries. One farmer may maintain habitat that improves pollination on a neighbour's field; another may rent hives that also forage elsewhere. Individual cost-benefit calculations can miss those spillovers. Better spatial and seasonal data could help estimate who bears the costs and who gains the crop benefits, an issue relevant to farm contracts, conservation incentives and agricultural policy.[1]

Yuan has suggested that technology refined for tiny beehive boxes might eventually help detect other small objects, including road debris relevant to autonomous driving. That is an example of possible transfer, not a documented self-driving deployment or evidence of improved road safety. The immediate research challenge is much more concrete: obtaining defensible ecological measurements without sending field teams to every location every week.[1]

The project is intended to make ecological observations more geographically and seasonally complete. Its economic conclusions will still depend on field verification and on separating pollination effects from weather, crop choice and other farming decisions. Detecting a hive box is an input to that work, rather than a direct measurement of the value produced by its bees.[1]

Report an error or suggest a correction