Managing Agriculture by Measuring Pollinating Insects

 

From Lab Bench to Orchard

A new millimeter-wave radar can identify a pollinator from the micro-Doppler signature of its wings. The interesting question for working agriculture is no longer whether the technique works, but how it is fielded—on poles, on UAVs, or stitched into the cellular fabric already going up around the farm.

A Special Report  |  May 2026

Bottom Line Up Front 

Pollinating insects support roughly 35 percent of global food crop production by volume.¹ Commercial growers of pollinator-dependent crops—almonds, blueberries, apples, cherries, melons, sunflower, canola, hybrid seed—already pay for pollination services as a major line item, but they have historically had no quantitative way to verify which insects are working their fields. A new millimeter-wave (mmWave) radar technique reported in PNAS Nexus in April 2026 by Antony et al. closes that gap by classifying pollinator species in flight from their wingbeat micro-Doppler signature.² The operational use case is straightforward: continuous, species-resolved pollinator activity maps that let a grower verify pollination contracts, time pesticide and irrigation events around peak foraging, validate hedgerow and cover-crop investments, and trigger contingency hive deployments when activity falls below the threshold for adequate fruit set. The hardware is feasible today on three platform classes—fixed pole-mounted nodes, tethered or free-flying small UAVs, and 5G/6G base stations operating dual-use as biodiversity sensors. UAV deployment is technically achievable with current 60–94 GHz automotive radar chipsets (the Size, Weight, and Power budget is approximately 50–150 g and under 10 W); the binding constraints are atmospheric attenuation, ego-motion clutter on multirotor platforms, and the very small radar cross-section of a flying insect, which limits practical detection ranges to roughly 5–15 meters. A hybrid architecture—dense fixed nodes in high-value zones, UAV scouting for broader survey coverage—offers the best near-term match to current grower economics.

---

What the Radar Actually Sees

The Trinity College Dublin / Technical University of Denmark system reported by Linta Antony, Adam Narbudowicz, Ian Donohue, Jane Stout and colleagues exploits a phenomenon any radar engineer will recognize from helicopter-versus-fixed-wing classification: micro-Doppler modulation. A continuously transmitted millimeter-wave signal scatters off an insect's body and wings; the wings, beating at species-characteristic frequencies on the order of 100–400 Hz for bees and faster for flies, impose a comb of sidebands on the return. A hierarchical machine-learning model extracts more than seventy harmonic, spectral, and temporal features from those sidebands and assigns the target to a species class. In controlled trials the system separated honeybees (Apis mellifera), buff-tailed bumblebees (Bombus terrestris), and common wasps (Vespula vulgaris) with high confidence, including pairs that are notoriously difficult to discriminate visually.²

The signal physics are not new. W-band coherent radar measurements of insect wingbeat frequencies were demonstrated as early as 2017 by Wang et al. at Beijing Institute of Technology, and harmonic radar tagging of individual bees by Rothamsted Research goes back to the 1990s.³ What is new in the 2026 work is the combination of (a) chip-scale 60–94 GHz transceivers that did not exist commercially a decade ago, (b) a sufficiently large training dataset, and (c) classifier architectures that can run on embedded silicon. Together these turn what used to be a laboratory measurement into a fieldable instrument.

The Operational Concept: How a Grower Uses the Data

Pollination is unusual among major crop inputs in that almost no grower currently measures it directly. Hive counts, placement maps, and visual scouting are the standard, and BeeHero CEO Omer Davidi summarized the resulting visibility gap bluntly to AgTech Navigator: a grower may know how many hives were placed in the orchard, but that does not tell them how the pollinators are actually behaving or whether they are meeting the orchard's requirements.⁴ Two commercial systems—BeeHero's in-hive acoustic sensors and AgriSound's Polly field-mounted acoustic monitors, the latter recently deployed across 73 hectares of Spanish and Portuguese almond orchards in a partnership with Importaco—have begun to close that gap with sound.⁴⁵ mmWave radar adds two capabilities those acoustic systems do not provide: species-level discrimination, and the ability to detect insects that are not making detectable sound at the sensor.

A working operational concept (CONOPS) for the grower has three phases keyed to the agricultural calendar:

Phase 1: Pre-Bloom Baseline

For roughly two weeks before bloom, the sensor network establishes a baseline of background insect activity—what wild pollinators are present, in what species mix, at what times of day, and in which zones of the field. This baseline sets expectations and identifies pollinator-poor zones that may need supplemental managed colonies. A grower investing in hedgerows, wildflower strips, or cover crops can see whether those investments have actually translated into wild bee residency before the crop needs them.

Phase 2: Bloom-Period Real-Time Monitoring

This is where most of the operational value lives. During the 5–14 day peak bloom window, the sensor network produces near-real-time, species-resolved activity maps with the following directly actionable outputs:

• Pollination contract verification. California almond growers paid roughly $200–$300 per hive in 2025–2026, and tight bee supplies after the 2024–2025 colony collapse have driven scrutiny of contract performance.⁶ Radar-confirmed honeybee foraging activity in the immediate orchard, distinguished from non-honeybee insects nearby, gives both grower and beekeeper an independent, third-party-quality record of whether contracted hives are actually working the trees.
• Foraging-aware pesticide and fungicide timing. EPA's 2017 policy already restricts agricultural spray applications during contracted bloom, but timing within the bloom window is at the grower's discretion.⁷ Real-time activity data lets a grower apply early morning or late evening when foraging is at its diurnal minimum, and skip applications entirely on days of unexpectedly high visitor activity.
• Wild-pollinator dependence assessment. For crops where wild bees do significant work alongside managed honeybees—blueberries, squash, watermelon, many tree fruits—the species mix tells the grower whether the crop is being pollinated mostly by paid hives or by free wild labor. That changes hive-rental decisions in subsequent years.
• Trigger thresholds for contingency hives. If activity drops below a calibrated threshold mid-bloom—weather event, pesticide drift from a neighbor, unexplained colony failure—the grower has 24–48 hours to deploy emergency hives. Without monitoring, the loss is typically discovered at harvest, when nothing can be done.
• Spatial heterogeneity mapping. AgriSound notes that "two orchards with the same hive density can behave very differently. Even within a single orchard, some areas may be highly active while others remain relatively quiet."⁵ Activity heatmaps direct hive relocations within the season and inform tree-spacing and variety placement in the next planting.

Phase 3: Post-Bloom Yield Correlation

After harvest, the season's pollinator activity record is overlaid on the yield map produced by the combine or hand-counts. Over two or three seasons this yields a quantitative dose–response curve linking species-specific pollinator visit-minutes per square meter to fruit set, kernel weight, and quality grade. That curve becomes the basis for next year's hive-stocking rate, contract pricing, habitat investment, and crop insurance underwriting.

UAV Deployment: Engineering Feasibility

The natural question for any precision-agriculture engineer is whether the radar can ride on a small UAV rather than (or in addition to) a fixed pole. The answer is a qualified yes, and a research program already exists to prove it: the University of Hawai'i Mānoa group led by Yao Zheng and Daniel Jenkins is developing a drone-mounted 12/24 GHz harmonic radar for tracking invasive coconut rhinoceros beetles and melon flies across Hawaiian agriculture, funded for 2026 by the Hawai'i Invasive Species Council in partnership with USDA-PBARC.⁸ That program uses harmonic radar (which requires a transponder tag) rather than micro-Doppler classification, but it establishes the airframe and RF-payload baseline that a passive mmWave classifier would inherit.

SWaP and Frequency Tradeoffs

A modern automotive cascaded mmWave radar transceiver—Texas Instruments AWR2243, NXP S32R45, or Infineon RXS81xx-class parts at 76–81 GHz—dissipates roughly 5–8 W and weighs under 25 g without the antenna array. Add a chip-scale patch antenna, an embedded GPU or ARM-class processor for the classifier, a stabilization gimbal, and minimal cabling, and the total payload is 100–200 g. That fits comfortably on any sub-2 kg multirotor with 25–35 minutes of hover endurance. The same payload at 60 GHz (the unlicensed V-band used for 5G fixed wireless) trades modest sensitivity loss for more available bandwidth and lower regulatory friction; at 94 GHz the SWaP tightens but Doppler resolution improves and the antenna gets smaller for a given gain.⁹

What Limits the Sensor

Three engineering constraints dominate the deployment trade space.

Insect radar cross-section. A honeybee at W-band presents an effective radar cross-section roughly five to six orders of magnitude smaller than a small drone—on the order of 10⁻⁵ to 10⁻⁴ m².³ By the fourth-power range dependence in the radar equation, this collapses practical detection ranges to roughly 5–15 m for the wingbeat micro-Doppler features the classifier needs. That is a constraint, not a fatal one: most pollination happens in a thin foraging layer 0.5–3 m above the canopy, which a UAV at 5–8 m altitude can easily cover.

Atmospheric attenuation. The 60 GHz V-band sits squarely in the oxygen absorption peak (~15 dB/km one-way), but that is irrelevant at 10 m ranges. Rain attenuation matters more for outdoor operation; mmWave systems lose significant range in heavy rain, which conveniently is also when pollinators stop flying.

Ego-motion clutter. A multirotor's rotors and airframe vibrate at frequencies that overlap the wingbeat band of interest. Mitigation requires either (a) a tethered hover platform that can be dynamically stabilized, (b) onboard motion compensation using IMU-derived ego-Doppler estimates, or (c) brief station-keeping pauses with rotors at idle—the last being non-trivial for safety. Fixed-wing UAVs avoid the rotor problem but cannot hover.

Three Deployment Architectures

Architecture	Best For	Coverage / Sensor	Capex Class	Key Tradeoff
Fixed pole nodes	High-value perennial crops (almond, cherry, blueberry orchards)	~80–300 m² each	Low per node, scales with hectares	Many sensors required for whole-field coverage
Tethered hover UAV	Mid-size operations needing one mobile zone of attention	~1 ha per platform	Medium; one platform per 25–50 ha	Cable management; weather-limited
Free-flying multirotor scout	Large diffuse fields (canola, sunflower, hybrid seed)	~1 ha per 30-min flight at 1.5 m/s	Medium; scales by flight tempo, not hectares	Battery cycles; ego-motion clutter; pilot/BVLOS waivers
5G/6G base-station co-host	Long-term, regional biodiversity monitoring	Cell-coverage scale	Effectively free if telecoms cooperate	Sensor placement optimized for comms, not entomology

The Antony et al. paper explicitly flags the fourth option: their 60 GHz hardware is "compatible with emerging mmWave communication and sensing infrastructures, such as 5G/6G and IoT," meaning that the same towers carrier networks are installing along rural highway corridors could one day report pollinator species data without dedicated sensors.² That is a long-horizon prospect, but it changes the unit economics of large-scale monitoring in a way that fixed entomological networks never could.

Recommended Hybrid Architecture

For a representative California almond grower running a 100-hectare operation, the operational picture that emerges is roughly this:

• Permanent fixed nodes at canopy-adjacent height every 30–50 m through the highest-value blocks, transmitting hourly aggregated species counts to a farm-management dashboard. Density: ~50–100 sensors. This is the steady-state monitoring layer.
• One or two tethered hover platforms deployed during peak bloom for about ten days, providing high-resolution coverage of any block where fixed-sensor data has flagged anomalies.
• Periodic free-flying scout flights—weekly during bloom, monthly off-season—for whole-orchard sweeps that map species mix at higher spatial resolution than any practical fixed network can deliver.
• Cellular co-located sensors at the carrier's nearest 5G mmWave node, where available, providing perimeter and regional-context data at no marginal cost.

This architecture answers the practical question of "how do I afford to monitor 100 hectares to species level" by acknowledging that the grower does not need uniform coverage—they need dense coverage where pollination is the binding yield constraint and survey coverage everywhere else.

Where This Fits Among Existing Tools

mmWave radar is not a replacement for in-hive monitoring (BeeHero, with roughly 25 million daily in-hive samples by mid-2024¹⁰) nor for acoustic field monitoring (AgriSound Polly, deployed at 120-plus sensors per orchard in commercial trials⁵). Each layer answers a different question:

• In-hive sensors tell the beekeeper and grower that the colony is alive, queen-right, and growing.
• Acoustic field sensors tell the grower that bees are flying in the orchard and roughly how active they are.
• mmWave radar tells the grower which species are flying, where, and when—information that neither of the others provides.

For a high-value perennial crop with a complex pollinator portfolio—blueberries pollinated by a mix of honeybees, bumblebees, mason bees, and native bees, for example—species-level data closes the last meaningful information gap in the pollination value chain.

The Backdrop That Justifies the Investment

The case for any of this rests on a recent and uncomfortable empirical record. U.S. commercial beekeepers lost an average of 62 percent of their colonies between June 2024 and February 2025—approximately 1.7 million hives, valued at roughly $600 million—and USDA's Agricultural Research Service has since traced the cause to amitraz-resistant Varroa mites vectoring deformed wing virus and acute bee paralysis virus.¹¹¹² The October 2025 IUCN European Red List update classified 172 of 1,928 wild bee species (about 10 percent) as threatened with extinction, more than double the 2014 figure.¹³ A May 2026 Nature study documented that wild pollinators provide 44 percent of farming income and over 20 percent of vitamin A, folate, and vitamin E intake in vulnerable smallholder communities.¹⁴

For a working grower, the practical implication is straightforward. The supply curve for managed pollination services has shifted hard to the left—fewer hives available, at higher prices, with greater year-to-year variance—at the same time that the wild-pollinator backstop is thinning. Quantitative pollination monitoring stops being a research curiosity under those conditions and starts being a risk-management tool.

Forward Roadmap

Three things must happen for mmWave pollinator radar to move from peer-reviewed proof of concept into routine commercial agriculture.

First, the species library has to expand. The Trinity team's published dataset covers a small set of European pollinators; a U.S. grower needs reference signatures for the local Bombus, Xylocopa, Osmia, Megachile, hoverflies, and several non-target species (yellowjackets, paper wasps, bee mimics). Each new species requires controlled training data with known ground truth, which is the limiting factor in classifier scope.

Second, outdoor classifier robustness must be validated. Wind-blown vegetation, blowing dust, ambient temperature drift, and overlapping insects in a single radar resolution cell all degrade classification accuracy. Field trials currently underway at Trinity, and parallel work at the Hawaii drone-radar program, will produce the data needed to characterize that degradation.

Third, the integration layer—from radar return to grower dashboard to action—has to come down to commodity pricing. BeeHero, AgriSound, and PollenOps have built that pipeline for acoustic and in-hive sensors,¹⁰¹⁵ and there is no obvious technical reason a radar feed should cost more to pipe through than any other IoT signal. The first commercial deployments will likely come from one of those existing precision-pollination platforms adding mmWave as an additional sensor modality, rather than from a clean-sheet radar startup.

The instrument has, in short, finally arrived. Whether the agricultural sector picks it up at the speed the colony-loss data suggests it should is the open question.

---

Verified Sources

1. Klein, A.-M., et al., "Importance of pollinators in changing landscapes for world crops," Proceedings of the Royal Society B, 274:303–313, 2007. Foundational figure: 87 of leading global food crops, ~35% of crop production volume. Cited in CABI Reviews (2024).
   https://www.cabidigitallibrary.org/doi/10.1079/cabireviews.2024.0016
2. Antony, L., White, C., Marchetti, N., Donohue, I., Stout, J. C., and Narbudowicz, A., "Harnessing mmWave signals and machine learning for noninvasive taxonomic classification of insects," PNAS Nexus, vol. 5, no. 4, pgag096, April 2026. DOI: 10.1093/pnasnexus/pgag096.
   https://academic.oup.com/pnasnexus/article/5/4/pgag096/8662959
3. Wang, R., Hu, C., Fu, X., Long, T., and Zeng, T., "Micro-Doppler measurement of insect wing-beat frequencies with W-band coherent radar," Scientific Reports, 7:1396, May 2017. DOI: 10.1038/s41598-017-01616-4.
   https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5431090/
4. "How precision pollination promises to increase crop yields," AgTech Navigator, March 2024 (BeeHero Pollination Insight Platform 2.0; California almond deployments).
   https://www.agtechnavigator.com/Article/2024/03/06/how-precision-pollination-promises-to-increase-crop-yields/
5. "AgriSound and Importaco put pollination under the microscope to de-risk almond sourcing," AgTech Navigator, April 2026 (Polly sensor deployment, Zurria Spain 50 ha and Freixo Portugal 23 ha).
   https://agtechnavigator.com/Article/2026/04/10/agrisound-and-importaco-put-pollination-under-the-microscope-to-derisk-almond-sourcing/
6. "Bees Are Gold: Almond Pollination in 2026," AgNet West, February 2026 (BeeHero interview at World Ag Expo on tight bee supplies and rising costs after 2024–25 colony losses).
   https://agnetwest.com/almond-pollination-2026-bee-hero/
7. U.S. Environmental Protection Agency, "EPA Actions to Protect Pollinators" (2017 spray-application policy during contracted bloom).
   https://www.epa.gov/pollinator-protection/epa-actions-protect-pollinators
8. Zheng, Y., Cai, H., and Jenkins, D., "Drone-Mounted mmWave Harmonic Radar for Invasive Insect Monitoring," University of Hawai'i Mānoa, funded 2026 by Hawai'i Invasive Species Council. Field validation with USDA-PBARC and NIWC Pacific.
   http://www2.hawaii.edu/~yaozheng/grant/2026_hisc_harmonic_radar/
9. Bauer, S., Tielens, E. K., and Haest, B., "Monitoring aerial insect biodiversity: a radar perspective," Philosophical Transactions of the Royal Society B, 2024. DOI: 10.1098/rstb.2023.0113. Reviews radar entomology and frequency-band tradeoffs.
   https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11070259/
10. "BeeHero Launches Pollination Insight Platform 2.0," BeeHero corporate release, July 2024 (25 million in-hive samples per day; species range identifiable: leaf-cutter bees, hover flies, carpenter bees, bumblebees, honeybees).
    https://www.beehero.io/research/beehero-launches-pollination-insight-platform-2-0
11. Auburn University College of Agriculture / Project Apis m. / American Beekeeping Federation / Apiary Inspectors of America, "U.S. Beekeeping Survey 2024–2025," reported June 2025. Annual losses 55.6%; commercial losses 62%; range 34.3%–90.5%.
    https://agriculture.auburn.edu/feature/u-s-beekeeping-survey-reveals-highest-honeybee-colony-losses-during-2024-2025/
12. USDA Agricultural Research Service, "USDA Researchers Find Viruses from Miticide Resistant Parasitic Mites are Cause of Recent Honey Bee Colony Collapses," 2 June 2025. Peer-reviewed publication: Lamas, Z. S., Evans, J. D., et al., PLOS Pathogens, 23 February 2026.
    https://www.ars.usda.gov/news-events/news/research-news/2025/usda-researchers-find-viruses-from-miticide-resistant-parasitic-mites-are-cause-of-recent-honey-bee-colony-collapses/
13. IUCN, "Mounting Risks Threaten Survival of Wild European Pollinators – IUCN Red List," 11 October 2025.
    https://iucn.org/press-release/202510/mounting-risks-threaten-survival-wild-european-pollinators-iucn-red-list
14. Timberlake, T. P., et al., "Pollinators support the nutrition and income of vulnerable communities," Nature, 6 May 2026. DOI: 10.1038/s41586-026-10421-x.
    https://www.nature.com/articles/s41586-026-10421-x
15. "Pollination Contract Software for Almond and Berry Growers," PollenOps, 2026 (grower portal for hive count verification, placement, and bloom alignment).
    https://pollenops.com/pollination-contract-software-for-growers