The study, led by Jianan Ding, Arjan Boonman, Ofri Eitan, Yuval Werber, David Troupin, Nir Sapir, and Yossi Yovel, deployed passive acoustic recorders at 27 sites, 12 urban and 15 rural, spread across nearly 300 kilometers of Israel. The sites ranged from temperate forests receiving roughly 800 millimeters of annual rainfall in the north to semi-arid desert zones receiving about 200 millimeters in the south. Each AudioMoth full-spectrum recorder was placed within 20 meters of a meteorological station, allowing the researchers to tie every recorded bat call to precise, locally measured weather conditions. From August 1 to October 13, the team logged 1,151 nights of recordings, capturing 95,473 bat occurrences in total.

The technical setup was demanding. Recorders operated on a 50 percent duty cycle, capturing 30-second audio sessions every minute for 12 hours each night from sunset to sunrise, generating nearly 720 audio files per site per night. A custom-written MATLAB program then detected echolocation calls and automatically classified them to species based on signal frequency, duration, and harmonic content. To validate the automated pipeline, the researchers manually reviewed complete recordings from 30 nights at each site. Automatic classification was used only where accuracy exceeded 90 percent; at sites with intermediate accuracy, every identification was manually verified, and at two sites with accuracy below 70 percent, all files were classified by hand. Three sites yielded no bat detections at all.

Of the 23 bat species detectable in the region with the equipment’s sampling constraints, the team recorded 11 species from 7 genera and 4 families across 24 sites. Rural sites proved richer, hosting all 11 species, while urban sites supported 8, all of which also occurred in rural areas. For seven of the species found in both habitat types, activity was significantly higher at rural sites, measured as encounters per hour. The sole exception was Tadarida teniotis, the European free-tailed bat, which was significantly more active in urban areas, where it typically roosts. Species richness also correlated positively with latitude, reflecting the region’s rainfall gradient, with the five most species-rich sites all clustered in the wetter north.

The study’s central insight emerged when the researchers examined how environmental variables influenced bat activity. They divided species into two functional groups: generalists, detected at both urban and rural sites, and rural specialists, found exclusively outside cities. Using zero-inflated negative binomial models, they tested 511 predictor combinations per species and selected the best-supported models by Akaike Information Criterion. The predictors included temperature, humidity, wind speed, ambient illumination, the proportion of built-up area, average building height, environmental noise, hours from sunset, and latitude. The pattern was strikingly consistent: rural specialists were the most sensitive to weather, generalists at rural sites showed intermediate sensitivity, and generalists at urban sites were the least affected of all.

On average, bat activity rose with temperature and humidity and fell with wind speed and illumination, but the strength of these effects varied dramatically by habitat. Rural bats strongly preferred low wind speeds, a preference the authors attribute to thermoregulatory costs, acoustic interference with echolocation, and the behavior of insect prey, whose own flight activity drops in windy conditions. Urban bats, by contrast, showed little to no wind aversion. When the analysis isolated moonlight as the sole illumination variable, activity at rural sites generally declined as moonlight intensified, consistent with heightened perceived predation risk, whereas most urban species showed no significant lunar response, likely because city lights wash out the moon’s signal entirely.

Comparing the environmental conditions under which bats actually flew revealed the buffering effect in vivid detail. Rural specialists were the most restricted animals in the dataset, active only at higher temperatures and lower humidities than generalist species foraging at the very same sites. Generalists, meanwhile, operated across a much wider range of conditions, but remarkably, the same species extended their activity to lower temperatures, higher humidities, and stronger winds when hunting in cities than when hunting in the countryside. A Kullback-Leibler divergence analysis, which quantifies how different the weather during bat activity was from the weather across the entire night, confirmed that this mismatch was greatest for rural specialists and smallest for urban generalists.

The physical environment itself helps explain the pattern. During the study period, urban sites averaged roughly 2 degrees Celsius warmer and about 5 percent more humid-differing in humidity by about 5 percent lower-than nearby rural sites, a signature of the urban heat island effect and the reduced weather variability that cities create. The authors suggest several mechanisms that might allow urban bats to tolerate harsher conditions: the proximity of reliable food, the availability of irrigated parks, ponds, and lakes that attract both insects and bats in a water-scarce region, and the microclimatic shelter provided by buildings. Seasonal trends reinforced the story, as species richness and activity declined markedly toward winter at rural sites but remained comparatively stable in cities, hinting that urban environments may extend the active season for city-dwelling bats.

The researchers are careful to note the study’s limitations. Acoustic monitoring counts call encounters rather than individual bats, so differences in activity could partly reflect differences in the number of animals present. The monitoring window covered only 2.5 months from midsummer to autumn, leaving other seasons unexamined. And because the work was entirely remote, it cannot definitively establish that urban buffering causes bats to broaden their activity windows; unmeasured factors such as prey availability and fine-scale habitat structure may also contribute. The microphone setup, sampling at 96 kilohertz, could reliably detect only relatively loud calls below 50 kilohertz, excluding nine local species including all horseshoe bats, though previous work consistently identifies those species as rural specialists unlikely to alter the conclusions.

Beyond its findings on bats, the study showcases the power of passive acoustic monitoring combined with automated species classification for biodiversity research at scales that would be impossible with traditional surveys. As urbanization accelerates worldwide, the results add nuance to a usually grim picture: while rural specialists avoid cities and overall diversity is lower in urban areas, the generalist bats that do persist there appear to gain a measure of climatic freedom, operating under conditions that would ground their rural relatives. In an era of rapid environmental change, the city, for some species at least, may be less a threat than a shelter.

Subject of Research: The effects of urbanization and weather conditions on the activity of insectivorous bats

Article Title: Urban habitats offer buffered environmental conditions for city-dwelling bats

Article References: Ding, J., Boonman, A., Eitan, O., Werber, Y., Troupin, D., Sapir, N., & Yovel, Y. (2026). Urban habitats offer buffered environmental conditions for city-dwelling bats. iScience, 29(10), Article 117767. https://doi.org/10.1016/j.isci.2026.117767

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117767

Keywords: bats, urbanization, passive acoustic monitoring, echolocation, urban heat island, weather, moonlight, wildlife behavior, Israel, iScience, biodiversity, urban ecology

The research, led by Xia Lin and Baojun Lin of the Aerospace Information Research Institute at the Chinese Academy of Sciences, demonstrates that adding just four satellites in elliptical lunar frozen orbits, known as ELFO, to China’s BeiDou-3 navigation constellation can eliminate the rotational drift that has plagued autonomous navigation concepts since their inception. The team combined twenty-four real BeiDou-3 Medium Earth Orbit, or MEO, satellites with four simulated lunar satellites in a centralized extended Kalman filter, processing sixty days of genuine onboard inter-satellite link ranging measurements alongside simulated Earth-moon links. The result was remarkable: the constellation’s orientation remained stable for the full two-month span, the mean user range error held at just 0.35 meters, and the lunar satellites themselves were located to within 2.3 meters in three dimensions.

To appreciate why this matters, it helps to understand how autonomous orbit determination works and why it fails. Modern navigation satellites are not solely dependent on ground stations; they carry inter-satellite links that let them range against one another and compute their own orbits onboard. This autonomy is a strategic asset, because ground stations can fail, be disrupted, or fall out of contact during crises. But inter-satellite ranging has a fundamental blind spot. The measurements capture only relative geometry, the distances between satellites, and relative distances are completely insensitive to a rigid rotation of the whole network. Mathematically, the rotation of the constellation is unobservable: no matter how the network turns as a whole, the ranging data remains unchanged. The system is rank-deficient, and the missing information must come from somewhere else.

Operators have long compensated for this with prediction-based corrections, using forecast orbits generated on the ground to nudge the constellation back toward its expected orientation. This slows the drift but never stops it, because the forecasts themselves carry errors that grow over time. The problem becomes especially acute during eclipse seasons, when satellites pass through Earth’s shadow and solar radiation pressure models become less reliable, accelerating the degradation of predicted orbits. The error accumulates relentlessly, and without periodic ground intervention the constellation’s absolute orientation eventually becomes unusable. What the field has needed is an external, dynamics-based reference, something in the measurement data itself that makes rotation visible without requiring constant contact with Earth.

The insight behind the new study is that the moon provides exactly that reference, through the physics of gravity rather than through any additional measurement. Satellites in elliptical lunar frozen orbits are governed overwhelmingly by lunar gravity, while BeiDou-3 MEO satellites are governed overwhelmingly by Earth’s gravity. Because these two gravitational environments are so different, the two sets of satellites respond differently to any rotation of the combined network. When the whole system rotates, the lunar satellites’ dynamics pull their computed positions in a way that is inconsistent with the ranging data, and that inconsistency makes the rotation detectable. In effect, the lunar satellites act as a dynamical anchor: their motion is so strongly constrained by the moon’s gravity that they pin down the absolute orientation of the entire constellation.

The numbers from the observability analysis make the transformation vivid. With MEO satellites alone, the condition number of the position information matrix, a measure of how ill-conditioned the estimation problem is, sat between roughly ten to the fifteenth and ten to the eighteenth, effectively a singular case in which the rotation simply cannot be resolved. Once four ELFO satellites were added, the condition number fell to about ten squared, a well-behaved problem that a filter can solve robustly. In the joint Earth-moon solution, the three-axis rotational biases over sixty days stayed within 13.59, 10.27, and 4.04 milliarcseconds, compared with more than 300 milliarcseconds when no correction was applied and about 18 milliarcseconds with traditional prediction-based correction. The BeiDou-3 user range error, the quantity that ultimately determines how accurately a user on the ground can be positioned, remained at 0.35 meters, against 7.85 meters without correction and 0.60 meters with prediction-based correction.

The lunar satellites did not merely serve as passive anchors; they achieved impressive accuracy in their own right. Their maximum radial errors stayed below 0.16 meters, while along-track and cross-track errors remained under 1.7 meters and 1.8 meters respectively. This dual benefit is significant for the future of cislunar space operations, because it means the same infrastructure that stabilizes Earth’s navigation constellation also delivers precise positioning for spacecraft operating near the moon. The ELFO itself is a particularly favorable orbit for this purpose, offering good coverage of the lunar polar regions while demanding only modest station-keeping budgets, a combination that has already attracted the attention of major space agencies.

Indeed, the orbital choice aligns closely with international plans for lunar infrastructure. Both the European Space Agency’s Moonlight program and NASA’s LunaNet initiative have selected elliptical lunar frozen orbits as the reference orbit for their planned lunar navigation and communication constellations, specifically to provide continuous coverage of the lunar south pole, where water ice deposits have made the region a prime target for future crewed and robotic missions. The Chinese study suggests that these lunar constellations could do double duty, serving lunar explorers while simultaneously anchoring Earth-orbiting navigation networks. A single architecture, spanning the Earth-moon system, could deliver resilient positioning, navigation, and timing services to users on both worlds.

A crucial strength of the work is its grounding in real operational data. Rather than relying entirely on simulations, the researchers used sixty days of genuine onboard inter-satellite link measurements from BeiDou-3 MEO satellites, pairing them with simulated Earth-moon links. This hybrid approach gives a far more realistic picture of what today’s operational systems can actually achieve, since real ranging data carries the noise, biases, and scheduling constraints that idealized simulations tend to smooth away. The authors note that the lunar satellites essentially act as an anchor for the entire network, and that the joint solution addresses a limitation that has persisted since the earliest autonomous navigation concepts were proposed decades ago.

The implications extend well beyond the Earth-moon system. Long-duration autonomous navigation for Earth-orbiting constellations would reduce reliance on vulnerable ground infrastructure and improve resilience against disruption, a concern of growing importance as navigation services underpin everything from aviation and shipping to financial transaction timestamps and power grid synchronization. The authors caution that this is still a preliminary analysis: future work will replace the simulated lunar links with real onboard observations once lunar satellite missions become operational, and will develop link scheduling strategies that simultaneously support the orbit determination solution and provide continuous positioning, navigation, and timing services to lunar users. But the core result stands as a proof of concept that the moon, the celestial body humanity has watched and navigated by for millennia, may soon serve as the fixed reference point that keeps Earth’s satellite navigation networks true.

Subject of Research: Joint autonomous orbit determination of BeiDou-3 MEO satellites and lunar ELFO satellites using inter-satellite links

Article Title: Lunar satellites could anchor Earth's GPS: study finds a fix for drifting navigation constellations

Article References: Lunar satellites could anchor Earth's GPS: study finds a fix for drifting navigation constellations. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: BeiDou-3, autonomous orbit determination, elliptical lunar frozen orbit, inter-satellite links, extended Kalman filter, GNSS, lunar navigation, user range error, observability, Moonlight, LunaNet, Satellite Navigation journal