The research, led by Facundo X. Palacio of Argentina’s Museo de La Plata and CONICET, tested whether a simple photogrammetric framework known as the pinhole camera model can produce reliable estimates of bird morphology from field photographs. The model describes how a three-dimensional scene is projected onto a two-dimensional camera sensor: light from a point in the scene passes through the aperture and maps to a corresponding point on the sensor, so the physical size of an object relates proportionally to its size in the image, its distance from the camera, and the focal length of the lens. In other words, if you know how far away the bird is and what lens you used, you can convert pixels into millimeters.

That idea had already proven itself on large mammals. Researchers using camera traps estimated the size of reindeer with a mean relative error of roughly minus nine percent, and similar approaches have been applied to leopards and other terrestrial species. But large animals are forgiving subjects. Small, mobile, wary creatures like birds pose a far harder problem: they flee, hide in dense vegetation, perch high in the canopy, and present constantly changing angles to the lens. Earlier attempts hinted at promise—one team estimated bill, head, tarsus, and body length in free-ranging chickens with estimates that did not differ from a one-to-one relationship with true values—yet those tests involved a single large-bodied species at short distances. A stereo-camera study of kiwi achieved remarkable accuracy on taxidermied specimens in daylight but produced biases exceeding twenty percent on live birds at night.

To find out whether the method could survive contact with real fieldwork, the team first built a controlled experiment. Four Neotropical species spanning roughly 25 to 300 grams—House Sparrow, Rufous-bellied Thrush, Campo Flicker, and Chimango Caracara—were represented by museum-mounted specimens posed in lifelike positions. Using a Canon DSLR with a 70–300 mm lens fixed at 1.50 meters on a tripod, the researchers photographed each specimen from three horizontal distances (3, 5, and 7 meters), eleven different heights along a ladder, and four orientations relative to the focal plane: 0, 20, 40, and 60 degrees. In total, 580 photographs were captured, with reference values established from images that included a metric scale. The design deliberately mirrored the geometry of real birdwatching: the distances chosen match those at which observers typically encounter birds.

The results were strikingly patterned. Under optimal conditions, with the specimen parallel to the focal plane, body and wing length showed the lowest biases—between one and six percent and one and nine percent respectively—while tarsus and bill length fared somewhat worse, at five to twelve and two to fifteen percent. When the researchers fitted generalized least squares models to explain the absolute relative bias, the three predictors—distance, viewing angle, and elevation angle—accounted for nearly all the variability, with pseudo-R-squared values between 0.974 and 0.995. The dominant factor was the angle of the bird relative to the focal plane: the further a bird rotated away from parallel, the more its projected length shrank, a classic foreshortening effect. Distance, counterintuitively, had only weak and variable effects, and moderate distances of three to ten meters actually emerged as a sweet spot, balancing geometric distortion against image resolution.

The trait differences have a logical anatomical basis. Bills and wings are typically fully exposed on a perched bird, making their outlines easy to trace in an image. The tarsus, by contrast, is often partially hidden by feathers at the joint with the tibia, forcing observers to guess where the structure begins. Body length is harder still, because it must be measured along a bird’s natural, often curved posture rather than the relaxed, flattened position used in museum protocols. The authors therefore recommend prioritizing bill or wing length when measuring birds photographically, and they note that the tarsus also showed the lowest repeatability across repeated photographs of the same individuals—an intraclass correlation of 0.73, compared with 0.94 for body length, 0.81 for wing length, and 0.80 for bill length.

Then came the real test. Over ten surveys between July and November 2024 in the talares forests and Ligustrum-invaded woodlands of the Parque Costero del Sur Biosphere Reserve in Buenos Aires province, the team photographed 99 individual birds from 37 species, recording the horizontal distance to each bird with a tape measure and its elevation angle with a clinometer. Mean shooting distance was about 9.5 meters, ranging from 1.55 meters for a Mottle-cheeked Tyrannulet to over 39 meters for a White-tipped Dove. Because capturing the same wild birds to verify measurements was impractical, the photographic estimates were benchmarked against two independent references: the global AVONET morphological database and caliper measurements of museum specimens of the same species from the Museo de La Plata collection.

The correlations were strong across the board. Photographic estimates correlated with AVONET values at 0.84 for tarsus length up to 0.92 for bill length, and with museum measurements at 0.82 to 0.95, all highly significant. Uncorrected estimates mostly fell below the one-to-one line—a systematic underestimation expected from foreshortening—but the relative ranking of species was preserved. That preservation matters enormously for community ecology, because many functional diversity metrics depend on the relative positions of species in trait space rather than their absolute values. When the researchers computed functional richness and functional divergence for their photographed assemblage and compared them against 500 bootstrap assemblages simulated from AVONET, the observed values fell comfortably within the simulated distributions, suggesting that photo-based trait data can reproduce community-level patterns.

The team also applied a trigonometric correction, dividing measured lengths by the cosine of the body’s inclination angle to compensate for out-of-plane rotation. The correction brought estimates closer to the one-to-one relationship on average, though with greater variability, and the authors caution that it is an approximation rather than a full three-dimensional reconstruction. A single two-dimensional image collapses depth information, so different combinations of yaw and pitch can produce indistinguishable projections, and the rotation angle of the whole body does not necessarily match the effective projection angle of each individual structure. Their practical advice: photograph birds as parallel to the focal plane as possible, at low elevation angles, from moderate distances, and take two or three shots of the same individual to average out error.

The implications reach well beyond ornithological curiosity. Photogrammetry could open morphological research to raptors, parrots, and other taxa that are difficult or risky to capture, and to environments like dense cities where trapping is impractical. The same geometry can quantify prey items carried to nests, as demonstrated in seabird diet studies, or measure the architecture of conspicuous but inaccessible structures like Rufous Hornero oven nests—all without handling a single feather. The authors are candid about the caveats: four experimental species cannot represent the full diversity of bird form, photographic sampling may favor bolder or more exposed individuals, and body length from a photo is an approximation of the standard total length rather than an equivalent. But the core message stands. With nothing more than a camera, a tape measure, and a clinometer, ecologists can now measure the birds they see—no nets required.

Subject of Research: Photogrammetric estimation of bird morphological traits from field photographs using the pinhole camera model

Article Title: Measuring Bird Morphology From Field Photographs

Article References: Palacio, F. X., Elso, E., & Montalti, D. (2026). Measuring Bird Morphology From Field Photographs. Ecology and Evolution, 16(10), Article e74422. https://doi.org/10.1002/ece3.74422

Image Credits: AI Generated

DOI: 10.1002/ece3.74422

Keywords: photogrammetry, pinhole camera model, bird morphology, morphometrics, ornithology, field photography, functional diversity, AVONET, non-invasive methods, community ecology, trait measurement, Ecology and Evolution

The research, carried out by Nasser Laouami of the National Applied Research Centre in Earthquake Engineering in Algeria, tackles one of the most persistent gaps in seismic hazard assessment. Many strong-motion stations worldwide were installed decades ago, often on the roofs or basements of buildings, with little or no accompanying subsurface investigation. Drilling new boreholes at each site is costly and time-consuming, so the average shear-wave velocity of the top 30 metres, known as Vs30, remains unknown or poorly constrained for a large fraction of the global seismic network. Yet Vs30 is the single parameter that most building codes, including the American NEHRP provisions and Eurocode 8, use to assign a site class and therefore to determine the design forces a structure must withstand. Without reliable Vs30 values, the ground-motion predictions that feed into hazard maps carry an extra layer of uncertainty.

Laouami’s approach exploits a phenomenon that seismologists have long used as a diagnostic tool: the horizontal-to-vertical spectral ratio, or H/V. When earthquake waves arrive at a station, the horizontal components of motion are typically amplified relative to the vertical component at frequencies where the shallow soil column resonates. That resonance frequency depends on the thickness and stiffness of the sedimentary layers, while the shape and amplitude of the H/V curve encode information about the impedance contrasts between layers. In principle, the entire H/V spectrum is a fingerprint of the subsurface velocity structure. The challenge has always been to invert that fingerprint, meaning to find a layered velocity model whose predicted H/V curve matches the observed one, without falling into the trap of non-uniqueness, where many different models fit the data equally well.

What makes the new method notable is that it requires no prior information about the subsurface. Most inversion schemes in geophysics lean heavily on initial guesses: an approximate depth to bedrock, a rough velocity for the top layer, or a fixed number of layers. When such constraints are wrong, the inversion can converge to a misleading solution. The new technique instead assumes that the strong-motion wavefield at the station is dominated by vertically incident body waves, a reasonable approximation for stations recording earthquakes at regional distances, and then searches for the velocity profile that best reproduces the observed H/V spectrum. The inversion is statistical in character, exploring a wide space of possible layered models and identifying the family of solutions consistent with the data rather than committing to a single profile that may be an artefact of the starting conditions.

To test the method rigorously, Laouami turned to one of the best-characterized seismic datasets on Earth: Japan’s KIK-NET network, operated by the National Research Institute for Earth Science and Disaster Resilience. KIK-NET stations are paired with boreholes that reach stiff engineering bedrock, and the shear-wave velocity profile at each site has been measured directly. That makes the network an ideal proving ground, because the true answer is known. The study used strong-motion records and measured velocity profiles from 66 KIK-NET stations, spanning a wide range of soil conditions from soft alluvial basins to relatively stiff sites. The inverted profiles were compared layer by layer against the borehole measurements, and the key classification parameter, Vs30, was compared station by station.

The results were evaluated in two configurations. In the first, the inversion ran completely unconstrained, with no prior information whatsoever about the subsurface. In the second, a small piece of shallow information was supplied: the average shear-wave velocity of the top five metres, or Vs5, which can be obtained cheaply from non-invasive surface measurements. The comparison revealed a clear and useful pattern. Without any prior information, the method produced reliable velocity profiles for sites whose Vs30 values reach approximately 600 metres per second, which covers the softer end of the soil spectrum, including many of the most seismically hazardous sedimentary settings. For stiffer profiles beyond that threshold, the unconstrained inversion began to lose accuracy, but adding the modest Vs5 constraint restored performance, extending reliable estimates to stiffer sites as well.

The quantitative agreement is impressive for a technique that requires no drilling. For most of the 66 stations, the relative error between the observed and predicted Vs30 was less than 20 percent, and the error in the predominant frequency of the H/V curve, the frequency at which the site resonates most strongly, was around 15 percent. In geotechnical and geophysical practice, both levels of accuracy are generally considered acceptable for preliminary site characterization. Perhaps more importantly, the method captured the overall shape of the velocity profiles, including the depth of major impedance contrasts, rather than merely matching a single averaged number. That matters because site response is not determined by Vs30 alone; the full layering controls how ground motion is amplified across the frequency range that affects buildings of different heights.

The implications extend well beyond Japan. Dense seismic networks exist in many earthquake-prone countries, but borehole measurements are far rarer, particularly in developing regions where instrumentation has expanded faster than site characterization budgets. A technique that extracts velocity profiles from recordings that stations are already producing, at no additional field cost, could rapidly fill the characterization gap. It could also serve as a first-pass screening tool: stations flagged by the H/V inversion as sitting on unusually soft or strongly resonant soils could be prioritized for follow-up borehole investigation. In regions governed by seismic codes that rely on Vs30-based site classes, such as the Algerian RPA2024 regulations or the NEHRP provisions, even a preliminary estimate with 20 percent accuracy can meaningfully improve hazard maps and design decisions.

The method also fits into a broader scientific conversation about what the H/V spectral ratio actually measures. Earlier theoretical work, including diffuse-field formulations of the H/V ratio developed by researchers such as Kawase, Sánchez-Sesma and Matsushima, showed that the ratio can be interpreted through the properties of the wavefield and the layered medium beneath the receiver. Laouami’s contribution is to demonstrate, against ground truth, that a body-wave assumption applied to strong-motion records is sufficient to recover the velocity structure without prior constraints, at least for a substantial range of site stiffness. The technique complements rather than replaces established non-invasive methods such as surface-wave analysis, which suffers from its own well-documented non-uniqueness problems, and it inherits the practical advantage of requiring only a single three-component station rather than an array of sensors.

Limitations remain, and the author is careful about them. The assumption of vertically incident body waves may weaken at sites with complex three-dimensional structure or strong lateral heterogeneity, and the accuracy degrades for very stiff sites unless shallow information is supplied. The validation set, while diverse, comes from a single national network with high-quality borehole data, so performance in other tectonic and sedimentary environments will need independent confirmation. Nevertheless, the core result stands: for a wide range of soil profiles, the ground beneath a seismic station can be mapped from its own earthquake recordings, with errors small enough to support site classification. In a field where every unknown parameter translates directly into uncertainty in shaking estimates and building safety margins, a reliable, non-invasive, zero-drilling method for recovering shear-wave velocity profiles is a genuinely valuable addition to the earthquake engineering toolkit, and one that could be applied to thousands of existing stations worldwide almost immediately.

Subject of Research: Non-invasive estimation of S-wave velocity profiles from single-station H/V spectral ratio inversion of strong-motion records

Article Title: Estimating S-wave velocity profiles from single-station horizontal-to-vertical spectral ratio inversion without prior information: application for KIK-NET accelerometric stations

Article References: Laouami, N. (2026). Estimating S-wave velocity profiles from single-station horizontal-to-vertical spectral ratio inversion without prior information: application for KIK-NET accelerometric stations. Bulletin of Earthquake Engineering. https://doi.org/10.1007/s10518-026-02660-0

Image Credits: AI Generated

DOI: 10.1007/s10518-026-02660-0

Keywords: shear-wave velocity, H/V spectral ratio, inversion, Vs30, site characterization, KIK-NET, strong-motion records, seismic station classification, earthquake engineering, geophysics, non-invasive methods, soil amplification