The challenge the team set out to solve is one familiar to anyone who manages a vast heritage site. Pompeii covers roughly 44 excavated hectares and contains approximately 1,500 buildings, and every one of them, along with its frescoes, statues, and structural elements, is slowly deteriorating under the open sky. Continuous monitoring is essential, but human inspection of every roof is slow, expensive, and sometimes dangerous. Drones offer a solution: they can fly over inaccessible areas and capture high-resolution, georeferenced photographs. Yet the raw images are only the beginning. Someone, or something, must interpret thousands of aerial photographs and decide which roofs need urgent attention.

HydraML’s answer is a two-stage pipeline that blends the strengths of deep learning with the efficiency of classical machine learning. In the first stage, a deep neural network, selected automatically by a deep AutoML process built on Amazon’s AutoGluon framework, examines each drone image and extracts a rich numerical representation, or embedding, of what it sees. Rather than using this network to make the final decision, HydraML passes those embeddings through a mathematical compression step based on the Fuzzy-Transform, a technique rooted in fuzzy set theory that reduces the dimensionality of the feature space while preserving the information that matters for classification. The compressed features then feed into a second AutoML system, which automatically selects and trains the best traditional classifier for the task.

This hybrid design is what sets HydraML apart from earlier approaches. Previous combinations of deep and classical methods required researchers to manually choose and tune both the feature extractor and the final classifier, a laborious process involving extensive hyperparameter optimization. By bringing both stages into the AutoML paradigm, HydraML becomes something close to plug-and-play: users specify only a training time budget for the deep component and a few parameters for the fuzzy compression, and the system handles the rest. The Fuzzy-Transform compression also pays dividends before training even begins, because smaller images mean less energy spent transmitting data from drone to cloud.

The mathematical heart of the compression step is elegant in its simplicity. The Fuzzy-Transform partitions an image or feature matrix using overlapping basic functions, which can be sinusoidal, Gaussian, or pi-shaped, each covering a neighborhood of the data. By projecting the original matrix onto these functions, the method produces a much smaller compressed representation. Crucially, prior theoretical work has shown that under reasonable conditions, a learning system trained on Fuzzy-Transform-compressed data can achieve error rates of the same order as one trained on the original data. In HydraML’s experiments, compressing 768-dimensional deep embeddings down to a 12-by-16 grid roughly halved both classifier training time and CPU energy consumption on the Pompeii datasets, with almost no loss in predictive performance.

To test the system, the researchers built two novel datasets from orthophotos of Pompeii captured between June 2023 and September 2024 by a DJI Matrice 300 RTK drone flying at 80 meters, equipped with a Zenmuse L1 camera. Nine orthophotos, each roughly 10 gigabytes, were georeferenced so that any pixel could be traced back to its exact location on the ground. With the help of domain experts, the team curated PompeiiTile, containing 1,327 images of tiled roofs labeled as normal, broken, or covered in weedy vegetation, and PompeiiFlat, containing 922 images of flat roofs labeled as normal, broken conduits, stained, or weedy. Careful data augmentation, applied separately to each cross-validation fold to avoid leakage, balanced the training and validation sets.

The results were striking. On PompeiiTile, HydraML using a cosine fuzzy partition achieved an F1-score of 0.9406, with an accuracy of 0.9404 and the lowest variance among all tested methods, signaling unusual stability. On PompeiiFlat it reached an F1-score of 0.9117. To probe generalization beyond cultural heritage, the team also evaluated HydraML on RescueNet, a publicly available dataset of 8,354 aerial images collected after Hurricane Michael struck Florida in 2018, categorized by FEMA-inspired damage grades from no damage to total destruction. There, HydraML achieved an F1-score of 0.8686, outperforming a DenseNet201-based hybrid approach and a VGG16-based transfer learning method, and beating the standalone deep AutoML system by 1.30 percentage points.

The energy savings may prove even more consequential than the accuracy gains. On the urban dataset, HydraML cut GPU energy consumption by about 62.5 percent compared with a DenseNet201-SVM hybrid, dropping from roughly 139,000 joules to 52,000 joules. On PompeiiTile, reducing training from 480 to 300 seconds saved around 40.92 percent of GPU energy, and on PompeiiFlat the system saved 43.64 percent of GPU energy and 12.28 percent of CPU energy relative to the standalone deep AutoML baseline. Notably, HydraML beat that baseline even though it added just 20 seconds for compression and classical classifier training. The team measured power directly, sampling CPU and GPU draw every 100 milliseconds during training and integrating over time, giving the comparison unusual rigor.

Beyond simply detecting damage, HydraML can grade its severity. Using the same deep embeddings without retraining the expensive neural stage, the system trains additional classifiers to sort each detected problem into small, medium, or high severity. It distinguished broken-tile severity with an F1-score of 0.97, and performed well on vegetation and conduit issues, though it struggled with damp patches on flat roofs, where F1 barely exceeded 0.75. Most classification errors occurred between adjacent severity levels, a difficulty the researchers note also plagues human experts defining the ground truth. This severity information matters because heritage managers use vulnerability indices to prioritize interventions, and accurate severity grading feeds directly into maintenance planning.

The system is not infallible, and the researchers are candid about its failure modes. Dirt and mold on tiles can mimic missing tiles, shadows on flat roofs can be mistaken for damp stains, and dry, reddish vegetation is harder to spot than lush green growth. In one telling example, a clearly broken roof section at the edge of an image was classified as normal because the model focused on the center. The team also acknowledges that the Pompeii datasets contain relatively few original samples and that the approach has not yet been tested at other archaeological sites. Still, the researchers point out that false positives are the safer error: a wrongly flagged roof gets checked by an expert, whereas a missed problem can quietly destroy irreplaceable frescoes and structures. With the code released openly and the Pompeii datasets available on request, HydraML offers heritage sites worldwide a template for monitoring that is simultaneously smarter, faster, and greener.

Subject of Research: A hybrid automated machine learning approach using deep learning, Fuzzy-Transform compression, and UAV imagery to classify damaged roofs at the archaeological site of Pompeii

Article Title: HydraML: a hybrid damaged roofs automated machine learning classifier for cultural heritage

Article References: Colace, F., De Santo, M., Gaeta, R., & Zuchtriegel, G. (2026). HydraML: a hybrid damaged roofs automated machine learning classifier for cultural heritage. Neural Computing and Applications, 38(19), Article 798. https://doi.org/10.1007/s00521-026-12499-w

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12499-w

Keywords: HydraML, automated machine learning, cultural heritage, Pompeii, deep learning, Fuzzy-Transform, roof damage classification, UAV imagery, sustainable AI, AutoGluon, energy efficiency, archaeological site monitoring

The significance of the work lies in the unusual nature of the calibration target. Most geological dating methods are validated against other laboratory measurements, creating a chain of inference in which small errors can compound. The Vesuvius eruption is different. Because Pliny the Younger recorded his first-hand observations of the disaster in extraordinary detail, scientists know the eruption date from written history rather than from any radiometric measurement. That makes the volcanic minerals erupted in 79 CE a kind of natural standard, a sample whose true age is known independently and against which the laboratory technique can be tested and refined.

Argon-argon dating relies on the slow, steady radioactive decay of potassium-40, a naturally occurring isotope, into argon-40. Because argon is a gas that normally escapes from minerals before they crystallize, any argon-40 trapped inside a volcanic mineral must have accumulated after the rock cooled. Scientists measure the ratio of argon-40 to argon-39, a proxy isotope created in the laboratory by irradiating the sample with neutrons, which converts a stable isotope of potassium, potassium-39, into argon-39. The more argon-40 relative to argon-39, the older the sample. The technique is prized because it spans an enormous range of time, from eruptions within living memory to meteorites that formed at the dawn of the solar system, but its usefulness depends entirely on knowing the decay rate of potassium-40 with precision.

The new study, led by Paul Renne, a Berkeley professor in residence of earth and planetary science and director of the independent Berkeley Geochronology Center, analyzed eight samples of sanidine, a potassium-rich volcanic mineral, from pumice ejected during the 79 CE eruption. When the minerals were analyzed in 2025, the argon-argon method pegged their age at 1,938 years, give or take 13 years. The actual age, derived from the historical record, is 1,946 years. That result corresponds to a precision of 0.7 percent and an accuracy of 0.4 percent, a level of performance that pushes the technique firmly into the historical realm. Accuracy, in this context, refers to whether a measurement reflects the true value, while precision describes how reproducible the measurement is; the Vesuvius benchmark shows the method now excels on both counts.

Getting to that level of performance required solving problems on two fronts: the laboratory and the archive. In the laboratory, the team benefited from better samples, an improved mass spectrometer, and updated neutron irradiation techniques. The samples themselves had an interesting history. In 1998, co-author Andrea Marzoli of the University of Padua collected new, more potassium-rich pumice from Oplontis, another Roman town buried by the same eruption. Crucially, these samples came from the earliest stage of the eruption. Magma chambers beneath stratovolcanoes like Vesuvius tend to stratify, with dense iron- and magnesium-rich magma settling to the bottom while soluble elements such as potassium concentrate near the top. Because the potassium-rich magma erupts first, it ends up at the base of the ash and pumice deposits, and Marzoli’s samples came from exactly those bottom layers. The samples were then shelved and never analyzed for nearly three decades.

Several years ago, graduate students Caroline Hasler, Anthony Fuentes, and Andy Tholt, working under postdoctoral fellow Jack Carter in Renne’s laboratory, proposed retrieving the old samples and attempting to improve on a 1997 analysis of Vesuvius sanidine that had predicted sub-one-percent precision was achievable. The effort dovetailed with a larger project: a Bayesian scheme, published in 2025, to intercalibrate the two most widely used geologic clocks, argon-argon dating and uranium-lead dating, which have historically produced results that do not always agree. The 1997 Vesuvius data had been included in that intercalibration but carried little statistical weight because of its lower precision. The new measurements change that, giving the historical benchmark real authority in reconciling the two methods. Renne also credited co-author Bill Cassata with a major role in developing the analytical strategy and data analysis, describing the success as a combination of better samples, instrumental advantage, and a more concerted effort.

Before any of the laboratory work could be interpreted, however, the team had to settle a long-running dispute among historians. Some scholars have argued that the eruption occurred later in the fall of 79 CE, citing a coin found at Pompeii that they claimed could only have been minted in September of that year. Hasler compared the coin with other contemporary Roman coinage and concluded that it was likely produced before September, supporting the traditional August 24 date derived from Pliny the Younger’s account. Adopting a liberal two-month uncertainty in the eruption timeline turned out to have little effect on calibrating the dating technique itself, but it proved essential for a second major result: a sharper determination of the half-life of potassium-40.

That half-life, the time it takes for half of any given quantity of potassium-40 to decay, is the fundamental constant underlying every argon-argon date ever published. The team now reports a value of 12.044 billion years, with an uncertainty of 0.088 billion years, twice as precise as the previous value derived from nuclear physics experiments. Because the decay constant links the measured argon ratios to elapsed time, improving it tightens every argon-argon age across the entire timescale the method covers, from recent volcanic eruptions to the formation of the planets.

The practical consequences extend well beyond Vesuvius. Renne noted that more accurate and precise argon-argon dating will be invaluable in calibrating radiocarbon dating, the dominant method for attaching dates to organic materials such as wood that are younger than about 55,000 years. It will also help ground-truth uranium-lead dating of billion-year-old rocks from the earliest days of planet Earth. Geologists studying active volcanoes stand to benefit directly as well, since reconstructing the eruptive history of a volcano that still threatens dense urban areas, such as those near Mexico City, Naples, or Yogyakarta in Indonesia, depends on precisely dating its past eruptions. As Renne put it, if you want to assemble the eruptive history of a volcano in relatively recent time, precision and accuracy really count, and the study shows that kind of useful precision can now be achieved into the historical realm.

The improved clock also sharpens scientists’ ability to connect cause and effect in the deep past. Renne pointed out that better precision allows researchers to more confidently infer causal relationships between events in the geologic record, such as a meteor impact structure and a mass extinction. A decade ago, he used argon-argon dating to establish precise dates for a meteor impact, massive volcanic eruptions in India, and the dinosaur extinction, all of which occurred within a few tens of thousands of years around 66 million years ago. Those dates bolstered the idea that the impact triggered intensified volcanism, delivering a one-two punch that wiped out all non-avian dinosaurs. Looking forward, Renne said the team hopes to unify as many geologic dating methods as possible using the same Bayesian mathematics, bringing more raw observations into a single coherent framework. But argon-argon dating, he emphasized, will always remain a standard and an important calibrant. The study also establishes a new benchmark for the method’s ability to date very recent eruptions with decadal accuracy, ensuring that the eruption that destroyed Pompeii will keep contributing to science for centuries to come. The work was funded by the National Science Foundation, the Ann and Gordon Getty Foundation, and the Berkeley Geochronology Center.

Subject of Research: Recalibration of the argon-argon geologic dating method using the historically dated 79 CE Vesuvius eruption

Article Title: The eruption that destroyed Pompeii is helping scientists date Earth's history

Article References: The eruption that destroyed Pompeii is helping scientists date Earth's history. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: argon-argon dating, Vesuvius, Pompeii, geochronology, potassium-40 half-life, sanidine, radiometric dating, Pliny the Younger, Science Advances, UC Berkeley, volcanic eruption dating, uranium-lead calibration

An international team led by Alejandro G. Sinner of the University of Victoria, together with colleagues from the University of Barcelona, the Catalan Institute of Classical Archaeology, and the University of Cincinnati, analyzed 57 ceramic individuals dated between the late fourth and early first centuries BCE. Crucially, the samples came not only from domestic consumption contexts but also from production areas within Pompeii itself, allowing the researchers to compare presumed local wasters and workshop debris with the fine tableware Pompeians actually used.

The team employed a three-pronged analytical battery. Wavelength-dispersive X-ray fluorescence, or WD-XRF, measured the elemental chemistry of each vessel, while powder X-ray diffraction, PXRD, identified the crystalline mineral phases preserved in the fired clay. Scanning electron microscopy coupled with energy-dispersive spectroscopy, SEM-EDS, then zoomed in on the famous black slip itself, revealing its microstructure and composition at the microscopic scale. Together these techniques provide chemical fingerprints that can distinguish clay sources and reconstruct ancient firing practices in ways that visual typology never could.

The chemical results revealed a polygenic assemblage, meaning the pottery found in Pompeii did not come from a single source. Instead, the material separated into several distinct compositional groups. One major group was interpreted as local or regional production associated with Pompeii or its immediate surroundings, lending quantitative weight to the idea that the city supported its own black gloss industry. Other groups corresponded to additional Campanian productions, including the celebrated Campanian A ware from the Bay of Naples, and one widely distributed group documented across the Italian peninsula and the western Mediterranean that shows striking compositional similarity to the Pompeian group. The data also identified a production compatible with materials linked to the Etruscan city of Chiusi in Tuscany, underscoring Pompeii’s connections to trade networks stretching well beyond Campania.

The mineralogical and microstructural analyses opened a window onto the potters’ kilns. Most of the productions used calcareous clays, rich in calcium compounds, and were fired at roughly 900 to 950 degrees Celsius, precisely the temperature window required to produce a durable, well-sintered ceramic body for this class of fine ware. By contrast, the low-calcareous Campanian A pottery had been fired at temperatures below or around 800 degrees Celsius, a distinctly different technological recipe that reflects the particular properties of the volcanic Bay of Naples clays used in that tradition.

The black slip itself, the defining feature of the ware, told an equally nuanced story. All of the slips proved to be iron-rich and illitic, consistent with the established Black Gloss manufacturing tradition in which a fine, iron-laden clay slurry is applied to the vessel and reduced in firing to produce the characteristic metallic black surface. But the study revealed considerable variability in slip thickness, vitrification, and adhesion, particularly within the Pompeian group. That variability reflects differences in technological control from workshop to workshop, and perhaps even from potter to potter, offering a rare glimpse of quality differences within a single production tradition.

Perhaps the most consequential finding concerns the mismatch between archaeological classification and chemical reality. Typological attributions, such as assigning a vessel to the famous Cales production on the basis of its shape and style, did not correspond to single production units in the compositional data. In other words, similar-looking vessels were being made in multiple places. The results therefore support a model of decentralized production in Campania, in which numerous workshops exploited similar calcareous clay resources while maintaining distinct compositional and technological identities. This picture replaces the older notion of a few dominant production centers flooding the market with a monolithic product.

The study forms part of the broader MedConTaCCt project, Mediterranean Connectivity: Economy, Trade and Commercial Circuits in the Roman West, funded by the Social Sciences and Humanities Research Council of Canada. As part of that project, nearly 600 Black Gloss samples have now been analyzed from consumption centers across the Italian peninsula, including Populonia, Pompeii and Gabii; from Sicilian sites such as Monte Iato and Morgantina; from southern French sites including Narbo, Loupian and Lattara; and from a string of Iberian ports and inland towns from Emporion to Valentia. The Pompeii dataset thus plugs into a vast comparative framework that allows individual vessels to be matched against reference groups spanning the entire western Mediterranean.

For archaeologists, the implications extend well beyond ceramics. Black Gloss pottery is one of the most common dating and diagnostic artifacts in Hellenistic and early Roman sites, and assumptions about its origin underpin reconstructions of trade routes, market systems and economic integration. Demonstrating that Pompeii participated actively in production, not merely consumption, positions the city as a node in regional supply networks decades before it became the famous Roman colony of the imperial era. The results also refine interpretations of how technological knowledge, particularly the demanding reduction-firing technique behind the black slip, spread among workshops that shared raw materials yet preserved their own manufacturing signatures.

The full dataset is being released in open access through the SMART database, the Server for Mediterranean Archaeometric Roman Tableware, ensuring that future researchers can test and extend these compositional groups as new samples accumulate. As the chemical atlas of Black Gloss pottery grows, the humble black-slippered cup on an ancient table is proving to be one of the most informative tracers of economic life in the Hellenistic and early Roman Mediterranean, and Pompeii, it turns out, was making its mark on that world long before the eruption preserved it in stone.

Subject of Research: Archaeometric provenance and technology analysis of Black Gloss pottery from Pompeii

Article Title: Made in Pompeii? New insights into the production, consumption and distribution of black gloss pottery

Article References: Made in Pompeii? New insights into the production, consumption and distribution of black gloss pottery. (n.d.). https://doi.org/10.1007/s12520-026-02542-1

Image Credits: AI Generated

DOI: 10.1007/s12520-026-02542-1

Keywords: Pompeii, Black Gloss pottery, archaeometry, WD-XRF, PXRD, SEM-EDS, Campanian A, ceramic technology, provenance studies, Hellenistic trade, decentralized production, Roman ceramics