Aria, clima, elettrificazione, acque e biodiversità. 39473 articoli raccolti da fonti istituzionali e specializzate, classificati per area ambientale e linkati al porto di riferimento.
Maxi rissa con coltelli e due feriti: otto giovani indagati. Nel casolare abbandonato droga e armi RavennaToday
Dalla maxi rissa con due feriti si sviluppa un'inchiesta che porta a indagare otto giovani, a sequestrare un verso arsenale di coltelli e a scoprire un covo zeppo di droga. Questo il risultato dell'operazione dei Carabinieri a seguito della violenta rissa che era scoppiata nella serata del 22 giugno nel centro storico di Lugo.
La rissa e le indagini
Tutto era iniziato alle 23.45, nei pressi del monumento a Francesco Baracca, in una delle aree più frequentate della città. Qui si era sviluppata una furibonda lite che si sarebbe conclusa nel giro di pochi minuti. Secondo le testimonianze degli avventori dei locali limitrofi, una decina di giovani aveva iniziato ad affrontarsi violentemente, scatenando un fuggi-fuggi generale. Durante i tafferugli, oltre a calci e pugni, sarebbero spuntati anche dei coltelli. Alcuni testimoni hanno inoltre riferito di aver udito un rumore simile a un colpo d'arma da fuoco.
L'intervento dell'Arma è stato immediato. Una pattuglia della stazione dei Carabinieri di Lugo, già in transito nei paraggi, ha notato il trambusto e si è diretta tempestivamente verso la piazza. Contemporaneamente l’operatore del 112, monitorando in diretta le telecamere di videosorveglianza cittadina, ha rilevato lo scontro in atto e ha fatto convergere sul posto ulteriori due pattuglie. Al loro arrivo, i Carabinieri si sono trovati di fronte a una situazione di forte scompiglio, ma sono riusciti a fermare e identificare due degli individui coinvolti. Questi ultimi avevano avuto la peggio durante la colluttazione, riportando ferite da taglio successivamente curate dai sanitari del Pronto Soccorso di Lugo.
Le indagini, avviate nell'immediato dalla Compagnia Carabinieri di Lugo, si sono concentrate sull'acquisizione e sull'attenta analisi delle immagini fornite dai sistemi di videosorveglianza pubblica e privata a presidio della piazza e delle vie d'accesso. Questo lavoro ha permesso di identificare altri giovani che avrebbero preso parte allo scontro: si tratta di otto ragazzi di origini marocchine ed egiziane, di età compresa tra i 20 e i 29 anni.
Sulla base degli elementi raccolti, la Procura della Repubblica di Ravenna ha aperto un fascicolo nei confronti degli otto indagati per i reati di rissa aggravata, lesioni aggravate e porto di armi od oggetti atti ad offendere.
Dall'arsenale di coltelli al covo della droga
Ma le indagini non si sono fermate. All’alba di oggi i militari della Compagnia di Lugo, con il supporto dei colleghi della Compagnia di Cervia-Milano Marittima e del Nucleo Carabinieri Cinofili di Pesaro per la ricerca di armi e stupefacenti, hanno dato esecuzione a una serie di perquisizioni personali e domiciliari a carico degli indagati, su delega della Procura.
Nel corso delle operazioni di perquisizione sono stati rinvenuti e sequestrati numerosi armi da taglio (coltelli di grosse dimensioni e machete) e gli indumenti presumibilmente indossati dagli indagati nel corso della rissa. A Cervia, all’interno di un casolare abbandonato, ritrovo abituale di persone senza fissa dimora, uno degli otto sarebbe stato trovato in possesso di oltre un chilo di sostanza stupefacente (hashish e cocaina), in parte già confezionata in dosi, due bilancini di precisione e denaro contante ritenuto provento dell’attività di spaccio. Per l’occasione è stato fondamentale l’intervento delle unità cinofile che hanno individuato il luogo dell’occultamento. Il giovane è stato quindi anche arrestato per detenzione di sostanze stupefacenti finalizzata allo spaccio.
U.S.-headquartered technology player SLB has been tasked by Brunei Shell Petroleum (BSP), a joint venture company between the Brunei government and the UK-headquartered Shell, to support the resumption of oil and gas production across mature offshore fields in Brunei, Southeast Asia. The post SLB lends Shell a hand to unlock oil & gas from shut-in wells appeared first on Offshore Energy .
U.S.-headquartered technology player SLB has been tasked by Brunei Shell Petroleum (BSP), a joint venture company between the Brunei government and the UK-headquartered Shell, to support the resumption of oil and gas production across mature offshore fields in Brunei, Southeast Asia. SLB has been awarded a contract by Brunei Shell Petroleum to support production restoration from shut-in wells across multiple offshore fields, spanning subsurface evaluation, well candidate selection, engineering and offshore execution. The U.S. firm emphasizes that project management, intervention services, monitoring, metering, and marine logistics are integrated within a single coordinated execution model designed to support efficient production restoration. Gokhan Yarim, Senior Vice President of Integration at SLB, commented:“Building on our long collaboration in the region, this contract reflects a shared commitment to maximizing value from existing offshore resources. “Together with BSP, we are combining SLB’s production expertise and integrated execution capabilities to help restore production from shut-in wells, increase recovery from existing infrastructure and support BSP’s long-term production objectives.” According to SLB, the project marks the first deployment of an integrated production restoration solution for BSP, bringing together multiple disciplines, services and workflows to optimize recovery from mature offshore assets. The program aligns with standard well, reservoir, and facility management methodology through coordinated surveillance, engineering, and intervention planning, supporting a structured approach to optimizing production while maintaining safe and efficient operations across mature offshore assets. The U.S. player emphasized:“As operators increasingly prioritize maximizing recovery from existing assets, mature offshore fields present both opportunity and complexity. “This award reflects a broader industry shift toward integrated production and recovery solutions that combine technical expertise, coordinated execution and production management to safely and efficiently improve asset performance.” SLB recently won multiple assignments, including a deal with Vår Energi to help the operatoradvance projectsfrom discovery to development and reduce time to first oil on the Norwegian Continental Shelf (NCS). Take the spotlight and anchor your brand in the heart of the offshore world! Join us for a bigger impact and amplify your presence at the core hub of the offshore energy community!
NEC Secures Japan Defense Contract to Build First AI That Hears Submarines and Earthquakes Tech Times
Japan's defense research agency has awarded NEC Corporation a contract to build what would be the world's first general-purpose foundation model for underwater acoustics — an AI system designed to hear the ocean the way large language models read text. The project applies self-supervised learning, the same training technique that made ChatGPT and GPT-4 possible, to a domain that has resisted automation for a simple reason: collecting labeled training data underwater is expensive, slow, expert-dependent, and in military contexts classified by default. NEC'sofficial announcement on August 18, 2026confirms the contract and the commencement of research. If NEC's bet pays off, the implications run in two directions simultaneously. The Japanese Maritime Self-Defense Force gains AI-accelerated sonar analysis — faster submarine detection, better acoustic classification in contested waters, less dependence on the specialist analysts who are the current bottleneck. And ocean science gains something arguably more transformative: a general-purpose acoustic intelligence that could function as a planetary passive sensor network, improving earthquake prediction, marine biodiversity monitoring, and environmental surveillance of ocean health across a domain that covers more than 70% of Earth's surface. Light and radio waves do not travel through seawater in any useful way. At the distances that matter for navigation, surveillance, and scientific observation, sound is the only information carrier the ocean permits. Sonar and hydrophones — essentially underwater microphones — are therefore the workhorses of everything from submarine detection to earthquake monitoring to whale population surveys, asthe physics of underwater acousticsmake clear. The data those instruments produce is vast, unstructured, and almost entirely unlabeled. Analyzing it requires trained specialists who understand multipath propagation (sound bouncing off the seafloor, the surface, and internal thermal layers), Doppler shifts from moving targets, and the extreme variability of ambient noise across ocean basins and seasons. At 1 kHz, the wavelength of sound in water is about 1.5 m (4.9 ft) — creating a physical environment that behaves very differently from the air-acoustic world that most audio AI was built to handle. Sound travels through seawater at approximately 1,450 to 1,500 m/s (4,757 to 4,921 ft/s), roughly 4.4 times faster than through air, which means the geometry of multipath returns is denser and more complex than anything above-surface sonar AI typically encounters. Conventional deep learning needs large volumes of carefully labeled training data. In underwater acoustics, that label is almost never available. Collecting it requires sea operations; classifying it requires expertise; and in military contexts, the most valuable data — recordings of actual submarine acoustic signatures — is classified and kept off any shared training corpus. The result, as NEC's own announcement bluntly states, is thatno general-purpose acoustic models exist, with development historically confined to narrow AI tuned for single, specific tasks. The approach NEC is using — self-supervised learning (SSL) — is specifically designed to sidestep the labeled-data bottleneck. Rather than requiring human-annotated examples, SSL trains models to find structure in raw, unlabeled data, using the data itself as the supervisory signal. A model trained to reconstruct masked portions of an acoustic spectrogram, or to predict the next acoustic segment from the current one, learns rich representations of underwater sound without any human ever labeling a single recording as "submarine" or "whale." This is exactly the training logic that produced BERT, GPT, and the family of large language models that have dominated AI since 2018. The reason it worked in NLP is that the Internet provided billions of unlabeled text documents; the model learned language by predicting words from context. NEC is betting the ocean can supply the analogous corpus — vast quantities of raw acoustic recordings from sonar systems, hydrophone networks, and scientific monitoring stations — and that SSL can extract general acoustic intelligence from it the same way. Academic research published in the past two years supports the bet. A 2024 study in the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing demonstrated that anSSL framework combined with mixup augmentationachieved 86.33% accuracy on the DeepShip vessel classification dataset, outperforming supervised baselines specifically in few-shot and low-signal-to-noise conditions — precisely the conditions that define real-world passive sonar. Research published in the Journal of the Acoustical Society of America in 2023 confirmed that SSL via mask modeling learns effective acoustic representations from unlabeled data. The pattern across these studies is consistent: SSL reduces but does not eliminate the need for labeled examples, which is exactly what NEC needs for a domain where labeled examples are scarce rather than absent. Read more:Japan Stakes $6.2 Billion on Physical AI Factory Built on NVIDIA Vera Rubin In language, generality is measurable: a model that can answer questions, translate, summarize, and code has demonstrated breadth. The acoustic equivalent would require a single architecture to perform competently across submarine detection, marine mammal recognition, passive sonar localization, seismic event characterization, ship noise classification, and environmental monitoring — each task governed by different signal physics, different frequency bands, and different environmental conditions. That is an ambitious benchmark. Submarines produce acoustic signatures in the 10 Hz to several hundred Hz range, driven by propeller cavitation and machinery vibration. Blue and fin whales vocalize in overlapping bands. Underwater seismic events produce characteristic T-phase arrivals at hydrophones thousands of kilometers away. Shipping noise is broadband and dominant in most of the world's major ocean basins. A model that is genuinely general must learn to distinguish all of these, in variable noise conditions, across radically different ocean environments from Arctic to tropical. NEC's90-plus years of sonar development historyrepresent an important asset here. The company has accumulated institutional knowledge about what passive sonar data looks like across real operational conditions — the kind of knowledge that is hard to reconstruct from published literature alone. Combined with the large-scale pretraining infrastructure developed for the cotomi LLM platform, NEC is positioning itself as the organization that can actually assemble the training corpus and run the pretraining at the required scale. One reason the ocean is an unusually promising domain for foundation-model AI is a physical property called the SOFAR channel (Sound Fixing and Ranging channel). At roughly 750 to 1,200 m (2,460 to 3,940 ft) depth — where water pressure and temperature conspire to create a minimum sound speed — low-frequency sound is naturally ducted horizontally and can travel thousands of kilometers with minimal energy loss. Thephysics of underwater sound propagationexplain why the U.S. Navy exploited this channel to build the Cold War's Sound Surveillance System (SOSUS), a network of seafloor hydrophones that tracked Soviet submarines across entire ocean basins. An acoustic foundation model trained on SOFAR-band signals could, in principle, function as a passive planetary sensor. A seismic event on the seafloor produces acoustic signals that propagate through the SOFAR channel and can be detected at hydrophone arrays thousands of kilometers away; correlating these signals across a distributed network is currently a labor-intensive analysis task. A general-purpose AI that can interpret SOFAR-channel recordings would compress that analysis dramatically, potentially giving earthquake early warning systems and ocean monitoring networks a capability that no previous AI architecture has been able to provide. This is the core of what NEC calls adigital twin of the ocean— a real-time AI-maintained representation of the underwater environment built from acoustic sensing data, continuously updated as hydrophones around the world report in. Whether that ambition is achievable by the end of fiscal year 2027 is a reasonable question; it is more likely to emerge from successive refinements of a model that begins as defense-focused and is progressively extended to civilian applications. The awarding body, the Defense Innovation Science and Technology Institute, is Japan's explicit answer to DARPA. Established in October 2024 within the Acquisition, Technology and Logistics Agency (ATLA) of Japan's Ministry of Defense, DISTI operates with a dedicatedbudget of approximately $140 millionand a staff of roughly 100, including specialists in AI and robotics. The institute was specifically designed to recruit program managers from outside Japan's Ministry of Defense — importing the DARPA model of technologists with commercial-sector experience running high-risk research programs — and isauthorized to fund projectsthat could, in its own framing, "fundamentally strengthen defense capabilities or transform society." NEC's contract falls under DISTI's "Innovative Breakthrough Research" track, which sits alongside a "Breakthrough Empirical Research" track that earlier funded Sakana AI's multimodal decision-making AI for Japan's command-and-control system. This is not the first time DISTI has moved into AI-for-defense territory, and it will not be the last. Japan's overall FY2026 defense budget stood at ¥9.04 trillion (approximately $56.6 billion USD), arecord twelfth consecutive year highand the fourth year of the five-year Defense Buildup Program that commits roughly ¥43 trillion (approximately $269 billion USD) through FY2027. Within that budget, maritime AI and autonomous underwater systems are explicitly prioritized. The SHIELD coastal defense system, which deploys unmanned air, sea, and underwater drones in an integrated network, is scheduled for completion in 2028 and represents the operational context into which NEC's acoustic AI would eventually feed. Read more:Japan's Cyber Chief Confirms U.S. AI Export Ban Hit Government Security Scanning The labeled-data bottleneck NEC is trying to break is not unique to underwater acoustics. It is the defining challenge of every domain where the environment is expensive to instrument, expert annotation is rate-limited, and the most valuable data is operationally sensitive or simply never recorded in usable form. Seismic monitoring, atmospheric acoustics, subsurface geology, and deep-space radio astronomy all share the same fundamental constraint: vast unlabeled data, nearly no labeled examples, and a history of narrow task-specific models. The SSL approach NEC is applying to ocean acoustics has already succeeded in one such domain: protein biology. AlphaFold's foundational insight — that self-supervised pretraining on protein sequences could learn structural relationships that no human annotator could label at scale — is architecturally analogous to what NEC is attempting with acoustic signals. If the ocean acoustic foundation model works, it would establish a template for SSL in physical sensing domains more broadly, potentially catalyzing similar projects in seismic interpretation, subsurface resource exploration, and ocean-atmospheric gas-exchange monitoring. That generalization potential — not just the sonar application — is the architectural reason this announcement matters beyond Japan's defense establishment. The "Part 1" designation in NEC's contract title is consequential. It signals that further research phases are already anticipated, and with them, a series of decisions that neither NEC nor DISTI has yet made public: What portions of the training data will be shared with the scientific community? Will model weights be released, either openly or to selected research partners? Will the architecture itself be published in refereed journals? And how will the model be governed if it is used by allied navies — given that the training data will include recordings from Japan Maritime Self-Defense Force operations that are, by definition, classified? Those governance questions are not merely procedural. The scientific value of an ocean acoustic foundation model is directly proportional to how widely it can be used and how thoroughly it can be validated. A fully classified model, trained on classified data and used only within Japan's defense establishment, would advance military sonar capability without returning anything to the ocean science community that developed many of the underlying techniques. A partially open model — with published architecture and publicly shared pretraining data drawn from non-classified sources, and fine-tuning capability reserved for classified applications — would be more scientifically productive and would also serve NEC's commercial interests in the maritime and environmental monitoring markets. NEC's press release frames the project's ambitions in civilian terms as readily as military ones, suggesting the company has every reason to maximize the scientific spillover. Whether that intent survives the classification decisions that will accompany any genuinely useful military sonar model remains the open question that this project will have to answer before its conclusion in fiscal year 2027. Currency conversions are approximate, based on exchange rates at the time of publication. Current AI systems for sonar are narrow: each model is trained for a specific task, such as classifying vessel types or detecting a particular submarine signature, and performs poorly when deployed outside the conditions it was trained on. A foundation model works differently — it is first pretrained on enormous quantities of unlabeled acoustic data to learn general representations of underwater sound, then fine-tuned for specific tasks with relatively little additional labeled data. The same architecture that serves submarine detection could also serve whale monitoring or seismic event identification. This "train once, apply broadly" approach is what made large language models like GPT so versatile, and NEC is betting the same logic applies to ocean sound. Labeling underwater acoustic data requires trained specialists, expensive sea operations to collect the recordings, and — in military contexts — security clearances to access the most valuable examples. Self-supervised learning sidesteps labeling entirely by training a model to predict missing portions of its own input data. A model that learns to reconstruct a masked acoustic spectrogram from its surrounding context is learning something real about how sounds in water are structured, without any human ever attaching a label. Once that general acoustic understanding is learned, a small number of labeled examples is often enough to fine-tune the model for a specific application. NEC specifically names earthquake prediction, marine life exploration, resource exploration, and environmental monitoring. The ocean's acoustic environment carries signals from seafloor seismic events, volcanic eruptions, whale migrations, shipping traffic, and even ocean temperature changes (which alter sound propagation speed). A general-purpose acoustic AI trained on hydrophone data from around the world could potentially provide real-time situational awareness of all of these simultaneously — what NEC calls a "digital twin of the ocean" — giving scientists a window into the deep ocean that no existing system provides. The labeled-data bottleneck that makes underwater acoustic AI so difficult is not unique to the ocean. It is the same fundamental challenge in seismic monitoring, subsurface geology, atmospheric acoustics, and other physical sensing domains where the environment is expensive to instrument and the most scientifically valuable data is nearly impossible to annotate at scale. If self-supervised learning succeeds in ocean acoustics — a domain with some of the worst data conditions imaginable — it would establish a template for applying the same approach across the full range of physical sensing environments where narrow task-specific AI has historically failed to generalize. ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
Picco vicino a 6mila abbonati, Spezia-Torres sfiora la top ten di Coppa Italia con un solo ospite Città della Spezia
Temi del giorno: Andrea Bonatti Più spettatori di campi di serie A tra cui Udinese-Padova, Parma-Catania, Frosinone-Juve Stabia e Monza-Avellino. I tifosi dello Spezia sono stati i più presenti di tutta la serie C nel primo turno di Coppa Italia di Lega Pro. Dei 4.718 spettatori che hanno sfidato il caldo afoso della domenica post Ferragosto per Spezia-Torres, uno solo risultava in arrivo dalla Sardegna. Non accomodato sulle roventi sedute del settore ospiti, ma per suo sollievo ospitato in tribuna ed entrato con un accredito richiesto dal club sassarese. Il Picco vestito di bianco batte anche le presenze di casa nel derby Salernitana-Scafatese, seguita dal vivo da 4.995 spettatori di cui 493 in arrivo dal comune al confine con Napoli. A seguire Pescara-Sambenedettese (4.174, senza ospiti) e via via tutte le altre fino ai 150 paganti di Vado-Inter U23 e i 110 di Juventus U23-Forlì. Allargando lo sguardo alla Coppa Italia tutta, Spezia-Torres sarebbe l’undicesima partita. Con più spettatori di campi di serie A tra cui Udinese-Padova, Parma-Catania, Frosinone-Juve Stabia e Monza-Avellino. E’ insomma un inizio stagione all’insegna dell’entusiasmo nell’ambiente aquilotto, con un effetto più da promozione che da retrocessione. Dato confermato dalla campagna abbonamenti che, a metà mattina di martedì 18 agosto, conta 5.828 tessere staccate. I seimila abbonati sono un obiettivo concreto. Chiusura della fase libera venerdì 21 agosto a mezzanotte.
Singapore-based global shipping company Eastern Pacific Shipping (EPS) announced the successful completion of its 700th LNG bunkering operation for the vessel Lake Victoria in the German port of Emden. IMAGE: Illustration of an LNG dual-fuel crude oil tanker. ENGINE via ChatGPT The company said its ships have consumed a total of 2.92 million cbm (1.31 million mt) of LNG since November 2020. EPS said in itsESG reportthat itmanagesa fleet of over 350 vessels, of which 179 are listed as dual-fuel, without specifying whether they are dual-fuel-capable or dual-fuel-ready. Of those 179, 92 are in operation. LNG remains the dominant fuel choice for shipping companies investing in alternative fuel technologies. There are around 929 LNG dual fuel vessels currently in operation globally, with a further 672 on order through 2033, according to DNV’s data. LNG is not the company's only alternative-fuel bet. EPS said alternative fuels, including LNG, LPG, ethane and biofuel, made up around a quarter of its total fuel consumption in 2025, cutting roughly 465,000 mtCO2e against conventional marine fuels. Additionally, EPS's firstammonia-fuelled vessel, a very large ammonia carrier, is under construction at HD Hyundai Heavy Industries and is scheduled for delivery from October 2026, marine engine maker Everllence said in an update earlier this year. By Nachiket Tekawade Please get in touch with comments or additional info to news@engine.online
Costa Rica is turning pineapple and coffee waste into syngas, then burying the carbon-rich material left behind instead of sending it back into the atmosphere Energies Media
Norwegian energy data and intelligence company TGS and U.S.-headquartered energy firm Chevron have established […] The post TGS and Chevron jointly pursuing next-gen seismic imaging technology appeared first on Offshore Energy .
Un repunte de grandes terremotos en Colombia, Venezuela, México, Indonesia y Granada ha desatado la alarma sísmica en 2026. Sin embargo, la ciencia confirma que no hay efecto dominó: la Tierra sigue su curso geológico habitual
The Urbanoid Booba is a 530-lb clamshell trailer that a Fiat 500 can tow, inflates into a two-person tent camper in minutes, and stores standing upright in a normal garage.