Medicanes (MEDIterraean HurriCANES) are meteorological events with the potential to cause devastating floods, storm surges, and windstorms, often leading to significant disruption and casualties. During their mature phase, they exhibit phenomenological features typical of tropical cyclones, notably the warm core (WC), a warm area spanning the mid-to-upper troposphere. The critical need for rapid, automated identification of this feature motivates this work.

This study details the development of an automated WC identification system in the context of the ESA MEDICANES project (https://medicanes.isac.cnr.it/) that exploits passive microwave measurements from Low Earth Orbit (LEO) satellites, using four channels within the oxygen absorption band as input. The data corpus for this study was derived from approximately 30,000 satellite overpasses across three instruments from 2000–2020, encompassing 770 Mediterranean cyclones.

The base algorithm is a Convolutional Autoencoder-based Semi-Supervised Anomaly Detection (AE-SAD) model, trained primarily on "normal" (non-WC) atmospheric cases, complemented by a limited set of labelled “anomalies” (WC cases). The model’s efficacy is founded on reconstruction error: normal inputs are reconstructed with minimal error, whereas the distinctive features of WC anomalies are characterized by a significantly higher error, utilized as the anomaly score. Preliminary results suggest the AE-SAD system provides robust differentiation, with high reconstruction error scores reliably flagging WC events.

To rigorously validate the model's performance, accuracy, and overall utility, the study compares systematically the AE-SAD model against a comparative algorithm, quantified using standard metrics such as the F1-score and Area Under the Curve (AUC), implicitly prioritizing Recall, as both metrics reward models that correctly identify the minority class (the rare medicanes). This research is anticipated to provide a critical, automatic tool for the near-real-time identification of medicanes' WC, significantly improving lead time for hazard prediction and risk assessment in the Mediterranean region.

Methane (CH₄) monitoring is a global priority for climate mitigation strategies, as highlighted by initiatives such as the Global Methane Pledge. While satellite observations have advanced our ability to quantify methane emissions across scales, current products do not provide information on the vertical distribution of CH₄, a key parameter for identifying emission sources and understanding atmospheric processes. In this context, Physics-Informed Neural Networks (PINNs) offer new opportunities by embedding physical constraints into machine learning models, thereby enhancing both accuracy and efficiency compared to traditional retrieval methods. The PRIN-MVP (Methane Vertical Profiling) project develops a PINN-based approach to retrieve CH₄ vertical profiles from hyperspectral infrared observations. The methodology was trained and tested on about one million synthetic spectra simulated under clear-sky conditions with the σ-IASI/F2N radiative transfer model, supported by auxiliary atmospheric parameters. To optimize the information content, only the most sensitive spectral channels, identified through Averaging Kernel analysis, were retained. Both spectral data and atmospheric profiles were compressed using principal component analysis (PCA). The model was validated using real spectra relating to two case studies: the Mediterranean basin in spring 2025, during episodes of Saharan dust intrusion, and the sabotage of the Nord Stream gas pipeline in the Baltic Sea. Results demonstrate that the PINN-based approach accurately identifies anthropogenic methane emissions and consistently reconstructs vertical profiles of CH₄. This work highlights the potential of PINNs for regional-scale methane monitoring and contributes to the development of innovative tools for atmospheric greenhouse gas analysis. In the future, the model could be extended to CO₂.

Each austral spring, the Antarctic ozone hole develops and reaches its maximum extent between October and November, before disappearing in December as stratospheric temperatures rise. The seasonal warming inhibits the formation of Polar Stratospheric Clouds (PSCs), which provide the surfaces that catalyze the reactions responsible for ozone destruction. PSCs appear when temperatures fall below 195 K, enabling the condensation of nitric acid and water vapor into crystalline particles (mainly HNO₃·3H₂O or NAT). Ozone depletion is routinely monitored from space by UV–visible instruments such as OMI (the Ozone Monitoring Instrument) and TROPOMI (Tropospheric Monitoring Instrument). However, their dependence on reflected sunlight limits their capability during the polar night and under persistent cloudy conditions, typical of the Antarctic winter, when the interior of the continent remains largely unobserved. In addition, these sensors are insensitive to nitric acid and water vapor in the gas phase. Microwave limb sounders, such as MLS/AURA, provide complementary HNO₃ data but at coarse spatial resolution and without information on the thermodynamic state of the Upper Troposphere–Lower Stratosphere (UT/LS). Recent improvements in radiative transfer modeling have made it possible to retrieve temperature, ozone, and nitric acid simultaneously from IASI (Infrared Atmospheric Sounding Interferometer) infrared spectra, even in cloudy scenes. The model we developed is specifically designed to exploit IASI’s high spectral resolution, enabling the retrieval of key atmospheric parameters over Antarctica in both clear and cloudy conditions, thus extending observational coverage during the polar night. Analysis of IASI observations from 2021–2023 reveals a deeper and more extended ozone hole than shown by ECMWF analyses that assimilate OMI and TROPOMI data. The results indicate a clear relationship between decreasing HNO₃ concentrations and temperatures below 195 K, confirming the formation of NAT particles. Spatial patterns of HNO₃ retrieved from IASI closely match those from MLS/AURA, highlighting the robustness of our approach in capturing ozone–nitric acid interactions in the cold and cloud-covered Antarctic atmosphere.

Nell’ambito della convenzione tra il CNR-ISAC e il Dipartimento della Protezione Civile (DPC) è in fase di sviluppo un nuovo strumento per la validazione e il confronto continuo, in near real time (NRT), di diversi prodotti di precipitazione provenienti sia da piattaforme satellitari sia da misure, stime e osservazioni a terra. L’obiettivo principale è fornire un sistema integrato, automatizzato e scalabile in grado di monitorare e valutare in maniera costante l’affidabilità, la qualità e la coerenza dei diversi prodotti disponibili, con particolare enfasi sull’area italiana, dove la complessità orografica e climatica rende particolarmente interessante l’analisi proposta.

La metodologia adottata trae origine dall’esperienza consolidata nel progetto europeo H SAF di EUMETSAT ed è stata ampliata per includere un ventaglio di analisi a diverse scale temporali: dal singolo evento, all’analisi giornaliera, mensile, stagionale e di lungo periodo. Attualmente il sistema integra oltre 20 prodotti di precipitazione provenienti da sensori satellitari, reti pluviometriche e misure radar internazionali e consente di effettuare più di 60 confronti sull’intero globo.

I risultati ottenuti dimostrano le potenzialità dello strumento nell’evidenziare differenze sistematiche, complementarità e criticità tra prodotti, fornendo statistiche dettagliate sui singoli confronti (indici di accuratezza, bias, correlazioni, distribuzioni spaziali e temporali) e una visione d’insieme utile al monitoraggio e alla valutazione comparativa dei prodotti. Tale approccio fornisce al tempo stesso una base solida per attività di calibrazione e miglioramento dei prodotti di stima della precipitazione, oltre che per supportare decisioni operative e strategie di gestione del rischio idro-meteorologico.

L’obiettivo a medio termine è l’ampliamento progressivo del numero di prodotti gestiti e delle tipologie di confronti effettuati, così da costruire un quadro sempre più completo e robusto delle performance dei diversi prodotti di stima delle precipitazioni. L’architettura, concepita come flessibile ed estendibile, potrà essere ulteriormente potenziata con l’integrazione di nuovi dati, metodologie avanzate di confronto basate anche su tecniche di machine learning, e strumenti di visualizzazione interattiva atti a facilitare l’esplorazione e la disseminazione dei risultati verso comunità scientifica, enti istituzionali e utenti finali.