Alberto Carrassi

A reanalysis can be considered the most complete representation of the atmospheric state given all available observations as well as the model background. Via data assimilation, a reanalysis allows assessing the non-local influence of observations on the time evolution of the modeled state, while aiming at conserving the physicality of the model. Here, we present the pioneering, 3-months convective-scale campaign reanalysis of the 'Swabian MOSES 2023' campaign, consisting of a control (CTRL) dataset without and a campaign (CMPG) dataset with additional field campaign observations from the mobile atmospheric measurement platform KITcube during June, July and August 2023. Both are computed at hourly resolution on a 2.2 km grid centered over Germany with the quasi-operational data assimilation framework by Deutscher Wetterdienst which is based on a Local Ensemble Transform Kalman Filter. CMPG incorporates additional ground-based remote-sensing and in situ observations from the southwest German mountain ranges of the Black Forest and Swabian Jura, a region with moderately complex terrain. The unprecedentedly dense network of 12 Doppler wind lidars, one X-band precipitation radar, two additional radiosounding sites, Global Navigation Satellite Systems receivers, and meteorological masts increases the number of observations in the approximately 105 km2 campaign area by 45 % compared to the operational measurement network. As expected, the assimilation reduces the difference between observations and model fields. Specifically, the standard deviation of the observation-to-model difference across all additional observations is reduced by 10 % to 35 % after assimilation. Updates of the model background are predominantly introduced in the lower troposphere and in the campaign region itself, but differences between CMPG and CTRL extend all over Germany, eastern France, and northern Italy. Mean absolute differences in the campaign region reach up to the order of 1.5 m s−1 in windspeed, 0.4 K in temperature, and 0.7 g kg−1 in humidity. This pioneering campaign reanalysis can be employed to quantify the influence of individual observations and different observation types and to diagnose potential model deficiencies in complex terrain. It further serves as a high-resolution reference for detailed case studies targeting high-impact weather and provides a basis for re-forecast experiments to quantify observation impact on forecasts.

Weather and climate shape every aspect of our lives, yet for decades they have been studied as separate worlds—one concerned with what happens tomorrow, the other with what awaits us over the coming decades. This long-standing divide has slowed progress in both fields. Here, we argue that artificial intelligence (AI) is helping to dissolve the boundary between them. By learning directly from vast streams of Earth observations, AI is starting to predict both tomorrow’s storm and next century’s climate within a single, unified framework. We show how this convergence is already underway, why weather and climate science have much to gain from each other, and what still stands in the way—from trust, transparency, and fairness to the energy cost of computation and the risk of widening global inequalities. Realizing this vision will demand not only better algorithms but a new culture of collaboration across scientific communities, operational centers, fields and disciplines. Done well, a unified, AI-driven Earth science could deliver faster, more reliable, and more actionable predictions—helping societies anticipate and adapt to a rapidly warming planet.

Accurate representation of atmospheric dynamics at convection scale remains a major challenge for numerical models and a critical aspect in operational weather predictions. In this work, the ICOsahedral Non-hydrostatic (ICON) model is run at convection-permitting scale over the Italian domain, in combination with the Local Ensemble Transform Kalman Filter (LETKF), following the operational configuration of Arpae Emilia-Romagna and ItaliaMeteo Agency. We focus on a poorly-predicted extreme convective storm in the Marche region, Italy, highlighting the crucial role of low-level moisture convergence in convection initiation and the significant undersampling of humidity in conventional data. To address this, we investigate the added value of humidity-sensitive microwave radiances from polar satellites. Assimilation of clear-sky observations from the Microwave Humidity Sounder (MHS) leads to notable improvements in precipitation forecasts compared to the current operational setup, based on conventional and radar observations only. Infrared all-sky radiances in water vapor channels from the geostationary Meteosat Second Generation SEVIRI instrument are further integrated, providing higher spatial and temporal resolution but limited cloud penetration capability. The joint assimilation of microwave and infrared satellite channels leads to improvements in both surface and upper-level variables, supporting the future operational assimilation of satellite radiances at convective-scale in the Arpae and ItaliaMeteo system. The relative contribution of each observation type is evaluated through an updated version of the Partial Analysis Increments algorithm (Diefenbach et al., 2023), which is corrected to account for posterior covariance inflation and localization.Building on the promising results of this work, we present ongoing developments towards ensemble-based data assimilation in a Machine Learning Limited Area Model (ML-LAM). Machine learning weather prediction models have demonstrated competitive forecast skill at coarse resolution, but convective-scale ML-LAMs remain much less explored. A recent study (Adamov et al., 2025) presents the development of a high-resolution ML-LAM, trained on a convective-scale analyses dataset over Switzerland. The potential reproducibility of ML-LAM across different regions makes it appealing for application in other LAM settings. We show here the implementation of ML-LAM over the Italian domain, evaluating the skill against the physics-based model in severe convection conditions. This serves as a first step towards the assimilation of satellite microwave radiances in a operational-like setting using an LETKF scheme. Diefenbach, T., Craig, G., Keil, C., Scheck, L. and Weissmann, M., QJRMS, 149(752), 740–756, (2023). Adamov, S., Oskarsson, J., Denby, L., Landelius, T., Hintz, K., Christiansen, S., Schicker, I., Osuna, C., Lindsten, F., Fuhrer, O. and Schemm,S., arXiv, (2025).

The development of summertime convection is sensitive to mesoscale atmospheric conditions, which, in particular in complex terrain, are not well captured by conventional observation systems. Here, we investigate whether the assimilation of substantial additional observations of lower-tropospheric thermodynamic and dynamic variables at high resolution has the potential to improve the resulting analysis used for forecast initialization, and ultimately the predictability of convective events.The high-resolution observations used here have been obtained during the ‘Swabian MOSES 2023’ campaign, conducted in June, July and August 2023 in the southwest German mountain ranges, a region that is particularly prone to hailstorms. A key component was an unprecedented network of 12 Doppler wind lidars (DWL), a configuration never used before in data assimilation experiments. We adopted the regional forecasting system of Deutscher Wetterdienst, the ICON model at 2 km resolution (ICON-D2) coupled to the Kilometer Scale Ensemble Data Assimilation system (KENDA) based on the Local Ensemble Transform Kalman Filter (LETKF) with an hourly assimilation step to produce a campaign and a control reanalysis. The campaign reanalysis uses, in addition to all operationally available observations used in the control reanalysis, (i) DWL-retrieved vertical profiles of the horizontal wind components, (ii) X-band radar reflectivity, (iii) targeted radiosoundings from two sites during intensive observation periods, (iv) ground-based zenith path-delay data from a domain-wide Global Navigation Satellite Systems receiver network, and (v) 2m-temperature and relative humidity from meteorological masts at the campaign sites.We show that the additional campaign observations have a substantial influence on the reanalysis. We focus specifically on the mesoscale flow in the Black Forest region where the dense wind lidar observations lead to systematic differences in wind speed and direction between the campaign and the reference analysis. We discuss possible sources of these differences by further stratifying the analysis increments by time of day, different flow regimes, and stability. Furthermore, we investigate the impact of additionally assimilated observations on reforecasts by comparing forecasts initialized from the campaign and control reanalysis. Our results motivate future reanalyses projects, such as a high-resolution reanalysis for the ‘Multi-scale transport and exchange processes in the atmosphere over mountains – programme and experiment’ (TEAMx) summer extensive observation period.

This study explores a physics–data‐driven hybrid approach for sea‐ice column physics models, in which a machine‐learning component acts as a state‐dependent parametrization of forecast errors. We examine how perturbations in snow thermodynamics and sea‐ice radiative properties affect forecast errors, and train dedicated neural networks (NNs) for each model configuration. The performance of the hybrid models is evaluated for long lead‐time forecasts and compared against a benchmark system based on climatological forecast‐error estimates. The NN‐based hybrids prove to be stable, robust to initial condition and atmospheric forcing errors, and consistently outperform their climatology‐based counterpart. To derive guiding principles for efficiently handling possible physical model updates, we perform transfer learning experiments to test whether pretrained NNs optimized for one model configuration can be successfully adapted to another. Results indicate that direct evaluation of pretrained networks on the target task provides useful insights into their adaptability, recommending transfer learning whenever performance exceeds a trivial baseline. Finally, a feature‐importance analysis shows that atmospheric forcing inputs have negligible influence on NN predictive skill, and ice‐layer enthalpies play a key role in achieving satisfactory performance.

Radar data assimilation (DA) is critical for convective-scale forecasting, as it provides real-time, high-resolution information on precipitation, wind, and convective system dynamics that is not captured by surface observations or satellite data. Polarimetric radar observations complement conventional reflectivity (Zh) and radial velocity (Vr) measurements by enabling the determination of hydrometeor types and particle size characteristics. Differential reflectivity (ZDR) is one of the key polarimetric radar variables, defined as the ratio between horizontal and vertical reflectivity, that provides direct information on hydrometeor shape and size. Despite its strong potential to better constrain storm microphysics and improve convective-scale forecasts, the assimilation of ZDR remains challenging. Challenges associated with observation operators, error characterisation, and data quality underscore the need for further research in this area.The current study investigates the direct assimilation of ZDR in an observing system simulation experiment (OSSE) of a long-lived supercell. The OSSE is conducted using the ICOsahedral Nonhydrostatic (ICON) model with a two-moment microphysics scheme and the Local Ensemble Transform Kalman Filter (LETKF), employing both hydrometeor mixing ratios and number concentrations as analysis variables. Synthetic observations are generated using the polarimetric radar forward operator EMVORADO developed at the Deutscher Wetterdienst. The synthetic ZDR observations are assimilated in addition to the non-polarimetric variables, namely Zh and Vr, while a reference experiment assimilating only non-polarimetric synthetic observations was conducted for comparison. A series of sensitivity experiments are performed to assess the impact of DA settings on assimilation performance, for varying observation error, localisation radius, and ensemble size. In addition, appropriate thresholds and no-reflectivity (clear air) equivalents for ZDR observations are examined.

Intermittent dynamics are a common feature of many Earth-system components that often interact across ample ranges of temporal and spatial scales. Our previous work shed light on the mechanism driving intermittency and identified precursors of its onset (Barone et al., 2025). This current study moves forward, and it investigates the processes by which an intermittent component in a coupled system influences other ones that, in the absence of the coupling, would evolve quasi stationarily. In particular, we investigate a prototypical fast–slow, e.g. atmosphere-ocean, setup in which fast intermittent systems act as a unidirectional forcing on slow components characterized by a stable limit cycle.Using a two-scale version of the Lorenz–63 model, we show that intermittent bursts in the fast dynamics induce deviations from the slow dynamics’s limit cycle, which, depending on the strength of the coupling and the timescale difference, can even fully destabilize the limit cycle and lead to a chaotic regime. We show that increasing the frequency of intermittent events does not necessarily affect the slow component response, which below a critical value retains its structural properties, highlighting the non-trivial nature of intermittent information transfer across scales. The induced transition from periodicity to chaos caused by the intermittent burst, is looked through the lens of the power spectrum decomposition (PSD) of the finite-time Lyapunov exponents, offering a unique view on the progressive loss of predictability in the slow component. The analysis is then extended to a spatially extended system based on unidirectionally coupled Kuramoto–Sivashinsky equations. As the coupling strength increases, the energy PSD of the slow and initially regular dynamics, progressively approaches that of the fast intermittent system, up to a regime in which the two become effectively indistinguishable. Remarkably, mutual information between subsystems reveals a clear latency in the slow response that increases with the degree of time-scale separation.Our study provides a robust framework to investigate similar dynamical configurations in Earth system models, whereby a fast intermittent atmosphere induces short-living, yet impactful, changes in a slow ocean. A. Barone, A. Carrassi, T. Savary, J. Demaeyer, S. Vannitsem; Structural origins and real-time predictors of intermittency. Chaos 1 October 2025; 35 (10): 103119. https://doi.org/10.1063/5.0287572

Marine biogeochemistry models are critical for forecasting, as well as estimating ecosystem responses to climate change and human activities. Data assimilation (DA) improves predictions from these models by aligning them with real-world observations, but marine biogeochemistry DA faces challenges due to model complexity, non-linearity, and sparse, uncertain observations. Existing DA methods applied to marine biogeochemistry struggle to update unobserved variables effectively, while ensemble-based methods are computationally too expensive for high-complexity marine biogeochemistry models. This study demonstrates how machine learning (ML) can improve marine biogeochemistry DA by learning statistical relationships between observed and unobserved variables. We integrate ML-driven balancing schemes into a 1D prototype of a system used to forecast marine biogeochemistry in the North-West European Shelf seas. ML is applied to estimate (i) state-dependent correlations from free-run ensembles and (ii), in an “end-to-end” fashion, analysis increments from an Ensemble Kalman Filter. Our results show that ML improves updates for previously not-updated variables when compared to univariate schemes akin to those used operationally, particularly in lead times smaller than 5 d. Furthermore, ML models exhibit some potential for transferability to new locations, a crucial step toward scaling these methods to 3D operational systems. We conclude that ML offers a clear pathway to overcome current computational bottlenecks in marine biogeochemistry DA and that refining transferability, optimising training data sampling, and evaluating scalability for large-scale marine forecasting, should be future research priorities.

Identifying the optimal strategy for initializing coupled climate prediction systems is challenging due to the spatio-temporal scale separation and disparities in the observational network. We aim to clarify when strongly coupled data assimilation (SCDA) is preferable to weakly coupled data assimilation (WCDA). We use a two-components coupled Lorenz-63 system, mimicking the atmosphere and the ocean, and the Ensemble Kalman Filter (EnKF) to compare WCDA and SCDA for diverse spatio-temporal scale separations and observational networks – only in the atmosphere, the ocean, or both components. In the fully observed scenario, SCDA and WCDA yield similar performances. However, little differences are present, and we conjecture these are due to the SCDA being more sensitive to the approximations at the basis of the EnKF present in the cross-update – linear analysis update and sampling error, and how they impact the cross-update between ocean and atmosphere. This sensitivity increases as the temporal scale separation increases and is stronger on the slow and large-scale components. When observations are only in one of the components, the spatio-temporal scale separation influences SCDA's performance. In this scenario, the largest improvements are found when the observed component has a smaller spatial scale. The fast-to-slow update has a larger benefit with a larger temporal scale separation. Meanwhile, with the slow-to-fast update, the improvement is limited to instances when the temporal scale separation is less than one-half. This suggests that SCDA of fast atmospheric observations can potentially improve the large and slow ocean component. Conversely, observations of the fine ocean can improve the large atmosphere at a comparable temporal scale. However, when both components are highly chaotic, and the observed component's spatial scale is the largest, SCDA does not improve over WCDA. In such a case, the cross-updates may become too sensitive to data assimilation approximations. We further validated that WCDA systematically outperforms uncoupled data assimilation (UCDA) in both components, legitimizing the transition toward WCDA.