Biases are often associated either to the presence of model structural errors or to a misrepresentation of the properties of initial condition errors (initial error biases or a bad representation of the initial error distribution). In the current work, the development of biases is addressed by considering a twin experiment in which the dominant initial condition uncertainties are imposed to the external forcing of a coupled ocean-atmosphere extratropical system in a perfectly controlled environment. The forcing is generated by a low-order 3-variable tropical model mimicking the dynamic of ENSO. No structural model errors are introduced and the statistical properties of the initial error are perfectly known. It is shown that even if this almost perfect setting, important biases are induced on seasonal-to-decadal forecasts, and hence unreliable (under-dispersive) ensembles.More specifically, three main types of ensemble forecast experiments are performed: with random perturbations along the three Lyapunov vectors of the tropical model; along the two dominant Lyapunov vectors; and along the first Lyapunov vector only. When perturbations are introduced along all vectors, important forecasting biases, inducing a mismatch between the error of the ensemble mean and the error spread, are produced. Theses biases are considerably reduced only when the perturbations are introduced along the dominant Lyapunov vector. Hence, perturbing along the dominant instabilities allows a reduced mean square error to be obtained at long lead times of a few years, as well as reliable ensemble forecasts across the whole time range. These very counterintuitive findings, reported in Vannitsem and Duan (2026), further underline the importance of appropriately controlling the initial condition error properties in the tropical components of models.ReferenceVannitsem, S., Duan, W. A Note on the Role of the Initial Error Structure in the Tropics on the Seasonal-to-Decadal Forecasting Skill in the Extratropics. Adv. Atmos. Sci. 43, 157–169 (2026). https://doi.org/10.1007/s00376-025-4521-7

The variability in ensemble forecasts can either be generated dynamically - as is usually done with Numerical Weather Prediction (NWP) models -, stochastically or by using new approaches such as AI generative techniques. As these approaches are in their infancy for geophysical applications, the properties of the ensembles of generative models are still far from clear, especially if those models are to be used in operational activities. This aspect is investigated here for nowcasting models.This work provides a predictability analysis over Belgium for the generative AI nowcasting model LDCast [1], as well as for the stochastic STEPS nowcasting algorithm (pysteps implementation [2]). Both models correctly estimate the error at almost all scales by means of their ensemble spread (i.e. good spread/error relationship), and they adapt the morphology of their ensembles depending on whether the event dynamics is convective or stratiform. Surrogate ensembles are also derived from the ensembles of STEPS and LDCast, and used as benchmarks with which to compare the spatial scores of the models. This reveals that both STEPS and LDCast ensembles struggle to provide added value for the spatial localization of the uncertainty associated with the growth and decay of rainfall. Therefore, STEPS and LDCast ensembles seem to be accurate statistically but not dynamically.[1] Leinonen, J., et al. (2023). Latent diffusion models for generative precipitation nowcasting with accurate uncertainty quantification. arXiv preprint arXiv:2304.12891.[2] Pulkkinen, S., et al. (2019). Pysteps: an open-source python library for probabilistic precipitation nowcasting (v1.0). GMD, 12(10):4185–4219.

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

The atmospheric tropical belt is believed to be more predictable than the extratropics. This question is revisited here by exploring the emergence of chaos in reduced-order model versions of the vorticity equation under the weak temperature gradient hypothesis, which provides a good description of the large-scale tropical atmosphere. The analysis reveals that under fairly realistic divergence forcing amplitudes, chaos may emerge, sometimes with Lyapunov time scales of less than a day. This result contrasts with the idea of a predictable tropical atmosphere, and opens important questions on the effective origin of predictability in the Tropics.

The predictability of the generative AI-based nowcasting model LDCast (trained on another region) is evaluated over Belgium, together with the pysteps implementation of the nowcasting algorithm STEPS. STEPS and LDCast are slightly underdispersive, but the ensemble spread provides an estimation of the error at almost all scales. Both models adapt the properties of their ensembles to the type of event, either convective or stratiform. The spatial scores of the STEPS and LDCast ensembles are compared with those of surrogate ensembles having some key properties, revealing that both STEPS and LDCast have very little ability to spatially localise the ensemble mean error vector through their ensemble members. This suggests that the content of STEPS and LDCast ensembles is informative in terms of statistics, but not in terms of dynamics.

Baroclinic instability is a fundamental mechanism driving atmospheric dynamics. In this work, we revisit Pedlosky's two-layer model for finite amplitude baroclinic waves - a seminal framework for studying the unstable growth of finite perturbations - leveraging modern nonlinear techniques and computational resources. We show that the geophysical state of the baroclinic wave exhibits a rich diversity of dynamical regimes governed by the level of dissipation induced by Ekman boundary layers. In the inviscid limit, we demonstrate that the model is integrable. Upon increasing dissipation, the system undergoes a complex sequence of bifurcations. On one hand, deterministic chaos, identified by means of the Lyapunov exponents, provides a genuine mechanism for destabilization of the wave. On the other hand, in regimes where the wave equilibrates, dependence on the initial condition is crucial, eventually leading to the coexistence of multiple attractors. We study the governing equations of the model and their truncation to a finite-dimensional system of ordinary differential equations, together with the minimal low-order truncated system which is structurally equivalent to the Lorenz model. Its bifurcation diagram allows for elucidating the transition of the wave amplitude from stable equilibration to periodic oscillations - terminating in homoclinic orbits - and, ultimately, deterministic chaos through a period-doubling route. We finally comment on the robustness of these features for higher-dimensional models.

Abstract. The Liang–Kleeman Information Flow (LKIF) framework has been increasingly used to identify causality in Earth–system dynamics in recent years. Yet, uncertainties remain regarding how the causal graph changes when mediators and confounders (conditioners) are introduced, i.e., divergence between the bivariate and conditioned forms. A controlled synthetic experiment shows that LKIF preserves correct causal directions under hidden confounding, while differences between bivariate and multivariate estimates still occur. We characterize these differences using time–varying LKIF (ΔIF) and introduce several quantifiers, including Mediator Dominance Index, Moderation Gain, Confounding Pressure, and Convergence Rate, to evaluate the contributions of conditioners. This is illustrated in an application to soil moisture (SM) and vapor pressure deficit (VPD) couplings with vegetation (Leaf Area Index (LAI) and Gross Primary Productivity (GPP)). We find that the influence of SM on LAI remains consistently direct and stable, whereas its influence on GPP and that of VPD on vegetation exhibit substantial divergence between bivariate and multivariate estimates. This divergence peaks under compound dry–hot conditions, with VPD and temperature emerging as dominant mediators. These results demonstrate that bivariate and multivariate LKIF can yield markedly different conclusions across time and space, especially under compound environmental stress. This highlights the need for multivariate conditioning when assessing vegetation responses.

From 8 to 10 December 2025, the EUMETNET - E-TREND Joint Workshop on Postprocessing, Forecasting and Nowcasting for Renewable Energy, also known as RenEUCast25, was held at Vrije Universiteit Brussel (VUB). RenEUCast25 was meant as a dedicated event for the exchange of knowledge focused on effective forecasting and postprocessing for the renewable energy sector, and to promote a better understanding of how end users integrate forecasts and current developments in their decision-making processes. Organized by the EUMETNET Postprocessing Module, the Belspo E-TREND consortium (led by the Royal Meteorological Institute of Belgium), the EUMETNET Nowcasting module (led by GeoSphere Austria), and VUB’s ETRO.RDI group, it gathered more than 50 participants from both the weather forecasting community and the renewable energy sector. 30 contributions were presented, organized in six different sessions, and including five keynotes. In addition, the event was streamed online with more than 20 additional participants.

Atmospheric blocking exerts a profound influence on midlatitude circulation, yet its predictability remains elusive, due to intrinsic nonlinearities and sensitivity to initial conditions. While blocking dynamics have been extensively studied, the impact of geographical positioning on predictability remains largely unexplored. This study provides a comparative assessment of the predictability of western and eastern North Pacific blocking events, leveraging analogue-based diagnostics applied to Coupled Model Intercomparison Project Phase 6 (CMIP6) Model for Interdisciplinary Research on Climate, version 6 (MIROC6) simulations. Blocking structures are identified using geopotential height gradient reversal, with their temporal evolution analyzed through trajectory tracking and error growth metrics. Results reveal that eastern blocks exhibit lower predictability, characterized by rapid error divergence and increased mean logarithmic growth rates compared with western blocks. Persistence analysis gives no significant difference between eastern and western North Pacific blocking events. Sensitivity analyses across varying detection thresholds validate the robustness of these findings.

Unstable Periodic Orbits (UPOs) were used to identify regimes, and transitions between regimes, in a reduced-order coupled atmosphere-land spectral model. In this paper, we describe how the chaotic attractor of this model was clustered using the numerically derived set of UPOs. Using continuation software, the origin of these clusters was also investigated. The flow of model trajectories can be approximated using UPOs, a concept known as shadowing. Here, we extend that idea to look at the number of times a UPO shadows a model trajectory over a fixed time period, which we call cumulative shadowing. This concept was used to identify sets of UPOs that describe different life cycles of each cluster. The different regions of the attractor that were identified in the current work, and the transitions between these regions, are linked to specific atmospheric features known as atmospheric blocks.

Low-frequency variability (LFV) encompasses atmospheric and climate processes on time scales from a few weeks to decades.​ This includes atmospheric blockings, heat waves, cold spells, and at longer time scales long-term oscillations like the MJO, the NAO, ENSO….. Better understanding of LFV, could contribute to improved long term forecasts​. In the results described in Xavier et al, 2023, who used a reduced order atmosphere-land model (Demaeyer et al, 2020), weather patterns that involve atmospheric blocking to the west of a given topographical feature tend to have reduced predictability and show instability when contrasted with blocking occurrences situated to the east of such topographical elements. This finding aligns with actual meteorological occurrences, such as the persistence of North Pacific blocking patterns (Breeden et al., 2020; Kim and Kim, 2019). The shape and characteristics of the identified blocking events closely resemble North Pacific blocks, where a high-pressure system exists either on the western or eastern side of the underlying topography. In the physical world, these positions correspond to Asian and American continents on either side of the Pacific. Despite quasi-geostrophic models being overly simplified, using such reduced order models in this study allowed us to undertake such mathematical analysis. Thus allowing for a comparison with the real world…In the current study, we aim to analyze these predictability differences based on the morphology of the blocking situations in the real-world scenario using the CMIP6 dataset. Blocking situations are identified using two different indices, the anomaly-based blocking index proposed by Sausen et al., 1995 and the local reversal of the meridional flow-based index proposed by Davini et al., 2012. Predictability is quantified using local dimension metrics and analogue studies in the identified blocking events. The findings are discussed from the perspective of the current literature on the predictability of blocking.ReferencesXavier, A. K., Demaeyer, J., and Vannitsem, S.: Variability and Predictability of a reduced-order land-atmosphere coupled model, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2023-2257, 2023.Demaeyer, Jonathan & De Cruz, Lesley & Vannitsem, S.: qgs: A flexible Python framework of reduced-order multiscale climate models. Journal of Open Source Software. 5. 2597. 10.21105/joss.02597, 2020.Breeden, M. L., Hoover, B. T., Newman, M., and Vimont, D. J.: Optimal North Pacific blocking precursors and their deterministic subseasonal evolution during boreal winter, Monthly Weather Review, 148, 739–761, 2020Kim, S.-H. and Kim, B.-M.: In search of winter blocking in the western North Pacific Ocean, Geophysical Research Letters, 46, 9271–9280, 2019.Sausen, R., König, W., & Sielmann, F. (1995). Analysis of blocking events from observations and ECHAM model simulations. Tellus A, 47(4), 421–438.Davini, P., Cagnazzo, C., Gualdi, S., & Navarra, A. (2012). Bidimensional diagnostics, variability, and trends of Northern Hemisphere blocking. Journal of Climate, 25(19), 6496–6509.

The first European Nowcasting and Weather Forecasting Conference (ENWFC-2024) dedicated to nowcasting, seamless forecasting, statistical postprocessing and ensemble prediction took place in Oslo, Norway (Oslo Science Park) from 4 to 8 November 2024, organized by EUMETNET (European National Meteorological and Hydrological Services Network) within the Weather Forecasting Cooperation Programme (E-WFC). More than 70 participants attended the conference in person; and ca. 100 more were joining via livestream. 67 conference's presentations (45 oral and 22 posters) were given. The sections in this conference report summarize key topics and discussions of the ENWFC-2024 within: Innovations in satellite and radar technology, early warnings, and crowdsourced data; Advancements in statistical and AI-based weather forecasting; Verification and societal impacts; and Applications. A particular focus was provided to the use of Machine Learning techniques and their applications in the context of the listed key topics.

There exist significant challenges in accurately predicting unusual tropical cyclone (TC) tracks. This study applies the orthogonal conditional nonlinear optimal perturbations (O-CNOPs) method to the Weather Research and Forecast (WRF) model to improve ensemble forecast reliability of unusual TC tracks. Ensemble forecast experiments were conducted for twenty-three forecast periods of five TCs, all of which exhibited sharp turns, to examine the effectiveness of O-CNOPs. Results demonstrate that the O-CNOPs method outperforms the singular vectors (SVs) and bred vectors (BVs) methods by providing more stable and reliable improvements in TC track forecasting skills, from both deterministic and probabilistic perspectives. Notably, the O-CNOPs shows a superior ability to generate ensemble members that accurately predict the sharp turns of TCs at lead times from one to five days. These results highlight the superiority of the O-CNOPs method over the SVs and BVs methods in enhancing the forecasting accuracy of TC tracks, particularly for forecasting unusual TC tracks. This study underscores the potential of O-CNOPs to be extended to real-time TC forecasting and to play an important role in operational track forecasts.

Intermittency has been initially linked to systems alternating regular and irregular states, while nowadays, it also encompasses systems that switch between two or more regimes. These include geophysical processes such as turbulence, convection, and precipitation patterns, not to mention applications in plasma physics, medicine, neuroscience, and economics. Traditionally, the study of intermittency has focused on global statistical indicators, such as the average frequency of regime changes under fixed conditions, or how these vary as a function of the system’s parameters or forcing, like in the case of climate change. However, these global indicators fail in capturing the local spatio-temporal nature of the regime transitions. In this work, we use global and local perspectives to analyze intermittent systems and uncover reliable indicators of regimes’ changes. In particular, using tools such as the Lyapunov exponents and Covariant Lyapunov Vectors (CLVs), we have been able to characterize both global dynamics and local transitions in five different systems, of various complexities, and for three types of intermittency. We identified some key indicators and precursors of the regime transition that are common, despite the differences in the intermittency mechanism and in the dynamical model properties. At the same time, intermittency-type related mechanisms have been unveiled. For instance, we discover very peculiar behaviors in the Lorenz 96 (L96; Lorenz, 1996) and in the Kuramoto-Shivanshinki models (KS; Kuramoto, Y. and Tsuzuki, T., 1976; G. Sivashinsky, 1977) that, despite their notoriety, have been so far unseen. In the L96 we identified crisis-induced intermittency with pseudo-periodic intermissions. In the KS equations we detected a spatially global intermittency which follows the scaling of type-I intermittency. In our local analysis, we uncover the relation between the CLVs mutual alignment and the regimes’ change, a connection that is present in all of the systems and for all types of intermittency considered. In the case of the on-off intermittency, the angle between CLVs is an effective precursor of the jump to the “on” regime. Furthermore, in all systems, the last CLV (the most stable), was found to carry important information about the dynamical features of the intermittency. In the case of type-I intermittency it allowed us to reconstruct the limit cycle around which the intermittency develops, in merging-crisis type of intermittency the last CLV successfully detected the pseudo-periodic intermissions, while in the case of on-off intermittency it allowed us to indicate the “off” state. The identification of these general and fundamental mechanisms driving intermittent behaviours, and in particular the detection of indicators spotting the regimes’ change, have the potential to be impactful in the study of turbulent geophysical fluids, rainfall patterns or atmospheric deep convection. In particular, they could be used to define a latent space of reduced dimension in which a neural network can be trained to automatically predict regime changes.

The trend in Antarctic sea-ice extent has been slightly positive between 1979 and 2015. However, a series of record lows in Antarctic sea ice have occurred since 2016, especially during summer, which could indicate a regime shift. In this context, it is of crucial importance to better understand the drivers of changes in summer Antarctic sea-ice extent. In our study, we make use of five different large model ensembles and compute cause-effect relationships between variables based on the Liang-Kleeman information flow method at the interannual time scale over the period 1970–2100. We find that proximate factors, such as the previous spring sea-ice extent, sea-surface temperature and surface air temperature, are generally the most important drivers of changes in summer Antarctic sea-ice extent at the pan-Antarctic scale, as well as in most Antarctic sectors. The Southern Annular Mode (SAM), Amundsen Sea Low (ASL), and teleconnection patterns associated with El Niño-Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) also have direct and indirect influences on summer sea-ice extent, but with a generally weaker impact compared to proximate factors. An exception is the Western Pacific Ocean, where ENSO and IOD play the largest role. The Indian and Western Pacific Oceans generally present fewer causal connections compared to the Bellingshausen-Amundsen, Weddell and Ross Seas, partly leading to smaller values of direct causal influences on summer sea-ice extent, which indicates a more unpredictable behavior of sea ice in these two sectors.

High-resolution gridded precipitation data is scarce, especially at time intervals shorter than daily. However hydrological applications for example benefit from a finer temporal resolution of rainfall information. In this context, we introduce an hourly precipitation dataset for Belgium, featuring a resolution of 1 km. An hourly high-resolution gridded precipitation product over Belgium can provide valuable insights into the dynamics of both short-term and long-term rainfall events, which can be used for wide-ranging applications such as flash flood forecasting and warning systems, studying precipitation extremes and trends, validating weather and climate models or detecting changes in precipitation patterns due to climate change. Similar products such as EURADCLIM (Europe) (Overeem et al., 2023) and RADKLIM (Germany) (Winterrath et al., 2018), both radar-based gauge-adjusted datasets, have already been created and published. Both datasets are high spatial resolution dataset (2 and 1 km, respectively). A high resolution precipitation grid of hourly precipitation data for Belgium covering the period from 1940 to 2016 using the analog technique, is created. The analogs are sampled from the period 2017–2022 for which high resolution radar data precipitation fields are available. The initial step involves identifying the criteria, i.e. atmospheric parameters such as atmospheric pressure, temperature and humidity, that can be used to determine analogous days. These atmospheric parameters are obtained from the ERA5 observational data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). In a second step, hourly precipitation data for suitable analog days are extracted from our radar database, and then used to create the high resolution grid of hourly precipitation for Belgium from 1940 to 2016. Data from rain gauges on the Belgian terrain were used for validation of the candidate precipitation analogs. The dataset for this project lists the top 25 analog days for 1940–2016 based on similarities in weather patterns. The analogs are ranked based on how closely they match to their target day. The database is relying on the Zarr archiving format and is composed of two archives. A first archive contains all target days together with the 25 best analogs. The second one provides a precipitation field for each hour of every day in the past. The Zarr format of the database allows slicing through the database. For example, it allows one to easily delimit a specific area of interest and a specific time frame for which the high resolution gridded median hourly precipitation fields are needed. The median field dataset is available on Zenodo (https://doi.org/10.5281/zenodo.14965710) (Debrie et al., 2025).

The new programs of EUMETNET have now been launched for 5 years (2024-2028). Within this context, new activities on statistical postprocessing have been proposed, supported by 16 National Meteorological Services. One key activity that will continue is the benchmark of different statistical techniques for correcting weather forecasts. Another one is the development of ready-to-use ensemble calibration techniques that will be available to the community, in particular using machine learning techniques. Finally, several workshops will be organized on the topic during the duration of the project. In this poster, we will discuss the past achievement on statistical postprocessing using the benchmark developed in the context of the previous phase of the project, the current activities, and the future plans in comparing the statistical methods.

Local dimension computed using extreme value theory (EVT) is usually used as a tool to infer dynamical properties of a given state ζ of the chaotic attractor of the system. The dimension computed in this way is also known as the pointwise dimension in dynamical systems literature and is defined using a limit for an infinitely small neighborhood in the phase space around ζ. Since it is numerically impossible to achieve such a limit, and because dynamical systems theory predicts that this local dimension is almost constant over the attractor, understanding the properties of this tool for a finite scale R is crucial. We show that the dimension can considerably depend on R, and this view differs from the usual one in geophysics literature, where it is often considered that there is one dimension for a given dynamical state or process. We also systematically assess the reliability of the computed dimension given the number of points to compute it. This interpretation of the R dependence of the local dimension is illustrated on the Lorenz 63 system not only for ρ=28, but also in the intermittent case where ρ=166.5. The latter case shows how the dimension can be used to infer some geometrical properties of the attractor in phase space. The Lorenz 96 system with n=50 dimensions is also used as a higher-dimension example. A dataset of radar images of precipitation (the RADCLIM dataset) is finally considered, with the goal of relating the computed dimension to the (in)stability of a given rain field.

The predictability of a coupled system composed by a coupled reduced-order extratropical ocean-atmosphere model forced by a low-order 3-variable tropical recharge-discharge model, is explored with emphasis on the long term forecasting capabilities. Highly idealized ensemble forecasts are produced taking into account the uncertainties in the initial states of the system, with a specific attention to the structure of the initial errors in the tropical model. Three main types of experiments are explored with random perturbations along the three Lyapunov vectors of the tropical model, along the two dominant Lyapunov vectors, and along the first Lyapunov vector, only. When perturbations are introduced along all vectors, forecasting biases are developing even if in a perfect model framework. Theses biases are considerably reduced only when the perturbations are introduced along the dominant Lyapunov vector. This perturbation strategy allows furthermore for getting a reduced mean square error at long lead times of a few years, and to get reliable ensemble forecasts on the whole time range. These very counterintuitive findings further underline the importance of appropriately control the initial error structure in the tropics through data assimilation.