Artificial intelligence (AI) has recently shown promising results in ENSO (El Niño–Southern Oscillation) forecasting, outperforming traditional models. Yet AI models deliver accurate predictions without showing the underlying mechanisms. Our study examines a specific AI model, the reservoir computer (RC). Our results show that the RC is less sensitive to initial perturbations than the traditional Zebiak–Cane (ZC) model. This reduced sensitivity can explain the RC's superior skills.

We present an alternative statistical methodology to detect whether the Atlantic Ocean’s circulation system is approaching a tipping point. Our approach separates natural variability from real early warning signals of tipping, reducing false alarms. When applied to a proxy of the Atlantic Ocean’s circulation strength, we find significant signs that the system is undergoing destabilization, suggesting that it may be approaching a tipping point, which could have major impacts on global climate patterns.

After major earthquakes, smaller shocks often follow, yet predicting where they will occur remains difficult. This study introduces a new method for tracking changes in underground stress after a large earthquake. Using the 2018 Hualien earthquake in Taiwan as a case study, we found that areas with strong stress differences provide clearer signals of future aftershocks than stress magnitude alone. This approach can improve short-term earthquake risk assessment and disaster response planning.

Existing regional climate model (RCM) emulators mainly provide deterministic emulations, while internal RCM variability is typically not represented. We develop MESMER-RCM, a probabilistic RCM emulator for annual 2-m temperature, using a simple and physically interpretable approach. We demonstrate its ability to emulate both RCM trends and internal variability in a high-dimensional spatial setting, where existing approaches typically struggle.

Sea ice is crucial in the climate system, especially in the Marginal Ice Zone (MIZ). As the MIZ expands, understanding its dynamics is essential for predicting climate impacts. This paper addresses the largely overlooked role of clouds, developing an idealized atmosphere–ocean–ice model with cloud effects, tackling both simulation and data assimilation. The results imply the potential of integrating idealized models with data assimilation for understanding Arctic dynamics and predictions.

 Unequal-weighted ensemble prediction (UWE) using outputs of the dynamic–statistic prediction is presented, and its possible impact factors are also analysed. Results indicate that the UWE has shown promise in improving the prediction skill of summer precipitation in China on account of the fact that UWE can overcome the shortcomings of the structural inadequacy of individual dynamic–statistic predictions, reducing formulation uncertainties and resulting in more stable and accurate predictions.

We investigate the stability characteristics of lightning discharge channels through their differential resistance properties. Our analysis reveals that lightning channels exhibit bistable behavior, transitioning between low- and high-conductivity states depending on channel length and electric-field conditions. This work provides new insights into lightning channel dynamics and could contribute to improved lightning protection strategies.

The accurate estimation of atmospheric state variables from radar observation in rapidly growing deep convection, which causes heavy thunderstorms, is a major challenge. This study examines the advantage of incorporating radar observation data with very high frequency such as 30 s compared with the conventional case of 5 min, from a theoretical perspective.

We present a novel approach to targeted covariance inflation (TCI) which aims to improve the assimilation of 3D radar reflectivity and, possibly, short-term forecasts of reflectivity and precipitation. Using an operational numerical weather prediction framework, our numerical results show that TCI makes the system accurately generate new reflectivity cells and significantly improves the fractional skill score of forecasts over lead times of up to 6 h by up to 10 %.

This work introduces and evaluates two new tandem data assimilation techniques. The first uses two synchronised forward model runs before a single adjoint model run to consistently increase the precision of the parameter estimation. The second uses a lower-resolution model with adjoint equations to drive a higher-resolution target model through data assimilation with no loss in precision compared to data assimilation without tandem methods.

We have developed a systematic approach to study the climate system at multiple scales using climate networks, which have been previously used to study correlations between time series in space at only a single scale. This new approach is used to upscale precipitation climate networks to study the Indian summer monsoon and to analyze strong dependencies between spatial regions, which change with changing scales.

In this work we address the change in rainfall predictability in Serbia due to climate change, using a novel entropy-based method that highlights both small and large fluctuations. The study is performed on data from 14 stations from 1961–2020. While rainfall average remains rather stable between two subperiods, the predictability of large and small fluctuations has changed, suggesting that climate change has affected rainfall dynamics in ways not observable by standard statistical methods.

Here we apply Bayesian methods to reconstructed and simulated climate model data over past decades to determine the role of long timescale phase dependencies, and extratropical teleconnections, on the major drivers of tropical climate variability.

We present a new framework based on Dynamic Mode Decomposition (DMD) to better detect outliers and model extremes. Unlike standard DMD, which focuses on average system behaviour, our approach targets rare, exceptional dynamics. Applied to climate data, it improves extreme event approximation and reveals meaningful spatiotemporal patterns. The method may generalise to other types of extremes.

Mathematical concepts and measures from dynamical systems theory are applied to identify commonalities across a diverse set of chaotic attractors to better understand the relationship between predictability, directions and rates of expansion and contraction of instabilities over finite time forecast horizons, and dimensionality. The patterns that emerge have broad implications for understanding many dynamical features of geophysical flows.

This work extends the Mapping Particle Filter to account for local dependencies. Two localization methods are tested: a global particle flow with local kernels, and iterative local mappings based on correlation radius. Using a two-scale Lorenz-96 truth and a one-scale forecast model, experiments with full/partial and linear/nonlinear observations show Root Mean Square Error reductions using localized Gaussian mixture priors, achieving competitive performance against Gaussian filters.

Using a simple day-night thermodynamic model based on surface energy balance, this study explains key mechanisms of the urban heat island (UHI): reduced diurnal temperature range (DTR) due to high heat capacity and increased mean temperature from low albedo. The model captures the stronger nighttime UHI and reproduces observed patterns, showing its value in understanding UHI dynamics and urban effects.

This study uses numerical simulations to investigate geometric nonlinearity in six-component seismic motions. Our results reveal the characteristic of these nonlinear effects: while their pattern in body waves is universal, their excitation efficiency shows a strong source-type dependency. It demonstrates that rotational components are highly sensitive to effects driven by S-waves, and that surface waves may be the primary carriers of these features.

Extreme weather poses serious risks, making prevention crucial. Using the Lorenz 96 model as a testbed, we propose a bottom-up approach to mitigate extreme events via local interventions guided by multi-scenario ensemble forecasts. Unlike control-theoretic methods, our approach selects the best control scenario from available options. It achieves a high success rate of 99.4% while maintaining reasonable costs, offering a practical strategy to reduce extremes under limited control.

We used a simple coupled model and a data assimilation method to find the correct initialisation for climate predictions. We aim to clarify when weakly or strongly coupled data assimilation (WCDA or SCDA) is best, depending on the system's dynamical characteristics (spatio-temporal) and data coverage. We found that WCDA is better in full data coverage. When we have a partially observed system, SCDA is better. This result depends on the temporal and spatial scale of the observed quantity.

Climate prediction is challenging due to systematic errors in traditional climate models. We addressed this by training a machine learning model to correct these errors and then integrating it with the traditional climate model to form an AI-physics hybrid model. Our study demonstrates that the hybrid model outperforms the original climate model on both short-term and long-term predictions of the atmosphere and ocean.

Particle-tracking simulations compute how ocean currents transport material. However, initializing these simulations is often ad hoc. Here, we explore how two different strategies (releasing particles over space or over time) compare. Specifically, we compare the variability in particle trajectories to the variability of particles computed in a 50-member ensemble simulation. We find that releasing the particles over 20 weeks gives variability that is most like that in the ensemble.

The Madden–Julian Oscillation (MJO) is a large-scale tropical wave of enhanced and suppressed rainfall, slowly moving eastward along the equator and influencing the weather and climate globally. We study the MJO using a simplified model designed to capture its large-scale features. We introduce new, more realistic model inputs, show that this enhanced model successfully replicates key characteristics of the MJO, and identify some of its limitations.

Controlling chaotic systems is a key step toward weather control. The control simulation experiment (CSE) modifies weather systems using small perturbations, as shown in studies with the Lorenz-63 model. However, the effectiveness of CSE compared to other methods is unclear. This study evaluates CSE against model predictive control (MPC). Simulations reveal that MPC achieves higher success rates with less effort under certain conditions, linking control theory and atmospheric science.

The hydrostatic approximation is the basis of most simulations of ocean and climate dynamics. It is evaluated here by using a projection method in the 4D Fourier space. The evaluation is analytic.

Turbulence is found in many natural systems. We study its statistics in terms of scaling laws of up to fourth-order normalized moments for the velocity or magnetic fields through their relative behavior. We show analytically using existing intermittency models that the physical dimension of structures becomes irrelevant. The strongest intermittent structure has a parabolic scaling as found previously and confirmed by long simulations. More analysis will be performed using simplified models. 

The contributing influence of SSWs (sudden stratospheric warming) on regional ionosphere through chaos theory is examined. We find that ionospheric chaos is more pronounced in the European sector compared to the African sector during an SSW. Evidence of orderliness behavior in regional ionosphere of African sector is observed. Finally, we notice that after the peak phase of an SSW, ionospheric chaos is found to be more pronounced.

Can multilevel Monte Carlo methods improve ensemble variational data assimilation without increasing its computational cost? By shifting part of the ensemble generation cost to coarser simulation grids, larger ensemble sizes become affordable. This gives smaller sampling errors without introducing any coarse-grid bias. Numerical experiments with a quasi-geostrophic model demonstrate the potential of the approach and highlight the challenges of operational implementation.

In recent years, there have been more and more floods due to intense precipitation, such as the July 2021 event in Belgium. Predicting precipitation is a difficult task, even just for the next few hours. This study focuses on a tool that assesses whether a given situation is stable or not (i.e., whether it is likely to stay as it is or could evolve in an unpredictable manner).

Fog can disrupt aviation by causing accidents and delays due to low visibility, yet it remains underresearched in Croatia. This study examined fog and mist at Pula Airport using 20 years of data and machine learning techniques. There is a declining trend in fog that is linked to changing weather patterns. Fog mainly occurs from October to March. These findings enhance knowledge about fog in Croatia and can improve weather forecasts, increasing safety at the airport.