Accurate representation of turbulent processes remains a critical challenge in atmospheric modelling. Large Eddy Simulations (LES) serve as valuable tools for understanding atmospheric turbulence by explicitly resolving energy-containing eddies while parameterizing smaller-scale motions through subgrid-scale (SGS) models. In their most complex forms, these SGS parameterizations can significantly influence LES performance and computational efficiency, making their improvement useful for advancing atmospheric modelling capabilities. Neural Network based emulation of such parametrizations have proven effective in reducing the computational cost, while maintaining accuracy and stability.

Building upon recent advances in physics-informed neural networks (NN) for atmospheric modelling and emulation of physics-based processes, we present a physics-guided NN architecture for emulation of the SGS turbulence parameterizations that introduces several key innovations. Our approach uniquely combines scale-specific normalization with multi-scale feature extraction through parallel convolutional paths, distinguishing it from existing physics-guided machine learning frameworks. The deep learning-based (DL) model also incorporates physically-motivated constraints across different spatial scales while simultaneously ensuring conservation of momentum and energy.

Unlike earlier studies that focus on single aspects of physical conservation, our architecture implements a comprehensive physics-informed framework that combines Richardson number gradient handling for stability constraints, with explicit treatment of diffusion and viscosity coefficients, and scale-specific normalization for different atmospheric variables. The model was trained on limited high-resolution Radiative-Convective Equilibrium (RCE) simulations from the Met Office-Natural Environment Research Council (NERC) Cloud model (MONC), employing physics-based loss functions that enforce both conservation laws and stability constraints.

The training dataset consisted of 3-D diagnostics data from the RCE simulations, with a 64x64 km² domain and 1 km grid spacing in the horizontal. While the original simulations had 99 vertical levels with varying vertical resolution, the DL model was trained on random slices (vertical levels) chosen from the original data volume. The inputs consisted of the resolved state variables like velocity components (u, v, w) from the previous time step, the perturbations to potential temperature, mixing ratios and Richardson number, whereas the targets for the DL model were the SGS tendencies of the model prognostic fields resulting from the Smagorisnky parameterization and the coefficients of viscosity and diffusion.

The DL model's cross-regime applicability was evaluated through multiple independent test cases: (200-second sampling frequency) and different atmospheric conditions from the Atmospheric Radiation Measurement (ARM) program. The simulations from ARM atmospheric settings were mainly targeted at simulating shallow convection, with different grid/domain configurations. Results from the off-line tests demonstrate promising performance in predicting SGS and transport coefficients (viscosity and diffusion) across these varied conditions, particularly in maintaining physical consistency during regime transitions.

Our preliminary findings indicate that this enhanced multi-scale, physics-informed architecture can effectively learn SGS parameterizations from limited training data while maintaining physical fidelity across different atmospheric conditions and spatio-temporal resolutions. This approach demonstrates the potential for the development of high-fidelity, generalizable parameterizations for weather and climate models, suggesting a route forward for reducing the greater computational costs associated with more complex SGS parameterization schemes.

How to cite: Panda, S. K., Jones, T., Shahzad, M., Ellis, A.-L., and Lawrence, B.: Physics-Guided Deep Learning-Based Emulation of Subgrid-Scale Turbulence Parameterization for Atmospheric Large Eddy Simulations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-13920, https://doi.org/10.5194/egusphere-egu25-13920, 2025.

Stochastic parameterisations have seen widespread use in atmospheric models. These schemes represent uncertainty by adding terms that include a random noise component directly to the equations that describe the time evolution of the model. Stochastic parameterisation development thus involves the following questions: what are the sources of uncertainty, how do we represent them, and how precisely should we formulate stochastic terms to quantify them? Common methods to quantify the uncertainty inherent in parameterisation include applying multiplicative perturbations to physics tendencies (such that the larger the tendency due to subgrid processes, the more uncertainty we should have in the tendency) or applying perturbations to physical parameters (our physics schemes often rely on parameters whose values we do not know precisely, and have complicated nonlinear responses to perturbing this set of parameters, so perturbing each one within its own specified range during the model run allows this uncertainty to feed back into the model state).

In this work we present a different approach to stochastic parameterisation. We perturb the thermodynamic profiles that constitute the inputs to parameterisation schemes. The perturbations are scaled by the degree of subgrid inhomogeneity. A representation of the subgrid inhomogeneity is estimated by a machine learning model which has been trained on coarse-grained high resolution (dx=~1.5km) model output from the Met Office Unified Model. The scheme is implemented in LFRic, the UK Met Office’s next generation modelling system, and we present results of experiments ran in single column model mode.

How to cite: Reid, H. and Morcrette, C.: A new stochastic physics scheme incorporating machine-learnt subgrid variability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11747, https://doi.org/10.5194/egusphere-egu25-11747, 2025.

The added value of machine learning for weather and climate applications is measurable through performance metrics, but explaining it remains challenging, particularly for large deep learning models. Inspired by climate model hierarchies, we propose that a full hierarchy of Pareto-optimal models, defined within an appropriately determined error-complexity plane, can guide model development and help understand the models' added value. We demonstrate the use of Pareto fronts in atmospheric physics through three sample applications, with hierarchies ranging from semi-empirical models with minimal parameters (simplest) to deep learning algorithms (most complex). First, in cloud cover parameterization, we find that neural networks identify nonlinear relationships between cloud cover and its thermodynamic environment, and assimilate previously neglected features such as vertical gradients in relative humidity that improve the representation of low cloud cover. This added value is condensed into a ten-parameter equation that rivals deep learning models. Second, we establish a machine learning model hierarchy for emulating shortwave radiative transfer, distilling the importance of bidirectional vertical connectivity for accurately representing absorption and scattering, especially for multiple cloud layers. Third, we emphasize the importance of convective organization information when modeling the relationship between tropical precipitation and its surrounding environment. We discuss the added value of temporal memory when high-resolution spatial information is unavailable, with implications for precipitation parameterization. Therefore, by comparing data-driven models directly with existing schemes using Pareto optimality, we promote process understanding by hierarchically unveiling system complexity, with the hope of improving the trustworthiness of machine learning models in atmospheric applications.

Preprint: https://arxiv.org/abs/2408.02161

How to cite: Beucler, T., Grundner, A., Shamekh, S., Ukkonen, P., Chantry, M., and Lagerquist, R.: Distilling Machine Learning's Added Value: Pareto Fronts in Atmospheric Applications, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15200, https://doi.org/10.5194/egusphere-egu25-15200, 2025.

Differentiable programming enables automatic differentiation (AD) tools to compute gradients through code without manually defining derivatives. AD tools can differentiate through entire software stacks, composing many functions and algorithms via the chain rule. With models that incorporate differentiable programming long-standing challenges like systematic calibration, comprehensive sensitivity analyses, and uncertainty quantification can be tackled, and machine learning (ML) methods can be integrated directly into the process-based core of earth-system models (ESMs) to incorporate additional information from observations. Through the advent of ML, several AD tools are gaining traction. A new generation of powerful tools like JAX, Zygote and Enzyme enable differentiable programming for models of varying complexity including highly complex coupled ESMs. Here we present an overview about the perspectives of differentiable programming for ESMs, using experience from two of our applications in atmospheric modelling. First off, we set up PseudoSpectralNet, a differentiable quasi-geostrophic model in Julia with Zygote. This is a hybrid model combining neural networks with a dynamical core, showcasing how the stability and accuracy of ML models is improved by integrating a process-based dynamical core into our model. Additionally, ongoing work uses Enzyme to achieve a differentiable version of the significantly more complex SpeedyWeather.jl atmospheric model. We will discuss advantages of both approaches and give an outlook into future possibilities with differentiable models. 

How to cite: Gelbrecht, M., Klöwer, M., and Boers, N.: Differentiable Programming for Atmospheric Models: Experiences and Perspectives , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-12601, https://doi.org/10.5194/egusphere-egu25-12601, 2025.

The training of machine learning models for weather and climate on multiple datasets, including local high-resolution reanalyses and level-1 observations, is one of the frontiers of the field. It promises to allow for models that are no longer constrained by the capabilities of equation-based models, as is currently still largely the case when training on global reanalyses. For example, level-1 observations contain the feedback from arbitrary scale processes and hence do not suffer from the closure problem of equation-based models. Training on observations might hence lead to machine learning-based Earth system models with reduced systematic biases, in particular for long-term climate projections. Local reanalyses are only available for a small set of regions, mainly over Europe and North America. Appropriate training might allow one to generalize the detailed process information in these to other regions or even globally. 

In this talk, we present results on the effective training with a combination of global and local reanalysis as well as level-1 observations. We consider different pre-training protocols to learn the correlations between datasets, which is critical to obtain a benefit through their combination. We use a forecasting task as baseline and study the effectiveness of different variants of masked-token modeling and more sophisticated approaches that exploit the latent space of the machine learning models. We also study different fine-tuning strategies to extract a best state estimate from multiple datasets and to generalize regional datasets globally. For this, we build on the extensive results on fine-tuning of large language models that have been developed in the last years. Our results aim to determine general principles which combination of datasets is beneficial. We also perform a detailed analysis of the physical consistency and physical process representation in the model output. Through this, we believe our work provides an important stepping stone for the next generation of machine learning-based models for weather and climate.

How to cite: Lessig, C.: Towards next generation machine learning-based Earth system models that exploit a wide range of datasets, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16817, https://doi.org/10.5194/egusphere-egu25-16817, 2025.

This study explores integrating reinforcement learning (RL) with idealised climate models to address key parameterisation challenges in climate science. Current climate models rely on complex mathematical parameterisations to represent sub-grid scale processes, which can introduce substantial uncertainties. RL offers capabilities to enhance these parameterisation schemes, including direct interaction, handling sparse or delayed feedback, continuous online learning, and long-term optimisation. We evaluate the performance of eight RL algorithms on two idealised environments: one for temperature bias correction, another for radiative-convective equilibrium (RCE) imitating real-world computational constraints. Results show different RL approaches excel in different climate scenarios with exploration algorithms performing better in bias correction, while exploitation algorithms proving more effective for RCE. These findings support the potential of RL-based parameterisation schemes to be integrated into global climate models, improving accuracy and efficiency in capturing complex climate dynamics. Overall, this work represents an important first step towards leveraging RL to enhance climate model accuracy, critical for improving climate understanding and predictions. Code accessible at https://github.com/p3jitnath/climate-rl.

How to cite: Nath, P., Moss, H., Shuckburgh, E., and Webb, M.: RAIN: Reinforcement Algorithms for Improving Numerical Weather and Climate Models, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5159, https://doi.org/10.5194/egusphere-egu25-5159, 2025.

Accurate oil spill predictions are crucial for mitigating environmental and socioeconomic impacts. Numerical models, like MEDSLIK-II (De Dominicis et al., 2013), simulate oil advection, dispersion, and transformation, but their performance depends heavily on the configuration of physical parameters, often requiring labor-intensive manual tuning based on expert judgment.

To address this limitation, we integrate MEDSLIK-II with a Bayesian Optimization (BO) framework to systematically identify the optimal parameter configuration, ensuring simulations closely match observed spatiotemporal oil spill distributions. Our optimization focuses on horizontal diffusivity and drift factor parameters, using the Fraction Skill Score as the objective metric to maximize, thus reducing the overlap between simulations and observations.

The approach is validated on the 2021 Baniyas (Syria) oil spill, demonstrating improved accuracy, reduced biases and lower computational costs compared to the standalone numerical model.

By integrating BO with the MEDSLIK-II numerical model, our method enhances oil spill prediction capabilities and provides a transferable, physically consistent optimization framework applicable to a wide range of geophysical challenges.

This work is conducted within the framework of the iMagine European project, which leverages Artificial Intelligence, including AI-assisted image generation, to advance a series of use cases in marine and oceanographic science.

References

De Dominicis, M., Pinardi, N., Zodiatis, G., & Lardner, R. (2013). MEDSLIK-II, a Lagrangian marine surface oil spill model for short-term forecasting – Part 1: Theory. Geoscientific Model Development, 6, 1851–1869. https://doi.org/10.5194/gmd-6-1851-2013

How to cite: De Carlo, M. M., Accarino, G., Atake, I., Elia, D., Epicoco, I., and Coppini, G.: Improving Oil Spill Numerical Simulations through Bayesian Optimization, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17333, https://doi.org/10.5194/egusphere-egu25-17333, 2025.

AI enhanced environmental modelling workflows: Towards Automated Scientific Exploration in Hydrology

Authors: Darri Eythorsson, Kasra Keshavarz, Cyril Thébault, Mohamed Ismaiel Ahmed, Raymond Spiteri, Alain Pietroniro and Martyn Clark

Modern hydrological modeling has evolved into a complex scientific endeavour requiring sophisticated workflows that span multiple scales, processes, and computational paradigms. While existing workflow solutions address specific technical challenges, the field lacks comprehensive frameworks that can support end-to-end modeling while maintaining reproducibility and scalability. This works introduces two complementary frameworks that aim to address these fundamental challenges: CONFLUENCE (Community Optimization Nexus For Large-domain Understanding of Environmental Networks and Computational Exploration) and INDRA (the Intelligent Network for Dynamic River Analysis).

CONFLUENCE implements a modular architecture that enforces workflow reproducibility through a unified configuration system while maintaining the flexibility needed to support diverse modeling applications. The framework provides comprehensive solutions for four critical workflow components: (1) flexible geospatial domain definition and discretization, (2) model-agnostic data acquisition and preprocessing, (3) extensible model setup and parameterization capabilities, and (4) comprehensive evaluation and optimization tools. This systematic approach enables efficient, reproducible, and transparent hydrological modeling across scales.

INDRA augments this foundation by implementing a network of specialized AI expert agents that support various components of the hydrological modeling workflow. Through structured dialogue between domain experts (including AI specialists in hydrology, hydrogeology, meteorology, data science, and geospatial analysis), INDRA provides context-aware guidance while maintaining complete provenance of modeling decisions and their justification. This AI-assisted approach helps address three critical challenges: (1) the growing complexity of modelling decisions, (2) the need for reproducible workflows and detailed documentation, and (3) the technical barriers limiting broader adoption of advanced modeling practices.

The integration of these frameworks aims to explore how automation and AI assistance can enhance rather than disrupt traditional modeling practices. By maintaining clear documentation of decisions and their justifications, these systems help build trust in model results while creating opportunities for recursive learning from previous modeling experiments. Our case studies, spanning scales from individual catchments to continental domains, showcase the frameworks' capabilities while highlighting their potential to transform how researchers’ interface with complex environmental modeling workflows.

This work aims to advance both operational and research oriented hydrological modeling practices, offering a foundation for reproducible, scalable, and interoperable modeling while maintaining scientific rigor and flexibility. The framework’s open-source nature and modular design create opportunities for community-driven development and extension, potentially accelerating scientific discovery in hydrological sciences.

How to cite: Eythorsson, D., Keshavarz, K., Thébault, C., Ahmed, M., Spiteri, R., Pietroniro, A., and Clark, M.: AI enhanced environmental modelling workflows: Towards Automated Scientific Exploration in Hydrology, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4565, https://doi.org/10.5194/egusphere-egu25-4565, 2025.

Vertical energy transport from the land surface into the atmosphere in the form of sensible and latent heat flux must be well represented in numerical weather prediction models to allow accurate estimates of near-surface atmospheric variables. Traditionally, these heat fluxes are parameterized relying on Monin-Obukhov Similarity Theory (MOST), which is based on differences in wind speed, air temperature, and humidity between adjacent measurement or model levels. Recently, Wulfmeyer et al. (2024) estimated heat flux with machine learning at much higher accuracy compared to MOST. Their ML model proposed the incorporation of additional predictor variables when estimating latent heat flux (such as solar radiation), which stands in contrast to the classical MOST approach. However, the analysis in Wulfmeyer et al. (2024) is based on a rather short data period in August 2017 at three nearby locations in Oklahoma, USA, which limits the generalizability of the results. Here, we replicate and expand the findings from Wulfmeyer et al. (2024) on a dataset from the boundary layer field site (GM) Falkenberg of the German Meteorological Service over a period of twelve years, covering various seasons and synoptic weather situations. Our findings support the role of incoming shortwave radiation not only for latent but also for sensible heat flux estimates, particularly for other parts of the year. The results thus underline the potential to develop more advanced flux parameterizations beyond MOST. In future research, we intend to investigate the role of other predictor variables, such as vapor pressure deficit or soil moisture, to assess the generalizability of the relations, to judge their performance under extreme conditions, and to derive simple but universally applicable parameterizations.

How to cite: Butz, M. V., Karlbauer, M., Beyrich, F., and Wulfmeyer, V.: Replicating Sensible and Latent Heat Flux Diagnosis with Multilayer Perceptrons on Multi-Year Falkenberg Tower Data , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17542, https://doi.org/10.5194/egusphere-egu25-17542, 2025.