Abstract. Hydrogen is a potential candidate for an alternate energy source and carrier. As usage of hydrogen in industry rises, leakages into the atmosphere may occur, causing an increase in the global atmospheric hydrogen concentration. Hydrogen is an indirect greenhouse gas, known to increase methane, stratospheric water vapour, and tropospheric ozone. Methane and hydrogen are closely coupled, with the main atmospheric destructive pathway of both species being via reaction with the hydroxyl radical (OH). Currently, most earth system models (ESMs) simulate hydrogen or methane with a prescribed lower boundary condition, which suppresses chemical feedbacks at the surface. In this work, we implement hydrogen emissions and a hydrogen soil uptake scheme into an ESM with free-running methane to demonstrate the capability of a fully interactive hydrogen and methane emissions-driven ESM. We show that the model is able to capture long term trends and seasonal cycles of both species when compared to observations, and find that the inclusion of both fluxes does not impact other chemical species in the model, such as tropospheric ozone. We show that the model can be used under pre-industrial conditions and present day scenarios. The ESM with fully coupled hydrogen and methane chemistry has great potential to be used in future scenarios and to estimate a more accurate global warming potential of hydrogen.

Abstract. High-resolution geodynamic simulations of mantle convection are essential to quantitatively assess the complex physical mechanisms driving the large-scale tectonic processes that shape Earth's surface. Accurately capturing small-scale features such as unstable thermal boundary layers requires global resolution on the order of 1 km. This renders traditional sparse matrix methods impractical due to their prohibitively high memory demands and low arithmetic intensity. Matrix-free methods offer a scalable alternative, enabling the solution of large-scale linear systems efficiently. In this work, we leverage the matrix-free Finite Element framework HyTeG to conduct large-scale geodynamic simulations that incorporate realistic physical models. We validate the framework through a combination of convergence studies of the Finite Element approximations against analytical solutions and through geophysical community benchmarks. The latter include test cases with temperature-dependent and nonlinear rheologies. Our scalability studies demonstrate excellent performance, scaling up to problems with about 100 billion (1011) unknowns in the Stokes system.

Abstract. We present a set of new Landlab numerical model components that allow users to track sediment properties across a landscape grid. The components use a mass-balance approach to partition the mass concentration of each property based on sediment fluxes calculated by various Landlab flux components. The methods are generic, allowing the user to assign any sediment property that can be expressed as a mass, volume, or number concentration (for example, mass of magnetite, volume of quartz, number of zircons, number of radiogenic 10Be atoms, “equivalent dose” of luminescence). Several properties can be tracked at once, each with concentration tracked in both sediment and bedrock at every location on the grid. Two ConcentrationTracker components have been formulated; one for distributed, space- and time-varying hillslope regolith movement and another for transport in fluvial networks, allowing for interaction between sediment in the water column and on the channel bed. These components can be used individually to study a single process or coupled to study the interactions of multiple processes acting on a dynamic landscape. We present two examples that illustrate the diverse uses of the ConcentrationTracker components: colour banding in hillslope regolith and provenance tracking of fluvial sediments.

Abstract. The growing availability of glacier observations poses a challenge for models to integrate this heterogeneous information in a dynamically consistent way. At the same time, estimates of current glacier volume and area remain uncertain, as many global inventories and thickness datasets date back to the early 2000s. We present the Open Global Glacier Data Assimilation Framework, named AGILE, a time-dependent variational method inspired by 4D-Var data assimilation. AGILE is built on a reimplementation of the OGGM flowline glacier evolution model in PyTorch, enabling full differentiability through automatic differentiation (AD). We test AGILE v0.1 in a series of idealized experiments designed to reflect common initialization and calibration scenarios in global glacier modeling. The goal is to recover glacier bed topography and distributed ice volume in 2020 through transient calibration, based on dynamical simulations starting in 1980. In these experiments, we assume a perfectly known mass balance and fixed ice dynamics parameters. While this setup simplifies real-world complexity, it allows us to isolate and evaluate the core functionality of the approach. Our results show that AGILE efficiently optimizes multiple control variables by leveraging AD-derived gradients, requiring only a few iterations to substantially improve upon initial guesses. We also examine the potential to reconstruct earlier glacier states (e.g., in 1980) without direct observations and find that this is fundamentally limited because glacier dynamics are governed by a diffusion equation, which leads to a loss of information about past states over time, even in an idealized setting. Overall, our experiments demonstrate AGILE’s potential as a flexible and efficient data assimilation framework. Its ability to integrate diverse datasets in a dynamically consistent manner makes it a promising tool for future real-world glacier modeling applications.

Abstract. Conceptual hydrological models, traditionally relying on simplified representations of physical processes governed by conservation laws remain widely used in operational hydrology due to their explainability and practical applicability. However, these process-based models inherently face structural uncertainties and a lack of scale-relevant theories – challenges that emerging artificial intelligence (AI) techniques may help address. In parallel, high-resolution models are crucial for predicting extreme events characterized by strong variability and short duration, making spatially distributed hybrid modeling critical in the current context. We introduce a hybrid physics–AI framework that embeds neural networks (NNs) seamlessly into a spatialized, regionalizable, and fully differentiable process-based model via universal differential equations (UDEs). The model integrates a state-dependent NN to refine internal water fluxes and an implicit resolution of the UDE system, followed by kinematic wave routing on a flow direction grid. Spatially distributed parameters are inferred through regionalization mappings including convolutional NNs, and adjoint-based gradients enable end-to-end training of the hybrid system. We implement this framework into the latest release of the smash platform, significantly extending its capabilities to comprehensively evaluate hybrid models at kilometric spatial and hourly temporal resolutions. The results show that hybrid approaches demonstrate consistently strong and stable performance in calibration and various validation scenarios. Additionally, the UDE structure exhibits a hybridization effect that modifies state dynamics and runoff flow, achieving more accurate streamflow simulations for flood modeling.

Abstract. Australian commonwealth, state and territory geological surveys possess information on over 3 million drillhole logs. In addition to mineral exploration drilling, extensive drillhole datasets exist from oil and gas exploration and hydrogeological studies. Other countries no doubt have similar data holdings. Together these legacy drillhole datasets have the potential to significantly enhance constraints on regional 3D geological models and improve our understanding of subsurface architecture, but have limited use in their current form as many if not most drill logs lack stratigraphic information, containing only lithological descriptions. This study develops open-source codes and methodologies for stratigraphy recovery (determining the ordered sequence of stratigraphic units) from drillhole lithological data by introducing a search algorithm that systematically explores all geologically plausible stratigraphic orderings for individual drillholes, combined with a solution correlation algorithm that compares the topological relationships of stratigraphic units across multiple drillholes to identify geologically consistent solutions and reduce uncertainty. The algorithms combine constraints from lithological descriptions with stratigraphic relationships automatically derived from regional maps. In addition, the method quantifies uncertainty by generating multiple plausible stratigraphic interpretations, providing critical insights for resource estimation, scenario analysis, and data acquisition strategies. The application of our method to a dataset of 52 drillholes from South Australia demonstrated its ability to make useful predictions of stratigraphic solutions and quantifying associated uncertainties. These results not only validate our approach but also highlight opportunities to refine current stratigraphic descriptions and provide a valuable new source for regional 3D geological modelling.

Abstract. We propose a stochastic framework for wildfire spread prediction using deep generative diffusion models with ensemble sampling. In contrast to traditional deterministic approaches that struggle to capture the inherent uncertainty and variability of wildfire dynamics, our method generates probabilistic forecasts by sampling multiple plausible future scenarios conditioned on the same initial state. As a proof-of-concept, the model is trained on synthetic wildfire data generated by a probabilistic cellular automata simulator conditioned on canopy cover, vegetation density, and terrain slope for two real fires, namely the Chimney Fire in 2016 and the Ferguson Fire in 2018, both in California. To assess predictive performance and uncertainty representation under an identical neural network architecture, we compare a conventional supervised regression training paradigm against a conditional diffusion framework that employs ensemble sampling, and evaluate both approaches on independent ensemble test datasets. Across independent ensemble test sets, the diffusion surrogate consistently outperforms the deterministic baseline. It delivers lower errors in standard accuracy metrics such as mean squared error (MSE), exhibits higher spatial coherence as reflected by improved structural similarity index measure (SSIM) values, and generates samples of superior distributional quality according to the Fréchet inception distance (FID). Moreover, the diffusion-based model shows stronger probabilistic capability, as evidenced by higher scores in the hit rate (HR) metric, which we introduce as an uncertainty-aware verification measure. These results demonstrate that diffusion-based ensemble modelling provides a more flexible and effective approach for wildfire forecasting and, by capturing the distributional characteristics of future fire states, supports the generation of fire susceptibility maps that convey probabilistic risk information useful for assessment and operational planning in fire-prone environments.

Abstract. Fires play an important role in the Earth system but remain complex phenomena that are challenging to model numerically. Here, we present the first version of BuRNN, a data-driven model simulating burned area on a global 0.5° × 0.5° grid with a monthly time resolution. We trained Long Short-Term Memory networks to predict satellite-based burned area (GFED5) from a range of climatic, vegetation and socio-economic parameters. We employed a region-based cross-validation strategy to account for the high spatial autocorrelation in our data. BuRNN outperforms the process-based fire models participating in ISIMIP3a on a global scale across a wide range of metrics. Regionally, BuRNN outperforms almost all models across a set of benchmarking metrics in all regions. Through explainable AI we unravel the difference in regional drivers of burned area in our models, showing that the presence/absence of bare ground and C4 grasses along with the fire weather index have the largest effects on our predictions of burned area. Lastly, we used BuRNN to reconstruct global burned area for 1901–2019 and compare the simulations against independent long-term historical fire observation databases in five countries and the EU. Our approach highlights the potential of machine learning to improve burned area simulations and our understanding of past fire behaviour.

Abstract. Climate change significantly threatens crop yields levels and stability. The complex interplay of factors at the local scale makes assessing these impacts difficult, requiring coupled climate-phenology models, which integrate climate data and crop information. Identifying suitable local management practices and crop varieties under future conditions becomes essential for developing effective adaptation strategies. This study presents the implementation and application of an integrated climate-phenology adaptation support modelling system. This is based on regional CORDEX climate models and the CERES Maize model from the DSSAT platform. Novel modules for optimal management and genotype identification under climate change have been developed in the system, employing a hybrid approach that combines deterministic modelling with machine learning (ML) techniques and genetic algorithms. This system was run as a regional pilot over Southern Romania, operating in real-time in interaction with users, performing agro-climate projections (combination of fertilization, sowing date, genotype) and providing best crop management simulated under climate change projections. Multi-model ensemble simulations were conducted for two radiative forcing scenarios RCP4.5 and RCP8.5 and twelve management scenarios, yielding novel results for the region. Results indicate a projected decrease in maize yields for the current genotype across all tested scenarios, primarily attributed to a shortened grain-filling period and reduced fertilization efficiency under warmer conditions. The analysis warns about a projected narrowing of the agro-management options for maintaining a high yield level. However, we find an added value from the impact of genotype selection in mitigating climate change impacts, even in extreme years. Genotype optimisation across six crossed cultivar dependent parameters revealed that while maximum yields decline, specific genotype windows exhibit increased intermediate yields under future climates compared to current conditions. Sensitivity analysis identified the thermal time requirements during juvenile and maturity stages as the most critical factors influencing genotype performance under warmer climates. This research demonstrates the added value of combining deterministic and data-driven modelling approaches within a coupled climate-crop system for developing effective adaptation strategies, including optimised fertilization pathways that contribute to climate change mitigation.

Abstract. SSH-aerosol is developed to represent the evolution of primary and secondary pollutants in the atmosphere by processes linked to gas-phase chemistry, aerosol dynamics (coagulation, condensation/evaporation and nucleation) and intra-particle reactions. The representation of process complexity can be adjusted based on the user's choices. The model uses a sectional size distribution, and offers the capability to discretize chemical composition to account for the mixing state of particles. The algorithms are designed to represent the evolution of ultrafine particles: conservation of mass and number during numerical resolution, taking into account the Kelvin effect, the condensation dynamics of nonvolatile compounds, and nucleation. Different parameterizations are provided for nucleation: binary, ternary, heteromolecular and organic nucleation depending on the compounds involved. For gas-phase chemistry, schemes of different complexities can be handled: from simple schemes to model ozone, oxidants and inorganic chemistry (e.g. CB05, RACM2, Melchior2), to more complex schemes, e.g. from the Master Chemical Mechanism (MCM). The complexity of the schemes used for secondary organic aerosol (SOA) formation may also be adjusted: from schemes built from chamber data to near-explicit schemes from MCM. SOA schemes reduced using the GENOA algorithm are also provided for several precursors (toluene, a sesquiterpene and three monoterpenes), together with their evaluation against chamber or flow-tube experiments. A wall-loss module has also been added for easier comparisons to chamber experiments. Specific developments were made in version 2.0 to automatically link the chosen gas-phase mechanism to SOA formation by using the SMILES structure of organic compounds, allowing for the determination of their hydrophilic and hydrophobic properties and for the partitioning in both organic and aqueous phases. The gas/particle partitioning may also be represented with different complexities. For the organic phase, viscosity may be modelled, adapting the aerosol viscosity to its composition, and coupling organic and inorganic thermodynamics. The dynamic evolution of the partitioning may be computed explicitly or thermodynamic equilibrium may be assumed. Different options are also provided to simulate the chemistry of organic compounds inside the particles with different types of reactions: irreversible 1st order reactions, bulk oligomerization, hydratation of aldehydes and reactions of organic compounds with inorganic ions. The SSH-aerosol model may be installed with a docker for standalone use. It has also been coupled to several 3D models to represent gas and aerosol concentrations: from the local scale with computational fluid dynamic and street network models to the regional scale with chemistry-transport models.