“Spin-up” is time needed for a model’s result to become effectively free of influence from initial conditions, and it is usually excluded from analysis. Here, spin-up is examined by comparing one decadal simulation to another initialized 20 years prior, in order to determine when their solutions converge. Differences lessen over the first fall and winter, but re-emerge over the following spring and summer, suggesting that at least 1 annual cycle is needed to spin up regional climate simulations.

We examined the two-moment cloud microphysics sedimentation schemes of the ICON (ICOsahedral Nonhydrostatic) weather model, comparing the default semi-implicit with an explicit method faster on graphics processing units. Using idealized setups and thunderstorm case studies, we find differences in numerical diffusion and extreme precipitation rates due to changed coupling with the remaining microphysics. Neither method develops alarming instabilities in full model setups; both can be safely used.

Thunderstorms can accelerate particles in the atmosphere, producing bursts of radiation at the ground. We investigated how strong the electric field inside a cloud must be to start such events. Using advanced computer simulations and comparing with measurements from mountain stations, we found that fields must be stronger than earlier theory suggested. Our results improve understanding of storm electricity and its role in natural radiation.

We present the implementation of a new climate-phenology integrated system for adaptation to climate change, using high-resolution scenarios and the Decision Support System for Agrotechnology Transfer crop model, with new modules developed for optimal agromanagement and genotypes identification using a hybrid deterministic/machine learning Genetic-Algorithms method. The system is user-interactive in real time. It has been implemented for South Romania and is applicable and extendable for Europe.

A major shift in Arctic sea ice occurred in 2007, transitioning from thicker, deformed ice to thinner, more uniform ice with reduced surface roughness. This abrupt change likely altered the dynamic and thermodynamic interactions between sea ice and ocean. We present a suite of regional coupled ocean-sea ice simulations designed to assess the potential impact of the regime shift on sea ice-ocean interactions, with a regional focus on the Atlantic sector of the Arctic Ocean.

The release of carbon from thawing permafrost soils could amplify future climate warming. However, this feedback is highly uncertain because most Earth system models (ESM) do not represent permafrost carbon. We have improved the Institut Pierre-Simon Laplace ESM by including permafrost physical, carbon and nitrogen processes to better represent Arctic ecosystems. The model more accurately represents past and present permafrost physics and biogeochemistry, paving the way for future projections.

Climate models are becoming more detailed and accurate by simulating weather at scales of just a few kilometers. Simulating at km-scale is computationally demanding requiring powerful supercomputers and efficient code. This work presents a refactored dynamical core of a state-of-the-art climate model using a Python-based approach. The refactored code has passed through a sequence of verification and validation demonstrating its usability in performing km-scale global simulations.

Ground-based microwave radiometers (GMWRs) offer high temporal resolution observations with strong sensitivity to the lower atmosphere, making them valuable for data assimilation. However, their assimilation has traditionally focused on retrieved profiles. This study implemented the direct assimilation of brightness temperatures from GMWRs with a machine learning-based bias correction scheme. The results show improvements in the low-level atmospheric structure and precipitation predictions.

The ICON climate and numerical weather prediction model was fully ported to Graphical Processing Units (GPUs) using OpenACC compiler directives, covering all components required for operational weather prediction. The GPU port together with several performance optimizations led to a speed-up of 5.6× when comparing to traditional Central Processing Units (CPUs) . Thanks to this adaptation effort, MeteoSwiss became the first national weather service to run the ICON model operationally on GPUs.

The quasi-biennial oscillation (QBO) is the main source of wind fluctuations in the tropical stratosphere, which can couple to surface climate. However, models do a poor job of simulating the QBO in the lower stratosphere, for reasons that remain unclear. One possibility is that models do not completely represent how ozone influences the QBO-associated wind variations. Here we propose a multi-model framework for assessing how ozone influences the QBO in recent past and future climates.

We pushed a global cloud-resolving model to a novel 100 m setup over the San Francisco Bay Area using a regionally refined mesh. The model captured fine-scale air motions over complex terrain and coastal regions at large-eddy scales with comprehensive global modeling configuration, enabled by scale-aware turbulence parameterization. Performance tests demonstrated that graphics processing unit (GPU) acceleration make such high-resolution simulations feasible within practical timeframes.

We developed a high-resolution ocean model to study marine debris dispersion across the Indian Ocean, from coastal scales to the open sea. Results show that model resolution and wave forcing, through both the Stokes-Coriolis force and Stokes drift, play a key role in shaping ocean circulation, seasonal energy budgets, and floating debris trajectories. High-resolution currents and wave forcing, especially during monsoons, increase the anisotropic spread and travel distances of drifting material.

Accounting for carbon and nutrients in rivers is essential for resolving carbon dioxide (CO₂) exchanges between the ocean and the atmosphere. In this study, we add the effect of present-day rivers to a pioneering global-ocean biogeochemistry model. This study highlights the challenge for global ocean numerical models to cover the complexity of the flow of water and carbon across the Land-to-Ocean Aquatic Continuum.

We present a new database of radar-relevant optical properties for a wide range of ice particle shapes, computed using the Discrete Dipole Approximation (DDA) at 5.6, 9.6, 35.6, and 94 GHz. The database is designed to support habit-evolving microphysical schemes, which predict continuous changes in ice particle properties rather than the traditionally assumed fixed categories. It includes over 2600 individual crystals and 450 aggregates with varying riming degrees and morphology.

Hydrogen usage will likely increase to achieve net zero emission targets. We undertook calculations with an Earth system model using a high-performance computer to explore hydrogen atmospheric dynamics. Simulations with our model yielded highly accurate results at global scale. Correctly representing hydroxyl radicals in the model is a critical requirement for predicting hydrogen concentrations well. Our hydrogen budget is also in very good agreement with bottom-up literature estimates.

We developed software to identify sea breezes from weather model output, using three different methods, and applied these to four models for a 6-month period over Australia. We tested each method using case studies and statistics of sea breeze occurrences, finding that a method that identifies atmospheric moisture fronts performs well. Some potential errors are demonstrated due to detection of other frontal systems, but this method could be useful for robustly analyzing sea breezes from models.

We introduce BuRNN (BUrned area modelling by Recurrent Neural Networks), a model which estimates monthly burned area based on satellite observations and climate, vegetation, and socio-economic data using machine learning. BuRNN outperforms existing process-based fire models, which improves our capabilities of modelling past and future burned areas.

This paper presents a new modelling framework that reconstructs the history of rocky shores using cosmogenic nuclide data. The model can be applied to both eroding and stable coasts, and allows users to test how factors like cliff retreat, down-wearing and sea-level change have shaped coastal landscapes. It provides a flexible tool for exploring long-term coastal evolution across diverse environments.

Millions of historical drillholes contain rock descriptions but lack stratigraphic information needed for subsurface modeling. We developed an automated method converting rock descriptions into stratigraphic interpretations by testing plausible sequences using regional maps. The approach quantifies uncertainty and correlates multiple drillholes. Testing on fifty-two South Australian drillholes successfully predicted correct sequences, unlocking legacy data value for geological surveys.

We introduce the first denoising diffusion model for wildfire spread prediction, a new kind of generative AI model that learns to simulate fires not just as one fixed outcome, but as a range of possible scenarios. This allows us to capture the inherent uncertainty of wildfire dynamics. Our model produces ensembles of forecasts that reflect physically meaningful distributions of where fire might go next.