Short summary

Mesoscale Convective Systems cause extreme weather and flash floods, yet they remain difficult to simulate in climate models. We developed the Evolution-based Mesoscale Convective System Model Assessment tool to identify these storms using standard model data. Our 27-year record for Europe shows these systems drive over 60 percent of heavy hourly rain. This benchmark allows us to evaluate climate model performance and investigate how these intense storms will change in a warming climate.

Short summary

This study introduces a verification method that accounts for differences in grid resolution when evaluating extreme event forecasts. We apply it to an artificial intelligence-based weather prediction model and a high-resolution numerical weather prediction model. Results show that, when assessed on equivalent neighborhood scales, the high resolution numerical weather prediction model only outperforms the AI system for short lead times in predicting extreme precipitation.

Short summary

Drought is reshaping Europe's forests. Using an uncalibrated process-based model across 149 German sites, we identified key drivers of tree mortality in European beech and Norway spruce forests. Our model captured both the timing and extent of mortality. A new bark beetle module improved predictions for spruce. High soil water capacity and heterogeneous soils reduced drought impacts. These findings offer new insights to anticipate forest responses in a warming, drying climate.

Short summary

Tropical cyclones come in various size and shape, which smoothes out key storm features in composite analyses. To address this, we developed a compositing method that symmetrizes storms and better aligns their eyewalls and horizontal extents prior to compositing. This approach preserves small-scale features in the composites, reduces within-group variance, and enhances the power of statistical testing. The method facilitates the investigation and understanding of tropical cyclone development.

Short summary

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.

Short summary

Climate systems can undergo abrupt, potentially irreversible changes with major impacts on ecosystems and societies, yet consistent tools to detect these transitions across different models are lacking. We present an open-source software package for systematically detecting where and when such changes occur in climate simulations and quantifying variation in transition timing. This enables robust comparison of abrupt changes across models and contributes to assessing climate-tipping risks.

Short summary

The LocalWeatherSampler generates 20–30 year weather scenarios at daily resolution using historical weather data. Wet/dry years can be defined by threshold or via the Standardized Precipitation Index and future weather sequences can be generated tailored to specific scenarios, like extremely dry or very wet sequences. This approach enables testing and analyzing precipitation patterns and temperature trends with models that rely on realistic, daily weather data, such as soil and crop models.

Short summary

This study presents a novel scheme for simulating Earth's free surface. Traditional methods like 'Sticky Air' face limitations such as increased computational costs and marker fluctuation issues. Our approach integrates the 'Sticky Air' concept into an Arbitrary Lagrangian–Eulerian framework using an internal boundary enabling a true free surface simulation, which reduces marker noise, enhances numerical stability.

Short summary

This study introduces the Modified Spatial Efficiency metric to more rigorously evaluate how well climate models reproduce observed spatial patterns, addressing a long-standing challenge in model assessment. It demonstrates robust performance across a wide range of conditions, capturing spatial structures in an intuitive and physically meaningful way. This new metric offers researchers an improved tool for evaluating and inter-comparing climate models.

Short summary

Complex models in environmental science typically have a lot of tuning parameters, which has to be set by the users depending on the application. This study presents a new method of objective tuning of a huge number of parameters, by combining expert judgement with automated tuning (LiMMo). The method is successfully applied to the regional climate model ICON-CLM over Europe.

Short summary

Deep learning models for Earth system prediction are often criticized as "black boxes" that lack physical interpretability. This study introduces a "glass box" dissection framework to analyze the internal logic of these systems. Using the OceanCastNet wave model, we demonstrate that the AI autonomously organizes its computations into modules analogous to physical source terms (wind input, dissipation, and propagation), proving that data-driven models can spontaneously learn physical laws.

Short summary

Climate change affect lakes by including not just the lakes themselves but also the land areas that drain into them. These surrounding areas influence how much water and nutrients flow into lakes which in turn impact water quality. Here, water fluxes from land, derived from a global hydrological model where water fluxes are modelled at the grid scale, were used to estimate streamflow inputs to lakes from their catchments. Using data from 70 Swedish lakes, we showed that our method works well.

Short summary

Can we be sure that different computing environments, that should not change the model climate, indeed leave the climate unaltered? In this article, we present a novel methodology that answers whether two model climates are statistically the same. Besides a new methodology, able to detect significant differences between two model climates 60 % more accurately than a similar recent state-of-the-art method, we also provide an analysis on what actually constitutes a different climate.

Short summary

Machine learning is increasingly used in environmental sciences for spatial predictions, but its effectiveness is challenged when models are applied beyond the areas they were trained on. We propose a Local Training Data Point Density (LPD) approach that considers how well a model's environment is represented by training data. This method provides a valuable tool for evaluating model applicability and uncertainties, crucial for broader scientific and practical applications.

Short summary

This study proposes a new method to improve climate simulation evaluation, tackling a key model error: selecting the best parameter combinations. Our "dynamic weighting" method automatically gives more importance to hard-to-predict variables, like precipitation and wind speed. When tested in two distinct climate regions, our approach identified model settings that produced more accurate and reliable forecasts than traditional equal-weighting methods, performing well in extreme weather years.

Short summary

We present the Potsdam Integrated Assessment Modeling validation tool, piamValidation, an open-source R package for validating climate scenarios. The tool enables comparison of model outputs with historical data, feasibility constraints and alternative scenarios. Designed as a community resource, validation configuration files can serve as a knowledge-sharing platform. The main objective is to improve the credibility of Integrated Assessment Models by promoting standardized validation practices.

Short summary

The Gregory method is a common approach for calculating the equilibrium climate sensitivity (ECS). However, studies which apply this method lack transparency in how model data is processed prior to calculating the ECS, inhibiting replicability. Different choices of global weighting, net radiative flux variable, anomaly calculation, and linear regression fit can affect the ECS estimates. We investigate the impact of these choices and propose a standardised method for future ECS calculations.

Short summary

Earth system models consist of many components, coupled in time and space. Standard coupling algorithms introduce a numerical error, which one can compute with iterative coupling methods. We use such a method for a coupled model of a single vertical column of the atmosphere, ocean, and sea ice. We find that coupling errors in the atmosphere and at the ice surface can be substantial and that discontinuous physics parameterizations lead to convergence issues of the iteration.

Short summary

Short summary

To understand how well current Earth system models simulate the natural world, we compared the models' outputs against measurements from satellites. Our results show these models struggle to accurately capture trends in variables related to carbon cycle, because the models can’t respond to human and environmental influences. This evaluation is crucial because improving these models will lead to more reliable forecasts of how ecosystems and the climate will change in the future.

Short summary

This work builds upon and extends previous multi-model ensemble studies by introducing four meta-modelling approaches to predict ecosystem-scale C fluxes. Our results show that meta-models consistently outperform both the multi-model median and the best individual process-based models, improving explained variance by up to 38.5 % and substantially reducing bias, even for challenging fluxes such as total ecosystem respiration and net ecosystem exchange.

Short summary

This study introduces a timescale-aware bias-correction framework to enhance Earth system model assessments, vital for the geoscience community. By decomposing model outputs into oscillatory components, we preserve critical information across various timescales, ensuring more reliable projections. This improved reliability supports strategic decisions in sectors such as agriculture, water resources, and disaster preparedness.

Short summary

This study assessed three methods for correcting daily precipitation data: Quantile Delta Mapping, Empirical Quantile Mapping (EQM), and Detrended Quantile Mapping (DQM) using 11 GCMs. EQM performed best overall, offering reliable corrections and lower uncertainty. The best bias correction method for each grid is selected differently depending on the weighting case. The best bias correction method can vary depending on factors such as climate and terrain.

Short summary

The diagnostics package is an open-source Python software package used for evaluating the Madden–Julian Oscillation teleconnections to the extratropics, as predicted by subseasonal-to-seasonal (S2S) forecast systems.

Short summary

Present study revisits model that represent urban effects in the Weather Research & Forecasting model. We propose methods to identify evaluable parameters via field measurements and found inconsistencies between physics and its code implementation. Simulations reveal small errors can significantly impact outputs.

Short summary

This study investigates changes in the Southwestern Atlantic Ocean with a high-resolution ocean model. Particularly in the Brazil-Malvinas Confluence (BMC), it finds that the southward movement of the BMC, induced by the warming trends in the region, is balanced by northward flow from the Malvinas Current and Pacific Waves, affecting the Atlantic. The results also comment on disparities observed in the simulation, especially concerning the North Brazil Current, which impacts its evolution.

Short summary

Forecast verification is essential for improving weather prediction models but faces challenges with traditionally used metrics. New spatial verification metrics like the Fraction Skill Score (FSS) perform better but are difficult to use in a global domain due to large computational cost. We introduce two new global smoothing methodologies that can be used with smoothing-based metrics in a global domain. We demonstrate their effectiveness with an analysis of global precipitation forecasts.

Short summary

Artificial intelligence (AI) is playing a bigger role in weather forecasting, often competing with physical models. However, combining AI models with data assimilation, a process that improves weather forecasts by incorporating observation data, is still relatively unexplored. This study explored the coupling of ensemble data assimilation with an AI weather prediction model, ClimaX, which succeeded in employing weather forecasts stably by applying techniques conventionally used for physical models.

Short summary

Here, we implement a basal ice shelf melt module (PICO – Postdam Ice-shelf Cavity mOdel) in an ice sheet model (GRISLI) and test six simple statistical methods to calibrate this module. We show that calculating the mean absolute error of bins best fits the observational datasets under multiple conditions and without using temperature corrections. Additionally, we show that calibration at a smaller scale than all Antarctic ice shelves is not needed. Finally, we assess the impact with future projections until 2300.

Short summary

Chemistry transport models (CTMs) are critical in environmental assessments, but they are computationally expensive and thus often not directly used to support decision-making. We evaluate the use of data-driven model discovery and model reduction methods to act as reduced-order models for CTM simulations, and show that they can reconstruct and forecast changes in the global ozone distribution for up to several years at a fraction of the cost of a CTM while also being more accessible.

Short summary

Accurately predicting long-term ocean temperatures is vital for climate science. We developed a new model that integrates multiple environmental variables using a novel spatiotemporal framework. Tests demonstrated consistent improvement over baseline models, delivering more accurate monthly temperature predictions up to a decade in advance.

Short summary

We developed a novel methodology for assessing whether a new physics parameterization should be added to a climate model based on its effect across a hierarchy of model dynamical constraints. Our study used this model hierarchy to evaluate the effect of a new cloud radiation parameterization on longwave radiation and determined that the parameterization should be added to climate radiation models, but its effect is not large enough to be a priority.

Short summary

Air pollution harms health and society. Understanding and predicting it is crucial. Various models have been developed to model air pollution. However, the consistency exhibited by a model in different areas is commonly neglected. Our study accounts for this and shows lower accuracy near busy roads but higher accuracy in less populated areas. Considering location characteristics in air pollution predictions is important in comparing statistical models and understanding the health–society–space relationship.

Short summary

We propose an approach to obtain weights for calculating averages of variables from Earth system models (ESM) based on concepts from information theory. It quantifies a relative distance between model output and reality, even though it is impossible to know the absolute distance from model predictions to reality. The relative ranking among models is based on concepts of model selection and multi-model averages previously developed for simple statistical models, but adapted here for ESMs.

Short summary

A new method called coarse-graining scale analysis is gaining traction as an alternative to Fourier analysis. However, it requires data to be on a regular grid. To address this, we present a high-performance Python package of the coarse-graining technique using discrete Laplacians. This method can handle any mesh type and is ideal for processing output directly from unstructured-mesh models. Computation is split into preparation and solving phases, with GPU acceleration ensuring fast processing.

Short summary

Land surface models are used for simulating how Earth's surface interacts with the atmosphere. As models grow more complex and detailed, researchers need better tools to evaluate their performance. OpenBench, a new software system that makes the evaluation process more comprehensive and efficient, stands out by incorporating various factors and working with data at any scale, enabling scientists to incorporate new types of models and measurements as our understanding of Earth's systems evolves.

Short summary

We present the new Multi-Model Large Ensemble Archive (MMLEAv2) and introduce the newly updated Climate Variability Diagnostics Package version 6 (CVDPv6), which is designed specifically for use with large ensembles. For highly variable quantities, we demonstrate that a model might perform evaluation poorly or favourably compared to the single realisation of the world that the observations represent, highlighting the need for large ensembles for model evaluation.

Short summary

We use total freeboard data from NASA’s Ice, Cloud and land Elevation Satellite-2 (ICESat-2) across both hemispheres and estimates of winter Arctic sea ice thickness to evaluate climate model simulations of sea ice, providing constraints beyond the traditional sea ice area metric. ICESat-2 provides accurate freeboard data, but its short observational record requires careful consideration of natural variability.

Short summary

This paper presents a flexible workflow using a unified data assimilation framework to evaluate atmospheric composition models. It enables comparison of observations with forecasts of trace gases and aerosols from different models. The system is consistent and adaptable, reducing repetitive work, supporting model validation and observation assessment, and aligning evaluation with operational data assimilation for research and practical applications.

Short summary

Magmatic dykes transport magma to the Earth's surface, sometimes causing eruptions. We advanced a model of dyking, treating it as plastic deformation in a porous medium, unlike the classic model that treats dykes as fractures in elastic solids. Comparing the two, we found the plastic model aligns with the fracture model in dyke speed and energy consumption, despite quantitative differences. This new method could be a powerful tool for understanding volcanic processes during tectonic activity.

Short summary

This study introduces a new method that helps improve the accuracy of climate models by automatically selecting the best parameters to match real-world observations. Instead of manually adjusting many parameters, the method uses a mathematical tool to find the most appropriate settings for the model. It can be especially helpful for researchers who introduce new physical parameters into climate models to assess their impact on model results and optimize them along with the old ones.

Short summary

Volatile organic compounds (VOCs) affect air quality and climate, but their behavior in the atmosphere is still uncertain. We launched a global research effort to compare how different models represent these compounds and to improve their accuracy. By analyzing model results alongside observations and satellite data, we aim to better understand the atmospheric composition of these compounds.

Short summary

With the growing use of machine learning for groundwater level (GWL) prediction, proper performance estimation is crucial. This study compares three validation strategies—blocked cross-validation (bl-CV), repeated out-of-sample (repOOS), and out-of-sample (OOS)—for 1D-CNN models using meteorological inputs. Results show that bl-CV offers the most reliable performance estimates, while OOS is the most uncertain, highlighting the need for careful method selection.

Short summary

This study presents a method to identify climate drivers of an impact, such as agricultural yield failure, from high-resolution weather data. The approach systematically generates, selects and combines predictors that generalise across different environments. Tested on crop model simulations, the identified drivers are used to create parsimonious models that achieve high predictive performance over long time horizons, offering a more interpretable alternative to black-box models.