Finite-element methods (FEMs) permit the use of more flexible unstructured meshes but are rarely used in full waveform inversions (FWIs), an iterative process that reconstructs velocity models of earth’s subsurface, due to computational and memory storage costs. To reduce those costs, novel software is presented allowing the use of high-order mass-lumped FEMs on triangular meshes, together with a material-property mesh-adaptation performance-enhancing strategy, enabling its use in FWIs.

We bring the state-of-the-science chemistry module GEOS-Chem into the Community Earth System Model (CESM). We show that some known differences between results from GEOS-Chem and CESM's CAM-chem chemistry module may be due to the configuration of model meteorology rather than inherent differences in the model chemistry. This is a significant step towards a truly modular Earth system model and allows two strong but currently separate research communities to benefit from each other's advances.

More accurate predictions of the Gulf's ocean dynamics are needed. We investigate the impact on the predictive skills of a numerical shelf sea model of the Gulf after changing a few key aspects. Increasing the lateral and vertical resolution and optimising the vertical coordinate system to best represent the leading physical processes at stake significantly improve the accuracy of the simulated dynamics. Additional work may be needed to get real benefit from using a more realistic bathymetry.

Atmospheric chemistry models must be able to operate both online as components of Earth system models and offline as standalone models. The widely used GEOS-Chem model operates both online and offline, but the classic offline version is not suitable for massively parallel simulations. We describe a new generation of the offline high-performance GEOS-Chem (GCHP) that enables high-resolution simulations on thousands of cores, including on the cloud, with improved access, performance, and accuracy.

Knowledge of the internal structures of the major continental ice sheets is improving, thanks to new investigative techniques. These structures are an essential indication of the flow behavior and dynamics of ice transport, which in turn is important for understanding the actual impact of the vast amounts of water trapped in continental ice sheets on global sea-level rise. The software studied here is specifically designed to simulate such structures and their evolution.

A lack of inventory data can be a limiting factor in developing landslide predictive models, which are crucial for supporting hazard policy and decision-making. We show how case-based reasoning and domain adaptation (transfer-learning techniques) can effectively retrieve similar landslide modeling situations for prediction in new data-scarce areas. Using cases in Italy, Austria, and Ecuador, our findings support the application of transfer learning for areas that require rapid model development.

A procedure combining numerical simulations, remote sensing, and statistical analyses is developed to detect large-scale current systems in large lakes. By applying this novel procedure in Lake Geneva, strategies for detailed transect field studies of the gyres and eddies were developed. Unambiguous field evidence of 3D gyre/eddy structures in full agreement with predictions confirmed the robustness of the proposed procedure.

In this study, we implement a hardening mortality scheme into CTSM5.0-FATES-Hydro and evaluate how it impacts plant hydraulics and vegetation growth. Our work shows that the hydraulic modifications prescribed by the hardening scheme are necessary to model realistic vegetation growth in cold climates, in contrast to the default model that simulates almost nonexistent and declining vegetation due to abnormally large water loss through the roots.

We developed a new simple climate model designed to fill a perceived gap within the existing simple climate models by fulfilling three key requirements: calibration using Bayesian inference, the possibility of coupling with integrated assessment models, and the capacity to explore climate scenarios compatible with limiting climate impacts. Here, we describe the model and its calibration using the latest data from complex CMIP6 models and the IPCC AR6, and we assess its performance.

A local data assimilation scheme (Local DA v1.0) was proposed to leverage the advantage of hybrid covariance, multiscale localization, and parallel computation. The Local DA can perform covariance localization in model space, observation space, or both spaces. The Local DA that used the hybrid covariance and double-space localization produced the lowest analysis and forecast errors among all observing system simulation experiments.

The performance of a chemical transport model in reproducing PM_2.5 concentrations and composition was evaluated at the finest scale using measurements from regulatory sites as well as a network of low-cost monitors. Total PM_2.5 mass is reproduced well by the model during the winter when compared to regulatory measurements, but in the summer PM_2.5 is underpredicted, mainly due to difficulties in reproducing regional secondary organic aerosol levels.

We examine the effects of spatially aggregated upstream information as input for a deep learning model forecasting near-surface ozone levels. Using aggregated data from one upstream sector (45°) improves the forecast by ~ 10 % for 4 prediction days. Three upstream sectors improve the forecasts by ~ 14 % on the first 2 d only. Our results serve as an orientation for other researchers or environmental agencies focusing on pointwise time-series predictions, for example, due to regulatory purposes.

Inspired by the success of deep learning in various domains, we test the applicability of video prediction methods by generative adversarial network (GAN)-based deep learning to predict the 2 m temperature over Europe. Our video prediction models have skill in predicting the diurnal cycle of 2 m temperature up to 12 h ahead. Complemented by probing the relevance of several model parameters, this study confirms the potential of deep learning in meteorological forecasting applications.