Formulating short-term precipitation forecasting as a video prediction task, a novel deep learning architecture (convolutional long short-term memory generative adversarial network, CLGAN) is proposed. A benchmark dataset is built on minute-level precipitation measurements. Results show that with the GAN component the model generates predictions sharing statistical properties with observations, resulting in it outperforming the baseline in dichotomous and spatial scores for heavy precipitation.

PyFLEXTRKR is a flexible atmospheric feature tracking framework with specific capabilities to track convective clouds from a variety of observations and model simulations. The package has a collection of multi-object identification algorithms and has been optimized for large datasets. This paper describes the algorithms and demonstrates applications for tracking deep convective cells and mesoscale convective systems from observations and model simulations at a wide range of scales.

We develop a generalized spatial autoregressive neural network model used for three-dimensional spatial interpolation. Taking the different changing trend of geographic elements along various directions into consideration, the model defines spatial distance in a generalized way and integrates it into the process of spatial interpolation with the theories of spatial autoregression and neural network. Compared with traditional methods, the model achieves better performance and is more adaptable.

FEOTS aims to enhance the value of data produced by state-of-the-art climate models by providing a framework to diagnose and use ocean transport operators for offline passive tracer simulations. We show that we can capture ocean transport operators from a validated climate model and employ these operators to estimate water mass budgets in an offline regional simulation, using a small fraction of the compute resources required to run a full climate simulation.

An accurate description of the wind energy input into ocean waves is crucial to ocean wave modeling, and a physics-based consideration of the effect of wave breaking is absolutely necessary to obtain such an accurate description, particularly under extreme conditions. This study evaluates the performance of a recently improved formula, taking into account not only the effect of breaking but also the effect of airflow separation on the leeside of steep wave crests in a reasonably consistent way.

C-Coupler3.0 is an integrated coupler infrastructure with new features, i.e. a series of parallel-optimization technologies, a common halo-exchange library, a common module-integration framework, a common framework for conveniently developing a weakly coupled ensemble data assimilation system, and a common framework for flexibly inputting and outputting fields in parallel. It is able to handle coupling under much finer resolutions (e.g. more than 100 million horizontal grid cells).

Internal solitary waves (ISWs) play crucial roles in mass transport and ocean mixing in the northern South China Sea. Massive numerical investigations have been conducted in this region, but there was no systematic evaluation of a three-dimensional model about precisely simulating ISWs. Here, an ISW forecasting model is employed to evaluate the roles of resolution, tidal forcing and stratification in accurately reproducing wave properties via comparison to field and remote-sensing observations.

We implemented a halogen volcanic chemistry scheme in a one-dimensional modelling framework preparing for further use in a three-dimensional global chemistry-transport model. The results of the simulations for an eruption of Mt Etna in 2008, including various sensitivity tests, show a good consistency with previous modelling studies.

A mathematical method known as common EOFs is not widely used within the climate research community, but it offers innovative ways of evaluating climate models. We show how common EOFs can be used to evaluate large ensembles of global climate model simulations and distill information about their ability to reproduce salient features of the regional climate. We can say that they represent a kind of machine learning (ML) for dealing with big data.

In this study, we developed a novel modeling system called iHydroSlide3D v1.0 by coupling a modified a 3D landslide model with a distributed hydrology model. The model is able to apply flexibly different simulating resolutions for hydrological and slope stability submodules and gain a high computational efficiency through parallel computation. The test results in the Yuehe River basin, China, show a good predicative capability for cascading flood–landslide events.

This study introduces an improved method of the Globally Resolved Energy Balance (GREB) model by the Bayesian network. The improved method constructs a coarse–fine structure that combines a dynamical model with a statistical model based on employing the GREB model as the global framework and utilizing Bayesian networks as the local optimization. The results show that the improved model has better applicability and stability on a global scale and maintains good robustness on the timescale.

A new method to simulate deterministic ice nucleation processes based on the differential activated fraction was evaluated against a cumulative approach. Box model simulations of heterogeneous-only ice nucleation within cirrus suggest that the latter approach likely underpredicts the ice crystal number concentration. Longer simulations with a GCM show that choosing between these two approaches impacts ice nucleation competition within cirrus but leads to small and insignificant climate effects.

The weather and climate physics suites used in GRIST-A22.7.28 are compared using single-column modeling. The source of their discrepancies in terms of modeling cloud and precipitation is explored. Convective parameterization is found to be a key factor responsible for the differences. The two suites also have intrinsic differences in the interaction between microphysics and other processes, resulting in different cloud features and time step sensitivities.

Using outputs of global biogeochemical ocean model and machine learning methods, we demonstrate that it will be possible to identify linkages between surface environmental and ecosystem structure and the export of carbon to depth by sinking organic particles using real observations. It will be possible to use this knowledge to improve both our understanding of ecosystem dynamics and of their functional representation within models.

Desert dust has significant impacts on climate, public health, infrastructure and ecosystems. An impact assessment requires numerical predictions, which are challenging because the dust emissions are not well known. We present a novel approach using satellite observations and machine learning to more accurately estimate the emissions and to improve the model simulations.

We document details of the regional climate downscaling dataset produced by a global variable-resolution model. The experiment is unique in that it follows a standard protocol designed for coordinated experiments of regional models. We found negligible influence of post-processing on statistical analysis, importance of simulation quality outside of the target region, and computational challenges that our model code faced due to rapidly changing super computer systems.

The emissions of volatile organic compounds from vegetation (BVOCs) influence atmospheric composition and contribute to certain gases and aerosols (tiny airborne particles) which play a role in climate change. BVOC emissions are likely to change in the future due to changes in climate and land use. Therefore, accurate simulation of BVOC emission is important, and this study describes an update to the simulation of BVOC emissions in the United Kingdom Earth System Model (UKESM).

This study conducts the first global evaluation of the latest generation of global climate models to simulate a set of fire weather indicators from the Canadian Fire Weather Index System. Models are shown to perform relatively strongly at the global scale, but they show substantial regional and seasonal differences. The results demonstrate the value of model evaluation and selection in producing reliable fire danger projections, ultimately to support decision-making and forest management.