WAVETRISK-2.1 is an innovative climate model for the world's oceans. It uses state-of-the-art techniques to change the model's resolution locally, from O(100 km) to O(5 km), as the ocean changes. This dynamic adaptivity makes optimal use of available supercomputer resources, and allows two-dimensional global scales and three-dimensional submesoscales to be captured in the same simulation. WAVETRISK-2.1 is designed to be coupled its companion global atmosphere model, WAVETRISK-1.x.

The ocean current flowing along the southeastern coast of Australia is called the East Australian Current (EAC). Using computer simulations, we tested how surface and subsurface observations might improve models of the EAC. Subsurface observations are particularly important for improving simulations, and if made in the correct location and time, can have impact 600 km upstream. The stability of the current affects model estimates could be capitalized upon in future observing strategies.

We present an idealized ocean model configuration and a set of simulations performed using varying horizontal grid spacing. While the model domain is idealized, it resembles important geometric features of the Atlantic and Southern oceans. The simulations described here serve as a framework to effectively study mesoscale eddy dynamics, to investigate the effect of mesoscale eddies on the large-scale dynamics, and to test and evaluate eddy parameterizations.

A new karst hydrological model (the QMG model) is developed to simulate and predict the floods in karst trough valley basins. Unlike the complex structure and parameters of current karst groundwater models, this model has a simple double-layered structure with few parameters and decreases the demand for modeling data in karst areas. The flood simulation results based on the QMG model of the Qingmuguan karst trough valley basin are satisfactory, indicating the suitability of the model simulation.

Coupled Earth system models require transfer of field data between multiple components with varying spatial resolutions to determine the correct climate behavior. We present the Metrics for Intercomparison of Remapping Algorithms (MIRA) protocol to evaluate the accuracy, conservation properties, monotonicity, and local feature preservation of four different remapper algorithms for various unstructured mesh problems of interest. Future extensions to more practical use cases are also discussed.

Parameter optimization can improve the accuracy of modeled carbon fluxes. Few studies conducted pixel-level parameterization because it requires a high computational cost. Our paper used high-quality spatial products to optimize parameters at the pixel level, and also used the machine learning method to improve the speed of optimization. The results showed that there was significant spatial variability of parameters and we also improved the spatial pattern of carbon fluxes.

Application of 1-D lake models coupled within earth-system prediction models will improve accuracy but requires accurate initialization of lake temperatures. Here, we describe a lake initialization method by cycling within a weather prediction model to constrain lake temperature evolution. We compared these lake temperature values with other estimates and found much reduced errors (down to 1-2 K). The lake cycling initialization is now applied to two operational US NOAA weather models.

This work demonstrates the use of modern machine learning techniques to enhance the resolution of atmospheric chemistry simulations. We evaluate the schemes for an 8 x 10 increase in resolution and find that they perform substantially better than conventional methods. Methods are introduced to target machine learning methods towards this type of problem, most notably by ensuring they do not break known physical constraints.

To understand the scientific consequence of perturbations caused by slave cores in heterogeneous computing environments, we examine the influence of perturbation amplitudes on the determination of the cloud bottom and cloud top and compute the probability density function (PDF) of generated clouds. A series of comparisons of the PDFs between homogeneous and heterogeneous systems show consistently acceptable error tolerances when using slave cores in heterogeneous computing environments.

We present a number of augmentations to the ecosystem model LPJ-GUESS, which will allow us to use it in studies of the interactions between the land biosphere and the climate. The new module enables calculation of fluxes of energy and water into the atmosphere that are consistent with the modelled vegetation processes. The modelled fluxes are in fair agreement with observations across 21 sites from the FLUXNET network.

Deep neural networks are used to produce downscaled regional climate change projections over Europe for temperature and precipitation for the first time. The resulting dataset, DeepESD, is analyzed against state-of-the-art downscaling methodologies, reproducing more accurately the observed climate in the historical period and showing plausible future climate change signals with low computational requirements.

When studying tropical cyclones in a large dataset, one needs objective and automatic procedures to detect their specific pattern. Applying four different such algorithms to a reconstruction of the climate, we show that the choice of the algorithm is crucial to the climatology obtained. Mainly, the algorithms differ in their sensitivity to weak storms so that they provide different frequencies and durations. We review the different options to consider for the choice of the tracking methodology.

This study investigates the nudging implementation in the EAMv1 model. We find that (1) revising the sequence of calculations and using higher-frequency constraining data to improve the performance of a simulation nudged to EAMv1’s own meteorology, (2) using the relocated nudging tendency and 3-hourly ERA5 reanalysis to obtain a better agreement between nudged simulations and observations, and (3) using wind-only nudging are recommended for the estimates of global mean aerosol effects.

Terrain horizon and sky view factor are crucial quantities for many geoscientific applications; e.g. they are used to account for effects of terrain on surface radiation in climate and land surface models. Because typical terrain horizon algorithms are inefficient for high-resolution (< 30 m) elevation data, we developed a new algorithm based on a ray-tracing library. A comparison with two conventional methods revealed both its high performance and its accuracy for complex terrain.

We present an implicit modeling method based on deep learning to produce a geologically valid and structurally compatible model from unevenly sampled structural data. Trained with automatically generated synthetic data with realistic features, our network can efficiently model geological structures without the need to solve large systems of mathematical equations, opening new opportunities for further leveraging deep learning to improve modeling capacity in many Earth science applications.

Our study provided a detailed description and a package of an individual tree-based carbon model, FORCCHN2. This model used non-structural carbohydrate (NSC) pools to couple tree growth and phenology. The model could reproduce daily carbon fluxes across Northern Hemisphere forests. Given the potential importance of the application of this model, there is substantial scope for using FORCCHN2 in fields as diverse as forest ecology, climate change, and carbon estimation.

The sea surface temperature (SST) is highly influenced by the transfer of energy driven by turbulent air–sea fluxes (TASFs). In the NEMO ocean general circulation model, TASFs are computed by means of bulk formulas. Bulk formulas require the choice of a given bulk parameterization, which influences the magnitudes of the TASFs. Our results show that parameterization-related SST differences are primarily sensitive to the wind stress differences across parameterizations.

A prototype data assimilation system for NOAA’s next-generation rapidly updated, convection-allowing forecast system, or Rapid Refresh Forecast System (RRFS) v0.1, is tested and evaluated. The impact of using data assimilation with a convective storm case study is examined. Although the convection in RRFS tends to be overestimated in intensity and underestimated in extent, the use of data assimilation proves to be crucial to improve short-term forecasts of storms and precipitation.

The traditional Carnegie–Ames–Stanford Approach (CASA) model driven by multisource data such as meteorology, soil, and remote sensing (RS) has notable disadvantages. We drove the CASA using RS data and conducted a case study of the Qinghai Lake basin alpine grassland. The simulated result is similar to published and measured net primary productivity (NPP). It may provide a reference for simulating vegetation NPP to satisfy the requirements of accounting carbon stocks and other applications.