Machine learning (ML) is an increasingly popular tool in the field of weather and climate modelling. While ML has been used in this space for a long time, it is only recently that ML approaches have become competitive with more traditional methods. In this review, we have summarized the use of ML in weather and climate modelling over time; provided an overview of key ML concepts, methodologies, and terms; and suggested promising avenues for further research.

We present an open-source Python framework for performing empirical-statistical downscaling of climate information, such as precipitation. The user-friendly package comprises all the downscaling cycles including data preparation, model selection, training, and evaluation, designed in an efficient and flexible manner, allowing for quick and reproducible downscaling products. The framework would contribute to climate change impact assessments by generating accurate high-resolution climate data.

Surface waves that propagate in oceanic or coastal environments get influenced by their surroundings. Changes in the ambient current or the depth profile affect the wave propagation path, and the change in wave direction is called refraction. Some analytical solutions to the governing equations exist under ideal conditions, but for realistic situations, the equations must be solved numerically. Here we present such a numerical solver under an open-source license.

It is important to know how well atmospheric models do in mountains, but there are not very many weather stations. We evaluate rain and snow from a model from 1987–2020 in the Upper Colorado River basin against the available data. The model works rather well, but there are still some uncertainties in remote locations. We then use snow maps collected by aircraft, streamflow measurements, and some advanced statistics to help identify how well the model works in ways we could not do before.

We present a new activity which aims to organize the analysis of biases in the Canadian Earth System model (CanESM) in a systematic manner. Results of this “Analysis for Development” (A4D) activity includes a new CanESM version, CanESM5.1, which features substantial improvements regarding the simulation of dust and stratospheric temperatures, a second CanESM5.1 variant with reduced climate sensitivity, and insights into potential avenues to reduce various other model biases.

Global climate models are susceptible to spurious trends known as drift. Fortunately, drift can be corrected when analysing data produced by models. To explore the uncertainty associated with drift correction, we develop a new method: Monte Carlo drift correction. For historical simulations of thermosteric sea level rise, drift uncertainty is relatively large. When analysing data susceptible to drift, researchers should consider drift uncertainty.

Kernel density estimators (KDE) approximate the probability density of a data set without the assumption of an underlying distribution. We used the solution of the diffusion equation, and a new approximation of the optimal smoothing parameter build on two pilot estimation steps, to construct such a KDE best suited for typical characteristics of geoscientific data. The resulting KDE is insensitive to noise and well resolves multimodal data structures as well as boundary-close data.

Most current studies on planetary boundary layer (PBL) parameterization schemes are relatively fragmented and lack systematic in-depth analysis and discussion. In this study, we comprehensively evaluate the performance capability of the PBL scheme in five typical regions of China in different seasons from the mechanism of the scheme and the effects of PBL schemes on the near-surface meteorological parameters, vertical structures of the PBL, PBL height, and turbulent diffusion. 

We developed a new modelling framework combining numerical methods with machine learning. Using this approach, we focused on understanding how ice moves within glaciers, and we successfully learnt a prescribed law describing ice movement for 17 glaciers worldwide as a proof of concept. Our framework has the potential to discover important laws governing glacier processes, aiding our understanding of glacier physics and their contribution to water resources and sea-level rise.

The PRIMAVERA project aimed to develop a new generation of advanced global climate models. The large volume of data generated was uploaded to a central analysis facility (CAF) and was analysed by 100 PRIMAVERA scientists there. We describe how the PRIMAVERA project used the CAF's facilities to enable users to analyse this large dataset. We believe that similar, multi-institute, big-data projects could also use a CAF to efficiently share, organise and analyse large volumes of data.

While several studies have shown that mixotrophs play a crucial role in the carbon cycle, the impact of environmental forcings on their dynamics remains poorly investigated. Using a biogeochemical model that considers mixotrophs, we study the impact of light and nutrient concentration on the ecosystem composition in a highly dynamic Mediterranean coastal area: the Bay of Marseille. We show that mixotrophs cope better with oligotrophic conditions compared to strict auto- and heterotrophs.

We present the development of a code to simulate simultaneously the dynamics of landscapes over geological time and the evolution of the concentration of cosmogenic isotopes in grains throughout their transport from the slopes to the river outlets. This new model makes it possible to study the relationship between the detrital signal of cosmogenic isotope concentration measured in sediment and the erosion--deposition processes in watersheds.

This study explores how the variation in the source profiles adopted in chemical transport models (CTMs) impacts the simulated results of chemical components in PM_2.5 based on sensitivity analysis. The impact on PM_2.5 components cannot be ignored, and its influence can be transmitted and linked between components. The representativeness and timeliness of the source profile should be paid adequate attention in air quality simulation.

Vegetation is a critical component of carbon storage in peatlands but an often-overlooked concept in many peatland models. We developed a new model capable of simulating the response of vegetation to changing environments and management regimes. We evaluated the model against observed chamber data collected at two peatland sites. We found that daily air temperature, water level, harvest frequency and height, and vegetation composition drive methane and carbon dioxide emissions.

Atmospheric models benefit from idealized tests that assess their accuracy in a simpler simulation. A new test with artificial mountains is developed for models on a spherical earth. The mountains trigger the development of both planetary-scale and small-scale waves. These can be analyzed in dry or moist environments, with a simple rainfall mechanism. Four atmospheric models are intercompared. This sheds light on the pros and cons of the model design and the impact of mountains on the flow.

In addition to the dominant role of the PBL scheme on the results of the meteorological field, many factors in the model are influenced by large uncertainties. This study focuses on the uncertainties that influence numerical simulation results (including horizontal resolution, vertical resolution, near-surface scheme, initial and boundary conditions, underlying surface update, and update of model version), hoping to provide a reference for scholars conducting research on the model.

In this study, to noticeably improve precipitation simulation in steep mountains, we propose a sub-grid parameterization scheme for the topographic vertical motion in CAM5-SE to revise the original vertical velocity by adding the topographic vertical motion. The dynamic lifting effect of topography is extended from the lowest layer to multiple layers, thus improving the positive deviations of precipitation simulation in high-altitude regions and negative deviations in low-altitude regions.

We present an along BGC-Argo track 1D modelling framework. The model physics is constrained by the BGC-Argo temperature and salinity profiles to reduce the uncertainties related to mixed layer dynamics, allowing the evaluation of the biogeochemical formulation and parameterization. We objectively analyse the model with BGC-Argo and satellite data and improve the model biogeochemical dynamics. We present the framework, example cases and routines for model improvement and implementations.

We propose a discrete multilayer shallow water model based on z-layers which, thanks to the insertion and removal of surface layers, can deal with an arbitrarily large tidal oscillation independently of the vertical resolution. The algorithm is based on a two-step procedure used in numerical simulations with moving boundaries (grid movement followed by a grid topology change, that is, the insertion/removal of surface layers), which avoids the appearance of very thin surface layers.