Despite advances in high performance computing, accurate numerical simulations of global atmospheric dynamics remain a challenge. The resolution required to fully resolve the vast range scales as well as the strong coupling with—often not fully-understood—physics renders such simulations computationally infeasible over time horizons relevant for long-term climate risk assessment. While data-driven parameterizations have shown some promise of alleviating these obstacles, the scarcity of high-quality training data and their lack of long-term stability typically hinders their ability to capture the risk of rare extreme events. In this work we present a general strategy for training variational (probabilistic) neural network models to non-intrusively correct under-resolved long-time simulations of turbulent climate systems. The approach is based on the paradigm introduced by Barthel Sorensen et al. (2024, https://doi.org/10.1029/2023ms004122) which involves training a post-processing correction operator on under-resolved simulations nudged toward a high-fidelity reference. Our variational framework enables us to learn the dynamics of the underlying system from very little training data and thus drastically improve the extrapolation capabilities of the previous deterministic state-of-the art—even when the statistics of that training data are far from converged. We investigate and compare three recently introduced variational network architectures and illustrate the benefits of our approach on an anisotropic quasi-geostrophic flow. For this prototype model our approach is able to not only accurately capture global statistics, but also the anistropic regional variation and the statistics of multiple extreme event metrics—demonstrating significant improvement over previously introduced deterministic architectures.
Land surface models (LSMs) are essential tools for simulating the coupled climate system, representing the dynamics of water, energy, and carbon fluxes on land and their interaction with the atmosphere. However, parameterizing sub-grid processes at the scales relevant to climate models (10–100 km) remains a considerable challenge. The parameterizations typically have a large number of unknown and often correlated parameters, making calibration and uncertainty quantification difficult. Moreover, many existing LSMs are not readily adaptable to the incorporation of modern machine learning (ML) parameterizations trained with in situ and satellite data. This article presents the first version of ClimaLand, a new LSM designed for overcoming these limitations, including a description of the core equations underlying the model, the results of an extensive set of validation exercises, and an assessment of the computational performance of the model. We show that ClimaLand can leverage graphics processing units for computational efficiency, and that its modular architecture and high-level programming language, Julia, allows for integration with ML libraries. In the future, this will enable efficient simulation, calibration, and uncertainty quantification with ClimaLand.
Langmuir turbulence affects turbulent mixing in the ocean boundary layers and its effects require parameterizations in ocean circulation models. Most existing Langmuir turbulence parameterizations focus on the surface boundary layer in open oceans. In the shallow waters of coastal oceans, a surface boundary layer may interact and even merge with a bottom boundary layer. It is unclear how existing Langmuir turbulence parameterizations perform under such complex conditions. Here we assess the performance of two recent Langmuir turbulence parameterizations in an idealized case of merging boundary layers against turbulence-resolving large-eddy simulations (LES). In addition to assessing the solutions of free runs of single-column model (SCM) simulations, in which errors in the mean fields and turbulent fluxes are entangled, we also compare the simulated turbulent fluxes in SCM simulations with their mean fields nudged to those of the LES. In doing so, we focus on the parameterized turbulent fluxes in different parameterizations given the perfect mean fields. Our comparison highlights the tendency of parameterizations to deviate from the LES at each time instance, and thereby reveals the deficiencies of parameterizations in an instantaneous sense. It is shown that both parameterizations overestimate the near-bottom turbulent momentum flux when velocity shear is correct, resulting in too weak near-bottom shear in a free run. Consistent with previous studies, a down-Stokes drift shear momentum flux is necessary for capturing the momentum flux due to Langmuir turbulence but still misses the nonlocal momentum flux when coherent Langmuir supercells form.
National Centers for Environmental Prediction turbulent kinetic energy (TKE)-based eddy-diffusivity mass-flux (EDMF) scheme is implemented in Geophysical Fluid Dynamics Laboratory atmospheric model (AM4.0) for improving the physical consistency of subgrid-scale planetary boundary layer (PBL) turbulence parameterization. The mass flux (MF) component represents vertically coherent convective structures responsible for countergradient transport in the upper PBL, which the original AM4.0's ED-only scheme cannot represent. Consequently, AM4.0 with EDMF produces a deeper and more well-mixed PBL, leading to better zonal-mean vertical temperature and humidity profiles and reduced near-surface wet bias over subtropical and midlatitude oceans. Other model performance changes are generally minor, such as similar biases in global top of atmosphere (TOA) net radiation and shortwave cloud radiative effects, small and compensating changes in low cloud amount and cloud liquid water path, improved low-level equatorial easterlies but deteriorated extratropical westerlies, slightly increased global-mean precipitation, and weaker TOA radiative response to uniform sea surface warming. Three adaptations of EDMF are important for its performance at AM4.0's relatively coarse vertical resolution: limiting the overshoot of MF updraft above PBL-top, reducing the ED-induced mixing across PBL-top, and disabling the MF transport of TKE. Low clouds and their radiative effects are also sensitive to four EDMF parameters that control the ED in the lower and upper PBL respectively, the TKE dissipation rate, and the lateral entrainment of MF updraft and downdraft. An automatic linear tuning of these parameters slightly improves the radiative bias, especially for the coastal stratocumulus. More substantial improvements likely require formulation updates of the EDMF scheme and its coupling with other AM4.0 model components.
Forward- and backward-in-time Lagrangian advection, used to determine fate and origin of material in the ocean, are mathematically consistent. However, their numerical computations are hampered by round-off and truncation errors. Trajectory calculations are stable to errors (i.e., errors are dampened) in zones of velocity convergence and unstable (errors are amplified) in regions of divergence. The stability to errors thus flips when time integration is reversed, which, depending on the numerical configuration, can lead to significant discrepancies between forward- and backward-in-time trajectories. As divergence statistics can be asymmetrical and may be inhomogeneously distributed in space, this can lead to what we call the “stability bias.” Using representative numerical set-ups, we show that already for timescales of less than half a year, there can be systematic basin-scale biases in which regions are identified as particle origins or sinks. While the stability bias is linked to divergence, it is not only limited to 2D trajectories in 3D flows, as we discuss how inappropriate treatment of surface boundary conditions in 3D Lagrangian studies can also introduce an effective non-zero divergence. Backtracking is typically applied to material that has accumulated in convergent zones, for which the stability bias especially impedes source attribution studies. Furthermore, we show how discrepancies between forward and backward trajectories can make a Bayesian approach to backtracking unsuitable. We advise modelers to routinely compare forward- and backward trajectories and assess the bias in different numerical set-ups to increase study robustness. Analytical integration methods are less error-prone and may be preferred over RK4.
Cloud microphysics—the collection of processes that govern the small-scale formation, evolution, and interactions of liquid droplets and ice crystals in clouds and precipitation—remains a major source of uncertainty in weather and climate models. Although too small in scale to be explicitly resolved in any large-eddy simulation, weather, or climate model, the representation of cloud microphysical processes has significant impact at the climate scale. Current microphysical schemes are limited by both parametric uncertainty, linked to uncertainty in physical parameter values, and structural uncertainty, arising from incomplete physical understanding of the processes at play or approximations made for computational efficiency. Recent advances in the application of machine learning (ML) to the physical sciences show significant potential for minimizing these limitations by leveraging high-fidelity simulations and observations. Here we outline the challenges that must be addressed to apply ML toward cloud microphysics scheme development. This perspectives paper synthesizes recent progress in using data-driven methods, including ML, to improve cloud microphysics parameterizations and highlights opportunities to address key uncertainties. We discuss the roles of aleatoric (irreducible, or statistical) and epistemic (reducible, or systematic) errors in contributing to microphysics parameterization uncertainty. ML can leverage observations to improve microphysical schemes via bottom-up and top-down constraints. Methods such as differentiable programming and ML-enhanced sampling strategies and the creation of large scale benchmark data sets promise to bridge the gap between observations and models and to improve the consistency of cloud microphysical representation across temporal and spatial scales.
This paper extends previous work (Groom et al., Artif. Intell. Earth Syst., 2024) in applying the entropy-optimal Sparse Probabilistic Approximation (eSPA) algorithm to predict ENSO phase, defined by thresholding the Niño3.4 index. Only satellite-era observational data sets are used for training and validation, while retrospective forecasts from 2012 to 2022 are used to assess out-of-sample skill at lead times up to 24 months. Rather than train a single eSPA model per lead, we introduce an ensemble approach in which multiple eSPA models are aggregated via a novel meta-learning strategy. The features used include the leading principal components from a delay-embedded EOF analysis of global sea surface temperature, vertical temperature gradient (a thermocline proxy), and tropical Pacific wind stresses. Crucially, the data is processed to prevent any form of information leakage from the future, ensuring realistic real-time forecasting conditions. Despite the limited number of training instances, eSPA avoids overfitting and produces probabilistic forecasts with skill comparable to the International Research Institute for Climate and Society (IRI) ENSO prediction plume. Beyond the IRI's lead times, eSPA maintains skill out to 20 months for the ranked probability skill score and 24 months for accuracy and area under the ROC curve, all at a fraction of the computational cost of a fully coupled dynamical model. Furthermore, eSPA successfully forecasts the 2015/2016 and 2018/2019 El Niño events at 24 months lead, the 2016/2017, 2017/2018, and 2020/2021 La Niña events at 24 months lead and the 2021/2022 and 2022/2023 La Niña events at 12 and 8 months lead.
Understanding the relation between large-scale (100 km) tropical atmospheric motions and small-scale convective circulations remains a challenge, despite such multiscale interactions playing a crucial role in the dynamics of large-scale circulations. In this study, a 40-day simulation made with a global storm-resolving model at 4 km horizontal resolution is used to simultaneously characterize large- and small-scale convective circulations and examine their relationships. The large-scale motions are characterized by the area-averaged vertical mass flux profile over tropical domains of similar size to the grid scale of contemporary climate models; small-scale motions are computed as deviations of vertical mass flux from the large-scale mean. We find that the simulated large-scale circulations tend to evolve in a systematic way that bears qualitative resemblance to the canonical evolution of tropical convective systems, with large-scale regimes progressing, on average, from weak but widespread subsidence, to shallow ascent/congestus clouds, to deep convection, to top-heavy stratiform anvils, and finally returning to a suppressed state. Utilizing the moisture-space framework, we compare the composite structures of the small-scale circulations in different large-scale regimes. We find that the structure of the large-scale vertical motions strongly constrains the shape of the embedded small-scale circulations. We further identify tropical anvil clouds as mediators of deep convection's up-scale influence on changes in the large-scale circulation.
The Community Earth System Model version 2 (CESM2) has a higher equilibrium climate sensitivity (ECS) than previous versions of CESM and many other Coupled Model Intercomparison Project models. Relatedly, CESM2 simulates too-cold ice-age and too-hot warm paleoclimates. An inappropriate ice number limiter in the CESM2 microphysics scheme was discovered, and some simulations indicate that the high ECS may be partially attributable to this inappropriate limiter. In light of those findings, we seek to provide users of CESM2 guidance on the fitness of CESM2 for a variety of applications. We find that despite concerns about its climate sensitivity and simulations of past climates, the transient climate response in CESM2 is moderate relative to the CMIP6 ensemble and robust across different versions of CESM. The changes made between CESM1 and CESM2 and the fixes to the microphysical issues of CESM2 have little impact on its simulated 20th and 21st century climates under SSP3–7.0. As a result, the simulated 20th and 21st century climates of CESM2 fall well within the range of the CMIP6 ensemble and agree well with observations over the historical record. However, hotter and colder paleoclimates simulated by CESM2 are inconsistent with paleoclimate evidence. A modified version of CESM2, PaleoCalibr CESM2, may be suitable for paleoclimate studies. Simulations past the end of the 21st century with default CESM2 and studies of microphysical processes in all GCMs should be analyzed with care.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：