Journal of Advances in Modeling Earth Systems最新文献

In machine learning (ML) algorithms, neural networks (NNs) can effectively learn the mapping between initial condition errors and system states through training. To reduce model errors in data assimilation, this work proposes an optimization strategy for the data assimilation (DA) process based on ML methods. A novel hybrid deep learning approach is introduced, which combines bidirectional long short-term memory (BiLSTM) networks and gated recurrent units (GRUs) in a serial processing manner. First, the BiLSTM-GRU hybrid method is trained by modeling the residuals between the system's observational data and the ensemble Kalman filter (EnKF) assimilation results, which are used as inputs to learn the error correction process. Second, the Huber loss function is used to quantify the error, and the network model's parameters are updated on the basis of the loss feedback to correct model errors in the data assimilation system. Additionally, the hybrid model incorporates nudging by introducing a certain amount of “relaxation” in each iteration, gradually approaching the actual solution. Ultimately, comprehensive experiments demonstrate that the BiLSTM-GRU hybrid, even after strengthening traditional EnKF with inflation, localization, and stabilization, consistently achieves superior robustness across varying dynamical regimes, ensemble sizes, observation densities, and noise levels, highlighting its scalability for nonlinear chaotic systems.

In this study, an Accelerated Discrete Ordinate Method (ADOM) is proposed to enhance the computational efficiency of multi-layer radiative transfer (RT) simulations while maintaining a high accuracy. ADOM applies the Discrete Ordinate Method (DOM) only in scattering layers, while the radiances for the adjacent clear-sky layers are merged and computed by non-scattering RT theory. The merging process significantly reduces the number of layers involved, enhancing computational efficiency while the vertical structure of both the Planck function and the Rayleigh scattering single-scattering albedo are fully accounted. This hybrid RT approach enables ADOM to be applicable across the visible to microwave spectrum. For satellite radiance assimilation, tangent linear and adjoint modules of ADOM are also developed to compute the Jacobians of all relevant parameters. Although ADOM merges adjacent clear-sky atmospheric layers during RT calculations, the Jacobians of properties in each merged clear-sky layer can still be accurately computed by constructing an adjoint module of the merging process. The accuracy of both the forward and adjoint modules of ADOM is evaluated against 128-stream DOM and the finite difference results based on DOM. Notably, the computational efficiency gain of ADOM is influenced by the ratio of clear-sky layers to cloud layers. As the number of cloud layers decreases, the efficiency of ADOM increases. In fully cloudy conditions, the runtime of ADOM converges to that of DOM.

A distributed multi-architecture design for massively parallel hyperbolic solvers is herein introduced and benchmarked. A unified object-oriented central processing unit (CPU) + graphics processing unit (GPU) approach is complemented with an inter-device communication layer, enabling both coarse and fine-grain parallelism on hyperbolic solvers. The approach involves the combination of three different programming platforms, namely OpenMP, CUDA and MPI. The efficiency of this distributed-heterogeneous approach is quantified under static and dynamic loads on consumer and professional grade CPUs and GPUs. An asynchronous communications scheme is implemented and described, showing very reduced overheads and a nearly linear scalability for multiple device combinations. For simulations (or systems) with non-homogeneous workloads (or devices) the domain decomposition algorithm incorporates a low-frequency load-to-device fitting function to ensure computational balance. A real-world application to high-resolution hydrodynamic modelling is presented: the propagation of a tsunami in the estuary a large river and its run-up in an urban mesh. The proposed implementation shows speedups of up to two orders of magnitude, opening new perspectives for solvers with high-demand requirements but relatively simple hardware in multi-architecture machines.

A global Four-Dimensional Ensemble Variational (4DEnVar) data assimilation system for the Atmospheric component of the Model for Prediction Across Scales (MPAS-A) is presented that uses the Joint Effort for Data assimilation Integration (JEDI). Dual-resolution cycling experiments with a 30 km analysis but an ensemble run at a coarser (60 km) resolution are shown to perform well, thereby reducing the computational cost. Month long global cycling data assimilation experiments show that 4DEnVar updates have lower mean errors in both observation and model space than comparable 3DEnVar experiments. Additional improvements over 4DEnVar are then demonstrated when using Hybrid-4DEnVar, which leverages the benefits of both flow-dependent ensemble covariance and a static climatological covariance, and when assimilating all-sky Advanced Microwave Sounding Unit-A (AMSU-A) radiance observations. Lastly, extended forecasts initialized from the four-dimensional analyses are compared with forecasts initialized from three-dimensional analyses. A particular focus is on the prediction of clouds and precipitation in forecasts initialized from Hybrid-4DEnVar versus Hybrid-3DEnVar analyses. Results from extended forecasts show that both forecasts of traditional meteorological fields and precipitation are improved through use of Hybrid-4DEnVar. However, improvements in precipitation forecasts from 4D methods are shown to be most significant in the southern hemisphere, consistent with where the largest improvements in other meteorological fields are found. Significant improvements in precipitation forecasts in the tropics are found in both 3D and 4D experiments assimilating all-sky AMSU-A radiance observations. In summary, 4DEnVar and Hybrid-4DEnVar capabilities are available through MPAS-JEDI—an open-source community developed tool—and perform well during continuous global cycling experiments across traditional verification metrics.

Tropical cyclone (TC) activity in the Seamless System for Prediction and EArth System Research (SPEAR) model large ensemble simulations is evaluated in this study. The climatological analysis indicates that though biases exist, the model overall captures the key TC characteristics, including track density, seasonality, landfall frequency, and precipitation. However, TC track density is overestimated over the Northwest Pacific (NWP) and underestimated over the Northeast Pacific. The model tends to underestimate TC landfall frequency across most continents. TC precipitation (TCP) bias varies regionally, and decomposition analysis reveals that it is primarily driven by biases in TC occurrence rather than by precipitation per TC, although the model tends to overestimate TCP in the inner-core area of TCs. SPEAR reasonably simulates the observed temporal and spatial pattern of natural variability, including the El Niño-Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO) and the Atlantic Multidecadal Oscillation (AMO), and their TC modulations. Specifically, SPEAR simulates the increase in TC activity over the NWP and decrease over the North Atlantic (NA) during El Niño events and captures the increase in TC activity over the tropical Pacific Ocean during the PDO positive phase and the increase over NA during the AMO positive phase. TCP anomalies between different phases of the climate modes follow closely the TC track density anomalies, as they are mainly caused by TC occurrence differences. While SPEAR reproduces the TC landfall frequency anomalies associated with ENSO and AMO, it exhibits the opposite signal for PDO-related anomalies compared to the observed signal.

Accurate simulation of the quasi-biennial oscillation (QBO) is challenging due to uncertainties in representing convectively generated gravity waves. We develop an end-to-end uncertainty quantification workflow that calibrates these gravity wave processes in E3SM for a realistic QBO. Central to our approach is a domain knowledge-informed, compressed representation of high-dimensional spatio-temporal wind fields. By employing a parsimonious statistical model that learns the fundamental frequency from complex observations, we extract interpretable and physically meaningful quantities capturing key attributes. Building on this, we train a probabilistic surrogate model that approximates the fundamental characteristics of the QBO as functions of critical physics parameters governing gravity wave generation. Leveraging the Karhunen–Loève decomposition, our surrogate efficiently represents these characteristics as a set of orthogonal features, capturing cross-correlations among multiple physics quantities evaluated at different pressure levels and enabling rapid surrogate-based inference at a fraction of the computational cost of full-scale simulations. Finally, we analyze the inverse problem using a multi-objective approach. Our study reveals a tension between amplitude and period that constrains the QBO representation, precluding a single optimal solution. To navigate this, we quantify the bi-criteria trade-off and generate a set of Pareto optimal parameter values that balance the conflicting objectives. This integrated workflow improves the fidelity of QBO simulations and offers a versatile template for uncertainty quantification in complex geophysical models.

Numerical studies of submesoscale ocean dynamics are restricted by several challenges, including its vast range of scales, nonhydrostatic features, and strong anisotropy. The Stratified Ocean Model with Adaptive Refinement (SOMAR) was developed to address many of these issues. Recent improvements to SOMAR incorporate Runge-Kutta time integration, Arakawa-C grids, new grid transfer methods, and error controllers in an effort to increase the model's fidelity and stability. In this paper, we detail these recent improvements, establish SOMARv2's accuracy, and demonstrate its utility as an efficient submesoscale model.

This study investigates the effects of urbanization, specifically land use change and anthropogenic emissions (AE), on convection, lightning, and surface precipitation for a case of summertime sea-breeze convection observed over the Houston metropolitan area. The unique capabilities of the NASA-Unified Weather Research and Forecasting model allows us to conduct a series of sensitivity experiments with complex configurations, in particular including multi-year land model spin-up simulations, treatment of aerosols and their precursors, and explicit cloud charging and lightning. The simulation results show that urban land use primarily alters the temporal evolution of convection, lightning, and surface precipitation, leading to late afternoon thunderstorm development. The decrease in latent heat flux from the land surface caused by urbanization weakens convection in the early afternoon, while a condition suitable for convection development is maintained in the late afternoon due to less stabilization of the lower troposphere by the weaker convection development and high sensible heat flux from the surface. On the other hand, anthropogenic aerosols directly enhance convection, lightning, and surface precipitation by increasing convective updrafts due to the aerosol-induced convective invigoration. The combined effects of urban land use and AE lead to even stronger thunderstorms in the late afternoon, mostly consistent with observations. These results indicate that urbanization increases the probability of late afternoon thunderstorms over the Houston area during the summer season. Advanced weather forecasting models that incorporate these urbanization effects might support sustainable urban planning to better mitigate the impacts of urbanization on local weather and public safety.

The development of the Simplified Cloud Resolving Energy Exascale Earth System Atmosphere Model (SCREAMv1) enables global storm-resolving simulations on modern GPU-based supercomputers. However, the high computational cost of SCREAMv1 limits its routine use for process-level studies, creating a need for efficient proxy configurations. This study addresses this gap by introducing DP-SCREAMv1, a doubly periodic cloud-resolving model designed to be fully consistent with SCREAMv1 while enabling high-resolution, long-duration simulations at significantly reduced computational expense by simulating a limited doubly periodic domain rather than the entire globe. Built on a C++/Kokkos architecture, DP-SCREAMv1 achieves exceptional performance scalability on GPU systems and includes a rich library of cases for validation and scientific exploration. In this work, we demonstrate short wall-clock times at SCREAMv1's default resolution and show that DP-SCREAMv1 supports routine execution of large-domain, high-resolution experiments that were previously challenging in practice. Furthermore, we show that DP-SCREAMv1 enables routine execution of “Giga-LES” style simulations and facilitates large-domain, high-resolution simulations that were recently considered burdensome to perform. These results document an efficient, fully consistent process-level configuration for SCREAMv1 (DP-SCREAMv1) and illustrate its use for long-duration and large-domain experiments at cloud-resolving to eddy-permitting resolution.