数值模式预报方法

No abstract available

No abstract available

Pure AI-based weather prediction (AIWP) models have made waves within the scientific community and the media, claiming superior performance to numerical weather prediction (NWP) models. However, these models often lack impactful output variables such as precipitation. One exception is Google DeepMind’s GraphCast model, which became the first mainstream AIWP model to predict precipitation, but performed only limited verification. We present an analysis of the ECMWF’s Integrated Forecast System (IFS)-initialized (GRAPIFS) and the NCEP’s Global Forecast System (GFS)-initialized (GRAPGFS) GraphCast precipitation forecasts over the contiguous United States and compare to results from the GFS and IFS models using 1) grid-based, 2) neighborhood, and 3) object-oriented metrics verified against the ECMWF Reanalysis v5 (ERA5) and the NCEP/EMC Stage IV precipitation analysis datasets. We affirmed that GRAPGFS and GRAPIFS perform better than the GFS and IFS in terms of root mean squared error and stable equitable errors in probability space, but the GFS and IFS precipitation distributions more closely align with the ERA5 and Stage IV distributions. Equitable threat score also generally favored GraphCast, particularly for lower accumulation thresholds. Fractions skill score for increasing neighborhood sizes shows greater gains for the GFS and IFS than GraphCast, suggesting the NWP models may have a better handle on intensity but struggle with location. Object-oriented verification for GraphCast found positive area biases at low accumulation thresholds and large negative biases at high accumulation thresholds. GRAPGFS saw similar performance gains to GRAPIFS when compared to their NWP counterparts, but initializing with the less familiar GFS conditions appeared to lead to an increase in light precipitation.

A wide range of important societal and economic applications on a national and international level strive for an integrated understanding and forecasting of weather and climate, at high spatial resolution ranging from days to decades. The global to regional model system ICON (Icosahedral Nonhydrostatic) has been applied to weather as well as to climate timescales with joint developments of the model infrastructure. However, ICON’s model configurations share the same dynamical core but differ substantially in their physical parameterization and the coupling of Earth System components, depending on whether they were designed for numerical weather prediction (NWP) or climate applications. Starting in 2020, a new modeling initiative has been launched as a joint project between climate modeling institutes and the Deutscher Wetterdienst. The initiative “vertically” integrates NWP, climate predictions, climate projections, and atmospheric composition modeling based on the ICON framework and targets a unified treatment of the respective subgrid-scale parameterizations. This initiative aims at the development of coupled model configurations of ICON to conduct operational weather and ocean forecasts for several days, climate predictions with timescales up to 10 years ahead as well as climate projections, and it provides a model baseline for joint research for NWP and climate. This paper illustrates the strategic direction of this modeling initiative, isolates key challenges, and reports on first results.

Precipitation nowcasting, which is critical for flood emergency and river management, has remained challenging for decades, although recent developments in deep generative modeling (DGM) suggest the possibility of improvements. River management centers, such as the Tennessee Valley Authority, have been using Numerical Weather Prediction (NWP) models for nowcasting, but they have been struggling with missed detections even from best-in-class NWP models. While decades of prior research achieved limited improvements beyond advection and localized evolution, recent attempts have shown progress from so-called physics-free machine learning (ML) methods, and even greater improvements from physics-embedded ML approaches. Developers of DGM for nowcasting have compared their approaches with optical flow (a variant of advection) and meteorologists’ judgment, but not with NWP models. Further, they have not conducted independent co-evaluations with water resources and river managers. Here we show that the state-of-the-art physics-embedded deep generative model, specifically NowcastNet, outperforms the High Resolution Rapid Refresh (HRRR) model, which is the latest generation of NWP, along with advection and persistence, especially for heavy precipitation events. Thus, for grid-cell extremes over 16 mm/h, NowcastNet demonstrated a median critical success index (CSI) of 0.30, compared with median CSI of 0.04 for HRRR. However, despite hydrologically-relevant improvements in point-by-point forecasts from NowcastNet, caveats include overestimation of spatially aggregate precipitation over longer lead times. Our co-evaluation with ML developers, hydrologists and river managers suggest the possibility of improved flood emergency response and hydropower management.

This article examines a small portion of G. I. Marchuk's multifaceted scientific and organizational work, which had a significant impact on the development of modern numerical weather prediction. Emphasis is placed on the initial development and application of the splitting method for solving systems of differential equations describing atmospheric processes, and, briefly, on the development of adjoint equation theory and its application.

Operational meteorological forecasting has long relied on physics-based numerical weather prediction (NWP) models. Recently, this landscape has faced disruption by the advent of data-driven artificial intelligence (AI)-based weather models, which offer tremendous computational performance and competitive forecasting accuracy. However, data-driven models for medium-range forecasting generally suffer from major limitations, including low effective resolution and a narrow range of predicted variables. This study illustrates the relative strengths and weaknesses of these competing paradigms using the physics-based GEM (Global Environmental Multiscale) and the AI-based GraphCast models. Analyses of their respective global predictions in physical and spectral space reveal that GraphCast-predicted large scales outperform GEM, particularly for longer lead times, even though fine scales predicted by GraphCast suffer from excessive smoothing. Building on this insight, a hybrid NWP-AI system is proposed, wherein temperature and horizontal wind components predicted by GEM are spectrally nudged toward GraphCast predictions at large scales, while GEM itself freely generates the fine-scale details critical for local predictability and weather extremes. This hybrid approach is capable of leveraging the strengths of GraphCast to enhance the prediction skill of the GEM model while generating a full suite of physically consistent forecast fields with a full power spectrum. Additionally, trajectories of tropical cyclones are predicted with enhanced accuracy without significant changes in intensity. Work is in progress for operationalization of this hybrid system at the Canadian Meteorological Centre.

No abstract available

No abstract available

Accurate forecasting of wind speed is essential for daily life and social production. While numerical weather prediction products are widely used, they rely on global data and mathematical models to solve atmospheric dynamics' equations, often failing to capture localized micrometeorological phenomena accurately. Factors such as surface conditions, land‐sea differences, and topography, particularly in coastal areas, further impact the accuracy of wind speed forecasts. This study presents a new method to enhance short‐term wind speed forecasting along China's coast by incorporating local and neighborhood spatiotemporal information. The approach integrates meteorological data from adjacent grid points as new inputs in the LightGBM, CatBoost, and XGBoost algorithms. Stacking ensemble technique is then employed to effectively combine with the aforementioned foundational models. Two sets of experiments are conducted: Experiments 1 exclude surrounding information, while Experiments 2 include it. Each set consists of five experiment groups: annual, spring, summer, autumn, and winter. Within each group, four models are tested: XGBoost, LightGBM, CatBoost, and stacking. Results show that incorporating surrounding site information improves forecast accuracy. In all five groups with added surrounding site information, the stacking model performs best. Compared to ECMWF forecast data, the stacking model improves wind speed forecast accuracy from 53.3%, 50.9%, 55.2%, 53.0%, and 54.0% to 77.2%, 73.1%, 76.7%, 78.2%, and 77.1%, respectively. These findings demonstrate the potential effectiveness of the proposed method for improving short‐term wind speed forecasts in China's coastal areas.

No abstract available

The international radio occultation (RO) community is conducting a collaborative effort to explore the impact of a large number of RO observations on numerical weather prediction (NWP). This effort, the Radio Occultation Modeling Experiment (ROMEX), has been endorsed by the International Radio Occultation Working Group, a scientific working group under the auspices of the Coordination Group for Meteorological Satellites (CGMS). ROMEX seeks to inform strategies for future RO missions and acquisitions. ROMEX is planned to consist of at least one three-month period during which all available RO data are collected, processed, archived, and made available to the global community free of charge for research and testing. Although the primary purpose is to test the impact of varying numbers of RO observations on NWP, the three months of RO observations during the first ROMEX period (ROMEX-1, September-November 2022) will be a rich data set for research on many atmospheric phenomena. The RO data providers have sent their data to EUMETSAT for processing. The total number of RO profiles averages between 30,000 and 40,000 per day for ROMEX-1. The processed data (phase, bending angle, refractivity, temperature, and water vapor) will be distributed to ROMEX participants by the Radio Occultation Meteorology Satellite Applications Facility (ROM SAF). The data will also be processed independently by the UCAR COSMIC Data Analysis and Archive Center (CDAAC) and available via ROM SAF. The data are freely available to all participants who agree to the conditions that the providers be acknowledged and the data are not used for commercial or operational purposes.

No abstract available

Numerical weather prediction (NWP) is crucial in the current short-term wind power forecasting (STWPF) based on data, but it is difficult for STWPF to achieve high accuracy due to the limited accuracy of NWP, which poses a serious challenge to the formulation of forward generation plans. In response to the above issues, this article conducts a traceability analysis of the error of STWPF and proposes a wind power prediction method based on NWP wind speed trend correction. Firstly, the causes of existing errors are analyzed to quantify the impact of NWP on prediction accuracy. Secondly, considering the process correlation between measured and predicted wind speeds, improved complete ensemble EMD with adaptive noise (ICEEMDAN) is used to decompose historical measured wind speeds and NWP wind speeds to construct the most relevant low-frequency trend components. Thirdly, a weighted dual constraint mechanism is proposed to select the most similar historical NWP trend segments to correct NWP wind speed. Finally, the corrected wind speed is used for power prediction and completing STWPF. Through the application of this method to a wind farm in Inner Mongolia Autonomous Region, China, which effectively improves the accuracy of NWP and reduces the average RMSE by 1.39% for power prediction, the effectiveness of this method is verified.

We develop a method to assess numerical weather prediction (NWP) model performances from the mid‐stratosphere to the mesosphere‐lower thermosphere ( ∼ ${\sim} $ 30–120 km), through comparisons between observed and simulated amplitude of oceanic infrasound known as microbaroms. We adapt a recently published array processing algorithm, the multichannel maximum‐likelihood (MCML), to the 360 ° ${}^{\circ}$ ‐observations of microbaroms. We simulate infrasound propagation using a source model and atmospheric specifications prescribed by NWP models. As this study paves the way for the assimilation of microbarom observations in these models, we assess the different components of our method. The sensitivity of the NWP model assessments to the acoustic propagation is investigated. We demonstrate the limitations of a parametrized attenuation solely driven by atmospheric fields at the infrasound station (a method previously used due to its computational efficiency) by comparing it with an explicit simulation retaining the whole 3D atmospheric fields. Importantly, in the microbarom simulations, we account for the array response to allow one‐to‐one comparisons with observations. We also highlight an observed intermittent semi‐diurnal periodicity, whose occurrence depends on middle‐atmospheric conditions, pointing at arrivals from the mesosphere and lower thermosphere. Hence, its variability needs to be accounted for in the simulations. We use a circular optimal transport metric to quantify differences between simulated and observed microbarom azimuthal distributions in a systematic way. We present NWP models relative performances diagnostics over periods of interest, including a sudden stratospheric warming in January 2021. We discuss how our approach provides insights into the model performances in the middle atmosphere.

The temporal variation of wind power is primarily influenced by wind speed, exhibiting high levels of randomness and fluctuation. The accuracy of short-term wind power forecasts is greatly affected by the quality of Numerical Weather Prediction (NWP) data. However, the prediction error of NWP is common, and posing challenges to the precision of wind power prediction. To address this issue, the paper proposes a NWP wind speed error correction model based on Residual Network-Gated Recurrent Unit (ResNet-GRU). The model corrects the forecasted wind speeds at different heights to provide reliable data foundation for subsequent predictions. Furthermore, in order to overcome the difficulty of selecting network parameters for the combined prediction model, we integrate the Kepler Optimization Algorithm (KOA) intelligent algorithm to achieve optimal parameter selection for the model. We propose a Convolutional Neural Network-Long and Short-Term Memory Network (CNN-LSTM) based on Attention Mechanism for short-term wind power prediction. Finally, the proposed methods are validated using data from a wind farm in northwest China, demonstrating their effectiveness in improving prediction accuracy and their practical value in engineering applications.

The scientific literature has many studies evaluating numerical weather prediction (NWP) models. However, many of those studies averaged across a myriad of different atmospheric conditions and surface forcings which can obfuscate the atmospheric conditions when NWP models perform well versus when they perform inadequately. To help isolate these different scenarios, we used observations from the U.S. Climate Reference Network (USCRN) obtained between 1 January and 31 December 2021 to distinguish among different near-surface atmospheric conditions (i.e., different near-surface heating rates (), incoming shortwave radiation (SWd) regimes, and 5-cm soil moisture (SM05)) to evaluate the High-Resolution Rapid Refresh (HRRR) model, which is a 3-km model used for operational weather forecasting in the U.S. On days with small (large) , we found afternoon T biases of about 2°C (−1°C) and afternoon SWd biases of up to 170 W m−2 (100 W m−2), but negligible impacts on SM05 biases. On days with small (large) SWd, we found daytime temperature biases of about 3°C (−2.5°C) and daytime SWd biases of up to 190 W m−2 (80 W m−2). Whereas different SM05 had little impact on T and SWd biases, dry (wet) conditions had positive (negative) SM05 biases. We argue that the proper evaluation of weather forecasting models requires careful consideration of different near-surface atmospheric conditions and is critical to better identifying model deficiencies which supports improvements to the parameterization schemes used therein. A similar, regime-specific model verification approach may also be used to help evaluate other geophysical models.

No abstract available

No abstract available

The operational Canadian Global Deterministic Prediction System suffers from a weak-intensity bias for simulated tropical cyclones. The presence of this bias is confirmed in progressively simplified experiments using a hierarchical system development technique. Within a semi-idealized, simplified-physics framework, an unexpected insensitivity to the representation of relevant physical processes leads to investigation of the model’s semi-Lagrangian dynamical core. The root cause of the weak-intensity bias is identified as excessive numerical dissipation caused by substantial off-centering in the two time-level time integration scheme used to solve the governing equations. Any (semi-)implicit semi-Lagrangian model that employs such off-centering to enhance numerical stability will be afflicted by a misalignment of the pressure gradient force in strong vortices. Although the associated drag is maximized in the tropical cyclone eyewall, the impact on storm intensity can be mitigated through an intercomparison-constrained adjustment of the model’s temporal discretization. The revised configuration is more sensitive to changes in physical parameterizations and simulated tropical cyclone intensities are improved at each step of increasing experimental complexity. Although some rebalancing of the operational system may be required to adapt to the increased effective resolution, significant reduction of the weak-intensity bias will improve the quality of Canadian guidance for global tropical cyclone forecasting.

No abstract available

Preemptively managing electrical circuits during periods of elevated wildfire risk, such as downslope windstorms in complex terrain, necessitates accurate, high-resolution forecasts days in advance. Currently available high-resolution operational guidance is limited with respect to the forecasting period and/or spatial detail it can provide. As a consequence, many public utilities support their own NWP systems, primarily WRF-based, with initial conditions and boundary forcings supplied by operational global models such as the GFS. This study demonstrates pairing Google’s AI-based GraphCast (GC) model, which operates on a 0.25-degree grid and produces global forecasts every six hours, with a “last mile” WRF (WRF-LM) downscaling framework to refine GC’s spatial and temporal resolution for utility forecasting needs. We illustrate the potential of this system through a case study involving the December 2021 Marshall fire windstorm in the Boulder, Colorado area. Our results suggest that GC-initialized WRF forecasts are competitive with those guided by NCEP and ECMWF operational products, offering a promising hybrid approach for utility weather forecasting.

No abstract available

No abstract available

As the first hyperspectral infrared sounder onboard a geostationary satellite, FengYun 4A (FY-4A) Geostationary Interferometric Infrared Sounder (GIIRS) plays an important role in high-impact weather forecasting with a high spatial and temporal resolution. Because hyperspectral infrared data are highly sensitive to clouds and these cloud-affected data are currently challenging to assimilate, cloud detection is a necessary step to assimilate hyperspectral infrared data better. FY-4A GIIRS has been operationally assimilated in the Chinese operational numerical weather prediction (NWP) system China Meteorological Administration Global Forecast System (CMA-GFS), which uses an imager-assisted cloud detection method to detect the clear-sky field of view. This article develops a new cloud detection method to detect clear channels for FY-4A GIIRS in CMA-GFS, extending the original method from two-dimensional (2-D) to three-dimensional (3-D) space. Compared to the original method, the 3-D cloud detection (3DCD) method increases the number of observations to two to three times for the middle- and lower-tropospheric channels and three to four times for the upper-tropospheric channels. The additional observations are of the same quality as those retained by the original cloud detection method, and the averaged observation minus background (OMB) biases after bias correction for upper-tropospheric channels decreased 54% compared to the original. Experiments results indicate that this method successfully achieves the cloud detection in 3-D for GIIRS data and shows the potential in improving data utilization. Also, the results show that 3DCD has a slightly positive impact on temperature and wind forecasts.

No abstract available

This paper provides an outlook on the future of operational weather prediction given the recent evolution in science, computing and machine learning. In many parts, this evolution strongly deviates from the strategy operational centres have formulated only several years ago. New opportunities in digital technology have greatly accelerated progress, and the full integration of computational science in numerical weather prediction centres is common knowledge now. Within the last few years, a vast machine learning research community has emerged for creating new and tailor-made products, accelerating processing and - most of all - creating emulators for the entire production of global forecasts that outperform traditional systems at the spatial resolution of the training data. In this context, the role of both numerical models and observations is changing from being equation to data driven. Analyses and reanalyses are becoming the new currency for training machine learning, and operational centres are in a powerful position as they generate these datasets based on decades worth of experience. This environment creates incredible opportunities to progress much faster than in the past but also uncertainties about what the strategic implications on defining cost-effective and sustainable research and operations are, and how to achieve sufficient high-performance computing and data handling capacities. It will take individual national public services a while to understand what to focus on and how to coordinate their substantial investments in staff and infrastructure at institutional, national and international level. This paper addresses this new situation operational weather prediction finds itself in through formulating the most likely"what if?"scenarios for the near future and provides an outline for how weather centres could adapt.

Weather prediction is critical for a range of human activities, including transportation, agriculture and industry, as well as for the safety of the general public. Machine learning transforms numerical weather prediction (NWP) by replacing the numerical solver with neural networks, improving the speed and accuracy of the forecasting component of the prediction pipeline1, 2, 3, 4, 5–6. However, current models rely on numerical systems at initialization and to produce local forecasts, thereby limiting their achievable gains. Here we show that a single machine learning model can replace the entire NWP pipeline. Aardvark Weather, an end-to-end data-driven weather prediction system, ingests observations and produces global gridded forecasts and local station forecasts. The global forecasts outperform an operational NWP baseline for several variables and lead times. The local station forecasts are skilful for up to ten days of lead time, competing with a post-processed global NWP baseline and a state-of-the-art end-to-end forecasting system with input from human forecasters. End-to-end tuning further improves the accuracy of local forecasts. Our results show that skilful forecasting is possible without relying on NWP at deployment time, which will enable the realization of the full speed and accuracy benefits of data-driven models. We believe that Aardvark Weather will be the starting point for a new generation of end-to-end models that will reduce computational costs by orders of magnitude and enable the rapid, affordable creation of customized models for a range of end users. Aardvark Weather, an end-to-end machine learning model, replaces the entire numerical weather prediction pipeline with a machine learning model, by producing accurate global and local forecasts without relying on numerical solvers, revolutionizing weather prediction with improved speed, accuracy and customization capabilities.