大气与海洋物理

• 交叉列表
• 替换

查看 最近的 文章

显示 2025年07月21日， 星期一 新的列表

交叉提交 (展示 1 之 1 条目 )

[1] arXiv:2507.13529 (交叉列表自 physics.ao-ph) [中文pdf, pdf, html, 其他]

标题： 预测稳定性对导航尾迹规避的影响 显示英文标题

标题： Impact of Forecast Stability on Navigational Contrail Avoidance

Thomas R Dean, Tristan H Abbott, Zeb Engberg, Nicholas Masson, Roger Teoh, Jonathan P Itcovitz, Marc E J Stettler, Marc L Shapiro

评论： 24页，9图，1表。提交至《Environmental Research: 基础设施与可持续性》

主题： 大气与海洋物理 (physics.ao-ph)

通过重新引导航班绕过形成尾迹云的区域，可以减轻尾迹云引起的变暖，这需要对对流层上部和下平流层的状态进行准确且稳定的预测。 预测稳定性（即不同提前时间的预测周期之间的一致性）对于“准战术”尾迹云规避策略尤为重要，这些策略基于提前时间长达24-48小时的预测来调整航线。 然而，迄今为止尚无研究系统地量化了预测稳定性在多大程度上限制了准战术规避策略的有效性。 本研究通过将使用ECMWF HRES天气预测生成的尾迹云预测（提前时间最长为48小时）与基于ECMWF ERA5再分析数据生成的尾迹云回溯预测进行比较，填补了这一空白。 对预测误差的分析表明，在持续形成尾迹云的区域中，预测与再分析数据之间的逐点一致性较低，逐点误差率与之前在再分析数据和现实情况中比较尾迹云形成区域时发现的误差率相似。 然而，我们还表明，当将预测与再分析数据进行比较，以及将再分析数据与现场测量数据进行比较时，尾迹云形成区域的位置空间误差相对较小。 最后，我们表明，设计一个能够利用相对较小的空间误差的轨迹优化器，使得基于尾迹云预测的飞行轨迹优化可以在最相关于飞行计划的8-24小时提前时间内，通过再分析数据评估，将尾迹云气候强迫减少80-90%，燃油成本增加不超过0.4%。 我们的结果表明，与飞行计划相关的预测具有足够的稳定性，可用于准战术尾迹云规避。 显示英文摘要

Mitigating contrail-induced warming by re-routing flights around contrail-forming regions requires accurate and stable forecasts of the state of the upper troposphere and lower stratosphere. Forecast stability (i.e., consistency between forecast cycles with different lead times) is particularly important for "pre-tactical" contrail avoidance strategies that adjust routes based on forecasts with lead times as long as 24-48 hours. However, no study to date has systematically quantified the degree to which forecast stability limits the effectiveness of pre-tactical avoidance. This study addresses this gap by comparing contrail forecasts generated using ECMWF HRES weather forecasts with lead times up to 48 hours to contrail hindcasts generated based on ECMWF ERA5 reanalysis. An analysis of forecast errors shows low pointwise consistency between persistent-contrail-forming regions in forecasts and reanalysis, with pointwise error rates similar to those found in previous comparisons of contrail-forming regions in reanalysis and reality. However, we also show that spatial errors in the locations of contrail-forming regions are relatively small, both when forecasts are compared to reanalysis and when reanalysis is compared to in-situ measurements. Finally, we show that designing a trajectory optimizer to take advantage of relatively small spatial errors allows flight trajectory optimizations based on contrail forecasts to reduce contrail climate forcing evaluated based on reanalysis by 80-90% at the 8-24 hour lead times most relevant to flight planning, with fuel penalties under 0.4%. Our results show that forecasts with lead times relevant to flight planning are stable enough to be used for pre-tactical contrail avoidance.

替换提交 (展示 3 之 3 条目 )

[2] arXiv:2412.02503 (替换) [中文pdf, pdf, html, 其他]

标题： VA-MoE：用于增量天气预报的变量自适应专家混合模型 显示英文标题

标题： VA-MoE: Variables-Adaptive Mixture of Experts for Incremental Weather Forecasting

Hao Chen, Han Tao, Guo Song, Jie Zhang, Yunlong Yu, Yonghan Dong, Lei Bai

评论： 这篇论文已被ICCV25接受

主题： 机器学习 (cs.LG) ; 大气与海洋物理 (physics.ao-ph)

本文提出了变量自适应专家混合（VAMoE），这是一种新颖的增量天气预报框架，能够实时动态适应实时数据中的演变时空模式。 传统的天气预测模型常常在高昂的计算开销以及随着新观测数据到达而持续更新预测的需求上面临挑战。 VAMoE通过利用专家的混合架构来解决这些问题，其中每个专家专门用于捕捉大气变量（温度、湿度、风速）的不同子模式。 此外，所提出的方法采用变量自适应门控机制，根据输入上下文动态选择和组合相关专家，从而实现高效的知识蒸馏和参数共享。 这种设计显著降低了计算开销，同时保持了高预测准确性。 在真实世界ERA5数据集上的实验表明，VAMoE在短期（1天）和长期（5天）预测任务中表现与最先进模型相当，仅使用约25%的可训练参数和50%的初始训练数据。 显示英文摘要

This paper presents Variables Adaptive Mixture of Experts (VAMoE), a novel framework for incremental weather forecasting that dynamically adapts to evolving spatiotemporal patterns in real time data. Traditional weather prediction models often struggle with exorbitant computational expenditure and the need to continuously update forecasts as new observations arrive. VAMoE addresses these challenges by leveraging a hybrid architecture of experts, where each expert specializes in capturing distinct subpatterns of atmospheric variables (temperature, humidity, wind speed). Moreover, the proposed method employs a variable adaptive gating mechanism to dynamically select and combine relevant experts based on the input context, enabling efficient knowledge distillation and parameter sharing. This design significantly reduces computational overhead while maintaining high forecast accuracy. Experiments on real world ERA5 dataset demonstrate that VAMoE performs comparable against SoTA models in both short term (1 days) and long term (5 days) forecasting tasks, with only about 25% of trainable parameters and 50% of the initial training data.

[3] arXiv:2504.19883 (替换) [中文pdf, pdf, html, 其他]

标题： 在风暴解析模拟中，潮汐锁定的类地系外行星上的闪电活动对于一系列地表压力的情况 显示英文标题

标题： Lightning activity on a tidally locked terrestrial exoplanet in storm-resolving simulations for a range of surface pressures

Denis E. Sergeev, James W. McDermott, Lottie Woods, Marrick Braam, Jake K. Eager-Nash, Ian A. Boutle

评论： 12页，7图；已接受发表于《皇家天文学会月报》

主题： 地球与行星天体物理学 (astro-ph.EP) ; 大气与海洋物理 (physics.ao-ph)

云层大气会产生电放电，包括闪电。 闪电反过来提供了足够的能量将空气分子分解成活性物质，并从而影响大气成分。 围绕红矮星运行的潮汐锁定的岩石系外行星的气候可能具有强烈且高度局部化的雷暴活动，这与它们白天一侧的湿对流有关。 产生闪电的对流云的分布和结构受到各种气候参数的影响，其中关键的一个是大气质量，即地表空气压力。 在本研究中，我们使用一个全球风暴解析气候模型，在不同地表压力下预测潮汐锁定系外行星的雷暴发生情况。 我们比较了两种闪电参数化方法：一种基于冰云微物理过程，另一种基于对流云的垂直范围。 我们发现，这两种参数化方法都预测闪电量随着地表压力增加而单调减少，这是由于对流较弱和冰云较少。 行星上的闪电分布随地表压力变化而变化，响应于大尺度环流和大气垂直分层的变化。 我们的研究为潮汐锁定类地系外行星的闪电活动提供了修订后的高分辨率估计，对全球大气化学有重要意义。 显示英文摘要

Cloudy atmospheres produce electric discharges, including lightning. Lightning, in turn, provides sufficient energy to break down air molecules into reactive species and thereby affects the atmospheric composition. The climate of tidally locked rocky exoplanets orbiting M-dwarf stars may have intense and highly localised thunderstorm activity associated with moist convection on their day side. The distribution and structure of lightning-producing convective clouds is shaped by various climate parameters, of which a key one is atmospheric mass, i.e. surface air pressure. In this study, we use a global storm-resolving climate model to predict thunderstorm occurrence for a tidally locked exoplanet over a range of surface pressures. We compare two lightning parameterisations: one based on ice cloud microphysics and one based on the vertical extent of convective clouds. We find that both parameterisations predict that the amount of lightning monotonically decreases with surface pressure due to weaker convection and fewer ice clouds. The spatial distribution of lightning on the planet changes with respect to the surface pressure, responding to the changes in the large-scale circulation and the vertical stratification of the atmosphere. Our study provides revised, high-resolution estimates for lightning activity on a tidally locked Earth-like exoplanet, with implications for global atmospheric chemistry.

[4] arXiv:2507.12144 (替换) [中文pdf, pdf, html, 其他]

标题： FourCastNet 3：一种用于大规模概率机器学习天气预报的几何方法 显示英文标题

标题： FourCastNet 3: A geometric approach to probabilistic machine-learning weather forecasting at scale

Boris Bonev, Thorsten Kurth, Ankur Mahesh, Mauro Bisson, Jean Kossaifi, Karthik Kashinath, Anima Anandkumar, William D. Collins, Michael S. Pritchard, Alexander Keller

主题： 机器学习 (cs.LG) ; 大气与海洋物理 (physics.ao-ph)

FourCastNet 3 通过实施可扩展的几何机器学习（ML）方法，推进了全球天气建模的概率集合预报。 该方法旨在尊重球面几何，并准确建模问题的空间相关概率性质，从而在多个尺度上实现稳定的谱和现实的动力学。 FourCastNet 3 提供的预测准确性超过了领先的传统集合模型，并与最佳基于扩散的方法相当，同时其预测速度比这些方法快8到60倍。 与其他ML方法相比， FourCastNet 3 展现出出色的概率校准，并在长达60天的延长预报时间下仍保持现实的谱。 所有这些进步都是通过一个专为球面几何设计的纯卷积神经网络架构实现的。 通过一种新颖的训练范式，结合模型并行和数据并行，1024块以上的GPU实现了可扩展且高效的大型训练，该范式受到经典数值模型中区域分解方法的启发。 此外，FourCastNet 3 在单个GPU上实现了快速推理，能够在不到4分钟内生成分辨率为0.25{\deg }，每6小时一次的60天全球预报。 其计算效率、中程概率技能、谱保真度以及亚季节时间尺度上的滚动稳定性，使其成为通过大规模集合预测改进气象预报和早期预警系统的有力候选。 显示英文摘要

FourCastNet 3 advances global weather modeling by implementing a scalable, geometric machine learning (ML) approach to probabilistic ensemble forecasting. The approach is designed to respect spherical geometry and to accurately model the spatially correlated probabilistic nature of the problem, resulting in stable spectra and realistic dynamics across multiple scales. FourCastNet 3 delivers forecasting accuracy that surpasses leading conventional ensemble models and rivals the best diffusion-based methods, while producing forecasts 8 to 60 times faster than these approaches. In contrast to other ML approaches, FourCastNet 3 demonstrates excellent probabilistic calibration and retains realistic spectra, even at extended lead times of up to 60 days. All of these advances are realized using a purely convolutional neural network architecture tailored for spherical geometry. Scalable and efficient large-scale training on 1024 GPUs and more is enabled by a novel training paradigm for combined model- and data-parallelism, inspired by domain decomposition methods in classical numerical models. Additionally, FourCastNet 3 enables rapid inference on a single GPU, producing a 60-day global forecast at 0.25{\deg}, 6-hourly resolution in under 4 minutes. Its computational efficiency, medium-range probabilistic skill, spectral fidelity, and rollout stability at subseasonal timescales make it a strong candidate for improving meteorological forecasting and early warning systems through large ensemble predictions.