


Editor | ScienceAI
In the era of large models, the effects of purely data-driven meteorological and climate models have gradually caught up with or even surpassed numerical models.
However, existing large-scale meteorological and climate models still have some problems. For example, the physical consistency in the model is not high enough, which limits the ability to predict complex weather and climate phenomena such as precipitation. In addition, the forecast effect of divergent wind is not satisfactory. These issues require further research and improvement to improve the prediction accuracy and reliability of the model.
At present, combining physics, atmospheric dynamics and deep learning models is an important way to solve the bottleneck problem.
Recently, the team of researcher Huang Gang from the Institute of Atmospheric Physics, Chinese Academy of Sciences, based on the data and computing power support of the Earth System Numerical Simulation Device (Huan), starting from the perspective of the coupling relationship of physical variables, combined with the figure The neural network imposes physical soft constraints on multiple variables, improves the precipitation forecasting skills of numerical models, and makes some attempts and explorations in the integration of physics and AI.
The research was titled "Coupling Physical Factors for Precipitation Forecast in China With Graph Neural Network" and was published in "Geophysical Research Letters" on January 18.
Paper link: https://doi.org/10.1029/2023GL106676
In response to the difficult issues of precipitation forecasting, especially the forecasting of heavy precipitation, Our team solved the problem by starting from the influencing factors and occurrence mechanism of precipitation, combining the omega equation and the water vapor equation, etc., to screen variables and build a variable coupling graph network.
The Omega equation and the water vapor equation describe vertical movement and water vapor changes respectively, both of which are important factors affecting precipitation. From the perspective of graph network, the aforementioned equation reflects the relationship between the nonlinear combination of basic physical quantities (temperature, wind, humidity, etc.) and the key elements of precipitation. Therefore, it can be abstracted into a graph network, and through the variables between the graph network (nodes) and the relationships between variables to represent the combination and coupling between different physical variables.
At the same time, taking into account the impact of climate factors on weather scales, especially the systematic differences in model errors under different climate backgrounds, this study combined sparse data such as season, ENSO and other climate factors and reporting time Use entity embedding technology to embed the correction model to distinguish errors in different backgrounds.
In addition, for the precipitation process, this study made local improvements to the graph neural network ChebNet, so that it can basically maintain the effect while avoiding global operations and greatly reducing the computational complexity.
Figure 1: Schematic of the omega-GNN model. (Source: paper)
Model comparison results show that the two physically constrained models omega-GNN and omega-EGNN proposed in this study significantly improve the precipitation forecasting skills of each category compared with the numerical model, and at the same time their performance Better than the current mainstream deep learning models without physical constraints (such as U-NET, 3D-CNN, etc.).
In addition, this study performed ten sets of perturbations on all deep learning models, allowing them to perform ensemble forecasts. Combining diagnosis and case analysis, it is found that the model with physical constraints is significantly better than the model without physical constraints. For the prediction of heavy precipitation, the consistency between the omega-GNN model and the omega-EGNN model ensemble is higher, and the forecasting skills are better.
Figure 2: Each model’s (a) TS score, (b-g) spatial distribution of TS differences relative to the numerical model (precipitation above the 20mm/6h threshold). (Source: paper)
Researcher Huang Gang, the corresponding author of the paper, said: "Our team has accumulated a lot in the direction of climate dynamics. In recent years, we have made some attempts to use AI to improve weather and climate predictions, and have achieved many related results. Won relevant competition awards for the first time. In the era of large AI models, how to integrate physics with AI is a big issue, and there are many ways and ideas for integration. We combined some thoughts on atmosphere and climate dynamics to soft-soften the model from the perspective of physical coupling. Constraints, some attempts have been made in this direction, hoping to provide some incremental information for related fields."
The research was conducted by master students Chen Yutong, Dr. Wang Ya, Researcher Huang Gang and the Institute of Atmospheric Physics, Chinese Academy of Sciences Completed in collaboration with Dr. Tian Qun from the Guangzhou Institute of Tropical Marine Meteorology, China Meteorological Administration.
Reference content: https://iap.cas.cn/gb/xwdt/kyjz/202401/t20240119_6959543.html
The above is the detailed content of Applying physically coupled graph neural networks to improve precipitation forecasting skills at the Institute of Atmospheric Physics, Chinese Academy of Sciences. For more information, please follow other related articles on the PHP Chinese website!

机器学习是一个不断发展的学科,一直在创造新的想法和技术。本文罗列了2023年机器学习的十大概念和技术。 本文罗列了2023年机器学习的十大概念和技术。2023年机器学习的十大概念和技术是一个教计算机从数据中学习的过程,无需明确的编程。机器学习是一个不断发展的学科,一直在创造新的想法和技术。为了保持领先,数据科学家应该关注其中一些网站,以跟上最新的发展。这将有助于了解机器学习中的技术如何在实践中使用,并为自己的业务或工作领域中的可能应用提供想法。2023年机器学习的十大概念和技术:1. 深度神经网

实现自我完善的过程是“机器学习”。机器学习是人工智能核心,是使计算机具有智能的根本途径;它使计算机能模拟人的学习行为,自动地通过学习来获取知识和技能,不断改善性能,实现自我完善。机器学习主要研究三方面问题:1、学习机理,人类获取知识、技能和抽象概念的天赋能力;2、学习方法,对生物学习机理进行简化的基础上,用计算的方法进行再现;3、学习系统,能够在一定程度上实现机器学习的系统。

本文将详细介绍用来提高机器学习效果的最常见的超参数优化方法。 译者 | 朱先忠审校 | 孙淑娟简介通常,在尝试改进机器学习模型时,人们首先想到的解决方案是添加更多的训练数据。额外的数据通常是有帮助(在某些情况下除外)的,但生成高质量的数据可能非常昂贵。通过使用现有数据获得最佳模型性能,超参数优化可以节省我们的时间和资源。顾名思义,超参数优化是为机器学习模型确定最佳超参数组合以满足优化函数(即,给定研究中的数据集,最大化模型的性能)的过程。换句话说,每个模型都会提供多个有关选项的调整“按钮

截至3月20日的数据显示,自微软2月7日推出其人工智能版本以来,必应搜索引擎的页面访问量增加了15.8%,而Alphabet旗下的谷歌搜索引擎则下降了近1%。 3月23日消息,外媒报道称,分析公司Similarweb的数据显示,在整合了OpenAI的技术后,微软旗下的必应在页面访问量方面实现了更多的增长。截至3月20日的数据显示,自微软2月7日推出其人工智能版本以来,必应搜索引擎的页面访问量增加了15.8%,而Alphabet旗下的谷歌搜索引擎则下降了近1%。这些数据是微软在与谷歌争夺生

荣耀的人工智能助手叫“YOYO”,也即悠悠;YOYO除了能够实现语音操控等基本功能之外,还拥有智慧视觉、智慧识屏、情景智能、智慧搜索等功能,可以在系统设置页面中的智慧助手里进行相关的设置。

人工智能在教育领域的应用主要有个性化学习、虚拟导师、教育机器人和场景式教育。人工智能在教育领域的应用目前还处于早期探索阶段,但是潜力却是巨大的。

阅读论文可以说是我们的日常工作之一,论文的数量太多,我们如何快速阅读归纳呢?自从ChatGPT出现以后,有很多阅读论文的服务可以使用。其实使用ChatGPT API非常简单,我们只用30行python代码就可以在本地搭建一个自己的应用。 阅读论文可以说是我们的日常工作之一,论文的数量太多,我们如何快速阅读归纳呢?自从ChatGPT出现以后,有很多阅读论文的服务可以使用。其实使用ChatGPT API非常简单,我们只用30行python代码就可以在本地搭建一个自己的应用。使用 Python 和 C

人工智能在生活中的应用有:1、虚拟个人助理,使用者可通过声控、文字输入的方式,来完成一些日常生活的小事;2、语音评测,利用云计算技术,将自动口语评测服务放在云端,并开放API接口供客户远程使用;3、无人汽车,主要依靠车内的以计算机系统为主的智能驾驶仪来实现无人驾驶的目标;4、天气预测,通过手机GPRS系统,定位到用户所处的位置,在利用算法,对覆盖全国的雷达图进行数据分析并预测。


Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

AI Hentai Generator
Generate AI Hentai for free.

Hot Article

Hot Tools

Zend Studio 13.0.1
Powerful PHP integrated development environment

EditPlus Chinese cracked version
Small size, syntax highlighting, does not support code prompt function

Dreamweaver Mac version
Visual web development tools

Atom editor mac version download
The most popular open source editor

mPDF
mPDF is a PHP library that can generate PDF files from UTF-8 encoded HTML. The original author, Ian Back, wrote mPDF to output PDF files "on the fly" from his website and handle different languages. It is slower than original scripts like HTML2FPDF and produces larger files when using Unicode fonts, but supports CSS styles etc. and has a lot of enhancements. Supports almost all languages, including RTL (Arabic and Hebrew) and CJK (Chinese, Japanese and Korean). Supports nested block-level elements (such as P, DIV),
