一种求解条件非线性最优扰动的快速算法及其在台风-中国气象学会

合集下载

基于非线性最优目标函数的红外云图台风中心自动定位

适合不易求导函数的求解问题。它将问题域的可
３．２．２变异操作
行解看作是群体的“ 染色体” ，将其置于问题的“ 环
境” 中，根据适者生存的原则，从中选择出适应环境的“ 染色体” 进行复制，通过交叉、变异两种基
为了防止有效基因的缺失，最大限度地避免早熟收敛，本文利用二元变异算子㈣（同或／异或）来进行变异操作，提高了遗传算法的优化速度。其变异概率为：
物遗传算法，求得了台风中心位置的最优数值解，有效地实现了台风中心的自动定位。多个台风中心定位仿真结果表明，该方法能够实现有眼和无眼台风的自动定位，并且理论可靠、实用性强、定位精度高。
关键词：非线性最优目标函数；台风中心；自动定位；红外云图
，、，，
函数
，
利用简单遗传算法进行求解；在此基础之
这些研究主要是综合利用数字图像处理技术和数
上
后
学形态学操作从云图中提取出边缘和骨架等形态
特征，，引，实现台风中心的自动定位，这些方
针对台风中心密闭云区的特点，对遗传算法的初始群体生成和遗传操作方面进行了改进
定量描述和参数的自动提取，因此，这些算法都存在定程度的通用性和鲁棒性问题；王燕燕等
一
台风中心的确定对于天气分析和台风预测具有重要的意义，而台风作为一种复杂的天气系统，
是在热带海洋上空发展起来的强烈气旋性涡旋，
具有很强的非线性结构特点，在大气动力学中用
一
［１０将台风运动看作是刚性物体的运动并需人工干预自动化程度不高，另外，由于气旋中心并

台风集合预报研究进展

台风集合预报研究进展作者：张璟李泓段晚锁张峰来源：《大气科学学报》2022年第05期摘要台风数值预报是防台减灾的关键，而集合预报是体现和减少数值预报不确定性的常用方法。

本文对近年来台风集合预报方法的研究进展进行了梳理和总结，涉及初值集合扰动、模式扰动技术以及基于统计的台风集合预报后处理技术。

对全球几个主要集合预报系统的发展及我国的区域台风集合预报系统做了回顾。

最后，在回顾的基础上，讨论和提出了关于台风集合预报仍存在的问题及未来可能的研究方向。

关键词台风集合预报;初值扰动;模式扰动;集合预报后处理技术;综述台风是发生在热带洋面的多时间多空间尺度相互作用的极端天气现象。

由于台风的发生通常伴随着极端灾害性事件：比如强风、暴雨、风暴潮等，给人类的生命和财产造成巨大损害。

因此，对台风及其伴随的极端事件进行及时准确地预报是防灾减灾的关键。

然而，由于台风的发生发展受到复杂的多时空尺度过程的相互作用的影响，其预报存在较大的不确定性。

单一确定性预报的形式无法对台风预测的不确定性进行定量度量，提供的预报信息较为有限，存在一定的局限性。

对此，国际上越来越多的业务预报机构开始将集合预报的思想和方法应用到台风预测中，尝试定量估计台风预报的不确定性，并以概率的形式给出台风指导预报。

集合预报从思想和方法的提出到业务应用的发展，已有近半个世纪。

目前，世界气象组织（WMO）更是将集合预报列为未来数值预报的三大发展战略之一。

集合预报思想最早是由Epstein（1969）和Leith（1974）提出。

其思路如圖1所示，对模式的初始分析场（红点）进行扰动，生成一组能反映初始状态不确定性的初始集合成员。

然后，再对这些扰动初始场分别进行数值积分，得到未来天气演变的多种可能性（多个黑色方框）。

业务预报中进一步将这些集合预报场进行后处理，得到集合预报产品，比如降水概率估计（填色）。

相比之下，针对台风的集合预报研究起步稍晚，始于20世纪90年代中期。

改进的粒子群算法在求解条件非线性最优扰动的应用

ａｅａｌｇｏａｒｌｌｂｌＣＮＯＰｉ０ｉｓｎ２０ｔｍｅ．Ｋｅｏｒｙｗｄｓ：ｐａｔｃｅｓｒｏｔｍｉａｉｎ，ｃｎｔｏａｏｌｎａｐｉａｒｕｒｔｏｒｉｌｗａｍｐｉｚｔｏｏｄｉｉｎｌｎｎｉｅｒｏｔｍｌｐｅｔｂａｉｎｎｎｓＯｈｏｍＯｔ
局ＣＮＯＰ。
关键词：粒子群算法；件非线性最优扰动；光滑条非
中图分类号：３文献标识码：文章编号：６１—７４（０１０Ｐ４５Ａ１７１７２１ｍｔｍｉａｉｎＡｌｏｉｈｓｆｒｍｒｖｄＰａｔｃｅＳｒＯｐｉｚｔｏｇｒｔｍｏ
ＣｏｎｄｉｉｎａｎｌｎｅｒＯｐｔｍａｒｕｒｔｏｎｔｏｌＮｏｉａｉｌＰｅｔｂａｉ
Ｙｅ—ｕ，ＺＮｕＧＮｉｎ — ＵＺＥＧＱｎＥＦｉｉｈＨＡＧＬ，ＯＧＸａｇＷ，ＨＮｉ
（ｎｔｕｅｏｃｅｃｓＰＡＵｉｅｉｆｃｅｃｎｅｈｏｏｙＮｎｉｇ１１１ＣｉａＩｓｉｔｆｉｎｅ，ＬｎｖｓｙｏｉｅａｄＴｃｎｌｇ，ａｊ０，ｈｎ）ｔＳｔＳｎｎ２１
Ａｂｔａｔ：ｓｒｃＤｕｅｔｔｉｏｉｓｓｍｐｌｏｅ，ｅｓｐｒｔｏｎａｔｃｎｅｇｎｃｅｃｄａｙｏｅａｉｎａｄｆｓｏｖｒｅｅｐａｔｃｅｓｒｏｉｉａｉｎｒｉｌｗａｍｐｔｍｚｔｏ

条件非线性最优扰动（CNOP）：简介与数值求解

条件非线性最优扰动（CNOP）：简介与数值求解孙国栋;穆穆;段晚锁;王强;彭飞【期刊名称】《气象科技进展》【年(卷),期】2016(6)6【摘要】This paper introduces the deifnition of conditional nonlinear optimal perturbation (CNOP), and the applications of the CNOP in atmosphere and ocean studies. The CNOP approach is expanded as that related to initial perturbation (CNOP-I), related to parameter perturbation (CNOP-P), and the combined both of CNOP-I and CNOP-P, according to the different perturbation types. The CNOP-I approach has been applied to the predictability studies of ENSO events, Kuroshio path anomalies, blocking, nonlinear stabilities of thermohaline circulation and grassland ecosystem. The CNOP-I has been further employed to explore the target observation of typhoon. The sensitive region could be identiifed by using the CNOP-I approach. The forecast skill may be improved by adding more adaptive observations in the sensitive region. The CNOP-P approach has been applied also to Kuroshio path anomalies, nonlinear stabilities of thermohaline circulation and grassland ecosystem. Here, we carried out a numerical simulation how to obtain the CNOP with the Burgers equation through building the tangent linear model and adjoint model. The result shows that the CNOP can be calculated by using the Burgers equation, the tangent linear model and the adjoint model with nonlinear optimizationalgorithm; It supplies a guide to a beginner to learn the CNOP and a reference for employing the CNOP to other applicable subjects.%介绍了条件非线性最优扰动（Conditional Nonlinear Optimal Perturbation，CNOP）的定义及其在大气和海洋等可预报性研究中的应用。

条件非线性最优扰动方法在黑潮目标观测研究中的应用

条件非线性最优扰动方法在黑潮目标观测研究中的应用张星;穆穆;王强;张坤【摘要】对近年来利用条件非线性最优扰动(Conditional Nonlinear Optimal Perturbation,CNOP)方法开展的黑潮目标观测研究进行了总结,主要包括日本南部黑潮路径变异的目标观测研究、黑潮延伸体模态转变的目标观测研究和源区黑潮流量变化的目标观测研究.通过计算这些事件的CNOP型扰动,发现这些事件的CNOP型扰动具有局地特征,可以作为实施目标观测的敏感区.理想回报试验结果表明,如果在由CNOP方法识别的敏感区内实施目标观测,则会大幅度提高上述事件的预报技巧.【期刊名称】《山东气象》【年(卷),期】2018(038)001【总页数】9页(P1-9)【关键词】目标观测;条件非线性最优扰动方法;黑潮【作者】张星;穆穆;王强;张坤【作者单位】青岛市黄岛区气象局,山东青岛266400;复旦大学大气科学研究院,上海200433;中国科学院海洋研究所,山东青岛266071;中国科学院海洋研究所,山东青岛266071【正文语种】中文【中图分类】P714.2引言在地球科学中，观测是发现科学事实和研究科学问题的重要手段。

尤其是在大气科学与海洋科学中，气象与海洋观测资料为大气与海洋模式提供初始条件，从而使科研工作者可以利用数值预报模式对将来发生的一些大气与海洋事件做出预报(或预测)。

因此，观测的准确程度将直接影响到数值预报结果的准确程度。

在数值天气预报发展的初期，气象学家就注意到某一区域数值预报的准确程度依赖于前期某一局部区域内初始条件的准确程度[1]。

但受限于当时的科学理论，研究人员往往是根据主观经验判断在哪些区域增加观测从而改善该区域的初始条件准确度。

直到20世纪90年代中期，“目标观测”的概念才被正式提出。

目标观测[2]又称适应性观测，是为了使将来关心的某一时刻(验证时刻)的某一区域(验证区域)的某个事件的预报结果更加精确，需在验证时刻之前的一个时间(目标时刻)在对关心的事件预报产生较大影响的区域(敏感区)进行更多的额外观测，来获得更多的观测资料。

条件非线性最优扰动台风目标观测框架系统

图 1 目标观测示意图（周菲凡，2009）。
条件非线性最优扰动（CNOP）台风目标观测框架系统是将 CNOP 方法（Mu et al. 2003）应用于台风目标观测进行敏感区识别的框架系统(如图 2)。目前，它主要基于美国国家大气研究中心（NCAR）与宾夕法尼亚大学共同发展的第 5 代中尺度模式（简称 MM5 模式； Dudhia 1993）及其伴随系统（MM5_ADJ; Zou et al. 1997）和大规模非线性优化算法谱投影梯度算法（SPG2；Birgin et al. 2001）。首先，由 MM5 模块（即 MM5 模式的前处理部分）提供初、边值条件，接着
1，对 cnop-target 文件夹的说明该文件夹是利用 MM5 模式及其伴随系统 MM5_ADJ 计算 CNOP、第一线性奇异向量（LSV）的文件夹。其中，子文件夹 MM5 为 MM5 模式的前处理部分；子文件夹 cnop、lsv、 mkobs、以及 readout 为新创建的文件夹。其余子文件夹采用的是 MM5_ADJ 里的子文件夹。 cnop 文件夹用以计算 CNOP；lsv 文件夹用以计算 LSV。 mkobs 文件夹用以产生 CNOP、LSV 计算需要的一些量，如初始扰动 pert.D，初始基本态 U0.IC, 预报时刻基本态 Uth.D(需改名为 Ut_BAS.SAVE)，以及计算所需 psdc.txt 文件。上述文件生成后需拷贝到 cnop 或 lsv 文件夹中。 readout 文件夹用以把计算结果转换成二进制的形式。
对 cnop‐target 文件夹的说明……..…….…….…….…….…….……………..4 CNOP 计算前准备工作……..…….…….…….…….…….………………….….4

条件非线性最优扰动及其在天气气候目标观测中的应用

CNOP’s applications
Spring prediction barrier
Mu et al., GRL,2007 Mu et al., JGR, 2007 Duan et al., JGR,2009 Yu et al., QJRMS,2009
et al., QJRMS, 2013）
（王斌和刘娟娟，2011，气象学报；Qin
目标观测
又称适应性观测：是指为了最大限度地提高未来某一时刻（验证时刻）某一地区（验证区域）的预报技巧，通过某些方法在此之前的某一时刻（目标观测时刻）某一区域（敏感区）增加观测。
初始扰动非线性预报模式初始基态初始时刻预报时刻
2
CNOP的物理意义
1. The initial error which have the largest effect on the uncertainty at prediction time (Mu et al.,
2007; Duan et al., 2005; 2009).
Blocking event, Mu and Jiang, 2008)
• 线性奇异向量(LSV)
⎧ ∂ (δ w) ∂F + |w=U ( w, x, t ; P )δ w = 0 ⎪ ∂w ⎨ ∂t ⎪δ w | = δ w t =0 0 ⎩
δ w( x , t N ) = M t
0 −tN
Methods to identify the sensitive areas
u SV (Palmer et al.1998)
u Adjoint Sensitivity (Ancell and Mass 2006) u ET (Bishop and Toth 1999) u EKF (Hamill and Snyder 2002)

数值分析在气象数值预报中的应用

数值分析在气象数值预报中的应用气象数值预报是一项重要的天气预测和天气研究工作，它通过数值模型对大气运动进行数值模拟和预报，为人们提供准确的天气信息。

而数值分析作为一种重要的分析方法，在气象数值预报中发挥着关键的作用。

一、数值模型的构建与分析气象数值预报通常使用大气环流模型来模拟和预测大气的运动。

这些模型基于方程组和初始条件，通过计算机模拟大气的运动和变化。

在数值模型的构建和分析过程中，数值分析起到了至关重要的作用。

首先，数值分析可以对观测数据进行插值和平滑处理，以减小观测数据的误差和扰动。

通过将观测数据插值到离散点上，再进行平滑处理，可以获得更加准确和稳定的初始场，提高数值模型的预报精度。

其次，数值分析还可以对数值模型中的方程进行数值求解。

大气运动方程是一个复杂的非线性方程组，很难通过解析的方法得到解。

数值分析通过离散化和迭代的方法，对方程进行数值求解，得到数值模型的数值解，从而预测未来一段时间内的天气变化。

此外，数值分析还可以对数值模型的预报结果进行评估和分析。

通过与实际观测数据进行对比，可以评估数值模型的准确性和可靠性，进一步完善和优化数值模型的参数和算法。

二、数值模型在气象数值预报中的应用数值模型在气象数值预报中的应用主要包括以下几个方面：1. 预报短期天气变化：通过对数值模型进行初始化和运行，可以获得未来数小时或数天的天气预报。

这些预报结果包括降水、温度、风速等天气要素的变化情况，为人们提供准确的短期天气预报。

2. 预报中期和长期气候趋势：数值模型还可以预报中期和长期的气候趋势。

通过对大气环流的模拟和预测，可以获得未来一周或一个月的气候变化情况，提供给农业、交通等部门参考和决策。

3. 预测灾害性天气事件：数值模型在气象灾害预测和预警中发挥着重要作用。

例如，利用数值模型可以预测台风、暴雨等灾害性天气事件的轨迹和强度，及时发布预警信息，减少灾害性天气事件带来的损失。

4. 优化决策和资源分配：气象数值预报可以为政府、企事业单位提供准确的天气信息，以便进行决策和资源的合理分配。

条件非线性最优扰动(CNOP)：简介与数值求解

黑潮和阻塞可预报性以及热盐环流和草原生态系统稳定性的研究。此外，ＣＮＯＰ — Ｉ方法也被应用于探讨台风目标观测的研究，利用ＣＮＯＰ一１：）５－法能够识别出台风预报的初值敏感区，通过观测系统模拟试验表明在初值敏感区增加观测能够有效改进台风的预报技巧。ＣＮＯＰ — Ｐ方法也在ＥＮＳＯ和黑潮可预报性以及热盐环流和草原生态系统稳定性研究中得到了应用。为了将ＣＮＯＰ方法应用于更多的领域，本文利用一个简单的Ｂｕｒｇｅｒｓ方程，介绍了如何通过建立Ｂｕｒｇｅｒｓ方程的切线性模式和伴随模式，从而利用非线性最优化算法计算获得ＣＮＯＰ。这一数值试验为将ＣＮＯＰ方法应用于更多的领域提供了借鉴。关键词：条件非线性最优扰动方法（ＣＮＯＰ），可预报性，目标观测
ＩｎｔｒｏｄｕｃｔｉｏｎａｎｄＮｕｍｅｒｉｃａｌＣｏｍｐｕｔａｔｉｏｎ
ＳｕｎＧｕｏｄｏｎｇ，ＭｕＭｕ，ＤｕａｎＷａｎｓｕｏ，ＷａｎｇＱｉａｎｇ，ＰｅｎｇＦｅｉ ’
４ＵｎｉｖｅｒｓｉｙｔｏｆＣｈｉｎｅｓｅＡｃａｄｅｍｙｏｆＳｃｉｅｎｃｅｓ，Ｂｅｉｊｉｎｇ１０００４９）
Ａｂｓｔｒａｃｔ：Ｔｈｉｓｐａｐｅｌｉｎｔｒｏｄｕｃｅｓｔｈｅｄｅｉｆｎｉｔｉｏｎｏｆｃｏｎｄｉｔｉｏｎａｌｎｏｎｌｉｎｅａｒｏｐｔｉｍａｌｐｅｒｔｕｒｂａｔｉｏｎ（ＣＮＯＰ）ａｎｄｔｈｅａｐｐｌｉｃａｔｉｏｎｓｏｆｔｈｅ

一种求解条件非线性最优扰动的快速算法及其在台风目标观测中的初步检验

一种求解条件非线性最优扰动的快速算法及其在台风目标观测中的初步检验王斌;谭晓伟【期刊名称】《气象学报》【年(卷),期】2009(067)002【摘要】条件非线性最优扰动(CNOP)是Mu等2003年提出的一个新的理论方法,它是线性奇异向量在非线性情形的推广,克服了线性奇异向量不能代表非线性系统最快发展扰动的缺陷,成为非线性系统可预报性和敏感性等研究新的有效工具.然而,由于以往CNOP的求解需要采用伴随技术,计算量相当巨大,限制了该方法的推广应用.为了克服这一困难,本文基于经验正交分解(EOF),提出了一种求解CNOP的快速算法,利用GRAPES区域业务预报模式实现了CNOP快速计算,并在台风"麦莎"的目标观测研究中得到初步检验,通过观测系统模拟实验(OSSE)检验了该方法确定敏感性区域(瞄准区)的有效性和可行性.试验结果表明,用快速算法求解的CNOP,其净能量随时间快速地发展,而且发展呈非线性.在台风"麦莎"个例的目标观测试验中,用快速算法得到的预报时间为24 h的CNOP可以有效地识别瞄准区,并通过瞄准区内初值的改善,可明显减少目标区域(检验区)内24 h累计降水预报误差.尤其,累计降水预报的这种改进效果能够延伸到更长时间(如72 h),尽管检验时间是设在第24小时.进一步分析发现,24 h累计降水预报误差的减少是通过利用瞄准区内改善的初值改进初始时刻台风暖心结构、高空相对涡度以及水汽条件等而得以实现的.【总页数】14页(P175-188)【作者】王斌;谭晓伟【作者单位】中国科学院大气物理研究所,LASG,北京,100029;中国科学院大气物理研究所,LASG,北京,100029【正文语种】中文【中图分类】P435【相关文献】1.条件非线性最优扰动方法在黑潮目标观测研究中的应用 [J], 张星;穆穆;王强;张坤;2.条件非线性最优扰动方法在适应性观测研究中的初步应用 [J], 穆穆;王洪利;周菲凡3.条件非线性最优扰动方法在黑潮目标观测研究中的应用 [J], 张星;穆穆;王强;张坤4.条件非线性最优扰动法在大气与海洋目标观测研究中的应用 [J], 穆穆;王强;段晚锁;姜智娜5.基于条件非线性最优扰动的目标观测中瞄准区不同引导性变量的影响试验研究[J], 谭晓伟;王斌;王栋梁因版权原因，仅展示原文概要，查看原文内容请购买。

条件非线性最优扰动在可预报性问题研究中的应用

差和模式参数误差同时存在的情形；（２）将拓展的ＣＮＯＰ方法分别应用于ＥＮＳＯ和黑潮可预报性研究，考察了初
始误差和模式参数误差对其可预报性的影响，揭示了初始误差在导致ＥＮＳＯ和黑潮大弯曲路径预报不确定性中的重要作用；（３）考察了阻塞事件发生的最优前期征兆（ＯＰＲ）以及导致其预报不确定性的最优增长初始误差（ＯＧＲ），揭示了ＯＰＲ和ＯＧＲ空间模态及其演变机制的相似性及其局地性特征；（４）研究了台风预报的目标观测问题，用ＣＮＯＰ方法确定了台风预报的目标观测敏感区，通过观测系统模拟试验（ＯＳＳＥｓ）和／或观测系统试验（ＯＳＥｓ），表
明了ＣＮＯＰ敏感区在改进台风预报中的有效性。具体地，台风ＯＧＲ的主要误差分布在某一特定区域，空间分布具有明显的局地性特征，在台风ＯＧＲ的局地性区域增加观测，有效改进了台风的预报技巧，该区域代表了台风
预报的初值敏感区。事实上，上述Ｅ１Ｎｉｌｆｏ事件、黑潮路径变异以及阻塞事件的ＯＧＲ的空间分布也具有明显的局地性特征，这些事件的ＯＧＲ刻画的局地性区域可能也代表了初值敏感区。关键词天气气候可预报性最优扰动中图分类号Ｐ４３３文献识别码Ａ文章编号１００６－９８９５（２０１３）０２ — ０２８１ — １６ｄｏｉ：１０．３８７８８．ｉｓｓｎ．１００６ — ９８９５．２０１２．１２３１９
第３７卷第２期２０１３年３月

条件非线性最优扰动(CNOP)方法穆穆

Editorial Committee of Appl.Math.Mech.,ISSN 0253_4827Article ID :0253_4827(2005)05_0636_11APP LICA TIONS OF NONLINEAR OP TIMIZA TION M ETH ODTO NUMERICAL STU DIES OF ATM OSPH ERICAND OCEANIC SCIENCESDUAN Wan _suo (段晚锁), MU Mu (穆穆)(LASG,Institute of Atmospheric Physics,Chinese Academy of Sciences,Beijing 100029,P.R.China)(Communicated by HUANG Dun,Original M e mber of Editorial Committee,AM M)Abstract :Linear singular vector and linear singular value can only describe the evolution ofsufficiently small perturbations during the period in which the tangent linear model is valid.With this in mind,the applications of nonlinear optimization methods to the atmospheric andoceanic sciences are introduced,which include nonlinear singular vector (NSV )andnonlinear singular value (NSVA),conditional nonlinear optimal perturbation (CNOP),andtheir applications to the studies of predictability in numerical weather and climate prediction.The results suggest that the nonlinear charac teristics of the motions of atmosphere and oceanscan be explored by NSV and CNOP.Also attentions are paid to the introduction of theclassification of predictability problems,which are related to the maximum predictable time,the maximum prediction error,and the maximum allowing error of initial value a nd theparameters .All the information has the background of application to the evaluation ofproducts of numerical weather and climate predic tion.Furthermore the nonlinear optimizationmethods of the sensitivity analysis with numerical model are also introduced,which can givea qua ntitative assessment whether a numerical model is able to simulate the observations andfind the initial field that yield the optimal simulation.Finally,the difficulties in the lack ofripe algorithms are also discussed,whic h leave future work to both computationalmathematics and scientists in geophysics.Ke y words :nonlinear optimization;wea ther;climate;predictability;sensitivity analysisChinese Library Classification :P456.7 Document code :A2000Mathematics Subject Classification :86A10IntroductionNu merical prediction of weather and climate essentially consists of solving a set of partial differential equ ations,which is usually referred as a model in ou r scientific literature,with proper initial and bou ndary values.Due to the great difficulties for the theoretical study posed by 636Applied Mathematics and Mechanics(English Edition) Vol.26No.5May 2005Published by Shanghai University,Shanghai,China Received Nov.21,2003;Revised Nov.4,2004P roject supported by the National Natural Science Foundation of China (Nos.40233029and 40221503)and the ANFA S P roject of Chinese Academy of Sciences (No.KZCX2_208)Corresponding author MU M u,Professor,E_mail:mumu@lasg.iap.ac.c nthe complexity and nonlinearity of atmospheric and oceanic motions,nu merical modeling has become a common approach in the world.Consequently,nu merical model shows its superiority in scientific background for the task of numerical weather and climate prediction.But the inevitable model deficiencies and initial errors will cause the uncertainties of forecast results.The studies of these uncertainties have become known as predictability problems ,which can be traced back to the last centu ry,some fu ndamental notions can be found in Refs.[1,2],and the latter Ref.[3].Until now,the predictability problems of numerical weather and climate prediction are still one of the important subjects, e.g.,the well_known international research programme on Climate Variability and Predictability (CLIVAR) .Many authors employed linear singular vector (LSV)and linear singular value (LSVA)to investigate the predictability of atmospheric and oceanic motions [4,5].Usually,it is assumed that the initial perturbation is su fficiently small such that its evolu tion can be governed by the tangent linear model (TLM )[6],which is derived by linearizing the corresponding nonlinear model about a basic solution.For a discrete TLM ,the forward p ropagator can be expressed as a matrix,and computing the LSV is reduced to calculate the singular vector corresponding to the maximum singular value of the matrix.LSV and LS VA were proposed by Lorenz [6]to stu dy the predictability of atmospheric motion.Then they were also utilized to study the finite _time linear stability [7]and to construct the initial perturbations field for the ensemble forecast at the Eu ropean Center for Medium Range Weather Forecast (ECMWF).Recently,this method has been used to find out the initial condition for optimal growth in a coupled ocean_atmosphere model of E1Nino_Southern Oscillation (ENSO),in an attempt to explore error growth and the predictability of the coupled model [8,9].The motions of atmosphere and ocean are nonlinear and complex and are usually governed by complicated nonlinear model.LSV and LS VA are established on the condition that the evolution of the initial perturbation can be described approximately by the linear version of the nonlinear model.This raises a few qu estions concerning the validity of TLM.Although there have been a few papers to add ress these concerns,their conclusions are accordant,that is,determining the validity of TLM in advance is difficult and essentially empirical [10-12].Therefore,in order to investigate the effect of nonlinearity on the predictability of atmospheric and oceanic motions,based on the research for several decades,Mu and Duan [13,14]proposed the concepts of nonlinear singular vector (NSV )and conditional nonlinear optimal perturbation,which are discussed in detail in Section 1and Section 2respectively.According to the factors causing the uncertainties of forecast results,the predictability problems are usu ally classified into two types,the first kind of predictability,which is related to the initial errors,and the second kind of predictability,which is to the model errors [15].The definition of model errors varies with the different authors [16].In this paper,we adopt the following one:If the initial value of the model is the true state,then the difference between the value of the forecast and the true state at the prediction time is called model error [17].From this definition,it is easily seen that there are many factors causing model errors.However,in this paper,we only consider the errors of the parameters in the model,which is generally regarded as one of the main problems in the model [16].On the basis of practical demands,M u e t al.[16]classified the three problems of predictability,which are referred to in Section 3.637Nonlinear Optimization Method in Atmospheric and Oceanic SciencesInitial errors and model errors are the dominant factors causing the uncertainties of forecast results.In order to reduce and identify them from the model,the meteorologist performed nu merous works with sensitivity ually,there are three approaches utilized in sensitivity analysis:numerical simulation,adjoint method,and LS V [18-20].For the approach of nu merical simulation,it is often blindfold and empirical in the process of choosing the control experiments [21].The adjoint method and LSV are based on the tangent linear model,which can only describe the evolution of small pertu rbations in the time period,in which the tangent linear model is valid [22].In Section 4,the authors will introduce the preliminary applications of nonlinear optimization method to the sensitivity analysis of the nu merical model.1 Nonlinear Singular Vector and Nonline ar Singular ValueThe model,which governs the motions of atmosphere and ocean,can be written as the following partial differential equations with initial and boundary values,w t +F (w )=0,w |t =0=w 0,(1)where w (x ,t )=(w 1(x ,t ),w 2(x ,t ),!,w n (x ,t )),x =(x 1,x 2,!,x n ),(x ,t )∀ #[0,T ],T <+∃,t is the time,F a nonlinear operator,w 0the initial state and a region in R n . Throughout this paper,the initial value problem (1)is assumed to be well posed,which means that for each initial value w 0,there exists a unique solution of Eq.(1)which depends on w 0continuously.Let U 0be the initial value of the basic state U (x ,t ).If u 0(x )is the initial pertu rbation superposed on U 0,U 0+u 0will evolve into U (x ,T )+u (x ,T )at time T .Then u (x ,T )is the nonlinear evolution of initial perturbation u 0(x ).The initial perturbation u *10is called the first nonlinear singular vector (NSV ),or the nonlinear optimal perturbation superposed on the basic state U (x ,t ),if and only ifI (u *10)=max u 0%u (T )%2%u 0%2.(2)The positive square root of I (u *10)is the nonlinear singular value (NSVA),which represents the growth rate of the first NSV in terms of the chose norm %&%.Here it is worthwhile to point out that there could exist several first NS Vs corresponding to the first NS VA [13].In addition to the first NSV defined above,we can also define the second NS V,u *20,by investigating the following conditional maximum problem:I (u *20)=max u 0 u *10%u (T )%2%u 0%2,(3)where u 0 u *10means u 0is orthogonal to all first NSV u *10,the positive square root of I (u *20)is called the second nonlinear singular value.Obviously,the n th(n =3,4,!,n )NSV and singular value can be defined in this manner step by step.The two_dimensional qu asi_geotropic model was used to study the NS V and NSVA [23].The nu merical results demonstrated that if the first NS V in terms of the chosen norm was sufficiently small,it could be approximated by the LSV.But for the large nonlinear optimal pertu rbation,the638DUAN Wan_suo and MU Mutangent linear model could not be used to describe its nonlinear evolution.Besides some types of basic states,there existed local nonlinear optimal perturbations,which corresponded to the local maximu m values of the fu nctional I (u 0).However there was no such phenomenon in the case of LSVs and LSVAs due to the absence of the corresponding TLM.These local nonlinear optimal perturbations usu ally possess larger norms compared with the first NSV,which corresponds to the global maximum value of fu nctional I (u 0).Although the growth rate of the local nonlinear optimal perturbations are smaller than the first NSVA,their nonlinear evolutions at the end of the time interval are considerably greater than that of the first NSV in terms of the chosen norm.In this case,the local nonlinear optimal perturbation could play a more important role than the global nonlinear optimal perturbation in the study of the predictability.Recently Durbiano [24]successfully computed the first six NS Vs of a shallow water model,and compared them with the LSV to explore the difference between them.It is clear from Refs.[23,24]that for the p redictability,we should find out all the local nonlinear optimal perturbations,then investigate their effects on the predictability.But this is inconvenient in practical applications.Besides,the local nonlinear optimal pertu rbations with large norm could be insignificant physically.All these weaknesses suggest that we should investigate the nonlinear optimal pertu rbation with constraint conditions.2 Conditional Nonlinear Optimal PerturbationLet U (x ,t )and U (x ,t )+u (x ,t )be the solutions of Eq.(1),of which the initial values are respectively U 0and U 0+u 0,u 0the initial perturbation on U (x ,t ).For the chosen norm %&%,the initial perturbation u 0 is conditional nonlinear optimal pertu rbation (CNOP)[26],if and only ifJ (u 0 )=max %u 0%∋%M T (U 0+u 0)-M T (U 0)%,(4)where M T is the nu merical model,or the propagator from 0to T ,the inequality %u 0%∋ is the constraint condition.It is easily shown from definition of CNOP that the nonlinear evolution of CNOP is the maximum among all the initial pertu rbations in %u 0%∋ .In the above,the constraint condition is simply expressed as a ball with the given norm %&%.Obviously,we can also investigate the situation that the initial perturbation belongs to some functional set,or satisfy some physical laws.2.1 A pp lication s of CNOP to predictab ility of ENSOEl Nino (La Nina)is a phenomenon of short_term climate variation happening in the tropical Pacific u ally the annual mean sea surface temperatu re (SST)over the equatorial Pacific takes on a strong asymmetry between the relatively warm western part of the basin,the region is called the warm pool,and the cooler eastern basin,called the cold tongue.In some years,the SST anomaly of the equatorial eastern Pacific is up to a few degrees,the phenomenon of which is usu ally as El Nino (La Nina).This phenomenon is associated with the atmosphere,and thus the term ENSO (El Nino_Sou thern oscillation)that incorporates the southern oscillation phenomenon is commonly used [25].Southern oscillation refers to a seesaw shift in su rface air pressu re at Darwin,Australia and the South Pacific Island of Tahiti.When the pressure is high at Darwin it 639Nonlinear Optimization Method in Atmospheric and Oceanic Sciences640DUAN Wan_suo and MU Muis low at Tahiti and vice versa.Though they originate in the tropical Pacific,they have an impact on weather and climate globally.ENSO has to be regarded as an inherently coupled atmosphere_ ocean mode.Conditional nonlinear optimal perturbation(CNOP)was used to study the optimal precu rsors of ENSO event within the frame of a simple coupled ocean_atmosphere model for ENSO[26].The results suggested that for the proper constraint condition,the CNOPs of the climatological mean state evolved into ENSO events more probably than the LSV.Consequently it was reasonable to regard CNOP as the optimal precursors of ENSO events.Observed anomalous monthly mean SST ((C)and depth of20(C isotherm(m)derived from NCEP ocean reanalysis for the equatorial eastern Pacific(5(S 5(N,150( 90(W)region verified the existence of these optimal precu rsors qu alitatively.The spring predictability barrier problem for ENSO event,which is one of the essential characteristics of ENSO and means that regardless of when a forecast is started,at the time of the year the anomaly correlation falls rapidly,was also investigated.In Ref.[14],the CNOPs of El Nino and La Nina events were compu ted.The results suggested that the error growth was enhanced in spring for El Nino event and su pp ressed in spring for La Nina event.That was to say,there was a tendency of spring predictability barrier for El Nino and not for La Nina.And the larger the initial pertu rbation was,the more notable the phenomenon of spring predictability barrier was.There was also evidence in the paper of M u and Duan[14]that LS V could only describe the evolution of the sufficiently small perturbation at the time interval when the tangent linear model was valid.For the large initial perturbation,LSV could not disclose that the effect of nonlinearity on the spring predictability barrier.Besides,LSV largely u nderestimated or overestimated the error growth of ENSO.To further investigate what causes the spring predictability barrier in the model,the CNOPs of El Nino and La Nina events with strong and weak_coupled ocean_atmosphere instability were also computed[14].It was shown that the strong_cou pled ocean_atmosphere instability during spring of the year was one of the causes of the spring predictability barrier.Some sensitivity experiments showed that the spring predictability barrier of ENSO event had the tendency for phase_locking to the annual cycle of the climate mean state.2.2 A pp lication s of CNOP to sensitivity analysis of thermohaline circulationThe sensitivity problem of the ocean thermohaline circulation has been one of the fundamental researches in climate variability.Due to the presence of several physically nonlinear feedbacks processes that govern the evolution of the thermohaline flow,the thermohaline circulation shows the strong sensitivity to the finite pertu rbations,which largely limits the predictability of the thermohaline flow.M any authors employed the LSV to study the sensitivity of the thermohaline to the initial perturbation[27].But the results obtained by LSV could not explain the asymmetry which responds to the sign of fresh water perturbation.In order to explore the effect of nonlinearity on the sensitivity of the thermohaline,M u et al.[28]adopted CNOP approach to analyze the different sensitivities of the thermohaline circulation to finite amplitude freshwater and salt perturbations.Within the frame of a simple model for the thermohaline circulation,the impacts of nonlinearity on the evolution of the finite amplitude freshwater and salinity perturbations were investigated.It was demonstrated that thereexisted an asymmetric nonlinear response to the sign of the finite fresh water perturbations.From the sensitivity analysis of the thermohaline circulation to the freshwater and salinity pertu rbations along the bifurcation diagram,it was shown that the system became unstable near the bifurcation diagram regime,and a finite pertu rbation could lead to a transition of the thermohaline circulation from an equilibriu m state to another one.3 Three Sub _Proble ms of PredictabilityWith the development of the hu man society and economy,people require to know the answers to the questions such as how large the prediction error is,and with given accuracy how long we can predict the weather or climate.With this in mind,M u et al.[16]classified three predictability problems in nu merical weather and climate prediction according to practical demands.Problem 1 Suppose that the initial observation u obs 0and the first given valu e of the parameter g are known,M t and M T are respectively the propagators from 0to t and T .At prediction time T ,the maximum allowing prediction error in terms of the chosen norm %&%A is%M T (u obs 0, g )-u t T %A ∋!,(5)where u t T is the true value of the state at time T .Then the maximu m predictable time,T !,can be estimated by the following nonlinear optimization problem:T !=max ∀%M t (u obs 0, g )-u t t %A ∋!,0∋t ∋∀.(6)Since the true state u t t cannot be obtained exactly,it is impossible to obtain the exact value of T !.However,if we know more information about the errors of initial value and the parameters,useful estimation on T !can be derived by using some methods.For example,assume that the errors of the initial value and the first given value of the parameters are known as follows:%u t 0-u obs 0%A ∋ 1, % t - g %B ∋ 2,(7)where %&%B is a norm measuring the parameters in the model.Then we can investigate the following nonlinear optimization problem:T !l =min u 0∀B 1, ∀B 2T u 0, |T u 0, =max ∀,%M t (u 0, )-M t (u obs 0, g )%∋!,0∋t ∋∀,(8)where B 1and B 2are the balls with centers at u obs 0, g ,and radius 1, 2,respectively.It is not difficult to prove thatT !l ∋T !.Thus T !l of Eq.(8)gives the lower bou nd of the maximum predictable time.Problem 2 Assu me that the initial observation u obs 0and the first value of the parameterg are given.At prediction time T ,the prediction error can be expressed as follows:E =%M T (u obs 0, g )-u t t %A .(9)The true state u t T cannot be obtained exactly,so E is not solvable.But if Eq.(7)holds,the prediction error E can be estimated by the following nonlinear optimization problemE u =max u 0∀B 1, ∀B 2%M T (u 0, )-M T (u obs 0, g )%A .(10)It is easy to prove that E ∋E u .Then E u gives the estimation of the upper bound of prediction 641Nonlinear Optimization Method in Atmospheric and Oceanic Scienceserror.Problem3 Assume that the initial observation u obs0and the first given value of the parameter g are available.At prediction time T,the allowing maximu m prediction error is Eq.(5).Then the allowing maximum initial error and the parameter error can be reduced into the nonlinear optimization problemmax =max%u-u obs0%A∋ 1,%-g%B∋ 2,if 1+ 2= ,then%M T(u obs0,g)-u t T%A∋!.Following the above idea,we can estimate .Investigating the problem-max =max%MT(u obs0,g)-M T(u0,)%A∋!,u0∀B1,∀B2, 1+ 2= ,(11)we can conclude that-ma x ∋max.In the abvoe problems,if the errors in the parameter can be ignored,and furthermore the model can be assumed to be perfect,the problems are the three ones of the first kind of predictability;on the other hand,if there exists no initial error,the problems become those of the second kind of predictability concerning the parameter error.Lorenz model[2]was adopted to demonstrate how to realize these above ideas nu merically[16].4 Nonlinear Optimization Me thod of Sensitivity Analysis with NumericalMode lThe nonlinear optimization method of the sensitivity analysis with numerical model[29]is discussed in detail in this section.Assume that Eq.(1)is a forecast model with model error,M t is its propagator,U obs0and U obs T are respectively the observational data with error at time0and T.How to determine the initial field U0that makes the results of the model at time T,U T=M T(U0),can optimally simulate the observational fields,U obs T.This problem can be formulated into the following nonlinear optimization problem:E=minUJ(U0),(12)where J(U0)=12(M T(U0)-U obs T)T W(M T(U0)-U obs T)is the objective function,W is theweighting coefficient matrix.Let E=min J(U0).For a given error bound!,there are two cases for E,E>!,0∋E∋!.(13)When testing a model,E>!that even if we get the optimal initial field U*,the model is not able to simulate the observation,U obs T properly in the given error bound!.Namely, no matter how we adjust U0,a satisfactory simulation for U obs T cannot be obtained.Then we can conclude that the model error is considerably large so that the model needs to be improved.In the 642DUAN Wan_suo and MU Mucase of 0∋E ∋!,the numerical solution U T =M T (U 0)and the observation U obs T have no apparent differences,which indicates that a satisfactory simulation can be obtained by adjusting the initial field U 0.It should be pointed out that,in this case,the model errors could be large too,which will be discussed later.With a given norm,defining a maximum allowable initial error !0,we have three cases now:%U *0-U obs 0% !0,(14a)%U *0-U obs 0%)!0,(14b)%U *0-U obs 0%!!0.(14c) When the model error is the model can simulate the movement of atmosphere very well and we can estimate the observation based on Eqs.(14a),(14b)and (14c).In Eq.(14a),a satisfactory simulation for the observation U obs T can be obtained from the existing observation U obs 0directly.We do not need to treat the initial field U 0of the model particularly,and some ordinary interpolation is enough.In Eq.(14b),a satisfactory simulation for U obs Tcannot be obtained from U obs 0directly.But if we improve the initial field U 0of the model (for example,by assimilation method),a satisfactory simulation can be obtained too.In Eq.(14c),the existing observation lacks enough information and cannot represent the real weather and climate states.If we want to obtain a satisfactory simulation for U obs T ,we should intensify the observational network to get more detailed observation than the existing one.In case that the observational error is small,the observation is close to the real development of atmosphere.In Eq.(14a),the model error is small and a satisfactory simulation for the observation U obs T is easily obtained by adjusting the initial field U 0.In Eq.(14b ),there are certain model errors but a satisfactory simulation can also be obtained by adjusting U 0in the allowable error bound.It is the case that an inaccurate model plus an inaccurate initial fieldproduces a satisfactory simulation.In Eq.(14c),the difference between U *0and U obs 0is too large,U obs 0is close to the real state,so U *0has no physical significance.We can conclude thatthe model error is large,a satisfactory simulation for U 0T is illusive and more work should be done to improve the numerical model.In practice,it is common to evaluate the model error by comparing the numerical simulation with a relatively accurate observation,or to assess the observation by comparing it with a relatively accurate numerical simulation.In these two cases,we can obtain some significant conclusions by above analysis method.However,due to some objective reasons,when the model and observation are both considerably inaccurate,the applicability of the nonlinear optimization method to sensitivity analysis is limited.In this case,when E >!,as mentioned above,the model error is considerably large,and the model needs to be improved.When 0<E ∋!,a satisfactory simulation can be obtained by adjusting the initial field U 0,but both model error and observational error may be large.Let !0be the observational precision that is usually known.If%U *0-U obs 0%!!0appears,the optimal initial field U *0is far from the real state ofatmosphere,and has no physical significance,which implies the model error is large.In other cases,we cannot obtain some significant conclusions by the nonlinear optimization method.In su mmary,we can get some instructive conclusions by applying the nonlinear optimization 643Nonlinear Optimization Method in Atmospheric and Oceanic Sciences644DUAN Wan_suo and MU Mumethod to sensitivity analysis of a numerical model.Xu et al.[30]performed a series of the sensitivity experiments with a two_dimensional quasi_geostrop hic model by using the nonlinear optimization methods.The results suggest that the nonlinear optimization method can give a qu antitative assessment whether a numerical model is able to simulate the observations and find the initial field that yield the optimal simulation,which are the main advantages of using the nonlinear optimization method.When the simulation results are apparently satisfactory,but the model error and the initial error may be large,under some conditions,the nonlinear optimization method can identify the type of the error that plays a dominant role.5 Conclusions and DiscussionsIn this paper,the authors review the applications of nonlinear optimization methods to the predictability study.Firstly,considering that linear singular vector(LS V)cannot describe the nonlinear motions of atmosphere and ocean,nonlinear singular vector(NS V)and conditional nonlinear optimal perturbation(CNOP)are successively used to investigate the predictability of atmosphere and ocean,and hope that science can gradu ally understand deeply the nonlinearity on the predictability.The results suggest that NS V and CNOP are useful tools in investigating the effects of nonlinearity on predictability.Secondly,the new classification of predictability problem provides an approach to estimating qu antitatively the maximum predictability time,the maximum prediction error and the allowing maximu m initial error and parameter errors of the model.All the information is of importance in the utilization of products of nu merical weather and climate prediction.Thirdly,nonlinear optimization method is applied to the sensitivity analysis with numerical model.It is shown that nonlinear optimization method can give a qu antitative assessment whether a numerical model is able to simulate the observations and find the initial field that yields the optimal simulation.The above results,which are related to the studies of the challenging problems of atmospheric and oceanic sciences,are obtained by using simple model.But it is clear from them that the nonlinear optimization methods can investigate the physical mechanism of atmospheric and oceanic motions,and further explore the nonlinear characteristic of atmospheric and oceanic motions.Therefore it is reasonable to expect that the nonlinear optimization methods will be well applied in complicated models.In fact a few difficulties will be faced in the utilization of nonlinear optimization methods. For the three su b_problems of predictability from the point of view of pure mathematics,there are still not rather ripe and effective algorithms.In Ref.[16],for the simple models,the lower bou nds of maximum predictable time and the maximu m allowable initial error are simply computed by filtration.However,the model governing the motions of atmosphere and ocean is generally complicated nonlinear model.For the complicated model,it is very difficult to use this method to solve the corresponding optimization problems.Therefore it is expected that more scientists can enjoy in the development of riper and more effective algorithms to work out the above optimization problems.The estimation of the maximum prediction error,which is equivalent to the calculation of CNOP,for the linear,and the simple quadratic equality or。

基于条件非线性最优扰动方法的集合预报试验

≤ Ε−1q, Ε−1q ≥ 1 V
V (Ε−1q)(Ε−1q)dV ,
(1) 总能量范数为
∫ ( ) q
2 TE
=
(q, q)TE
≤
q, −Ε−1q ≥ − 1 V
V
q(Ε−1q) dV ,
(2)
积分区域为整个大气 V. 其中, q 是位势涡度扰动. 定
义 q=E·ϕ, ϕ 是流函数扰动, E 为把流函数转化为位势
细节可参考Farrell和Moore[26].
1.3 试验设计
本研究采用“完美模式”假定. 已有的资料为 ECMWF1982/1983 年冬至 1993/1994 年冬(12 月至次年 2 月)每日 00 时流场的业务分析资料, 可用来作为积分模式的初始资料. 首先, 从已有资料中随机选择某日的流场资料, 此处定义为真实初始场. 用非线性模式积分这个给定的准确的初始场, 从而产生一个参考的大气态. 在该研究中, 这个大气态被认为是一个真实态, 所有的预报场都与这个真实态作比较. 接着, 采用三维变分同化方法来产生控制试验的初始
子 MTE* 可由 M* 推导得到 , 即 ΜTE* = Ε ⋅ Μ* ⋅ Ε−1 . 于
是,
δ J1(q0 ) = −(2Ε ⋅ Μ* ⋅ (Q0 + q0 ) ⋅ Ε−1
⋅[MT (Q0 + q0 ) − MT (Q0 )],δ q0 )TE
= (∇J1,δ q0 )TE .
(7)
采用的优化算法为谱投影梯度法(SPG2)[25], 该
涡度的线性算子; E−1 为把位势涡度转化为流函数的
线性算子.
条件非线性最优扰动为这样的初始扰动 q0*σ , 其
使目标函数

2017年两院院士推荐候选人信息表

附件1
2017年两院院士推荐候选人信息表一、个人信息
二、主要学历（6项以内）
三、主要学术经历（10项以内）
—2—
四、重要学术组织（团体）或重要学术刊物等的任（兼）职（6项以内）
—3—
五、在科学技术（或工程科技）方面的主要成就和贡献（3000字以内）
填写2-3项反映被推荐人系统的、创造性的学术成就和体现重大贡献的学术水平的主要工作。

说明在学科领域所起的作用、在学术界的影响和评价，以及（或）在国民经济和社会发展中的作用和贡献。

—4—
六、10篇（册）以内代表性论文、著作（包括教材）、研究技术报告、重要学术会议邀请报告
希望10篇（册）中含国内刊物发表的文章，每篇（册）应说明被推荐人的主要贡献，包括：提出的学术思想、创造性、研究工作的参与程度、学术刊物中的主要引用及评价情况等（200字以内）。

按以下顺序填写：
论文：作者（按原排序），题目，期刊名称，卷（期）（年），起止页码；
著作：作者（按原排序），著作名称，出版社，出版年份，出版地；
研究技术报告（未公开发表的重要报告）：作者（按原排序），报告题目，完成年份；
—7—
—8—
—9—
—10—
七、发明专利情况（10）项以内
请按顺序填写专利申报人（按原排序），专利名称，申请年份，申请号，批准年份，专利号。

并分别简述专利实施情况和被推荐人在专利发明和实施中的主要贡献（100字
—11—
八、重要科技奖项情况（10项以内）
按顺序填写全部获奖人姓名（按原排序），获奖项目名称、获奖年份、类别及等级（如：1999年国家自然科学二等奖，1998年中国科学院科技进步一等奖等），并简述
—12—
—13—。

条件非线性最优扰动在南海台风中的应用研究

条件非线性最优扰动在南海台风中的应用研究
王晓雷;朱克云;周菲凡
【期刊名称】《成都信息工程学院学报》
【年(卷),期】2010(025)006
【摘要】为了提高南海台风的数值预报,利用条件非线性最优扰动(CNOP)方法寻找对南海台风预报影响最大的区域(敏感区),并对该区域中会导致较大预报误差的初始扰动场进行了详细地分析.对两个南海台风个例地研究发现,CNOP的风场和温度场结构在各σ层上不完全一致.计算了CNOP各σ层上的干能量,并在垂直方向上做了积分,将积分后的能量大值区确定为敏感区.进一步对敏感区内温度场的垂直剖面进行分析,发现敏感区内,温度场呈现一定的斜压结构.研究还发现,敏感区与台风的对流活动区域不完全重合,这有利于台风适应性观测地实施.
【总页数】7页(P640-646)
【作者】王晓雷;朱克云;周菲凡
【作者单位】成都信息工程学院大气科学学院,四川,成都,610225;成都信息工程学院大气科学学院,四川,成都,610225;中科院大气物理研究所云降水物理和强风暴实验室,北京,100029
【正文语种】中文
【中图分类】P456.7
【相关文献】
1.条件非线性最优扰动方法在黑潮目标观测研究中的应用 [J], 张星;穆穆;王强;张坤;
2.条件非线性最优扰动方法在黑潮目标观测研究中的应用 [J], 张星;穆穆;王强;张坤
3.条件非线性最优扰动在热带气旋调控减灾中的应用初探 [J], 彭跃华; 张卫民; 郑崇伟; 项杰
4.条件非线性最优扰动方法在台风“风神”和“凤凰”相互作用过程中的应用研究[J], 王晓雷;周菲凡;朱克云
5.条件非线性最优扰动在长江中下游地区冬季暴雨中的应用研究 [J], 刘段灵;孙照渤;彭世球
因版权原因，仅展示原文概要，查看原文内容请购买。

气象集合预报的研究进展

气象集合预报的研究进展彭勇;王萍;徐炜;周惠成;王本德【摘要】Ensemble prediction is an important development direction of numerical prediction. Compared with the single deterministic prediction, the ensemble prediction considers the uncertainties associated with the initial conditions and models,and its results can reflect different possible weather conditions in the future and therefore the ensemble prediction can provide the information which the deterministic prediction cannot (e.g. ,the reliability of the forecast). Precipitation information is the core input of the hydrological forecast. The uncertainty of the single precipitation forecast leads to the uncertainty of the hydrological forecasting, which restricts the scientific values of the regulation of the water conservancy project However, the ensemble forecast can solve the uncertainty to some extent,and thus the weather ensemble prediction is valuable for the practical application in the hydrological field In order to help the hydrologists better apply the ensemble prediction, this paper reviews the current research status and achievements of the initial disturbance,model disturbance,and products of ensemble prediction based on the point of view of prediction system. The application of the products of precipitation ensemble prediction in the field of hydrological forecast is then analyzed Finally, the outlook of the application of ensemble forecast in the field of hydrological prediction is presented%集合预报是数值预报发展的一个重要方向,相对于单一的确定性预报,它考虑了初值及模式的不确定性,其结果反映了未来天气的多种状况,能为用户提供确定性预报所不能提供的信息(如预报结果的可信度等).降水信息是水文预报系统的核心输入,单一的降水预报存在着不确定性,从而使得水文预报也存在不确定性.这种不确定性制约了水利工程调度的科学性.集合预报在一定程度上能解决这种不确定性.从而在水文领域,气象集合预报具有较高的应用价值.为了水文工作者能够更好地应用集合预报,本文首先基于系统的角度,对集合预报系统中的初值扰动、模式扰动和集合预报产品等研究现状及成果进行了评述,然后分析了降水集合预报产品在水文预报领域中的应用情况,最后对集合预报在水文领域中的应用进行了展望.【期刊名称】《南水北调与水利科技》【年(卷),期】2012(010)004【总页数】8页(P90-96,126)【关键词】集合预报;扰动方法;集合预报产品;水文集合预报【作者】彭勇;王萍;徐炜;周惠成;王本德【作者单位】大连理工大学建设工程学部水利工程学院,辽宁大连116023;大连理工大学建设工程学部水利工程学院,辽宁大连116023;大连理工大学建设工程学部水利工程学院,辽宁大连116023;大连理工大学建设工程学部水利工程学院,辽宁大连116023;大连理工大学建设工程学部水利工程学院,辽宁大连116023【正文语种】中文【中图分类】P339;P444数值预报模式可以提供短时、短期、中期以及气候等各个时间尺度的预报，是制作气象预报不可缺少的基础和手段。

QuikSCAT风场在台风风暴潮计算中的应用

QuikSCAT风场在台风风暴潮计算中的应用付翔;仉天宇;于福江【摘要】Model wind is usually used as forcing field in the operational storm surge simulation. The maximum wind speed radius is one of the parameters which is the hardest to be determined in most wind models. By using QuikSCAT wind data fitting maximum wind speed radii, the storm surge forcing field is assimilated. As a result, the simulated surge is more approximate to the observations.%业务化风暴潮计算中多采用模型风场的算法来给出风暴潮强迫场.最大风速半径是模型风场最难确定的参数之一.利用QuikSCAT卫星风场数据拟合台风最大风速半径是确定该参数的有效方法.将拟合得到的最大风速半径代入模型风场计算风暴潮的驱动场,模拟的沿岸风暴增水与实况更为接近.【期刊名称】《海洋预报》【年(卷),期】2012(029)004【总页数】5页(P18-22)【关键词】QuikSCAT风场;模型风场;最大风速半径;风暴潮【作者】付翔;仉天宇;于福江【作者单位】国家海洋环境预报中心,国家海洋局海洋灾害预报技术研究重点实验室,北京100081;国家海洋环境预报中心,国家海洋局海洋灾害预报技术研究重点实验室,北京100081;国家海洋环境预报中心,国家海洋局海洋灾害预报技术研究重点实验室,北京100081【正文语种】中文【中图分类】P444QuikSCAT卫星是美国太空总署（NASA）于1999年6月发射的一颗海洋科学探测卫星，它上面所携带的SeaWinds（洋面风场散射探测仪）能够探测洋面10 m 的风向风速[1]。

条件非线性最优扰动方法在台风＂风神＂和＂凤凰＂相互作用过程中的应用研究

在台风多发的季节，西太平洋上常常会同时
出现两个甚至多个台风的情况，当同时出现的台
风之间距离较小时，往往产生相互作用的现象，
如相互旋转、吸引以及合并、逃离等，导致台风
路径异常而难以预报。很长时间以来，双台风相互作用的问题一直为气象学者所关注，早在１９２１
第２９卷第２期２０１３年０４月
热
带
气
象
学
报
Ｖｏ１．２９．ＮＯ．２Ａｐｒ．，２０１３
ＪＯＵＲＮＡＬＯＦＴＲＯＰＩＣＡＬＭＥＴＥＯＲＯＬ０ＧＹ
王Ｉ毙宙，周菲凡，朱克．条件非线性最优扰动方法在台风 “ 风神” 和“ 凤凰 ” 相互作用过程中的应用研究［Ｊ］．热带气象学报，２０１３，２９（２）：２３５－２４４
文章编号：１００４．４９６５（２０１３）０２．０２３５．１０
条件非线性最优扰动方法在台风 “ 风神＂和 “ 凤凰＂
相互作用过程中的应用研究
王晓雷一，周菲凡。，朱克云
（１成都信息工程学院大气科学学院，四川成都６１０２２５；２．武汉市气象局，湖北武汉４３００４０；３．中国科学院大气物理研究所云降水物理与强风暴实验室，北京１０００２９）
年，Ｆｕｊｉｗｈａｒａ［１】就对一般双涡旋的相互作用问题进行了理论分析和试验研究，后来这种双涡旋之

【国家自然科学基金】_条件非线性最优扰动_基金支持热词逐年推荐_【万方软件创新助手】_20140730

2008年序号 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
科研热词条件非线性最优扰动风生涡旋集合预报运筹学约束最优化热盐环流滤子最大值原理收敛性年代际变率奇异向量变分法中期 qp-free方法 ncp函数
推荐指数 3 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2013年序号 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
2013年科研热词条件非线性最优扰动高影响天气预报非脆弱非线性切换系统解析解观测系统模拟试验蒙特卡罗模拟草原生态系统色噪声网络化控制系统瞬态概率密度相互作用目标观测气候模拟结果不确定性最优有界控制最优扰动敏感区振动与波容错控制天气学天气可预报性双台风参数不确定性动态输出反馈公共李雅普诺夫函数保性能控制伽辽金法 t-s模糊模型 h∞控制推荐指数 3 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
推荐指数 2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2011年序号 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
2011年科研热词推荐指数条件非线性最优扰动 2 预测控制 1 非光滑 1 陆面过程模式 1 遗传算法 1 连续搅拌罐反应器 1 观测系统模拟实验 1 经济性能 1 粒子群算法 1 目标观测 1 条件非线性最优扰动(cnop) 1 最大预报误差上界 1 台风 1 参数优化 1 动态实时优化 1 分层控制 1 伴随方法 1 "on-off"开关 1

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

基于月动力延伸预报最优信息的中国降水降尺度预测模型
顾伟宗1,2陈丽娟1张培群1李维京1李想1刘绿柳1
1. 国家气候中心，中国气象局气候研究开放实验室，北京，100081
2. 山东省气候中心，济南， 250031
摘要
利用国家气候中心月动力延伸预报结果、NCEP/NCAR再分析资料和中国160个站观测资料，通过计算两次相关的方法，获取最优预报信息作为建立降尺度预测模型的预测因子，提取的最优预测因子同时满足既是观测环流要素场影响降水的关键区域，又是模式要素场预报的高技巧区域两个条件。

结合挑选出的最优预测因子，利用最优子集回归建立月平均降水的降尺度预测模型。

文中设计了消除预测因子和预测量的线性趋势值后建立预测模型（方案1）和直接利用原始资料建立预测模型（方案2）两种方案。

经过独立样本检验，发现这两种方案建立的预测模型都能够提高月尺度降水预测，方案1对月尺度降水预测的距平相关系数平均可达0.35。

利用该方案对超前时间分别为0、5、10 d的月动力延伸预报产品进行月降水的降尺度预测表明，模式初值信息不仅影响月动力延伸预报结果，也影响降尺度应用效果，利用超前时间为0和5 d的月动力延伸预报结果进行降水降尺度预测可在业务中参考。

此外，降尺度预测模型中选取的预测因子不仅在统计上是显著的，同时也具有清楚的物理意义。

关键词月动力延伸预报, 最优信息, 降水, 降尺度预测
资助课题：国家自然科学基金项目(40675039)、国家科技部科技支撑计划项目(2006BAC02B04)和中国气象局业务建设项目（气候模式产品在区域气候预测业务中的应用）。

作者简介：顾伟宗，研究方向：气候预测研究和应用。

Email: longmarch529@
通讯作者：陈丽娟，研究方向：气候预测研究和应用。

Email:chenlj@
中图法分类号P456.2 P456.7
2007-09-04收稿,2007-11-07改回.
Downscaling precipitation prediction in China based on Optimization Information Extracted from Monthly Dynamic Extended Range Forecast.
Gu Weizong1,2, Chen lijuan1, Zhang peiqun1, Li Weijing1, Li Xiang 1and Liu Lvliu1
1. National Climate Center, Laboratory for Climate Studies,, CMA, Beijing .100081
China
2. Shandong Climate Center, Jinan, 250031 China
Abstract
Using monthly dynamic extended range forecast (DERF) products, NCEP/NCAR reanalysis data and 160 station data in China, we have extracted optimal predictors which significantly influence precipitation. The predictors are selected from high skill regions of DERF and high correlation between observing precipitation and other variables. Downscaling monthly anomaly precipitation are predicted using the optimal subset regression from selected predictors. Two schemes are designed in the paper. One uses the detrending data to establish the prediction equations (scheme 1), the other uses the original data to establish the prediction equations (scheme 2). The results show that the two schemes can improve the skills of monthly precipitation forecast whereas scheme 1 has higher skills than scheme 2. The anomaly correlation coefficient between scheme 1 and observation can reach 0.35. Downscaling tests indicate that initial fields could have great influence on model results and downscaling precipitation prediction. Using scheme 1 and DERF results from different leadtime including 0 day, 5 days and 10 days, it shows that when lead time is 5 days and 0 day, the downscaling precipitation
prediction could be used as a good reference in monthly climate prediction. In the downscaling model, selected predictors can pass statistic tests and have clear physical meanings. The downscaling methods can be applied in operational climate prediction.
Key words Dynamic extended range forecast, Optimization information, Precipitation, Downscaling。