三阶龙格-库塔时间分裂显式算法的误差分析

分析了新一代非静力中尺度数值模式中常用的三阶龙格-库塔时间分裂显式算法(RK3)的稳定性和误差性质,特别是分析了空间中央差分和迎风偏斜两种不同情况下该算法不同的稳定性和误差性质。运用数学软件先进的符号计算功能,分析了该算法涉及的复杂高阶、高次幂振幅矩阵的特征值性质;并通过一维线性声波-平流方程组的数值模拟实验,检验了时间分裂算法的模拟效果。对振幅矩阵特征值模的表达式进行高阶的级数展开,得到了该算法的分裂误差项的公式;而且,由于特征值模的公式保留了较高阶项,可以同时分析迎风偏斜和中央差两种空间差分格式的分裂误差性质。根据分裂误差项公式,定量地比较了三阶和二阶龙格-库塔格式(RK2)的分裂误差大小以及误差与小时间步数的关系,发现迎风格式RK3的分裂误差明显小于RK2的误差,并具有更好的稳定性质。空间中央差格式的分裂误差项具有更高阶数,比迎风格式具有更小的时间分裂误差。对于各种不同波长的特征值分析和采用中央差格式的数值模拟,也进一步证实空间差分采用中央差时,RK3时间分裂显式算法在不同方向传播的声波振幅几乎没有差别。另外,误差公式以及数值试验结果说明RK3的分裂误差也略小于Adams-Bashforth-Moulton分裂显式法的分裂误差。     

三维指数型迎风格式以及在环境数值模拟中的初步应用

为了适应大气、海洋的环境数值模拟需要,在前人工作基础上提出了三维随流格式和指数型迎风格式.随流格式能使方程对流项中的对流速度误差明显减小.三维指数型迎风格式能有效抑制对流占优问题的数值振荡,并用简单的七点差分格式获得O(h2)精度的三维模式数值结果.因此,所建的模式具有格式简单,计算稳定性良好和计算精度较高的优点.初步计算表明,这个模式能有效应用于强迫-耗散型大气、海洋运动和环境问题的数值模拟.     

平流方程的数值研究

二维平流方程分别用二阶差分方案、四阶差分方案、一维三次样条函数方案及谱导数方案(假谱方案)进行长时间的数值积分,其中风场是给定的均匀旋转风场或形变风场,并分别以不同大小的圆锥状的物理量分布作为初值。数值试验的结果表明,谱导数方案的精度最高,没有位相误差;其次是一维三次样条函数方案;再其次是四阶差分方案;二阶差分方案的精度最差,畸变严重,位相误差大。 对上述四种方案给出用权重系数形式表示的统一的导数计算公式,看到高精度的导数计算公式是非局地性的。     

非均匀网格下的高阶精度中央差分格式:理论推导和理想试验

传统的高阶精度有限差分格式通常是在均匀网格的基础上推导得到的,在非均匀网格的情况下它会出现精度退化的问题。基于泰勒展开方法构造了一种适用于非均匀网格的2阶、4阶和6阶精度中央有限差分方案,利用Burgers方程和一维平流方程对新方案的性能进行测试,着重分析新方案对其误差大小及分布形态的改进效果。数值模拟结果表明:在非均匀网格下,提高差分方案的精度可明显减小数值解误差(降低了70%~88%),特别是当差分精度从2阶提高到4阶的时候。同时,高阶精度方案在梯度变化较大或者网格距较粗区域的模拟结果更有优势,4阶和6阶精度方案在以上区域的误差远小于2阶精度方案。方案可用于提高数值天气预报模式中非均匀分层模式的垂直差分计算精度。      

一类计算稳定性好的显式平流差分格式

通常的显式平流差分格式，如迎风格式，Lax-Wendroff格式等，均是有条件稳定的,其稳定条件与差分网格的时、空步长有关。本文对线性和拟线性平流方程分别构造了一种计算稳定性好的显式差分格式。对前一格式，本文严格证明了它的无条件稳定性及收敛性，并具有一阶精度；对后一格式，由于非线性方程的限制，本文用数值试验研究了它的计算稳定性。     

中尺度模式MM4对高精度水汽输送#C$算法的敏感性试验

水汽输送的算法精度对中尺度模式的预报能力起有重要作用。本文在MM4的基础上用高精度的平流格式改进模式的水汽输送格式，使改进后的模式包含了对多种高精度的水汽输送格式的选择，如Prather格式和Bott格式。这些格式都是正定的，在理想风场的数值实验中几乎无数值耗散，无计算频散。我们在B网格中央差格式（原MM4中的格式）、Upstream格式和 Prather格式等模式选择项下，对1982年6月的一次梅雨期暴雨过程进行了数值对比实验。结果表明:1）模式的预报能力对水汽输送格式的精度较敏感， 2）高精度的水汽输送格式（Prather格式） 提高了模式对暴雨区、降水场结构、雨团活动的预报能力，较真实地模拟出了尺度较小的降水系统的活动。 高精度的水汽输送算法有助于提高中尺度模式的预报精度。     

几种数值插值方法的试验比较

该文首先引进３种精度较高的边界条件，借助于Ｓｐｌｉｎｅ函数，近似计算离散网格上的一阶空间微商，并通过数值试验进行比较，然后用Ｌａｇｒａｎｇｉａｎ、Ｓｐｌｉｎｅ和Ｈｅｒｍｉｔｅ方法对不同相对位置的插值点作插值计算，并给出一套较为简单适用的Ｈｅｒｍｉｔｅ方案计算公式。结果表明，这３类插值方法都能达到一定的精度，但其中以取周期性边界条件，用Ｈｅｒｍｉｔｅ方法算得的插值函数最为精确，其误差约为采用二阶或三     

高阶精度有限差分方案下的非跳点网格试验:基于浅水波方程

非跳点网格在模式动力—物理过程的耦合方面具有独特的优势,但是由于二阶精度差分方案下非跳点网格频散误差较大而很少被使用于数值天气预报模式。随着近年来数值模式计算精度的不断提高,非跳点网格在频散关系方面的计算误差是否会发生变化还有待研究。本文在高阶精度差分格式下通过浅水波方程对跳点网格和非跳点网格的频散关系进行理论分析和数值试验,主要得到以下结论:（1）在低波数区跳点网格的频散关系基本不随计算精度的提高而变化,但是非跳点网格下的频散关系则随着计算精度的提高而更加接近真实解。在四阶精度下,非跳点网格的频散关系已经非常接近跳点网格。（2）差分精度提高以后,在高波数区非跳点网格仍然存在频率极大值,而且极值中心随着计算精度的提高而逐渐向更高波数区移动。跳点网格在计算精度提高以后高波数区的频率仍然随波数单调增加,且更接近真实解。（3）在高阶精度非跳点网格模拟试验的基础上,结合高阶扩散项对高频短波进行滤除,可以得到与二阶精度跳点网格相接近的模拟结果。总之,在高阶精度有限差分方案下利用非跳点网格构造模式动力框架是一种比较可行的做法。     

应用于双向套网格模式的一种变格距差分解法

本文提出了一种用于双向套网格模式的变格距差分计算方案。该方案在不同格距的网格区采用不同精度的差分格式，它自然地连接粗细网格，避免了一般套网格方案在粗细网格相重合点上进行的重复计算。用解析法和数值试验证明了:它与其它一些变格距差分格式相比，对短波的穿透能力有明显改进，虚假的反射也较小。应用该方案建立了正压原始方程双向套网格模式，并采用空间分解和时间分解计算方法。这不仅使二维问题转化为二个一维问题，而且二维套网格也可简化为一维均匀网格和一维套网格两部分，从而使计算和程序简化。用理想场为初值所做的一系列数值试验表明，该模式中的波可以自由进出粗细网格区，计算稳定。最后，还用该模式做了台风路径预报试验，给出了一些试验结果。     

变步长显式完全平方守恒差分格式

综合隐式完全平方守恒差分格式和显式瞬时平方守恒差分格式的优点，针对一类非线性发展方程构造了一种通过自动调节时间步长来保持平方守恒性的显式差分格式。它基于加小耗散的思想，但又与小耗散法有所不同。本文取的耗散项不是一般的人工耗散，而是取能够弥补由于时间离散所产生的截断误差的所谓（时间）协调耗散。因此，该格式具有较高的时间精度。在数值试验中，该类格式可取得满意的效果。     

A Two-Step Shape-Preserving Advection Scheme

This paper proposes a new two-step non-oscillatory shape-preserving positive definite finite difference advection transport scheme, which merges the advantages of small dispersion error in the simple first-order upstream scheme and small dissipation error in the simple second-order Lax-Wendroff scheme and is completely different from most of present positive definite advection schemes which are based on revising the upstream scheme results. The proposed scheme is much less time consuming than present shape-preserving or non-oscillatory advection transport schemes and produces results which are comparable to the results obtained from the present more complicated schemes. Elementary tests are also presented to examine the behavior of the scheme.     

On the forward-in-time upstream advection scheme for non-uniform and time-dependent flow

Summary The classical forward-in-time upstream advection scheme for uniform flow field has been extended to include non-uniform and time-dependent advective flow. This generalised scheme is described in one dimension for an advective flow which varies both in time and in space. The classical upstream advection scheme is only first-order accurate both in time and in space if the advective flow is not uniform. Higherorder accuracy in both time and space, however, can be easily obtained in the generalised scheme.This generalised scheme with third-order accuracy is applied to the one-dimensional inviscid Burgers equation (socalled self-advection problem), two-dimensional steady flow, and to a time-split shallow water equation model. The results are compared with those obtained from the Takacs' (1985) scheme and from a standard third-order semi-Lagrangian scheme, and also with those obtained from the fourth-order Lax-Wendroff scheme of Crowley (1968) in the time-split shallow water equation model. It is shown that the generalised scheme performs as well as, but is more efficient than, the standard semi-Lagrangian scheme with same order. It is much more accurate than the Takacs' scheme which has large dissipation errors, especially for the flow with strong deformation. In contrast, the generalised scheme has very weak dissipation and has much better dispersion and shapeconserving properties. Although the fourth-order Lax-Wendroff scheme has higher accuracy and can give more accurate numerical solutions for uniform advective flow or solid rotational flow (Crowley, 1968), it is inferior to the generalised third-order scheme for non-uniform flow with strong deformation or large spatial gradients. This generalised scheme, therefore, has considerable application potential in different numerical models, especially for the models using time-split algorithms.With 8 Figures     

PRM标量平流方案在GRAPES全球预报系统中的应用

如何更好地模拟水物质的分布,对于数值天气预报效果的改进,特别是对于更好地模拟降水过程,具有重要的意义。半拉格朗日模式中的标量平流计算要求做到高精度、守恒、正定和保形,但GRAPES_GFS (Global-Regional Assimilation and PrEdiction System, Global Forecast System) 中采用的QMSL(Quasi-Monotone Semi-Lagrangian)平流方案在水汽的强梯度、不连续区域计算精度较低,且不能做到严格守恒。本研究借鉴计算流体力学领域的研究进展,将一个基于分段有理函数的物质平流方案PRM(Piecewise Rational Method)引入GRAPES_GFS中,按照通量形式求解水汽方程,并对极区进行了混合等技术处理。通过一系列理想试验对两种平流方案进行了对比,证明了PRM方案精度较高,特别是在水汽梯度大的区域优势明显,频散、耗散误差较小,守恒、保形性也要好于QMSL方案。通过对GRAPES_GFS中批量预报试验效果的检验,验证了PRM方案可以有效地改进模式对水物质分布的模拟,提高了降水的预报效果,对模式综合预报性能的提升也有明显作用。     

A comparison of high-order explicit and non-oscillatory finite difference advection schemes for climate and weather models

Summary The Euler equations govern the behavior of a fluid in motion. They have long been used as a test-bed for assessing the accuracy and efficiency of numerical schemes for solving them. This study focused on advection dominated flows so all other terms in the equations are omitted. The study examined only explicit schemes and does not address the many alternative approaches such as semi-Lagrangian and implicit formulations. The schemes examined here were applied to test cases of increasing complexity as it is well-known that a particular scheme may work very well on some test problems but fail on others. To avoid such problems, we began with the standard tests such as the advection of sine waves and various other one-dimensional shapes, then moved on to two-dimensional problems and finally tested cases that have many scales.Summarizing, the results from the range of test cases, it was found that the schemes that perform best are high-order upwind schemes, with diminishing returns above fifth- and sixth-order.Finally, the schemes were applied to cases where scalar advection requires that constraints such as positive-definiteness be satisfied. For these problems, the Flux Corrected Transport (FCT) and weighted essentially non-oscillatory (WENO) schemes were also applied. Similar conclusions were drawn, namely, that fifth- and sixth-order FCT and WENO schemes produced excellent simulations. However, these schemes were more computationally expensive than the standard high-order upwind and centred schemes.     

Numerical simulation of plume-advection over a hill with coupled models. Part I: Numerical solution of the transport-equation

Summary Four numerical methods suitable for the calculation of transport are compared on a linear two-dimensional advection equation. It is found that a mass-compensation algorithm which removes negative values has also a beneficial effect on the accuracy of schemes which are of higher order but are not positive. It is shown by numerical experiments, that the upstream spline advection scheme combined with a mass compensation is prefereable for some applications rather than other high accurate positive schemes like the flux-corrected transport scheme.With 3 Figures     

用氚的模拟比较被动示踪物平流输送方案

使用一个太平洋海盆尺度环流模式模拟氚的分布,并分别采用二阶中央差(CD)、通量修正输送法(FCT)、分段抛物线法(PPM)、多维正定平流输送法(MPDATA)及二阶矩法(SOM)计算流场对氚的平流输送,通过结果比较考察算法间的差异.模拟试验设定为完整物理过程情形( CASEl)及仅含平流、对流情形(CASE2).CAS...     

Application and numerical experiments with a highly-accurate advection scheme in a regional transfer model

Summary This paper examines a 19-level regional transfer model, EM3. The sensitivity of the results to different transfer schemes and effectiveness of using a high-precision scheme in raising modeling accuracy is examined. The use of a good scheme to improve the ability to simulate horizontal diffusion close to reality is also explored. EM3 was run with different precision schemes, i.e., a high-precision second-order moment conservation prather scheme (SOM for short) and an anti-diffusion Smolarkiewicz scheme (Smolar for short). The model was run 70 hrs. Results indicate that EM3 is much more sensitive to the precision of the schemes. SOM and Smolar, were compared, the former showing very weak numerical diffusion and computational dispersion so that positive solutions were kept. It is shown that i) numerical diffusion in SOM is one order of magnitude less compared to the latter; ii) the central intensity of transferred SO₂ increases by 4–5 times and the average motion trajectory of the concentration center is much improved in SOM; iii) SOM allows EM3 to precisely reflect the real horizontal diffusion. Therefore, choice of a high-precision advection scheme for EM3 will contribute greatly to an increase in its simulation accuracy.With 4 Figures