BP模型双样本训练方法

根据气象预报应用情况，分析ＢＰ模型训练人工神经元网络中存在的问题，并提出双样本训练方案。试验表明，采用双样本训练的网络，实际预报能力较高。     

A modeling of the structure and favorable H-docking sites and defects for the high-pressure silica polymorph stishovite

Employing first-principles methods, the docking sites for H were determined and H, Al, and vacancy defects were modeled with an infinite periodic array of super unit cells each consisting of 27 contiguous symmetry nonequivalent unit cells of the crystal structure of stishovite. A geometry optimization of the super-cell structure reproduces the observed bulk structure within the experimental error when P1 translational symmetry was assumed and an array of infinite extent was generated. A mapping of the valence electrons for the structure displays mushroom-shaped isosurfaces on the O atom, one on each side of the plane of the OSi₃ triangle in the nonbonded region. An H atom, placed in a cell near the center of the super cell, was found to dock upon geometry optimization at a distance of 1.69 Å from the O atom with the OH vector oriented nearly perpendicular to the plane of the triangle such that the OH vector makes a angle of 91° with respect to [001]. However, an optimization of a super cell with an Al atom replacing Si and an H atom placed nearby in a centrally located cell resulted in an OH distance of 1.02 Å with the OH vector oriented perpendicular to [001] as observed in infrared studies. The geometry-optimized position of the H atom was found to be in close agreement with that (0.44, 0.12, 0.0) determined in an earlier study of the theoretical electron density distribution. The docking of the H atom at this site was found to be 330 kJ mol^–1 more stable than a docking of the atom just off the shared OO edge of the octahedra as determined for rutile. A geometry optimization of a super cell with a missing Si generated a vacant octahedra that is 20% larger than that of the SiO₆ octahedra. The valence electron density distribution displayed by the two-coordinate O atoms that coordinate the vacant octahedral site is very similar to those displayed by the bent SiOSi angles in coesite. The internal distortions induced by the defect were found to diminish rather rapidly with distance, with the structure annealing to that observed in the bulk crystal to within about three coordination spheres.     

两种垂直累积液态含水量产品的应用对比分析

结合雷达体扫工作模式以及风暴单体垂直结构分布，利用CINRAD气象产品软件包提供的基于单体和格点的VIL产品，对滨州SC雷达监测到的几例强对流风暴单体个例进行分析发现，两种VIL产品的变化趋势基本与风暴变化同步，对灾害性天气发生的预警具有较好的指示意义；一般情况下，基于单体的VIL数值大于基于格点的VIL数值，尤其是发展成熟、孤立的强风暴单体表现更加明显，可以作为判断风暴单体强度的定性指标；对于多单体风暴，由于风暴质心识别等因素会造成VIL数值误差增大，而雷达体扫模式的限制可能造成VIL数值失真。     

气候系统可预报性的全局研究

胞映射方法是对非线性系统进行全局分析的强有力工具。文中基于一个最大简化的地－气耦合模式,引入胞映射全局分析方法,对气候系统的可预报性进行了全局的研究。气候系统存在最大可预报期限,对于超过逐日预报可预报期限后的预报,用胞映射思想证明了平均值的可预报性,并得到了定量的结果。研究表明,耦合机制具有延长可预报期限的作用,观测精度的提高亦可延长可预报期限     

广东省后汛期强对流天气潜势预报方法研究

利用2007—2008年两年7—10月广东后汛期强对流天气出现时的雷达资料、对应的GRAPES模式资料以及地市台站上报的强对流天气发生的实况,把瞬时大风〉17.2 m·s^-1、冰雹、龙卷作为强对流发生的依据,对上述数据进行整理。根据广州热带海洋研究所中尺度模式的输出GRAPES资料,结合雷达CAPPI数据,计算单体的各层风速、温度、湿度、有效位能等环境特征量,将单体特征和模式计算的单体环境场要素以及强对流发生实况,通过多元逐步回归方法建立后汛期强对流天气潜势预报方程,据此对发生于广东省后汛期强对流天气（如雷雨大风、冰雹和龙卷风）进行0～1小时临近预报。用预报成功率（POD）、虚假警报率（FAR）和关键成功指数（CSI）衡量方法的预报性能。共有5540个有效样本参与回归计算,31个因子中有12个引入了回归方程,建立的预报方程在阈值取为0.26时,得到的预报成功率POD为0.73,虚假警报率FAR为0.61,关键成功指数CSI为0.338,各项指标均要好于前汛期预报性能;从实际的预报能力来看,在后汛期强对流潜势预报中,后汛期强对流潜势预报方法得到的空报率和漏报率都要低于前汛期,预报效果较好,可用于广东后汛期的强对流天气潜势预报中。     

无磷条件诱导铜绿微囊藻(Microcystis aeruginosa)释放挥发性有机化合物对莱茵衣藻(Chlamydomonas reinhardtii)的影响

为了揭示磷(P)营养缺乏对蓝藻释放挥发性有机化合物(VOCs)的影响及其对其他藻类的化感作用,以形成蓝藻水华的主要种类铜绿微囊藻(Microcystis aeruginosa)为材料,在无P培养条件下对其释放的VOCs进行分析,同时测定VOCs对莱茵衣藻(Chlamydomonas reinhardtii)生长、光合色素含量和光合性能的影响.结果表明,采用无P培养基培养铜绿微囊藻24 h后,其释放的VOCs种类和含量均明显增加,与标准培养基培养相比,VOCs总释放量增加了73.4%,并出现7种新化合物.将铜绿微囊藻释放的VOCs通入莱茵衣藻溶液中,在标准培养基中铜绿微囊藻释放的VOCs对莱茵衣藻生长无显著影响,而无P条件下释放的VOCs则明显抑制莱茵衣藻生长,其响应指数(RI)为-0.25.此外,莱茵衣藻光合色素含量、光系统II(PSII)最大光化学量子产量(Fv/Fm)、有效光化学量子产量[Y(II)]、光化学淬灭系数(q P)和光合电子传递速率(ETR)也明显降低,而非光化学淬灭系数(NPQ)则明显升高,其RI为0.26.由此可见,蓝藻在富营养化水体中大量繁殖以及P自身沉降特性导致的P缺乏会促进蓝藻释放VOCs,同时这些VOCs在保持蓝藻营养竞争优势和水体藻类多样性减少中具有化感抑制作用.     

基于图像区域特征的细胞识别方法及实现

先提取目标物体的面积、中心矩、圆形度、细长度、核浆比等区域特征。然后采用单原型模式表征方法，利用集群分析技术中最小距离分类器，构造判别函数，并依此作为识别类型的依据。最后，通过对鳞细胞的区域特征进行计算比较，选择出能够较明显地辨别出类别的特征向量，建立判别函数并对结果的可信度进行了讨论。     

多模式集合预报技术及其分析与检验

基于国家气象中心天气预报业务平台,对德国、日本、欧洲中心数值预报模式和我国T213模式的夏季预报产品进行检验,在此基础上,通过不同模式对目标区域预报能力的分析,分别应用神经元网络预报技术和基于Ts评分的客观多模式权重系数法(ME),建立了4个模式的集合预报方法,并应用于2005年汛期业务运行。结果表明:ME对短期降水预报技巧高于简单集合平均,因此具有一定的业务应用前景。     

高压下天然菱铁矿的压缩性和电子结构研究

采用同步辐射光源和金刚石对顶砧(DAC)技术,对天然菱铁矿的压缩性和电子结构进行了原位X射线衍射(XRD)和X射线吸收近边结构谱(XANES)测试研究。在室温下随着压力逐渐升高至50.2 GPa,菱铁矿保持方解石型结构不变,但是逐渐向Na Cl型结构转变;刚性[CO3]2-基团平行于ab-平面定向排列使c轴的压缩性大于a轴。菱铁矿在44.6~47.1 GPa之间发生电子由高自旋态(HS)向低自旋态(LS)的转变,表现为体积塌陷8%。HS菱铁矿的等温状态方程参数为K0=112(5)GPa和K'0=4.6(3)。首次采用XANES技术对菱铁矿中Fe2+的电子结构进行了研究,结果表明:随着压力升高至37.3 GPa,Fe2+的配位和局域对称并未发生明显变化;此后电子结构开始转变,Fe2+的3d轨道分裂能降低,电子跃迁概率增大,呈现LS特性。     

Determination of Ge, As, Se and Te in Silicate Samples Using Isotope Dilution-Internal Standardisation Octopole Reaction Cell ICP-QMS by Normal Sample Nebulisation

A method for the determination of Ge, As, Se and Te in silicate samples using isotope dilution-internal standardisation (ID-IS) octopole reaction cell (ORC) ICP-QMS by normal sample nebulisation was developed. The method does not involve either hydride generation or ion exchange. Germanium, Se and Te were determined by isotope dilution (ID), and As was determined by ID-IS. A silicate sample with an added Ge-Se-Te spike was digested with an HF-HNO₃-HBr mixture, dried, re-dissolved with HF and the supernatant liquid was directly aspirated into an ORC-ICP-QMS instrument with He or H₂ gas. No matrix effects were observed down to a dilution factor (DF) of ∼ 70 for Ge, Se and Te and DF of ∼ 1000 for As, which resulted in 3s detection limits in silicates of 2, 1, 0.1 and 4 ng g⁻¹, respectively. Advantages of the method are the simple sample introduction as well as a capability of determining S, Ti, Zr, Nb, Mo, Sn, Sb, Hf and Ta by ID-IS-ICP-QMS/SFMS from the same solution. Furthermore, the total sample solution consumption was only 0.253 ml with DF = 2000. Therefore, only a 0.13 mg test portion was required. To demonstrate the applicability of this technique, Ge, As, Se and Te in eight silicate reference materials were determined, as well as S, Ti, Zr, Nb, Mo, Sn, Sb, Hf and Ta in four carbonaceous chondrites.