基于主分量神经网络的降水集成预报方法研究

运用人工神经网络与主分量分析(PCA)相结合的方法,对同一降水预报量的各种数值预报产品进行集成预报研究.结果表明:主分量人工神经网络方法所构造的集成预报模型,不仅对历史样本的拟合精度好于个各子预报产品,独立样本的实验预报结果也显示出更好的预报准确率及稳定性.业务应用前景良好.     

PCA-BP神经网络在SO2浓度预报中的应用

将基于主成分分析(PCA)的BP神经网络预报方法引入大气污染预报,建立SO2浓度预报模型.结果表明:应用主成分分析对数据进行前处理,以原始预报因子的主成分作为BP神经网络的输入,降低了数据维数,消除了样本间存在的相关性,大大加快了BP神经网络的收敛速度.对模型进行预报验证,预报值与实际值之间的绝对误差为0.0098,预报值与实际值的相关系数达到0.885,得到较好的预报效果.并且比一般的BP神经网络模型具有较高的拟合和预报精度.     

用均生函数预测模型预报降水的若干试验

基于均生函数的处理原则,对目前常用的均生函数逐步筛选模型,主成分模型进行预报试验,提出带逐步筛选的主成分模型,并用变化幅度较大的降水序列对三种模型进行若干预测试验,结果表明,逐步筛选的主成分模型优于主成分模型,在对不同预报步长的预测试验中也表明,逐步筛选的主成分模型在预测效果以及模型的稳定性方面都是较好的。     

基于人工神经网络的集成预报方法研究和比较

用人工神经网络方法对同一预报量的各个子预报方程进行集成预报研究,并以同样的子预报方程进行回归、平均和加权预报集成。对神经网络集成预报模型与各个子预报方程及其它集成预报方法进行了对比分析研究。结果表明,人工神经网络方法所构造的集成预报模型不仅对历史样本的拟合精度比各个子预报方法及其它集成预报方法更好,独立样本的试验预报结果也显示出更好的预报准确性。并且,采用神经网络方法进行预报集成,可以避免以往集成预报方法难以确定权重系数的困难     

支持向量机在雷暴预报中的应用

论文利用2002--2006年AREM模式产品和常规观测报文资料,综合运用改进的K平均聚类和主成分分析等方法,基于MOS原理逐月建立了最小二乘支持向量机和线性规划支持向量机的单站雷暴释用预报模型,并针对海口站2007年58月进行了具体的预报。结果表明：支持向量机结合AREM模式产品进行雷暴的释用预报是合适、有效的,而且主成分分析对预报结果的提高也起到了积极的作用。     

几种降水集成预报方法的对比分析

采用多数表决集成法、评分权重集成法、多元回归集成法对承德市两来系统形成的降水进行晴雨和分级集成预报试验.结果表明:在每种单一的数值预报方法TS值均较高,总体预报质量相差不大的情况下,多数表决集中集成法可大幅度提高预报准确率;如果几种数值预报对于降水分级预报的预报准确率相差较大时,可采用评分权重集成法来提高总体预报质量;在样本足够的条件下,细化的预报结果多元回归集成法是晴雨预报和分级降水预报中比较可行的集成方法之一.     

集对分析在云南强降水预报集成方法中的应用

将集对分析中的联系度概念用于云南省强降水过程预报集成 ,首先对强降水天气过程进行分型 ,确定MM5模式预报、省台客观指导预报、预报员经验预报与云南强降水过程的联系度 ,用权重法将强降水预报进行集成 ,从而给出集不同预报于一体的云南全省性强降水过程预报集成方法     

滁河流域汛期多模式集成面雨量预报检验评估

对2013—2016年汛期ECMWF、JMA及中尺度WRF模式的预报结果进行检验,基于合理的时间尺度,制作了动态权重集成面雨量预报(DWI)、分步集成面雨量预报(SI)和等权重集成面雨量预报(EWI)产品,并对2017—2020年汛期降水期间多种集成面雨量预报产品和ECMWF、JMA、WRF的单个模式面雨量预报产品进行对比评估.结果表明:3种集成面雨量预报效果整体上优于单一数值预报模式,尤其是预报致灾严重的暴雨等级降水优势明显;DWI和WRF对强降水的面雨量预报正确率最高,其他2种集成面雨量预报结果次之,ECMWF及JMA较差;模式集成预报弱降水过程的优势不明显.     

集成方法在热带气旋路径和强度预报中的应用

利用多元线性回归、递归正权决策和算术平均方法,分别建立热带气旋路径和强度多模式集成预报方程,并用2004年西北太平洋和南海生成的热带气旋资料进行预报试验,同时对各家预报和集成预报进行了检验分析。结果表明,用集成预报方法作热带气旋路径和强度预报是可行的,预报年总平均误差有一定程度的减小,最大预报误差也有所减小,出现较大预报误差的概率明显减小。集成方法的预报能力一般好于集成方法中采用的任意单一预报,体现了集成方法的优越性。     

遗传算法优化的BP神经网络在地面温度多模式集成预报的应用研究

基于TIGGE资料集中的ECMWF、CMA和JMA的数值预报产品,利用加权集成、回归集成和消除偏差集成等线性集成方式与遗传算法优化的BP神经网络(GABP)集成,对我国大部开展地面2 m温度在24 h、48 h和72 h预报时效的多模式集成预报试验。通过对2013年1—6月的预报检验,结果表明:GABP集成预报效果有较大提升,均方误差明显小于各单一模式预报。GABP集成的误差分布在新疆和华北均方误差较大,但是在预报效果改进上GABP集成在西部地区相对单一模式的误差减小更加明显。在进行几种多模式集成方式时,GABP集成相比线性方法预报结果更加精准。对于天气过程个例的预报,GABP集成预报出预报量的变化趋势,预报效果优于单一模式和线性集成预报。无论是较长时间段还是短时间的天气过程,在改进预报效果上GABP集成都起到了最佳的作用。     

Influence of meteorological variability upon aerosol mass size distribution

Aerosol mass size distribution has been measured by using an optical particle counter. The measurements were done in an urban background location in the western Mediterranean during winter 2006. The study has been focused in determining the mass size distribution under special meteorological conditions like moderate rain, considerable winds and high atmospheric stability. The results obtained showed a mass predominance of accumulation mode during rain and high stability periods although for different reasons. In the case of rain, it is due to greater atmospheric cleansing effectiveness that rain has upon coarse mode particles. However, during stagnant periods, the meteorological situation favored coagulation processes among nucleation mode particles giving like result a mass increase in the accumulation mode. Finally, strong winds favor the resuspension of the largest particles and the dispersion of particles with sizes inferior to 7.5 μm. Similar results have been reproduced using principal component analysis (PCA). In this way, three components were identified. The first (PC1) represents particles in the accumulation mode. The second component (PC2) is constituted by coarse particles to 7.5 μm, and the third (PC3) corresponds to coarser particles. The contribution of each group to the overall average concentration was determined: 27.2% corresponds to particles with sizes belonging within the first component, 35.4% to PC2 and 37.3% to PC3. Important percentage variability for each component under meteorological episodes has been obtained. Results obtained showed an important increase of PC1 during Rainy Days (53.8%) and High Pollution Days (40.2%). Contrary to this on Windy Days this component decreases to 7.4%. However, during this kind of day PC3 increases to 64.6%.     

Winter daily precipitation variability over Cumbria, Northwest England

Daily precipitation totals at 55 sites were used to investigate geographic variability in winter (DJF) rainfall over Cumbria, NW England, over an 11-year period. Winter is the wettest season (>800?mm in the mountainous Lake District), with rainfall mechanisms closely linked to North Atlantic forcing. The Lamb weather type catalogue was used to identify rainfall distributions under different wind directions. Precipitation magnitude over Cumbria is much more sensitive to a change in wind direction than the geographic pattern in rainfall, with southwesterly (easterly) winds producing the highest (lowest) spatially averaged daily rainfall totals of 8.2?mm (0.6?mm). S-mode principal components analysis was used to identify the main patterns of precipitation variability. Three principal components (PCs) were retained as being statistically significant (cumulative explained variance for unrotated PCs?=?84.3%), with a correlated PC structure (direct oblimin rotation) best describing the spatial variance in rainfall. PC 1 has a very high index of strength (variance measure?=?40.9), indicating that there is one dominant rainfall pattern. PC 1 shows a gradient between wetter conditions in southwest Cumbria and over the central Lake District and drier conditions in NE Cumbria, and is usually caused by active zonal west to southwest flows. Almost of equal importance to PC 1 is PC 3 (variance measure?=?39.3), which has a more uniform rainfall distribution than PC 1 and is usually caused by fronts stalling over the region. PC 2, which shows an east to west decline in rainfall totals, is much less important than PCs 1 and 3 (variance measure?=?18.6). PC 2??s rainfall pattern can be caused by easterly flows with high pressure over Scandinavia and low pressure over the Continent, or by strong southwesterly flows, with depressions often centred over Scotland. Finally, cluster analysis was carried out to identify precipitation regions for all days and for each wind direction. Clusters were found to be largely stable to changes in wind direction, with stations in the central Lake District often clustered together, thus highlighting the importance of orographic enhancement of rainfall in this region.     

Winter daily precipitation variability over the South West Peninsula of England

Summary Daily precipitation totals for five consecutive winters (1995–99) were obtained for 127 stations in Devon and Cornwall to explore spatial variations in rainfall. This dataset was assembled with the explicit aim of assessing the appropriateness of current arrangements for daily rainfall forecasts in the SW Peninsula of England. Firstly, the extent to which fundamental geographic variables determine precipitation was investigated by correlating each station’s mean wet day amount (WDA) and percentage of wet days (PERWET) with altitude, latitude, longitude and distance from the coast. Altitude emerged as the most important control on precipitation, with a two-variable multiple linear regression model containing altitude and latitude being able to explain 39.3% (29.8%) of the variance in WDA (PERWET) values. The main spatial modes of variability in the region’s precipitation field were identified by using S mode principal components analysis (PCA). Six PCs were statistically significant and explained 83.4% of the geographic variance in precipitation over Devon and Cornwall. The components were interpreted physically by examining the synoptic flow environment (pressure and wind anomalies) on days with high positive and negative PC scores. Explaining 25.1% of the variance, the most important pattern (PC1) depicts a location’s degree of exposure or shelter in a moist, unstable W–NW airflow. The higher PCs describe modes of variability that accentuate rainfall in East Devon (PC2), Cornwall (PC3), Dartmoor and Bodmin Moor (PC4), South Devon (PC5), and North Cornwall and NW Devon (PC6) relative to other areas of the Peninsula. Finally, a winter precipitation regionalisation was derived by applying agglomerative hierarchical cluster analysis to the PC loadings of the significant components. In most cases, the six coherent precipitation regions do not reflect the familiar administrative or topographic areas used for forecasting, suggesting that forecasts issued on such a basis are likely to be insufficiently detailed and misleading.     

我国秋季降水的年际变化及与热带太平洋海温异常分布的关系

利用我国160个站点58年(1951～2008年)的降水资料、NCEP/NCAR再分析环流资料和Hadley海表温度资料,对我国秋季降水年际变化的特征和可能成因进行了分析。结果显示,秋季降水前两模态分别反映长江流域及以南地区和长江以北的江淮、黄淮、华北、四川盆地北部至河套等地降水的变化,两降水模态的变化都以年际尺度为主,年代际变化特征不明显。就环流形势而言,第一模态的年际变化主要与西太平洋副热带高压的强度及相应的对流层低层菲律宾群岛附近的异常气旋/反气旋联系紧密,第二模态的年际变化则可能受到副热带高压的南北位置和相应的日本岛附近的异常气旋/反气旋的影响。同时,两模态及相应的异常环流还分别与热带东印度洋和热带西太平洋附近的异常垂直运动关系密切,热带地区的异常垂直运动可能通过经圈方向的异常环流影响到东亚地区。此外,两降水模态不仅与热带地区的异常环流关系密切,而且与热带海温异常也存在紧密的联系。与两模态相关联的热带太平洋海温异常显示出不同的分布特征,当热带东太平洋偏暖/冷,西太平洋偏冷/暖时,长江以南地区降水偏多/少。而当热带东太平洋和中太平洋一致偏暖/冷时,长江以北地区降水易偏少/多。两降水模态与热带海温及热带地区异常环流之间的密切关系显示热带太平洋海温异常的不同分布可能通过激发不同的热带地区异常垂直环流形势而对降水产生影响。     

Evaluation of cloud vertical structure simulated by recent BCC_AGCM versions through comparison with CALIPSO-GOCCP data

ABSTRACT The abilities of BCC-AGCM2.1 and BCC_AGCM2.2 to simulate the annual-mean cloud vertical structure （CVS） were evaluated through comparison with GCM-Oriented CALIPSO Cloud Product （CALIPSO-GOCCP） data. BCC-AGCM2.2 has a dynamical core and physical processes that are consistent with BCC-AGCM2.1, but has a higher horizontal resolution. Results showed that both BCC-AGCM versions underestimated the global-mean total cloud cover （TCC）, middle cloud cover （MCC） and low cloud cover （LCC）, and that BCC_AGCM2.2 underestimated the global-mean high cloud cover （HCC）. The global-mean cloud cover shows a systematic decrease from BCCA-GCM2.1 to BCC_AGCM2.2, especially for HCC. Geographically, HCC is significantly overestimated in the tropics, particularly by BCC_AGCM2,1, while LCC is generally overestimated over extra-tropical lands, but significantly underestimated over most of the oceans, especially for subtropical marine stratocumulus clouds. The leading EOF modes of CVS were extracted. The BCC_AGCMs perform well in reproducing EOF1, but with a larger variance explained. The two models also capture the basic features of EOF3, except an obvious deficiency in eigen- vector peaks. EOF2 has the largest simulation biases in both position and strength of eigenvector peaks. Furthermore, we investigated the effects of CVS on relative shortwave and longwave cloud radiative forcing （RSCRF and RLCRF）. Both BCC_AGCM versions successfully reproduce the sign of regression coefficients, except for RLCRF in PC1. However, the RSCRF relative contributions from PC1 and PC2 are overestimated, while the relative contribution from PC3 is underes timated in both BCC_AGCM versions. The RLCRF relative contribution is underestimated for PC2 and overestimated for PC3.     

El Nino的周期变化和空间特征——兼应用PC和CPC分析太平洋海温距平场

本文联合利用实主成分(以下简称PC)分析和复主成分(以下简称CPC)分析,探讨太平洋SST演变的时空特征.结果表明:这两种分析方法得到的第一个主成分都展示El Nino随时间周期性变化,在PC中表现为峰谷的变化,在CPC中表现为振幅和位相的变化;相应的荷载向量主要反映El Nino和La Nina事件时期空间分布特征,大振幅区域位于主要冷暖洋流活动中心.第一个复荷载向量可以认为在驻波上附加一个南北向传播的行波.第二个荷载向量为一驻波,第三个复荷载向量看上去像一个东西向传播的行波,后两个特征型各自强调不同El Nino和La Nina时期的空间特征.驻波峰、谷区域之间呈现反相关关系;这里虽然没有研究波群和群速度,但就行波而言,也初步看出能量传播的路径.与复荷载向量对应的实荷载向量仅显示驻波特征.     

Principal Components Analysis of Rainfall and Associated Synoptic Models of the Southwest Monsoon over India

Summary  The summer monsoon circulation shows various spatial and temporal oscillations and often a combination of systems produces an integrated effect. In this study phases of the southwest (SW) monsoon have been identified in an objective manner with the help of T-mode principal component analysis (PCA) of weekly rainfall anomalies. Mean composite charts have been prepared utilising all available upper air data (1977–1986) for each category of the SW monsoon epochs identified by the PCA. These sets of charts have been constructed for both the strong and weak phases associated with the first four significant principal components (PCs). A well defined east-west oriented trough system, extending from about 28° N Latitude/65° E Longitude to 20° N Latitude/90° E Longitude, in the lower levels, has been the main feature associated with the strong phase of the monsoon corresponding to PC I. The trough in the lower levels is more marked in the eastern half compared to the western half in both the sets of charts associated with strong phases of the monsoon related to the PC II and PC III. With PC II, the position of the troughs in the lower levels is further north of its location in PC III. The east-west trough system associated with the strong phase of PC IV has a large southward tilt with height. The charts corresponding to the weak phases of these PCs have synoptic features, such as the position of the trough close to the foothills of the Himalayas, and the shifting of middle and upper tropospheric anticyclones to the south. The study suggests an objective method of interpretation of principal components by utilising synoptic data. In addition, synoptic models and data sets corresponding to different phases of the monsoon can also be prepared in an objective manner by such PCA. Received July 18, 1997 Revised April 30, 1999     

气候变暖背景下全球海温对中国东部夏季降水年代际转折的影响

针对1970年代末及1990年代初中国东部夏季降水（ECP）的年代际变化格局,采用EOF分解、相关分析、回归分析等统计方法诊断了全球海表面温度对ECP年代际变化前两个模态（EOF1、EOF2）的影响。发现大西洋多年代际振荡（AMO）序列,印度洋偶极子（DMI）序列,太平洋年代际振荡（PDO）序列与ECP前两个模态时间系数（PC1、PC2）相关性较好,结合各海温指数的年代际变化特征,发现ECP在1970年代末受PDO及DMI的影响在低纬及中纬度地区分别呈现EOF1、EOF2的正位相分布特征;而在1990年代初受AMO及PDO的影响主要呈现EOF1的特征。由各海温指数及PC1、PC2重建的ECP分布特征可知,AMO及DMI与PC1重建的ECP型相近,对ECP的影响范围集中在低纬地区。除去变暖影响的DMI及PC1回归的高度场中发现一个源起大西洋的波列,黄河以北为异常反气旋中心,以南为异常气旋中心,低层南风异常,水汽被输送到北方,导致中国北方降水增加,南方降水减少。PDO与PC2重建的ECP型相近,对ECP的影响集中在中纬度地区。二者回归得到中国东部低层北风异常,水汽在长江流域辐合,长江流域降水增加。      

Integrated Analysis of Physical and Biological Pan-Arctic Change

We investigate the recent large changes that have occurred in the Arctic over the period of 1965–1995 through examination of 86 regionally-dispersed time series representing seven data types: climate indices, atmosphere, ocean, terrestrial, sea ice, fisheries, and other biological data. To our knowledge, this is the first semi-quantitative analysis of Arctic data that spans multiple disciplines and geographic regions. Although visual inspection and Principal Component Analysis of the data collection indicate that Arctic change is complex, three patterns are evident. The temporal pattern of change calculated as the first Principal Component (PC1), representing 23% of the variance, has a single regime-like shift near 1989 based on a large number of time series, which include projections from a strong stratospheric vortex in spring, the Arctic Oscillation, sea ice declines in several regions, and changes in selected mammal, bird, and fish populations. The pattern based on the second Principal Component (PC2) shows interdecadal variability over the Arctic Ocean Basin north of 70° N; this variability is observed in surface wind fields, sea ice, and ocean circulation, with the most recent shift near 1989. Contributions to PC1 cover a larger geographic area than PC2, and are consistent with a recent amplification of the interdecadal mode due to polar processes such as increased incidence of cold stratospheric temperature anomalies or internal feedbacks. Most land processes – such as snowcover, greenness, Siberian runoff, permafrost temperatures – and certain subarctic sea ice records show a third pattern of a linear trend over the 30-year interval, which is qualitatively different than either PC1 or PC2. These variables are from lower latitudes and often integrate the atmospheric or oceanographic influence over several seasons including summer. That more than half of the data collection projects strongly onto one of the three patterns, suggests that the Arctic is responding as a coherent system over the previous three decades. However, no single index or class of observations exclusively tracks change in the Arctic, a conclusion that emerges from a multivariate analysis.     

The Extreme Summer Precipitation over East China during 1982–2007 Simulated by the LASG/IAP Regional Climate Model

The extreme summer precipitation over East China during 1982-2007 was simulated using the LASG/IAP regional climate model CREM(the Climate version of a Regional Eta-coordinate Model).The results show that the probability density functions(PDFs) of precipitation intensities are reasonably simulated,except that the PDFs of light and moderate rain are underestimated and that the PDFs of heavy rain are overestimated.The extreme precipitation amount(R95p) and the percent contribution of extreme precipitation to the total precipitation(R95pt) are also reasonably reproduced by the CREM.However,the R95p and R95pt over most of East China are generally overestimated,while the R95p along the coastal area of South China(SC) is underestimated.The bias of R95pt is consistent with the bias of precipitation intensity on wet days(SDII).The interannual variation for R95p anomalies(PC1) is well simulated,but that of R95pt anomalies(PC2) is poorly simulated.The skill of the model in simulating PC1(PC2) increases(decreases) from north to south.The bias of water vapor transport associated with the 95th percentile of summer daily precipitation(WVTr95) explains well the bias of the simulated extreme precipitation.