基于BP神经网络的单站总云量预报研究

云是能够对飞行活动产生影响、甚至危及飞行安全的气象要素之一.为了对北京单站总云量进行预报,利用人工神经网络的BP模型,针对不同时次进行关键因子的选取,尝试用多种预报因子的组合,建立了总云量预报模型.试验结果表明:所建立的模型具有较好的拟合、预报精度,并且没有出现"过拟合"现象,对新样本具有较好的泛化能力.     

近50年来东莞气温年际变化的长期特征分析

运用福建省25个代表站汛期降水量资料，得到了能够反映全省旱涝状况指标，以此指标为预报量，运用相关分析和逐步回归分析方法，从前期海温场、大气环流场中选取了预报因子，并据此建立了福建汛期旱涝的多元线性回归和RBF神经网络预测模型。结果表明，RBF神经网络模型在历史样本拟合精度上、样本交叉检验和模型的实际预测能力上都明显优于传统的线性回归方法，该模型在实际预测中具有良好的应用能力和推广价值。     

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

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仪在统计上是显著的,同时也具有清楚的物理意义.     

基于BP神经网络的单站总云量预报研究

云是能够对飞行活动产生影响、甚至危及飞行安全的气象要素之一。为了对北京单站总云量进行预报,利用人工神经网络的BP模型,针对不同时次进行关键因子的选取,尝试用多种预报因子的组合,建立了总云量预报模型。试验结果表明：所建立的模型具有较好的拟合、预报精度,并且没有出现＂过拟合＂现象,对新样本具有较好的泛化能力。     

流域年均含沙量的B—P网络预测模型及其效果检验

应用误差反向传播算法的人工神经网络，建立了流域年均含沙量的预测模型，该模型用于某流域年均含沙量预测的拟合率达９０％以上，预留检验预报的准确率为７５％。     

影响广西热带气旋年频数的神经网络预测模型

利用人工神经网络方法，结合均生函数和逐步回归分析方法对影响广西的热带气旋年频数序列进行神经网络的预测模型研究，该模型既考虑了预报量自身显著周期变化，也考虑了海温场，高度场等因子对预报量的影响，给预报检验，考虑多种影响因子的预测模型比只考虑单一影响因素的预测模型效果好。     

人工神经网络西太平洋副热带高压预报业务系统

本文应用常规500hpa资料,结合具有较稠密、高分辨的“TOVS”资料（卫星信息资料）,分析、归纳了西太平洋副热带高压大幅度北移、南蓝（120°E脊线）的经验规则,在此基础上选择了具有最优下降因子的BP算法。在人工神经网络的结构方面设计了神经网络的前处理功能,将众多的单因子经基于专家系统的前处理模块,归纳的规则,作为网络的输入端,经网络学习、训练、迭代获得了较快的收敛速度和较高的历史拟合率,建立了“人工神经网络西太平洋副热带高压预报业务系统”     

模型参数对赤霉病人工神经网络预报模型精度的影响

运用最优化相关普查方法,选取确定了对江苏省太湖地区小麦赤霉病发生程度有着显著指示意义的预报因子,将其作为输入变量经多层前馈神经网络的ＢＰ算法进行学习训练,建立了赤霉病病穗率的人工神经网络预报模型。分析了结构参数对模型效果的影响情况,发现训练的总体误差平方和对模型的效果最为显著,历史样本的拟合率随着总体误差平方和的拳小而稳定上升,但总体误差平方和取值偏小时模型对独立样本的预报精度下降;当总体误差平方     

区域降水数值预报产品人工神经网络释用预报研究

利用T213、日本细网格降水预报等数值预报产品,采用人工神经网络方法进行预报释用。通过聚类分析方法对广西自治区测站进行分类,简化预报对象,对数量众多的T213数值预报产品采用自然正交分解(EOF)方法,浓缩大量因子的有效信息,并结合日本降水预报因子建立广西5~6月区域降水量级的逐日人工神经网络预报模型。运用与实际业务预报相同的方法进行逐日预报试验。结果表明,用这种数值预报产品释用方法建立广西3个预报区域的B-P人工神经网络预报模型对中雨以上降水量级预报的TS评分分别为0.55、0.5和0.26,比目前业务预报中参考使用的T213和日本数值预报产品降水预报具有更好的预报效果。     

高血压病发病率预报的人工神经网络方法研究

通过统计分析，选取影响银川地区高血压病发病率的主要气象因素，将其作为输入变量经多层前馈型神经网络的BP(Back Propagation)算法进行学习训练，建立了疾病发病率的人工神经网络(ANN，Artifical Neural Net)预报模型。结果表明：该方法计算简便、误差较小，为疾病发病率预报提供了一种新的预报方法。     

全球海气耦合模式对我国极端强降水模拟检验

以1961—1999年我国地面观测逐日降水资料作为观测基础, 初步分析了18个全球海气耦合模式对我国20世纪极端强降水的模拟能力。分析模式对不同级别降水的模拟发现, 各模式模拟的我国1~10 mm小雨日数普遍明显偏多; 10~25 mm中雨日数的模拟结果总体上也以偏多为主, 虽然部分模式能够模拟出我国南方存在的高值中心, 但位置偏北至长江中下游地区; 25~50 mm大雨日数在我国南方明显偏少, 并且大值中心的位置基本都没能模拟出来; 50 mm以上暴雨日数的模拟结果也明显偏小, 除MIROC3.2(hires) 外大部分模式在长江以南地区的结果都未超过2 d; 大部分模式不能正确模拟出我国东部地区大雨日数变化趋势的空间分布。进一步分析各模式对极端强降水的模拟发现:各模式极端强降水阈值明显低于观测; 半数左右的模式模拟出了1961—1999年西北西部极端降水增加的趋势, 个别模式趋势系数的大小与观测相当, 大部分模式对东北和长江中下游地区的模拟结果呈与观测反向的变化趋势, 没有模式能够模拟出我国东部地区存在的东北—华北与华中—长江中下游—华南存在的极端强降水日数增加-减少-增加-减少的空间分布; 大部分模式模拟的极端强降水日数标准差与观测结果比较接近, 这可能主要是由于对观测和各模式使用了同样的判定极端强降水发生的方法。总的来看, 全球海气耦合模式对我国极端强降水的模拟能力还有待进一步改进。     

陆地生态系统模型及其与气候模式耦合的回顾

陆地生态系统和气候系统通过能量通量、水汽通量、物质交换相互影响、作用。作者对陆地生态系统模型及其与气候模式耦合的研究进行综述和讨论,总结了当代5类主要全球陆地生态系统模型,即生物地理模型、生物地球化学模型、森林林窗模型、陆面生物圈模型和动态全球植被模型,以及它们与气候模式耦合的研究进展。阐述了动态全球植被模型及其与气候模式耦合研究在全球变化研究的重要作用。最后,对未来模拟研究的方向进行了分析。     

Progress in the Development and Application of Climate Ocean Models and Ocean-Atmosphere Coupled Models in China

A review is presented about the development and application of climate ocean models and oceanatmosphere coupled models developed in China as well as a review of climate variability and climate change studies performed with these models. While the history of model development is briefly reviewed, emphasis has been put on the achievements made in the last five years. Advances in model development are described along with a summary on scientific issues addressed by using these models. The focus of the review is the climate ocean models and the associated coupled models, including both global and regional models, developed at the Institute of Atmospheric Physics, Chinese Academy of Sciences. The progress of either coupled model development made by other institutions or climate modeling using internationally developed models also is reviewed.     

The Bolund Experiment,Part II: Blind Comparison of Microscale Flow Models

Bolund measurements were used for a blind comparison of microscale flow models. Fifty-seven models ranging from numerical to physical were used, including large-eddy simulation (LES) models, Reynolds-averaged Navier–Stokes (RANS) models, and linearized models, in addition to wind-tunnel and water-channel experiments. Many assumptions of linearized models were violated when simulating the flow around Bolund. As expected, these models showed large errors. Expectations were higher for LES models. However, of the submitted LES results, all had difficulties in applying the specified boundary conditions and all had large speed-up errors. In contrast, the physical models both managed to apply undisturbed ‘free wind’ boundary conditions and achieve good speed-up results. The most successful models were RANS with two-equation closures. These models gave the lowest errors with respect to speed-up and turbulent kinetic energy (TKE) prediction.     

CMIP5全球气候模式对青藏高原地区气候模拟能力评估

青藏高原是气候变化的敏感和脆弱区,全球气候模式对于这一地区气候态的模拟能力如何尚不清楚。为此,本文使用国际耦合模式比较计划第五阶段（CMIP5）的历史模拟试验数据,评估了44 个全球气候模式对1986~2005 年青藏高原地区地表气温和降水两个基本气象要素的模拟能力。结果表明,CMIP5 模式低估了青藏高原地区年和季节平均地表气温,年均平均偏低2.3℃,秋季和冬季冷偏差相对更大;模式可较好地模拟年和季节平均地表气温分布型,但模拟的空间变率总体偏大;地形效应校正能够有效订正地表气温结果。CMIP5 模式对青藏高原地区降水模拟能力较差。尽管它们能够模拟出年均降水自西北向东南渐增的分布型,但模拟的年和季节降水量普遍偏大,年均降水平均偏多1.3 mm d-1,这主要是源于春季和夏季降水被高估。同时,模式模拟的年和季节降水空间变率也普遍大于观测值,尤其表现在春季和冬季。相比较而言,44 个模式集合平均性能总体上要优于大多数单个模式;等权重集合平均方案要优于中位数平均;对择优挑选的模式进行集合平均能够提高总体的模拟能力,其中对降水模拟的改进更为显著。     

The South Pacific Convergence Zone in CMIP5 simulations of historical and future climate

The South Pacific Convergence Zone (SPCZ) is evaluated in historical simulations from 26 Coupled Model Intercomparison Project Phase 5 (CMIP5) models, and compared with previous generation CMIP3 models. A subset of 24 CMIP5 models are able to simulate a distinct SPCZ in the December to February (DJF) austral summer, although the position of the SPCZ in these models is too zonal compared with observations. The spatial pattern of SPCZ precipitation is improved in CMIP5 models relative to CMIP3 models, although the spurious double ITCZ precipitation band in the eastern Pacific is intensified in many CMIP5 models. All CMIP5 models examined capture some interannual variability of SPCZ latitude, and 19 models simulate a realistic correlation with El Niño–Southern Oscillation. In simulations of the twenty-first century under the RCP8.5 emission scenario, no consistent shift in the mean position of the DJF SPCZ is identified. Several models simulate significant shifts northward, and a similar number of models simulate significant southward shifts. The majority of CMIP5 models simulate an increase in mean DJF SPCZ precipitation, and there is an intensification of the eastern Pacific double ITCZ precipitation band in many models. Most models simulate regions of increased precipitation in the western part of the SPCZ and near the equator, and regions of decreased precipitation at the eastern edge of the SPCZ. Decomposition of SPCZ precipitation changes into dynamic and thermodynamic components reveals predominantly increased precipitation due to thermodynamic changes, while dynamic changes lead to regions of both positive and negative precipitation anomalies.     

A GCSS Boundary-Layer Cloud Model Intercomparison Study Of The First Astex Lagrangian Experiment

Three single-column models (all with an explicit liquid water budget and compara-tively high vertical resolution) and three two-dimensional eddy-resolving models (including one with bin-resolved microphysics) are compared with observations from the first ASTEX Lagrangian experiment. This intercomparison was a part of the second GCSS boundary-layer cloud modelling workshop in August 1995.In the air column tracked during the first ASTEX Lagrangian experiment, a shallow subtropical drizzling stratocumulus-capped marine boundary layer deepens after two days into a cumulus capped boundary layer with patchy stratocumulus. The models are forced with time varying boundary conditions at the sea-surface and the capping inversion to simulate the changing environment of the air column.The models all predict the observed deepening and decoupling of the boundary layer quite well, with cumulus cloud evolution and thinning of the overlying stratocumulus. Thus these models all appear capable of predicting transitions between cloud and boundary-layer types with some skill. The models also produce realistic drizzle rates, but there are substantial quantitative differences in the cloud cover and liquid water path between models. The differences between the eddy-resolving model results are nearly as large as between the single column model results. The eddy resolving models give a more detailed picture of the boundary-layer evolution than the single-column models, but are still sensitive to the choice of microphysical and radiative parameterizations, sub-grid-scale turbulence models, and probably model resolution and dimensionality. One important example of the differences seen in these parameterizations is the absorption of solar radiation in a specified cloud layer, which varied by a factor of four between the model radiation parameterizations.     

Simulated Relationship between Wintertime ENSO and East Asian Summer Rainfall: From CMIP3 to CMIP6

El Ni?o-Southern Oscillation(ENSO)events have a strong influence on East Asian summer rainfall(EASR).This paper investigates the simulated ENSO-EASR relationship in CMIP6 models and compares the results with those in CMIP3 and CMIP5 models.In general,the CMIP6 models show almost no appreciable progress in representing the ENSO-EASR relationship compared with the CMIP5 models.The correlation coefficients in the CMIP6 models are relatively smaller and exhibit a slightly greater intermodel diversity than those in the CMIP5 models.Three physical processes related to the delayed effect of ENSO on EASR are further analyzed.Results show that,firstly,the relationships between ENSO and the tropical Indian Ocean(TIO)sea surface temperature(SST)in the CMIP6 models are more realistic,stronger,and have less intermodel diversity than those in the CMIP3 and CMIP5 models.Secondly,the teleconnections between the TIO SST and Philippine Sea convection(PSC)in the CMIP6 models are almost the same as those in the CMIP5 models,and stronger than those in the CMIP3 models.Finally,the CMIP3,CMIP5,and CMIP6 models exhibit essentially identical capabilities in representing the PSC-EASR relationship.Almost all the three generations of models underestimate the ENSO-EASR,TIO SST-PSC,and PSC-EASR relationships.Moreover,almost all the CMIP6 models that successfully capture the significant TIO SST-PSC relationship realistically simulate the ENSO-EASR relationship and vice versa,which is,however,not the case in the CMIP5 models.     

夏季南海台风移动路径的一种客观预报方法

以1960—2003年共44a夏季的7月、8月、9月西行进入南海海域的台风样本为基础,综合考虑南海台风移动路径的气候持续因子和数值预报产品物理量因子,运用条件数方法选取因子并建立回归方程,进行台风路径预报模型的预报建模研究。通过对比分析发现,基于条件数方法的南海台风移动路径模型具有较好的预报效果,7月、8月、9月3个月24h台风路径预报的平均距离误差为153．9km,预报能力明显高于目前国内外的其他一些台风路径客观预报方法。该方法的预报精度相对于逐步回归方法有了很大的提高,相对于气候持续法也为正的预报技巧水平。     

The CLIVAR C20C project: selected twentieth century climate events

We use a simple methodology to test whether a set of atmospheric climate models with prescribed radiative forcings and ocean surface conditions can reproduce twentieth century climate variability. Globally, rapid land surface warming since the 1970s is reproduced by some models but others warm too slowly. In the tropics, air-sea coupling allows models to reproduce the Southern Oscillation but its strength varies between models. We find a strong relationship between the Southern Oscillation in global temperature and the rate of global warming, which could in principle be used to identify models with realistic climate sensitivity. This relationship and a weak response to ENSO suggests weak sensitivity to changes in sea surface temperature in some of the models used here. In the tropics, most models reproduce part of the observed Sahel drought. In the extratropics, models do not reproduce the observed increase in the North Atlantic Oscillation in response to forcings, through internal variability, or as a combination of both.