A Scalable High-Performance I/O System for a Numerical Weather Forecast Model on the Cubed-Sphere Grid

The design and implementation of a high-performance Input/Output (I/O) library for the Korean Integrated Model (KIM, KIM-IO) is described in this paper. The KIM is a next-generation global operational model for the Korea Meteorological Administration (KMA). The horizontal discretization of KIM consists of the spectral-element method on the cubed-sphere grid. The KIM-IO is developed to be a consistent and efficient approach for input and output of essential data in this particular grid structure in a multiprocessing environment. The KIM-IO provides three main features, comprising the sequential I/O, parallel I/O, and I/O decomposition methods, and adopts user-friendly interfaces similar to the Network Common Data Form (NetCDF). The efficiency of the KIM-IO is verified using experiments to analyze the performance of its three features. The scalability is also verified by implementing the KIMIO in the KIM at a resolution of approximately 12 km using the 4th supercomputer of KMA. The experimental results show that both regular parallel I/O and sequential I/O undergo performance degradation with an increasing number of processes. However, the I/O decomposition method in the KIM-IO overcomes this degradation, leading to improvement in scalability. The results also indicate that with using the new I/O decomposition method, the KIM attains good parallel scalability up to Ο (100,000) cores.     

大规模数据并行问题的可扩展性分析

大规模数据并行处理的性能受到处理机数量、I/O速度、通信速度等多方面因素的制约。增加处理机数量或提高处理机的计算速度，可以提高计算机的整体处理速度，但是通信和I/O会成为影响并行效率的主要因素。为了综合分析这些因素对计算性能的影响，用一种比较典型的大规模数据并行的计算模型，具体分析了处理机数量、处理机速度与处理机间的通信延迟、通信速率以及输入输出速度之间的关系。得到了大规模并行机的通信和I/O性能与处理机速度与数量之间存在的关系。指出，增加处理机数量、提高单节点处理速度的同时，必须按照一定的关系相应增加节点间的通信性能和I/O性能。单纯以增加处理机数量、提高单处理机速度提高计算机峰值速度的方法会降低系统的计算效率，不能达到计算速度与计算机处理能力同步增长的目的。     

LASG Global AGCM with a Two-moment Cloud Microphysics Scheme:Energy Balance and Cloud Radiative Forcing Characteristics

Cloud dominates influence factors of atmospheric radiation, while aerosol–cloud interactions are of vital importance in its spatiotemporal distribution. In this study, a two-moment(mass and number) cloud microphysics scheme, which significantly improved the treatment of the coupled processes of aerosols and clouds, was incorporated into version 1.1 of the IAP/LASG global Finite-volume Atmospheric Model(FAMIL1.1). For illustrative purposes, the characteristics of the energy balance and cloud radiative forcing(CRF) in an AMIP-type simulation with prescribed aerosols were compared with those in observational/reanalysis data. Even within the constraints of the prescribed aerosol mass, the model simulated global mean energy balance at the top of the atmosphere(TOA) and at the Earth's surface, as well as their seasonal variation, are in good agreement with the observational data. The maximum deviation terms lie in the surface downwelling longwave radiation and surface latent heat flux, which are 3.5 W m-2(1%) and 3 W m-2(3.5%), individually. The spatial correlations of the annual TOA net radiation flux and the net CRF between simulation and observation were around 0.97 and 0.90, respectively. A major weakness is that FAMIL1.1 predicts more liquid water content and less ice water content over most oceans. Detailed comparisons are presented for a number of regions, with a focus on the Asian monsoon region(AMR). The results indicate that FAMIL1.1 well reproduces the summer–winter contrast for both the geographical distribution of the longwave CRF and shortwave CRF over the AMR. Finally, the model bias and possible solutions, as well as further works to develop FAMIL1.1 are discussed.     

长短期记忆神经网络(LSTM)模型在低能见度预报中的应用

利用浙江省义乌市2015—2019年逐小时气象观测数据(相对湿度、风速、地气温差、能见度)和空气质量指数(Air Quality Index, AQI)数据, 分析了义乌地区低能见度天气(观测能见度lt; 10 km)的分布特征和气象要素条件。利用长短期记忆神经网络(Long Short Term Memory Neural Network, LSTM)模型对逐小时能见度进行模拟, 分别对比了观测能见度作为输入变量与否的模拟效果; 根据义乌地区低能见度天气条件的特征, 将模拟时段分为三个时期(11月至翌年2月, 3—6月, 7—10月), 对比了分时期模拟的效果; 以及评估了模型的预报步长。结果表明: 高湿、高污染、气温高于地温和低风速是义乌地区低能见度天气的主要特征。LSTM模型对单站能见度有较好的模拟效果, 当输入参数中加入历史观测能见度时, 能大幅提高模拟准确度, 日均能见度模拟结果均方根误差RMSE=0.63 km, 平均绝对误差MAE=0.51 km, 拟合优度R²=0.99;分时期进行模拟能得到更精准的模拟结果。本研究中选用的输入要素在冬季(11月至翌年2月)模拟效果最好, RMSE=2.35 km, MAE=1.46 km, 低能见度均方根误差RMSE_10 km=1.81 km, 低能见度平均绝对误差MAE_10 km=1.13 km, R²=0.83; 3—6月的模拟中, 输入变量中不加AQI模拟效果更好, 这意味着3—6月义乌地区的低能见度天气以雾天气为主导, 加入过多变量并不一定能提高模型准确度; 随着预报步长增大, 模型预报效果变差, 预测步长等于3 h, R²=0.71, 预测结果已不具备实际应用意义。     

基于卫星资料进行热带气旋强度客观估算

利用日本MTSAT (multi-functional transport satellite) 红外亮温资料,提取热带气旋云团中云顶较高、对流较旺盛的深对流信息,根据提取的对流核数量、对流核距热带气旋中心距离、对流核亮温极值等信息表征热带气旋强弱,初步建立了热带气旋强度估测模型;并根据该估算模型的误差分布对强度 (用最大风速表示) 大于40 m·s-1和小于18 m·s-1的样本结果进行了线性修正,修正后的结果与中国气象局《热带气旋年鉴》热带气旋最佳路径资料比较得到非独立样本和独立样本的强度平均绝对误差分别为5.5 m·s-1和5.9 m·s-1, 均方根误差分别为6.9 m·s-1和7.7 m·s-1;对于热带低压、强台风及以上的估计平均绝对误差分别降至4.9,4.7 m·s-1,准确度较好。试验表明:利用热带气旋云团中的对流核数量、分布、冷暖与其强度建立的统计关系模型是可行的,该算法的估算精度与Dvorak方法、AMSU (advanced microwave sounding unit) 定强算法相当。     

Sensitivity of Precipitation in Aqua-Planet Experiments with an AGCM

The sensitivity of precipitation was studied by conducting control aqua-planet experiments(APEs) with a model to determine atmospheric general circulation.The model includes two versions: that with a spectral dynamical core(SAMIL) and that with a finite-volume dynamical core(FAMIL).Three factors were investigated including dynamical core,time-step length,and horizontal resolution.Numerical results show that the dynamical core significantly affects the structure of zonal averaged precipitation.FAMIL exhibited an equatorial precipitation belt with a single narrow peak,and SAMIL showed a broader belt with double peaks.Moreover,the time step of the model physics is shown to affect the zonal-averaged tropical convective precipitation ratio such that a longer time step leads to more production and consumption of convective available potential energy and convection initiated away from the equator,which corresponds to equatorial double peaks of precipitation.Further,precipitation is determined to be sensitive to horizontal resolution such that higher horizontal resolution allows for more small-scale kinetic energy to be resolved and leads to a broader probability distribution of low-level vertical velocity.This process results in heavier rainfall and convective precipitation extremes in the tropics.Abstract The sensitivity of precipitation was studied by conducting control aqua-planet experiments(APEs)with a model to determine atmospheric general circulation.The model includes two versions:that with a spectral dynamical core(SAMIL)and that with a finite-volume dynamical core(FAMIL).Three factors were investigated including dynamical core,time-step length,and horizontal resolution.Numerical results show that the dynamical core significantly affects the structure of zonal averaged precipitation.FAMIL exhibited an equatorial precipitation belt with a single narrow peak,and SAMIL showed a broader belt with double peaks.Moreover,the time step of the model physics is shown to affect the zonal-averaged tropical convective precipitation ratio such that a longer time step leads to more production and consumption of convective available potential energy and convection initiated away from the equator,which corresponds to equatorial double peaks of precipitation.Further,precipitation is determined to be sensitive to horizontal resolution such that higher horizontal resolution allows for more small-scale kinetic energy to be resolved and leads to a broader probability distribution of low-level vertical velocity.This process results in heavier rainfall and convective precipitation extremes in the tropics.     

BCC_CSM气候系统模式移植优化及其气候模拟验证

为了提高BCC_CSM气候系统模式运行效率,保障业务科研工作的顺利开展,进行BCC_CSM气候系统模式在IBM高性能计算系统的移植工作;通过性能优化使BCC_CSM模式运行效率显著提高,通过气候要素形势场分布和相对误差量化指标对BCC_CSM气候系统模式模拟性能进行验证。结果表明：移植优化后,BCC_CSM气候系统模式计算效率提高为原来的1.4倍;基于CMIP5 piControl试验,完成531-540年10 a的气候模拟,年平均地表气温形势场分布合理,相对误差小于0.5%,BCC_CSM气候系统模式计算和模拟性能均能满足应用需求。     

不同分辨率和微物理方案对飑线阵风锋模拟的影响

为研究不同分辨率和微物理方案对飑线阵风锋模拟的影响,利用WRF中尺度数值预报模式,对2009年6月5日发生在上海的一次飑线过程分别进行了3、1、0.5 km水平分辨率和一、二阶矩微物理方案的理想试验。结果表明,模式水平分辨率和微物理方案对模拟飑线阵风锋有明显的影响。随模式水平分辨率的提高,模式模拟的飑线弓状回波结构更精细。与3 km分辨率相比,1和0.5 km分辨率模式能很好模拟出飑线后部下沉气流和前部上升气流,模拟的冷池前沿最大风速相对更接近实况。二阶矩微物理方案更能模拟出飑线弓状回波前强后弱的结构特征和飑线过境地面降温幅度,模拟的飑线移动速度、冷池面积和强度、冷池前沿最大风速和雨水蒸发率等均小于一阶矩微物理方案的模拟值。采用1和0.5 km模式水平分辨率及二阶矩微物理方案模式模拟的飑线与WSR-88D多普勒天气雷达探测实况更接近。模式分辨率的提高有利于模拟飑线的维持。对业务数值预报模式模拟飑线阵风锋而言,在计算条件允许的情况下,模式水平分辨率达1 km并采用二阶矩微物理方案可能是需要的。结果还表明,冷池前沿最大风速、冷池强度、模式底层降温幅度、飑线移动速度与雨水蒸发率存在对应的变化趋势,飑线移动速度的变化对飑线阵风锋地面大风的预报有指示意义。改善数值模式对飑线阵风锋预报性能除需关注模式水平分辨率和微物理方案外,还需关注数值模式对雨水蒸发率的模拟能力。     

A down-valley low-level jet event during T-REX 2006

A prolonged down-valley flow and low-level jet were observed throughout the Enhanced Observing Period 4 (April 28–29) of the 2006 Terrain-induced Rotor Experiment, held in the Owens Valley of California near the town of Independence. The low-level jet was strongest during the nocturnal hours, and special field observations captured important details of the event lifecycle. High-resolution simulations using the Weather Research and Forecasting numerical weather prediction model were generated, with underlying assumptions being that model resolution, boundary layer physics, and nesting configuration would be dominant controlling factors in reproducing the jet. The large-scale conditions were dry throughout the event, so moist physics were not a significant forcing consideration. For the control simulation, a two-nest (4.5 and 1.5 km grid spacing) configuration with 90 vertical levels was applied. Additionally, the Quasi-Normal Scale Elimination planetary boundary and surface layer option were selected due to its published performance under conditions of stable stratification. Three other sensitivity simulations were run for comparison, differing from the control just in the choice of vertical resolution (60 versus 90 levels with Quasi-Normal Scale Elimination) and planetary boundary/surface layer physics (90 levels/Mellor-Yamada-Jancic; 90 levels/Yonsei State University). Although the gross evolution (location, height, and timing) of the low-level jet is captured by all model runs (with the 1.5 km inner nest providing the more accurate details), there were at times large underestimations of the nocturnal jet speed max in each simulation (approaching 100 % error, or up to almost 10 m s^?1). Overall, the variations of vertical resolution and planetary boundary/surface physics against the control seemed to (1) yield little overall improvement to statistical or subjective evaluations; (2) do little to improve deficiencies in reproducing the magnitude strength of the nocturnal down-valley low-level jet. Since the cold-start simulations spanned 36 h (including a 12-h spin-up period), it was suspected that the lateral boundary conditions imposed on the outermost 4.5 km nest might negatively impact the interior model solutions in the Owens Valley. To investigate this possibility, an additional simulation was executed by adding two extra nests to the control configuration: an outer 13.5 km and an inner 500 m. This simulation produced a better evolution of the nocturnal low-level jet and especially the speed max. The addition of the larger 13.5 km nest appears more critical to this improvement than that of the extra spatial resolution provided by the inner 500 m nest, which supports the idea that accurate capturing of the large-scale synoptic condition was critical in reproducing important details of this down-valley low-level jet event. The extra 500-m resolution did seem to improve the morning valley cold pool forecast.     

Transport of Snow by the Wind: A Comparison Between Observations in Ad��lie Land, Antarctica, and Simulations Made with the Regional Climate Model MAR

For the first time a simulation of blowing snow events was validated in detail using one-month long observations (January 2010) made in Adélie Land, Antarctica. A regional climate model featuring a coupled atmosphere/blowing snow/snowpack model is forced laterally by meteorological re-analyses. The vertical grid spacing was 2 m from 2 to 20 m above the surface and the horizontal grid spacing was 5?km. The simulation was validated by comparing the occurrence of blowing snow events and other meteorological parameters at two automatic weather stations. The Nash test allowed us to compute efficiencies of the simulation. The regional climate model simulated the observed wind speed with a positive efficiency (0.69). Wind speeds higher than 12 m s ^?1 were underestimated. Positive efficiency of the simulated wind speed was a prerequisite for validating the blowing snow model. Temperatures were simulated with a slightly negative efficiency (?0.16) due to overestimation of the amplitude of the diurnal cycle during one week, probably because the cloud cover was underestimated at that location during the period concerned. Snowfall events were correctly simulated by our model, as confirmed by field reports. Because observations suggested that our instrument (an acoustic sounder) tends to overestimate the blowing snow flux, data were not sufficiently accurate to allow the complete validation of snow drift values. However, the simulation of blowing snow occurrence was in good agreement with the observations made during the first 20 days of January 2010, despite the fact that the blowing snow flux may be underestimated by the regional climate model during pure blowing snow events. We found that blowing snow occurs in Adélie Land only when the 30-min wind speed value at 2 m a.g.l. is >10 m s ^?1. The validation for the last 10 days of January 2010 was less satisfactory because of complications introduced by surface melting and refreezing.     

山西省大雾的中尺度数值模拟

利用中尺度模式MM5（The Fifth—Generation NCAR／Penn State Mesoscale Model）对山西省2009年发生的3场典型雾个例进行了数值模拟,探讨了物理过程参数化方案对雾数值模拟的影响,确定了基于模式模拟数据的雾判别指标,为该地区大雾数值预报系统的研制提供了理论基础。结果表明,综合考虑边界层方案和辐射方案对地表温度、高空温度、2m温度及相对湿度、10m风速、雾的空间分布、雾的生消过程、雾的发展高度等要素数值模拟的影响,边界层方案选用high-resolution planetary boundary layers cheme（HIR）方案、辐射方案选用Cloud方案时,雾数值模拟的结果与实况更为一致。综合分析多个典型雾个例的模拟结果,山西省境内雾的预报指标为：20m液态水含量为0．13～0．6g·kg^（-1）,20～1500m高度大气层存在逆温层,10m风速小于4m·s^（-1）。     

Evaluation of a micro-scale wind model’s performance over realistic building clusters using wind tunnel experiments

The simulation performance over complex building clusters of a wind simulation model(Wind Information Field Fast Analysis model, WIFFA) in a micro-scale air pollutant dispersion model system(Urban Microscale Air Pollution dispersion Simulation model, UMAPS) is evaluated using various wind tunnel experimental data including the CEDVAL(Compilation of Experimental Data for Validation of Micro-Scale Dispersion Models) wind tunnel experiment data and the NJU-FZ experiment data(Nanjing University-Fang Zhuang neighborhood wind tunnel experiment data). The results show that the wind model can reproduce the vortexes triggered by urban buildings well, and the flow patterns in urban street canyons and building clusters can also be represented. Due to the complex shapes of buildings and their distributions, the simulation deviations/discrepancies from the measurements are usually caused by the simplification of the building shapes and the determination of the key zone sizes. The computational efficiencies of different cases are also discussed in this paper. The model has a high computational efficiency compared to traditional numerical models that solve the Navier–Stokes equations, and can produce very high-resolution(1–5 m) wind fields of a complex neighborhood scale urban building canopy(～ 1 km ×1km) in less than 3 min when run on a personal computer.     

一种偏差订正方法在平昌冬奥会气象预报的应用

为了提高GRAPES_3 km（Global/Regional Assimilation and Prediction System）模式在2018年平昌冬奥会气象服务中的预报能力,采用一阶自适应的卡尔曼滤波方法对GRAPES_3 km模式的2 m气温、2 m相对湿度和10 m风开展偏差订正。结果表明:偏差订正方法明显提高了地面要素的预报效果,其中2 m气温的均方根误差整体减小到2℃左右,站点订正改善率为10%~60%;10 m风速的均方根误差减小到2 m·s-1左右,站点订正改善率为10%~45%;2 m相对湿度减小到20%以下,站点订正改善率为0~20%。与韩国气象厅LDAPS（Local Data Assimilation and Prediction System）及美国宇航局NU-WRF（NASA-Unified WRF）模式相比,GRAPES_3 km模式的风速预报表现更为优异,各站点整体预报效果明显优于LDAPS和NU-WRF模式。偏差订正方法可有效改善模式在复杂地形条件下的预报能力,是提高精细化预报准确率的重要手段。     

华南区域高分辨率数值模式前汛期预报初步评估

基于华南地区自动站逐小时观测资料, 采用传统站点评分、邻域法等评估华南区域高分辨率数值模式(包括GRAPES_GZ_R 1 km模式和GRAPES_GZ 3 km模式)对降水、地面温度和风场等要素的预报能力。结果表明: GRAPES_GZ_R 1 km模式的降水预报技巧优于GRAPES_GZ 3 km模式, 模式预报以正偏差为主。对于不同起报时间的预报, 00时(世界时, 下同)起报的预报效果优于12时。GRAPES_GZ_R 1 km模式的TS评分是GRAPES_GZ 3 km模式的两倍以上, 对不同降水阈值的评分均较高。分数技巧评分(FSS)显示GRAPES_GZ_R 1 km模式6 h累计降水预报在0.1 mm、1 mm及5 mm以上的降水均可达到最低预报技巧尺度, 对所检验降水对象的空间位置把握能力更好。2 m气温和10 m风速检验结果表明两个模式均能较好把握广东省温度的分布特征, GRAPES_GZ_R 1 km模式对2 m气温预报结果优于GRAPES_GZ 3 km模式, 预报绝对误差更小; 两个模式对风速的预报整体偏强, 预报偏差在1~4 m/s之间, 但相比之下GRAPES_GZ 3 km模式在风场预报上表现更好。GRAPES_GZ_R 1 km模式的2 m气温和10 m风速预报偏差随降水过程存在明显波动, 强降水过后温度预报整体偏低, 风速预报偏强, 在模式产品订正、使用等需要考虑模式对主要天气系统的预报情况。总的来说, GRAPES_GZ_R 1 km模式的预报产品具有较好的参考价值。      

台风麦莎登陆后粘性摩擦对正压特征波动的影响

本文推导出柱坐标系下含有粘性摩擦项的正压方程组。选取2005年台风麦莎登陆浙江过程中的8月6日15时的WRF(Weather Research and Forecasting)模式输出资料,利用数值差分方法对该正压方程组求特征波解,分析粘性摩擦对台风麦莎内部正压特征波动的影响。结果表明,重力惯性外波在粘性摩擦的影响下,最不稳定波的波数为45左右,波动在摩擦的影响下衰减,波动沿逆时针传播,在半径1000 km处,1波波速为47.43 m/s,在半径r>800 km的范围内,径向风分量扰动加大,辐合辐散运动增强;而摩擦影响下的涡旋Rossby波,2波最不稳定,波动增长率减小,在半径r=200 km处波动相速度为4.282~29.172 m/s,扰动涡度大值区范围减小,涡旋Rossby波的波动区域沿着径向向台风中心收缩。分析包含所有波动时,考虑摩擦后,最不稳定波数在45左右且波动衰减,1波波速在r=1000 km处(外螺旋雨带)为26.374 m/s;在半径r=200 km(内螺旋雨带)为5.275 m/s,考虑径向基本气流后,最不稳定波的波数保持不变,半径r=1000 km处的波速增加为30.324 m/s,r=200 km(内螺旋雨带)处波速为6.065 m/s,摩擦使得径向风分量扰动明显增大,辐合辐散运动加强。     

催化对降水区域外效应的中尺度数值模拟

The Fifth-Generation NCAR/Penn State Mesoscale Model (MM5) has been used to investigate the extra-area effects of silver iodide (AgI) seeding on stratiform clouds performed at the supercooled layer.A bulk two-moment microphysical scheme and the new software package for silver iodide are incorporated in MM5.Extra conservation equations are applied to trace the seeding agent,which is transported along the flow field and interacts with the supercooled cloud fields.In this study,the model was run using three nested grids,with 3.3 km × 3.3 km horizontal resolution in the finest grid.The model results showed that seeding with AgI at the 5 to 15℃ levels had microphysical effects on the simulated clouds and that the simulation produced a longer-lasting seeding effect because of the transport of the seeding agent by upper-level winds.Most of the AgI particles acted as deposition nuclei,and the deposition nucleation process contributed mostly to additional cloud ice formation in this study.The results showed that more precipitation results from seeded than unseeded case,and the precipitation was redistributed downwind of the target.Augmented precipitation (varying from 5% to 25% downwind) was confined in space to within 250 km of the seeding target and in time to the 3-h period after initial seeding.     

Impact of refined land surface properties on the simulation of a heavy convective rainfall process in the Pearl River Delta region,China

The location and occurrence time of convective rainfalls have attracted great public concern as they can lead to terrible disasters. However, the simulation results of convective rainfalls in the Pearl River Delta region often show significant discrepancies from the observations. One of the major causes lies in the inaccurate geographic distribution of land surface properties used in the model simulation of the heavy precipitation. In this study, we replaced the default soil and vegetation datasets of Weather Research and Forecasting (WRF) model with two refined datasets, i.e. the GlobCover 2009 (GLC2009) land cover map and the Harmonized World Soil Database (HWSD) soil texture, to investigate the impact of vegetation and soil on the rainfall patterns. The result showed that the simulation patterns of convective rainfalls obtained from the coupled refined datasets are more consistent with the observations than those obtained from the default ones. By using the coupled refined land surface datasets, the overlap ratio of high precipitation districts reached 36.3% with a variance of 28.5 km from the observed maximum rainfall position, while those of the default United States Geological Survey (USGS) dataset and Moderate Resolution Imaging Spectroradiometer (MODIS) dataset are 17.0%/32.8 km and 24.9%/49.0 km, respectively. The simulated total rainfall amount and occurrence time using the coupled refined datasets are the closest to the observed peak values. In addition, the HWSD soil data has improved the accuracy of the simulated precipitation amount, and the GLC2009 land cover data also did better in catching the early peak time.     

Atmospheric Rivers and Mei-yu Rainfall in China:A Case Study of Summer 2020※

Atmospheric rivers (ARs) are long, narrow, and transient filaments of strong horizontal water vapor transport that can lead to extreme precipitation. To investigate the relationship between ARs and mei-yu rainfall in China, the mei-yu season of 2020 in the Yangtze-Huaihe River basin is taken as an example. An adjusted AR-detection algorithm is applied on integrated water vapor transport (IVT) of the ERA5 reanalysis. The JRA-55 reanalysis and the data from Integrated Multi-satellite Retrievals for GPM (IMERG) are also utilized to study the impacts of ARs on mei-yu rainfall in 2020. The results reveal that ARs in East Asia have an average length of 5400 km, a width of 600 km, a length/width ratio of 9.3, and a northeastward orientation of 30°. ARs are modulated by the western North Pacific subtropical high. The IVT core is located at the south side of low pressure systems, moving eastward with a speed of 10° d^?1. For the cross sections of ARs in the Yangtze-Huaihe River basin, 75% of the total flux is concentrated below 4 km with low-level jets near AR cores. Moreover, ARs occur mainly in the mei-yu period with a frequency of 20%–60%. The intensity of AR-related precipitation is 6–12 times that of AR-unrelated precipitation, and AR-related precipitation contributes about 50%–80% to total mei-yu precipitation. As shown in this case study of summer 2020, ARs are an essential part of the mei-yu system and have great impacts on mei-yu rainfall. Thus, ARs should receive more attention in research and weather forecast practices.     

基于J-Earth领域编程框架的高效并行大气环流模式(H-GAMIL)构建

高性能并行计算,一直是气候系统模式研究中的重要科学前沿。然而,随着计算机技术的发展,计算机体系结构越来越复杂,以及气候系统模式本身的复杂性使得开发高性能气候系统模式成为一个难题。采用了一种全新的并行程序设计方法:基于针对气象科学中的计算特点发展出的领域编程框架J-Earth,构建了高性能并行大气环流模式H-GAMIL。在对原模式GAMIL分析和解构以后,结合J-Earth的特点,采用面相对象的编程方式,对H-GAMIL进行重构。重构后的H-GAMIL具有现代软件标准化模块化的特点,并具有自动二维剖分、MPI(Message Passing Interface)和Open MP(Open Multi-Processing)混合并行实现、并行输出等新功能。解决了原模式一维剖分受到处理器限制,串行IO等问题。在此基础上,本文对H-GAMIL进行了并行效率测试,结果显示,模式水平分辨率为1°(纬度)×1°(经度)时,可使用上千处理器核,并行效率达40%以上且负载平衡达70%。测试结果表明了H-GAMIL具有较好的并行效率和可扩展性;同时对H-GAMIL模拟性能进行对比分析,其结果表明H-GAMIL保持了原模式的守恒性,并且具有与原模式相当的模拟性能,能够满足气候研究的计算需求,达到实用程度。     

Hydrostatic and non-hydrostatic simulations of moist convection: Review and further study

Summary Quantitatively comparative experiments of moist convection using hydrostatic and non-hydrostatic models are reviewed and a further study is made of the suitability of the hydrostatic approximation for a high-resolution model when the grid size falls below 20 km. Idealized moist convection is treated, and then the torrential rain that occurred on 6 August 1993 in Kagoshima, southern Kyushu, Japan is simulated by each model. An explicit warm-rain process predicting cloud water and rainwater and the scheme of moist convective adjustment are individually or conjunctively employed in the model. The effect of hydrostatic water loading is also examined in detall.For the simulation of idealized convection, the hydrostatic simulation tends to overestimate and overexpand precipitation in comparison with the non-hydrostatic counterpart, and the drag effect of hydrostatic water loading is more significant for convective development than the non-hydrostatic effect. In the 20-km simulations, however, the hydrostatic simulation with hydrostatic water loading produces results that are comparable to the nonhydrostatic counterpart. For the simulation with real data, the comparative results well correspond to those of idealized convection. Furthermore, the 5 km hydrostatic simulation overestimates total precipitation more than that of dealized convection. On the basis of these results, when developing 1020 km numerical weather prediction (NWP) models, hydrostatic water loading should be evaluated in preference to adopting non-hydrostatic models, and a non-hydrostatic model with hydrostatic water loading is thought to be recommendable for a high-resolution NWP model.With 7 Figures