Vol.28, NO.1, 2017

Development and Operation Transformation of GRAPES Global Middle-range Forecast System
Shen Xueshun, Su Yong, Hu Jianglin, Wang Jincheng, Sun Jian, Xue Jishan, Han Wei, Zhang Hongliang, Lu Huijuan, Zhang Hua, Chen Qiying, Liu Yan, Liu Qijun, Ma Zhanshan, Jin Zhiyan, Li Xingliang, Liu Kun, Zhao Bin, Zhou Bin, Gong Jiandong, Chen Dehui, Wang Jianjie
2017, 28(1): 1-10. DOI: 10.11898/1001-7313.20170101
The developing history of GRAPES global middle-range numerical weather prediction system (GRAPES_GFS) of China Meteorological Administration is reviewed. Important progresses in recent years are summarized and their contributions to GRAPES_GFS operation are introduced.From the aspect of dynamic frame aspect, an algorithm for vertical advection of temperature and the polar filter scheme are improved. New algorithms are introduced, including terrain filtering algorithm, scalar advection scheme with conservation and high accuracy, w-damping noise suppression algorithm, and Rayleigh friction in the stratosphere, etc. Besides, horizontal and vertical resolutions are enhanced. These improvements significantly improve the stability, accuracy and mass conservation of the dynamic core.From the aspect of physical process, the RRTMG radiation program is upgraded, the CoLM land surface process scheme is introduced, the cumulus convective scheme and boundary layer scheme are improved, and a two-parameter cloud physics scheme is developed. On these basis, the prediction cloud scheme is further developed, the interface between dynamic and physics is adjusted, the calculation of sea ice and surface albedo are also optimized. These improvements and optimizations improve the prediction ability of the physical package.From the aspect of global three-dimensional variational assimilation (3DVar), the model space 3DVar is developed to avoid the interpolation error of the analysis space to the model space, fine quality control and deviation correction techniques are developed to achieve high quality observation data assimilation, and more satellite data assimilation techniques are adopted especially using satellite hyperspectral infrared detector as the focus.At the same time, the prediction ability of GRAPES_GFS2.0 is being evaluated based on results of two-year assimilation forecast cycle test, and compared with T639. Generally speaking, the forecast indicators of the system are fully beyond the GRAPES_GFS 1.0 version. Model outputs of isobaric elements in the troposphere forecast, including precipitation and 2 m temperature, have obvious advantages comparing with T639.
Improvements and Performances of the Operational GRAPES_GFS 3DVar System
Wang Jincheng, Lu Huijuan, Han Wei, Liu Yan, Wang Ruichun, Zhang Hua, Huang Jing, Liu Yongzhu, Hao Min, Li Juan, Tian Weihong
2017, 28(1): 11-24. DOI: 10.11898/1001-7313.20170102
In recent years, the capability and stability of GRAPES (Global/Regional Assimilation and PrEdiction System) three-dimensional variation data assimilation system (3DVar) is upgraded and improved gradually in Numerical Weather Prediction Center of China Meteorological Administration. Improvements in analysis scheme and assimilating data technique for GRAPES 3DVar in the past two years are overviewed. Then the capability and performance of G-M3DVar latest version are evaluated by two-year length experiments. The accuracy and precision of G-M3DVar analyses is evaluated against radiosonde observation and ERA-Interim reanalysis and is compared with NCEP FNL and T639 analysis.Taken radiosonde data as a reference, the root mean square error and bias of pressure analyses of G-M3DVar are smaller than ERA-Interim reanalysis and NCEP FNL analysis data in all domains in both winter and summer seasons. The root mean square error and bias of u wind analysis of G-M3DVar are larger than ERA-Interim reanalysis and NCEP FNL analysis in the Tropics. However, in the Northern Hemisphere, the root mean square error and bias of u wind of G-M3DVar are similar to ERA-Interim below 250 hPa. In the Southern Hemisphere, the root mean squared error of u wind of G-M3DVar is the largest compared to EAR-Interim reanalysis and NCEP FNL analysis. For humidity field, the bias of G-M3DVar analysis is smaller than EAR-Interim reanalysis and NCEP FNL analysis in the middle and high troposphere, which means that the humidity analysis of G-M3DVar is much drier than ERA-Interim and NCEP FNL data especially in the middle and high troposphere. Taken ERA-Interim reanalysis data as a reference, the root mean square error of G-M3DVar analysis is smaller than the T639 analysis but larger than NCEP FNL analysis data for all fields excluded the humidity.In conclusion, the quality of G-M3DVar analysis is better than T639 analysis and satisfies requirements of operational run. In recent years, the gap of analyses between G-M3DVar and advanced numerical weather centers such as ECMWF keeps growing, although the accuracy of G-M3DVar analysis is improved significantly in the past two years. Much more focus and works should be paid in the following aspects. First, the background error covariance (BE) is estimated by National Meteorological Center of USA (NMC) method, which is static and climatological. The static and climatological BE is far from meeting requirements of the modern numerical weather prediction. Second, the quality control scheme for all observations in G-M3DVar is still relatively inexactly and incapable. Third, the bias correction scheme for microwave radiance in G-M3DVar is still static which has been proved to have some shortcomings.
Main Technical Improvements of GRAPES_Meso V4.0 and Verification
Huang Liping, Chen Dehui, Deng Liantang, Xu Zhifang, Yu Fei, Jiang Yuan, Zhou Feifei
2017, 28(1): 25-37. DOI: 10.11898/1001-7313.20170103
After operational implementation of GRAPES_Meso V3.0 in March 2013, some problems are found, which include over-prediction of precipitation, integration instability, large 2 m temperature forecast errors, insufficient observations assimilated, and coarser resolution. To deal with these problems, a lot of changes are made, mainly including introducing variational quality control scheme, applying the bias correction for sounding humidity observation, assimilating GPS/PW data, FY-2E cloud drift wind and radio occultation observation, increasing resolution of the model, using the fourth horizontal diffusion scheme, adjusting the coupling scheme between dynamic core and WSM6 microphysics parameterization, optimizing land surface model, and improving diagnostic algorithm of composite radar reflectivity. GRAPES_Meso is also upgraded from Version 3.0 to Version 4.0 by integrating all of the progresses mentioned above. One month hindcast experiments are implemented and results show that, compared with GRAPES_Meso V3.0, ETS scores of precipitation forecasts for GRAPES_Meso V4.0 are obviously higher for all five thresholds of 24 h accumulated precipitation, and the bias is largely decreased for light, moderate and heavy rainfall thresholds. The monthly mean precipitation pattern and intensity are both closer to observation, and the detail precipitation distribution can be reproduced better. Daily time evolutions of root mean square errors for 2 m temperature forecasts are very similar, while the amount of V4.0 is much less than that of V3.0. Monthly mean errors are reduced about 1-2℃ over most region of China and even 3-5℃ over some region for 24 h forecast. It is apparent that GRAPES_Meso V4.0 performs better for height, temperature and wind fields, as anomaly correlation coefficients of these fields at 500 hPa are larger and root mean square errors of these fields at 850 hPa are less than those by GRAPES_Meso V3.0. The forecast skill of GRAPES_Meso is largely improved from Version 3.0 to Version 4.0. Also, the unified process control has been implemented for GRAPES_Meso and GRAPES_RAFS (Rapid Analysis and Forecast System), which can reduce the system maintenance and management costs significantly. GRAPES_Meso V4.0 is transitioned into operational run at China National Meteorological Center with horizontal resolution of 0.1°×0.1° and vertical resolution of 50 levels from July 2014 and the whole system running is stable.
Three-dimensional Cloud Initial Field Created and Applied to GRAPES Numerical Weather Prediction Nowcasting
Zhu lijuan, Gong Jiandong, Huang Liping, Chen Dehui, Jiang Yuan, Deng Liantang
2017, 28(1): 38-51. DOI: 10.11898/1001-7313.20170104
In order to get more accurate cloud initial fields in GRAPES_Meso model, the ARPS cloud analysis scheme is introduced. With some modifications or improvements based on the rational law in cloud macro-characteristic and micro-characteristic, the cloud analysis scheme is used to set up a local cloud analysis scheme which suitable for domestic numerical weather prediction model and local synoptic observations. Based on the background field, it integrates data sources from Doppler weather radar three-dimensional mosaic reflectivity data, geostationary meteorological satellite data, and surface observation. The cloud initial information is analyzed based on cloud physical laws of thermodynamics and dynamics and the observed empirical relationship. After the cloud analysis, analyzed three dimensional fields which include information of cloud hydrometeors are introduced by nudging technique for initialization of GRAPES_Meso model. One-month (15 Jul 2014-14 Aug 2014) time serial of experiments in different horizontal resolution (0.03°×0.03°, 0.1°×0.1°) are designed to verify the performance of the cloud analysis scheme. Case study shows that cloud macro-characteristic and cloud initial hydrometeors of synoptic system, such as typhoon, squall line etc., could be represented better by using cloud analysis scheme. The satellite cloud simulation technology of university of Wisconsin is adopted to produce the satellite cloud simulation product, which is convenient to compare the cloud product of model output with FY-2 meteorological satellite cloud products. The comparing result shows that 1 h nowcasting cloud of GRAPES_Meso model with cloud information initialized is close to satellite measurment in cloud macro-characteristic and cloud spatial distribution, while the one without cloud information initialized is missing and the brightness temperature is higher than satellite measurment. Until 6 hours, the nowcasting cloud of cloud analysis scheme is more similar to satellite measurment than the one without cloud analysis scheme, and the brightness temperature simulation is reasonable. As for performance of precipitation forecast, it is found that forecast with the cloud analysis has a significant positive impact on short range precipitation forecast. The 1-hour precipitation forecast with cloud analysis is closer to observation, and the positive effect can last for over twelve hours, which meets the demand for the short time nowcasting operational system. Furthermore, the spin-up time is also shortened. In long time experiments, the statistical variable of equitable threat score (ETS) of the precipitation forecast is calculated. At the first 6 h forecast in horizontal resolution of both 0.03°×0.03° and 0.1°×0.1°, the ETS of the precipitation forecast with cloud analysis is obviously increased compared with the one without cloud analysis. In the following three 6 h forecast, the positive effects decrease as forecast time increasing.
Boundary Layer Coupling to Charney-Phillips Vertical Grid in GRAPES Model
Chen Jiong, Ma Zhanshan, Su Yong
2017, 28(1): 52-61. DOI: 10.11898/1001-7313.20170105
It is an important challenge in numerical weather and climate prediction to obtain accurate coupling between physical parameterization and high resolution dynamic framework. Increased resolution in models and the use of large time-steps in semi-Langrangian advection stress the need for an equally accurate computation in time of the corresponding physical parameterizations and the physics-dynamics coupling on the temporal aspects. Physics-dynamics coupling on spatial aspects also plays a very important role on the accuracy of model predictions, for there is a choice for how to vertically arrange the predicted variables, namely, the Lorenz and Charney-Phillips grids.The physics-dynamics coupling on spatial aspects in GRAPES model is studied. As the Charney-Philips grid is used, the horizontal velocity is staggered relative to potential temperature, which means potential temperature and water substances are calculated at full levels, while horizontal velocity is calculated at half levels. In Lorenz physics scheme, all variables are set at half levels and the correspondent tendencies are estimated at half levels. The interpolation has to be used between full and half levels in physics-dynamics coupling before and after physics scheme package is called. The interpolation error is unavoidable and an unexpected zigzag noise appears because of the second-order difference in PBL (planetary boundary layer) scheme.In C-P PBL scheme, the momentum diffusivity KM is required at full levels and the heat diffusivity KH is required at half levels. It is easy to compute KM and KH in unstable PBL because KM and KH depend on the PBL height and surface variables. For local scheme in stable PBL and free convective atmosphere, diffusivities are functions of local Richardson number which has relation with both potential temperature and horizontal velocity. Here potential temperature gradient is averaged so that Richardson number is calculated at full levels. KM can be calculated at the full level and KH can be averaged at the half level. The boundary condition is given by the surface flux according to the constant flux layer. C-P PBL parameterization is developed to assure the accurate coupling of PBL physics and vertical Charney-Phillips grid.Improvements are detected using C-P PBL parameterization spatial physics-dynamics coupling in GRAPES_GFS model. The zigzag noise of temperature and moisture in PBL is removed and the correspondent profiles appear to be smooth with C-P PBL parameterization. The accuracy of PBL and dynamics coupling is improved, and an overall enhancement is found in the forecast of height and temperature.
The Optimization of GRAPES Global Tangent Linear Model and Adjoint Model
Liu Yongzhu, Zhang Lin, Jin Zhiyan
2017, 28(1): 62-71. DOI: 10.11898/1001-7313.20170106
Adjoint models are widely applied in numerical weather prediction. For instance, in four-dimensional variational data assimilation (4DVar), they are the best method to efficiently determine optimal initial conditions. The minimization of the 4DVar cost function is solved with an iterative algorithm and is computationally demanding. Though the minimization is usually performed with a much lower resolution than in forecast model, obtaining the optimal model state requires dozens of iterations, and the model parallel efficiency must be fast enough. However, the parallel efficiency of GRAPES global tangent linear model and adjoint model version 1.0 based on GRAPES global non-linear model 1.0 is so low that it seriously impacts the development of GRAPES_4DVar. In order to reduce the computational cost, a new tangent linear model and adjoint model version 2.0 are re-designed and re-developed based on GRAPES global model version 2.0. By optimizing the program structure of tangent linear model, the calculating time of GRAPES tangent linear model can be best controlled within 1.2 times of GRAPES non-linear model's consumption with only dynamic framework. And by methods transferring the model base state and trajectory to the adjoint model, the calculating time of GRAPES adjoint model can be best controlled within 1.5 times of GRAPES non-linear model's consumption. Therefore, the new GRAPES tangent linear model and adjoint model version 2.0 are very successful in terms of computational efficiency to speed up the development of GRAPES_4DVar.In practical applications, the tangent linear model and adjoint model is run at a lower resolution than the non-linear model, since the dynamics is already simplified through the reduction in horizontal resolution, the linearized physics doesn't necessarily need to be exactly tangent to the full physics. In principle, physical parameterizations can already behave differently between non-linear and tangent-linear models due to the change in resolution. In order to reduce computational cost, it is often necessary to select different set of simplified linearized parameterizations with the full physical processes of GRAPES forecast model. By decoupling base states calculation in GRAPES and the perturbation calculation in the tangent linear and adjoint model, the computational cost of GRAPES tangent and adjoint model with simplified physical parameterizations increases only a little than no physical parameterizations versions, and the computational efficiency is higher than GRAPES forecast model with full physical parameterizations.
Diurnal Variations of Summer Precipitation in Xinjiang
Chen Chunyan, Wang Jianjie, Tang Ye, Mao Weiyi
2017, 28(1): 72-85. DOI: 10.11898/1001-7313.20170107
Fundamental diurnal variation features of summer precipitation over Xinjiang are investigated based on hourly precipitation data of 1991-2014 from all 16 national stations in the target region, and distinct characteristics on the diversity of diurnal variations of summer precipitation between Northern and Southern Xinjiang are revealed. Results show that the diurnal variation of the precipitation amount in Northern Xinjiang presents nearly a single peak and trough, with the maximum at dusk (1600-2000 LST) and minimum about one-third proportion of peak value at noon time, and the precipitation amount almost decrease monotonically during the whole night. Differently, the diurnal variation of the precipitation amount in Southern Xinjiang shows triple peaks and troughs, with peaks at about dusk (1700-1800 LST), after the midnight (0000-0100 LST) and near the noon (1000 LST), respectively. The difference of single peak pattern between Northern Xinjiang and central-east China is the accumulation of different situations instead of similar peak overlay. In addition, the morning when precipitation peak value appears in Southern Xinjiang is opposite to that weaker or weakest time in all day in central-east China.The occurrence time of maximum average precipitation intensity and maximum hourly accumulative precipitation is more consistent with the counterpart of maximum hourly accumulative precipitation frequency, and more remarkable with the region where daily variation of precipitation presents multi-peak patterns, showing more contribution of precipitation intensity than frequency to accumulative precipitation, which is also an important characteristic of Xinjiang, especially in Southern Xinjiang. Short-duration precipitation within 6 hours are dominated events in summer (the average value is 85 percent which is significantly higher than that in central-east China), and precipitation events lasting longer than 12 hours happen rarely. Contribution rates of short-duration precipitation events within 6 hours in total precipitation are up to 54% which is higher than that in central-east China in most region except the east side of Tianshan in Xinjiang.Except in the south edge of Tarim Basin, there are close relations between the precipitation diurnal cycle and its duration in most areas of Xinjiang, but not the same as those in central-east China. Namely, 2-3-hour short-duration precipitation events contribute to the peak value of the total precipitation diurnal cycle in western and Northern Xinjiang, whereas every duration event within 12 hours has nearly equal contribution to the maximum of the total precipitation diurnal cycle in central to eastern Tianshan Mountains.
Decadal Variation Characteristics of South China Pre-flood Season Persistent Rainstorm and Its Mechanism
Chen Si, Gao Jianyun, Huang Lina, You Lijun
2017, 28(1): 86-97. DOI: 10.11898/1001-7313.20170108
Based on the daily precipitation data of 243 stations in Fujian, Guangdong, Guangxi during 1961-2012, indices (intensity of daily rainstorm IRD, intensity of persistent rainstorm IRS, intensity of South China pre-flood season persistent rainstorm IRZ) are calculated and persistent rainstorm processes are selected. Indices reveal decadal variation characteristics of pre-flood season persistent rainstorm by kinds of mutation test and percentile method. It indicates that the decadal variation of pre-flood season persistent rainstorm can be divided into three sections, multiple (1961-1972)-fewer (1973-1991)-multiple (1992-2012). The present stage is in multiple sections with stronger intensity and longer duration; pre-flood season precipitation exhibits a significant low-frequency oscillation characteristic. The decadal variation of duration and intensity of persistent rainstorm during pre-flood season are closely related to that of configuration and intensity of low-frequency periodic signal. There is a positive correlation between the intensity of pre-flood season persistent rainstorm and the low-frequency intensity of 10-60 days with the intensity strongest in 10-20 days, the second in 30-60 days' and the third in 20-30 days'. The low-frequency signal of Guangdong and Fujian is more obvious than Guangxi in all sections. Frequent, strong and long persistent rainstorm is more likely to occur in multiple sections as the precipitation intensity of low-frequency signal oscillating acutely, and vice versa. The decadal meridional propagation features of tropical intraseasonal oscillation and mechanism and its possible impact on the decadal variation of pre-flood season persistent rainstorm are also studied by using East Asia-western NorthPacific ISO index (EAWNP ISO). The position of the meridional propagation of the East Asia-Pacific Northwest Pacific with different phases can be seen according to different phases' reconstruction of 850 hPa wind speed and OLR anomaly field. The decadal variation of tropical low-frequency northward movement signal cycle and intensity may probably be one of the main causes that lead to the decadal variation of pre-flood season persistent rainstorm. Since tropical low-frequency signal has various cycles, it may cause low-frequency variation of circulation system advantage to the precipitation of pre-flood season, which corresponds to the low-frequency variation characteristics of precipitation there and leading to changes of the intensity and duration of persistent rainstorm when low-frequency convection propagates northward to southeast coast of China. The longer (shorter) cycle and stronger (weaker) intensity of tropical low-frequency signal northward to southeast coast of China is, the longer (shorter) duration and stronger (weaker) intensity of pre-flood season persistent rainstorm may be.
Variation Characteristics of Soil Temperature & Moisture and Air Parameters in the Source Region of the Yellow River
Chen Jinlei, Wen Jun, Liu Rong, Jia Dongyu, Wang Zuoliang, Luo Qi, Xie Yan
2017, 28(1): 98-108. DOI: 10.11898/1001-7313.20170109
The source region of the Yellow River (SRYR) located in the northeast of the Tibetan Plateau, is the crucial water conservation area. Soil temperature & moisture variations and associated climate effects have important implications to the change of runoff. Three kinds of frequently used reanalysis datasets, ERA-Interim, CFSR, and JRA-55 are tested using field observations of Maqu Soil Temperature & Moisture Network so as to find the optimal one for SRYR. Combining with observations of Maqu Station, the climate changes in recent 35 years and the temporal variation of soil moisture & temperature are analyzed. In addition, their spatial variations are depicted by reanalysis datasets and CLM4.5(Community Land Model 4.5). Main results are as follows.CFSR is the best dataset to depict the soil moisture variation, and ERA-Interim is better on soil temperature, while JRA-55 is unsuited. Soil temperature has an indication to the climate change, but its response is less significant than air temperature. Soil moisture has an increasing trend, because freezing time becomes shorter and melting time is extending. Air temperature, soil temperature & moisture, except for precipitation, have abruptions in the last 35 years. Air temperature starts to abrupt during 1997-2000, after that it shows significant upward trend. Precipitation decreases from 1987 to 2004 and increases after 2005. Abrupt change of soil temperature takes place during 1985-1986, and beyond the belief line after 1994 with prominent rising. It means soil temperature is more sensitive than air temperature to climate warming. Soil moisture has an upward abruption in 2002. Soil temperature & moisture in 10 cm depth become warm and dry in recent years. Lakes and the Yellow River are the cold and wet centers in warm season, and turn warm and dry in cold season. CLM4.5 has high simulation accuracy, and is capable of describing detailed changes of soil in SRYR. All in all, it is better than reanalysis dataset in simulating the spatial variation of soil temperature & moisture, but still has a long way comparing with observations.
Fine Forecast of High Road Temperature Along Jiangsu Highways Based on INCA System and METRo Model
Feng Lei, Wang Xiaofeng, He Xiaofeng, Gao Jingjing
2017, 28(1): 109-118. DOI: 10.11898/1001-7313.20170110
The Integrated Nowcasting through Comprehensive Analysis (INCA) system is an observation-based analysis and forecasting system, in which measurements from automatic weather stations, radar data, satellite data, 3D fields from the operational numerical weather prediction (NWP) model and elevation data with high-resolution are incorporated. INCA system adds value to the classical NWP forecast by providing high-resolution analyses, nowcasts and improved forecasts both within and beyond the nowcasting range. A coupling of INCA system and the Environment and Temperature of Roads (METRo) Model is used to study the forecast of high road temperature during summer along highways in Jiangsu Province.Result shows that the highest road temperature forecasting is improved significantly through the coupling of the forecast with observations during the overlap period by adjusting the radiative fluxes to local conditions. The absolute error of the road temperature at 1400 BT is reduced by about 3℃ compared with results with non-coupling. The daily variation and the spatial distribution of the highest road temperature at 1400 BT can be captured by this forecasting system. The average absolute error for the daily highest road temperature is 4.1℃, and the average relative error is about 10.8%. Percentages of stations with absolute error below 5℃ is about 64.5%, and percentages of stations with relative error below 15% is about 74.6%, which is higher than that of the conventional method (a statistic road temperature forecast model driven by atmospheric forecasting from BJ-RUC) by about 23.1%, 25.3%. Analysis for the hot day shows that the level and of the highest road temperature and the location of the highest road temperature center can be reproduced well by this system. The level of the highest road temperature on more than 90% of stations is in accordance with the observation. The absolute error of the road temperature forecasting is below 5℃ for most stations. While the road temperature forecasting for the hottest points are lower than the observation. The distribution of the forecasting error of the road temperature at 1400 BT shows scattered features and there is no obvious positive (negative) error center. Besides, the forecast skill of small fluctuations of road temperature needs to be improved.The source of the road temperature forecasting errors are from both the road temperature forecast model and the numerical weather prediction products. The METRo model is also driven by the station weather observation to test which part is the main error. It estimates that about 70% of error is from the model itself, and notes that the cloud amount and surface pressure observations are from nearby weather stations.
Near-surface Gust Factor Characteristics in Several Disastrous Winds over Zhejiang Province
Zhou Fu, Jiang Lulu, Tu Xiaoping, Shen Huayu, Zheng Zheng
2017, 28(1): 119-128. DOI: 10.11898/1001-7313.20170111
Studies on near-surface gust characteristics in high winds are necessary for weather services. Using the daily 10 min data from automatic weather stations in Zhejiang Province during 2011-2013, characteristics of near-surface gust factors in several kinds of high winds caused by cold air masses, tropical cyclones and abruptly severe convections, are investigated over the offshore and in-land areas of Zhejiang Province. Spatial distributions of wind velocities and gust factors are especially considered, as well as the relationship among gust factors, geographical elements and mean wind speeds. The fuzzy cluster mean (FCM) and stepwise regression methods are applied as well to do the weather station clusters under different weather patterns and set up gust factor forecast models. Result shows that the gust factor distribution displays similarly both in cold air and tropical cyclone strong winds although spatial speed distributions might be different from each other, and wind directions show no effects on gust factor distributions. Disastrous winds usually happen over the offshore seas and coastal areas, with gust factors less than 1.5 and the isolines paralleling to the coastline and descending eastwards. However, over in-land areas of Zhejiang Province, gust factors are generally greater than 2.0 and even more than 3.0 over the hilly regions with gentle wind speeds, indicating enhancing effects of hilly terrain. The average gust factor is more than 1.8 under severe convective systems, which is greater than operational regulations. The convective gale events could occur at any locations within Zhejiang Province, but stations with occurring probabilities more than 10% mainly lay in the coastal and offshore Zhejiang Province, and the terrain roughness doesn't show much influence. Gust factors perform well related to 10 min mean wind speeds and altitudes in high winds by cold air masses and tropical cyclones. FCM analysis indicates that there are few differences in station distributions between clod air mass and tropical cyclone gale events, stations located in the northern and coastal regions often differ from those in the middle and southern areas in Zhejiang Province, and stations with altitudes more than 400 m are different from those with altitudes lower than 70 m. Stepwise regression is carried out to set up forecasting models between gust factors and mean winds and station altitudes before and after FCM clusters, verifications imply that FCM could help improve forecast ability of the models. The regression model for type Ⅰ tends to overestimate gust factors at stations with relative high altitudes, on the contrary, the model for type Ⅱ tends to underestimate gust factors at stations with relative low altitudes.