A new framework for multi-site weather generator: a two-stage model combining a parametric method with a distribution-free shuffle procedure

Keeping the spatial correlation of synthetic precipitation data is of utmost importance for hydrological modeling; however, most present weather generators are single-site models and ignore the spatial dependence in daily weather data. Multi-site weather generator is an effective method to solve this problem. This study proposes a new framework for multi-site weather generator denoted as two-stage weather generator (TSWG), in which the first stage generates the single-site precipitation occurrence and amount with a parametric chain-dependent process, and the second stage rebuilds the spatial correlation of the synthetic data using a post-processing, distribution-free shuffle procedure. Results show that TSWG reproduces the statistical parameters of the parametric stage quite well, such as wet days and precipitation amount, and it almost perfectly preserves the inter-station correlations of precipitation occurrence and amount as well as their dependences. Most important, it matches the input requirement of hydrological model and gives satisfactory hydrological simulations. There are several advantages for this new framework: (1) only one correlation matrix and two simple steps, no more input variables or iterative optimizations, are needed to rebuild the spatial correlation; (2) the statistical parameters of the observed data can be easily preserved; (3) the inter-station correlations can be satisfactorily rebuilt. As a post-processing method, the shuffle procedure used to reconstruct the spatial correlation has some potential extensions, such as turning current single-site weather generator into multi-site models and generating future multi-site climate scenarios.     

Fitting the distribution of dry and wet spells with alternative probability models

The development of the rainfall occurrence model is greatly important not only for data-generation purposes, but also in providing informative resources for future advancements in water-related sectors, such as water resource management and the hydrological and agricultural sectors. Various kinds of probability models had been introduced to a sequence of dry (wet) days by previous researchers in the field. Based on the probability models developed previously, the present study is aimed to propose three types of mixture distributions, namely, the mixture of two log series distributions (LSD), the mixture of the log series Poisson distribution (MLPD), and the mixture of the log series and geometric distributions (MLGD), as the alternative probability models to describe the distribution of dry (wet) spells in daily rainfall events. In order to test the performance of the proposed new models with the other nine existing probability models, 54 data sets which had been published by several authors were reanalyzed in this study. Also, the new data sets of daily observations from the six selected rainfall stations in Peninsular Malaysia for the period 1975–2004 were used. In determining the best fitting distribution to describe the observed distribution of dry (wet) spells, a Chi-square goodness-of-fit test was considered. The results revealed that the new method proposed that MLGD and MLPD showed a better fit as more than half of the data sets successfully fitted the distribution of dry and wet spells. However, the existing models, such as the truncated negative binomial and the modified LSD, were also among the successful probability models to represent the sequence of dry (wet) days in daily rainfall occurrence.     

Rainfall characterisation by application of standardised precipitation index (SPI) in Peninsular Malaysia

The interpretations of trend behaviour for dry and wet events are analysed in order to verify the dryness and wetness episodes. The fitting distribution of rainfall is computed to classify the dry and wet events by applying the standardised precipitation index (SPI). The rainfall amount for each station is categorised into seven categories, namely extremely wet, severely wet, moderately wet, near normal, moderately dry, severely dry and extremely dry. The computation of the SPI is based on the monsoon periods, which include the northeast monsoon, southwest monsoon and inter-monsoon. The trends of the dry and wet periods were then detected using the Mann–Kendall trend test and the results indicate that the major parts of Peninsular Malaysia are characterised by increasing droughts rather than wet events. The annual trends of drought and wet events of the randomly selected stations from each region also yield similar results. Hence, the northwest and southwest regions are predicted to have a higher probability of drought occurrence during a dry event and not much rain during the wet event. The east and west regions, on the other hand, are going through a significant upward trend that implies lower rainfall during the drought episodes and heavy rainfall during the wet events.     

Fitting optimum order of Markov chain models for daily rainfall occurrences in Peninsular Malaysia

The analysis of the daily rainfall occurrence behavior is becoming more important, particularly in water-related sectors. Many studies have identified a more comprehensive pattern of the daily rainfall behavior based on the Markov chain models. One of the aims in fitting the Markov chain models of various orders to the daily rainfall occurrence is to determine the optimum order. In this study, the optimum order of the Markov chain models for a 5-day sequence will be examined in each of the 18 rainfall stations in Peninsular Malaysia, which have been selected based on the availability of the data, using the Akaike’s (AIC) and Bayesian information criteria (BIC). The identification of the most appropriate order in describing the distribution of the wet (dry) spells for each of the rainfall stations is obtained using the Kolmogorov-Smirnov goodness-of-fit test. It is found that the optimum order varies according to the levels of threshold used (e.g., either 0.1 or 10.0 mm), the locations of the region and the types of monsoon seasons. At most stations, the Markov chain models of a higher order are found to be optimum for rainfall occurrence during the northeast monsoon season for both levels of threshold. However, it is generally found that regardless of the monsoon seasons, the first-order model is optimum for the northwestern and eastern regions of the peninsula when the level of thresholds of 10.0 mm is considered. The analysis indicates that the first order of the Markov chain model is found to be most appropriate for describing the distribution of wet spells, whereas the higher-order models are found to be adequate for the dry spells in most of the rainfall stations for both threshold levels and monsoon seasons.     

干湿持续期随机模拟

该文应用数据建模技术, 实现干湿期随机建模。主要包括:利用历史气象资料, 从中采集干湿期数据; 应用实测数据, 创建干湿期经验分布函数; 应用Monte Carlo方法和经验分布参数, 随机生成干湿期序列, 通过和Markov链模型输出的对比分析, 讨论生成序列的统计误差, 测试结果显示, 与两状态Markov链方法相比, 所建模型性能更好。     

Evaluation of the weather generator CLIGEN with daily precipitation characteristics in Korea

The ability of the CLImate GENerator (CLIGEN) weather generator to reproduce daily precipitation characteristics for Korea was assessed on the basis of 55-year long historical daily precipitation records from eight weather stations (Seoul, Incheon, Daegu, Ulsan, Gwangju, Busan, Kangneung, and Jeonju) representing different parts of the Korean peninsula. The basic statistics of daily precipitation (mean, standard deviation, skewness of daily precipitation, number of rainy days, and the lengths of wet/dry period), probability distribution characteristics of daily precipitation (percentiles and maximum value), and the spatial covariance statistic generated by CLIGEN were compared with those derived from the observed weather series. Significance tests were conducted on the difference between the historical and generated statistics with the 1% significance level. The results show that CLIGEN simulates most of the daily precipitation characteristics satisfactorily with a tendency to slightly underestimate the mean and variability of daily precipitation. Especially, the number of rainy days is perfectly reproduced with mean relative error of 0.4% across all the stations. It is also found that the spatial covariance statistic from eight different stations is well reproduced by CLIGEN with respect to the leading EOF mode of summer season daily precipitation.     

基于统计降尺度方法的夏季降水精细化预报

利用1999—2009年安徽省淮河以南地区60个县市站夏季逐日降水资料和安庆市探空站逐日资料,研究了中低层不同风向配置下局地降水与大尺度降水场之间的关系,以3种不同预报对象及相应的预报因子分别采用神经网络和线性回归方法设计6种预报模型对观测资料进行逼近和优化,从而实现空间降尺度.分析对比6种预报模型46站逐日降水量的拟合和预报效果,结果表明:采取相同的预报对象及预报因子的BP神经网络模型在拟合和预报效果上均好于线性回归模型,可见夏季降水场之间以非线性相关为主;神经网络模型预报结果同常用的Cressman插值预报相比,能很好地反映出降水的基本分布及局地特征;预报对象为单站降水序列的神经网络模型在以平原、河流为主要地形的区域预报效果较好,预报对象为REOF主成分的神经网络模型则在山地和丘陵地形区域预报效果较好.     

基于EOF的江西省秋季降水时空分布特征

基于1980—2020年秋季江西省83个气象观测站逐月降水数据,利用EOF方法分析了该地区秋季降水的时空分布特征。结果表明,江西省秋季降水场主要有4种类型,分别为全区型、北湿(干)南干(湿)型、西湿(干)东干(湿)型、中心湿(干)南北干(湿)型,累计贡献率为86.7%。1980—2020年,全区型和中心湿(干)南北干(湿)型降水呈增加趋势,而北湿(干)南干(湿)型和西湿(干)东干(湿)型降水呈下降趋势。其中全区型降水分布的年份占比75.6%,主要受大尺度大气环流的影响。北湿（干）南干（湿）型降水分布的年份占比17.1%,这是由于赣北地区受地形抬升作用,降水较多,而中南部在背风坡,降水较少,同时秋季赣北处于副热带高压边缘,且受到台风外围的影响,易发生降水,使得南北降水呈反相位变化。     

贵州夏季降水异常的区域特征

利用贵州省19个测站1951～2000年夏季逐月降水资料,计算了降水量的月平均（区域平均）标准化距平。并进行模糊聚类分析、经验正交函数分解（EOF）和小波分析,研究了贵州夏季降水异常的区域特征。结果表明,贵州夏季降水在近50 a中存在5个明显的气候段：20世纪50年代前期为多雨期;50年代中期到60年代前期为少雨期;60年代中后期为多雨期;70～80年代为少雨期;90年代以后进入多雨期;降水呈增多的趋势。全省一致性是贵州夏季降水的最主要特征,同时还存在区域差异。贵州夏季降水异常有5种空间分布型,即：全省旱（涝）型、东旱（涝）西涝（旱）型、南旱（涝）北涝（旱）型、中东旱（涝）西南涝（旱）型和西南旱（涝）其余涝（旱）型。各型降水具有多时间尺度振荡的特点,存在10～12 a、4～5 a、2～3 a的周期。     

MV-WG: a new multi-variable weather generator

This paper presents a new stochastic multi-variable weather generator (MV-WG) and compares its performance with LARS-WG version 4.0. Daily data of 109 meteorological stations from a North American database were used in a twofold comparison of the two generators: (1) the capability of reproducing the mean and variance of annual, seasonal and monthly values, and (2) the capability of reproducing extreme weather events were compared. Both generators did very well on imitating the mean and the variance of the monthly values of the investigated variables, but both showed a more moderate performance as far as the generation of extreme events was concerned. The three-parameter Weibull function, which is first introduced in MV-WG, was found to be a powerful tool to describe not only the distribution of the daily precipitation amounts, but also the distribution of dry and wet spell lengths, as well.     

浙江省短时强降水的时空分布特征

利用2010～2019年浙江省基准气象站和自动气象站逐小时降水的观测资料,对浙江省短时强降水的时空分布特征进行了统计分析,结果表明:1）2010 ～2019年浙江短时强降水累计发生频次为72601站次,随雨强增大呈指数式衰减。2）短时强降水空间分布不均匀,沿海向内陆发生频次减少,出现频次最高的地区位于温州西南部。夏半年随时间推进和影响系统演变,短时强降水的空间分布亦存在差异:5～6月浙西地区短时强降水多发,7月短时强降水全省分散分布无明显的区域集中特征,8～10月则主要在沿海地区多发。3）总体而言短时强降水的日变化峰值出现在17:00（北京时间,下同）,且高强度短时强降水更倾向发生在午后到傍晚时段。夏秋季节短时强降水在午后到傍晚最为多发,峰值出现在17:00至18:00,这与副热带高压强盛,午后到傍晚热力和不稳定条件好,易触发强对流天气有关;春季除午后到傍晚外夜间和凌晨亦为短时强降水多发时段,可能与低空急流多在夜间和早晨发展加强有关。短时强降水的月变化特征呈现类双峰型分布,8月最为多发（26.0%）（主要由台风降水造成）,其次为6月和7月。不同强度的短时强降水月变化特征存在较明显差异。而短时强降水的年际分布不均,2015年之后年际变化幅度增大,其中 2016 年短时强降水发生频次最高达8728站次,2017 年为发生频次最低仅5581站次。     

Multisite statistical downscaling model for daily precipitation combined by multivariate multiple linear regression and stochastic weather generator

This study provides a multi-site hybrid statistical downscaling procedure combining regression-based and stochastic weather generation approaches for multisite simulation of daily precipitation. In the hybrid model, the multivariate multiple linear regression (MMLR) is employed for simultaneous downscaling of deterministic series of daily precipitation occurrence and amount using large-scale reanalysis predictors over nine different observed stations in southern Québec (Canada). The multivariate normal distribution, the first-order Markov chain model, and the probability distribution mapping technique are employed for reproducing temporal variability and spatial dependency on the multisite observations of precipitation series. The regression-based MMLR model explained 16?%?~?22?% of total variance in daily precipitation occurrence series and 13?%?~?25?% of total variance in daily precipitation amount series of the nine observation sites. Moreover, it constantly over-represented the spatial dependency of daily precipitation occurrence and amount. In generating daily precipitation, the hybrid model showed good temporal reproduction ability for number of wet days, cross-site correlation, and probabilities of consecutive wet days, and maximum 3-days precipitation total amount for all observation sites. However, the reproducing ability of the hybrid model for spatio-temporal variations can be improved, i.e. to further increase the explained variance of the observed precipitation series, as for example by using regional-scale predictors in the MMLR model. However, in all downscaling precipitation results, the hybrid model benefits from the stochastic weather generator procedure with respect to the single use of deterministic component in the MMLR model.     

Preserving spatial linear correlations between neighboring stations in simulating daily precipitation using extended Markov models

Most stochastic weather generators have their focus on precipitation because it is the most important variable affecting environmental processes. One of the methods to reproduce the precipitation occurrence time series is to use a Markov process. But, in addition to the simulation of short-term autocorrelations in one station, it is sometimes important to preserve the spatial linear correlations (SLC) between neighboring stations as well. In this research, an extension of one-site Markov models was proposed to preserve the SLC between neighboring stations. Qazvin station was utilized as the reference station and Takestan (TK), Magsal, Nirougah, and Taleghan stations were used as the target stations. The performances of different models were assessed in relation to the simulation of dry and wet spells and short-term dependencies in precipitation time series. The results revealed that in TK station, a Markov model with a first-order spatial model could be selected as the best model, while in the other stations, a model with the order of two or three could be selected. The selected (i.e., best) models were assessed in relation to preserving the SLC between neighboring stations. The results depicted that these models were very capable in preserving the SLC between the reference station and any of the target stations. But, their performances were weaker when the SLC between the other stations were compared. In order to resolve this issue, spatially correlated random numbers were utilized instead of independent random numbers while generating synthetic time series using the Markov models. Although this method slightly reduced the model performances in relation to dry and wet spells and short-term dependencies, the improvements related to the simulation of the SLC between the other stations were substantial.     

北京城市和郊区夏季降雨日变化特征及对排水设计的影响分析

分别选取观象台和密云站作为北京市城区及郊区代表站,应用两站1961—2013年逐分钟雨量观测资料,比较北京城区和郊区夏季降雨量、降雨频次及降雨强度的日变化特征,利用耿贝尔分布拟合的年最大值法推求城区和郊区暴雨强度公式,比较其空间适用性。结果表明,北京地区降雨具有明显的日变化特征：城市和郊区的夜雨比重均大于日雨,降雨量、频次、降雨强度午后至次日清晨为高值区;郊区夏季降雨总量、短历时降雨和降雨雨强均比城区偏大。暴雨强度公式计算结果表明应用城区一站的降雨资料计算得出的公式在全市并不适用,在市政排水设计时应考虑城郊差异,采用不同的标准。     

Simulating Climate over Western North America Using Stochastic Weather Generators

The climatologies of daily precipitation and of maximum and minimum temperatures over western North America are simulated using stochastic weather generators. Two types of generator, differentiated only by their method of modeling precipitation occurrence, are investigated. A second-order Markov model, in which the probability of the occurrence of precipitation is modeled as contingent upon its occurrence on the previous two days, is compared with a spell-length model, in which mass functions of wet- and dry-spell lengths are modeled. Both models are able to reproduce the observed annual and monthly climatology in the region to a high degree of accuracy. However, there is considerable over-dispersion in annual precipitation, resulting primarily from an underestimation in the interannual variability of precipitation intensity. The interannual variability of temperatures is similarly underestimated, and is most severe for minimum temperatures. There is a severe problem in estimating minimum temperature extremes, which can be attributed to the negatively skewed distribution of daily minimum temperatures. Non-normality in the distribution of daily temperatures is shown to be a problem in simulating extreme temperature maxima as well as of minima. It is suggested that the normal distribution used in the generation of daily temperatures in the widely used Richardson (1981) generator, and its derivations, be supplanted by a more appropriate distribution that permits skewness in either direction.     

云台山景区降水资源调查分析

利用景区水库的降水和水文观测历史记录,采用对邻近气象测站测值进行同比气候差订正的方法来弥补少量资料的断缺,以形成完整的历史资料链条。运用与周边气象站对比分析的统计方法,对景区降水、泉瀑汇流水量的年度分布、多年变化趋势进行分析,结果表明:云台山景区降水资源较周边地区显著偏多,降水主要集中在汛雨阶段,多局地性的暴雨、大暴雨日;景区库区来水量分为丰、平、枯年,与汛期降水关系密切,其降水有明显的年代际变化特征,近期正在向少雨年代过渡。     

浑太流域降水极值的统计分布特征

基于浑太流域1966-2006年73个雨量站的日降水资料,建立了逐站年最大日降水量(AnnualMaximum,AM)序列和汛期4-9月日降水量<1.27mm.d-1的最长持续干旱天数(Munger Index,MI)序列,并对其时空分布规律进行了分析。采用广义极值(General Extreme Value,GEV)分布、广义帕雷托(General Pareto,GP)分布、韦布尔(Weibull,WB)分布、约翰逊SB(Jonhson SB,J-SB)分布、Burr分布和对数逻辑(Log-Logistic,L-LG)分布等6种极值分布函数对AM和MI序列进行了逐站分布拟合,结果表明,广泛应用的GEV分布整体拟合程度最好,有50个测站的KS检验统计量Dn<0.09,而未曾推广使用的Burr分布的拟合效果也非常好,有36个测站Dn<0.09。用GEV分布对50年一遇的AM和MI进行了估算,发现流域中心地区极端强降水和极端干旱的程度较高,分别为>208mm.d-1和>47d。     

年径流总量控制率优化研究

年径流总量控制率是海绵城市建设中的重要指标,在实际应用中受到业界高度关注。为了提高年径流总量控制率指标的针对性与规范性,有必要根据城市降水气候特征,并结合地形条件对该指标进行优化。为此,利用重庆市主城区4个国家气象站(北碚、渝北、巴南和沙坪坝站)1981-2018年日降水数据和225个区域自动气象站2013-2018年日降水资料,采用《海绵城市建设技术指南(试行)》(下称《指南》)推荐方法,分别计算4个国家气象站的年径流总量控制率及其对应的设计降雨量;根据主城区降水空间分布和各月变化特征,优化年径流总量控制率指标。结果表明:沙坪坝、北碚、渝北站采用19812018年资料计算的60%-85%年径流总量控制率对应的设计降雨量较《指南》中的偏大,而巴南站的则相反。重庆市主城区年降水量主要集中在4-10月,占全年总量的85.5%,采用每年410月日降水资料对年径流总量控制率对应的设计降雨量推算优化,推算结果较优化前的偏大,如在85%的年径流总量控制率条件,北碚、渝北、巴南、沙坪坝站优化后设计降雨量分别偏大3.2、3.2、2.9和2.8 mm。根据主城区降水空间分布特征,划定了年径流总量控制率指标在重庆主城区的适用范围。     

A new daily weather generator to preserve extremes and low-frequency variability

This paper addresses deficiencies of stochastic Weather Generators (WGs) in terms of reproduction of low-frequency variability and extremes, as well as the unanticipated effects of changes to precipitation occurrence under climate change scenarios on secondary variables. A new weather generator (named IWG) is developed in order to resolve such deficiencies and improve WGs performance. The proposed WG is composed of three major components, including a stochastic rainfall model able to reproduce realistic rainfall series containing extremes and inter-annual monthly variability, a multivariate daily temperature model conditioned to the rainfall occurrence, and a suitable multi-variate monthly generator to fit the low-frequency variability of daily maximum and minimum temperature series. The performance of IWG was tested by comparing statistical characteristics of the simulated and observed weather data, and by comparing statistical characteristics of the simulated runoff outputs by a daily rainfall-runoff model fed by the generated and observed weather data. Furthermore, IWG outputs are compared with those of the well-known LARS-WG weather generator. The tested characteristics are a variety of different daily statistics, low-frequency variability, and distribution of extremes. It is concluded that the performance of the IWG is acceptable, better than LARS-WG in the majority of tests, especially in reproduction of extremes and low-frequency variability of weather and runoff series.