A descriptor for the local dust storm occurrence probability constituted by meteorological factors

WanYuan Li , ShiHua Lü , ZhiBao Dong , ShiGong Wang ,ZhiBao Shen , YuChun Chen , Ye Yu , YinHuan Ao

1. Key Laboratory of Land Surface Process and Climate Change in Cold and Arid Regions, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, Lanzhou, Gansu 730000, China 2. Key Laboratory of Desert and Desertification, Cold and Arid Regions Environmental and Engineering Research Institute,Chinese Academy of Sciences, Lanzhou, Gansu 730000, China 3. Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, Lanzhou University, Lanzhou, Gansu 730000,China

A descriptor for the local dust storm occurrence probability constituted by meteorological factors

WanYuan Li1,2,3*, ShiHua Lü1, ZhiBao Dong2, ShiGong Wang3,ZhiBao Shen1, YuChun Chen1, Ye Yu1, YinHuan Ao1

1. Key Laboratory of Land Surface Process and Climate Change in Cold and Arid Regions, Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences, Lanzhou, Gansu 730000, China 2. Key Laboratory of Desert and Desertification, Cold and Arid Regions Environmental and Engineering Research Institute,Chinese Academy of Sciences, Lanzhou, Gansu 730000, China 3. Key Laboratory for Semi-Arid Climate Change of the Ministry of Education, Lanzhou University, Lanzhou, Gansu 730000,China

Based on daily data sets of 17 meteorological factors during the period of 1954–2005 for 60 gauge stations distributed over Gansu Province of China and the corresponding dust storm records, the dust storm probabilities related to different classes of each factor have been calculated and analyzed. On the basis of statistical analysis, a meteorological descriptor quantifying the daily dust storm occurrence probability for each station, which is referred to as the Dust Storm Occurrence Probability Index(DSOPI), has been effectively established. According to the statistical characteristics of DSOPI for each station, a feasible judging criterion for a dust storm event has been determined, which can greatly contribute to forecasting dust storms and completing the unavailable historic dust storm records. Meanwhile, the average daily dust storm probability related to each factor on the dust storm day for each station has been specially analyzed in detail, finally disclosing that, in general, the more significant one factor’s influence on dust storms, the greater its contribution to them; and each factor’s contribution clearly varies from place to place. Moreover, on average, maximum and mean wind speeds, maximum-speed wind direction, daily sunny hours, evaporation, mean and lowest relative humidity, lowest surface air pressure and vapor pressure contribute to dust storm events in Gansu Province most greatly in order among all the 17 involved factors.

meteorological factor; dust storm; occurrence; significance; probability index; contribution

1. Introduction

The especially dangerous effects exerted on meteorological, environmental and ecological situations of dust storm events have been deeply understood by research institutes, governments and the public. The ultimate goal of research on dust storms is to immediately and accurately forecast and assess dust storm events and the dangers to humanity brought by them (Huang and Gao, 2001; Sunet al.,2001; Qianet al., 2002; Kurosaki and Mikami, 2003;Natsagdorjet al., 2003; Liu Cet al., 2004; Liu CMet al.,2004; Dinget al., 2005; Liuet al., 2005; Shenet al., 2005a;Takemiet al., 2005; Zhouet al., 2005; Hayasakiet al., 2006;Liet al., 2006; Shao and Dong, 2006; Yukariet al., 2006;Yamamotoet al., 2007; Yanget al., 2007; Mao and Gong,2008; Qiuet al., 2008; Sunet al., 2008; Xiaoet al., 2008; Liet al., 2010). Researching the relevant history is an indispensable and very important key to prediction and assessment. Soil, ground surface and meteorology conditions in favor of dust storm occurrence can be determined by comparing the soil, ground surface and meteorological situations before, during and after dust storm events as well as those for historical non-dust-storm events. Generally speaking, sufficient sand and dust sources, strong enough winds and unstable atmospheric stratification are three necessary preconditions for dust storms to take place(Huang and Gao, 2001; Sunet al., 2001; Qianet al., 2002;Kurosaki and Mikami, 2003; Natsagdorjet al., 2003; Liu Cet al., 2004; Liu CMet al., 2004; Dinget al., 2005; Duet al., 2005; Liuet al., 2005; Shenet al., 2005a; Takemiet al., 2005; Zhouet al., 2005; Hayasakiet al., 2006; Liet al.,2006; Shao and Dong, 2006; Yukariet al., 2006; Yamamotoet al., 2007; Yanget al., 2007; Mao and Gong, 2008; Qiuet al., 2008; Sunet al., 2008; Xiaoet al., 2008; Liet al., 2010).As far as certain places are considered, dust and sand source are usually quasi constant, so the meteorological condition becomes a determinative factor of the local dust storm occurrence. It is clear that forecasting whether a dust storm event will happen according to the meteorological situation is an important and necessary method for predicting dust storms (Xuanet al., 2000; Qiuet al., 2001; Shao, 2001; Sunet al., 2001; Kara and Takeuchi, 2004; Qianet al., 2004;Wanget al., 2004; Chunet al., 2005; Dinget al., 2005;Jimmyet al., 2005; Miyata, 2005; Wanget al., 2005; Yanget al., 2005; Zhaoet al., 2005; Zhao and Sun, 2005; Shao and Dong, 2006; Zhanget al., 2007).

Owing to global warming and intensified drought and desertification, dust storm events have gained considerable attention of many scientists, who focus their work mostly upon the spatial and temporal variation characteristics of dust storm frequency and their geographical and climatic causes (Huang and Gao, 2001; Qiuet al., 2001; Sunet al.,2001; Qianet al., 2002; Kurosaki and Mikami, 2003;Natsagdorjet al., 2003; Liu Cet al., 2004; Liu CMet al.,2004; Wanget al., 2004; Zou and Zhai, 2004; Chunet al.,2005; Dinget al., 2005; Duet al., 2005; Kimet al., 2005;Wanget al., 2005; Wang and Ta, 2005; Shao and Dong,2006; Xu, 2006; Yukariet al., 2006; Wanget al., 2007;Zhanget al., 2007; Xiaoet al., 2008; Yanet al., 2009),along with numerical simulations of almost all aspects of dust storm events, such as their geographical and meteorological conditions, the entrainment, transportation and deposition of dust and sand and their environmental, meteorological and climatic effects (Shao, 2001; Fanget al.,2004; Gonget al., 2004; Kara and Takeuchi, 2004; Kimet al., 2005; Shen YBet al., 2005; Shen ZBet al., 2005; Shao and Dong, 2006; Xu, 2006; Kim and Kai, 2007; Zhanget al., 2007; Yanet al., 2009). However, previous results are always qualitative and difficult to directly apply to practical prediction, with a lot of information useful for forecasting hidden in the large quantity of historical materials but not accessed.

Therefore, it is necessary to establish a bridge to connect directly and quantitatively meteorological factors characterizing meteorological condition with the probability of a dust storm event in order to judge directly by the meteorological factors whether a dust storm event will happen under a certain meteorological condition. In favor of building such a bridge, a meteorological descriptor wholly constituted by meteorological factors will be established in this paper to characterize the real dust storm occurrence probability on a certain day, which is later referred to as the Dust Storm Occurrence Probability Index(DSOPI). The following properties should be guaranteed of the descriptor: (1) All typical factors characterizing the meteorological condition should be accounted for. (2) The influence and significance of each important factor upon the dust storm occurrence should be considered. (3) The dust storm occurrence probability under a certain meteorological condition should be validly quantified. (4) It must validly express the quantitative relationship between the factors and the dust storm occurrence probability. It is clear that the descriptor can be used to assess the dust storm occurrence probability directly all by meteorological factors, being meaningful theoretically and practically.

In China, dust storm events mostly happen in spring(from March to May) and in arid and semiarid areas of the Northwest. Gansu Province is an example that is prone to attacks of dust storms, especially in its northwest and middle regions (Figure 1a) (Qiuet al., 2001; Sunet al., 2001;Qianet al., 2002; Kurosaki and Mikami, 2003; Gonget al.,2004; Liu CMet al., 2004; Qianet al., 2004; Wanget al.,2004; Zou and Zhai, 2004; Kimet al., 2005; Wanget al.,2005; Shao and Dong, 2006; Wanget al., 2007; Yanget al.,2007; Zhanget al., 2007; Sunet al., 2008). There exist relatively complete data of meteorological factors and dust storm records in this province. As a convenient representative example, this paper has selected the existed spring dust storms in Gansu Province for study in order to establish an integrated meteorological descriptor based on abundant meteorological factor and dust storm historical records. This can be used to quantitatively characterize and evaluate the dust storm occurrence probability on a certain day and make its daily value well connected to reality regarding if a dust storm event takes place on the same day.On the basis of the above steps, a meteorological judging criterion can be created to effectively predict whether a dust storm event will happen at a certain place on a certain day. Thus, a new procedure for forecasting dust storms will be found and the historical blank records can be truly filled in by this criterion.

As known to all, there are many factors characterizing the meteorological condition, which impact dust storms in different ways, such as wind speed (Kurosaki and Mikami,2003; Fanget al., 2004; Liu CMet al., 2004; Hayasakiet al., 2006; Liet al., 2006; Yukariet al., 2006; Kim and Kai,2007; Qiuet al., 2008; Xiaoet al., 2008), air pressure and stability of the atmospheric stratified structure, all deter-mining entrainment and transportation of dust and sand(Kara and Takeuchi, 2004; Liu CMet al., 2004; Dinget al.,2005; Yamamotoet al., 2007), soil moisture, evaporation,precipitation, air humidity, vapor pressure, and lasting periods of drought and precipitation, all determining the surface threshold for entrainment of sand and dust (Xuanet al., 2000; Huang and Gao, 2001; Shao, 2001; Gonget al.,2004; Liu CMet al., 2004; Wanget al., 2004; Zou and Zhai, 2004; Duet al., 2005; Xuet al., 2005; Yanget al.,2005; Liet al., 2006; Shao and Dong, 2006; Xu, 2006; Xuet al., 2006; Kim and Kai, 2007; Zhanget al., 2007; Guanet al., 2008; Zhanget al., 2008; Yanet al., 2009). All factors mentioned above are directly related to dust storm events. Cloud fraction, sunny hours, air temperature and ground temperature can indirectly influence dust storm occurrence by creating or destroying its thermodynamic conditions (Shao, 2001; Qianet al., 2002; Natsagdorjet al.,2003; Masudaet al., 2005; Sassen, 2005; Wang and Ta,2005; Yamamotoet al., 2005; Hayasakiet al., 2006; Shao and Dong, 2006; Yukariet al., 2006; Wanget al., 2007;Yanget al., 2007). Moreover, the dust storm event can directly cause variation in some meteorological factors such as an increase in late precipitation and a reduction in sunny hours (Shao, 2001; Qianet al., 2002; Natsagdorjet al., 2003; Masudaet al., 2005; Sassen, 2005; Wang and Ta,2005; Yamamotoet al., 2005; Hayasakiet al., 2006; Shao and Dong, 2006; Yukariet al., 2006; Wanget al., 2007;Yanget al., 2007). In order to fully characterize the meteorological conditions for dust storm events entirely and uniformly, all the available meteorological factors are selected as long as they have sufficient statistical record lengths, not accounting for their initiative or passive relations to dust storms because there exist practical interactions between any factor and dust storms, and thus unclear distinction between the initiative and the passive.

2. Data and method

2.1. Data introduction

The 60 gauge stations in Gansu Province (Figure 1b)and their spring daily data or records during 1954–2005 for 17 meteorological factors (including daily mean and maximum wind speeds (hereafter WS and FWS, respectively), direction of maximum wind (WD), evaporation (E),daily mean and minimum relative humidity (RH and SRH,respectively), vapor pressure (VP), number of sunny hours(SH), daily mean, lowest and highest surface air pressures(P, LP and HP, respectively), mean, highest and lowest surface air temperatures (T, HT and LT, respectively), precipitation of 20:00–08:00, 08:00–20:00 and 20:00–20:00(R2008, R0820 and R2020, respectively)), and dust storm events have been selected from the Chinese Meteorology Science Data Web.

Figure 1 (a) Spring dust storm frequency distribution over Gansu, China (The shaded area is the northeast corner of the Qinghai-Xizang Plateau); (b) the selected 60 gauge stations’ positions (the two lines divide Gansu Province into three regions: northwestern Gansu, NWP; southeastern Gansu, SEP; middle part of Gansu, MP).

The selected data have the following statistical characteristics: (1) The starting and ending years of data for different stations are not always the same. The earliest beginning year is 1954 and the latest ending year is 2005. The temporally shortest data length is 27 years and the longest is 52 years. (2) No daily blank records exist in the dust storm data for each station from its beginning year (day) to its ending year (day), but more or less blank records exist in the meteorological data for nearly all factors, and the spatial (from station to station) and temporal (from day to day) distributions of blank records are not uniform for all of the factors. For example, there are no records of seven factors on all spring days in 1967 at the station of Dunhuang, and thus Dunhuang and 1967 are, respectively, the station and year with most unavailable factors. (3) For the following, a spring day with no blank dust storm record for a certain station is called one available day for that station,and any record for a meteorological factor on a certain day at a certain station is considered as one sample. In order to keep the validity of the following conclusions in the statistical sense, data subsets are selected only when the total number of samples for each station and factor is guaranteed no less than some threshold value: it is 1,000 for WD,600 for precipitation, and 1,500 for any other factors. Thus,a few factors will be kept out for some stations and the abandoned factors are not common for all stations. There are ten stations with the most of four factors abandoned,meaning they have 13 available factors, 17 stations with two factors abandoned, 8 stations with one factor abandoned, and the other 26 stations all have the 17 factors available. (4) Due to the aforementioned abandonment,there remain four factors having no less than 60 available stations (Table 1). (5) In spite of the blank records and some occasionally abandoned factors, among all the available days and all the stations, only 54% of the stations by days have no less than 15 available factors, and 99.92%have no less than ten available factors, only a few stations by days have three to nine available factors. (6) On any spring days during 1954–2005, the total number of stations with no blank dust storm records can be determined as the available station number of that day. Of all the 4,784 spring days during the 52 years, 21.15% have 60 available stations, 42.31% have 52 to 59, other 21.15% have 42 to 49,and there are totally 86.54% with no less than 35 available stations. What is more, the days with 22 to 28 available stations occupy 5.77%, and those with 15 to 19 occupy 7.69%. Critically speaking, blank data and their non-uniform distributions can partly influence the comparability of some results between different stations or different years, but the conclusions in this paper will not be distorted by this disadvantage.

2.2. Method of establishing the Dust Storm Occurrence Probability Index (DSOPI)

By these following steps we can create the Dust Storm Occurrence Probability Index (DSOPI):

(1) Divide all the daily samples during the selected period for each station and factor into seven classes according to the low (or small) to high (or large) sort order, trying the best to make each class have the same number of samples, thus the threshold (boundary) values between one class and another can be found. It can be imagined that the boundaries and the mean values of the same classes for a certain factor are both not equal and even quite distinct among different stations because the geographical and climatic situations are quite different. Especially for R2008,R0820 and R2020, all the samples for each station are divided into two classes, that is, zero precipitation and non-zero precipitation, because the samples with precipitation are always by far less than the samples with zero precipitation. Wind direction cannot be considered in terms of magnitude, and it will be classified by its available types for each station. In order to guarantee in the statistical sense the reality and the comparability of the relevant results, for a certain station the wind direction type with samples less than 100 will be abandoned, and the remaining types are the ultimate classes for it. Hereafter, the ultimate number of classes of wind direction for a certain station is referred aswi, whereirepresents the code of that station.

(2) Based on historical records, calculate the dust storm occurrence probabilities corresponding to different classes of each factor respectively:

wherekandkkrespectively denote the order number and the total number of classes of thenth factorXn. IfXnis precipitation,kk=2; and if it is wind direction,kk=wi.Mi,k(Xn) andSi,k(Xn) respectively represent the total sample number and the number of the samples with a dust storm event happening simultaneously corresponding to thekth class of thenth factorXnfor theith station, and thuspi,k(Xn) is just the dust storm occurrence probability of theith station whenXnlies in thekth class. The smallest and largest sample numbers of all the classes of each selected meteorological factor for all available stations are also listed in Table 1, which indicates that the sample number of any class of any certain factor for each station is always no less than 100, that is:

whereizandixrespectively represent the numbers of available stations for the factorXnand of the available factors for theith station. The large quantity of samples for each factor and station will completely keep the reality and validity ofpi,k(Xn) in the statistical sense.

(3) So, the dust storm occurrence probability associated with a certain factor at a certain station on a certain day can be obtained from the class that this factor belongs to at that station on that day. If the class code ismfor the factorXnat theith station on thejth day, the dust storm occurrence probability associated withXnatith station on thejth day is justPi,j(Xn)=pi,m(Xn) (note the difference between the capital and lowercase letters).

(4) The impacting significance of one factor upon the dust storm occurrence is different from that of another. The impacting significance of a certain factor can be determined as the difference of the largest and smallest dust storm occurrence probabilities associated with that factor,that is:

It is apparent that the impacting significance determined as above has sufficient comparability either between one factor and another or between one station and another.

?

(5) Taking the impacting significance of each factor as its weight, the weighted average of the dust storm occurrence probabilities respectively associated with the available factors at a certain station on a certain day can be calculated and the result is defined as the Dust Storm Occurrence Probability Index (DSOPI) at that station on that day:

whereL(i,j) represents the number of available factors at theith station on thejth day,l(i,j,Xn) is the code of the class the factorXnbelongs to at theith station on thejth day, andPIi,jis just the DSOPI at theith station on thejth day.

2.3. Method of determining the meteorological judging criterion for dust storm occurrence

Based on DSOPI, an idiographic meteorological criterion for judging whether there is a dust storm event at a certain station on some day can be developed by the following steps:

(1) Seek the averages of DSOPI on the dust storm and non-dust-storm days respectively, at each station, denoted asPIA1iandPIA0i:whereJ1(i) andJ0(i) represent the total number of spring dust storm days and non-dust-storm days during the selected period of 1954–2005 at theith station respectively.

(2) Seek the standard deviations of DSOPI for dust storm and non-dust-storm days respectively at each station,denoted asSDPI1iandSDPI0i:

(3) Finally, the criterion can be given as follows:

wherenfi,j=1 indicates there is a dust storm event at theith station on thejth day;nfi,j=0 means no dust storm event happens; andnfi,j=1/2 means there is no meaningful signs for either a dust storm or non-dust-storm event.

The accuracy of the criterion can be evaluated by using it to judge if there is a dust storm event at each selected available station on each selected available day firstly, and then compare the results with the corresponding historical records. It is necessary to further disclose the accuracy of the criterion when it is directly applied on the dust storm events in the years after 2005, but the meteorological data and dust storm records after 2005 had not been obtained till composing this paper. However, the large quantity of samples used when developing the criterion should be able to make it effectively to expand over the several years surpassing the sampling period, and it can be reasonably believed that the criterion can be effectively used to judge the dust storm events in recent years and in the near future.

3. Validity and applications of DSOPI

3.1. Average difference of DSOPI between the dust storm

and non-dust-storm days

The average difference of DSOPI between dust storm and non-dust-storm days can be calculated as follows:

It can be reasonably believed that the dust storm occurrence probability on the dust storm day should be generally larger than that on the non-dust storm day. So, only when the average DSOPI on the dust storm day is larger than that on the non-dust-storm day, the meteorological index DSOPI is sufficiently rational, and the larger positiveDPIAican more determinatively indicate thatPIi,jcan reasonably and efficiently characterize the real dust storm occurrence probability at theith station on thejth day.

Figure 2 displays the distribution ofDPIAiover the whole province of Gansu.DPIAiis uniquely positive at each station in the province and its smallest value of 0.22% lies at the station of Lintao (52,986; 35.37°N, 103.87°E), whose spring dust storm frequency averages 0.3 through 1954–2005, being the second fewest of all the selected 60 stations. The largestDPIAiof 11.36% lies at the station of Minqin (52,681; 38.63°N, 103.08°E), whose average spring dust storm frequency of 12.1 is the highest of the whole province. The similarity coefficient between the two geographical distributions over the whole province ofDPIAiandADSFi(representing the spring dust storm frequency averaged through 1954–2005 at theith station) reaches a high value of 0.97, namely, the largerADSFi, the largerDPIAi. Of all 60 stations the average ofDPIAiis 1.7%, and its standard deviation is 2.04%. Such large differences inDPIAibetween different stations are caused partly by the large differences inADSFi, which may indicate that the same meteorological condition in favor of dust storm events can readily lead to the real dust storm occurrence at those stations with high dust storm frequencies.

The average difference of the dust storm occurrence probabilities associated with each factor on dust storm and non-dust-storm days is also uniformly positive at each available station (Table 1), and its average over the whole province reaches 10.32% for FWS and 4.6% for WS, being the largest two of the selected 17 factors. All those facts indicate that the 17 factors have clear influences upon the dust storm occurrence, and it is reasonable and feasible to use the dust storm occurrence probabilities associated with them to characterize the real dust storm occurrence possibility.

Figure 2 The difference (%) of mean dust storm occurrence probability descriptor between spring dust storm and non-dust-storm days for each gauge station in Gansu Province of China. The shaded area is the Qinghai-Xizang Plateau’s northeast corner.

3.2. The accuracy of the judging criterion for dust storm occurrence

With the aforementioned judging criterion for the dust storm occurrence, we can obtain the judged results of whether a dust storm event would happen at each selected station on each selected spring day. In the following, the ratio of the number of days with the judged result matching with the historical record to the total number of available days during 1954–2005 at a certain station is named as the judging accuracy rate of the same station and denoted asFri. Also, the ratio of the number of stations at which the judged result matches with the real fact to the total number of available stations on a certain spring day during the period from March 1, 1954 to May 31, 2005 is named as the judging accuracy area of that day and denoted asFrj.

Figure 3a gives the geographical distribution ofFriover the whole province of Gansu. In summary, for most areas in the northwestern part, the northern areas of the middle part, and some areas in the southern part of Gansu,the judging accuracy rateFriis higher than that for the northeastern boundary areas of the Tibetan Plateau.Frireaches the smallest value of 62% at the station of Zhengning (53,935; 35.5°N, 108.35°E) in the southern part,and obtained the largest value of 87% at the station of Gaolan (52,884; 36.35°N, 103.93°E). The smallest number of available spring days (or samples) is 2,484 of all the 60 stations, and the largest is 4,784 and the average is 3,924.The smallest number of days with the true judged result is 1,945 of all the 60 stations, and the largest and the average are respectively 4,072 and 3,181. The average accuracy rate of all the stations is 81% and the standard deviation ofFriis only 4%. Among the 60 stations there are 43 with an accuracy rate beyond 80% and 58 beyond 73%, showing that the judging criterion developed above can be well applied to the dust storm events at each station in the province of Gansu.

Of all the 4,784 spring days, the smallest, largest and average number of available stations are 15, 60 and 49,respectively; the smallest, largest and average number of stations at which the judged result matches with the real fact are 4, 60 and 40, respectively; the smallest, largest and average value ofFrjare 23%, 100% and 81%, with its standard deviation being 14%. The days withFrjmore than 70% occupy the percentage of 80%, those withFrjmore than 80% occupy 62%, those withFrjmore than 90% occupy a little more than 34%, and those withFrjequal to 100% occupy 5%. All above listed facts tell us that the criterion can be well used to judge the expanding area affected by the dust storm event on a certain day. As a reference, Figure 3b gives theFrjvalues on all the 4,784 spring days.

Figure 3 (a) Efficiency of the judging criterion for spring daily dust storm occurrence based upon dust storm occurrence probability descriptor for each gauge station in Gansu Province of China. The shaded area is the Qinghai-Xizang Plateau’s northeast corner. (b) The fraction of the whole province with correct judgment by this criterion for each spring day during the period from March 1, 1954 to May 31, 2005.

3.3. Capability of the DSOPI to characterize the spring dust storm frequency

If the indexPIi,jcan reasonably characterize the real dust storm occurrence probability at theith station on thejth day,the accumulation ofPIi,jon all spring days in a certain year(coded asN) should be able to characterize the spring dust storm frequency:

whereJJ(i,N) represents the number of available spring days in theNth year at theith station. Only when the positive correlation coefficient between the time series ofAPIi,NandDSFi,N(representing the real spring dust storm frequency at theith station in theNth year) is sufficiently large, it can be said thatPIi,jcan reasonably and even truly characterize the real dust storm occurrence probability.However, the non-dust-storm days are always by far more than the dust storm days at any station in any spring, which causesAPIi,Nto contain a constituent made up of the non-dust-storm days prior or by far prior to another constituent made up of the dust storm days, and unable to characterize the dust storm frequency and its year-to-year variation. So,APIi,Nshould be corrected as the accumulation ofPIi,jon all dust storm days:

whereJJ1(N) is referring to the total number of available spring dust storm days in theNth year.

Only the sufficiently high positive similarity coefficient between the two geographical distributions ofAPI1i,NandDSFi,Ncan clearly tell that,API1i,Nhas the ability to effectively characterize the spatial variation of the spring dust storm frequency in theNth year, andPIi,jcan satisfyingly describe the real dust storm occurrence probability.

Figures 4a and 4b respectively give the geographical distributions of the correlation coefficients ofAPIi,NversusDSFi,NandAPI1i,NversusDSFi,N. Generally speaking,APIi,Nis poorly correlated withDSFi,N(Figure 4a); among the 60 stations, there are only seven with the correlation coefficient above 0.5, 14 above 0.4, 31 below 0.2, and 12 with weak negative correlation; 5 stations with the coefficient above 0.5 lie in the southeastern part of Gansu.However, the correlation coefficient between the temporal series ofAPI1i,NandDSFi,Nis uniformly very large at each of the 60 stations (Figure 4b); the smallest, largest and average coefficient of the whole province are respectively 0.7, 1 and 0.96, with the standard deviation being very small at 0.06; all indicating thatAPI1i,Ncan aptly characterize the year-to-year variation of the dust storm frequency at each station.

Figure 5 gives the year-to-year variations of the similarity coefficients of the geographical distributions of bothAPIi,NandAPI1i,Nwith the spatial distribution ofDSFi,Nduring the period of 1954–2005. It is apparent that the similarity coefficients ofAPIi,NandAPI1i,NwithDSFi,Nare both remaining above 0.62, averaging 0.86 and 0.89, respectively;among the 52 years there are 20 with the similarity coeffi-cient ofAPIi,NversusDSFi,Nabove 0.9; 22 years with the coefficient ofAPI1i,NversusDSFi,Nabove 0.9; and in the other years it is uniformly above 0.8 except for two years; all indicating thatAPIi,NandAPI1i,Nboth can aptly characterize the spatial variation of the spring dust storm frequency in theNth year, and the latter is prior to the former.

Figure 4 Correlation coefficient (%) distribution between spring dust storm frequency and accumulated dust storm occurrence probability descriptor for all spring days (a) and all spring dust storm days (b) over Gansu Province of China.The shaded area is the Qinghai-Xizang Plateau’s northeast corner.

Figure 5 Yearly variation of the similarity coefficients between the spatial distributions of spring dust storm frequency (DSF)and accumulated dust storm occurrence probability descriptors for all spring days (API) and all spring dust storm days (API1) over Gansu Province of China

3.4. Comparison of the qualified contributions of different factors to dust storm occurrence

The temporally varying range of one factor and its impacting significance upon the dust storm occurrence at a certain station are both different from those of another factor,so the contributions to the dust storm occurrence at the same station are also quite different from one factor to another.The contribution of one factor to the dust storm occurrence at a certain station can be represented by the average dust storm occurrence probability associated with that factor on the dust storm day at the same station:

The largerCi(Xn) means the larger contribution ofXnto the dust storm occurrence at theith station. The smallest,largest and average values ofCi(Xn) of all the 60 stations are all listed in Table 2 with the standard deviation ofCi(Xn). It is shown that for each factor, the largest value ofCi(Xn) is uniquely larger than the smallest by two orders of magnitude,and the standard deviation is equivalent to the average ofCi(Xn), both indicating that the contribution of each factor has a large varying range within the whole province.

Averagely speaking, the first nine factors among all the 17 factors, which greatly contribute to the dust storm occurrence in Gansu are FWS, WS, WD, SH, E, RH, SRH,LP and VP in turn, and their contributions are respectively 12.54%, 6.86%, 4.43%, 3.92%, 3.41%, 3.2%, 3.18%,3.08% and 3.01%. It is interesting that the above nine factors also occupy the first place in their impacting significance upon the dust storm occurrence. The sort orders of factors respectively according to the averages ofCi(Xn) andFi(Xn) over the whole province of Gansu are listed in Table 3, and they are fundamentally consistent with each other,indicating that, the larger impacting significance of one factor to the dust storm occurrence, the larger factor’s contribution to it.

The averages ofCi(Xn) respectively over the northwestern, middle, and southern areas of Gansu are listed in Table 4. Except for the factor of WD, both the average ofCi(Xn)and the averaged mean DSOPI on the dust storm day reach the maximum in the northwestern area and the minimum in the southern area, which is consistent with or explainable by the average spatial distribution of the spring dust storm frequency. The relative standard deviation ofCi(Xn) (denoted as D/A in Table 4) is always reaching the minimum also in the northwestern area for each factor, indicating that, their contributions are all uniformly large at all stations there, but the clear differences from one station to another exist in their contributions for the other two areas.

Averagely speaking, the first eight factors who contribute most greatly to the dust storm occurrence for the northwestern area of Gansu are in turn FWS, WS, SH, E, RH, HT,VP and SRH, but for the middle area of Gansu they are FWS,WS, SH, WD, E, HT, LT and T in turn, and for the southern area they are FWS, WS, WD, SRH, RH, E, VP and SH (Table 4). For each area, the sort order of factors according to their average contributions is also generally consistent with that according to their average impacting significances, further indicating that the latter are determinative to the former.What is more, the sort order of factors according to their contributions is partly different from one area to another; it can be believed that the similarity in the sort orders of the three areas is caused by the inherent properties and consolidated roles of the factors in the earth-atmosphere systems, and the discrepancy can be attributed to the quite different geographical,ecological and climatic situations of different areas.

3.5. Possible significance of other impacting factors upon dust storm occurrence

As mentioned above, there are many other factors impacting on the dust storm occurrence besides the involved 17 factors. The possible composite significance of other impacting factors upon the dust storm occurrence can be evaluated in the following. Firstly, the parameterPI0i,jis introduced as below:

which means the non-dust-storm occurrence probability associated with the available factors (among the involved 17 factors) at theith station on thejth day, and can be temporally regarded as the composite capability of the factors resisting the dust storm occurrence. The average ofPI0i,jfor the dust storm days is denoted asPI0A1i, and the sum ofPI0A1iandPIA1ican be approximately considered as the total effect on the dust storm occurrence by the above factors. So, the total effect of other factors independent of the involved ones can be approximately parameterized as follows:

Figure 6 gives the geographical distribution ofPI0iover the province of Gansu.PI0ireaches the smallest value of 0.83% at the station of Longxi (56,092; 35°N, 104.65°E),and the largest value of 29.96% at the station of Wuwei(52,679; 37.92°N, 101.67°E), and its average and standard deviations over the whole province are, respectively, 12.98%and 7.04%. Among all the 60 stations there are 33 withPI0i<15%, indicating thatPI0iis very small all over the province.In addition, the similarity coefficients between the geographical distributions of each involved factor and the dust storm frequency in spring averaged through 1954–2005 are listed in Table 5, together with the correlation coefficients between the temporal series of each factor and dust storm frequency averaged over Gansu. There are eight factors with a similarity coefficient of no less than 0.7 and four other factors above 0.5, and there are nine factors with an average correlation coefficient above 0.2, both indicating that the selected 17 factors are significant in influencing dust storm occurrence in Gansu.

4. Conclusion

On the basis of the dust storm occurrence probabilities corresponding to different classes of each of the 17 involved meteorological factors at 60 gauge stations distributed over Gansu Province of China, a composite meteorological index(DSOPI) has been developed to describe the actual daily dust storm occurrence probability at each station. According to the statistical characteristics of DSOPI at each station, a feasible judging criterion for dust storm occurrence has been established, which can greatly contribute to forecasting dust storm events and complete the historical blanks in the dust storm records.

?

?

Figure 6 Other factors’ contribution to the spring daily dust storm occurrence over Gansu Province of China.The shaded area is the Qinghai-Xizang Plateau’s northeast corner.

The average dust storm occurrence probability related to each factor on dust storm days has been specially analyzed for all the stations, indicating that, generally speaking, the more intense one factor’s impact on the dust storm occurrence, the greater its contribution to it; each factor’s contribution greatly varies from place to place. By further analysis,the following points have been disclosed: (1) It is more appropriate to use multi factors to compositely describe the meteorological condition for dust storm occurrence than to use any single factor. (2) The selected 17 factors seem to be more determinative for dust storm occurrence at those places with a higher dust storm frequency. (3) In terms of the spatial distribution and temporal variation, the accumulated DSOPI on all spring dust storm days can effectively characterize the real spring dust storm frequency. (4) On average, over the whole Gansu Province, the first 9 factors that contribute to dust storm occurrence most greatly are the daily maximum and mean wind speeds (FWS and WS), direction of the daily maximum wind (WD), daily sunshine duration (SD), daily evaporation (E), daily lowest surface air pressure (LP), daily mean and smallest relative humidity (RH and SRH), and daily mean vapor pressure (VP) in turn. (5) There are still other possible factors independent of the available ones considerably impacting upon dust storm occurrence in Gansu Province.(6) On the one hand, the sort order of the factors in their relative contribution to dust storm occurrence is different from one part to another; on the other hand, it is similar to another,which may be caused by the composite effects of the inherent properties and consolidated roles of the factors in the earth-atmosphere system and the differences in the geographical, ecological and climatic environments among different areas or places.

Acknowledgment:

This work is jointly funded by National Program on Key Basic Research Project (973 Program, Grant No.2009CB421402), the open foundation from Key Laboratory for Semi-Arid Climate Change of the Ministry of Education,Lanzhou University, and National Natural Science Foundation of China (Grant No. 40975007). We are very grateful to all the colleagues and anonymous editors who have given us very useful suggestions on this manuscript.

Chun Y, Nam JC, Chung HS, 2005. Monitoring and prediction of Asian dust(Hwangsa) in Korea. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January,O2-2.

Ding R, Li J, Wang S, 2005. Decadal change of the spring dust storm in northwest China and the associated atmospheric circulation. Geophys.Res. Lett., 32: L02808. DOI: 10.1029/2004GL021561.

Du MY, Yonemura S, Shen ZB, Shen YB, Wang WF, Yamada Y, Maki T,Kawashima S, Inoue S, 2005. Relationship between ground surface conditions and Aeolian dust outbreak in Dunhuang, China. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact.Nagasaki, Japan, 26–28 January, O1-2.

Fang XM, Han YX, Ma JH, 2004. Dust storm events and loess accumulation on the Tibetan Plateau: A case study of dust event on 4 March 2003 in Lhasa. Chin. Sci. Bull., 49(9): 953–960.

Gong SL, Zhang XY, Zhao TL, 2004. Sensitivity of Asian dust storm to natural and anthropogenic factors. Geophys. Res. Lett., 31: L07210. DOI:10.1029/2004GL019502.

Guan XD, Guo N, Huang JP, 2008. Applicability analysis of VCI to monitoring Northwest China drought. Plateau Meteorology, 27(5):1046–1053.

Hayasaki M, Sugata S, Tanaka HL, 2006. Interannual variation of cold frontal activity in spring in Mongolia. JMSJ, 84: 463–475.

Huang F, Gao Q, 2001. Climate controls upon dust storm occurrence in Maowusu Desert, Inner Mongolia, North China. Journal of Environmental Sciences, 13(1): 14–21.

Jimmy E, Quilang P, Maki T, Du MY, 2005. Monitoring and evaluation of dust storm outbreak at Dunhuang, China. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp.1–14.

Kara A, Takeuchi K, 2004. Yellow dust: an overview of research and felt needs. Journal of Arid Environments, 59: 167–187.

Kim HS, Kai K, 2007. Recent dust outbreaks in the Taklimakan Desert and their relation to surface wind and land surface condition. SOLA, 3:69–72.

Kim YJ, Lee KH, Von W, Huene H, 2005. Satellite monitoring of atmospheric aerosol in fine spatial resolution over Northeast Asia. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp.1–21.

Kurosaki Y, Mikami M, 2003. Recent frequent dust events and their relation to surface wind in East Asia. Geophys. Res. Lett., 30(14): 1736. DOI:10.1029/2003GL017261.

Li N, Gu W, Du ZX, Li ZJ, Song PG, 2006. Observation on soil water content and wind speed in Erlianhot, a dust-source area in northern China.Atmospheric Environment, 40: 5298–5303.

Li WY, Lv SH, Dong ZB, Yu Y, Shen ZB, Chen YC, 2010. The fitted spring dust storm frequency by multi meteorological factors for north and northwest China. Plateau Meteorology, 29(5): 1302–1313.

Liu CM, Qian ZA, Wu MJ, 2004. A composite study of the synoptic differences between major and minor dust storm springs over China-Mongolia regions. TAO, 15(5): 999–1018.

Liu MZ, Wei WS, Sadayo Y, 2005. Effects of climate change on the condition related to the occurrence of dust storm events in Xinjiang. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp.1–26.

Liu X, Yin ZY, Zhang X, Yang X, 2004. Analyses of the spring dust storm frequency of northern China in relation to antecedent and concurrent wind, precipitation, vegetation, and soil moisture conditions. J. Geophys.Res., 109: D16210. DOI: 10.1029/2004JD004615.

Mao R, Gong DY, 2008. Correlation analysis of interannual variability between spring dust storm frequency and atmospheric circulation factors.Plateau Meteorology, 27(5): 1023–1030.

Masuda K, Mano Y, Ishimoto H, 2005. Mineral dust optical thickness derived from the GMS-5 visible data. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan,26–28 January, O2-4.

Miyata K, 2005. Relation of Aeolian dust to upper-air winds observed with wind profiler. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp.2–30.

Natsagdorj L, Jugder D, Chung YS, 2003. Analysis of dust storm events observed in Mongolia during 1937–1999. Atmospheric Environment, 37:1401–1411.

Qian WH, Quan LS, Shi SY, 2002. Variations of the dust storm in China and its climatic control. J. Climate, 15(10): 1216–1229.

Qian WH, Tang X, Quan LS, 2004. Regional characteristics of dust storm events in China. Atmospheric Environment, 38: 4895–4907.

Qiu XF, Zeng Y, Miao QL, 2001. Sand-dust storm events in China, temporal-spatial distribution and tracks of source lands. Journal of Geographical Sciences, 11(3): 253–260.

Qiu YJ, Niu SJ, Zhou XY, 2008. The relationship between dust event frequency and meteorological factors. Plateau Meteorology, 27(3):637–643.

Sassen K, 2005. Polarization Lidar studies of indirect desert dust effects on clouds. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, O2-7.

Shao Y, 2001. Physics and Modeling of Wind Erosion. Kluwer Academic Publishers, pp.145–222.

Shao Y, Dong CH, 2006. A review on East Asian dust storm climate, modeling and monitoring. Global and Planetary Change, 52: 1–22.

Shen YB, Shen ZB, Du MY, Wang WF, 2005. Modeling of dust emission in Northwest China. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp.2–25.

Shen ZB, Shen YB, Du MY, Wang WF, 2005. Numerical simulation and forecasting of dust emission in Dunhuang Area. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, O1-3.

Sun GW, Tang X, Li JP, 2008. Relationship between northern edge of the East Asian summer monsoon and sand-dust storm. Plateau Meteorology,27(5): 1088–1093.

Sun J, Zhang M, Liu T, 2001. Spatial and temporal characteristics of dust storm events in China and its surrounding regions, 1960–1999: Relations to source area and climate. J. Geophys. Res., 106(D10): 10325–10333.

Takemi T, Yasui M, Zhou JX, Liu LC, Itabe T, Mizutani K, Aoki T, 2005.The role of dry convection on the vertical transport of mineral dust over arid regions. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp. 2–27.

Wang SG, Wang JY, Zhou ZJ, 2005. Regional characteristics of three kinds of dust storm events in China. Atmospheric Environment, 39: 509–520.

Wang T, Ta WQ, 2005. Recent dust events and their relation to wind and precipitation in the strong dust source of Hexi Corridor in West China.Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, O5-8.

Wang X, Dong Z, Zhang J, 2004. Modern dust storm events in China: An overview. Journal of Arid Environment, 58: 559–574.

Wang YQ, Zhang XY, Gong SL, 2007. Surface observation of sand and dust storm in East Asia and its application in CUACE/dust. Atmos. Chem.Phys. Discuss., 7: 9115–9138.

Xiao FJ, Zhou CP, Liao YM, 2008. Dust storm events evolution in Taklimakan Desert and its correlation with climatic parameters. J. Geogr. Sci.,18: 415–424.

Xu J, 2006. Sand-dust storm events in and around the Ordos Plateau of China as influenced by land use change and desertification. Catena, 65:279–284.

Xu X, Li B, Zhou Z, 2005. Demonstration of afforestation for control of sand drift along base highway oil field in hinterland of Taklimakan Desert—A study on irrigation technology of saline water at sandy strata. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp. 1–8.

Xu XK, Levy JK, Lin ZH, 2006. An investigation of sand-dust storm events and land surface characteristics in China using NOAA NDVI data. Global and Planetary Change, 52: 182–196.

Xuan J, Liu GL, Du K, 2000. Dust emission inventory in northern China.Atmospheric Environment, 34: 4565–4570.

Yamamoto T, Yoshino M, Suzuki J, 2007. The relationship between occurrence of dust events and synoptic climatological condition in East Asia,1999–2003. JMSJ, 85: 81–99.

Yamamoto T, Yoshino M, Suzuki J, Chiba M, 2005. The relationship between occurrence of precipitation events and dust events under strong wind condition in East Asia. Proceedings of Fourth ADEC Workshop:Aeolian Dust Experiment on Climatic Impact. Nagasaki, Japan, 26–28 January, pp. 1–16.

Yan CZ, Zhou YM, Song X, 2009. Estimation of regions of sand and dust emission in the Hexi Corridor from a land cover database, an approach that combines remote sensing with GIS. Environ. Geol., 57: 707–713.

Yang B, Brauning A, Zhang ZY, Dong ZB, Esper J, 2007. Dust storm frequency and its relation to climate changes in Northern China during the past 1000 years. Atmospheric Environment, 41: 9288–9299.

Yang X, Zhang K, Jia B, 2005. Desertification assessment in China: An overview. Journal of Arid Environment, 63: 517–531.

Yukari H, Uno I, Wang ZF, 2006. Long-term variation of Asian dust and related climate factors. Atmospheric Environment, 40: 6730–6740.

Zhang B, Tsunekawa A, Tsubo M, 2007. Contributions of sandy lands and stony deserts to long-distance dust emission in China and Mongolia during 2000–2006. Glob. Planet. Change, 60(4): 487–504.

Zhang TR, Song ZX, Wang JY, 2008. The impact of variations of underlying surface and vegetation index on numerical simulation of dust emission. Plateau Meteorology, 27(2): 392–400.

Zhao JF, He Q, Lei JQ, Nagashima H, Yamada Y, 2005a. Analysis of the effective winds on dust weather in South Taklimakan. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact.Nagasaki, Japan, 26–28 January, pp. 1–5.

Zhao L, Sun J, 2005b. Coupling an wind erosion scheme with the MM5 Modeling systems and its applications in North China. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact.Nagasaki, Japan, 26–28 January, pp. 1–6.

Zhou HF, Wei WS, Gao WD, Xu GQ, 2005. Correlation analysis on dust events and ground meteorological factors in Xinjiang. Proceedings of Fourth ADEC Workshop: Aeolian Dust Experiment on Climatic Impact.Nagasaki, Japan, 26–28 January, pp.1–4.

Zou XK, Zhai PM, 2004. Relationship between vegetation coverage and spring dust storm events over northern China. J. Geophys. Res., 109:D03104. DOI: 10.1029/2003JD003913.

10.3724/SP.J.1226.2012.00140

*Correspondence to: Dr. WanYuan Li, Assistant Researcher of Cold and Arid Regions Environmental and Engineering Research Institute, Chinese Academy of Sciences. No. 320, West Donggang Road, Lanzhou, Gansu 730000, China. Tel: +86-931-4967681;Email: ywl@lzb.ac.cn

June 26, 2011 Accepted: October 10, 2011