An Exploratory Study of the Relationship between Annual Frequency of Invaded Typhoons in Taiwan and El Nino/Southern Oscillation

In this study, a nonlinear association between the characteristics of invaded typhoons (hereafter, referred to as ITYs) in Taiwan and sea surface temperature anomalies (SSTAs) over the equatorial eastern Pacific is identified. The relation is such that the years with a warmer-than-normal SSTA condition in the late season generally correspond to fewer annual ITYs (＜4),while those with a cooler SSTA are basically associated with more ITYs (≥4).When the SSTA becomes too cold(≤-1.0℃), however, the annual total number of ITYs is, in fact, greatly reduced (~1-2) instead of being increased. It is suggested that such a phenomenon is caused by different large-scale atmospheric responses to the underlying sea surface temperature conditions. A lower-level anti-cyclonic circulation with strong divergence suppresses the usual activities of tropical cyclones over the western North Pacific (WNP) in the cold years. An opposite mechanism occurs in the early season of the cool years. During the warm years, the typhoon activities over the WNP are not necessarily suppressed but are rather displaced eastward in response to an anomalous cyclonic circulation over the eastern part of WNP. Under such a configuration, however, Taiwan is not in the path of the preferential storm track. This observation appears to be useful in the long term forecasting of typhoon activities in the Taiwan area, as long as the SSTAs can be predicted.


INT RODUCTION
The in\1asion of typhoons is one ot� the. most important \Veather e\1ents in Taiwan t' o1-its direct link to one of the. most important climatic (or weather) parame.ters, namely', rainfall.
Summer rainf' all from inv(1ded typhoons (ITY s) is the 1najor \Vater resource for most parts of  TAO, \1ol . 7, No.l, ftvfarch 1996 the island. Naturally, an excessive amount can result in floods, while the reverse can cause droughts, with the most recent examples being the years of 1994 (7 ITYs) and 1993 (I ITYs), respectively. Either one of these extremely anomalous climate situations (i.e., much above or below the averaged number of ITYs) may cause a lot of damage to both the agricultural activities and the socioeconomic system of the island. In fact, the loss of property and human li· fe due to typhoon invasions is the greatest among all disasters from weather, as reported by Shieh and Chen ( 1985). Accordingly, we would have a better stand from which to prepare ourselves for such disasters, were we able to understand the characteristics of these in advance.
From the viewpoint of short-term climate predictions, the actual number of ITYs there . \vill be during the upcoming typhoon season is probably the most practical question that one ,.
would ask. Indeed, the interannual variability of yearly total ITYs in Taiwan is pronounced (Figure 1). Ac . cording to the figure, while the average number of ITYs is about 3-4 per year, there can be as many as 8 or even none at all. It should be noted that although the typhoon season in Taiwan is generally from July to September, it can begin as early as April and last until late November. Thus, some other practical questions may be asked are: What is the onset month of the season? When does the season end? How many ITYs are there in the main season, which includes July, August and September (JAS), the early season, April, May and June (AMJ), or the late season, October and November (ON), respectively? As a starting point, only the first question is addressed here. It should also· be noted that the typhoon tracking problem as well as the understanding of the preferentially large-scale dynamical condition for typhoon for1nation over the western North Pacific (WNP) are tackled by confining the scope of this study to local interest (see Section 3.1.2)), since the Taiwan area only occupies a smalJ part of the WNP basin.  The development and maintenance of typhoons (or more generally, tropical cyclones) primari ly depends on the heat supply from the ocean (e.g., Emanuel 1986). It is also known that the heat flux is closely re . lated to the sea surface temperature (SST), in addition to the atmospheric conditions. On acknowledging that the El Nino/Southern Oscillation (ENSO) event is by far the most noticeable, interannual, large-scale, air-sea interaction system, it is tempting to assume that the yearly characteristics of typhoons may be related to variations in tropical SSTAs over the eastern and central Pacific. The formation region of tropical cyclones over the WNP has been verified as being generally associated with ENSO in that the region spreads into the tropical central Pacific in El Nino ye.ars, while it mostly confines itself to the west of 160°E in La Nina years (e.g., Lander 1994;Wu and Lau 1992). Thus, the possible relationship between the annual frequency of ITYs in Taiwan and the SSTAs of the equatorial eastern Pacific is explored first.
Researches on relating local climate statistics in Taiwan to ENSO had not been em phasized unti1 recently. Follo\ving the findings of Halpert ( 1987, 1989) and Halpe11 and Ropelewski (1992) in global scale (in space) and in seasonal scale (in time), Hsu and Chen ( 1994) also investigated the empirical relati onship bet\veen short-ter·m climate vari ability (primarily on the surface temperature and precipitation) in the Ta iwan area and the variation of tropical SSTAs in monthl)' scales. Their results show that the temperature in September, and the precipitation in March tend to be abo \ 1e (below) normal fo llowing the El Nino (La Nina) winters. Except fo r that, studies at the Central Weather Bureau (CWB) of Taiwan generally suggested that the. local climate variations are more connected to the SSTAs over the North Pacifi c area centered at (20°N, I 50°E), rather than to the SSTAs over the equatorial eastern Pacific (Mong-Ming Lu, personal communication). In this work, however, it is pointed out that there is an intriguing relationship between the characteri stics of the local ITY events in the Tai\\1an area and the ENSO cycles.
The flow of this paper is outlined as follows. In Section 2, the data sets used are presented. The characteristics of annual ITY s, the selection of suitable ENSO index, and the way to reconstruct the mean surface wind fi eld from the pseudo-stress data are described in Section 3. The methodology applied to detect the relation ship between two parameters is briefly discussed in Section 4. In Section 5 the main results of this study are reported, while in Section 6 one possible physical explanation of the findings from the percept of mean large-scale circulations. Concluding remarks are then drawn in Section 7.

DATA
The ITY data set used in this work is provided by Hsiao ( 1992, his Table 6), whose study was based on historical typhoon reports of the CWB in Taiwan. With their records in Tai \ \>·an having been collected since 1897, an ITY at the CWB is defined as 1) a typhoon whose center had passed through Tai wan or the nearby maritime areas within a 200-km lin1it (about the averaged 10-m/s-wind rad ius of storrn), or 2) one that is beyond the 200-km limit but that results in any near sea-level stations in the Taiwan area recording a maximal mean wind over 10 mis or receiving a total rainfall of over 100 mm during its influence period (Hsu, 1949), or 3) one whose path of center approaches the nearb)1 maritime regions and causes loss of lives and/or great damage to properties on either Ta iwan or the nearby maritime areas (one c . ase in 1983), including Kin men Island (one case in 1973) (Chi 1978). Here, Figure l shows the time series of the annual total of ITYs in Taiwan updated to 1993 . However, only data from the period of· 1949-93 are used for the reason which becomes c . lear later. It may seem 1�4-0, v'ol. 7, l'lo.1, "l\1a. J"('. h 1996 that the definition of an ITY in Taiwan thus given is subjective . . However, on ree\1aluation of the records from 1958 to 1987, Hsiao et al. ( 1989) fo und only a slight difference ( 4 out of 101 storms) in the new list of ITYs created with a more objecti \ 1e definition (i.e., according to sea-level pressure, mean wind and precipitation recorded in the near sea-level stations) from the old one. The1·et' ore, the. result in this study should not be subject to strong statistical bias.
For compari son a11d completeness, records of tropical cyclone activities over the WNP (including South China Sea) area are also required. The data are taken from Lander ( 1994; see his 1able 1), which originally came from the Annual Tropic<1l Cyclone Report (ATCR) issued by the Joint Typhoon Warning Center (JT . WC) in Guam. The data only cover the period from 1960 to 1991, however. Note that tropical cyclones include tropical depressions (i.e., maximum \\i'ind < 17 mis), tropical storms (i.e., 17 m/s < maximum wind � 34 m/s) and typhoons (i.e., maximum wind > 34 mis). The ITYs considered in this work include both tropical storms and typhoons. Inf, or1nation about the ENSO cycle is provided by the Long-range Forecast Div·ision of the Japanese Meteorological Agency (LFD/JMA). The JMA has compiled four SST indices, since January 1949, by averaging monthly SSTAs over fo ur diffe rent regions mostly located in the equatorial eastern and central Pacific ( Figure 2). Focus here is on the SSTA index in the so-calle-d Nin.o 3 region that covers the area of 4 ° S-4 °N, 150° -90° W in that it is supposed to be one of the most representative ENS O indice . s since it generally covers the core region of the warm pool during the major ENSO events (Rasmusson and Carpenter� 1982).
The computed monthly pseudo-stress anomalies at sea leve1 (actually at 10 m above) in the 1961-93 period are used as a bridge to link the relationship between the typhoon activities and the ENSO phenomena. The data, \\1hich cover the domain of 29°S-29°N, 12 . 4°E-70°W with a 2° x 2° resolution, were compiled by Florida State University (f. SLT) from ship reports (e.g. Goldenberg and O'Brien 1981 ). Although this data set only contains the information of lower-level tropospheric circulation, it is the only available data that cover the earliest observations. Since the upper-Jayer circulation (e.g., 200-hpa) is generally in an opposite phase to the lo\ver-level circulation over the tropical region, a good view of the whole general circulation is still permitted over this period.

PRELil\'IIN ARY ANALYSIS
3.1 Typhoon Activities in Taiwan 3.1.1 Interannual and ultra-low-frequency '\tariabilit)1 It is apparent from Figure I that there is a pronounced interannual fluctuation in the annual number of ITY s (for shortness, denoted as <"fY> t\ v ). This can also be understood from the distribution of years categoriz . ed with the same <TY> tw (Figure 3). According to the upper panel, <TY> tw ranges from none to at most eight in one year. Most years generally record 3-4 ITY s, \\l hile the number of years drops considerably when <TY> tw is beyond 5 events, or whe-n it is belo \ v 2. The overall shape of the distribution in the lower panel ( 1949-93) is some . how similar to the longer record ( 1897-93 ), except that it is less like a normal distribution. This may be caused by either sample bias or a long-term trend. Fro1n the perspective of' a normal-distribution� Figure 3(b) indicates that the number of years with <TY> t \ \' = 1, 3 or 5, respectively, are slightly higher, while those with <TY> tw = 2, or 4, Jen-Cher1g Joseph Chang· .  ; 1940 1950 1960 1970 1980 1990 2000 . 1940 1950 1960 1970 1980 1990 2000  . 1940 1950 1960 1970 1980 1990 2000 Time (

87
respectively, are a little fe �'er than anticipated for the period of 1949-93. The long-terrn trend seems to exist as obser\'ed from Figure 1. The periods of mid-l 900s, midl 920s, early 60s and 80s appe . ar to have more <TY> tw ' while other periods show the opposite. It can also be noted that there were more <TY> t\ :V before 1930, but fewer after 1960. In this study, the years with <TY> tw 2: 5 are classifi e .d as above-normal years, whereas those with <TY> tw ::::; 2 below-nonnal years. Such a definition can be justified on the gr . ound that the probabilities of <TY> tw being less than 3 or greater than 5 are roughly equal ( rv 30% ), as deduced from the .time series (Figure 4 ). With this classification in mind, the shifts to ward lower <TY> tw in each category (as suggested by the comparison between both panels of 88 , TA O, Vo l. 7, No .l� !vfarch 1996 (a ) Distribution of Annual Invaded Typhoons in Taiwan ( 1897-1993) 30 .- <1> 2 0'-: : : : : : . . .
. .    Jen-Cheng Joseph Chang 89 Figure 3) should be viewed as further evidence of the somewhat decreasing long-terrn trend in the annual total number of ITY s in Taiwan. Nevertheless, such a trend may be artificial on account of the earlier observations being less reliable.

Dynamical relevance?
It is expected that there would be more <TY> tw in those years when with more tropical storms over the WNP. If this is true, what we really need to look at is just the relationship between the more large-scale typhoon activities over the WNP and the ENSO cycle. Several studies have explored the existence of such a correlation (e.g., Ramage and Hori 1981;Chan 1985, 1990� Dong, 1988Li, 198 8;Lander 1994 ), and it is unanimously accepted that a seesaw pattern for the location of storm formation is closely correlated to the ENSO events. In other words, the genesis region of tropical cyclones over the WNP shifts eastward when the Southern Oscillation index (SOI) is low and the SSTs in the equatorial central and eastern Paci fic are warmer than normal, while it shifts westward when the SOI is high and the SSTs are colder than nor1nal (see e.g., Lander 1994). Nevertheless, the issue on the annual frequency of typhoon activity has not yet been settled. To illustrate, the annual report of the JMA ( 1991) suggested that El Nino events generally correspond to a decrease in the annual totals of tropical storms in the WNP area. The result of Wu and Lau (1992) also indicated a significant ENSO-related signal in the annual total of simulated tropical cyclones from a 15-year (1962-76) GCM simulation with observed SSTs as boundary conditions. On the other hand., Ramage and Hori ( 1981) and Lander ( 1994) did not fi nd any significant correlation between tropical storm frequency and ENSO. Regardless of the above-mentioned debate, Table 1 suggests that the deciding factor as to how the tendency of the ITY s in Taiwan would be (i.e., more, nor1nal, or less) is not just associated with the annual total number of tropical storms occurring over the WNP (denoted as <TY> wnp ). From the table, the years with above-normal <TY> tw do in general tend to be associated with higher <TY> wnp (say 2: 27), and vice versa. It is however not always true, as one would notice, that several years (say 1989, 1972 and 1964) with more <TY> wnp did not result in more ITYs in Taiwan but in fac t the opposite occurred. Such an observation is not surprising since whether an individual typhoon over the WNP influences the Taiwan area or not strongly depends on its track, which is determined mostly by the background steering flow and the transient extratropical upper layer trough. Therefore, by fo cusing on the annual frequency of the ITYs in Taiwan, not only are the large-scale dynamical conditions favorable to the fo rmation of tropical sto1·1ns over the ·wNP considered, but the large-scale background flow that is preferential to direct typhoons which passing nearby Taiwan is also. Since the atmospheric response associated with the underlying SST boundary forcing is generally reflected in the mean large-scale circulation, only the mean large-scale steering circulation regarding to the typhoon tracking issue is emphasized.

ENSO Indices
Since the purpose of this study is to relate the annual activities of typhoons to tropical SSTA conditions over the equatorial eastern Pacific, it is obvious that a suitable yearly parameter fr om the monthly Nino 3 index needs to be compiled. The easiest one is perhaps to choose the yearly average of the SSTA index as an ENSO parameter. However, by doing so, the ENSO signal would be greatly reduced since the extreme phases generally do not persist throughout the whole calender year. Therefore, it is worth examining the time series of the SSTA first in order to select a representative parameter. 90 TAO, Vol. 7, 1'.rc). 1, Ma1,ch 1996 Tab le 1. Ranking of years according to the corresponding annual total of tropical storms over the WNP ( <TY> wnp ) in the period of 1960-9 1. Also indi cated in the fo urth column are the associated categories (classified as + above-no1ma], and -below-nonna1) of the.annual total of ITYs in Taiwan ( <TY> tw ) in each year.  Figure 5 shows the e\'Olution of the five-month running-mean SSTA index at Niiio 3 within a two-year window. It is noted that a significant magnitude of warm/cold phases tends to occur fr om June of one year through February of the following. The maxima were generally reache . d in the . period of September through December. Based on this observa tion, it is reasonable to use the September-December mean of each year as the desired ENSO

91
index, which is referred to as <SSTA> sond . The only exception is perhaps the extremely strong ENSO event that occurred in the winter of 1982-83. During that episode, the warmest phase arrived in December 1982. However, the period with SSTA higher than 1 ° C persisted from June 1982 through July 1983 for more than a one-year span. As a result, the typhoon season in 1983 might have actually been significantly affected by the ENSO event although <SSTA> sond of 1983 was merely -0. 175°C. Tab le 2 gives the compiled <SSTA> sond for each year.

Derivation of Weighted-Wind Anomalies
The FSU pseudo-stress vector is defined as where · r denotes the . pseudo-stress, subscripts indicate the x, ' Y components respectively, lV is the wind speed� and · u , 1J are zonal and meridional wind components, respectively (see Goldenberg and O'Brien 1981 ) . The over bar stands for the . monthly mean. Therefore, the pseudo-stress field can be used to substitute the horizontal wind field if the circulation is one's main interest. To also examine the corresponding divergence field, it is better to revert the FSU pseudo-stress into a field that has a unit equivalent to velocity. Assuming that the nonlinear effect among diffe rent te . mporal scales can be neglected, ' the monthl)' mean wind speed on each grid point can be estimated as: As such, the monthly-mean horizontal wind field can be reconstructed by : (3) In thi s study, Eqs. (2) and (3) were applied to the FSU monthly pseudo-stress anoma lies to recover the monthly wind anomalies. Prior to this procedure, both temporal and spatial 92 TA O, Vo l. 7, No .l, March 1996  smoothing were applied to the original pseudo-stress data since the seasonal-scale, large scale circulation is of prime interest here. The temporal filtering is a 3-point smoothing .with weights of ( 1, 2, 1 )/4. A 9-point spatial smoothing is employed with weights equivalent to applying a filter of (1, 2, 1)/4 simultaneously in both the x and y directions.

Linear Correlation Analysis
The purpose of this study is first to verify whether the <TY> tw can be related to the SSTAs that occurred over the equatorial eastern Pacific. The simplest way to identify any linear relation between two variables is to calculate their correlation coefficient. However, if the relation between two parameters is not linear, it cannot be detected in this manner, as, it is shown later, is the case in this study. A more general analysis is thus desired to reveal the association be.tween two variables .

Contingency Ta ble Analysis
Although <TY> tw is a discrete integer, it can always be classified into different cate gories defined by certain ranges. This is also true for the SSTA index. Therefore, if there is any association between the two parameters, it can be detenT1ined by examining the distribu tion of data in eac . h category. Such an analysis is the so-called Contingency Table Analysis (CTA) with the significance of the association measured by the X2 test.
Suppose that it is decided to have I categories in the SSTA index, and J categories in <TY> tw · The X2 can thus be calculated as : Note that . 1Vi , j is the number of observed data points that fall into both the i-th category of the SSTA and the j -th category of <TY> tw ' ni,j is the expected value of Ni ,j , Ni denotes the of data points that belongs to the J -th category of <TY> tw , and 1 is the total number of data points considered. The strength of the association between two variables is measured either by the Cramers vr ' which is defined as : x2 (6) _rv · min( I -1, J -1) ' or by the contingency coefficient C, where: c-x2 (7) • Table 3. Ranking of years  according to the <TY> tw of each year. The corresponding <SSTA> sond (°C) is also given in the last column . •

Rank
Year <TY> tw <SSTA>sond the relation betwe . en <TY> tw and the SSTA index is a nonlinear one, and one such that <TY> tw is low when the SSTA is warmer than usual, but high when the SSTA is slightly cooler than normal. However, when the SSTA becomes too cold (say roughly 1.0°C below normal), <TY> tw in fact would be reduced, instead of increased.
To fu rther demonstrate the nonlinear relationship, the scatterp lot of <TY> tw versus <SSTA> sond was graphed in Figure 7. It is apparent that the warm years are generally associated with <TY> tw < 4, which is also true for years with SSTA 1.0 (°C) below normal. On the other hand, the cool years [SSTA below normal, yet not extremely cold (2: -1.0°C)] are roughly associated with <TY> tw �3. With this diagram as a reference, the data points may be classified into three categories in terms of the SSTA index, i.e., the warm years (SSTA 2 0°C), the cool years (-1 .0 ::.:; SSTA < 0°C), and the cold years (SSTA < -l.0°C). (In fact, this has already been done in previous discussions here.) Together with previous definitions of above-, normal-, and below-year in terms of <TY> tw ' a contingency table can be formed as listed in Table 4. From this table, it is evident that the majority of cold years are associated with below-normal <TY> tw ' 50% of the cool years are with above-normal <TY> tw , while more than 50% of the warm years mostly correspond to a normal range of ITYs in Taiwan. The test, as forrr1ulated in (4), indicates the association between the . SSTA index, and <TY> tw is very significant (surpassed 99% confidence level). The strength of the association is 0.4 as measured by Cramers V (6), or 0.49 by the c . ontingency coefficient cv (7).  Fig. 7. Scatterplot for <TY> tw versus <SSTA> sond in the Nino 3 region.
From these . points, it is concluded that the annua] total ITY s in Tai wan is related to the SSTAs over the equatorial eastern Pacific. The association between the two parameters is not a simple linear correspondence, but, on the contrary, a rather nonlinear one. It appears that there are fe \ ver <TY> tw in warm years, whereas there are more in cool years. However, when the SSTA gets too c . old, instead of being enhanced, the <TY> tw is actually severely reduced. It thus seems that there is a threshold, or a turning point in below-normal SSTAs that reverses the correlation between the SSTA index and <TY> tw · In the previous section, a statistical association has been identified between the tropical SSTAs and the annual number of ITYs in Taiwan. It is speculated that there may exist a certain dynamical linkage between these two par_ am eters, and the data indicated that they are indeed related to each other.
Since many observational and modeling studies suggest that the atmospheric responses are quite different to various equatorial SSTA conditions, it is reasonable to assume that the tropical SSTAs could influence the interannual variation in <TY> tw through the mean large-scale circulation over the WNP associated with the ENSO cycle. To explore such a possibility, it is therefore instructi ve to examine the composite circulati ons under different SSTA and <TY> tw conditions.
To address the main concerns of this study, the composites of three clusters, namely the cold years with below-nor1nal <TY> tw (1988,1973,1970,1964), the cool years with above normal <TY> tw (1985,1984,1981,1962,1961), and the war1n years with below-normal <TY> tw (1979,1976,1972,1963) are highlighted. For easy reference., the three groups are referred to as COLD years, COOL years and WA RM years . With the FSU pseudo-stress data starting in 1961, it is therefore not possible to include cases prior to 1960 in each group. However, it is the only data set covering the earliest observation that could be found, as mentioned in Section 2, so as to have as many cases in each cluster as possible. Another point to keep in mind is that the year 1993 is not included in the WA RM composite due to its different evolution toward the end of the calendar year from that of other years that have been chosen.
In Figure 8 would greatly suppress the generation of tropical stor1ns over the WNP. Except for 1964, the other three years considered here are confinned to be with a low <TY> wnp · Furthermore, according to Table 1, the lower <TY> wnp generally corresponds to a reduced <TY> tw · It is thus understandable why the statistics indicate that <TY> tw tends to be low during the cold years. The anti-cyclonic flow (and the corresponding divergence field) remains there in the late season, but with a smaller convergence region to its south. Those circulation fe atures (in both the main and the late seasons) are significant as confir1ned from a Student t-test (95o/o confidence level; not shown). In the early season, a weak anti-cyclonic circulation can be · fo und to the east of the Philippines, and a large cyclonic flow is found mainly north of 15°N, extending from 120°E to 160°E. However, the t-test (not shown) shows only a few points over the WNP (mainly to the east of 160°E, or to the south of S0N) that are significant (95% ), indicating the circulation in the early season is either quite different from case to case, or close to normal in each case. The composite of COOL years is shown in Figure 9. Significant cyclonic circulation together with a large-scale convergence field over the WNP can be found in the early season. It is located over (20°N, 140°E), which spreads from at least 124°E (the western boundary of the data domain) to the international dateline, and from S0N to beyond 30°N. This is almost opposite to the findings revealed in the main season of COLD years, though it covers a broader area, albeit of weaker intensity. Such a large-scale dynamical condition is undoubtedly preferential to the generation of tropical storms over the WNP in the early season. Although that does not necessarily result in the earlier occurrence of ITY s in Taiwan, it at least increases the possibility from a statistical point of view. By comparing this with the data from the JMA (1993), it is noted that ten out of eighteen years · since 1960, when there were four (or more) typhoons over the WNP in AMJ (averaged 3.51), did correspond to at least one ITY in Taiwan in the early season. Furthermore, the data here indicate that there tends to be more <TY> tw if the typhoon season come earlier (9 out of 14 cases) in Taiwan. Therefore, this feature at least provides a preferential large-scale dynamical background that is consistent with more <TY> t \\l in those COOL years. The large-scale pattern seems to propagate eastward in the main season, as suggeste . d from the middle panel of Figure 9. In the meantime, a hint of an anti-cyclonic circulation seems to appear over the Philippine area. The circulation in the late season is dominated by an anti-cyclonic flow to the west of 140°E, while the cyclonic fe ature is primarily located to the east of 140°E with weakened intensity.
It is interesting to note that there is a cyclonic circulation located to the east of 140° E through the WA RM years (Figure I 0). This is not surprising since the strong anomalous equatorial westerlies in the central Pacific are expected toward the boreal ENSO winters (e.g., Rasmusson and Carpenter 1982;Rasmusson and Wallace 1983). There is also a strong convergent zone over the climatologica l location of tropical sto1·m genesis as shown in the main season. It appears that such a large-scale dynamical condition is not unf a vorable to the generation of tropical cyclones, though the location of origin may be shifted eastward accordingly . Indeed, the annual total number of typhoons over the WNP in the WARM years are not necessary in the lo\\l·er end (e.g., in 1972, see Table I). Then, the question is to ask why the circulation is associated with an reduction in <TY> tw · Fi gure 11 shows the large scale background circulation (on 700 hPa) for recurving-so uth tropical storms over the WNP compiled b) ' Harr and Elsberry ( 1991, their Figure4 ). One should note that the pattern within the domain that is of interest is similar to the correspond ing patterns presented in Figure 1   flows centered near along 150°E generally steer the tropical sto1·ms to recurve northward . As such, the occurrence of ITYs in Tai wan is reduced. In other words, Ta iwan is probably not in the preferential storm track with such a large-scale circulation feature. It is now clear that different responses of atmospheric circulation to diffe rent underlying surface conditions, in tum, at least partially control the annual frequency of ITYs in Tai wan. Generally speaking, the large-scale circulation in the cool years is such that it is favorable to the storrn formation over the WNP, especially in the early season. This suggests that the annual number of ITYs in Taiwan are likely to be higher than normal in the cool years, which is consistent with the observation. On the other hand, the unfavorable large-scale dynamical condition in the cold ye . ars suppre . sses the acti vities of tropical cyclones over the WNP in both the main and late seasons. As such, the chance for Ta iwan and the nearby regions to be affected by typhoons is greatly reduced. In the warm years, the typhoon activities over the WNP are not necessarily weakened by the large-scale flows, except for a perhaps eastward displacement. However, the preferential storm track is generally not directed to Tai wan or nearby areas, which results in a mostly mediate or less than normal number of ITYs in Tai wan.  In this study, the possible relati onship betwee11 the interannual variability of ITY activ ities in Taiwan and the SSTAs over the equatorial eastern Paci fic has been explored. It was found that the annual frequency of ITYs in Tai wan is likely related to the ENSO (SSTA) index. Their association is not a simple linear correlation, but rather a nonlinear one. The nonlinear relationship is such that the annual number of ITYs in Tai wan is about average to less-than-normal when the SSTA index is warme. r, while the number increases when SSTA is cooler. However, when the SSTA becomes too cold (� -1.0°C), the annual total actually gets smaller than normal instead of getting larger. This observation appears to be beneficial in improving the long-term forecast on t)'phoon activities in the Tai wan area, provided that the SSTAs over the equatorial eastern Pacific can be accurately predicted.
The composite analysis suggested that such an intriguing relationship can be mainly attri buted to different atmospheric responses over the WNP region to changes in underlying SSTAs over the equatorial eastern Pacific. In the cold years, there is an associated lower tropospheric large-scale anti-C)'Clonic circulation accompanied by a large divergence field over the climatological formation zone of tropical cyclones over the WNP. Such an unfavorable large-scale feature would suppress the generation of tropical storms in the main and late typhoon seasons, thereby reducing the probability of visiting typhoons in the Taiwan area. In contrast, there is large-scale cyclonic circulation with convergence located over the WNP in the cool years , especially in the early typhoon season, which is more favorable to the fo rmation of tropical cyclones. Consequently, the annual total number of ITYs in Taiwan is likel)' to increase. As for the warm years, the induced cyclonic circulation over the eastern part of the WNP (e . ast of 140°E) is actually favorable to storm-genesis over the same region. However, the associated anti-cyclonic circulation to the north is in such a position that it may drive most of the storms to recurve northward before approaching Taiwan or nearby areas. In other words, Taiwan tends to be out of the prefe1·ential storrn track under such a large-scale configuration. As a result, the annual number of ITYs is generally no more than the average. It is interesting to note that the acti \ 1ities of typhoon invasion in a wa1·1n year are not necessarily as inactive as in a cold year because the typhoon acti,rities over the WNP in the former period have mostly not been suppressed as in the latter, but only been displaced eastward. Such a displacement of the preferred stoc·m-genesis region may also contribute to the reduction in the number of annual ITYs for the storms being generally farther from Tai wan than otherwise.
Since this is an exploratory work, the search for a possible statistical relationship be t. w een the characteri stics of invaded typhoons in Tai wan and the ENSO indices is by no mean comprehensive. Other ENSO parameters, such as the Nino 4 index or SSTA changes, should also be investigated. As a matter of fact, a preliminary analysis suggests that yearly SSTA trends could be a better indicator of the annual typhoon activities in Tai wan (Chang, 1996). It should be noted that the proposed interpretation of the stati stical results with mean large-scale circulations is far from complete since other factors are also important in determining the track of typhoons. Since only surface anomalous circulations on a restricted domain ha\'e been investigated, a further investigation of 3-dimensional flows on a larger domain is desired to avoid any bias arising from this study. The finding may also shed some light on the annual activity of tropical storms over the WNP and its relati onship with the ENSO cycle.