Physical mechanisms of the Australian summer monsoon: 2. Variability of strength and onset and termination times

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2005jd006808, 2006 Physical mechanisms of the Australian summer monsoon: 2. Variability of strength and onset and termination times Kwang-Yul Kim, 1,2 Katherin Kullgren, 1 Gyu-Ho Lim, 3 Kyung-On Boo, 3,4 and Baek-Min Kim 3 Received 25 October 2005; revised 28 February 2006; accepted 22 May 2006; published 27 October [1] Variability of the Australian monsoon onset, termination, and strength was investigated in terms of the first five major modes of monsoon precipitation found via cyclostationary empirical orthogonal function analysis. They are the El Niño Southern Oscillation (ENSO) mode, the seasonal cycle, the ENSO transition mode, and the two Madden-Julian oscillation (MJO) modes, respectively. Whereas the seasonal cycle defines the fixed onset (5 January) and termination (5 March) dates, the presence of the other modes alters these dates. The impact of each mode was investigated after each mode, with a strength corresponding to 1 standard deviation of its variability, was added to or subtracted from the mean seasonal cycle. It is shown that the contribution of each mode to the monsoon precipitation is geographically complex and varies significantly throughout the monsoon period. One striking feature is that the impact of each mode is highly asymmetric with respect to its phase. The negative modes generally affect more significantly the onset, the termination, and the amount of monsoon precipitation. Although each of the four modes makes a unique and tangible contribution, the ENSO mode contributes most significantly to the overall mean and variance of the monsoon precipitation variability. While the positive ENSO (El Niño) mode does not seriously alter the onset and termination times, the negative ENSO (La Niña) mode prolongs the duration of the Australian summer monsoon significantly. Although the contribution to the overall mean is small, the two MJO modes are the most dominating factor controlling the onset and termination times of the Australian summer monsoon. Citation: Kim, K.-Y., K. Kullgren, G.-H. Lim, K.-O. Boo, and B.-M. Kim (2006), Physical mechanisms of the Australian summer monsoon: 2. Variability of strength and onset and termination times, J. Geophys. Res., 111,, doi: /2005jd Introduction [2] In part 1 of this companion set [Kullgren and Kim, 2006], physical evolution in the normal Australian summer monsoon (the seasonal cycle) was investigated along with a description of the onset and termination mechanisms. Having discussed the normal picture of the Australian summer monsoon, part 2 of this companion set (this paper) will examine the variability of the onset, termination, and the strength of the monsoon. [3] The onset of the Australian summer monsoon is characterized by a reversal of lower level wind anomaly from easterly to westerly and the opposite in the upper 1 Department of Meteorology, Florida State University, Tallahassee, Florida, USA. 2 Now at Environmental Forecast and Value-Oriented Research Services Inc., Tallahassee, Florida, USA. 3 School of Earth and Environmental Sciences, Seoul National University, Seoul, Republic of Korea. 4 Now at Climate Research Laboratory, Meteorological Research Institute, Seoul, Republic of Korea. Copyright 2006 by the American Geophysical Union /06/2005JD atmosphere over northern Australia. This is accompanied by a rapid increase of rainfall over northern Australia [Troup, 1961; Nicholls et al., 1982; Nicholls, 1984; Holland, 1986; Hendon and Liebmann, 1990]. Despite these characteristic features, they alone cannot provide a unique definition of the onset of the Australian summer monsoon. [4] The definition of onset of the Australian monsoon has been stated in many different ways. One of the earliest thoughts came from Troup [1961], who defined the onset to have occurred when 4 out of 6 stations near Darwin experienced their first rainfall event simultaneously after 1 November and the area-averaged rainfall over N days exceeded 0.75(N + 1) inches. Others used a large-scale perspective approach to determine the onset. Davidson et al. [1983], for example, depicted the onset as directly resulting from a flare-up of convective activity associated with anticyclones over northern Australia, which was determined by satellite imagery. Keenan and Brody [1988] found that the convective bands were directly linked to a 200 mbar trough/ ridge system and a subtropical jet streak. Yet still others looked at rain and wind parameters at the station level. Nicholls et al. [1982] used the monsoon onset defined by Troup [1961] to verify that onsets can be predicted, to a certain degree, by the amount of rainfall at a single station. 1of17

2 Table 1. R 2 Values of Regression of Key Physical Variables Onto the First Five CSEOF Modes of Precipitation CSEOF Mode Variable First Second Third Fourth Fifth Sea level pressure Surface temperature OLR hpa wind hpa wind hpa W hpa W Moisture transport Nicholls [1984] concluded that the onset of the monsoon had occurred at a station in northern Australia after 15% of the mean annual rainfall was recorded. Also using station data, Holland [1986] deduced that the monsoon onsets occur at a station with the passing of the first westerly winds at the 850 mbar level. Davidson and Hendon [1989] also stated that the monsoon onset was associated with an acceleration of the lower level winds. Hendon and Liebmann [1990] used the combination of rainfall and wind to define the onset; onset is determined by the first detection of wet (having rainfall of at least 7.5 mm d 1 ) westerlies. Drosdowsky [1996] criticized that the wet westerly as described by Hendon and Liebmann [1990] was misleading since there was no clear correlation of westerly winds Figure 1. ENSO mode (first CSEOF) of 5-day averaged austral summer precipitation anomalies with respect to the summer (2 December to 31 March) mean. This mode explains 16.1% of total variability. Only every fourth pattern is shown here. The amplitude is mm d 1. 2of17

3 Figure 2. Principal component (PC) time series (with symbols) of (a) ENSO mode, (b) ENSO transition mode, (c) first MJO mode, and (d) second MJO mode. The time interval is from December 1979 through March The dashed lines represent the level of 1 standard deviation for each PC time series. The solid curves in Figures 2a and 2b are the first two PC time series of the sea surface temperature anomalies over the tropical Pacific Ocean, respectively. See text for more information. and rainfall. Using mean zonal wind from surface to 500 hpa and cross-spectral analysis, he found that the rainfall led the maximum west wind by one quarter of a cycle (a cycle being approximately 40 days) in the intraseasonal band. [5] It is difficult to come up with an accurate definition of the monsoon onset because it is affected by many physical mechanisms. Holland [1986] and Nicholls et al. [1982], for example, found the onset to be in direct correlation with the Southern Oscillation Index (SOI). They concluded that the correlation between the SOI and the onset is as follows: When a negative SOI (warm El Niño event) occurs, there will be a delay in the onset of the Australian monsoon. Drosdowsky [1996] also found a strong relationship between the Australian summer monsoon onset and the El Niño and the Southern Oscillation signals. He concluded that the onset of the Australian summer monsoon is strongly correlated with the Southern Oscillation Index (SOI) in the austral spring (SON) prior to the onset but is not correlated in the spring following the onset, which was in contrast to what Holland [1986] suggested. He also concluded that the interannual variability and the intensity of the monsoon are significantly related to the El Niño Southern Oscillation (ENSO) cycle. Hung et al. [2004], on the basis of the strong correlation in precipitation from Southern Asia and Northern Australia with the Niño 3.4 SST index, were able to create an Asian-Australian monsoon index, which was used to determine strong and weak monsoon years. Joseph et al. [1991] additionally found a strong correlation between the Indian summer monsoon rainfall (ISMR) from June through September and the onset of the Australian summer monsoon. They found that a deficiency of ISMR is associated with a delayed onset, and an excessive ISMR is associated with an early onset. Vincent [1994] suggested the south Pacific convergence zone (SPCZ), which lies angled across the Pacific from the equator to 20 S and from about 130 E to 140 W during the austral summer (November March), may have an impact on the rainfall associated with the monsoon in northern Australia. He claims that the maximum rainfall is found to the south of the SPCZ, which encompasses northern Australia during the austral summer. Hendon and Liebmann [1990] reported that the onset of the monsoon coincides with the first occurrence of the convectively active 40- to 50-day oscillations, and McBride [1987] found a periodicity of 40 days in cloud bands that were associated with convective activity. [6] As hinted above, one obvious difficulty arises in coming up with an unambiguous definition of the Australian monsoon onset because of the distinct contributions of individual physical mechanisms to the variability of the monsoon precipitation. Each physical mechanism may alter the evolution patterns of precipitation and other physical variables in a unique way such that a general definition of the monsoon onset may not necessarily be possible. Thus it would be wise to identify first the primary physical mechanisms exerting influences on the Australian summer pre- 3of17

4 Figure 3. Same as Figure 1 but for the ENSO transition mode (third CSEOF). This mode represents ENSO transition from a warm phase to a cold phase. This mode explains 8.8% of total variability. cipitation and have an orderly account of how each of these mechanisms affects the monsoon precipitation. As addressed in part 1 [Kullgren and Kim, 2006], the primary physical modes are extracted from observational data via cyclostationary empirical orthogonal function (CSEOF) analysis. The present study delves deeper into the various influences governing the monsoon precipitation so that the onset variability is better understood. In addition, the variation of the termination and the strength of monsoon will also be addressed in the context of individual physical mechanisms. [7] This paper is divided into 6 sections. Section 2 briefly summarizes the results of analysis. The methods of analysis were explained in detail in part 1 [Kullgren and Kim, 2006]. Section 3 describes the four major modes other than the seasonal cycle extracted from the precipitation data; they are the physical mechanisms introducing variability in the onset, strength, and termination of the Australian summer monsoon. A new definition of the onset and termination of the Australian summer monsoon is introduced in section 4, with the normal onset and termination being defined in the context of the seasonal cycle. Then, section 5 describes how each of the four main modes affects the onset and termination times, along with the strength of the monsoon, followed by summary and concluding remarks in section Results of Analysis [8] First, empirical orthogonal function (EOF) analysis was conducted on all variables including the Xie-Arkin pentad precipitation data [Xie and Arkin, 1997]. Note that 4of17

5 Figure 4. Same as Figure 1 but for the first MJO mode (fourth CSEOF). This mode describes eastward propagating oscillations with 50-day period. This mode explains 7.6% of total variability. the analysis was conducted only for the austral summer (December March). Then, CSEOF analysis was carried out in the resulting EOF space using the first 30 EOF modes for each variable (see part 1 [Kullgren and Kim, 2006]). The amount of variance explained by the first 30 EOFs varies from one variable to another; they explain about 68% of the total variability for the precipitation [see Kullgren and Kim, 2006, Table 1]. CSEOF analysis on the precipitation data identified five major modes; they are the ENSO mode, the seasonal cycle, the ENSO transition mode, and the two Madden-Julian oscillations (MJO). They explain 16.1%, 12.4%, 8.8%, 7.6%, and 7.0% of the total precipitation variability, respectively. [9] Regression analysis was then conducted to extract the evolutions of other physical variables in consistence with those of the first five CSEOF modes of precipitation (see part 1 [Kullgren and Kim, 2006]). The first 10 PC time series of the predictor variables were used for this exercise. The R 2 value of regression (1 regression error variance/ signal variance) is generally high for all the variables despite the small number of PC time series employed in the analysis (Table 1). Thus all the regressed patterns have essentially the same evolution history as the PC time series of the corresponding CSEOF modes of precipitation. 3. Major Modes of Australian Monsoon Precipitation [10] This section describes the first five major modes of the CSEOF analysis except for the seasonal cycle, which has been investigated in part 1 [Kullgren and Kim, 2006]. Only a brief description of each mode is included here. 5of17

6 Figure 5. Same as Figure 1 but for the second MJO mode (fifth CSEOF). This mode describes eastward propagating oscillations with 50-day period. This mode explains 7.0% of total variability. Detailed features of their physical, dynamical and thermodynamical nature can be found at the authors anonymous ftp Web site. (The detailed evolution features of precipitation, outgoing longwave radiation, surface temperature, sea level pressure, low-level wind, upper level wind, moisture transport, etc., for the first five modes (aumon1.mov, aumon1a.mov,..., aumon5a.mov) can be acquired at ftp://pacific.met.fsu.edu/pub/kkim/nfs/aus in QuickTime movie format and will not be shown in the present paper.) [11] The ENSO mode (first CSEOF mode) describes the evolution of physical variables during a typical El Niño event. As shown in Figure 1, the patterns of the precipitation anomalies unmistakably reflect the typical condition of El Niño, and so do the patterns of surface temperature anomalies (figure not shown). As shown in the figure, the ENSO mode exerts a significant influence over northern Australia particularly during the early (December) and terminal (March) stages of the Australian summer monsoon. Whereas the tropical central Pacific exhibits excessive precipitation, the western Pacific including northern Australia experiences drought conditions during an El Niño event. The drought condition associated with an El Niño event is due to the downward motion of the anomalous Walker circulation over the western Pacific Ocean. Not only does the corresponding PC time series clearly show major El Niño and La Niña events in the data period, but it also coincides well with the PC time series of the first CSEOF mode of the sea surface temperature anomalies (Figure 2a). [12] The third CSEOF mode is the ENSO transition mode (ETM), which describes the transition between a warm 6of17

7 Figure 6. Composite of five-mode reconstruction over the Australian summer monsoon period (2 December to 1 April) of (a) precipitation (mm d 1 ) averaged over the S band and (b) 850 hpa zonal wind (m s 1 ) averaged over the S band. Dark shade represents values greater than 7.0 for precipitation and 1.0 for zonal wind, and light shade represents values less than 3.0 for precipitation and 1.0 for zonal wind. The data period is from December 1979 through March The dash-dotted lines across Figures 6a and 6b represent the normal monsoon onset time (5 January). (wet) phase and a cold (dry) phase of ENSO over the tropical Pacific Ocean (Figure 3). The pattern of surface temperature anomalies for this mode exhibits positive anomalies over the central Pacific in the premonsoon stage and negative anomalies expanding from western to central Pacific in the termination stage of the Australian summer monsoon (figure not shown). The corresponding PC time series are highly correlated with the time series of the second CSEOF mode of the sea surface temperature anomalies over the tropical Pacific domain (Figure 2b). This mode also exerts significant influences on precipitation over northern Australia specifically in the premonsoon and terminal stages (Figure 3). Figure 7. Same as Figure 6 except that summertime (December March) mean has been removed both from precipitation and zonal wind. Dark shade represents values greater than 0.5, and light shade represents values less than of17

8 Figure 8. Mean precipitation anomalies (mm d 1 ) over northern Australia (10 20 S) in the seasonal cycle of the Australian summer monsoon. Dark shade represents values greater than 1.0, and light shade represents values less than 1.0. The dash-dotted lines represent the normal monsoon onset time (5 January) and termination time (5 March). [13] The MJO modes (fourth and fifth CSEOF modes) describe the evolution of Madden-Julian oscillations with alternating positive and negative phases [Madden and Julian, 1972, 1994]. They are tied with eastward propagating oscillations with a period of 50 days as shown in Figures 4 and 5. These eastward propagating signals pass the northern part of Australia and appear to influence significantly the onset and the ensuing evolution of the Australian monsoon [Hendon and Liebmann, 1990]. Because of the varying phase of MJO, it cannot be captured as one mode in CSEOF analysis [Seo and Kim, 2003]; the phases of the two MJO modes are separated by approximately one quarter of the MJO period with the first MJO mode leading the second MJO mode (see Figures 4 and 5). [14] Figure 2 shows the PC time series of the four primary modes described above for the Australian summer monsoon precipitation together with the 1 standard deviation level of each PC time series. As shown, there are several years when the strengths of individual modes exceed the 1 standard deviation level over certain stages or the entire period of the monsoon. During these years, the impacts of individual modes may be strongly felt; as a result, the evolution of the Australian summer monsoon deviates from its normal course affecting the onset and termination dates along with the strength. The next two sections will describe how individual modes modify the normal evolution of the monsoon (i.e., the seasonal cycle). 4. Definition of Onset and Termination [15] Hendon and Liebmann [1990] defined the onset date of the Australian summer monsoon as the first arrival of wet westerlies at Darwin (130 E, 12 S); they noted that the average onset date of the Australian summer monsoon is Figure 9. (a) Precipitation anomalies in the seasonal cycle plus ENSO mode over northern Australia (10 20 S). (b) Precipitation anomalies in the seasonal cycle minus ENSO mode over northern Australia. Here, the seasonal cycle has the mean amplitude, and the amplitude of the ENSO mode is at the level of 1 standard deviation. The dash-dotted lines across Figures 9a and 9b represent the normal monsoon onset time (5 January) and termination time (5 March). 8of17

9 Figure 10. Contribution of main modes to the precipitation over northern Australia: (a) 125 E, (b) 135 E, and (c) 145 E. The left column is for the positive modes, and the right column is for the negative modes. The amplitude of each mode is at the level of 1 standard deviation. around 25 December with a standard deviation of 16 days. This is consistent with the average onset date of 28/29 December reported by Drosdowsky [1996]. Indeed, there seems to be, on average, a close agreement between the initiation time of positive zonal wind at Darwin and the timing of precipitation exceeding 7.5 mm d 1 over northern Australia. Figure 6 shows a composite image of the fivemode reconstruction of the S averaged precipitation and the S averaged zonal wind over the Australian summer monsoon period (December March); it represents the average. Confirming their finding, the 7.5 mm d 1 precipitation contour averaged over the northern Australian longitude ( E) initiates roughly at the same time as the 1 m s 1 isotach 9of17

10 Figure 11. Precipitation over northern Australia after the contribution of individual modes: (a) western part ( E), (b) central part ( E), and (c) eastern part ( E). The left column is for the positive modes, and the right column is for the negative modes. The amplitude of each mode is at the level of 1 standard deviation. of the zonal velocity in the five-mode reconstructions of precipitation and zonal velocity. As noted by Drosdowsky [1996], however, zonal wind and precipitation are not in phase at Darwin on an intraseasonal timescale. Therefore the onset definition by Hendon and Liebmann [1990] is not accurate when the intraseasonal component of variability is strong. [16] In the present study, therefore, the onset date of the Australian summer monsoon is defined as the occurrence of positive precipitation anomalies. Namely, with the onset of the monsoon, local precipitation exceeds the austral summer mean value at that location. This seems to be a sensible definition since one precipitation value (say, 7.5 mm d 1 ) alone cannot uniformly define the monsoon onset over all of 10 of 17

11 Figure 12. Same as Figure 9 except for the ENSO transition mode. Australia. When comparing Figures 6 and 7, the initiation of the 7.5 mm d 1 precipitation contour roughly coincides with the zero contours of the corresponding precipitation anomalies and zonal wind anomalies over northern Australia. Likewise, the termination date of the Australian summer monsoon is defined as the disappearance of positive precipitation anomalies. According to the five-mode reconstruction, the average onset date during is approximately 5 January, and the average termination date is approximately 5 March with an uncertainty of ±5 days since 5-day averaged data were employed. In contrast, Hendon and Liebmann [1990] noted that the average onset date of the Australian summer monsoon is around 25 December with a standard deviation of 16 days. Drosdowsky [1996] determined the onset (termination) to be the first (last) day of the first (last) active rainfall period and found it to be 28/29 December (13 March). The onset and termination times in this study define a significantly shorter monsoon duration than that of Drosdowsky [1996]. See Kullgren and Kim [2006] for further discussion. [17] Since the amplitude of the seasonal cycle is generally positive, the timing of positive precipitation anomalies as reflected in the seasonal cycle averaged over northern Australia (10 20 S) does not vary (Figure 8). Thus the variation in the onset and termination dates arises from the disturbance of the zero contour of the precipitation anoma- Figure 13. Same as Figure 9 except for the first MJO mode. 11 of 17

12 Figure 14. Same as Figure 9 except for the second MJO mode. lies due to the influence of modes other than the seasonal cycle. It should be noted that the long-term average contribution of each mode is not quite zero although small. That is why the zero contour of the precipitation anomaly in Figure 7 slightly differs from that of the seasonal cycle in Figure Variability of Onset, Termination, and Strength of Australian Monsoon [18] In this section, variability of the monsoon onset date will be investigated in lieu of different physical subsystems comprising the Australian summer monsoon system. As addressed earlier, the present analysis identified five primary modes of austral summer precipitation: the ENSO mode, the seasonal cycle, the transition mode, and the two MJO modes. Each of these modes uniquely contributes to the variability of the onset and termination of the Australian summer monsoon. We will investigate how each mode affects the onset and termination dates of the monsoon. For this end, each mode is added/subtracted to the seasonal cycle. While the amplitude of the seasonal cycle is set to be its mean value, the amplitude of the mode added/subtracted is set to the ±1 standard deviation level of its variability ENSO Mode [19] The positive ENSO (El Niño) mode slightly delays the onset of the monsoon on the western part of northern Australia while it expedites the onset in the eastern part as seen by the zero contour in Figure 9a. Its impact on the onset of the Australian summer monsoon is, in general, small except at the far eastern area of the continent. During El Niño events, the negative precipitation anomalies in the seasonal cycle are fortified because of the downward motion over northern Australia. There is no significant onset delay during big El Niño events, because by the normal monsoon onset time the Walker circulation anomalies over northern Australia weaken considerably. The termination time is also not seriously disturbed by the El Niño mode because the sign of the precipitation anomalies associated with the ENSO mode is the same as that of the seasonal cycle near the normal termination time. Thus the positive ENSO mode does not significantly alter the duration of the monsoon. [20] During the positive (El Niño) phase, precipitation over northern Australia is significantly reduced before and especially after the normal monsoon period. During the monsoon, however, the amount of precipitation does not change significantly maintaining a maximum near early February (see Figures 10 and 11). Only in the eastern part of the continent is the maximum precipitation delayed toward the end of February. [21] Generally, an opposite situation emerges during the negative (La Niña) phase (Figure 9b). In early December, the increased precipitation due to the upward motion over northern Australia favors an expedited onset. During weak La Niña events, the increased precipitation may not completely offset the negative anomalies in the seasonal cycle and henceforth may not significantly affect the onset time. During big La Niña events (amplitude 1 standard deviation), however, positive precipitation anomalies can start much earlier (as early as late November or early December) as suggested in Figure 9b [see also Troup, 1961]. The termination time is also strongly affected by the negative ENSO mode; the western side of the continent specifically experiences lingering precipitation. This is due to the upward motion over northern Australia associated with La Niña while the seasonal cycle is weak during the transition from positive to negative precipitation anomalies. Thus the negative ENSO mode means extended precipitation, specifically in the western part of the continent. [22] During the negative (La Niña) phase, northern Australia experiences increased precipitation well ahead of the normal onset time especially in the western area (see also 12 of 17

13 Table 2. Summary of the Contribution of Each Mode to the Precipitation Over the Eastern, Central and Western Parts, and the Entire Area of Northern Australia in Comparison With the Mean Seasonal Cycle a Region Mode Western Australia ( E) Central Australia ( E) Eastern Australia ( E) Northern Australia ( E) Premonsoon (02 22 Dec) Onset (27 Dec to 11 Jan) Active Termination (01 11 Mar) Postmonsoon (16 26 Mar) Mean Contributions Positive Mode Negative Mode Mean Contributions Positive Mode Negative Mode Early (16 26 Jan) Middle (31 Jan to 10 Feb) Late (15 25 Feb) Mean Contributions Positive Mode Negative Mode Mean Contributions Mean Variance Seasonal ENSO ETM MJO MJO Seasonal ENSO ETM MJO MJO Seasonal ENSO ETM MJO MJO Seasonal ENSO ETM MJO MJO a The first number in each column represents the mean contributions (mm day 1 ) over the specified interval; it represents the contributions by the positive mode with the amplitude at its one standard deviation level. For a negative mode, sign should be reversed. The bold-faced entries indicate that their magnitudes are greater than the corresponding contribution by the mean seasonal cycle. The second and third numbers in each column represent the acceleration (+) or delay ( ) of precipitation in number of days for the positive and negative modes, respectively. The last column shows the noncentered variance (variance with respect to zero instead of its mean). 13 of 17

14 Figure 15. The ±1 pentad averaged precipitation over northern Australia (solid curves) during the summer monsoon period (December March) in comparison with the reconstructed precipitation based on the first 5 modes (dotted curves) and the first 10 modes (dashed curves). Figure 10). Around the normal onset time, however, precipitation is reduced so that there is a local minimum precipitation anomaly. During the normal monsoon cycle, the amount of precipitation varies only slightly by the negative ENSO mode. As the termination time approaches, the western side of the continent has increased precipitation whereas the eastern side experiences decreased precipitation. During the postmonsoon period, northern Australia experiences significantly increased precipitation ENSO Transition Mode [23] The positive ENSO transition mode slightly decreases the precipitation amount during the premonsoon stage because atmospheric and oceanic conditions are similar to those of the El Niño mode. During the active monsoon period, this mode describes the transition of the atmospheric and oceanic conditions from a warm phase to a cold phase of ENSO. As a result, the onset of the Australian summer monsoon in the western and central parts of the continent is accelerated slightly (Figures 12a and 10). In the eastern part of the continent, however, monsoon onset is delayed because of the lingering presence of the El Niño condition in the western Pacific Ocean. Once the monsoon initiates in January, precipitation is slightly increased over northern Australia. This leads to a precipitation maximum in the middle of January instead of the middle of February (as observed in the seasonal cycle) on the northwestern side of Australia. During the termination stage, precipitation decreases and henceforth expedites the termination time, although the atmospheric and oceanic conditions in central and eastern Pacific remain similar to those of La Niña events. It appears that the Walker circulation is not well established during the transition stage resulting in no sign of strong upward motion over the Maritime continent. Generally, the duration of the monsoon is shortened slightly in the presence of the positive transition mode. [24] The impact of the transition mode on the onset time is not symmetric with respect to the phase of the mode, with negative phase being more significant in its impact than the positive phase. The negative transition mode at the 1 14 of 17

15 Table 3. Yearly Mean and Variance of ±1 Pentad Moving Averaged Precipitation Over Northern Australia and Those of the 5-Mode and 10-Mode Reconstructions a Mean Standard Deviation Year Observed 5-Mode 10-Mode Observed 5-Mode 10-Mode a Numbers in boldface represent greater than 1.0 mm d 1 difference in the mean and in the standard deviation. Year in the first column is from December of the previous year through March of the current year (e.g., 1980 equals December 1979 to March 1980). standard deviation level increases precipitation during the premonsoon stage (Figures 12b and 10). This effect is exceptionally strong in the eastern part of the continent. This increased precipitation is due to the La Niña like condition in the premonsoon stage. Around the normal onset time, however, an El Niño like condition sets in delaying the monsoon onset. This El Niño like condition decreases the amount of precipitation in the early lifecycle of the monsoon so that the precipitation is more skewed toward the latter half of the monsoon period. During the termination stage, precipitation is increased; as a result, the termination of the monsoon is delayed by 7 days. Thus the presence of the negative transition mode implies that precipitation is increased in the premonsoon stage while the main monsoon period is delayed and shortened over northern Australia MJO Modes [25] The last two modes representing the Madden-Julian oscillation (hereafter MJO-1 and MJO-2) exert a significant influence on the onset, termination, and the strength of the Australian monsoon (Figures 13 and 14). The positive modes of MJO-1 and MJO-2 decrease the amount of precipitation during the premonsoon stage (Figures 13a and 14a). Around the normal monsoon onset time, however, both modes fortify the positive precipitation anomalies in the seasonal cycle and henceforth expedite the monsoon onset. At 1 standard deviation level, each mode expedites the onset by a few days over northern Australia. The termination of the monsoon is also delayed over northern Australia; the delay is particularly significant in the eastern part of the continent, where the two modes, particularly the MJO-2 mode, exert a strong influence (Figure 10). Thus the positive MJO modes tend to increase the duration of the monsoon over northern Australia. [26] Both modes increase the precipitation in the early stage of the monsoon (Figures 13a and 14a). Then, the MJO-1 mode exhibits negative precipitation anomalies around the middle of February so that the normal precipitation maximum in mid-february decreases significantly. As a result, maximum precipitation shifts toward late January or early February. The MJO-2 mode exhibits negative precipitation in early February so that the precipitation anomalies show two peaks before and after early February. Near and shortly after the termination date, the precipitation increases because of the MJO modes specifically in eastern Australia. [27] The negative MJO modes increase precipitation during the premonsoon stage (Figures 13b and 14b). In the presence of strong negative MJO modes, December precipitation is above the summertime mean. At 1 standard deviation level, the MJO-1 mode results in positive precipitation anomalies 15 days before the onset time and the MJO-2 mode by more than 10 days on both sides of the continent. Around the normal onset time, both modes Table 4. Yearly Onset Date of the Australian Summer Monsoon From the ±1 Pentad Moving Averaged Precipitation Over Northern Australia and Those of the 5-Mode and 10-Mode Reconstructions a Onset Date Year Observed 5-Mode Difference 10-Mode Difference Average SD a The onset date (in days) is with respect to the mean onset date (5 January) in the seasonal cycle. The positive sign means an expedited onset, and the negative sign means a delayed onset by the designated number of days. The positive sign in the difference means that the onset in the reconstruction is earlier than in the observation, and the negative sign in the difference means that it is later by the designated numbers. The numbers in boldface represent a greater than 10-day difference between the onset in the observational data and that in the reconstruction. Year in the first column is from December of the previous year through March of the current year (e.g., 1980 equals December 1979 to March 1980). 15 of 17

16 Figure 16. Correlation and normalized RMS error between the ±1 pentad averaged precipitation over northern Australia and the reconstructed precipitation as a function of the number of CSEOF modes used for the reconstruction. exhibit negative precipitation anomalies resulting in a decrease of precipitation in the early stage of the monsoon. Thus the appearance of the main monsoon period is delayed by more than 10 days, particularly on the eastern side of the continent. The positive precipitation anomalies of these oscillation modes roughly coincide with the normal maximum precipitation so that the precipitation anomalies increase significantly in the latter part of the monsoon cycle. The termination of the monsoon is expedited by a few days with regard to both MJO-1 and MJO-2. Thus, in the presence of the negative MJO modes, the primary monsoon period is shorter than normal with later onset and earlier termination, and precipitation more skewed toward the later time of the monsoon. 6. Summary and Concluding Remarks [28] This study investigated how each physical mode of variability as represented by a CSEOF affects the evolution of the summer monsoon precipitation over northern Australia specifically the onset and termination times and the amount of precipitation. Figure 10 summarizes how each mode affects the precipitation over northern Australia. As can be seen, the contributions of individual modes to the precipitation over northern Australia are by no means uniform regionally and throughout the monsoon period (see also Table 2). It is shown that the precipitation changes introduced by individual modes fluctuate significantly over the monsoon period; their impacts vary significantly over the course of the monsoon. Also, the relative importance of individual modes is significantly different between the eastern and the western sides of the continent. [29] Table 2 summarizes detailed accounts of the contributions each mode makes over the seven different phases of the monsoon. Figure 11 describes the precipitation patterns in the presence of primary modes and contains essentially the same information as in Table 2. It is shown that the mean seasonal cycle is typically bigger than the contributions of other modes except when the former is small, i.e., during the onset and termination stages (see Table 2). During the onset and termination stages, the respective dates may change significantly in the presence of the primary modes, especially when they are negative. [30] Table 2 shows that the ENSO mode makes the biggest contribution to the overall mean and variance. While the positive ENSO (El Niño) mode does not seriously affect the onset and termination times, the negative ENSO (La Niña) mode significantly affects each particularly the termination time (see also Figure 11). The negative ENSO mode prolongs the duration of the Australian summer monsoon and the amount of precipitation increases, particularly in northwest Australia, during the normal cycle of the monsoon. [31] During the onset and termination stages, the two MJO modes also make a significant contribution (Table 2). Specifically, the role of the two MJO modes as the major factor for setting the onset, termination, and the monsoon duration is important in understanding the monsoon variability. The positive MJO modes prolong the monsoon duration while the negative modes shorten it; the effects, however, are not quite symmetric with a much more dramatic impact by the negative MJO modes. The overall change in the magnitude of precipitation by the two MJO modes is small compared to that of the ENSO or the ETM modes but the corresponding variances are comparable to each other. This is due to the fact that the MJO modes are oscillations and henceforth positive and negative anomalies tend to cancel each other yielding a small mean. [32] Figure 15 shows the ±1 pentad average of the observed precipitation over northern Australia against the five-mode reconstruction (dotted curves). The general evolution patterns match reasonably between the two time series; correlation between the two is 0.69 (0.62 with the raw data); 50% of the total variance is explained by the first five modes. This indicates that the evolution patterns of precipitation can be explained reasonably in terms of the five primary modes. Undoubtedly, however, some detailed evolution patterns were not captured in the five-mode reconstruction of the precipitation data notably in , , , , , and The total amount of precipitation is erroneous particularly in , , , and (see also Table 3). Onset and termination times were also erroneous in some years, notably in and (see also Table 4). This indicates that more than 5 modes are needed for a more accurate description of the monsoon precipitation. As the number of modes is increased, the reconstruction converges gradually to the observational data. Figure 16 summarizes how the correlation and normalized RMS error vary with the number of modes. Figure 15 also shows the 10-mode reconstruction (dashed curves), which is correlated at 0.85 with the ±1 pentad averaged observational data (0.80 with the raw data) with a normalized RMS error of 27% (36% with the raw data). Using the 10-mode reconstruction, the evolution of precipitation is generally improved (see Table 3) and the onset and the termination dates are also more accurate (Table 4). [33] The systematic accounting of how the primary physical mechanisms affect the precipitation in the present study seems to add a valuable insight into understanding the variability of the Australian summer monsoon. Further, 16 of 17

17 the decomposition of precipitation variability into a number of deterministic modes (CSEOFs) offers an intriguing possibility of forecasting the monsoon precipitation by predicting the amplitudes of these modes [Lim and Kim, 2006]. Specifically, if prediction had been carried out perfectly with the first 5 (10) CSEOFs, the prediction result would be consistent with the dotted (dashed) curves in Figure 15. [34] Acknowledgments. The authors are thankful for the useful and constructive comments from reviewers. We gratefully acknowledge the support by NSF (ATM ) for this research. References Davidson, N. E., and H. H. Hendon (1989), Downstream development in the Southern Hemisphere monsoon during FGGE/MONEX, Mon. Weather Rev., 117, Davidson, N. E., J. L. McBride, and B. J. McAvaney (1983), The onset of the Australian monsoon during winter MONEX: Synoptic aspects, Mon. Weather Rev., 111, Drosdowsky, W. (1996), Variability of the Australian summer monsoon at Darwin: , J. Clim., 9, Hendon, H. H., and B. Liebmann (1990), A composite study of onset of the Australian summer monsoon, J. Atmos. Sci., 47, Holland, G. J. (1986), Interannual variability of the Australian summer monsoon at Darwin: , Mon. Weather Rev., 114, Hung, C.-W., X. Liu, and M. Yanai (2004), Symmetry and asymmetry of the Asian and Australian summer monsoons, J. Clim., 17, Joseph, P. V., B. Liebmann, and H. H. Hendon (1991), Interannual variability of the Australian summer monsoon onset: Possible influence of Indian summer monsoon and El Niño, J. Clim., 4, Keenan, T. D., and L. R. Brody (1988), Synoptic scale modulation of convection during the Australian summer monsoon, Mon. Weather Rev., 116, Kullgren, K., and K.-Y. Kim (2006), Physical mechanisms of the Australian summer monsoon: 1. Seasonal cycle, J. Geophys. Res., 111, D20104, doi: /2005jd Lim, Y.-K., and K.-Y. Kim (2006), A new perspective on the climate prediction of Asian summer monsoon precipitation, J. Clim., 19, Madden, R. A., and P. R. Julian (1972), Description of a day oscillation in the zonal wind in the tropical Pacific, J. Atmos. Sci., 29, Madden, R. A., and P. R. Julian (1994), Observations of the day tropical oscillation: A review, Mon. Weather Rev., 122, McBride, J. L. (1987), The Australian summer monsoon, in Reviews of Monsoon Meteorology, edited by C. P. Chang and T. N. Krishnamurti, pp , Oxford Univ. Press, New York. Nicholls, N. (1984), A system for predicting the onset of the north Australian wet season, J. Clim., 4, Nicholls, N., J. L. McBride, and R. J. Ormerod (1982), On predicting the onset of the Australian wet season at Darwin, Mon. Weather Rev., 110, Seo, K.-H., and K.-Y. Kim (2003), Propagation and initiation mechanisms of the Madden-Julian oscillation, J. Geophys. Res., 108(D13), 4384, doi: /2002jd Troup, A. J. (1961), Variations in upper tropospheric flow associated with the onset of Australian summer monsoon, Indian J. Meteorol. Geophys., 12, Vincent, D. G. (1994), The South Pacific Convergence Zone (SPCZ): A review, Mon. Weather Rev., 122, Xie, P., and P. A. Arkin (1997), Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates and numerical model outputs, Bull. Am. Meteorol. Soc., 78, K.-O. Boo, Climate Research Laboratory, Meteorological Research Institute, Shindaebang-dong, Dongjak-gu, Seoul , Republic of Korea. B.-M. Kim and G.-H. Lim, School of Earth and Environmental Sciences, Seoul National University, Seoul , Republic of Korea. K.-Y. Kim, Environmental Forecast and Value-Oriented Research Services Inc., 6232 Whittondale Drive, Tallahassee, FL 32312, USA. (kwang56@gmail.com) K. Kullgren, Department of Meteorology, Florida State University, 404 Love Building/Meteorology 4520, Tallahassee, FL , USA. 17 of 17