1. Introduction

Recently, Mo and Higgins (1998a, 1998b, referred to hereafter as MH98a and MH98b, respectively) examined the precipitation regimes over the western United States. They found that precipitation over California is linked to tropical convection in both interannual and intraseasonal bands. In the interannual band, precipitation is influenced by El Niño–Southern Oscillation (ENSO; Cayan and Webb 1992; Cayan and Redmond 1994; Schonher and Nicholson 1989; MH98a). In addition to ENSO, California rainfall is also modulated by tropical intraseasonal oscillations (ISOs). Based on persistent rainfall events in California, MH98a and MH98b found that wet episodes are likely to occur when enhanced convection associated with the tropical ISO reaches the central Pacific and dry episodes are favored when enhanced convection is located in the western Pacific.

The strongest mode associated with the tropical ISO is the Madden–Julian oscillation (MJO, Madden and Julian 1972, 1994) or 30–60 day oscillation (Lau and Chan 1985; Ghil and Mo 1991). For example, during the 1992/93 winter, the precipitation regimes over the western United States were regulated by the MJO. There is a good correspondence between the MJO in the Tropics and California rainfall (MH98b). In addition to the MJO related rainfall, California often experiences alternating wet and dry episodes with periods shorter than the MJO timescales. An example is given in Fig. 1, which shows 5-day running mean precipitation averaged over nine California stations during the 1996/97 winter. Five coastal stations (Brookings, Los Angeles, Pendleton, San Diego, and San Francisco) and four inland stations (Blue Cayan, Fresno, Stockton, and Thermal) contribute to the mean. Figure 1 indicates that there were four wet episodes between 15 November 1996 and 5 February 1997. These episodes were roughly 20 days apart with breaks in between. These alternating wet and dry events occur often in California during winter. Additional examples are given in Fig. 2, which shows daily precipitation averaged over a box centered in California (32.5°–40°N, 118°W to the coast) based on precipitation analysis by Higgins et al. (1996). Recurrent heavy rainfall events over California on submonthly scales were also observed during the winter of 1979/80 and 1985/86 (Fig. 2) and these events contributed to the American river flooding (Maurice Roos 1998, private communications).

Lau et al. (1994) and Ghil and Mo (1991) found a 20–30-day oscillation in addition to a 30–60 day oscillation in midlatitudes when they examined multiscale circulation modes in the global atmosphere. These are also preferred timescales for the tropical ISO. While there are many studies examining the impact of the MJO, the circulation and convection patterns associated with oscillations on the submonthly scales are less understood. Branstator (1987) reported a westward-traveling pattern in the Northern Hemisphere with a period of 25 days during the 1979/80 winter. Kushnir (1987) linked westward propagating waves in the North Pacific to the occurrence of blocking events. In the Southern Hemisphere, submonthly convective variations of the South Atlantic convergence zone (SACZ) were documented by Leibmann et al. (1999). In this paper, the life cycle of the 20–25-day oscillatory mode is examined. In addition to the MJO, the 20–25-day mode also has impact on California rainfall and it is related to tropical convection.

In addition to tropical forcing, one important feature related to California rainfall is dryness in the Pacific Northwest and rainfall deficiency over the subtropical eastern Pacific (MH98a). The area just north of the ITCZ plays an important role in regulating California rainfall during ENSO (MH98a). It is the area known to have strong tropical–extratropical interactions (Kiladis and Weickmann 1992a,b). Meridionally oriented cloud bands often extend from that region into midlatitudes. The role played by cloud bands in that area in the intraseasonal range will be examined.

While MH98a and MH98b recognized the linkages between tropical ISO and precipitation regimes over the western United States, they did not identify various oscillatory modes associated with California rainfall. This paper extends previous work to examine multiscale California rainfall variations on intraseasonal timescales and isolates the circulation and outgoing longwave radiation (OLR) anomalies (OLRA) patterns associated with dominant oscillatory modes using singular spectrum analysis (SSA; Vautard and Ghil 1989; Vautard et al. 1992). SSA identifies the quasi-oscillations in short noisy time series. There are two dominant modes of oscillation associated with California rainfall with periods near 36–40 and 20–25 days. They are referred to as the 40-day mode and the 22-day mode, respectively. After two modes are identified, their linkages to enhanced tropical convection are examined. The atmospheric conditions associated with these two modes are studied using the dataset from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis Climate Data Assimilation System (CDAS).

The datasets used in this paper are described in section 2. Results from SSA are given in section 3. The evolution of the 40-day mode and its linkages to the MJO are discussed in section 4. The evolution of the 22-day mode is discussed in section 5. The above analyses are based on California rainfall. It is unclear whether the 22-day mode is a local mode related only to California rainfall or a global mode having impact on California. Evidence that the 22-day mode has a tropical component is presented in section 6. Conclusions are given in section 7.

2. Data

The data used in this study are daily mean global gridded analyses from the NCEP–NCAR reanalysis CDAS for the period from 1979 to 1997 (Kalnay et al. 1996). The data are on a 2.5° lat × 2.5° long grid and 28 levels in the vertical. Daily averages of the National Oceanic and Atmospheric Administration (NOAA) satellite OLR field are used as a proxy for cloudiness. The OLR data (Liebmann and Smith 1996) started from June 1974. Because of a gap in 1978, the period from 1 January 1979 to 31 December 1997 was used. To obtain the ISO signal, OLRs were filtered using the minimum bias window developed by Papoulis (1973) to retain periods in the range of 10–90 days. This is referred to as the 10–90-day band. To study fluctuations associated with the 22-day mode, data also were filtered to retain periods in the range of 7–30 days. This is referred to as the 7–30 day band. Because California receives rainfall from November to March, the period from 1 November to 30 March was defined as the winter season.

Over the United States, daily observed precipitation derived from gridded hourly station data (Higgins et al. 1996) is used to obtain precipitation composites. The data are on a 2° lat × 2.5° long grid covering the period from 1979 to 1995. This is the same dataset used by MH98a and MH98b for their rainfall composites. Daily precipitation anomalies are defined as departures from the mean daily values for the entire period.

3. Multiscale variability

The California OLRA index was constructed by averaging OLRA over a box in California (32.5°–40°N, 118°–123°W). This index was used as a proxy for precipitation for the period from 1 October 1979 to 31 September 1997. Rainfall anomalies were not used to identify modes of oscillation but were used for verification. Rainfall does not follow a normal distribution and the dataset is too short to obtain a stable distribution function needed to calculate covariance matrix for SSA.

SSA was performed on the California OLRA index in the 10–90-day band to identify oscillatory modes. SSA is basically a statistical technique related to EOF analysis, but in the time space. SSA is more effective than conventional spectral analysis because SSA is not restricted to periodic cycles (Vautard and Ghil 1989; Vautard et al. 1992). Quasi-periodic signals appear as pairs of degenerate eigenmodes and their corresponding eigenfunctions in the time domain (T-EOFs) are in quadrature with each other. For details, readers are referred to Vautard and Ghil (1989). A window length of 61 days was used to highlight oscillations on intraseasonal timescales. Results are not sensitive to the particular window length used. The original time series can be projected onto T-EOFs to obtain principal components in the time domain (T-PCs). Since the SSA modes are not pure sines and cosines, the dominant periods of T-PCs are estimated using a Blackman–Tukey analysis with a bandwidth of 0.0074.

The first two eigenmodes are degenerate with eigenfunctions (T-EOFs); (Fig. 3a) in quadrature with each other so they represent the ISO with a period of about 36–40 days. Together they explain 27% of the variance in the 10–90-day band. The third and fourth eigenmodes (Fig. 3b) have a period of about 20–25 days. They explain about 25.6% of the variance in the 10–90-day band. These modes are referred to as the 40-day mode and the 22-day mode, respectively. The 40-day mode explains about 12% of the variance of the total unfiltered California OLRA index, and the 22-day mode explains only about 9%.

T-EOFs associated with the 40-day and the 22-day modes were selected. The California OLRA index in the 10–90-day band was projected onto these selected T-EOFs to obtain T-PCs. The time series corresponding to the 40-day and the 22-day modes can be reconstructed based on T-PCs and their corresponding T-EOFs (Vautard et al. 1992). The reconstructed time series for the 40-day and the 22-day modes were plotted together with the California OLRA index for selected winters in Fig. 4, which should be compared with rainfall averaged over the same box in California (Figs. 1 and 2). The OLRA index plotted is the 5-day running mean time series for the unfiltered data. These winters were selected because of their strong 22-day mode (1979/80, 1985/86, and 1996/97) or 40-day mode (1992/93)

There is a very good correspondence between the daily rainfall time series and the California OLRA index (thick dark dots). During these winters, large contributions were from the intrasesaonal band. Overall, negative OLRAs coincide with heavy rainfall periods and positive OLRAs coincide with rainfall breaks. This suggests that the OLRA index is a good proxy for rainfall. When the 22-day mode was strong, the OLRA index followed the reconstructed 22-day mode time series closely. They corresponded well with alternating wet and dry episodes over California (Figs. 1 and 2) during January and February 1980, February and March 1986, and November 1996–February 1997. This indicates that the 22-day mode is real and it is not the artifact of SSA. When the 22-day (crosses) and the 40-day modes (solid line) are in phase, the maxima and minima of the OLRA index are most likely to occur. When two modes are opposite in phase, the total contribution is smaller. For example, the dominant mode was the 40-day mode during the 1992/93 winter. However, there were periods that rainfall was delayed or weakened by the 22-day mode. The total variance of the unfiltered OLRA index explained by these two modes is not large (about 21%). However, for certain years, they were the dominant modes controlling the characteristics of California rainfall.

To examine properties of the 40-day and the 22-day modes, composites of 200-hPa streamfunction anomalies with zonal means removed and OLRAs in the 10–90-day band and daily rainfall were produced for winter. The standard deviation of the reconstructed time series corresponding to a given mode was computed. For a given mode, a positive event starts when the reconstructed time series for that mode is larger than 1.8 standard deviations. That date is counted as the onset date. The event ends when the time series is below 1.8 standard deviations. The negative event can be selected the same way but with a sign reversal. Lagged composites of variables were produced from 20 days before onset to 18 days after onset for both positive and negative events. Since composites for positive and negative anomalies are similar except for a sign reversal, the composite differences are presented. The statistical significance was assessed by assuming that anomalies obey a normal distribution and that each event is one degree of freedom. There are roughly 40–50 events forming each composite. Areas where values are statistically significant at the 95% level are shaded. Rainfall anomalies were not filtered and rainfall maps are used for verification, so no statistical significance is given. Events vary slightly and magnitudes of anomalies in the composites change, when the threshold varies from 1.2 to 2.0 standard deviations. However, conclusions reported here are not sensitive to the threshold used. The OLRA composites (not shown) were also verified using a pentad rainfall dataset analyzed the same way as Xie and Arkin (1996). All major features in the OLRA composites (Fig. 5 and later Fig. 8) were reproduced.

4. The 40-day mode and linkages to the MJO

Figure 5 shows the OLRA composite differences between negative and positive events in the 10–90-day band for winter. The corresponding daily rainfall composite differences are given in Fig. 6. Again, over the United States, there is a good correspondence between rainfall and OLRA differences. The composite starts from day −8 when negative OLRAs are located in the central Pacific with positive OLRAs in the western Pacific (Fig. 5a). There are weak negative rainfall anomalies centered in Oregon (Fig. 6a). Four days later (day −4), rainfall increases in California (Fig. 6b). At day −4 (Fig. 5b), the OLRA composite exhibits a three-cell pattern along western North America with positive anomalies located in the Pacific Northwest and Mexico and negative anomalies over California. The rainfall difference shows a dipole with rainfall in California and dryness in the Pacific Northwest (Fig. 6b). When enhanced convection (negative OLRAs) in the Tropics moves out of the central Pacific at day −4, it reappears in the Indian Ocean at day 0 (Fig. 5c). At onset (day 0), rainfall in California continues to intensify and extends northward. At day +2 (Fig. 5d), negative OLRAs are located in the Indian Ocean and at 20°N near the dateline with positive OLRAs located in the western Pacific. The three- cell pattern over western North America intensifies and rainfall over California reaches a maximum (Fig. 6d). At that time, dryness covers Alaska with wetness extending eastward to the southeastern United States. At day +6 (Fig. 6e), rainfall in California starts to diminish. When negative OLRAs reach the tropical western Pacific at day +10 (Fig. 5f), positive anomalies can be found in the tropical central-eastern Pacific (180°–120°W). The anomaly pattern (Fig. 5f) is similar to that of day −8 (Fig. 5a) with a sign reversal and it corresponds well with the rainfall signal reversal over the western region (Figs. 6a and 6f). That completes a half-cycle. Figure 5f also resembles the composite of positive MJO cases in the equatorial central Pacific by Lau and Chan (1985, Fig. 11b, therein). That supports the hypothesis that precipitation over the western region of North America is modulated by the MJO.

The evolution of the 40-day mode can be illustrated by the Hovmoeller diagrams of OLRA and 200-hPa streamfunction composites (Fig. 7). Even through composites are keyed to the 40-day mode associated with California rainfall, the OLRAs in the Tropics (Fig. 7a) show both standing and eastward propagating nature of tropical convection associated with the MJO similar to that reported by Lau and Chan (1985), Kiladis and Weickmann (1992a), and many others. Over the western region of North America, the OLRA composites show the familiar three-cell pattern observed by MH98a and anomalies have preferred geographic locations to amplify: 12°N and 38°N. The standing component of the oscillation dominates but weak northward propagation is also visible (Fig. 7b). Notice that the OLRA composites in the Tropics (Fig. 7a) are noisier than anomalies over the western region (Fig. 7b) because composites are based on the California OLRA index. The 200-hPa streamfunction anomaly composites exhibit many characteristics of the MJO (Weickmann et al. 1985). The response of the rotational flow in the Tropics is an anticyclonic couplet straddling the equatorial convection. This is illustrated by Fig. 7c. As enhanced convection moves eastward, 200-hPa streamfunction anomalies propagate eastward smoothly from the Indian ocean to the western Pacific and to the central Pacific. Results are in general agreement with the results of Kiladis and Weickmann (1992a). When enhanced convection moves to the central Pacific, the response in the Northern Hemisphere resembles the Pacific–North American teleconnection pattern (PNA, Wallace and Gutzler 1981; Weickman et al. 1985). The 200-hPa streamfunction anomalies over the western region of North America (Fig. 7d) show a standing oscillation of a dipole, which is a part of the PNA pattern.

The 40-day mode has slightly shorter periods (36–40 day) than the typical tropical MJO (40–48 days). This suggests that the 40-day mode may not be purely a response to the tropical MJO. The tropical MJO and the 40-day mode may or may not have the same origins but they can interact episodically. When the tropical MJO is strong, it creates a favorable situation for the 40-day mode to intensify.

5. The 22-day mode

Figures 8 and 9 show the OLRA differences between negative and positive events in the 10–90-day band and total rainfall differences associated with the 22-day mode for winter, respectively. Large rainfall anomalies are located along the west coast of North America. The OLRAs in the Tropics are weaker than the OLRA composites associated with the 40-day mode (Fig. 5). At day −4, negative OLRAs (Fig. 8a) extend from the eastern Pacific just north of the ITCZ northeastward to the Gulf of Mexico and positive OLRAs over California correspond well with dryness in that area (Fig. 9a). In the Tropics, there is a well-organized three-cell pattern with positive OLRAs centered at 5°N just west of the date line and negative OLRAs over the Philippines and in the central Pacific.

The dipole along the west coast of North America moves northward. At day −2, negative OLRAs strengthen as they shift to California, while positive OLRAs can be found in the Pacific Northwest (Fig. 8b). The same dipole pattern appears in the total rainfall difference map (Fig. 9b). At this time, weak negative anomalies start to appear in the eastern Pacific close to Central America. At day 0, a three-cell pattern appears along the west coast of North America (Fig. 8c) and rainfall over California intensifies. Notice that negative OLRAs start to appear at 15°N, 150°W at day −2, and together with the three-cell pattern, they form a quadrupole at day 0. At day 2, rainfall over California reaches a maximum. Rainfall also extends from the eastern Pacific through the Gulf of Mexico to the central and eastern United States. The three-cell pattern continues to move northward. At day 6, the OLRA pattern resembles that at day −4 with a sign reversal and this signals the completion of a half-cycle. Overall, there is a good correspondence between OLRAs and rainfall anomalies.

In the Tropics, the OLRAs show standing oscillation. However, westward propagation along the longitude bands just north of the equator (10°–20°N) is also visible. For example, positive OLRAs located just west of the date line (160°–180°E) at day −4 (Fig. 8a) shift westward and reach the western Pacific (140°E) at day 4 (Fig. 8e), while the negative OLRAs located near 10°–20°N, 160°W show standing oscillation. In the Southern Hemisphere, the OLRA composites from day −4 to onset show enhanced convection (negative OLRAs) in the area of the South Pacific convergence zone (SPCZ). The OLRA composites are keyed to the 22-day mode associated with California rainfall, so the tropical features are not as strong. However, it is clear that in addition to the northward propagation of OLRAs along the west coast of North America, the OLRAs show standing oscillation and westward propagation with the centers of action along 10°–20°N. The 22-day mode has distinct properties and it is not just a rapidly propagating MJO.

The 200-hPa streamfunction composites show westward propagating waves (Fig. 10) in the Northern Hemisphere. At day −4 (Fig. 10a), there is a wave train from Japan through the Gulf of Alaska, California to the eastern Pacific. The dipole with negative anomalies over California and positive anomalies in the eastern Pacific straddles the cloud bands over the eastern Pacific just north of the ITCZ (Fig. 8a). At day −2, a quadrupole starts to form in the North Pacific (Fig. 10b) corresponding well with the OLRA composite (Fig.8b). The whole pattern shifts westward. The negative anomalies originally located in California at day −4 move to the Pacific Northwest at day 0 and reach the Gulf of Alaska at day 6. The positive anomalies located near the Gulf of Alaska at day −4 weaken as they move to the eastern hemisphere near Japan at day 4. The flow pattern continues to move westward. By day 6, the anomaly pattern is similar to that of day −4 with a sign reversal and that signals the completion of a half-cycle. The composites presented here bear resemblances to the westward traveling pattern observed by Branstator (1987). In midlatitudes, zonal wavenumber 2 is the strongest. Because these composites are keyed to 22-day mode associated with California rainfall, many centers of maxima and minima are located along the West Coast.

The comparison between Figs. 8 and Fig. 10 shows consistency between OLRA and 200-hPa streamfunction composites. For example, over the western coast of North America, both show a three-cell pattern propagating northward. In the Tropics, the negative OLRAs in the western Pacific near 10°N, 160°E reach a minimum at day 6 (Fig. 8f). The streamfunction composite at day 6 shows a wave train from this area of enhanced convection through the North Pacific, the Gulf of Alaska, California, and to the eastern Pacific (Fig. 10f).

The 22-day mode can be summed up in Fig. 11, which shows the Hovmoeller diagrams of OLRAs and 200-hPa streamfunction composites in the 10–90-day band. In the Tropics, the OLRAs show both standing oscillation and westward propagation. OLRAs propagate along (10°–20°N) from the eastern Pacific just north of the ITCZ to the western Pacific near 120°E (Fig. 11a). Anomalies have fixed geographic locations to amplify:120°W, 160°W, and 140°E. The northward propagation of OLRAs along the west coast of North America is more smooth (Fig. 11b). The OLRAs start from 10°N and propagate northward to California. They also have preferred latitudes to amplify: 12°–15°N and 38°N.

The 200-hPa streamfunction composites also show anomalies propagating westward in midlatitudes. This can be demonstrated by the time–longitude plot of 200-hPa streamfunction anomalies averaged from 50° to 60°N, where the magnitudes of anomalies are large. Anomalies start from Europe near 60°E and propagate westward to 120°E (Fig. 11c). The dominant zonal wave number is wave 2 and the propagation completes a cycle in about 20–25 days. The 200-hPa streamfunction composites averaged over longitudes (118°–123°W) along the western region show anomalies propagating northward from the equator to California and to the Pacific Northwest (Fig. 11d). They have preferred latitudes to amplify: 3°N, 27°N, and 47°N.

6. The tropical 22-day mode

The above analyses are based on composites keyed to the California OLRA index. It is clear that the 40-day mode is associated with the MJO in the Tropics. The 22-day mode could be a local mode in the Pacific North American sector. It also could be a global mode having large impact on rainfall over the west coast of North America. If the 22-day mode is a global mode and has a tropical component, it should appear as a mode of large-scale variability of OLRAs in the Tropics.

Spatial EOF analysis was performed on OLRAs in the 10–90-day band as well as in the 7–30-day band for winter. To compare with the 22-day mode composites (Fig. 8), the domain includes the Pacific and the Indian Ocean. To keep the matrix small, the horizontal resolution was reduced to 5° × 5° and only data from December–February went into the EOF analysis. Anomalies were not normalized but a latitudinal cosine weighting factor was used in computing the covariance matrix. The first two EOFs in the spatial domain (S-EOFs) in the 7–30 day band (Fig. 12) is similar to the third and fourth S-EOFs in the 10–90-day band. They explain 4.6% and 3.9% of the total variance in the 10–90-day band and 4.2% and 3.8% of the variance in the 7–30-day band. The quadrature relationship suggests that they together represent an oscillation

S-EOF 1 (Fig. 11a) shows a four-cell pattern with alternating positive and negative loadings in the Tropics. The mean distance between centers of positive and negative loadings is smaller than the spatial distance of the MJO (Lau and Chan 1985). Negative loadings extend from convective area in the central Pacific to California. A weak dipole can be found along the west coast of North America with positive loadings at 10°–20°N and negative values over California. S-EOF 1 is similar to the OLRA composite at day −4 or day 6 keyed to the 22-day mode associated with California rainfall (Fig. 8f). S-EOF 2 is nearly in quadrature with S-EOF 1. It shows negative loadings in the central Pacific just north of the equator and alternating positive and negative loadings south of the equator. This pattern resembles the OLRA composite at day 0 (Fig. 8c) keyed to the 22-day mode associated with California rainfall.

Principal components in the spatial domain (S-PCs) were obtained by projecting OLRAs onto S-EOFs. SSA performed on both S-PC 3 and 4 from the 10–90-day band shows a leading mode of 40–48 days. The second pair has a period about 22–25 days. For 7–30-day-filtered S-PCs, the leading oscillatory modes for both S-PC 1 and S-PC 2 have a period about 20–22 days. The T-EOFs for the tropical 22-day mode obtained from 10–90-day filtered S-PCs 3 and 4 and the T-EOFs obtained from the 7–30-day-filtered S-PCs 1 and 2 are identical (Fig. 3c). They are similar to T-EOF 3 and 4 obtained from SSA analysis on the California OLRA index (Fig. 3b). S-PC 1 can be projected onto T-EOFs to obtain T-PCs and the time series corresponding to 22-day mode in the Tropics can be reconstructed based on T-EOFs (Fig. 3c) and their corresponding T-PCs. Composites can be obtained based on the reconstructed time series. The procedures are the same as before.

Figures 13 and 14 show, respectively, the OLRA and 200-hPa streamfunction composites in the 10–90-day band based on reconstructed time series corresponding to the 22-day mode in the Tropics. Composites start at day 0 so they are in phase with composites keyed to the 22-day mode associated with the California OLRA index (Fig. 8 and Fig. 10). For OLRA composites, both Fig. 8 and Fig. 13 show anomalies traveling from the eastern Pacific (120°–90°W) near 10°N to California to the Pacific Northwest. It takes about 10–12 days to complete a half-cycle. In the Tropics, there is a well-defined three-cell pattern extending from the eastern Pacific to the western Pacific. They show both standing oscillation and westward propagation. The main difference is that anomalies in the Tropics are stronger for composites keyed to the tropical mode and anomalies over the west coast are stronger for composites keyed to the 22-day mode associated with the California OLRA index. The pattern correlations for OLRA between the composites keyed to the 22-day mode in the Tropics and the composites based on the California OLRA index are between 0.52 to 0.61. Basically, they represent the same phenomenon.

The composites for 200-hPa streamfuncion anomalies keyed to the reconstructed time series (Fig. 14) for the tropical 22-day mode are also strikingly similar to anomalies keyed to the 22-day mode based on the California OLRA index (Fig. 10). The major difference is that anomalies over Europe are weaker for composites keyed to the tropical 22-day mode. Both Figs. 10 and Fig 14. show a westward propagating traveling pattern. Anomalies propagate from the eastern Pacific, through California, the Pacific Northwest, and the Gulf of Alaska to Japan. Large anomalies are located in the Western Hemisphere. The exact locations of maxima and minima may differ because of different indices used, but they clearly represent the same traveling pattern. Figures 10 and 14 can be compared with the traveling pattern of Branstator (1987, Fig. 4 therein). Figure 14b shows positive anomalies centered at 60°N, 170°W, and negative anomalies over the United States and in the subtropics at 25°N, 160°E similar to Branstator’s EOF 1 (his Fig. 4a). Figure 14e shows a PNA teleconnection pattern (Wallace and Gutzler 1981) similar to Branstator’s EOF 2 (his Fig. 4b). The above results suggest that the 22-day mode is a global mode, which influences rainfall over the western region of the United States. In the Northern Hemisphere, it resembles the large-amplitude traveling pattern described by Branstator (1987).

7. Summary and discussion

SSA is used to identify variability associated with California rainfall in the intraseasonal band. SSA provides a powerful tool for the examination of oscillatory behavior in a time series (Vautard and Ghil 1989). Results indicate the presence of two oscillatory modes, with periods between 36–40 and 20–25 days. These modes have very different properties and together, they describe major characteristics of winter California rainfall in the intraseasonal band.

The 40-day mode is related to the MJO in the Tropics. Composites keyed to the 40-day mode show that OLRAs propagate eastward from the western Pacific to the central-eastern Pacific. When the OLRAs diminish from the central Pacific, they reappear in the Indian Ocean. The standing and propagating features are similar to those of the tropical MJO (e.g., Kiladis and Weickmann 1992a; Lau and Chan 1985). In the western United States, a three-cell pattern of OLRAs starts to form about 4 days after enhanced convection reaches the central Pacific and rain starts in California

The 22-day mode is a global mode and is related to tropical convection. It is also the leading mode in the 7–30-day band. For many years, the strength of the 22-day mode exceeded the 40-day mode and the associated anomalies were responsible for alternating wet and dry episodes in California. Along the west coast of North America, the OLRAs propagate northward from the eastern Pacific to California to the Pacific Northwest. The 200-hPa streamfunction composites show waves propagating westward similar to those observed by Branstator (1987) and Kushnir (1987). These are not observed features of the MJO.

In the Tropics, the 22-day mode shows standing oscillation and westward propagation with centers of action located north of the equator at 10°–20°N. When enhanced convection is located in the central Pacific, clouds extend from the convective area to California. In the Tropics, oscillations have preferred geographic locations to amplify but these locations are different from the preferred locations for the MJO. The spatial scales of the waves are smaller than these associated with the MJO. Therefore, the 22-day mode is not a rapidly moving MJO. Ghil and Mo (1991) and Lau et al. (1994) in their studies of oscillations in the global atmosphere identified 30–60-day and 21–25-day oscillations as dominant modes in the extratropics and in the Tropics. However, modes in midlatitudes and in the Tropics may have different origins (Jin and Ghil 1990), but they do interact under favorable conditions. More work in this area is needed to understand the interactions between the tropical and extratropical modes.

For both modes, California rainfall is related to cloud bands in the eastern Pacific about 120°W just north of the ITCZ. The three-cell pattern with positive (negative) OLRAs in the Pacific Northwest and in the eastern Pacific and negative (positive) anomalies over California appears in decadal (Dettinger et al. 1998), interannual (MH98a), 40-day, and 22-day modes. The inverse relationship between California and the Pacific Northwest is supported by the moisture flux transport. When the southern branch is enhanced, California receives more moisture from the North Pacific, hence more rainfall. The situation reverses when the northern branch is active (MH98a). The downward branch of the Hadley cell associated with convection in the eastern Pacific just north of the ITCZ is located in California and that may explain the inverse relationship of rainfall centered at the eastern Pacific and California. The three-cell pattern is likely due to local influences.

It is interesting that the MJO and 22-day mode can both excite a wave train similar to the PNA pattern in the Northern Hemisphere. They appear in both interannual and intraseasonal bands. They can be forced by tropical convection but they can also occur independently (Jin and Ghil 1990). These patterns with geographically fixed and temporal recurrent anomalies seem to be preferred modes in the atmosphere.

Many aspects of the 22-day mode remain unexplained. They include the origins of the 22-day mode and its relationship to the MJO. These issues will be the focus of future studies.