This study investigates El Niño precursors in a high-resolution version of CCSM3.5. First, using an Empirical Orthogonal Function analysis of all non-ENSO tropical Pacific variability, we find that the Pacific Meridional Mode (PMM) acts as an ENSO trigger 7–9 months prior to large El Niño events in the model, which is consistent with previous model and observational studies. However, because not every PMM event triggers an ENSO event, we also find that PMM appears to be an effective trigger when the western-to-central Pacific is preconditioned (i.e., anomalously high sea surface heights or heat content). Second, this study looks at the contribution of western Pacific variability, namely westerly wind bursts (WWBs), as well as all other non-ENSO variability in the tropical Pacific. We find that the relative importance of low-frequency climate variability associated with PMM dominates over other non-ENSO variability between 15°N and 15°S, including high-frequency atmospheric variability and WWBs, in acting as a precursor to El Niño events.
 Precursors can be defined as any phenomenon that tends to occur prior to ENSO events, including variability on timescales of high-frequency atmospheric noise [Kirtman and Schopf, 1998; Kug et al., 2008, 2010], synoptic and intraseasonal westerly wind bursts [WWBs; McPhaden, 1999; Zhang and Gottschalk, 2002; Kug et al., 2008], Indian summer monsoons [Kirtman and Shukla, 2000], midlatitude atmospheric variability [Vimont et al., 2003a, 2003b], and low-frequency climate variability like the Pacific Meridional Mode (PMM) [Chang et al., 2007; Zhang et al., 2009]. WWBs, monsoonal variability, and PMM are sometimes considered ENSO triggers [Kessler et al., 1995; Zhang and Gottschalk, 2002; Kirtman and Shukla, 2000; Chang et al., 2007; Zhang et al., 2009]. These precursors are typically classified as part of the stochastic noise forcing of ENSO, because they are believed to be due to the internal variability of the system (e.g., WWBs); however, there is evidence that WWBs are, in part, ENSO state-dependent [Seiki and Takayabu, 2007; Tziperman and Yu, 2007].
 High-frequency precursors have been extensively studied in the literature. Enhanced atmospheric noise can add to the irregularity of ENSO events and modulate the decadal signal [Kirtman and Schopf, 1998]. Enhanced atmospheric noise variability in the western Pacific tends to lead El Niño by 7–10 months, whereas variability in the central Pacific is correlated contemporaneously with ENSO, supporting the theory that atmospheric noise is ENSO state dependent [Kug et al., 2008; Kirtman et al., 2005]. Kug et al.  show that low-frequency western Pacific zonal wind, high-frequency zonal wind variance, and equatorial sea surface height (SSH) correlate highly with ENSO at a 9 month lag, contemporaneously with each other, and are part of a coupled process over the western Pacific occurring prior to ENSO events. In late boreal spring in the western Pacific, WWBs tend to be more frequent during ENSO onset years [Seiki and Takayabu, 2007], and the strength of Madden-Julian Oscillation induced WWBs is linked to the strength of the subsequent El Niño [Hendon et al., 2007]. A modeling study by Lopez et al.  shows that the slow sea surface temperature (SST) state-dependent component of WWBs potentially has a stronger impact on ENSO variability than its state-independent counterpart.
 Precursors also include low-frequency coupled variability like PMM, which is associated with peak SST anomalies (SSTA) in boreal spring coupled with anomalous westerlies in the central Pacific subtropics [Chiang and Vimont, 2004]. Positive phase PMM tends to precede El Niño events and potentially acts as an ENSO trigger [Chang et al., 2007]. In this study, an Empirical Orthogonal Function (EOF) analysis and composites are used to investigate the dominant structure of atmospheric noise, which we define as all non-ENSO variability in the tropical Pacific and its relationship to the ENSO signal, and also identify the contribution of key precursors to this relationship.
2 Data and Analysis Methods
 The data used in this study are daily and monthly means of SST, zonal wind stress (τx), and SSH from an NCAR Community Climate System Model version 3.5 (CCSM3.5) current-day climate simulation named HRC06 and discussed in Kirtman et al. . The horizontal resolution of the ocean component is eddy-resolving at 0.1°, and the last 40 years of simulation output are considered here. Monthly anomalies are computed by removing the annual cycle, and daily anomalies are computed by removing the annual cycle at each calendar day.
 The τx anomaly noise (hereafter, atmospheric noise) is defined as linearly removing the contemporaneous ENSO signal from the full τx anomaly field. The ENSO signal is defined as the NINO3.4 index (SSTA averaged over 5°S–5°N, 170°W–120°W). Therefore, the atmospheric noise is found by
where α is the regression coefficient of the NINO3.4 index onto τx. It is important to note that the term “noise” may seem like a misnomer because it intuitively implies randomness. However, the way noise is defined here, it can have spatial and temporal statistics (i.e., phenomenological signatures) as well as include coupled variability (e.g., PMM) and high-frequency atmospheric variability. This residual definition is similar to the procedure used by Blanke et al.  and assumes that noise is any variability that is not linearly and contemporaneously related to NINO3.4. Previous studies, including Kug et al. , apply a band pass filter to the data to isolate the noise component; however, no such filtering is applied here. This point is important considering that low-frequency climate variability may contribute to the stochastic forcing and phase-locking of ENSO [Zhang et al., 2009; Chang et al., 2007]. This method differs from that by Chiang and Vimont , where the authors also linearly remove the ENSO signal but instead use the cold tongue index (SSTA averaged over 6°S–6°N, 180°–90°W) and from ensemble averaging including the interactive ensemble approach by Kirtman and Shukla .
 An EOF analysis is applied to isolate the dominant structure of the so-called atmospheric noise in the tropical Pacific region of 15°S–15°N, 160°E–80°W. It is important to note that the following results are insensitive to restricting the latitudinal extent of the domain anywhere between 5°S–5°N and 15°S–15°N. There is slight sensitivity to the longitudinal extent of the domain; however, the sensitivity does not change the major conclusions presented here. In fact, extending the domain, to include either Atlantic or Indian Ocean variability, yields very similar results. This suggests that the dominant non-ENSO variability in the tropics is in the Pacific Ocean but does not rule out the possible impact from the Atlantic or Indian Ocean [Ham et al., 2013]. The normalized time expansion coefficients of the first EOF mode (PC1) and NINO3.4 (Figure 1a) show a distinct lead/lag relationship with the strongest magnitude correlation occurring when NINO3.4 lags PC1 by 7 months with correlation coefficient −0.37 (Figure 1c).
 The timescale of the lead/lag relationship is interesting considering that Kug et al.  show that the variance of the atmospheric noise, which they define with a 2–180 day band pass filter and averaged over the equatorial Pacific waveguide tends to be stronger 7–10 months prior to an El Niño event. Despite the different definitions of atmospheric noise, both studies indicate the existence of a distinct relationship between atmospheric noise and the subsequent El Niño at a similar timescale. Interestingly, PC1 (red line) in Figure 1a suggests an underlying low-frequency signal of ENSO-like timescale that would not be included in the filtered noise component as in Kug et al. . Figure 1b shows enhanced off-equatorial atmospheric noise between 180°–150°W and 5°–15°N and resembles the anomalous zonal wind field associated with positive phase PMM [see Chiang and Vimont, 2004, Figure 1a] and the seasonal footprinting mechanism [see Vimont et al., 2003a, 2003b, Figure 3].
 To identify which frequencies are most important to the variability of EOF1, a low-pass filter is applied to the atmospheric noise data to remove all variability with periods less than 24 months, and an additional EOF analysis is computed. The spatial structure is similar (not shown), and PC1-low pass (Figure 1a; blue line) resembles a smoothed version of PC1 from the unfiltered case (Figure 1a; red line). The two PC1s are highly correlated (0.70). To further verify that higher frequencies are not significantly contributing to the variability of EOF1, the EOF analysis is repeated but after only allowing variability of periods less than 24 months (not shown). The lead/lag correlations between NINO3.4 and PC1-high pass are near zero and statistically insignificant for lags 0–16 months, and the spatial structure does not resemble Figure 1b. This filtering analysis indicates that high frequencies are not a main contributor to the spatial structure or lead/lag relationship shown in Figure 1.
 Composites are created to show the SSTA, atmospheric noise, and SSH anomalies (SSHA) structures in the tropical Pacific at lags −9, −7, and 0 months where lag 0 corresponds to the peak of a significant ENSO event. Significant ENSO events are defined as when NINO3.4 meets or exceeds 1.0°C in December. Ten El Niño years satisfy the definition. While the maximum correlation between PC1 and NINO3.4 is at lag −7 (Figure 1c), composites at lag −9 are included to show the state of the tropical Pacific at the onset of the peak lead/lag relationship. Composites of the 10 El Niño events are shown in Figure 2. SSTA and SSHA at lag 0 are as expected for peak El Niño conditions with strong positive anomalies extending from the equatorial eastern Pacific toward the dateline (Figure 2e). The composite of τx anomalies at lag 0 (not shown) shows strong (greater than 0.3 dyne/cm2) anomalous westerlies in the western-to-central equatorial Pacific; however, the anomalies are much weaker (less than 0.2 dyne/cm2) in the atmospheric noise composite at lag 0 (Figure 2f) thus confirming that linearly removing the ENSO signal did, in fact, remove the robust zonal wind signatures typical during El Niño events.
4 Central Pacific and PMM
 Figure 2a shows that warmest SSTA at lag −9 occur off of the equator with a structure that is consistent with positive phase PMM [see Chiang and Vimont, 2004, Figure 1a]. Considering that lag −9 is in March, results are consistent with the timing of peak PMM SSTA in boreal spring. Anomalous westerlies in the central Pacific subtropics are consistent with previous studies that show distinct anomalous southwesterly winds in this region during PMM (Figure 2b). Therefore, positive phase PMM is identified at lag −9. Also present at this time are weak cool SSTA extending from the eastern equatorial Pacific toward the dateline, a structure consistent with the decay phase of La Niña. Another key feature seen at lag −9 (Figure 2a) is the positive preconditioning of the western-to-central Pacific (i.e., high SSHA in the western-to-central equatorial Pacific). It is hypothesized that at this time, atmospheric noise triggers the Kelvin wave that sets up El Niño 2 months later at lag −7 (Figures 2c and 2d), which is when the maximum correlation between NINO3.4 and PC1 occurs. These results are similar to the findings from Vimont et al. [2003a, 2003b], which discuss the projection of τx onto the equatorial wave via the seasonal footprinting mechanism.
 A few studies [Chang et al., 2007; Zhang et al., 2009; Alexander et al., 2008] point out that PMM conditions tend to look like the optimal initial conditions for rapid ENSO onset described in Penland and Sardeshmukh, 1995. To confirm that PC1 is, in fact, PMM, PC1 is regressed onto SSTA noise and atmospheric noise (Figures 3a and 3b). The SSTA noise is defined in the same way as the atmospheric noise. Previous studies calculate PMM by computing a maximum covariance analysis between SST and horizontal winds after linearly removing the ENSO signal from each field. Therefore, our so-called noise components can easily be compared with these previous methods. The regression of SSTA noise onto PC1 (Figure 3a) closely resembles the SSTA composite at lag −9 (Figure 2a) in the subtropical eastern Pacific, which we identified as positive phase PMM. Additionally, there is some evidence of the nonlinear El Niño component in the eastern equatorial Pacific that is not removed by our linear approach. The off-equatorial, anomalous westerlies seen between 180° and 120°W in the PC1 regression (Figure 3b) closely resemble the atmospheric noise composite (Figure 2b), which were also identified as a PMM signature. Overall, the PC1 regressions show the PMM signatures that are also identified in the composites (Figures 2a and 2b) at lag −9. It is also worth noting that the increased atmospheric noise in the western Pacific in Figure 3b may not only be PMM state dependent but is also possibly an artifact of the seasonality of WWBs. Therefore, in the following section, we compare the relative importance of PMM and WWBs as ENSO precursors.
5 Western Pacific Variability and the ENSO Precursor
 The results thus far suggest an interesting question, namely does western Pacific variability contribute to the ENSO precursor shown in Figure 1? In other words, is the ENSO precursor dominated by low-frequency coupled variability associated with PMM or does western Pacific variability primarily associated with WWBs also contribute? Since we have stated that high frequencies are not a main contributor to the EOF1 variability, we can assume that the fast SST state-independent component of WWBs, as well as other high-frequency atmospheric variability, does not significantly contribute. However, since WWBs also include a slow SST state-dependent component [Seiki and Takayabu, 2007; Tziperman and Yu, 2007; Lopez et al., 2013], it is possible that some variability associated with WWBs remains in the low-pass filtered EOF case and may be contributing to the EOF1 variability. To show the relative importance of PMM and WWBs to the variability of ENSO precursors, τx anomalies are averaged over the regions of interest and correlated with PC1 of the unfiltered EOF at lags 0–16 months (Figure 4). For PMM, τx anomalies are averaged over the central Pacific subtropics (TAUX-PMM; 180°–160°W, 10°N–15°N) and for WWBs, over the western equatorial Pacific (TAUX-WWB; 130°E–160°E, 2.5°S–2.5°N). Figure 4b shows that TAUX-PMM and PC1 are highly correlated (0.87) and the correlation is maximized at lag 0. In comparison, Figure 4d shows that TAUX-WWB and PC1 are only modestly correlated at (0.34). For this reason, we can conclude that the relative importance of WWBs to the variability of ENSO precursors is less than PMM and that, overall, PMM dominates over high-frequency atmospheric variability as well as variability in the western Pacific in acting as a precursor to El Niño events. To be clear, we are not stating that western Pacific variability, namely WWBs, are not important to the triggering of ENSO events, only that the relative importance of WWBs is less than for PMM, according to the analysis of this model.
 First, we find that the dominant variability of tropical Pacific atmospheric noise in a high-resolution coupled model is PMM or low-frequency coupled variability, and acts as a trigger for ENSO 7–9 months prior to El Niño events. It is important to remember that the definition of atmospheric noise varies between studies and that here we include all non-ENSO variability. Because not every PMM event triggers an El Niño event, composite analysis suggests that PMM appears an effective trigger when the western-to-central equatorial Pacific is preconditioned.
 Second, after repeating the analysis with high- and low-pass filtered data, extending the domain to include other ocean basins, and isolating the variability in regions of interest, we are able to conclude that the relative importance of low-frequency coupled variability associated with PMM dominates over other variability between 15°N and 15°S, including high-frequency atmospheric variability and WWBs, in acting as a precursor to El Niño events.
 We thank Hosmay Lopez and two reviewers for helpful suggestions.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.