Abstract

In past research the Southern Oscillation index has often been used as an indicator of the tropical Pacific climate, notably for El Niño and La Niña event occurrences. This study identifies calendar monthly teleconnection signals in central and eastern North American precipitation associated with an alternative tropical Pacific indicator, sea surface temperature anomaly (SSTA) patterns. Using an approximate 1° resolution set of monthly precipitation totals for 1950–92, the work identifies monthly teleconnection relationships and their intraseasonal evolution. This builds upon previous studies that were limited to seasonal timescales. Here, a unique two-way statistical analysis is used to delineate linear SSTA–precipitation teleconnection patterns. First, a principal component analysis (PCA) is performed on a monthly tropical Pacific SSTA dataset for 1950–92 to identify the coherent modes of variability. The principal component (PC) score time series representing the most significant modes of SSTA variability (see below) are then correlated on a calendar-monthly basis with station precipitation anomalies, yielding associated “correlation-based” precipitation coherencies. In the second approach, PCA is applied to the precipitation anomaly data for each calendar month. Then, the resulting “PC-based” precipitation coherencies most central to each of the major correlation-based precipitation regions are identified, and their associated PC score time series are subsequently correlated with the tropical Pacific SSTA grid-cell data, yielding correlation-based SSTA coherencies that are then compared (generally favorably) with their PC-based forerunners.

The three SSTA PC patterns used to seek teleconnection signals in central and eastern North American precipitation are the first unrotated PC (UPC1, emphasizing central tropical Pacific variability), and the first (VPC1, eastern tropical Pacific) and second (VPC2, western to north central tropical Pacific) Varimax-rotated PCs. The strongest such signals to emerge were for precipitation in the southeastern United States (positive association with UPC1 and VPC1 in November–March), Texas (positive association with UPC1 and VPC1 in November–March), the Great Lakes/Ohio River region (negative association with UPC1 and VPC1 in January–March), the southeastern United States (negative association with UPC1 and VPC1 in July–August), the southern Canadian prairie (negative association with UPC1 and VPC1 in November–January), and along a northern storm track (positive association with VPC2 in September–October). These results, derived from new datasets using a unique statistical approach, both broadly confirm and significantly clarify previous findings and present striking new associations.

1. Introduction

One of the major sources of interannual variability in worldwide climate comes from the strong coupling of the ocean and atmosphere across the equatorial Pacific in El Niño–Southern Oscillation (ENSO) events. The teleconnections of this phenomenon to worldwide climate have now been seriously researched for almost 30 years. Some of the earlier studies include those of Bjerknes (1966), who proposed that in the eastern and central Pacific a weakening in the equatorial surface easterlies caused an anomalous sea surface temperature (SST) warming, and Berlage (1966), who linked variations in surface pressure at Djakarta with SSTs at Puerto Chicama, Peru. Following these works, Bjerknes (1969) related the Southern Oscillation to equatorial Pacific SST warming by showing the latter’s impact on the intensity of the Walker circulation and midlatitude circulation in the Northern Hemisphere via westerly angular momentum transport by the Hadley cell. Wyrtki (1975) first linked El Niño and the Southern Oscillation from the oceanographic standpoint. He demonstrated how water piled up in the western equatorial Pacific by abnormally strong tropical easterlies in the central and western Pacific preceding an El Niño event may cause the eastward propagation of an internal equatorial Kelvin wave across the Pacific when those easterlies weaken and, subsequently, produce anomalous SST warming along the Ecuador–Peru coast. The above unusually strong equatorial easterlies reflected a peak in the pressure anomaly difference between Darwin, Australia, and Easter Island, a Southern Oscillation index used by Quinn (1974), who showed that a peak exceeding 13 mb led previous coastal El Niño events by 9–13 months. Later, Rasmusson and Carpenter (1982) fully documented the Pacific atmosphere–ocean characteristics preceding and accompanying El Niño events by compositing six different episodes from 1949 to 1975 to develop what is known as the canonical El Niño event. The foregoing seminal works stimulated the subsequent explosion of scientific interest in the coupled ENSO phenomenon (reviewed in, e.g., Philander 1990; Glantz et al. 1991), some of which is briefly discussed below.

Many studies that have sought to identify the worldwide climate anomalies associated with ENSO have used the standard, normalized, Tahiti-minus-Darwin “Southern Oscillation index” (SOI), the time series of which now extends back to 1882 (Ropelewski and Jones 1987). Some of the more notable SOI-based teleconnection studies include Ropelewski and Halpert (1986, 1987, 1989) and Halpert and Ropelewski (1992), which used a harmonic dial analysis of precipitation based on low- and high-index SOI (i.e., El Niño and La Niña) years to delineate associated worldwide precipitation and temperature patterns. Their North American precipitation results (Ropelewski and Halpert 1986, 1989) indicate a tendency toward wet conditions in the Great Basin region for April–October during the year of maximum sea surface temperature anomalies (SSTA) across the tropical Pacific basin and wet (dry) conditions over the southeastern United States and along the Gulf of Mexico coast during October–March (October–April) following warm (cold) SSTA initiation. Pisciottano et al. (1994) used the same technique to find ENSO teleconnections in a high-density rainfall network in Uruguay. In addition, Kiladis and Diaz (1989) composited seasonal worldwide temperature and precipitation anomalies based on variations in the SOI and eastern tropical Pacific SST. More recently, Richman et al. (1991) used Procrustes Target analysis and principal component analysis (PCA) to find monthly SOI-related North American precipitation patterns in the same high-resolution precipitation database used here (see section 2), building upon the semestral and seasonal results presented by Ropelewski and Halpert (1986, 1989), Kiladis and Diaz (1989), and others, while also identifying additional relationships. This study will contribute in the same vein, but use tropical Pacific SST data instead of the SOI.

With the coupled nature of ENSO, it is important that parameters associated with both El Niño and the Southern Oscillation continue to be examined to understand their respective influences on worldwide climate. The need for a balanced approach is underlined by the relatively moderate correlations between the SOI and SSTs over the entire tropical Pacific region affected by El Niño warming (Fig. 1). Since the strongest correlation in this region is only approximately −0.7, over 50% of the SST variance is not linearly associated with the SOI. Moreover, Deser and Wallace (1987) demonstrated that warm (cold) SSTAs off the Peruvian coast have occurred without anomalously high (low) Darwin sea level pressure anomalies, supporting similar previous conclusions by Streten (1983). Furthermore, other studies have shown that all El Niño events are not identical (e.g., Fu et al. 1986; Trenberth 1993), thereby encouraging the use of ENSO measures beyond the narrowly based SOI to characterize each event. Some authors have already used Pacific SSTs to find associated global teleconnections (e.g., Angell 1981; Pan and Oort 1983), while others have used the same parameters to focus on regional-scale climate, such as United States temperature (Angell and Korshover 1981), Australian rainfall (Streten 1983; Nicholls 1989), Brazilian rainfall (Ward and Folland 1991), and western United States streamflow (Kahya and Dracup 1994).

Fig. 1.

Correlation between monthly values of Tahiti–Darwin SOI and SSTAs for 1950–92. Correlation contours are drawn at intervals of 0.2, with negative contours dashed. SSTA data used are described in section 2a.

Fig. 1.

Correlation between monthly values of Tahiti–Darwin SOI and SSTAs for 1950–92. Correlation contours are drawn at intervals of 0.2, with negative contours dashed. SSTA data used are described in section 2a.

Consistent with this situation, the present study seeks to build upon current knowledge of ENSO teleconnections by focusing on linear relations between North American precipitation and tropical Pacific SSTAs. Unlike previous seasonal studies (e.g., Ropelewski and Halpert 1986; Kiladis and Diaz 1989), this work identifies monthly regional-scale patterns using a high-resolution (1° latitude–longitude) central and eastern North American precipitation database containing quality-controlled data for 766 stations on a nearly equal-area grid. Here, PCA is used to relate modes of tropical Pacific SSTAs and North American precipitation variability, using a unique two-way approach. Specifically, PCA-based spatial modes of SSTAs (precipitation) variability are correlated with station precipitation (grid-cell SST) anomalies for each calendar month. Since this research focuses on linear relationships between SSTAs and North American precipitation, both analyses have been performed for all years during 1950–92. The assumption that precipitation relations associated with the cold La Niña are the linear inverse of those associated with El Niño is supported by Wright et al. (1988) and Halpert and Ropelewski (1992).

2. Data

a. Sea surface temperatures

The individual monthly SSTA data used here consist of anomalies from calendar monthly means for 1950–92 for 2° × 2° latitude–longitude cells centered between 19°N and 19°S and from 139°E to the west coast of the Americas (Fig. 2a). During this time period, several separate El Niño (1951, 1953, 1957–58, 1963, 1972–73, 1976–77, 1982–83, and 1986–87) and La Niña (1950, 1955–56, 1964, 1970–71, 1973, 1975, and 1988) events occurred, according to the available catalogs (e.g., Fu et al. 1986; Quinn et al. 1987), thus providing an adequate sample of SSTA patterns associated with both types of events.

Fig. 2.

Locations of (a) grid-cell centers for SST data and (b) North American precipitation stations.

Fig. 2.

Locations of (a) grid-cell centers for SST data and (b) North American precipitation stations.

The 2° × 2° individual monthly SST values for the region in Fig. 2a were taken from a “reconstructed” SST dataset obtained from the Climate Prediction Center (CPC) of the National Oceanic and Atmospheric Administration and described in Smith et al. (1996). This SST set was recently computed by the CPC to provide physically realistic SST fields without erroneous extremes or abnormally stretched gradients that might result from data deficiencies in some areas. The data subset relevant to this study was reconstructed from a set of empirical orthogonal functions (EOFs, Davis 1976) extracted from the 1982–93 optimum-interpolation (OI) SST dataset of Reynolds and Smith (1994) for tropical Pacific SSTAs from 30°N to 30°S, with anomalies computed with respect to the calendar-monthly adjusted OI climatology of Reynolds and Smith (1995). Using least squares techniques, the first 24 of those EOF modes (accounting for 89% of the SSTA variance) were fit to the median 2° × 2° grid-cell SST values in the Comprehensive Ocean–Atmosphere Data Set (COADS, Slutz et al. 1985) for each month from 1950 to 1992 to yield smoothed SST values over the domain (Smith et al. 1996). The resulting anomaly data were then converted to full fields by adding the adjusted OI climatology. For this study, these full fields were converted to anomalies relative to 1950–92.

To ensure that the reconstructed nature of the CPC SST data did not adversely affect this research, the analyses described in section 3 were also performed using individual monthly SST data from a gridded dataset obtained from the CPC for 1950–87 (Halpert and Ropelewski 1989). This dataset was computed from the aforementioned COADS and also had a resolution of 2° × 2° before an averaging procedure reduced it to a 2° × 4° grid. The results obtained using this 1950–87 dataset were quite similar to those derived from the reconstructed set. Consequently, the reconstructed set was chosen in order to include in the analysis data from the warm events and cold event that occurred during 1988–92.

b. North American precipitation data

The individual monthly precipitation data used here for 1950–92 were totaled from a set of daily values for 766 sites spanning the eastern two-thirds of North America south of approximately 56°N (Fig. 2b). This dataset constitutes an extension (from the western edge of the Appalachian Mountains to the Atlantic coast, to all months of the year, and through 1992) of a previously developed May–August 1949–80 daily precipitation dataset whose construction is described in Richman and Lamb (1985, 1987). This expanded set was also used by Richman et al. (1991) and Gong and Richman (1995). The spatial resolution of the data is approximately 1° latitude–longitude with no missing data, due to the substitution of daily values from a nearby secondary station when observations from a primary station were unavailable. The longitudinal station separation is increased in the northern part of the domain to approximate an equal-area grid and thereby avoids problems associated with computing principal components for irregularly spaced data (Karl et al. 1982).

c. Southern Oscillation index

To help place the SSTA principal component patterns described below in perspective, time series of their scores were correlated with the standard Tahiti–Darwin SOI time series described earlier (Ropelewski and Jones 1987). The latter was obtained from the CPC for 1950–92.

3. Methods

To find the linear associations between monthly North American precipitation patterns and tropical Pacific SSTAs, this study has used single-field principal component analysis (described in, e.g., Richman 1986; Preisendorfer 1988). Coherent modes in each of the SSTA and precipitation datasets will be extracted and related to the other set in complementary analyses, as described below.

a. PCA of SSTAs for all calendar months combined

This exploratory portion of the analysis is illustrated in Fig. 3a. First, a PCA is performed on all calendar months of the SSTA dataset combined. All calendar months were used in the PCA in order to focus on a few robust SSTA modes common to all months. The SSTAs were analyzed in an S-mode (spatial) sense, with the unrotated principal components (UPCs) being derived from the inter-grid-cell correlation matrix. Several tests were applied to determine the number of UPCs containing nonrandom signals, including the scree test (Cattell 1966), the eigenvalue separation criterion (North et al. 1982), and the comparison of each UPC spatial loading pattern with its associated “point teleconnection pattern” in the raw data (Richman and Lamb 1985; Richman 1986). The point teleconnection patterns corresponding to each UPC were constructed by selecting the grid cell with the maximum absolute UPC loading value and linearly correlating the time series of SSTAs for that cell with the time series of SSTAs for all other grid cells in the domain. This permitted a comparison of the UPC loading patterns with the spatial variations in the correlation data from which they were derived, which contributed to the assessment of whether the UPCs represented true modes of variability. While the first UPC has been generally shown to have potential physical meaning (Richman 1987), this has not been the case with UPCs beyond the first mode. Thus, the UPCs so determined to contain some nonrandom signal were also rotated to the Varimax criterion (VPCs, Kaiser 1958), as principal component (PC) rotation has been demonstrated to yield spatial loading patterns that are more representative of the modes of variability in the data than UPC patterns beyond the first UPC (e.g., Horel 1981; Richman 1986). Following the computation of the UPCs and VPCs, the time series of UPC and VPC scores (individual-monthly resolution) for the modes considered to best capture key aspects of tropical Pacific SSTA variability were correlated with the North American station precipitation anomalies on a calendar-monthly basis. Use of the monthly timescale here was due to the month to month variability in the nature of precipitation, such as there being more regionalized convection in the spring and summer months (e.g., Easterling 1991). This yielded spatially coherent calendar monthly precipitation patterns linearly associated with each mode of SSTA variability (Fig. 3a, right-hand side).

Fig. 3.

Schematic diagrams illustrating the two-way analysis used in this study: (a) PCA of SSTA for all calendar months combined and correlation of scores with monthly station precipitation anomalies, and (b) PCA of monthly station precipitation anomalies and correlation with monthly grid-cell SSTAs.

Fig. 3.

Schematic diagrams illustrating the two-way analysis used in this study: (a) PCA of SSTA for all calendar months combined and correlation of scores with monthly station precipitation anomalies, and (b) PCA of monthly station precipitation anomalies and correlation with monthly grid-cell SSTAs.

b. PCA of monthly precipitation totals

To further qualify the North American precipitation patterns associated with tropical Pacific SSTA variability, additional analyses were performed using sets of rotated PCs extracted from the station precipitation anomalies for each calendar month, as illustrated in Fig. 3b. (A separate PCA was performed for each calendar month in order to identify monthly precipitation modes that could be compared to the results of the first analysis above.) These PCAs were also performed in an S-mode sense using the interstation correlation matrix (Fig. 3b, left-hand side), with each UPC set considered to contain a nonrandom signal being rotated to the oblique Harris–Kaiser (HK) B′B criterion (Harris and Kaiser 1964). The Harris–Kaiser B′B criterion (rather than Varimax) was used for these precipitation PCs (HKPCs) since it was found to yield more robust regions of precipitation coherence for the relatively large number (12–18 per calendar month) of UPCs rotated. The number of PCs rotated for each month was determined by a subjective test described in Richman et al. (1992a), in which the rotated spatial loading patterns based on several different truncation points were examined to see which set contained the largest and most robust regionalized modes. This resulted in the retention, for January–December, respectively, of 16, 16, 18, 13, 12, 14, 13, 13, 17, 16, 14, and 13 PCs. For individual calendar months, the HKPC score time series for the coherent region(s) most spatially central to each area of precipitation coherence delineated by the procedure in section 3a (Fig. 3a, lower right) was/were then linearly correlated with the corresponding SSTA grid-cell data to determine the monthly SSTA pattern(s) associated with each monthly precipitation HKPC. Unlike the above PCA of SSTAs, the precipitation UPCs from which the rotated (i.e., HK) PCs were derived were not used further, due to their less representative delineation of precipitation, as demonstrated by Richman and Lamb (1985).

As Fig. 3 shows, the analyses in sections 3a and 3b thus yielded two delineations of both spatial SSTA and precipitation anomaly coherence. For each parameter, one such delineation is “straight PC” based, while the other rests on a correlation analysis with a PC index of the second parameter. These coherencies are henceforth referred to as being “PC based” and “correlation based”, respectively. The PC-based precipitation coherencies (Fig. 3b, left-hand side) subjectively determined to be the most geographically coincident with each monthly correlation-based precipitation anomaly region (Fig. 3a, right-hand side) were identified as noted above, and their corresponding correlation-based monthly SSTA regions (Fig. 3b, right-hand side) were examined for similarities with the original PC-based SSTA coherent areas obtained for all calendar months combined (Fig. 3a, left-hand side). This two-way procedure ensured that the coherent precipitation regions identified in the exploratory analysis (Fig. 3a) were also supported in the quasi-confirmatory analysis (Fig. 3b), thus strengthening the links between tropical Pacific SSTAs and North American precipitation revealed by this study.

To further assess the robustness of the results presented below, statistical significance procedures were applied to both the correlation-based precipitation anomaly and SSTA coherency patterns yielded by the above procedures (Figs. 3a,b, respectively). For these purposes, a field significance test was applied to both types of correlation analysis. A test assessing the global field significance, as opposed to the local significance, was used in response to the findings of Livezey and Chen (1983), which show how field significance tests are not susceptible to errors that may affect local significance tests when the fields are spatially correlated. Full details on the test applied are provided in the appendix.

As previously mentioned in the introduction, using a linear correlation analysis extracts coherencies in which the observed relationships during cold SSTA events are similar in structure and opposite in sign to those based on warm events. Since the correlation analysis is applied to all years here and not just those in which a warm or cold event occurred, it is also assumed that the years during which no substantial SSTA event occurred do not influence the correlation results. To ensure that including these normal years did not influence the results, the correlations were recomputed using only years during which warm and cold SSTAs were present. This analysis yielded patterns (not shown) with structures nearly identical to those in section 4b. Thus, the normal years did not have a heavy influence on the results presented here.

Single-field PCA is only one of a number of methods that can be used to find coupled patterns in climate datasets. Other common techniques (reviewed in Bretherton et al. 1992) include combined-field PCA (CPCA), canonical correlation analysis (CCA), and singular value decomposition (SVD). Although CPCA, CCA, and SVD have been shown to be successful in identifying coupled modes of variability in two datasets (e.g., Barnett and Preisendorfer 1987; Wallace et al. 1992), single-field PCA was selected for this study in order to permit each branch of the analysis to focus on one dataset, identify coherent modes, and relate those modes to the other dataset without constraining the coupled modes to be orthogonal. Thus, “pure” modes of SSTA variability are used to find related precipitation patterns in the exploratory analysis, and pure modes of precipitation variability are used to find associated SSTA patterns in the quasi-confirmatory analysis.

4. Results and discussion

a. SSTA principal component analysis

Some basic statistics from the SSTA PCA are shown in Table 1. They suggested that the first 6 UPCs contained a nonrandom signal and, therefore, warranted further treatment, including through rotation to the Varimax criterion. While the North et al. (1982) eigenvalue separation criterion was satisfied for all 10 UPCs, the scree test (Cattell 1966; Richman and Lamb 1985) indicated a truncation between UPC6 and UPC7. This test uses the last significant drop in the plot of eigenvalue with increasing root number before leveling off as the indicator of modes containing a nonrandom signal. Truncation at this point is also supported by the correlations with SSTA point teleconnection patterns, as the values for UPC 1–6 are above +0.62, while those for UPC 7–10 are below +0.46. These 6 UPCs together explain 77.8% of the total domain SSTA variance.

Table 1.

Eigenvalues, percentage of domain variance explained, and correlations with point teleconnection patterns and the SOI for the first 10 UPCs, and the Varimax rotation of the first 6 of those UPCs.

Eigenvalues, percentage of domain variance explained, and correlations with point teleconnection patterns and the SOI for the first 10 UPCs, and the Varimax rotation of the first 6 of those UPCs.
Eigenvalues, percentage of domain variance explained, and correlations with point teleconnection patterns and the SOI for the first 10 UPCs, and the Varimax rotation of the first 6 of those UPCs.

The spatial loading patterns for the first 6 SSTA UPCs are presented in Fig. 4a, and their corresponding point teleconnection patterns appear in Fig. 4b. UPC1 (explaining ∼ 42% of the total domain variance) has the most potential for use as an index of El Niño and La Niña events. It emphasizes a broad tongue of SSTA variability centered on the equator and extending westward from the central Pacific to beyond the date line, with lesser variability in the eastern Pacific. This pattern closely resembles a composited warm SSTA pattern found to characterize the “transition” and “mature” phases of El Niño evolution of Rasmusson and Carpenter (1982, hereafter RC82). Hsuing and Newell (1983) also found a similar tropical Pacific SST mode for the first unrotated EOF of 1949–79 SSTA, although its tongue of maximum variability was centered approximately 5° south of the equator. While most of the meridional gradient at the western end of the warm SSTA tongue in RC82 lies east of the date line, in contrast to the present UPC1 loading pattern, which places this gradient west of 180°, the validity of UPC1 is strongly confirmed by its remarkably high correlation (+0.99) with its point teleconnection pattern. Further, the time series of UPC1 scores (Fig. 4c) also correlates well with the SOI (−0.71, Table 1), and almost all of its extremes coincide with documented El Niño and La Niña events (Kiladis and van Loon 1988). Thus, UPC1 will be examined further below for possible relationships with North American precipitation.

Fig. 4.

(a) Loading patterns for SSTA UPCs 1–6, with the percentage of total domain variance explained given in the lower-right corner. (b) Point teleconnection patterns for each of the UPCs in (a), with spatial correlation with loading patterns in (a) given in lower-right corner. Contour interval is 0.2 in (a) and (b), with negative contours being dashed. (c) Score time series for UPC1 with year 0 of El Niño (La Niña) events listed in Kiladis and van Loon (1988) denoted with an “E” (“L”).

Fig. 4.

(a) Loading patterns for SSTA UPCs 1–6, with the percentage of total domain variance explained given in the lower-right corner. (b) Point teleconnection patterns for each of the UPCs in (a), with spatial correlation with loading patterns in (a) given in lower-right corner. Contour interval is 0.2 in (a) and (b), with negative contours being dashed. (c) Score time series for UPC1 with year 0 of El Niño (La Niña) events listed in Kiladis and van Loon (1988) denoted with an “E” (“L”).

UPCs 2–6 have little potential for capturing major SSTA variations and are, therefore, not considered further. Their loading pattern morphologies tend to be similar to the domain-dependent sequence shown in Buell (1975, his Figs. 5 and 9; 1×2 rectangle) and Richman and Lamb (1985, their Fig. 6), and do not adequately represent any phase of El Niño in RC82. Although UPC2 was initially considered to have some promise in this regard because of the partial resemblance of its loading pattern (Fig. 4a) morphology to that of RC82’s “peak” El Niño phase composite (with an equatorially centered tongue contour structure in the eastern Pacific), the equatorial loading values are relatively low (mostly between ±0.4), indicating that the mode does not represent SSTA variations in that region well, especially compared to UPC1. Moreover, unlike the SSTA in the RC82 peak phase pattern, the highest loading values for UPC2 are centered in the north-central part of the domain. While a similar feature also characterized an SSTA mode identified by Ropelewski et al. (1992) via a rotated PCA (with only two PCs rotated) for the years 1982–87, it is better captured by VPC2 (see below). Concerning UPCs 3–6, none of their score time series correlate well with the SOI (−0.04 to +0.08), while the spatial correlations with their point teleconnection patterns (+0.62 to +0.69) are also relatively low compared to UPC1 (+0.99), as was the percentage of the total domain variance that they explained (4.3% to 7.4%).

Fig. 5.

(a) Same as in Fig. 4a but for SSTA VPCs 1–6. (b) Same as in Fig. 4b but for VPCs. (c) Same as in Fig. 4c but for VPC1. (d) Same as in Fig. 4c but for VPC2.

Fig. 5.

(a) Same as in Fig. 4a but for SSTA VPCs 1–6. (b) Same as in Fig. 4b but for VPCs. (c) Same as in Fig. 4c but for VPC1. (d) Same as in Fig. 4c but for VPC2.

Fig. 9.

As in Fig. 6 but for February–March and SSTA Varimax-rotated PC 2.

Fig. 9.

As in Fig. 6 but for February–March and SSTA Varimax-rotated PC 2.

Fig. 6.

Calendar-monthly correlation patterns between the time series of unrotated SSTA PC 1 (left column) and Varimax-rotated SSTA PC 1 (right column) scores and monthly North American station precipitation anomalies for November–March. Contour interval is 0.2, with negative values denoted by dashed lines and zero contour omitted. Field significance levels are listed near the right-hand side of each plot. Correlation values are shaded according to legend. See section 3 and Fig. 3 for further explanation.

Fig. 6.

Calendar-monthly correlation patterns between the time series of unrotated SSTA PC 1 (left column) and Varimax-rotated SSTA PC 1 (right column) scores and monthly North American station precipitation anomalies for November–March. Contour interval is 0.2, with negative values denoted by dashed lines and zero contour omitted. Field significance levels are listed near the right-hand side of each plot. Correlation values are shaded according to legend. See section 3 and Fig. 3 for further explanation.

The spatial loading patterns for the Varimax rotation of the first 6 SSTA UPCs are presented in Fig. 5a and their corresponding point teleconnection patterns in Fig. 5b. For each VPC, Table 1 gives the explained variance fraction and the correlations with its point teleconnection pattern and the SOI. The spatial loading pattern for VPC1 emphasizes an equatorially centered tongue of SSTA variability that spans 10°N/S and stretches westward from the coast of the Americas to nearly 160°E, with loadings greater than 0.8 extending to 160°W. This pattern matches very closely with its corresponding point teleconnection pattern (Fig. 5b, +0.99 correlation) and is strongly analogous to RC82’s transition El Niño phase, which depicts a warm pool off the Ecuador–Peruvian coast and an associated warm tongue stretching coherently to 160°W and then weakening by 160°E. This loading pattern also corresponds closely to the first nonseasonal EOF identified by the tropical Pacific SST EOF analysis of Weare et al. (1976). Additionally, VPC1’s score time series (Fig. 5c) shows maxima (minima) occurring during almost all of the documented El Niño (La Niña) events, as in the time series associated with SSTA UPC1, to which the VPC1 series is correlated relatively well (+0.80). The variance explained by VPC1 is also relatively high (29.1%) for a rotated PC, and its scores are moderately correlated with the SOI (−0.59). Because of these characteristics of VPC1, its associations with North American precipitation will be investigated below. Although VPC1 explains less domain variance and is more weakly correlated with the SOI than UPC1, VPC1 usefully complements UPC1 (which primarily reflects central tropical Pacific SSTA variability) by emphasizing the eastern tropical Pacific.

For several reasons, VPC2 was also considered to warrant further investigation in relation to North American precipitation. First, the morphology of its spatial loading pattern does not represent a standard El Niño phase as identified by RC82 —the loading pattern emphasizes SSTA variability between 10°N/S west of 170°W and in a strip extending northeastward from there to 10°–20°N/120°–160°W (Fig. 5a). This is reflected in the lack of extremes in the VPC2 score time series during documented El Niño/La Niña events (Fig. 5d) and the relatively low correlation with the SOI (−0.25, Table 1). However, as already intimated, Ropelewski et al. (1992) obtained a similar spatial loading pattern from a more limited rotated SSTA PCA (with only two PCs rotated) for 1982–87. Additionally, Table 1 shows that the VPC2 loading pattern correlates well (+0.93) with its teleconnection pattern in the raw SSTA data and explains a moderate amount of variance for a VPC (16.3%). VPC2 will therefore be used to delineate North American precipitation patterns that may accompany SST variations in an area that is henceforth termed the western to north central tropical Pacific, as it complements UPC1 and VPC1, which emphasize the tropical central and eastern Pacific, respectively.

VPCs 3–6 were not used further in this investigation. None of their loading patterns represent El Niño signals, as shown in RC82. Also, the time series of the scores for these PCs (not shown) are not well correlated with the SOI (−0.18 to 0.27, Table 1). Furthermore, they neither explain a large fraction of the total domain SSTA variance nor account for any sizable local SSTA variability, although the correlations between their loading patterns and corresponding point teleconnection patterns all have a magnitude ≥ 0.75 (Table 1).

b. Relation to North American precipitation

As explained in section 3, the relationship of North American precipitation to tropical Pacific SSTAs was investigated using two sets of analyses. First, as shown schematically in Fig. 3a, the North American station precipitation anomalies for individual months were correlated with the score time series of the three selected SSTA PCs (UPC1, VPC1, and VPC2) obtained for all calendar months combined to yield areas of precipitation coherence associated with the SSTA variability emphasized by each PC (Fig. 6; also see Figs. 9, 13, 16a, and 17a in this paper). In this regard, UPC1 reflects SSTA variation in the central tropical Pacific, VPC1 reflects that in the eastern tropical Pacific, and VPC2 reflects that in the western to north central tropical Pacific. The procedure was reversed in the second analysis (Fig. 3b), in which the precipitation coherence was delineated through an obliquely rotated PCA (yielding HKPCs) of monthly station precipitation totals. The score time series of the precipitation PCs that were most geographically coincident with the above correlation-based areas of precipitation coherence were then correlated with grid-cell SSTA values on a calendar-monthly basis to yield areas of SSTA variability associated with the HKPC-based coherent precipitation regions (Figs. 7, 8; also see Figs. 10, 11, 12, 14, 15, 16b, and 17b in this paper). The statistical field significance of the results of both analyses was tested using the method described in the appendix, with the percentage of simulations having worse results indicated on each panel of the above figures. A field significance value greater than 90% was considered to justify the presence of a monthly relationship between SSTA and precipitation. In addition, patterns having a structure similar to that from a previous or following month that was field significant are also presented as part of that relationship.

Fig. 13.

As in Fig. 9 but for September–October.

Fig. 13.

As in Fig. 9 but for September–October.

Fig. 16.

(a) As in Fig. 6 but for April SSTA VPC1. (b) As in Fig. 7 but for April mid-Atlantic coast precipitation.

Fig. 16.

(a) As in Fig. 6 but for April SSTA VPC1. (b) As in Fig. 7 but for April mid-Atlantic coast precipitation.

Fig. 17.

(a) As in Fig. 6 but for June SSTA VPC1. (b) As in Fig. 7 but for June southeastern United States precipitation.

Fig. 17.

(a) As in Fig. 6 but for June SSTA VPC1. (b) As in Fig. 7 but for June southeastern United States precipitation.

Fig. 7.

Calendar-monthly correlation patterns between HKPCs for the southeastern United States (November–March) and tropical Pacific SSTAs. Regions of HKPC loading values greater than 0.4 are shaded in the left column. Associated correlation-based SSTA coherency patterns appear in the right column, with spatial field significance level indicated in the upper-left corner. Correlations between precipitation HKPC scores and SSTA VPC2 (western to north central tropical Pacific), UPC1 (central Pacific), and VPC1 (eastern Pacific) scores are respectively indicated at bottom of each right-hand panel, with boldface (asterisk) indicating significance of 90% (95%). Contouring and shading are as in Fig. 6. See section 3 and Fig. 3 for further explanation.

Fig. 7.

Calendar-monthly correlation patterns between HKPCs for the southeastern United States (November–March) and tropical Pacific SSTAs. Regions of HKPC loading values greater than 0.4 are shaded in the left column. Associated correlation-based SSTA coherency patterns appear in the right column, with spatial field significance level indicated in the upper-left corner. Correlations between precipitation HKPC scores and SSTA VPC2 (western to north central tropical Pacific), UPC1 (central Pacific), and VPC1 (eastern Pacific) scores are respectively indicated at bottom of each right-hand panel, with boldface (asterisk) indicating significance of 90% (95%). Contouring and shading are as in Fig. 6. See section 3 and Fig. 3 for further explanation.

Fig. 8.

As in Fig. 7 but for Texas (November–March) precipitation.

Fig. 8.

As in Fig. 7 but for Texas (November–March) precipitation.

Fig. 10.

As in Fig. 7 but for Great Lakes–Ohio River Valley (January–March) precipitation.

Fig. 10.

As in Fig. 7 but for Great Lakes–Ohio River Valley (January–March) precipitation.

Fig. 11.

As in Fig. 7 but for southeastern United States (July–August) precipitation.

Fig. 11.

As in Fig. 7 but for southeastern United States (July–August) precipitation.

Fig. 12.

As in Fig. 7 but for southern Canadian prairies (November–January) precipitation.

Fig. 12.

As in Fig. 7 but for southern Canadian prairies (November–January) precipitation.

Fig. 14.

As in Fig. 7 but for precipitation along a northern storm track (September–October).

Fig. 14.

As in Fig. 7 but for precipitation along a northern storm track (September–October).

Fig. 15.

As in Fig. 7 but for central Plains March precipitation.

Fig. 15.

As in Fig. 7 but for central Plains March precipitation.

1) Associations of at least 2-months duration

The above procedures suggested six major associations between North American precipitation and tropical Pacific SSTAs having significant spatial patterns in at least 2 consecutive months: 1) the southeastern United States (November–March), 2) Texas (November–March), 3) the Great Lakes–Ohio River Valley (January–March), 4) the southeastern United States (July–August), 5) the southern Canadian prairies (November–January), and 6) a northern storm track (September–October). The evidence for these associations is now summarized.

(i) Southeastern United States (November–March, UPC1 and VPC1)

Figures 6 and 7 suggest that precipitation in the southeastern United States is positively related to SSTAs in the central and eastern tropical Pacific for November–March. The patterns of correlation between monthly station precipitation anomalies and both SSTA UPC1 (which emphasizes the central tropical Pacific) and VPC1 (which emphasizes the eastern tropical Pacific) exhibit positive coherencies stretching along the Gulf of Mexico and the southern Atlantic coasts in November (Fig. 6a) and January–March (Figs. 6c–e), and from eastern Texas to Tennessee and Kentucky in December (Fig. 6b). The field significances for each of these patterns are consistently above 95%, with the only exception being December for UPC1 (90.0%). Additionally, for November and January–March (but not December), the PCA of monthly precipitation anomalies delineate coherencies (Figs. 7a,c–e, left-hand side) with similar morphologies to the patterns depicted in Figs. 6a,c–e in the southeastern United States. Each of these precipitation HKPCs are positively related to equatorially centered, correlation-based SSTA coherencies in the central and eastern tropical Pacific with field significances greater than 95%. Moreover, the correlations of their associated score time series with the SSTA UPC1 and VPC1 scores (hereafter termed “interscore correlations”) range from +0.27 to +0.51 and are significant at least at 90% in every case. The lack of a December precipitation–SSTA linkage is also shown by the relatively small-magnitude, correlation-based SSTA coherencies associated with a December precipitation HKPC in northern Florida (Fig. 7b).

These results, including the relative absence of SSTA-associated precipitation coherence for December, are similar to the simultaneous SOI–precipitation findings of Richman et al. (1991). They also support monthly relationships for Kiladis and Diaz’s (1989) identification of wet (dry) conditions for the December–February following the appearance of strong positive (negative) SSTAs in the central and eastern tropical Pacific. Additionally, the present study suggests associations in November and March, which do not emerge from their seasonal results for the preceding September–November and following March–May. Furthermore, the present study is not only consistent with Ropelewski and Halpert’s (1986, 1989) findings that wet (dry) conditions prevailed in this region during the October–March (October–April) following the initiation of strong positive (negative) central-eastern tropical Pacific SSTAs, but also extends the precipitation signal farther north along the Atlantic coast and demonstrates that the strongest monthly associations occur in November and January–March, with November’s region stretching farther inland than that for the latter 3 months.

(ii) Texas (November–March, UPC1 and VPC1)

Texas precipitation during November–March is also positively related to central and eastern tropical Pacific SSTAs (Figs. 6 and 8). The monthly precipitation coherence patterns over Texas associated with UPC1 (central Pacific SSTAs) and VPC1 (eastern Pacific SSTAs) are remarkably identical and are separated (November excepted) from the southeastern United States precipitation coherence areas discussed above in every month (Figs. 6a–e). After spanning much of the state in November, the UPC1- and VPC1-related areas of precipitation coherence are largely confined to southwestern and western Texas for December–February and then extend over central and northern Texas in March, with each month having a field significance above 95%, except for the December UPC1. Also supporting these Texas relations is the monthly precipitation PCA, which identified separate monthly coherencies over Texas for November–March (Fig. 8). The November, December, and February precipitation HKPCs (Figs. 8a,b,d) are associated with positive equatorially centered tongues of correlation-based SSTA regions in the areas of the SSTA UPC1 and VPC1 loading patterns (with interscore SSTA UPC1 and VPC1 correlations from +0.25 to +0.37) with field significances greater than 90%. The Texas precipitation HKPCs identified for January and March (Figs. 8c,e) are also linked with positive correlation-based SSTA regions, but with slightly lower field significances (85.4% and 88.5%, respectively) and interscore correlations with UPC1 and VPC1 (+0.22 to +0.26).

Ropelewski and Halpert (1986) were unable to isolate a low SOI (i.e., El Niño) signal in Texas precipitation (separate from the relation over the southeastern United States) using their relatively coarse spatial (only ∼10 Texas stations) and temporal (October–March semester) resolutions. However, within their Gulf and Mexican region, Ropelewski and Halpert (1989) indicated dry conditions during high SOI (i.e., La Niña) years over much of Texas, which is supported by the present study. Furthermore, the results presented here not only broadly confirm Kiladis and Diaz’s (1989) finding of wet (dry) El Niño (La Niña) Texas conditions for the September–November and December–February seasons following event initiation, but also suggest that the areas of strongest precipitation coherence associated with SSTAs vary during November–March. Such a specification could not be resolved using the comparatively coarse (seasonal, 3 widely dispersed stations) resolution in that study. This intraseasonal spatial evolution also did not emerge from the analyses of Richman et al. (1991), in which the relation between low SOI values (i.e., El Niño) and positive Texas precipitation anomalies was confined to the eastern part of the state and, again unlike the present results, was connected to the area of coherent precipitation anomalies in the southeastern United States.

(iii) Great Lakes/Ohio River Valley (January–March, UPC1, VPC1, and VPC2)

Figures 6, 9, and 10 indicate that precipitation in large areas of the Great Lakes–Ohio River Valley region is negatively related to tropical Pacific SST during January–March. In January (Fig. 6c), SSTA VPC1 (eastern Pacific) is negatively correlated with a precipitation coherency from Tennessee to the Great Lakes, as is SSTA UPC1 (central Pacific), although its relation is largely confined to Kentucky–Tennessee and southwestward to northern Louisiana. While SSTA VPC2 (western to north central tropical Pacific) is not significantly associated with January precipitation (not shown), it is negatively linked in February (Fig. 9a), as is UPC1 (Fig. 6d), with spatially extensive precipitation coherencies eastward of Arkansas to Minnesota having a field significance of 100%. The core of these regions is also negatively related to VPC1 in February (Fig. 6d). In March, the sizes of the coherencies associated with all three SSTA PCs are decreased (Figs. 6e and 9b), but are still continuous with respect to the previous 2 months. Interestingly, this negative January–March relationship occurs in the same season, and is as strong, as the previously discussed positive relationship in the southeastern United States. Further support is given by the monthly PCA of precipitation anomalies, which identified coherent regions coinciding with the negative correlation-based areas in Figs. 6c–e, as shown in Fig. 10 (left-hand side). The January and February precipitation HKPCs are most strongly related to SSTAs in the central and eastern tropical Pacific, as their interscore SSTA UPC1 and VPC1 correlations range from −0.25 to −0.42 and their associated correlation-based SSTA regions (with field significances greater than 95%) lie within the loading patterns of those SSTA PCs. The two March precipitation HKPCs, however, are related to SSTAs in the central and western to north central tropical Pacific, with interscore correlations with SSTA UPC1 and VPC2 from −0.26 to −0.31 and associated SSTA-based coherencies (with field significances exceeding 90%) spanning the area covered by the loading patterns for those SSTA modes (Figs. 4a and 5a).

In contrast to the above results, Ropelewski and Halpert (1986, 1989) were unable to detect any relationship between Great Lakes–Ohio River Valley precipitation and the tropical Pacific in their harmonic dial analysis. However, a negative association between tropical Pacific SSTAs and precipitation in the above region was obtained by Kiladis and Diaz (1989) for 6 stations extending from Indiana–Kentucky to just north of Lake Erie for RC82’s mature December–February composite period. The present results build upon this seasonal relationship by suggesting that January–March has the strongest relations and documenting the precipitation regions most exhibiting monthly associations with SSTAs in three different regions of the tropical Pacific. The monthly SSTA-related precipitation evolution presented here is also broadly supported by the SOI-related study of Richman et al. (1991).

(iv) Southeastern United States (July–August, UPC1 and VPC1)

Figure 11 suggests that July–August precipitation in the southeastern United States is negatively related to central and eastern tropical Pacific SSTAs. Although only discontinuous, negative, correlation-based precipitation coherencies are associated with SSTA UPC1 (central Pacific) and VPC1 (eastern Pacific) in July and August (not shown), stronger support is given by the monthly PCA of station precipitation anomalies (Fig. 11). The precipitation PCA identified HKPC coherencies along the mid-Atlantic coast in July and across Arkansas–Tennessee in August, each of which was negatively related to central and eastern tropical Pacific SSTAs with field significances greater than 95%. In addition, the interscore correlations with SSTA UPC1 and VPC1 range from −0.25 to −0.48 and are significant at the 95% level in each case, except for VPC1 in July.

This July–August relationship was not identified by Ropelewski and Halpert (1986, 1989) in their harmonic dial analysis. However, Kiladis and Diaz (1989) detected dry (wet) El Niño (La Niña) conditions for 3 Atlantic coastal stations (one each in Massachusetts, Virginia, and South Carolina) for June–August following the appearance of the respective SSTAs in the eastern tropical Pacific. This study confirms the relation for the southernmost 2 of those stations for July and also extends the relationship southwest into Alabama–Mississippi–Arkansas for August. The July–August SOI-based results of Richman et al. (1991) also generally agree with the SST-based connections presented here.

(v) Southern Canadian prairies (November–January, UPC1 and VPC1)

November–January precipitation in southern Saskatchewan and Manitoba is negatively related to central and eastern tropical Pacific SSTAs (Figs. 6 and 12). The correlation-based monthly precipitation anomaly coherencies associated with SSTA UPC1 (central Pacific) and SSTA VPC1 (eastern Pacific) are generally similar during these months, except for the November extension into Montana and North Dakota related to SSTA VPC1 (Figs. 6a–c). The regions linked to both SSTA PCs are slightly discontinuous in December, but exhibit substantial month to month continuity during this period and also lie along a January storm track identified by Whitaker and Horn (1984). Additionally, the very similar monthly PC-based precipitation anomaly coherencies in this region for November–January are shown to be negatively associated with central and eastern tropical Pacific SSTAs with field significances from 85%–90% (Fig. 12). The interscore correlation magnitudes involved are particularly modest for SSTA UPC1 for all 3 months (−0.22 to −0.25), but are stronger for SSTA VPC1 in November (−0.34 correlation).

This monthly SSTA–precipitation association was not identified in the aforementioned seasonal studies of Ropelewski and Halpert (1986, 1989) and Kiladis and Diaz (1989). Additionally, although Richman et al. (1991) used the same precipitation dataset as the present study, their SOI-based results did not feature this relationship, except for just north of North Dakota in November.

(vi) Northern storm track (September–October, VPC2)

According to Figs. 13 and 14, September–October precipitation along the northern United States is positively related to western to north central tropical Pacific SSTAs (i.e., VPC2). The correlation map linking September precipitation anomalies and SSTA VPC2 is characterized by a positive coherency arcing from Manitoba to South Dakota to the Great Lakes, with a field significance of 96.0% (Fig. 13a). This region coincides with a storm track identified by Whitaker and Horn (1984) and Richman et al. (1992b). For October, positive correlation-based precipitation coherencies lie over the Great Lakes and New Brunswick–Newfoundland–Maine (Fig. 13b) and are continuous with respect to the September map. Additional support is given by a discontinuous September PC-based coherency stretching from extreme southern Alberta into the Great Lakes, which correlates highest with western to north central tropical Pacific SSTAs (+0.37 correlation with SSTA VPC2) with field significance of 92.7% (Fig. 14a). Similarly, although the field significance of the correlations between October precipitation over the whole domain and SSTA VPC2 is low (68.6%, Fig. 13b), an October precipitation HKPC over New Brunswick–Newfoundland–Maine is positively correlated with western tropical to mid-North Pacific SSTAs (+0.39 interscore correlation with SSTA VPC2) with field significance of 90.6% (Fig. 14b). However, an October precipitation HKPC mode for the Great Lakes region (not shown) is not significantly linked to tropical Pacific SSTAs.

This connection between SSTAs and September–October precipitation was not part of any of the seasonal relations presented in Ropelewski and Halpert (1986, 1989) or Kiladis and Diaz (1989). Richman et al. (1991) did present a similar October SOI–precipitation linkage in New Brunswick–Newfoundland, but did not identify the above September relation.

2) Single-month associations

All of the relationships discussed thus far have involved significant spatial patterns in at least 2 consecutive calendar months. However, the analyses also revealed other strong associations between North American precipitation and tropical Pacific SSTAs that characterized single, discrete calendar months. The shorter duration of these relationships clearly raises questions about their robustness and physical plausibility since tropical Pacific SSTA fields generally do not change appreciably from month to month (as mentioned in section 3 and shown in Halpert and Ropelewski 1989). However, some of these 1-month relationships have strong spatial coherence and relatively high correlation magnitudes, thereby inviting initial identification as a prelude to further investigation. The most striking 1-month associations are listed below, along with a few descriptive comments pertaining to each. Except for the first two subsections below, the results presented had not been reflected by the relations obtained by previous authors.

(i) Central Plains (March, UPC1 and VPC1)

March precipitation in the central Plains is positively related to central and eastern tropical Pacific SSTAs (Figs. 6e and 16). The correlation patterns between March precipitation anomalies and SSTA UPC1/VPC1 score time series both exhibit sizable positive coherencies stretching from northern Texas to Nebraska (Fig. 6e) with field significances over 99%. Furthermore, the PCA of March precipitation (Fig. 15, left-hand side) identified two modes with HKPC loading patterns lying completely or partially within the above correlation-based region in Fig. 6e. These precipitation modes are associated with positive correlation-based SSTA coherencies (with field significances of 99.8% and 96.0%) centered on the equator and stretching from the Central and South American coast westward to the international date line (Fig. 15), which is reflected in their moderate interscore correlations with SSTA UPC1 (+0.43 and +0.30) and VPC1 (+0.52 and +0.38). Interestingly, this relation occurs in the last month of the aforementioned Texas signal [see section 4b(1)ii], perhaps suggesting the latter extends northward into the plains at the end of its evolution. This March central Plains association geographically coincides with 6 stations scattered through Oklahoma–Nebraska–Iowa that Kiladis and Diaz (1989) found to have wet (dry) El Niño (La Niña) conditions for the March–May season immediately following RC82’s Mature (December–February) El Niño composite, although the present study does not support any monthly relations in this region for April (Fig. 16a) or May (not shown).

(ii) Mid-Atlantic coast and New England (April, VPC1)

Figure 16a indicates that April precipitation from Maine to North Carolina and west to Missouri is positively related to the SSTA mode captured by VPC1 (eastern Pacific), with a field significance of 93.7%. In support of this finding, the April precipitation PCA identified an HKPC (Fig. 16b, left-hand side) with similar morphology to the eastern portion of the above correlation-based region and also to an April storm track identified by Whitaker and Horn (1984). Moreover, its associated SSTA coherence region (Fig. 16b, right-hand side) closely resembles the loading pattern of SSTA VPC1 (Fig. 4a), has a corresponding HKPC–VPC1 interscore correlation of +0.49, and has a spatial field significance of 100%. Interestingly, this relationship immediately follows the positive SSTA association with the southeastern United States precipitation for November through March, possibly indicating a move of the latter up the Atlantic coast. Ropelewski and Halpert’s (1986) preliminary analyses found a “mid-Atlantic” coherency that tended to experience positive precipitation anomalies during the Aprils following a mature El Niño phase, but the 1-month duration precluded its further consideration.

(iii) Southeastern United States and mid-Atlantic coast (June, VPC1)

Figure 17 suggests a positive relationship between June precipitation in this region and eastern (VPC1) tropical Pacific SSTAs. Although this SSTA–precipitation linkage is opposite to that of the following July–August [see section 4b(1)iv above], the spatial extent of the June correlation-based coherency is remarkable, as it arcs from Arkansas–Louisiana to North Carolina and has a field significance of 93.7% (Fig. 17a). Furthermore, Fig. 17b shows that a June precipitation HKPC is similar to the southern half of this correlation-based coherency and is positively associated with a correlation-based SSTA region from the Central and South American coasts to 160°W with a field significance of 86.7% (and +0.36 interscore correlation with VPC1).

(iv) Mississippi River Basin (December, UPC1 and VPC1)

Precipitation from eastern Texas up the Mississippi River Basin to Nebraska–Minnesota is positively related to central (UPC1) and eastern (VPC1) tropical Pacific SSTAs in December (Fig. 6b), with field significances of 90.9% and 97.4%, respectively. Interestingly, this occurs during the winter month when the positive SSTA–southeastern United States precipitation relationship [see section 4b(1)i above] is least apparent.

5. Summary

In this study, principal component analysis has been used to identify North American precipitation anomaly patterns associated with SSTAs in different regions of the tropical Pacific for each calendar month. Relations with significant spatial patterns in 1 and up to 5 consecutive months were identified, with the multiple-month signals summarized in Table 2. While previous seasonally focused and SOI-focused investigations included some broad indications of the first four associations in Table 2, the present research has provided the first detailed documentation of the intraseasonal spatial variation of such precipitation anomaly patterns in relation to tropical Pacific SSTAs. Additionally, this study has yielded two multiple-month relations (for the southern Canadian prairies in November–January and a northern storm track in September–October) not detected by earlier research, along with several possible new 1-month relationships. In almost all cases, the unique two-way PCA approach used strengthened the results obtained. The areas of precipitation coherency varied somewhat with the location of the tropical Pacific SSTAs—those associated with SSTA variability in the central and eastern tropical Pacific were generally consistent, while those linked to SSTA variability in the western to north central tropical Pacific differed somewhat.

Table 2.

Summary of relationships between tropical Pacific SSTAs and central and eastern North American precipitation variability. For SSTA pattern locations, “C” denotes central tropical Pacific, “E” is eastern tropical Pacific, and “WNC” indicates western to north central tropical Pacific (see section 4a).

Summary of relationships between tropical Pacific SSTAs and central and eastern North American precipitation variability. For SSTA pattern locations, “C” denotes central tropical Pacific, “E” is eastern tropical Pacific, and “WNC” indicates western to north central tropical Pacific (see section 4a).
Summary of relationships between tropical Pacific SSTAs and central and eastern North American precipitation variability. For SSTA pattern locations, “C” denotes central tropical Pacific, “E” is eastern tropical Pacific, and “WNC” indicates western to north central tropical Pacific (see section 4a).

The monthly statistical relationships presented here have established a framework that will now permit an efficient, focused, physical–dynamic investigation of the tropical–extratropical teleconnection mechanisms of the regional climate anomalies involved. In building on this foundation, future work will involve both synoptic climatological analyses that exploit the basic daily resolution of the present precipitation dataset (section 2b) and the combined use of GCM and regional climate model simulations. It will also relax the present assumption that North American precipitation teleconnections from cold tropical Pacific SSTs are the linear inverse of those linked to warm SSTs, in order to investigate possible differences in the teleconnection mechanisms associated with each of these phenomena, including the individuality of separate El Niño and La Niña events.

Acknowledgments

This work was funded by National Science Foundation Grant ATM 92-96119 and United States Environment Protection Agency Cooperative Agreement CR-819646010. The computing was performed at the Pittsburgh Supercomputing Center and also at the National Center for Supercomputing Applications at the University of Illinois. The author is indebted to Professor Michael Richman for much assistance, encouragement, and guidance on this research, as well as to Professor Peter Lamb for extensive comments and assistance in the writing of the manuscript. Bob Livezey at the Climate Prediction Center (CPC) also provided generous help and suggestions regarding statistical field significance techniques in his informal review of the paper. The contributions of the following individuals are also gratefully acknowledged: Chet Ropelewski of CPC for initial guiding comments, Thomas Smith and Richard Reynolds of CPC for information on the SST data used here, Neil Ward for statistical advice, and an anonymous reviewer for helpful suggestions on shortening the paper and including more comprehensive statistical significance techniques.

REFERENCES

REFERENCES
Angell, J. K., 1981: Comparison of variations in atmospheric quantities with sea surface temperature variations in the equatorial eastern Pacific. Mon. Wea. Rev., 109, 230–243.
——, and J. Korshover, 1981: Comparison between sea surface temperature in the equatorial eastern Pacific and United States surface temperature. J. Appl. Meteor., 20, 1105–1110.
Barnett, T. P., and R. Preisendorfer, 1987: Origins and levels of monthly and seasonal forecast skill for United States surface air temperature determined by canonical correlation analysis. Mon. Wea. Rev., 115, 1825–1850.
Berlage, H. P., 1966: The Southern Oscillation and world weather. K. Ned. Meteor. Inst., Meded. Verh., 88, 1–152.
Bjerknes, J., 1966: A possible response of the atmospheric Hadley circulation to equatorial anomalies of ocean temperature. Tellus, 18, 820–829.
——, 1969: Atmospheric teleconnections from the equatorial Pacific. Mon. Wea. Rev., 97, 163–172.
Bretherton, C. S., C. Smith, and J. M. Wallace, 1992: An intercomparison of methods for finding coupled patterns in climate data. J. Climate, 5, 541–560.
Buell, C. E., 1975: The topography of empirical orthogonal functions. Preprints, Fourth Conf. on Probability and Statistics in Atmospheric Sciences, Tallahassee, FL, Amer. Meteor. Soc., 188–193.
Cattell, R. B., 1966: The scree test for the number of factors. Multivar. Behav. Res., 1, 245–276.
Davis, R. E., 1976: Predictability of sea surface temperature and sea level pressure anomalies over the North Pacific Ocean. J. Phys. Oceanogr., 6, 249–266.
Deser, C., and J. M. Wallace, 1987: El Niño events and their relation to the Southern Oscillation. J. Geophys. Res., 92, 14 189–14 196.
Easterling, D. R., 1991: Climatological patterns of thunderstorm activity in the south-eastern USA. Int. J. Climatol., 11, 213–221.
Fu, C., H. F. Diaz, and J. O. Fletcher, 1986: Characteristics of the response of sea surface temperature in the central Pacific associated with warm episodes of the Southern Oscillation. Mon. Wea. Rev., 114, 1716–1738.
Glantz, M. H., R. W. Katz, and N. Nicholls, Eds., 1991: Teleconnections Linking Worldwide Climate Anomalies. Cambridge University Press, 535 pp.
Gong, X., and M. B. Richman, 1995: On the application of cluster analysis to growing season precipitation data in North America east of the Rockies. J. Climate, 8, 897–931.
Halpert, M. S., and C. F. Ropelewski, 1989: Atlas of Tropical Sea Surface Temperature and Surface Winds. National Oceanic and Atmospheric Administration, 212 pp.
——, and C. F. Ropelewski, 1992: Surface temperature patterns associated with the Southern Oscillation. J. Climate, 5, 577–593.
Harris, C. W., and H. F. Kaiser, 1964: Oblique factor analytic solutions by orthogonal transformations. Psychometrika, 29, 347–362.
Horel, J. D., 1981: A rotated principal component analysis of the interannual variability of the Northern Hemisphere 500 mb height field. Mon. Wea. Rev., 109, 2080–2092.
Hsuing, J., and R. E. Newell, 1983: The principal nonseasonal modes of variation of global sea surface temperature. J. Phys. Oceanogr., 13, 1957–1967.
Kahya, E., and J. A. Dracup, 1994: The influence of type 1 El Niño and La Niña events on streamflows in the Pacific Southwest of the United States. J. Climate, 7, 965–976.
Kaiser, H. F., 1958: The Varimax criterion for analytic rotation in factor analysis. Psychometrika, 23, 187–200.
Karl, T. R., A. J. Koscielny, and H. F. Diaz, 1982: Potential errors in the application of principal component (eigenvector) analysis to geophysical data. J. Appl. Meteor., 21, 1183–1186.
Kiladis, G. N., and H. van Loon, 1988: The Southern Oscillation. Part VII: Meteorological anomalies over the Indian and Pacific sectors associated with the extremes of the oscillation. Mon. Wea. Rev., 116, 120–136.
——, and H. F. Diaz, 1989: Global climatic anomalies associated with the Southern Oscillation. J. Climate, 2, 1069–1090.
Livezey, R. E., 1995: Field intercomparisons. Analysis of Climate Variability: Applications of Statistical Techniques, H. von Storch and A. Navarra, Eds., Springer-Verlag, 159–176.
——, and W. Chen, 1983: Statistical field significance and its determination by Monte Carlo techniques. Mon. Wea. Rev., 111, 46–59.
Mendenhall, W., and T. Sincich, 1992: Statistics for Engineering and the Sciences. 3d ed. MacMillan Publishing Company, 963 pp.
Nicholls, N., 1989: Sea surface temperatures and Australian winter rainfall. J. Climate, 2, 965–973.
North, G. R., T. L. Bell, R. F. Calahan, and F. J. Moeng, 1982: Sampling errors in the estimation of empirical orthogonal functions. Mon. Wea. Rev., 110, 699–706.
Pan, Y. H., and A. H. Oort, 1983: Global climate variations connected with sea surface temperature anomalies in the eastern equatorial Pacific Ocean for the 1958–73 period. Mon. Wea. Rev., 111, 1244–1258.
Philander, S. G. H., 1990: El Niño, La Niña, and the Southern Oscillation. Academic Press, 289 pp.
Pisciottano, G., A. Diaz, G. Cazes, and C. R. Mechoso, 1994: El Niño–Southern Oscillation impact on rainfall in Uruguay. J. Climate, 7, 1286–1302.
Preisendorfer, R. W., 1988: Principal Component Analysis in Meteorology and Oceanography. Elsevier, 425 pp.
Quinn, W. H., 1974: Monitoring and predicting El Niño invasions. J. Appl. Meteor., 13, 825–830.
——, V. T. Neal, and S. E. Antunez de Mayolo, 1987: El Niño occurrences over the past four and a half centuries. J. Geophys. Res., 92, 14 449–14 461.
Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev., 110, 354–384.
Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7, 929–948.
——, and ——, 1995: A high-resolution global sea surface temperature climatology. J. Climate, 8, 1571–1583.
Richman, M. B., 1986: Rotation of principal components. J. Climatol., 6, 293–333.
——, 1987: Rotation of principal components: A reply. J. Climatol., 7, 511–520.
——, and P. J. Lamb, 1985: Climatic pattern analysis of three- and seven-day summer rainfall in the central United States: Some methodological considerations and a regionalization. J. Climate Appl. Meteor., 24, 1325–1343.
——, and ——, 1987: Pattern analysis of growing season precipitation in southern Canada. Atmos.–Ocean, 25, 137–158.
——, ——, and J. R. Angel, 1991: Relationships between monthly precipitation over central and eastern North America and the Southern Oscillation. Preprints, Fifth Conf. on Climate Variations, Denver, CO, Amer. Meteor. Soc., 151–158.
——, J. R. Angel, and X. Gong, 1992a: Determination of dimensionality in eigenanalysis. Proc. Fifth Int. Meeting on Statistical Climatology, Toronto, ON, Canada, Environment Canada, 229–235.
——, P. J. Lamb, A. Zangvil, and J. R. Angel, 1992b: Multivariate statistical analysis of precipitation associated with North American storm tracks. Preprints, 12th Conf. on Probability and Statistics in the Atmospheric Sciences, Toronto, ON, Canada, Amer. Meteor. Soc., 277–283.
Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO). Mon. Wea. Rev., 114, 2352–2362.
——, and ——, 1987: Global and regional scale precipitation patterns associated with the El Niño/Southern Oscillation. Mon. Wea. Rev., 115, 1606–1626.
——, and P. D. Jones, 1987: An extension of the Tahiti–Darwin Southern Oscillation index. Mon. Wea. Rev., 115, 2161–2165.
——, and ——, 1989: Precipitation patterns associated with the high index phase of the Southern Oscillation. J. Climate, 2, 268–284.
——, M. Chelliah, T. Smith, and D. Marsico, 1992: What can COADS tell us? Potential and limitations. Proc. Int. COADS Workshop, Boulder, CO, 295–302.
Slutz, R. J., S. J. Lubker, J. D. Hiscox, S. D. Woodruff, R. L. Jenne, D. H. Joseph, P. M. Steurer, and J. D. Elms, 1985: Comprehensive Ocean–Atmosphere Data Set release 1. Climate Research Program, Environmental Research Laboratories, Boulder, CO, 268 pp.
Smith, T. M., R. W. Reynolds, R. E. Livezey, and D. C. Stokes, 1996: Reconstruction of historical sea surface temperatures using empirical orthogonal functions. J. Climate, 9, 1403–1420.
Streten, N. A., 1983: Extreme distributions of Australian rainfall in relation to sea surface temperature. J. Climatol., 3, 143–153.
Trenberth, K. E., 1993: The different flavors of El Niño. Proc. 18th Annual Climate Diagnostics Workshop, Boulder, CO, 50–53.
Wallace, J. M., C. Smith, and C. S. Bretherton, 1992: Singular value decomposition of wintertime sea surface temperature and 500-mb height anomalies. J. Climate, 5, 561–576.
Ward, M. N., and C. K. Folland, 1991: Prediction of seasonal rainfall in the north Nordeste of Brazil using eigenvectors of sea-surface temperature. Int. J. Climatol., 11, 711–743.
Weare, B. C., A. R. Navato, and R. E. Newell, 1976: Empirical orthogonal analysis of Pacific sea surface temperatures. J. Phys. Oceanogr., 6, 671–678.
Whitaker, L. M., and L. H. Horn, 1984: Northern Hemisphere extratropical cyclone activity for four mid-season months. J. Climatol., 4, 297–310.
Wright, P. B., J. M. Wallace, T. P. Mitchell, and C. Deser, 1988: Correlation structure of the El Niño/Southern Oscillation phenomenon. J. Climate, 1, 609–625.
Wyrtki, K., 1975: El Niño—The dynamic response of the equatorial Pacific Ocean to atmospheric forcing. J. Phys. Oceanogr., 5, 572–584.

APPENDIX

Statistical Significance Test Applied

To examine the statistical field significance of the correlation-based precipitation (Figs. 6, 9, 13, 16a, and 17a) and SSTA (Figs. 7, 8, 10, 11, 12, 14, 15, 16b, and 17b) coherency patterns derived from source time series of SSTA PC scores and precipitation PC scores, respectively, a permutation test based on Livezey (1995, 164–166) was used. Here, a t value was computed from the absolute value of the correlation between each source time series and each data anomaly time series, according to Eq. (A1), from Mendenhall and Sincich (1992, 444):

 
formula

where r is the calculated correlation and Ne is the effective number of degrees of freedom. Following Davis (1976), Ne was computed for each calendar month from the corresponding autocorrelation functions of the source time series and the data anomaly time series using

 
formula

where N is the total number of samples (43), Δt is the time interval between samples (12 months), and Css and Cdd are the autocorrelation functions for a particular calendar month of a source time series and a data anomaly series, respectively. This process was repeated for each calendar month for each precipitation station or SST grid cell in the respective analyses, with the average t value for the entire correlation field then being computed. Next, the 43 values in the source time series were randomly reshuffled 1000 times, the correlations were recomputed for each of these reshufflings, and the mean t value for each resulting correlation field was recorded. The statistical field significance was then determined by calculating the percentage of the reshufflings for which the mean t value was less than that for the actual field. This percentage is indicated on each panel in Figs. 6–17. This test differs from that presented in Livezey and Chen (1983) in that the local significance was not directly assessed for each data location. Instead, a single number—the average t value, which is influenced equally by information from each data location—is computed and compared with the counterpart values computed from the reshuffled series.

Footnotes

Corresponding author address: David L. Montroy, CIMMS, School of Meteorology, University of Oklahoma, 100 E. Boyd, Room 1110, Norman, OK 73019.

* Based on a B.S. (Honors) thesis that was awarded first prize in the 1994 Father James B. Macelwane Awards in Meteorology, sponsored by the American Meteorological Society.