Recent ENSO–PDO precipitation relationships in the Mediterranean California border region

The Mediterranean California Border Region (MCBR) rainfall's relationship with El Niño‐Southern Oscillation (ENSO) and the Pacific Decadal Oscillation (PDO) is reexamined for the period 1951–2014. When stratifying data by ENSO events we found that strong events of either sign yield the highest ENSO–rainfall correlation; but when the stratification was done by rainfall the wet seasons yield the highest ENSO–rainfall correlation. Most strong ENSO events have the same sign as PDO; but the ENSO–rainfall correlation for all ENSO–PDO same‐sign events is almost undistinguishable from the full‐record's correlation. Timewise stratification shows that 30‐year climatological values (MCBR precipitation, PDO and ENSO) and ENSO–rainfall correlations have decreased in recent years.


Introduction
The influence of the El Niño-Southern Oscillation (ENSO) phenomenon on the rainfall of southern California is well-known (e.g. Schonher and Nicholson, 1989;Fierro, 2014). The situation is similar in northwestern Baja California as rainfall regimes are comparable on both sides of the United States-Mexico border (Pavia and Graef, 2002). The southwestern California-northwestern Baja California region is part of the Mediterranean Californias (Pavia and Badan, 1998), because it is characterized by dry summers and wet winters, with usually wetter winters during warm ENSO (El Niño) events than during cool ENSO (La Niña) events; therefore, this region hereinafter shall be referred to as the Mediterranean California Border Region (MCBR). Furthermore decadal variability (e.g. Pacific Decadal Oscillation, PDO) is known to modulate the ENSO-rainfall relationship in this region in a constructive interference way (Gershunov and Barnett, 1998;Gershunov and Cayan, 2003;Pavia et al., 2006). However in the last decade the ENSO-rainfall relationship in the MCBR has weakened, for example after the last 1997-1998 strong El Niño wet year, the only two recent wet seasons were the 2004-2005 (a neutral event) and the 2010-2011 La Niña. Climatologically this resulted in a reduction of both: 30-year ENSO-rainfall correlations and 30-year regional precipitation climatological values; the latter coinciding with a similar decline in ENSO and PDO climatologies. Therefore the study of these circumstances is of utmost importance for the MCBR, where several major cities are located and where precipitation averages only around 240 mm year −1 (see Figure 1). Within the MCBR the situation is even more critical for Baja California than California, because the former lacks the level of hydraulic infrastructure that the latter possesses (MacDonald, 2007). Elsewhere variations in ENSO-rainfall and ENSO-PDO-rainfall relationships have been noted as well (e.g. Rocha, 1999;Krishna Kumar et al., 1999;McCabe and Dettinger, 1999;Chang et al., 2001;Cai et al., 2001;Sarkar et al., 2004;Gao et al., 2006;Zubair and Ropelewski, 2006;Kucharski et al., 2007); thus in this paper we study the changes in ENSO, PDO and rainfall in the MCBR (Figure 1), focusing on the recent 1951-2014 period. Our goal is not to scrutinize single or particular events, but to examine different subsets of our record (stratified in different ways) searching for rather general relationships of the MCBR rainfall with ENSO and PDO.

Data and methods
We construct ENSO, PDO and MCBR rainfall indices to obtain baseline correlations among them (for the entire period) and subset correlations for different stratifications (by index values and timewise). We also compute 30-year climatological values for these three variables plus 30-year ENSO-rainfall and PDO ′ -PDO correlations; where PDO ′ is an 'ENSO-forced' model (see Newman et al., 2003;Pavia, 2009). Finally we construct global maps of SST-ENSO and SST-regional precipitation correlation patterns for different periods to substantiate our results.

Original data
For ENSO we select the monthly El Niño 3.4 index from NOAA's Earth System Research Laboratory/Physical Sciences Division (ESRL/PSD) because the region of the El Niño 3.4 index is closely related to extra-tropical teleconnections; however, using other ENSO indices yields similar results. For decadal variability we select the PDO index (obtained from the University of Washington's Joint Institute for the Study of the Atmosphere and Ocean). For regional rainfall we select the monthly precipitation data from seven stations: Hemet (PHE), Laguna Beach (PLB), San Diego (PSD) and Chula Vista (PCV), California, obtained from the National Weather Service (NWS) among many others, and from Tijuana (PTJ), Ensenada (PEN) and San Quintin (PSQ), Baja California, obtained from the Mexican Water Authority (CNA) (see Figure 1). We also use global SST data from the Hadley Center (HadISST, Rayner et al., 2003), and gridded precipitation data from Livneh et al. (2013) within the region between 115 ∘ and 120 ∘ W and 30 ∘ and 34.5 ∘ N to perform further correlation analyses to verify our results. A preliminary analysis of the seven stations and gridded precipitation data gave comparable results; thus the stations dataset is considered typical of MCBR precipitation.

Data preparation and correlation analyses
We begin by calculating the average of annual total (July to June) precipitation of the seven stations: PHE, PLB, PSD, PCV, PTJ, PEN and PSQ; we consider the standardized average to be representative of our region's rainfall and call it CPI (CPI and a similar index calculated from the gridded data of Livneh et al. (2013) are correlated above 0.9). Similarly for the ENSO and PDO indices we use annual averages of the El Niño 3.4 index and the PDO index (also July to June), and call them ENI and PDI ( Figure 2). The correlations of ENI and PDI with CPI are performed for the entire 63-year series, as a first assessment of their relationship, and for 30-year running periods. The ENSO-PDO relationship is assessed thru the use of a model: PDI ′ = f (PDI, ENI). We then stratify data by index value to obtain subset correlations; similarly we stratify data timewise by periods and by seasons and calculate climatological values in addition to subset correlations. Finally four series of global SST-ENSO-rainfall spatial correlation maps (1951-1981, 1965-1995, 1983-2013 and 1951-1965 plus 1999-2014) are calculated to examine the details of our results. We selected the mentioned periods in order to show the differences in SST-regional precipitation and SST-ENSO correlation patterns depending on the dominant ENSO phase in each period. The period 1983-2013 includes the two El Niño events with the overall highest amplitude (1983 and 1998), while the 1951-1965 plus 1999-2014 combined-period excludes the four strongest ENSO events during winter (1965-1966, 1972-1973, 1982-1983 and 1997-1998) (see Figure 3).

ENI-CPI correlation
The baseline correlation for our entire record is R all = 0.58; thus only about one third of recent MCBR rainfall variability may be explained by ENSO.

Stratification by ENSO events
To further investigate the correlation between ENI and CPI, we stratified the ENI data into strong ENSO events (|ENI| > 0.8) and neutral events (−0.8 ≤ ENI ≤ 0.8). This yields 25 strong ENSO (12 positive) and 38 neutral ENSO events. The correlation R strong = 0.83, is statistically significant as the probability of getting a correlation as large as the observed value by random chance is 10 −7 , lower and upper bounds for a 95% confidence interval for R strong are [0.64, 0.92]. All but 1 year out of 12 with strong positive ENSO have CPI > 0; that is, rain above the mean. Also all but 2 years out of 13 with strong negative ENSO (ENI < −0.8) have CPI < 0; that is, rain below the mean (see Figure 4). This is the most robust result we could find in these data. With the definition more commonly used in the literature: |ENI| > 1.0, we obtained 20 strong and 43 neutral 282 E. G. Pavia, F. Graef and R. Fuentes-Franco events, and it yields R strong = 0.82, R neutral = 0.41, correlations which are statistically indistinguishable from those obtained using |ENI| > 0.8.

Stratification by regional rainfall events
Similarly when we stratified by wet (CPI > ), dry (CPI < − ) and neither (− ≤ CPI ≤ ) events (using different values for , from 0.5 to 1.0), the correlations were always higher for R wet (∼0.5, and significant) than for R dry (∼0.0, and non-significant). For R neither we also found values comparable with R wet , but only marginally significant.

Seasonal analysis
We also averaged monthly El Niño 3.4 Index data (ENSO) by season; that is, winter: January, February and March; spring: April, May and June; summer: July, August and September; autumn: October, November and December. For precipitation data (RAIN) we calculated totals for winter and autumn.
After an extensive statistical analysis we found that seasonal correlations are in all cases lower than R all = 0.58. The highest correlation found was R(ENSO winter , RAIN winter ) = 0.55; when using ENSO lagged data we found that ENSO's previous year gives R(ENSO autumn (n−1) , RAIN winter ) = 0.50. We did not find correlations RAIN winter (or RAIN autumn ) with any other ENSO season higher than 0.50. In general correlations increase as the ENSO season gets closer to RAIN winter .

SST-ENSO-rainfall correlation maps
We correlated all regional seasonal winter precipitation with the gridded global SST at all grid points (Figure 3). Similarly we correlated the ENSO index with the gridded global SST. All ENSO-SST correlation maps (Figure 3(b), (d), (f), (h)) show similar patterns among them for positive correlations in the Pacific, in contrast with the precipitation-SST correlation maps. The 1983-2013 period shows the highest correlation patterns (Figure 3(e)), it includes the strongest El Niño events (1983 and 1998), and furthermore this period exhibits the greatest similarity between ENSO-SST correlation (Figure 3(f)) and precipitation-SST correlation (Figure 3(e)). The period with the lowest correlation patterns is 1951-1965 plus 1999-2014, which excludes the four strongest ENSO events during the wintertime within our record: 1965-1966, 1972-1973, 1982-1983 and 1997-1998 (Figure 3(g)), and which also shows the greatest difference with the ENSO-SST correlation patterns (Figure 3(h)).

ENI-CPI correlation
Considering the stratification done in Section 3.1.1 we found that PDI and ENI have the same sign in 76% of strong ENSO events; but the ENI-CPI correlation for all ENSO-PDO same-sign events is R = 0.62. To further investigate the constructive interference of Gershunov et al. (1999) we varied the criterion parameter ( ) that we used to stratify strong and neutral ENSO years from 0.5 to 1.0 and also stratified the years as follows: (i) PDI ≥ 0 and ENI ≥ , (positive phase of PDO and strong El Niño). (ii) PDI < 0 y ENI < − (negative phase of PDO and strong La Niña). (iii) (i) and (ii). For example, for = 0.5 (0.8) we get 13 (8), 16 (11) and 29 (19) years for cases (i), (ii) and (iii), respectively. Comparing the corresponding correlation matrices, the correlations (ENI, PDI) and (PDI, CPI) are higher in general (some statistically significantly) in case (iii) than in case (i) but more so than in case (ii). None of the correlations in case ii) is significant and the highest correlation is R(ENI, PDI) = −0.54 for = 1.0, with 95% CI being [−0.89, 0.19]. In contrast, all correlations in case iii) are significant. With respect to R(ENI, CPI), it increases from 0.32 for = 0.5 to 0.90 for = 1.0 in case (i), and from 0.72 to 0.82 (not a significant increase) in case (ii); there is also an increase in case (iii), with respect to (i) and (ii), but only for values of = 0.5, 0.6 and 0.7. These results are in line with those found in Pavia et al. (2006): El Niño and the positive phase of PDO show constructive interference, but La Niña and a negative phase of PDO do not. Correlations for negative PDI and negative ENI (23 years) are low and non-significant, compared with the cases positive PDI and positive ENI (20 years).

30-year analysis
The most notorious feature here is that climatological values have decreased most recently: precipitation, since the 1975-2004 period, PDO since the 1976-2005 period, and ENSO since the 1977-2006 period; while the ENI-CPI correlations (R 1957 − 1986 = 0.53 < R(ENI, CPI) < R 1970 − 1999 = 0.73) have experienced a similar decrease, that is R 1984 − 2013 = 0.54 (see Figure 5(a)). This may be related to the fact that within our period there was a low-frequency positive PDO phase during the last quarter of the 20th century. There is also tree-ring evidence that during this period the region was unusually (in a 1000-year record) wet (Swetnam and Betancourt, 1998).

PDO ENSO-dependency
In this case we investigated the variability of the ENSO-PDO relationship within our period thru the use of a 'forced' model: PDI n ′ ∼ × PDI n−1 + × ENI n (see Pavia, 2009) for the entire record and for 30-year running periods. For the entire record we found R(PDI ′ , PDI) all = 0.70, and for the running correlations: Figure 5(b)).

Discussion and conclusions
The ENSO influence and the PDO-ENSO constructive interference on annual MCBR precipitation have been confirmed for the period 1951-2014. However the latter is clearer for the case of strong ENSO events of either sign (in 19 out of 25 events ENI and PDI have the same sign and the ENI-CPI correlation is R strong = 0.83) than for any other case: neutral, strong-positive, strong-negative, etc. (see Figure 4). Nevertheless the ENI-CPI correlation for all ENI-PDI same-sign events (R same-sign = 0.62) is statistically undistinguishable from the baseline correlation (R all = 0.58). That is, apart from the strong ENSO case, only about one third of regional annual rainfall variability may be explained by ENSO. Also in this period dry years are significantly less related to ENSO (R dry = −0.04) compared with wet years (R wet = 0.47), and 85% (92%) of strong negative (positive) ENSO events correspond to years with precipitation below (above) the mean; this is the opposite of what has been suggested: i.e. that La Niña is more consistently and predictably related to dry winters in the region than El Niño is related to wet winters (e.g. Barnett, 1998, Gershunov andCayan, 2003). Seasonally winter precipitation is better related to winter ENSO than to other season's ENSO values; however, the heavy 1977-1978 regional precipitation (see Figure 2) was not related to a strong positive ENSO event, but rather to a climate shift in the Pacific Ocean (Miller et al., 1994), neither does the recent MCBR drought seem to be associated with La Niña (Seager et al., 2014): the 2010-2011 La Niña was a wet year. This explains the recent decrease in 30-year ENI-CPI correlations (Section 3.2.2), which is almost concurrent with the recent decrease in 30-year climatological values for PDI, ENI and CPI (see Figure 5(a)). This may be related to a somewhat weaken influence of ENSO on the PDO (Figure 5(b)) during the peak of the 30-year PDI value ( Figure 5(a)), or perhaps exceptional years, such as 1977-1978 and 2010-2011, weaken the ENSO-MCBR rainfall relationship in general compared with the only strong ENSO case. Furthermore SST-ENSO-regional precipitation maps show that MCBR annual rainfall is clearly influenced by ENSO during periods with strong warm events, as revealed by conspicuous positive correlation patterns (see Figure 3). That is, in the absence of strong El Niño events regional precipitation is not well correlated to ENSO, and frequent rainy seasons usually concur with repeated warm ENSO events, but several years of low precipitation do not always relate with frequent cool ENSO events. This is in line with recent findings, for example: of the last three dry seasons, only the 2011-2012 season was found to be a response to La Niña, whereas the 2012-2013 and 2013-2014 seasons were related to a warm tropical west Pacific sea surface temperature anomaly (Seager et al., 2014). Consequently for the last 60 years or so wet (dry) periods correspond to higher (lower) ENSO-regional rainfall correlations.
Therefore in general, that is irrespective of particular events, strong ENSO events of either sign are good predictors of rainfall in the Mediterranean Californias: strong positive (negative) events implying wet (dry) years; but otherwise ENSO (in general: that is irrespective of particular years) is not a good predictor for rainfall in our region. Similarly constructive interference is only effective for strong ENSO events; otherwise the general coincidence of ENI and PDI signs impacts ENI-CPI correlations very slightly. For dry years the correlations (ENI, CPI) and (PDI, CPI) were found to be negative or not significantly different from zero.
In the last 15 years, besides the 2010 El Niño and the 2011 La Niña, which did not result in expected precipitation anomalies, the lack of strong ENSO events may explain why ENSO has not been a good predictor for MCBR precipitation recently. It can also be that what we call the past positive phase of the PDO during the last quarter of the 20th century was an exceptional period, as shown by a 1000-year tree-ring record (Swetnam and Betancourt, 1998), and the last 15 years is just a return to the normal conditions, similar to the pre-1977 situation.
Finally considering the 2015 El Niño which is expected to influence weather and climate patterns for the 2015-2016 winter: timewise this event is occurring during a dry and lower ENI-CPI correlation period; nevertheless our analysis of ENSO strength stratification indicates that strong events such as the 2015 El Niño are very likely to correspond to a very rainy season in our region. The October 2015 US Winter Outlook, issued by NOAA's Climate Prediction Center (http://www.noaanews.noaa.gov/stories2015/), indicates wetter than average conditions for southwestern California and other southern US states. Also the Mexican Meteorological Service is forecasting precipitations above normal in northwestern Baja California in particular and in northwestern Mexico in general. These statements are both in line with our ENSO-stratification results, with the caveat of our timewise climatological results indicating the recent weakening of the ENSO-rainfall relationship in our region.