^{1}

^{2}

^{*}

^{3}

^{3}

^{4}

^{1}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: JCB. Performed the experiments: JCB. Analyzed the data: JCB LG FFY MGG. Contributed reagents/materials/analysis tools: JCB LG FFY RR MGG. Wrote the paper: JCB LG FFY RR MGG.

We explored the possible effects of the North Atlantic Oscillation (NAO) and Arctic Oscillation (AO) on interannual sea surface temperature (SST) variations in the Alborán Sea, both separately and combined. The probability of observing mean annual SST values higher than average was related to NAO and AO values of the previous year. The effect of NAO on SST was negative, while that of AO was positive. The pure effects of NAO and AO on SST are obscuring each other, due to the positive correlation between them. When decomposing SST, NAO and AO in seasonal values, we found that variation in mean annual SST and mean winter SST was significantly related to the mean autumn NAO of the previous year, while mean summer SST was related to mean autumn AO of the previous year. The one year delay in the effect of the NAO and AO on the SST could be partially related to the amount of accumulated snow, as we found a significant correlation between the total snow in the North Alborán watershed for a year with the annual average SST of the subsequent year. A positive AO implies a colder atmosphere in the Polar Regions, which could favour occasional cold waves over the Iberian Peninsula which, when coupled with precipitations favoured by a negative NAO, may result in snow precipitation. This snow may be accumulated in the high peaks and melt down in spring-summer of the following year, which consequently increases the runoff of freshwater to the sea, which in turn causes a diminution of sea surface salinity and density, and blocks the local upwelling of colder water, resulting in a higher SST.

The most important mechanism responsible for interannual climate variability in South-West Europe is the North Atlantic Oscillation (NAO), particularly in winter

The NAO is not the only climatic index correlated with interannual climate variability in the Northern Hemisphere. The Arctic Oscillation (AO) is a climate index that may range from positive to negative values according to pressure anomalies in the Arctic region. Thompson and Wallace

Since the AO was introduced, there has been a vivid debate about its physical reality, and its connection with the NAO

There is a great interest in researching the ocean responses, such as variations in sea surface temperature (SST), to the atmospheric oscillations, although much debate exists on the mechanisms of how they are interacted.

Frias et al.

In this line of reasoning, the aim of this study was to explore the possible combined, differential and delayed effects of the NAO and AO on the inter-annual SST variation in the Alborán Sea, and to discuss possible mechanisms for these effects based on their impact on freshwater runoff. The principal novelty is that we analysed the effects of NAO and AO separately and together.

From an oceanographic point of view, the Mediterranean is a peculiar sea, because the important oceanic events are occurring at a small scale

Monthly SST averages for the Alborán Sea, for twenty-nine years (since 1982 to 2010, the whole period available) were extracted from the Extended Reconstruction Sea Surface Temperature (ERSST.v3b) dataset, freely available from the National Oceanic and Atmospheric Administration (NOAA website. Available:

Like Frias et al.

Monthly NAO index values were taken from the website of the National Oceanic and Atmospheric Administration (NOAA website. Available:

Monthly AO index values were taken from the website of the National Oceanic and Atmospheric Administration (NOAA website. Available:

Changes in NAO trend have a delayed effect on aquatic ecosystems, arguably due to ecosystem inertia

The NAO and AO present strong inter-annual and intra-annual variability [2, 8–7], with a strong NAO pattern in cold seasons, primarily from November to March. However, given that annual SST is also influenced by seasonal downwelling and upwelling water masses, it is advisable to analyse the complete year.

Given that the mean annual SSTs in the Alborán Sea only varied about one degree Celsius between extreme values during the study period, it offers little possibility for sound statistical analysis to look for patterns or trends. However, a probabilistic analysis may be introduced by taking a year at random and calculating the probability that the average annual temperature of that year is higher or lower than the average SST for all the years.

Binary logistic regression is widely used for es lishing relationships between environmental independent variables and the probability of response of target variables

To evaluate the models we assessed their parsimony, goodness-of-fit, and discrimination capacity. We assessed the parsimony of the models using the second-order correction of the Akaike information criterion (AICc)

Hence, we obtained a predictive model and ranked the 29 analysed years according to their probability of having a mean SST higher than the average. In order to turn the predictive model into an explanatory model, it is necessary to disentangle the different roles of the NAO and AO in the final predictive model, because they are correlated ^{2}_{NAO} and R^{2}_{AO}) was obtained by correlating the ranks obtained from the final model and the partial models, using Spearman rank correlation coefficient and the squared correlation values. Then, the pure independent effect of each variable (R^{2}_{pNAO} and R^{2}_{pAO}) was assessed by subtracting from 1 (the whole variation) the variation explained by the other variable (R^{2}_{pNAO} = 1-R^{2}_{AO}, R^{2}_{pAO} = 1-R^{2}_{NAO}). The variation attributable to both factors acting collaborately (R^{2}_{NAO+AO}) may be obtained by subtracting from 1 the pure effect of the two factors (R^{2}_{NAO+AO} = 1-(R^{2}_{pNAO}+R^{2}_{pAO}))

To test the seasonal variability in SST, NAO and AO we calculated the average of each of these variables for the winter season (January, February and March), spring (April, May and June), summer (July, August and September) and autumn (October, November and December). In a first step, we decomposed each yearly variable analyzed (SST, NAO and AO) in its main seasonal components. Thus, we calculated, based on forward-backward stepwise binary logistic regression, the probability of getting a SST value of a particular year greater than the average SST for all the years, using as explanatory variables: mean winter SST (SSTwinter), mean spring SST (SSTspring), mean summer SST (SSTsummer), and mean autumn SST (SSTautumn). We did the same analysis for the NAO and the AO. Given that SST fluctuates intra-annually, we also used logistic regression to investigate the seasonal NAO and AO values which may affect SST of subsequent years and seasons.

The differential effects of NAO and AO on the Alborán Sea SST could be due to two different aspects in which NAO and AO indices differ: the statistical way of reducing the signal (EOF

Mean annual SST during the study period was 18.715°C, with a range of 1.063°C (18.097°C –19.16°C), and 0.284°C of standard deviation. When we analysed the effect of NAO and AO separately on SST we obtained a significant partial logistic regression model only for the NAO (χ^{2} = 4.044, df = 1,

When we analysed the effect of NAO and AO together on SST we found that the probability of observing mean annual SST values higher than average was significantly (χ^{2} = 11.694, df = 2,

All parameters in the function were significant according to the Wald test (p<0.05). Model goodness-of-fit statistics indicated a good fit of the model to the data, as no significant difference (Hosmer & Lemeshow

The values of the NAO and AO indices during the study period were positively correlated (Spearman correlation coefficient NAO-AO = 0.682;

Values shown in the diagrams are the percentages of variation of the final model explained by the partial models based on the two variables separately.

We show in

Key: GPCP, Global Precipitation Climatology Project; NCAR-NCEP, the National Center for Atmospheric Research and the National Centers for Environmental Prediction (NCEP).

When decomposing SST in seasonal values, we observed that the variation in mean annual SST was due mainly to variation in the mean winter SST, and then to variation in the mean summer SST (in this order), with no additionally significant contribution of the mean autumn SST and mean spring SST, according to the logit equation that we show in

Logit function | χ^{2} (p<0.05) |
AUC | AICc |

y_{SSTannual} = 338.674*SSTwinter+191.538*SSTsummer – 9609.926 |
39.336 | – | – |

y_{SSTannual} = −1.9*NAOautumnpy+0.362 |
7.119 | 0.787 | 32.679 |

y_{SSTwinter = }−1.669*NAOautumnpy+0.182 |
5.995 | 0.748 | 34.359 |

y_{SSTsummer = }−1.45* AOautumnpy−0.079 |
5.878 | 0.752 | 34.752 |

y_{NAOannual} = 143.043*NAOautumn+148.712*NAOwinter+122.009*NAOspring – 72.143 |
40.168 | – | – |

y_{AOannual} = 81.221*AOautumn+71.85*AOwinter+81.195*AOspring+ 79.861*AOsummer+4.428 |
40.168 | – | – |

However, the variation in mean annual NAO was explained by variation in the mean autumn NAO, mean winter NAO, and mean spring NAO (in this order) (see

According to AUC value, the best model to explain the annual SST variability was the model that includes the annual NAO and AO. The AO effect on SST appears to be totally obscured by that of the NAO (

NAO and AO tend to be correlated. However, NAO and AO were not correlated (r = 0.413, p = 0.235, n = 10) during the five years with the highest probability and the five years with the lowest probability of having an annual SST higher than average. Moreover, in the five years with highest probability of having a higher than average annual SST, there were two years with negative mean NAO and positive mean AO (both with a mean SST higher than average,

Year | SST | SSTbinary | NAOpy | AOpy | Probability |

1987 | 18.589 | 0 | 0.503 | 0.085 | 0.069 |

1985 | 18.707 | 0 | 0.248 | −0.192 | 0.127 |

2005 | 18.712 | 0 | 0.243 | −0.192 | 0.132 |

2000 | 18.785 | 1 | 0.391 | 0.113 | 0.189 |

1993 | 18.115 | 0 | 0.581 | 0.437 | 0.217 |

1984 | 18.097 | 0 | 0.310 | 0.032 | 0.231 |

1983 | 18.547 | 0 | 0.430 | 0.298 | 0.322 |

2001 | 18.810 | 1 | 0.207 | −0.046 | 0.323 |

1995 | 18.943 | 1 | 0.576 | 0.532 | 0.333 |

1988 | 18.470 | 0 | −0.123 | −0.544 | 0.338 |

1986 | 18.399 | 0 | −0.183 | −0.519 | 0.501 |

1992 | 18.228 | 0 | 0.268 | 0.197 | 0.526 |

1994 | 18.594 | 0 | 0.179 | 0.079 | 0.555 |

1996 | 18.731 | 1 | −0.081 | −0.275 | 0.622 |

1990 | 19.012 | 1 | 0.702 | 0.950 | 0.640 |

1997 | 19.134 | 1 | −0.214 | −0.456 | 0.654 |

1982 | 18.445 | 0 | −0.213 | −0.435 | 0.678 |

2008 | 18.755 | 1 | 0.173 | 0.269 | 0.794 |

2004 | 18.960 | 1 | 0.098 | 0.152 | 0.795 |

2003 | 19.022 | 1 | 0.039 | 0.072 | 0.804 |

2006 | 19.162 | 1 | −0.268 | −0.375 | 0.828 |

2010 | 18.935 | 1 | −0.243 | −0.330 | 0.834 |

1989 | 18.796 | 1 | −0.013 | 0.040 | 0.845 |

1991 | 18.432 | 0 | 0.594 | 1.024 | 0.875 |

2002 | 18.833 | 1 | −0.183 | −0.162 | 0.885 |

1998 | 18.812 | 1 | −0.157 | −0.040 | 0.924 |

2007 | 18.981 | 1 | −0.208 | 0.138 | 0.981 |

1999 | 18.754 | 1 | −0.481 | −0.271 | 0.983 |

2009 | 18.986 | 1 | −0.378 | 0.177 | 0.997 |

Our seasonal analyses showed that our annual values reflect mainly the variation at the beginning of the year for SST and at the end of the year for NAO and AO. In this way, the lag in the effect of climatic conditions on SST is in fact shorter than a year. In fact, in terms of the difference in AICc our model based on the NAO in autumn of the previous year was better than that based on the NAO and AO of the whole previous year, although the calibration, the amount of variability in SST explained, and the discrimination capacity were all higher in the model including NAO and AO. In addition, the effect of the seasonal AO on the seasonal SST seems to be more delayed than that of the NAO.

There are positive correlations for both AO and NAO with 1000 (

The question now lies in identifying a meteorological system producing precipitation in our area of interest (SE Iberian Peninsula and Alborán Sea) with negative anomalies of geopotential in mid-troposphere but without significant geopotential anomalies at surface levels. The single structure that answers this question is the so-known cut-off low system. A cut-off low pressure system represents a closed low in the upper and mid troposphere that has become completely detached (or “cut off”) from the characteristic westerly current of the jet stream

The high positive correlation of NAO index with 500 hPa geopotential over the Gulf of Cadiz in

Recent papers discussed large-scale climate variability for several marine ecosystems and suggested types of ecosystem responses to climate

The Alborán basin presents the peculiar shape of a funnel, surrounded by a rugged coastline, with the highest peaks of the Iberian Peninsula (eg Mulhacen and Veleta peaks over 3000 meters high), and where the mountains accumulate snow. Thus, the accumulated snow is an important fresh-water reservoir. The one year delay in the effect of the NAO and AO on the SST could be partially related with the amount of accumulated snow. To test this hypothesis we correlated the total snow in the North Alborán watershed for a year with the annual average SST of the subsequent year. To do this we used the values of snow gauge

Previous-years | Accumulated snow | Mean SST in subsequent year |

1995 | 312.2 | 18.731 (year 1996) |

1996 | 1385.2 | 19.134 (year 1997) |

1997 | 742.2 | 18.812 (year 1998) |

1998 | 478 | 18.754 (year 1999) |

1999 | 706.4 | 18.785 (year 2000) |

2000 | 978.7 | 18.8096 (year 2001) |

2001 | 736.9 | 18.833 (year 2002) |

2002 | 977.4 | 19.0218 (year 2003) |

2003 | 1379.9 | 18.959 (year 2004) |

2004 | 748 | 18.712 (year 2005) |

2005 | 802.6 | 19.162 (year 2006) |

2006 | 1069.2 | 18.981 (year 2007) |

2007 | 865.3 | 18.755 (year 2008) |

2008 | 1725.2 | 18.986 (year 2009) |

2009 | 1998.8 | 18.935 (year 2010) |

The snow thaw could modify the SST in the Alborán Sea due to its effect on the mixed layer, as an increase in freshwater runoff from snowmelt could maintain the mixed layer at a higher depth. Thus, we related the depth of the mixed layer with the amount of snow in the southern basin of the Iberian Peninsula. We calculated the mixed layer depth based on the changes of temperature and density with reference values

Year | SST | MLD | Snowautumn | Snowwinter |

1996 | 18.731 | −12.343 | 184.3 | 1117.8 |

1997 | 19.134 | −14.510 | 340 | 365.1 |

1998 | 18.812 | −12.677 | 255.6 | 240.9 |

1999 | 18.754 | −10.454 | 235.8 | 452.6 |

2000 | 18.785 | −18.343 | 419 | 167.5 |

2001 | 18.810 | −10.788 | 289 | 395.5 |

2002 | 18.833 | −14.288 | 275.3 | 421 |

2003 | 19.022 | −22.010 | 513.1 | 728.8 |

2004 | 18.960 | −25.066 | 383.9 | 119.5 |

2005 | 18.712 | −8.232 | 126.7 | 628.9 |

2006 | 19.162 | −7.732 | 211.2 | 727 |

Consequently, an explanation for the role of the AO in the final model could be that a positive AO, which implies a colder atmosphere in the Polar Regions, could favour occasional cold waves over the Iberian Peninsula which, when coupled with precipitations favoured by a negative NAO, may result in snow precipitation. This snow may be accumulated in the high peaks and melt down in spring-summer of the following year, which consequently increases the runoff of freshwater to the sea, causing diminution of sea surface salinity and density, and blocking the local upwelling of colder water. This could also help in explaining the opposite effect of NAO and AO on the SST of the following year. In fact, 75% of the years (9 years out of 29) with SST lower than average were preceded by years with positive NAO that caused low precipitation and low snow accumulation. In addition, 3 of the years with SST higher than average (1989, 2007 and 2009) were preceded by years with negative NAO and positive AO that in the latter two years resulted in a high quantity of snow accumulated in peaks. Note that we did not consider the effect of snow for the year of 1988, due to lack of data.

A remarkable feature of the NAO is that its centre of action, the Icelandic Low and the Azores High, has shifted considerably to the northeast

Many embryonic and larval phases of animals are correlated with high sea temperatures

We thank Andy Patterson for his help with the English and his comments on a previous version of the manuscript. Two anonymous reviewers provided helpful comments on earlier versions of the manuscript.