Trends in seasonal surface air temperature in mainland Portugal, since 1941

ABSTRACT

This study provides a thorough assessment of recent changes in near‐surface air temperature in mainland Portugal at both the local and regional level, focusing on extreme events (maximum and minimum) at a seasonal scale. It examines trends in selected specific indices that are calculated from daily air temperature data from 23 measuring stations scattered across the territory, recorded between 1941 and 2006. The results show overall warming trends over mainland Portugal that are consistent with the dominant global warming and reflect an increase in both maximum and minimum air temperature. When we split the study period into two sub‐periods, 1945–1975 and 1976–2006, the partial trend analysis reveals that the first sub‐period is mostly characterized by cooling followed by an even stronger tendency towards warmer benchmarks in all the indices evaluated, in particular for the warm‐related temperature extremes in spring and summer.

The changes observed in seasonal patterns confirm the well‐known asymmetries in the climate in mainland Portugal and suggest that they are likely to be aggravated. There are changes associated with extreme temperatures, in particular, the significant increase in the frequency and duration of heat waves, and the increase in the frequency of hot days and tropical nights, especially in spring and summer; moreover there is a significant decrease in the frequency of cold waves and frost days.

Teleconnections associated with changing patterns of temperature are also investigated. The results show that, over mainland Portugal, cold‐related air temperature extremes have been associated with the East Atlantic mode in autumn, whereas warm‐related extremes have been associated with the Scandinavia teleconnection pattern in spring, summer and autumn. However, the most prominent Northern Hemisphere pattern, the North Atlantic Oscillation, exerts limited influence, which is felt mostly in winter and spring.

1. Introduction

The effect of climate change on society and the environment is a major concern worldwide. Changes in climate variables such as surface air temperature and/or precipitation could have great socio‐economic impacts at the regional and local scales, and have been reported globally in different areas: health, food security, agriculture, water availability, and water and other natural resources are only examples from a long list (e.g. IPCC, 2007b, García‐Herrera et al. , 2010, Trigo et al. , 2010).

For near‐surface air temperature, a dominant warming trend has been noted in recent decades in many geographical locations (e.g. Easterling et al. , 1997, Manton et al. , 2001, DeGaetano and Allen, 2002, Frich et al. , 2002, Klein Tank et al. , 2005, Vincent et al. , 2005, Alexander et al. , 2006, Moberg et al. , 2006, IPCC, 2007a). Although there was a break in Europe around the mid‐1970s, Walsh and Oliver (2011) found that the observed temperature trends seem to be too large to be due only to changes in CO₂. So the correspondence of CO₂ and surface temperature breakpoints in 1975 may not represent cause and effect but a response to a decadal shift in atmospheric behaviour. This warming, observed since the mid‐1970s, was associated with the prominent role played by large‐scale oceanic oscillations in the Pacific (PDO, Hurrell, 1996, Guilderson and Schrag, 1998) and Atlantic (AMO, Schlesinger and Ramankutty, 1994). Additionally, other authors have mentioned the increasing role of global climate forcing and human‐made greenhouse gases, GHG (e.g. Meehl et al. , 2007, Hansen et al. , 2011, Christidis et al. , 2012), possibly associated with the decreasing role of aerosols (e.g. Booth et al. , 2012b) and evolution of the dimming effect (e.g. Wild, 2009).

Human activities and the related increment of GHG have increased the risk of certain extreme weather events, such as the 2003 and 2010 heat waves in Europe (e.g. Barriopedro et al. , 2011). But we must acknowledge that they have also reduced the risk of other extreme phenomena (such as cold waves), and may have not affected the risk of some events substantially (e.g. Stott et al. , 2012).

The global mean near‐surface air temperature has increased 0.74 °C in the 1906–2005 period (ranging from 0.56 to 0.92 °C) (IPCC, 2007a). This warming was found not to be uniform over time (i.e. air temperature increasing at a constant rate). Instead, a relative cooling phenomenon was observed roughly from the 1950s to the 1970s that was followed over approximately the last 40–50 years by a period registering a relative (linear) warming trend of 0.13 °C (ranging from 0.10 to 0.16 °C), which is nearly twice that for the 100 years from 1906 to 2005 (IPCC, 2007a).

The fact that many studies are reporting different patterns and magnitudes of climate change highlights the need for better insight into both empirical findings, resulting from long‐term, high resolution, homogenized ground‐based data analyses, as well as from numerical models of future climate projection. Moreover, specific methodologies are required to study changes in mean values of the probability distribution function and particularly of extreme events that can occur at different temporal and spatial resolutions (e.g. Barriopedro et al. , 2011; Lovejoy and Schertzer, 2012).

After the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC, 2007a), the investigations on observed changes in climate extremes have significantly improved our understanding of temperature and precipitation extremes at the global (e.g. Brown et al. , 2008, IPCC, 2012) and regional scales for: North America, Hawaii, the Caribbean and US Pacific Islands (e.g. Kunkel et al. , 2008); the United States (e.g. Portmann et al. , 2009); Western North America (e.g. Booth et al. , 2012a); Western Central Africa, Guinea Conakry and Zimbabwe (e.g. Aguilar et al. , 2009); the Indo‐Pacific region (e.g. Caesar et al. , 2011); the Asia‐Pacific region (Choi et al. , 2009); countries located in the western Indian Ocean (e.g. Vincent et al. , 2011); and Europe (e.g. Della‐Marta et al. , 2007, van den Besselaar et al. , 2010). There are also several studies of climate variability focusing on the Mediterranean Basin, which were reported in Hertig et al. (2010) and Efthymiadis et al. (2011), and elsewhere.

Regarding the analysis of temperature extremes in the Iberian Peninsula several studies can be found in the literature: e.g. García‐Herrera et al. (2005), Rodríguez‐Puebla et al. (2010), Fernández‐Montes and Rodrigo (2012) and Sánchez‐Lorenzo et al. (2012). However, these studies have relied on very few stations in western Iberia (less than 5), where Portugal is located.

For mainland Portugal, empirical observations show that from the mid‐1970s the air temperature in mainland Portugal rose at a rate of about 0.5 °C decade⁻¹ (e.g. Ramos et al. , 2011), which is a considerably higher rate than the world average. Since then, seven out of the ten warmest years occurred after 1990 (e.g. Miranda et al. , 2006; Ramos et al. , 2011). Changes in the air temperature regime could affect not only the mean temperatures but also extremes' variability, and these changes have still not been fully explored.

Contributions to the characterization of extreme temperature events in mainland Portugal have come from e.g. Miranda et al. (2002, 2006) and recently by Ramos et al. (2011) and de Lima et al. (2013), in some cases by examining linear trends in a few selected indices of daily temperatures and focusing on the year as a whole. Nevertheless these studies did not fully explore seasonal trends. For example, at the seasonal scale Ramos et al. (2011) only investigated trends in percentile‐based indices (warm/cold days/nights). We are now reporting new insights into and supplementing those studies by analysing the surface air temperature data from mainland Portugal in more detail, at the seasonal scale.

By assessing seasonal variability in the intensity, frequency, and duration of extreme events, at the regional scale, this study examines seasonal trends in specific selected indices computed from daily data recorded at 23 locations in mainland Portugal, in the period 1941–2006. Trend analyses are conducted over the full period of the records and also for two sub‐periods 1945–1975 (a relative cooling period) and 1976–2006 (a relative warming period). The added value of this seasonal scale study can be used to find out if the warming revealed by the annual indices (e.g. Ramos et al. , 2011) can be observed in all seasons or attributed in particular to just one or two seasons.

The following section presents the surface air temperature datasets that were used in this study while Section 3. presents the different methodologies employed in the analyses. Section 4. provides the results of the analyses of the seasonal variability, variations in the intensity, frequency, and duration of extreme events in Portugal, and of the large‐scale influence on seasonal temperature extremes. The results and conclusions are discussed in Section 5.

2. Description of the study area and air temperature data

2.1. Study area

Mainland Portugal is located in western Europe between latitudes 36°56′ and 42°09′N and longitudes 6°10′ and 9°34′W. The highest altitude above mean sea level is roughly 2000 m; a relief map of mainland Portugal is shown in Figure 1.

Figure 1

Relief map of mainland Portugal and location of the climatological stations used in this study (Table 1).

The study area is in the transitional region between the sub‐tropical anticyclone and the sub‐polar low‐pressure zones. The climate, whose predominant characteristics are those of a mild Mediterranean climate, is greatly influenced by the latitude, orography, and proximity of the Atlantic Ocean (e.g. Miranda et al. , 2002, Lionello et al. , 2006). These characteristics, which are more pronounced in the south, include a warm dry summer period, strong seasonality and large inter‐annual variability. Annual average temperature varies between about 7 °C in the highlands of the northern and central interior and about 18 °C at the south coast.

2.2. Data

The surface air temperature data (maximum and minimum) for mainland Portugal used in this study are daily records from 23 climatological weather stations from the Portuguese Sea and Atmosphere Institute (IPMA) network. The location of these stations is shown in Figure 1; and their names, locations, and altitudes are given in Table 1 together with the period of the records and mean seasonal temperature. The data cover the period 1941–2006, but there are a few time series that start after 1941.

Table 1. Climatological weather stations used in this study: names and coordinates of the stations and period of surface air temperature records. Mean seasonal temperature for the reference period 1961–1990 is also given

Station	Lat. (°N)	Long. (°W)	Alt. (m)	Period		Mean seasonal temperature (°C)
				from	to	spring	summer	autumn	winter
Montalegre	41.82	07.78	1005	1941	2006	8.4	16.7	11.0	4.2
Bragança	41.80	06.73	690	1941	2006	10.9	20.1	13.0	5.1
Braga	41.55	08.40	190	1941	2006	13.4	20.2	15.5	9.3
Régua	41.17	07.80	65	1941	2006	15.0	22.9	16.7	9.0
Pinhão	41.17	07.55	130	1941	2006	14.9	24.1	16.8	8.4
Porto/S. Pilar	41.13	08.60	93	1941	2006	13.7	19.6	15.9	9.8
Penhas Douradas	40.42	07.55	1380	1941	2006	7.1	16.4	10.2	3.3
Coimbra/Bencanta	40.15	08.47	141	1941	2006	14.4	20.7	16.6	10.2
Alvega	39.47	08.05	51	1949	2006	14.4	22.3	16.7	9.4
Cabo Carvoeiro	39.35	09.40	32	1941	2006	14.2	18.1	16.6	12.0
Portalegre	39.28	07.42	597	1942	2006	13.7	22.7	16.7	9.1
Elvas	38.88	07.15	208	1941	2006	14.7	23.7	17.4	9.1
Cabo da Roca	38.78	09.50	142	1941	2006	13.9	17.8	16.6	11.9
Lisboa/Geofísico	38.72	09.15	77	1941	2006	15.8	22.1	18.3	11.8
Pegões	38.60	08.60	64	1941	2006	14.8	21.9	17.5	10.6
Évora	38.53	07.88	200	1941	2006	14.3	22.4	17.2	9.8
Setúbal	38.52	08.90	35	1949	2006	15.0	21.9	17.4	10.8
Amareleja	38.22	07.22	192	1963	2006	14.8	24.1	17.7	9.7
Beja	38.02	07.87	246	1957	2006	14.7	23.2	17.7	10.2
Alvalade	37.95	08.40	61	1941	2006	14.7	22.2	17.3	10.1
V. R. S. António	37.20	07.40	7	1949	2006	16.1	23.4	18.6	11.6
Sagres	37.02	08.95	40	1952	2006	15.2	19.8	17.9	13.1
Faro	37.02	07.98	8	1966	2006	15.9	22.9	18.7	12.5

• The location of these stations in mainland Portugal is shown in Figure 1.

Spatial scatter of the temperature series over mainland Portugal and data length, completeness, quality, and homogeneity were the main criteria for the selection of the dataset. We have restricted the analysis to those stations with fewer than 5% of missing values.

Basic quality controls were carried out for the time series individually. Anomalous values, ‘outliers’, in daily maximum and minimum temperature were identified as values that are above/below four standard deviations from the climatological daily average. Such daily temperature values were marked as a potential error and they were manually checked on a case‐by‐case basis. The few outliers that found were accepted because they were geographically consistent and could be explained by unusual synoptic circulation that gave rise to these exceptional events (e.g. summer 2003 heat wave). Standard homogeneity tests were applied to the time series of temperature monthly means, using the RhtestsV3 software package (e.g. Wang et al. , 2007, Wang, 2008, Wang and Feng, 2010). This procedure allows us to detect change‐points in the series that could affect the analysis of trends in the data. Ramos et al. (2011) identified a few time series that revealed non‐negligible change points, which were checked against the stations' metadata records. In fact, the available metadata for the 23 stations is restricted to the minimum requirements, including the geographical data/locations and changes in location, observing practices or data programmes. Additional information such as external changes (e.g. land‐use change) is not adequately reported. So, the few change points found previously (e.g. Ramos et al. , 2011) do not correspond to any known changes in geographical locations or instrumentation. Therefore, heeding other works that analysed the same temperature dataset, we decided not to correct the data (Ramos et al. , 2011, and de Lima et al. , 2013).

The major limitation to extending the study beyond 2006, while maintaining a good compromise between lack of missing values and a good spatial coverage, is that the number of conventional weather stations available fell significantly as many of them were either closed down or replaced by automatic weather stations.

3. Methodology

Changes in climate extremes are likely to be observed in different climate variables, (including in the mean and extreme values). In this study we are especially interested in exploring changes in extreme near‐surface air temperature data, related to both hot and cold extreme events.

Although there are different approaches to evaluating changes in extreme events in climate variables such as surface air temperature, here we are focusing on trends in specific extreme indices derived from daily maximum, and minimum air temperature data at the seasonal scale.

3.1. Indices

The standard climate indices that were selected for characterizing air temperature extremes are described in Table 2 (e.g. Peterson et al. , 2001). These indices were first proposed for assessing the many aspects of a changing climate and include threshold, absolute, duration, and other indices. Some of them are related to events that belong to the tails of the probability distributions, having very high return periods (e.g. Alexander et al. , 2006), whereas others represent events that are expected to occur on average several times per year (or season).

Table 2. Selected indices of maximum and minimum temperature extremes (e.g. Peterson et al ., 2001)

Element	Index	Description	Definition	Unit
TX	Mean TX	Maximum temperature	Mean of maximum temperature	°C
	TX25	Summer days	Number of days with daily maximum temperature > 25 °C	days
	TX35	Very hot days	Number of days with daily maximum temperature > 35 °C	days
	TXx	Warmest day	Maximum value of daily maximum temperature	°C
	TXn	Coldest day	Minimum value of daily maximum temperature	°C
	HWDI	Heat wave duration	Number of days in intervals of at least 6 consecutive days with TX > mean + 5 °C calculated for each calendar day (for reference period 1961–1990) using running 5‐d window	days
	TXHW90	90th percentile of heat wave duration	Maximum number of consecutive days with TX > 90th percentile calculated for each calendar day (1961–1990 reference period) using running 5‐d window	days
TN	Mean TN	Minimum temperature	Mean of minimum temperature	°C
	FD	Frost days	Number of days with daily minimum temperature < 0 °C	days
	TN20	Tropical nights	Number of days with daily minimum temperature > 20 °C	days
	TNn	Coldest night	Minimum value of daily minimum temperature	°C
	TNx	Warmest night	Maximum value of daily minimum temperature	°C
	CWDI	Cold wave duration	Number of days in intervals of at least 6 consecutive days with TN < mean‐5 °C calculated for each calendar day (1961–1990 reference period) using running 5‐d window	days
	TNCW10	10th percentile of cold wave duration	Maximum number consecutive days with TN < 10th percentile calculated for each calendar day (1961–1990 reference period) using running 5‐d window	days
Both: TX, TN	DTR	Diurnal temperature range	Monthly mean difference between TX and TN	°C
	ETR	Extreme temperature range	Difference between TXx and TNn	°C

• Symbol TX denotes daily maximum temperature and TN denotes daily minimum temperature. The indices were computed at the seasonal scale.

These climate indices have been adopted by the World Meteorological Organization and the Intergovernmental Panel on Climate Change (e.g. Peterson et al. , 2001) and are being widely used in many studies. The main advantages of this approach are that the computation of these indices is relativity simple and they can be compared directly with other studies in different regions of the world that use the same indices (e.g. Klein Tank et al. , 2009). For a discussion on the evolution over time of the indices of extremes, see e.g. Zhang et al. (2011).

The 16 indices selected were calculated at the seasonal scale, for each individual station and year, even those indices that occur with high frequency (e.g. TX25 in summer) and those that hardly ever occur (e.g. TX35 in winter).

For practical purposes, we have considered here that the different seasons of the year coincide with these quarters: September, October and November—SON (autumn); December, January and February—DJF (winter); March, April and May—MAM (spring); and June, July and August—JJA (summer). In addition, the reference period considered in this work is the climatological normal period 1961–1990.

3.2. Regional average series

To compile an overview of climate variability and regional trends, we have also computed, for every index, the regional average for mainland Portugal. The regional averaged indices are computed as the mean of the indices at individual stations, relative to the climate normals for the 1961–1990 reference period. This simple procedure was adopted because the widespread distribution of the 23 selected stations enables (in this aggregated index) all different areas of the country to be taken into account.

3.3. Trend tests

Time series of seasonal indices of surface air temperature extremes were tested for linear trends. The analyses were carried out for the full 66‐year record period (i.e. 1941–2006) and for two consecutive 31‐year sub‐periods, 1945–1975 and 1976–2006. The full monotonic trends and partial trends were both estimated by ordinary least squares (OLS) fitting. This approach was adopted based on the finding reported by several other authors that, in general, the magnitude of the trends estimated by the OLS method and by a more robust nonparametric method is very similar (e.g. Moberg and Jones, 2005). Moreover this methodology has been used in other studies (e.g. Klein Tank et al. , 2002, Klein Tank and Können, 2003, Klein Tank et al. , 2006, Moberg et al. , 2006, Efthymiadis et al. , 2011, Ramos et al. , 2011). The statistical significance of the trends was evaluated using Student's t test.

It is worth mentioning that, for air temperature indices, positive (increasing) trends might not necessarily indicate warmer conditions, nor would negative (decreasing) trends indicate cooler conditions. That is why both minimum and maximum extremes are being studied.

The (standard) breakpoint from a cooling period to a warming period found by Karl et al. (2000) for global air temperature was used to select the sub‐periods 1945–1975 and 1976–2006. According to Tomé and Miranda (2011), worldwide, the turning point from a cooling to a (recent) warming period is not always exactly the same (i.e. around the mid‐1970s); their study found marked differences in the behaviour of air temperature climatology across the globe, with some regions having changed to such a warming phase only in the 21st century, contrary to what has been observed elsewhere.

We used the method proposed by Tomé and Miranda (2004, 2005) for the partial trend analysis, since it allows a piecewise linear fitting model be applied to climate data; this model gives better results than the linear trend model (e.g. Seidel and Lanzante, 2004). The method can detect change points in temperature trends without predefining the number of breakpoints. We made an initial trend analysis of annual temperature time series based on this approach, i.e. not constraining the change point in the trend but allowing it to be detected objectively by the this robust method (Tomé and Miranda, 2004). For the majority of stations this breakpoint fell around the early to mid‐1970s, which agrees with what was found by Ramos et al. (2011). Thus, the selection of the breakpoints was initially based on a statistical approach. But this breakpoint (for local stations) was broadly similar to the one reported by Karl et al. (2000), at global scale. Therefore, we have chosen to impose fixed breaks in the time series to estimate the trends (e.g. Klein Tank et al. , 2002, Miranda et al. , 2002, van den Besselaar et al. , 2010, Ramos et al. , 2011). The possible physical mechanisms associated with this breakpoint have already been mentioned briefly in the introduction and will be discussed more in the last section.

3.4. Modes of low frequency variability

In Europe, large‐scale atmospheric patterns are known to drive inter‐annual variability of seasonal‐average temperatures in winter and also, to a lesser extent, in summer (e.g. Efthymiadis et al. , 2011). Of these patterns the North Atlantic Oscillation (NAO) is the most important teleconnection pattern in the Northern Hemisphere and the only one evident throughout the year (e.g. Wallace and Gutzler, 1981, Barnston and Livezey, 1987); it mostly affects winter temperatures, though it does so differently depending on geographical location (for the north‐western and south‐eastern Europe‐Mediterranean area see e.g. Hurrell and van Loon, 1997). Other modes that generally affect the region are the East‐Atlantic (EA), East Atlantic/Western Russia (EA/WR), and Scandinavia (SCAND) pattern (Barnston and Livezey, 1987). Several studies on the impacts of the modes of low frequency variability on temperature in different regions in Europe can be found in the literature (e.g. Xoplaki et al. , 2003, Lionello et al. , 2006, Kyselý, 2008, Rodríguez‐Puebla et al. , 2010, Cattiaux et al. , 2011, Efthymiadis et al. , 2011).

For the 1951–2006 period, these four indices were obtained from the Climate Prediction Center (CPC) of the National Oceanic and Atmospheric Administration (NOAA) of the United States of America. These modes of low‐frequency variability indices were computed from the 500 hPa geopotential height field for the entire Northern Hemisphere (20–90°N), using rotated principle component analysis (PCA) (Barnston and Livezey, 1987).

The NAO index that was used in this work for the entire period 1941–2006 (Section 4.2.) was obtained from the Climatic Research Unit (CRU) (Jones et al. , 1997). This index will be indicated by NAO‐CRU. The correlation between the two NAO indices (i.e. from the two different sources) is higher than 0.9 for the period 1951–2006 (significant at the 1% level).

The influence of these modes of low frequency variability on seasonal temperature indices from mainland Portugal was examined with the standard Pearson correlation coefficient. The statistical significance of the correlations was checked using the two‐tailed Student's t ‐test at the 5% significance level.

4. Results

4.1. Observed seasonal changes in near‐surface air temperature

Specific air temperature indices (Table 2) were computed at the seasonal scale to investigate the temporal and spatial variability in the data and the existence (or not) of significant trends in the 1941–2006 period. We focused on the frequency and duration of air temperature extremes. The indices were analysed in different groups: four fixed temperature threshold indices [in number of days: TX25, TX35, TN20, and FD (frost days)]; four absolute indices (in °C: TXx, TXn, TNn, and TNx); four duration indices (in days: HWDI, CWDI, TXHW90, and TNCW10); and other indices (in °C: TX, TN, DTR, and ETR).

4.1.1. Overview

A summary of the overall results of the analysis of seasonal trends in air temperature indices is given in this sub‐section. For the 1941–2006 period Table 3 shows the number of stations exhibiting positive/negative trends and the corresponding number of significant trends at the 5% level; and Table 4 gives the average trend for the aggregated regional seasonal temperature indices and confidence intervals, together with the significance level of the results. For the sub‐periods 1945–1975 (cooling period) and 1976–2006 (warming period), Table 5 shows the seasonal trends in the regional indices and Table 6 gives the number of stations exhibiting positive/negative trends in extreme temperature indices, and the corresponding number of significant trends at the 5% level.

Table 3. Number of time series of air temperature seasonal indices with positive (+) and negative (−) trends in the period 1941–2006 and the respective number of statistically significant trends at the 5% level (in brackets)

Index	Spring		Summer		Autumn		Winter
	+	−	+	−	+	−	+	−
Mean TX	19 (7)	4 (0)	19 (11)	4 (0)	11 (2)	12 (1)	22 (14)	1 (0)
TX25	20 (3)	3 (0)	19 (7)	4 (0)	8 (1)	15 (1)	6 (0)	11 (0)
TX35	9 (0)	6 (0)	17 (7)	7 (0)	14 (0)	5 (0)	n/a	n/a
TXx	23 (2)	0 (0)	14 (3)	9 (0)	16 (0)	7 (0)	18 (3)	5 (0)
TXn	12 (0)	11 (1)	20 (9)	3 (0)	14 (1)	9 (0)	23 (11)	0
HWDI	16 (2)	7 (0)	18 (7)	5 (0)	10 (0)	10 (1)	9 (0)	6 (0)
TXHW90	20 (5)	5 (0)	19 (8)	4 (0)	9 (0)	14 (0)	18 (7)	5 (0)
Mean TN	19 (11)	4 (2)	23 (20)	0	22 (14)	1 (0)	20 (15)	3 (0)
FD	14 (0)	6 (0)	n/a	n/a	4 (0)	13 (4)	2 (0)	20 (10)
TN20	10 (3)	4 (0)	23 (13)	0	21 (5)	2 (0)	n/a	n/a
TNn	12 (1)	11 (1)	22 (12)	1 (0)	20 (7)	3 (0)	23 (13)	0
TNx	23 (9)	0 (0)	22 (10)	1 (0)	22 (7)	1 (0)	22 (8)	1 (0)
CWDI	1 (0)	8 (0)	n/a	n/a	3 (0)	13 (2)	4 (0)	18 (5)
TNCW10	7 (1)	16 (3)	4 (0)	19 (10)	3 (0)	20 (11)	8 (0)	15 (8)
ETR	21 (2)	2 (0)	6 (0)	17 (7)	8 (0)	15 (2)	6 (0)	17 (8)
DTR	11 (6)	12 (3)	10 (5)	13 (10)	5 (0)	18 (11)	9 (3)	14 (7)

• The indices are described in Table 2. ‘n/a’ means not applicable.

Table 4. Seasonal trends per decade (with 95% confidence intervals in brackets) for the regional indices of air temperature, in the 1941–2006 period

Index	Unit	Spring	Summer	Autumn	Winter
Mean TX	°C	0.18 (0.02 to 0.34)	0.21 (0.08 to 0.34)	0.08 (−0.06 to 0.22)	0.25 (0.17 to 0.33)
TX25	days	0.42 (−0.33 to 1.17)	1.26 (0.55 to 1.97)	−0.03 (−0.84 to 0.79)	n/a
TX35	days	0.01 (−0.05 to 0.06)	0.61 (0.08 to 1.14)	0.09 (−0.10 to 0.28)	n/a
TXx	°C	0.27 (−0.04 to 0.59)	0.07 (−0.12 to 0.26)	0.12 (−0.13 to 0.38)	0.13 (−0.04 to 0.30)
TXn	°C	0.10 (−0.09 to 0.29)	0.28 (0.11 to 0.45)	0.17 (−0.01 to 0.34)	0.43 (0.26 to 0.61)
HWDI	days	0.03 (−0.70 to 0.76)	0.31 (−0.12 to 0.74)	−0.18 (−0.53 to 0.16)	−0.02 (−0.16 to 0.11)
TXHW90	days	0.15 (−0.26 to 0.56)	0.14 (−0.12 to 0.39)	−0.10 (−0.39 to 0.18)	0.14 (−0.09 to 0.37)
Mean TN	°C	0.13 (0.03 to 0.22)	0.27 (0.18 to 0.36)	0.21 (0.11 to 0.31)	0.20 (0.06 to 0.35)
FD	days	−0.02 (−0.19 to 0.16)	n/a	−0.12 (−0.29 to 0.04)	−0.91 (−1.55 to −0.28)
TN20	days	0.02 (−0.00 to 0.05)	1.07 (0.67 to 1.47)	0.21 (0.08 to 0.33)	n/a
TNn	°C	−0.01 (−0.20 to 0.18)	0.27 (0.14 to 0.39)	0.21 (−0.00 to 0.42)	0.32 (0.15 to 0.49)
TNx	°C	0.27 (0.08 to 0.46)	0.23 (0.09 to 0.38)	0.22 (0.09 to 0.35)	0.23 (0.09 to 0.36)
CWDI	days	−0.03 (−0.06 to 0.01)	n/a	−0.12 (−0.23 to −0.01)	−0.18 (−0.40 to 0.05)
TNCW10	days	−0.04 (−0.18 to 0.10)	−0.20 (−0.33 to −0.07)	−0.26 (−0.43 to −0.09)	−0.20 (−0.45 to 0.05)
ETR	°C	0.28 (−0.06 to 0.63)	−0.20 (−0.38 to −0.01)	−0.09 (−0.41 to 0.23)	−0.18 (−0.40 to 0.04)
DTR	°C	0.05 (−0.05 to 0.16)	−0.06 (−0.13 to 0.01)	−0.13 (−0.24 to −0.02)	0.05 (−0.08 to 0.17)

• Significance levels: 5% bold; 25% italics. The indices are described in Table II . ‘n/a’ means not applicable.

Table 5. Seasonal trends per decade for the regional indices of air temperature, for the cooling (1945–1975) and warming (1976–2006) periods

Index	Unit	Spring		Summer		Autumn		Winter
		1945–1975	1976–2006	1945–1975	1976–2006	1945–1975	1976–2006	1945–1975	1976–2006
Mean TX	°C	−0.51	0.80	−0.26	0.74	−0.03	0.19	0.21	0.28
TX25	days	−2.47	3.07	−0.65	3.42	−0.35	0.65	n/a	n/a
TX35	days	−0.06	0.09	−1.43	3.08	0.10	0.15	n/a	n/a
TXx	°C	−0.29	1.41	−0.61	0.76	−0.07	0.42	−0.08	0.30
TXn	°C	−0.24	0.36	−0.38	0.89	0.14	0.11	0.73	0.15
HWDI	days	−2.31	2.36	−1.39	2.37	−0.12	0.02	−0.35	0.32
TXHW90	days	−1.30	1.53	−0.92	1.42	−0.42	0.32	−0.35	0.71
Mean TN	°C	−0.37	0.66	−0.12	0.70	−0.12	0.51	0.23	0.09
FD	days	0.43	−0.53	n/a	n/a	0.38	−0.42	−0.74	−0.71
TN20	days	−0.05	0.10	−0.68	3.02	0.18	0.25	n/a	n/a
TNn	°C	−0.24	0.34	−0.20	0.75	0.14	0.59	0.33	0.27
TNx	°C	−0.30	0.79	−0.24	0.77	0.13	0.37	0.17	0.19
CWDI	days	−0.07	0.00	n/a	n/a	0.16	−0.20	−0.33	−0.02
TNCW10	days	0.14	−0.33	0.14	−0.59	0.28	−0.62	−0.55	0.19
ETR	°C	−0.05	0.44	−0.42	0.01	0.28	−0.17	−0.41	0.03
DTR	°C	−0.14	0.14	−0.14	0.04	0.09	−0.32	−0.03	0.19

• Significance levels: 5% bold; 25% italics. The indices are described in Table II . ‘n/a’ means not applicable.

Table 6. Number of time series of extreme air temperature seasonal indices with positive (+) and negative (−) trends for the cooling (1945–1975) and warming (1976–2006) periods in the period 1941–2006 and the respective number of statistically significant trends at the 5% level (in brackets)

Index		Spring		Summer		Autumn		Winter
		+	−	+	−	+	−	+	−
TXx	1945–1975	2 (0)	19 (0)	1 (0)	20 (11)	6 (0)	15 (3)	4 (0)	17 (3)
	1976–2006	23 (21)	0	21 (11)	2 (0)	21 (0)	2 (0)	20 (0)	3 (0)
TXn	1945–1975	3 (0)	18 (6)	1 (0)	20 (9)	3 (0)	18 (2)	18 (0)	3 (0)
	1976–2006	23 (0)	0	23 (14)	0	16 (1)	7 (0)	23 (2)	0
HWDI	1945–1975	1 (0)	20 (3)	3 (0)	18 (5)	8 (0)	10 (2)	4 (0)	8 (1)
	1976–2006	23 (6)	0	22 (14)	0	8 (0)	11 (0)	13 (1)	2 (0)
TXHW90	1945–1975	0	21 (7)	0	21 (12)	8 (0)	13 (6)	6 (0)	15 (4)
	1976–2006	23 (5)	0	23 (13)	0	19 (1)	4 (0)	22 (1)	1 (0)
TNn	1945–1975	1 (0)	20 (8)	3 (0)	18 (5)	1 (0)	20 (7)	14 (0)	7 (3)
	1976–2006	17 (2)	6 (0)	23 (14)	0	23 (7)	0	23 (6)	0 (0)
TNx	1945–1975	2 (1)	19 (0)	3 (0)	18 (5)	11 (0)	10 (1)	17 (0)	4 (3)
	1976–2006	23 (22)	0	23 (10)	0	21 (4)	2 (0)	23 (0)	0
CWDI	1945–1975	5 (0)	3 (0)	n/a	n/a	7 (0)	4 (0)	4 (0)	16 (0)
	1976–2006	n/a	n/a	n/a	n/a	2 (0)	7 (0)	7 (0)	12 (0)
TNCW10	1945–1975	15 (5)	6 (0)	15 (3)	6 (2)	16 (6)	5 (0)	2 (0)	19 (3)
	1976–2006	5 (0)	18 (6)	0	23 (12)	1 (0)	22 (5)	11 (0)	12 (0)

• The indices are described in Table II . ‘n/a’ means not applicable.

The results in Tables 3 and 4 show a general warming trend in the full 66‐year period, for all the seasons, and it is associated with positive trends in the majority of the indices. But, overall, the trends in extreme indices do not have the same magnitude in all seasons. The overview of the regional average trends in all air temperature seasonal indices and the 95% confidence intervals is illustrated in Figure 2 for the 1941–2006 period (see also Table 4 for additional information). The results show warming trends in all the relevant indices in summer.

Figure 2

Overview of trends in regional temperature indices in the 1941–2006 period: averages over the 23 stations (black dots), with their 95% confidence intervals (grey shading), for summer (top) and winter (bottom). Definitions of indices, whose labels are on the horizontal axes, are given in the main text and Table II .

The results of the partial trend analysis reported in Tables 5 and 6 reveal that in the sub‐period 1976–2006 and over the four seasons, all the warm‐related indices show positive trends that are statistically significant in summer. Moreover, the cold‐related indices exhibit decreasing trends, that only in winter are not statistically significant.

4.1.2. Temperature means

4.1.2.1. Maximum and minimum mean temperature (mean TX and mean TN )

The intra‐annual variability in surface air temperature in mainland Portugal for all seasons is examined first in Figure 3, which shows the time series of regional average anomalies of the seasonal mean of daily maximum and minimum air temperatures. It is evident that from the mid‐1970s there is a general trend towards an increase in maximum and minimum temperatures (less evident in winter), even though the maximum temperature increases faster in recent decades (excluding autumn). The trends are very consistent and are also found in the time series of individual stations.

Figure 3

Inter‐annual variability of regional average maximum and minimum temperature anomalies (relative to 1961–90 reference period) for spring, summer, autumn, and winter. Superimposed are the piecewise trends calculated for the cooling (1945–1975) and warming (1976–2006) periods.

In the sub‐period 1945–1975, the maximum and minimum temperatures exhibit a decrease in all seasons, except winter, but the trend is only statistically significant in spring. In 1976–2006, all seasons show an increase in both temperatures that is statistically significant in spring and summer. The increasing trend is also statistically significant for the minimum temperature in autumn and for the maximum temperature in winter. Notably, four out of the five warmest summers occurred after 2000, with 2003 and 2005 ranking among the warmest three summers since 1941, and being associated with several heat waves that induced significant mortality (e.g. Trigo et al. , 2009) and total burned area from forest fires (e.g. Trigo et al. , 2006, Gouveia et al. , 2010).

All the stations record rising trends in the mean of daily minimum air temperature in summer, from 0.1 to 0.85 °C decade⁻¹, that are statistically significant for 87% of the stations; in the other seasons, the majority of stations indicate significant positive trends ranging from 0.11 °C decade⁻¹ (winter) to 0.83 °C decade⁻¹ (spring) for 57, 61, and 65% of the stations, respectively in spring, autumn, and winter.

For the mean daily maximum air temperature, significant increasing trends are observed in 30% of the stations in spring and 61% in winter; in autumn only two stations have significant positive trends.

4.1.2.2. Diurnal and extreme temperature range (DTR and ETR)

For the 1941–2006 period, the regional average anomalies series of the DTR index (Table 4) show a small and statistically non‐significant increasing trend in spring and winter (0.05 °C decade⁻¹), a statistically significant decrease in autumn (‐0.13 °C decade^‐1) and a non‐significant decreasing trend in summer. The individual stations show either positive or negative trends in DTR, except in autumn; in that season around 80% of the stations have decreasing trends that are significant for about 50% of the stations. For the 1976–2006 period the regional DTR index exhibits an increasing trend (not statistically significant) in spring, summer and winter, whereas autumn has a statistically significant decreasing trend (Table 5).

At the annual scale for the country as a whole there is a decrease in the DTR index due to the larger increase of the minimum temperature when compared with the maximum temperature; this has also been reported by Ramos et al. (2011). Because some of the weather stations are located in urban areas, we have examined whether the growth of the urban heat island effect could be responsible for a substantial fraction of the observed mean temperature increase; this matters because it is well known that local warming associated with the urban effect has greater importance at night (i.e. for minimum temperature). However, we found that positive significant trends of mean annual temperature in small cities and rural areas are similar to or larger than those observed in the metropolitan areas of Lisbon and Porto. Similarly, negative trends in the DTR index obtained for Lisbon and Porto data are, with a few exceptions, comparable in magnitude to those observed for such rural areas or small cities.

These findings are consistent with other studies. For example, Jones et al. (1990) and Easterling et al. (1997) concluded that urban effects on the 20th century globally and hemispherically averaged land‐air temperature time series do not exceed about 0.05 °C century⁻¹. In addition, Parker (2004, 2006) noted that warming trends in minimum night temperatures in the 1950–2000 period (for a set of about 270 stations worldwide), were not enhanced on calm nights, which would be the time most likely to be affected by urban warming. Thus, the global land warming trend is very unlikely to be influenced significantly by increasing urbanisation (e.g. Parker, 2006).

Tables 3 and 4 show that for 1941–2006 the ETR index (difference between the warmest maximum temperature and the coldest minimum temperature) reveals an increasing trend in spring, and the negative trends in other seasons that are only statistically significant in summer. But for the 1976–2006 period (Table 5), the ETR index exhibits a significant upward trend in spring and displays a negative trend only in autumn. Overall, the results suggest that there is a greater increase in the extreme minimum temperatures than in the extreme maximum temperatures.

In general, the signs of the ETR index trends match those found by Ramos et al. (2011) for cold days (TX10) and cold nights (TN10), and for warm nights (TN90) and warm days (TX90p). Our results are also consistent with the warming observed in the night time temperature by Alexander et al. (2006), Donat and Alexander (2012), and IPCC (2012).

4.1.3. Fixed threshold temperature events

4.1.3.1. Frost days (FD )

The results for the 1941–2006 period (Table 3) show a decreasing trend in the FD winter index in 87% of the stations (statistically significant for about 43% of the stations). For 57% of the stations a decrease is also observed in autumn although the trend is statistically significant only in four data stations, located in the northern region of the study area.

The trends in the FD winter index across mainland Portugal are shown in Figure 4 (bottom left), which also shows the corresponding inter‐annual variability in the regionally averaged anomaly series (Figure 4, bottom right). The regional trends reveal a decrease of frost days in all seasons, and are statistically significant (at 5% level) in winter (Table 4).

Figure 4

Trends in the TN20 summer index (top left) and FD winter index (bottom left) and corresponding inter‐annual variability of regional average of anomalies (right), for the 1941–2006 period. On the left, the dots are scaled according to the magnitude of the trend, expressed in days decade⁻¹: blue for decreasing trends (FD index) and red for increasing trends (TN20 index). On the right, superimposed on the time series are the piecewise trends calculated for the sub‐periods 1945–1975 (cooling) and 1976–2006 (warming).

4.1.3.2. Tropical nights (TN20 )

The most relevant season for the TN20 index is summer (Tables 3 and 4; and Figure 4). All weather stations have warming trends in this season, statistically significant for 57% of the stations; in autumn almost all the stations have positive trends, but statistically significant only for 22% of them.

Figure 4 shows, for summer and the 1941–2006 period, the inter‐annual variability in the regionally averaged anomaly TN20 series; the associated regional trend suggests an increase in the TN20 summer index of about 1.1 days decade⁻¹ (Table 4). But the time series are dominated by a very significant increase of tropical nights after 1976 (see also Table 5); in the 1976–2006 period there is an overall statistically significant positive trend in TN20 that in summer is more than two‐and‐a‐half times the 66‐year trend. In the 1941–1975 period (Table 5) the TN20 index shows non‐significant negative trends in both summer and in spring.

4.1.3.3. Summer days, very hot days (TX25 , TX35 )

For the period 1941–2006, Figure 5 shows the trends in the TX25 and TX35 indices across mainland Portugal and the inter‐annual variability and trends for the regionally‐averaged anomaly series.

Figure 5

Trends in the TX25 and TX35 summer indices (top) and corresponding inter‐annual variability of regional average of anomalies (bottom) for the 1941–2006 period. On top, the dots are scaled according to the magnitude of the trend, expressed in days decade⁻¹: blue for decreasing trends and red for increasing trends; there are no negative trends in these plots. At the bottom, superimposed on the time series are the piecewise trends calculated for the sub‐periods 1945–1975 (cooling) and 1976–2006 (warming).

As with the TN20 index, the largest significant positive trends in the TX25 and TX35 indices are in the summer with more than 70% of the stations having positive trends, which are statistically significant at the 5% level in about 30% of them (Table 3).

In summer, the regional trends suggest increases per decade of 1.3 summer days and 0.6 very hot days (Table 4) in the 1941–2006 period. In the 1976–2006 sub‐period (Table 5) the increasing trends are very strong and reach 3.4 days decade⁻¹ for TX25 and 3.1 days decade⁻¹ for TX35. These rates are considerably higher than the long‐term trend for the full period.

4.1.4. Temperature extremes

4.1.4.1. Warmest day, coldest night (TXx , TNn )

Overall, for the 1941–2006 period the changes are more pronounced in the extreme low temperatures (TNn) than in the extreme high temperatures (TXx), except in spring (Table 4). The regional average of the TXx seasonal index shows an increasing trend in all seasons, not statistically significant at the 5% level, while the regional average of the TNn seasonal index suggests a significant increase of 0.27 and 0.32 °C decade⁻¹ in summer and winter, over the 66‐year period. But for the 1976–2006 sub‐period the high extremes of daytime temperatures increased slightly more than the low extremes of night‐time temperatures, in particular in spring; autumn is an exception, when the reverse happens (Table 5).

Figure 6 shows the spatial distribution of station trends in the spring and summer TXx indices, for the 1945–1975 and 1976–2006 sub‐periods (see also Tables 5 and 6 for additional information). Before the 1976 change point there was a statistically significant decreasing trend in the TXx summer index with as many as 95% of the stations exhibiting decreasing trends (statistically significant in 52% of them). A statistically significant positive trend in the 1976–2006 sub‐period is found in spring (1.4 °C decade⁻¹) and summer (0.8 °C decade⁻¹).

Figure 6

Trends in the warmest day index, TXx , related to the maximum value of daily maximum temperature for spring (top) and summer (bottom). The plots are for the sub‐periods 1945–1975 (left) and 1976–2006 (right). The dots are scaled according to the magnitude of the trend, in °C decade⁻¹: red for increasing trends and blue for decreasing trends. There are only negative trends in the bottom left plot.

For the 1976–2006 sub‐period, Figure 7 shows the trend in the TNn index for all seasons: the trend is positive for all 23 series in summer, autumn, and winter (Table 6), and statistically significant at the 5% level respectively for 60, 30, and 26% of the stations. For the 1945–1975 sub‐period, more than 78% of the stations revealed negative trends in spring, summer, and autumn, which are significant in more than 20% of them (Table 6).

Figure 7

Trends in the seasonal coldest night index, TNn , related to the minimum value of daily minimum temperature, in the period 1976–2006: (a) spring (MAM ), (b) summer (JJA ), (c) autumn (SON ), and (d) winter (DJF ). The dots are scaled according to the magnitude of the trend, in °C decade⁻¹: red for increasing trends and blue for decreasing trends. All dots are red.

4.1.4.2. Coldest day, warmest night (TXn , TNx )

For the 66‐year period, the regional average of the TXn index suggests significantly increasing rates of 0.28 and 0.43 °C decade⁻¹, respectively in summer and autumn, while the regional average of the TNx index shows significantly increasing trends in all seasons (Table 4).

For the 1976–2006 sub‐period the 23 data series show warming trends in the TXn and TNx indices in spring, summer, and winter. No statistically significant TXn trends are found in spring, but almost all stations exhibit significant positive trends for the TNx index. In summer significant positive trends are found in 61% of the stations for TXn and 43% for TNx. In 1976–2006 the warming trends were larger in summer for TXn (0.89 °C decade⁻¹) and TNx (0.77 °C decade⁻¹) and in spring for TNx (0.79 °C decade⁻¹).

4.1.4.3. Heat waves, cold waves (HWDI , CWDI )

Relevant results of the seasonal trend analysis of heat and cold waves' indices are shown in Figures 8 and 9 for the period 1941–2006. Figure 8 shows the HWDI and TXHW90 spring and summer indices, and Figure 9 the CWDI and TNCW10 autumn and winter indices. The opposite sign of the trends in HWDI and CWDI both indicate relative warming in the region: positive trends in the HWDI index (and similarly, e.g. for TX25 and TX35, in Figure 5) and negative trends in the CWDI index (e.g. likewise for FD). These results suggest changes over time that lead to more intense and longer heat waves. At the same time, cold waves seem to have become shorter. As expected the decrease in the CWDI index occurred mostly in autumn and winter.

Figure 8

Trends in the seasonal heat wave duration index (HWDI ) and 90th percentile of heat wave index (TXHW90 ), in the period 1941–2006: left, spring; and right, summer. The dots are scaled according to the magnitude of the trend, in days decade⁻¹: blue for decreasing trends and red for increasing trends. All dots are red.

Figure 9

Trends in the cold wave duration index (CWDI ) and 10th percentile of cold wave index (TNCW10 ), in the period 1941–2006, for autumn (left) and winter (right). The dots are scaled according to the magnitude of the trend, expressed in days decade⁻¹: blue for decreasing trends and red for increasing trends. All dots are blue.

For the sub‐periods 1941–1975 and 1976–2006, the HWDI and TXHW90 regional indices show significant trends in spring and summer, albeit with opposite signs (Table 5): negative trends in 1941–1975 and positive trends in 1976–2006. For the relative warming period 1976–2006 and for spring and summer, an increasing trend was noted in the regional HWDI index of 2.4 days decade⁻¹ and in the regional TXHW90 index of more than 1.4 days decade⁻¹ (Table 5). The decreasing trend in the regional CWDI and TNCW10 indices was statistically significant at the 5% level only for autumn.

4.2. Large‐scale influence on seasonal temperature extremes

4.2.1. Regional average of the modes of low frequency variability correlations

Additional analysis was undertaken aiming to study the relationship between large‐scale modes of atmospheric circulation (described in Section 3.4.) and the regional averages of the air temperature seasonal indices (Section 3.2.). The analysis focused on those teleconnection indices that have been identified as producing most impact on temperatures in Portugal (e.g. Ramos et al. , 2010). Note that the correlations between the regional seasonal indices and the seasonal EA, EA/WR, and SCAND patterns were computed for the 1951–2006 period (56 years) because these indices are only available from 1951, whereas for the NAO‐CRU index the correlations are for the 1941–2006 period (66 years).

The results for all the seasons are given in Figure 10 where, for each mode, the indices based on the daily minimum temperature are grouped on the left‐hand side of the plot and the daily maximum temperature related indices are on the right.

Figure 10

Mean correlations between the EA , SCAND , EA /WR , and NAO‐CRU teleconnections patterns and 14 air temperature indices for all the seasons. For EA , SCAND , and EA /WR the period is 1951–2006 and for NAO‐CRU it is 1941–2006. The horizontal dashed lines indicate the levels for 5% significance. The air temperature indices on the horizontal axes are defined in Table II .

It is notable that the magnitude of the seasonal correlations is generally stronger for the indices focusing on the mean values of temperature and weaker for those that are related to extremes (i.e. that take into account the tails of the distributions). It is also evident that the magnitude of the correlations between the teleconnection indices and the temperature indices is generally weak and that the majority of the correlations are not statistically significant at the 5% level. But there are some exceptions for cold and warm extremes. These results are explored further in the following paragraphs.

The highest correlations (r  ∼ −0.6) were found between the mean maximum temperature index and the SCAND index, in spring and autumn. For this teleconnection pattern there is a statistically significant anti‐correlation between warm‐related indices (i.e. air maximum temperature extremes) in spring, summer, and autumn, with the correlation's coefficients ranging between –0.4 and –0.6. It is also worth noting that the correlation coefficients between the SCAND index and the tropical nights and warmest night indices in summer and autumn are r  < –0.45.

There is a significant anti‐correlation in autumn between cold‐related indices (FD, CWDI, and TNCW10) and EA, with the correlation coefficients varying between −0.4 and –0.5. On the other hand, the correlation between this pattern and some mean minimum temperature indices is positive and statistically significant (mean TN, TNn, TNx, and TN20). The most significant correlation coefficients between air temperature indices and the NAO‐CRU index are found in spring, in particular for warm‐related indices (mean TX, TXx, and TX25), with the highest values (r  > 0.45) being for mean TX and TX25.

To summarize, in general, the overall correlations between the modes of low frequency variability and the air temperature indices are relatively weak and non‐significant in winter. But significant correlations are found for cold extremes in autumn for the EA index, warm extremes in spring, summer and autumn for the SCAND index, and warm extremes in spring for the NAO‐CRU index.

4.2.2. Correlation between NAO‐CRU , EA , and SCAND and seasonal extreme temperature indices

The correlations between the NAO‐CRU, EA, and SCAND indices and the seasonal extreme temperature indices computed for the individual stations are discussed in detail in this section, particularly for the SCAND pattern. Table 7 shows the percentage of weather stations that have a statistically significant correlation at the 5% level between the air temperature indices and the NAO‐CRU, EA and SCAND indices.

Table 7. Percentage of the stations exhibiting a statistically significant correlation, at the 5% level, between air temperature indices and the NAO‐CRU , EA , and SCAND modes of variability

Index	NAO‐CRU (1941–2006)				EA (1951–2006)				SCAND (1951–2006)
	Spring	Summer	Autumn	Winter	Spring	Summer	Autumn	Winter	Spring	Summer	Autumn	Winter
Mean TX	83	4	65	61	26	9	61	35	96	91	96	0
TX25	83	4	30	n/a	4	0	17	n/a	74	74	100	n/a
TX35	0	9	5	n/a	7	23	21	n/a	0	59	53	n/a
TXx	78	0	4	30	13	9	26	9	22	78	78	4
TXn	35	0	17	26	17	30	13	9	48	30	9	0
HWDI	57	0	10	33	0	36	11	0	83	36	58	7
TXHW90	61	0	4	35	0	48	22	30	96	74	91	0
Mean TN	30	0	13	43	78	26	83	4	91	96	48	4
FD	10	n/a	0	22	5	n/a	59	0	10	n/a	0	4
TN20	14	4	0	n/a	7	17	18	n/a	0	57	41	n/a
TNn	22	0	9	13	9	17	52	17	35	17	0	9
TNx	17	4	4	17	17	17	35	9	13	65	70	13
CWDI	0	n/a	0	9	29	n/a	29	0	0	n/a	0	0
TNCW10	9	4	9	9	26	4	70	13	9	57	0	0
DTR	78	17	91	78	22	30	13	22	65	39	70	17

The NAO‐CRU is the teleconnection pattern that best explains the correlation between air temperature indices in winter, although the correlation coefficients were weak. This pattern also assumes some significance in spring, particularly for the TX25 and mean TX indices; the correlation coefficients were greater than +0.45 in more than 50% (TX25) and 40% (TX) of the stations.

The EA pattern also affects mainland Portugal, but the correlations are modest and affect mostly the cold extremes. We found that this index has a major impact on the air temperature indices in autumn: significant correlations are found between the EA index and the Mean TX, mean TN, FD, TNn, and TNCW10 indices for more than 50% of the locations.

On average the SCAND index is the mode that correlates best with the air temperature indices (see also Section 4.2.1. and Table 7) in summer, autumn, and spring. For the mean TN, mean TX, TXHW90, and TX25 seasonal indices, more than 70% of the stations have statistically significant correlations with the SCAND index, which are negative in all seasons; the exception here is the mean TN autumn index, with only 48%. Apart from a few exceptions, the correlations between the SCAND index and the air temperature indices are in general weak.

Figure 11 shows the correlations between the SCAND index and the mean TX, TX25 and TXHW90 indices for all the seasons in the 1951–2006 period.

Figure 11

Correlations between the SCAND index and the Mean TX , TXHW90 , and TX25 for all the seasons, in the 1951–2006 period. Red/blue dots indicate that positive/negative correlations are statistically significant at the 5% level. All dots are blue. The TX25 winter index plot is not shown because this case is not applicable.

There is a significant anti‐correlation in spring and autumn between the mean TX and SCAND indices (Figure 11): the correlation coefficients for the individual stations vary between −0.26 and −0.69, and are lower than −0.45 in more than 80% of the stations. For the duration of warm events (HWDI and TXHW90 indices) the correlations are predominantly negative and significant at the 5% level; however, for TXHW90 only in spring are the correlation coefficients lower than –0.45 in more than 50% of the stations.

There are also significant anti‐correlations between the TX25 and SCAND indices (Figure 11): in spring and summer these correlations are significant for 74% of the stations; in autumn all the stations show negative and significant correlations; in spring and autumn, the correlation coefficients are smaller than −0.45 in more than 40% of the stations.

The weak correlation between the extreme temperature indices for Portugal and the different teleconnections indices can be associated with the specific location of Portugal (western Iberian Peninsula). Most of the precipitation that occurs in the region is related to frontal systems associated with the storm track in the North Atlantic Ocean (Trigo, 2006), which depends highly on the NAO index (e.g. Trigo et al. 2008); this occurs especially in the extended winter, which is when most of the precipitation occurs in Portugal (e.g. Miranda et al. , 2002). Extreme temperatures are less influenced by large‐scale teleconnection patterns; in particular, it is well known that the NAO plays a minor role in shaping maximum and minimum temperatures in western Iberia (e.g. Trigo et al. , 2002).

The strongest results obtained here are for the SCAND pattern, which is compatible with the occurrence (or absence) of blocking patterns that control the SCAND phase (Barriopedro et al. , 2006). In fact, blocking events are known to control the occurrence of extreme temperature episodes in the western Mediterranean for both winter (e.g. Trigo et al. , 2004) and summer temperature (e.g. Carril et al. , 2008). Furthermore, sea surface temperature (SST) anomalies can also influence extreme temperature in Europe, including Iberia (e.g. Xoplaki et al. , 2003, Cattiaux et al. , 2011).

We have performed a preliminary analysis correlating the different teleconnections' annual indices with the geopotential height at 500 hPa at each grid cell of the NCEP/NCAR reanalysis for sub‐sets of 30 year periods. Results were compared by computing the difference between the second sub‐period (1976–2010) and the first sub‐period (1950–1975). Overall, this simple assessment shows that there is intensification in one of the centres of action of the SCAND pattern (not shown). The relevance of this pattern for extremes (Figure 10) and the location of a centre over the Iberian Peninsula may help to explain the changes in the temperature extremes recorded in the last few decades. In any case this result is only preliminary and further investigation will be conducted by the authors in the future to evaluate it.

5. Discussion and conclusions

Previous studies have focused on the analyses of trends in air temperature datasets for the Iberian Peninsula (e.g. Miranda et al. , 2002, Rodríguez‐Puebla et al. , 2010, Ramos et al. , 2011, Fernández‐Montes and Rodrigo, 2012, Sánchez‐Lorenzo et al. , 2012, de Lima et al. , 2013). In general, they have found a warming signal across the area that is consistent with the dominant global warming, with this behaviour being more intense after the mid‐1970s.

But topics such as seasonal trends and regional variations across mainland Portugal were not fully explored by those studies and required further attention. Thus, the main purpose of this study was to provide a more comprehensive discussion using many near‐surface air temperature indices and more stations than in some of the previous studies, and allowing the visualization of spatial patterns of trend.

In general, our results confirm the previous findings of a warming trend across mainland Portugal; however, our more in‐depth analysis has provided additional information and clarified the behaviour observed. For example, our study found a consistent pattern of trends in daily temperature extremes over the study area for the 1941–2006 period, which suggests increases in the maximum and minimum surface air temperatures; these trends are significant in all seasons except autumn for maximum temperature. Unlike the other three seasons, in recent decades autumn has recorded a cooling period in Europe (e.g. Klein Tank et al. , 2005). But when the analyses were conducted over the shorter 31‐year sub‐periods different patterns of variability were identified for the majority of the Portuguese territory. Negative trends in warm‐related temperature indices were dominant in the 1945–1975 period, whereas cold‐related temperature indices showed positive trends. In the subsequent 1976–2006 sub‐period this behaviour changed: overall, there were positive trends in warm‐related indices and negative trends in cold‐related indices. In particular, the most significant changes in air temperature in mainland Portugal were after 1976, when warming trends were in fact stronger, which is consistent with other studies focusing on the same region (e.g. Miranda et al. , 2002, Miranda et al. , 2006, Ramos et al. , 2011, de Lima et al. , 2013). The data indicate an increasing number of warm extremes, summer days, very hot days and tropical nights, and a decrease in the number of frost days. This is accompanied by an increasing trend in heat wave duration and intensity, for the majority of the stations, whereas at the same time cold wave duration decreased in this period.

The temperature increase in mainland Portugal seems to be accompanied by positive shifts in both tails of the temperature distribution, in particular in the last few decades. Empirical evidence of this finding is also provided by the diurnal temperature range index and the intra‐annual extreme temperature range index. But, overall, the changes in air temperature data are more pronounced in the warm‐related extremes than in the cold‐related extremes. This type of an ‘asymmetric’ warming, with the ‘warm tail’ warming faster than the ‘cold tail’, has been reported by Klein Tank and Können (2003) for Europe, for a similar observation period. These authors point out that this finding is more noticeable for Europe‐wide average, daily maximum and minimum temperatures in summer, and daily maximum temperatures in winter.

It is important to highlight that trend analyses of temperature extremes' indices for the 1976–2006 period indicate a widespread increase of warm extremes in all seasons, in particular in spring and summer, with more intense and longer heat waves, and concurrently a significant increase in the frequency of hot days and tropical nights; this may lead, for example, to an increase in the occurrence of forest fires and heat‐related mortality, and to a decline in water quality and crop yields. At the same time, a decreasing trend in cold extremes is found in all seasons, but it is not as strong as the increasing trend of warm extremes. The significant decrease in the frequency and duration of cold waves and in the frequency of frost days in the cold season will have a positive impact by reducing mortality and energy consumption and by increasing crop yields. These findings should be considered in local climate change mitigation and preventive measures.

With respect to the large‐scale influence on seasonal near‐surface air temperature extremes over mainland Portugal, the results confirm that cold‐related extremes have been associated with the East Atlantic pattern in autumn, whereas warm‐related temperature extremes have been associated more strongly with the SCAND teleconnection pattern. On the other hand, the most prominent Northern Hemisphere pattern, the NAO, that represents the most important mode of low‐frequency variability in the Northern Hemisphere, seams on the whole to exert a limited influence on mainland Portugal: it influences spring temperature, but not markedly, and it affects winter temperatures more than the other teleconnection patterns. The response to the SCAND pattern is stronger for spring, summer, and autumn.

The physical understanding of the intensified warming after 1976 is not yet well established. Walsh and Oliver (2011) found that temperature trends seem to be too large to be due only to changes in CO₂. Different physical mechanisms that can explain this warming in the last few decades can be found in the literature. For example, Compo and Sardeshmukh (2009) stated that recent worldwide land warming has occurred largely in response to a worldwide warming of the oceans rather than as a direct response to increasing GHG over land. Several authors believe that this warming, observed since the mid‐1970s, was associated with the prominent role played by large‐scale oceanic oscillations in the Pacific (PDO, Hurrell, 1996, Guilderson and Schrag, 1998) and Atlantic (AMO, Schlesinger and Ramankutty, 1994). In addition, Booth et al. (2012b) show that volcanic and aerosol processes can drive pronounced multi‐decadal variability and trends in historical north Atlantic SST patterns. Moreover, changes in the amount of sunlight received at the Earth's surface (commonly known as global dimming and brightening) can also enhance warming in both summer and winter (Wild, 2009). Some studies reported that since the mid‐1980s there is an increase in the surface solar radiation (brightening) at numerous stations in Europe and in the United States, as well as in parts of East Asia (e.g. Wild et al. , 2009).

At the European regional scale, the persistence of some specific atmospheric circulation patterns (i.e. over Europe) can play a role in the year‐to‐year variations of extreme indices and their trends (e.g. Kyselý and Domonkos, 2006; Kyselý, 2008). However, according to Efthymiadis et al. (2011), it is unlikely that atmospheric circulation variability alone can explain the enhanced trends in warm/hot extremes in the Mediterranean basin over the past few decades. Taking this into account, other authors have searched for possible feedbacks regarding the warming of SSTs (e.g. Xoplaki et al. , 2003, Cattiaux et al. , 2011, Hoerling et al. , 2013) and the relationship between extreme events and soil moisture in summer (e.g. Fischer et al. , 2007, Vidale et al. , 2007).

The empirical variation of trends in surface air temperature observed in this study confirms and highlights the climate variability in mainland Portugal. But extrapolations beyond the periods investigated should be approached carefully: on the one hand, the results obtained empirically might depend strongly on the period analysed; on the other hand, such extrapolations require a comprehensive understanding of larger scale climate dynamics and forcing. It is also pertinent that the findings discussed in this study confirm the well‐known asymmetries in the climate of mainland Portugal and suggest that they are likely to be aggravated.

Acknowledgements

The authors wish to thank Álvaro Silva and Sofia Cunha (Portuguese Sea and Atmosphere Institute, Portugal) for their help in processing the maps in Figures 2, 4-9, and 11. Alexandre M. Ramos was supported by the Portuguese Foundation for Science and Technology (FCT) through grant FCT/DFRH/SFRH/BPD/ 84328/2012. Comments by two anonymous referees are acknowledged.