Seasonal temperature trends on the Spanish mainland: A secular study (1916–2015)

RESEARCH ARTICLE Seasonal temperature trends on the Spanish mainland: A secular study (1916–2015) Dhais Peña-Angulo 1 | José Carlos Gonzalez-Hidalgo 2,3 | Leire Sandonís 2,3 | Santiago Beguería 4 | Miquel Tomas-Burguera 4 | Joan Albert López-Bustins 5 | Marc Lemus-Canovas 5 | Javier Martin-Vide 5 1 Instituto Pirenaico de Ecología, Consejo Superior de Investigaciones Científicas (IPE-CSIC), Zaragoza, Spain 2 Departamento de Geografía, Universidad Zaragoza, Zaragoza, Spain 3 IUCA, Universidad Zaragoza, Zaragoza, Spain 4 Estación Experimental Aula Dei, Consejo Superior de Investigaciones Científicas (EEAAD-CSIC), Zaragoza, Spain 5 Grupo de Climatología, Universidad de Barcelona, Barcelona, Spain Correspondence José Carlos González-Hidalgo, Departamento de Geografía, University Zaragoza, Campus San Francisco, Zaragoza 50009, Spain. E-mail: [email protected] Funding information Spanish Ministry of Economy and Competitiveness, MEC; Regional Government of Aragon, Geoenvironmental and Global Climate Change Research Group, Grant/Award Number: E02-17R; Spanish Government, Grant/Award Numbers: CGL2017-83866-C3-3-R, CGL2017-83866-C3-2-R, CGL2017-83866-C3-1-R Abstract Trends in seasonal mean values of maximum and minimum temperature are analysed in the Spanish mainland from the new MOTEDAS_century database. This new data set has been developed combining the digitalized archives from the Spanish Meteorological Agency (AEMET) with information retrieved from Annual Books published by the former Meteorological Agency dating back to 1916, and covers the period 1916–2015. In all four seasons, mean seasonal temperature of maximum (Tmax) and minimum (Tmin) increased. The raising occurred in two main pulses separated by a first pause around the middle of the 20th century, but differed among seasons and also between maximum and minimum temperature. Analysis of the percentage of land affected by significant trends in maximum temperature reveals two increasing phases in spring and summer for Tmax, and in spring, summer, and autumn for Tmin. However, winter Tmax only rose during the recent decades, and autumn Tmax in the first decades. Negative significant trends were found in extended areas in spring Tmax, and in spring, autumn, and summer Tmin, confirming the first pause around the 1940's–1960's. Trends of seasonal mean values of Tmax and Tmin are not significant for at least the last 25–35 years of the study period, depending on the season. The areas under significant positive trend are usually more extended for Tmin than Tmax at any season and period. Areas with significant trend expand and contract in time according to two spatial gradients: south-east to north-west (east-west) for Tmax, and west to east for Tmin. We hypothesize a relationship between atmospheric prevalent advection and relief as triggering factors to understand spatial and temporal differences in seasonal temperatures at regional scale during the 20th century in the Iberian Peninsula. KEYWORDS MOTEDAS_century database, seasonal trends, Spain, temperature maximum, temperature minimum Received: 15 September 2020 Revised: 7 January 2021 Accepted: 8 January 2021 DOI: 10.1002/joc.7006 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2021 The Authors. International Journal of Climatology published by John Wiley & Sons Ltd on behalf of the Royal Meteorological Society. Int J Climatol. 2021;1–14. wileyonlinelibrary.com/journal/joc 1 1|INTRODUCTION Most temperature trend analyses face the problem of generalized poor data availability, which becomes particularly noticeable for the first half of the 20th century. To better understand current warming, however, a secular context or longer must be adopted better frame the current trends in perspective. Several studies have shown that temperature evolution during the 20th century was not monotonic but characterized by periods of rising temperature and periods in which temperature remained stable or even decreased, usually referred to as ‘pauses’or ‘hiatus’(Easterling and Wehner, 2009; Fyfe et al., 2013; Kosaka and Xie, 2013; Dong and McPhaden, 2017). Temperature rise periods are usually identified between circa 1910–1940 and 1976–1997, while pauses have been reported between circa 1941–1975 and 1997–2013 (Folland et al., 2018). In the Western Mediterranean basin, and particularly in the Spanish mainland, there are only a few secular analyses, mostly carried out with a very low station density. Brunet et al. (2007) analysed the seasonal temperature evolution in 22 stations for the period 1850–2005. They found that the highest contribution to the annual mean increasing trend was in winter and autumn, with the rates of maximum temperature (diurnal time, Tmax) being stronger than those of minimum temperature (night-time, Tmin). They analysed the first rising period of the 20th century (1901–1949), also the first pause (1950–1972), and the second rising period (1973–2005). However, they did not provide any information regarding the second pause. They have shown, however, that temperature trends were not significant in several periods: in winter for Tmax and Tmin, and in summer and autumn for Tmin over 1901–1949. Furthermore, negative and significant seasonal trends were found from 1950 to 1972, and during the final period analysed (1951–2005), autumn Tmax and winter Tmin trends were not significant. Different results were found by Staudt et al. (2007) who detected and corrected the urban effect in 43 stations, and suggested that Tmin trend rates were higher than those of Tmax. Unfortunately, these authors neither study sub-periods, nor combined the stations into a regional series. Other articles analysing the secular trends at sub-regional scale (see review in Gonzalez-Hidalgo et al., 2015) agree with Brunet et al. (2007). This situation contrasts with the large body of research covering post-1950 years, either using weather stations (del Río et al., 2011, 2012), grids (GonzalezHidalgo et al., 2015; Herrera et al., 2015; GonzalezHidalgo et al., 2018) or average regional and sub-regional series (Guijarro, 2013; Gonzalez-Hidalgo et al., 2016). Generally speaking, these papers analyse the period since the end of the first pause, thus covering only the second period of temperature rise, and in some cases detect the second pause. Still, they do not provide the secular of the evolution of temperature in the region. Gonzalez-Hidalgo et al. (2015) presented a more thorough review of these studies. This paper presents an analysis of seasonal mean maximum and minimum temperature trends over the Spanish mainland, covering the period 1916–2015, and based on high spatial resolution information provided by the new MOTEDAS_century database. The main objective is to explore the spatial variations of seasonal trends of Tmax and Tmin along the study period, by analysing their behaviour at different temporal windows. 2|DATA AND METHODS The recently presented MOnthly TEmperature DAtaset of Spain (acronym MOTEDAS_century) combines information digitalized from the Annual Books (Libros Resúmenes Anuales, LRA) edited by the former Meteorological Agency of Spain, and digital data from the Banco Nacional de Datos del Clima (BNDC) of the National Meteorological Agency of Spain (AEMET). The data includes monthly maximum (Tmax, diurnal-time) and minimum (Tmin, nigh-time) temperature, and covers the 1916–2015 period. A general description of data rescue, station matching, quality control, and grid calculation are described in Gonzalez-Hidalgo et al. (2020), although the most critical information will be provided here. The resulting grid has been therefore obtained using the highest spatial information currently available, in particular for the period 1916–1949. Notwithstanding, the number of stations varies largely during the period of study, with a minimum of 228 in 1939 (end of the Spanish Civil War), and a maximum of 2030 in 1994. MOTEDAS_century offers a spatial resolution of 10 ×10 km. Data rescued from the LRA filled gaps and extended back the BNDC series, because about 30% of data rescued from the LRA were not included in the BNDC. Thus, LRA data increased significantly the information available for the pre-1950 period. Nevertheless, data series from both, BNDC and LRA, present some drawbacks because there are a large number of stations with very short temporal records. For instance, 3,969 (75.5%) series from a total of 5,259 have less than 30 years (see tables 1 and 2 in Gonzalez-Hidalgo et al., 2020). This characteristic is accentuated in the 1916–1950 period, when 13.2% of all stations (156) recorded data for only 1 year; 22% for only 2 years (259 stations); and 63% of the station (748) 2D. PEÑA-ANGULO ET AL. for less than 10 years. Particularly, the LRA series tend to be very short: 351 stations from a total of 1107 (31.7%) recorded less than 3 year, and more than 70% (807) recorded less than 10 years. The development of the grid was done considering that: (i) the number of stations changed year by year; and also (ii) that there were many changes in the location of some stations. In order to maximize the amount of information we decided to compute monthly fields independently using the total amount of data available each month, instead of reconstructing a selected number of time series. This approach is similar to the one described and used in the reanalysis project (Slivinski et al., 2020), and we applied it after validating its adequacy by comparing alternative procedures using filled (reconstructed) and non-filled series. The interpolation procedure was a modification of the model suggested by Brunetti et al. (2006) that combines distance weighting and azimuth, after converting the original data into anomalies to avoid elevation effects. To do so, the difference between the observed values and the mean climatology was computed, using the 1951–2010 climatology computed by Peña-Angulo et al. (2016). Preliminary quality control of raw data was done to avoid suspicious data. Detailed information on the advantages and drawbacks of the procedure can be found in GonzalezHidalgo et al. (2020). Seasonal time series were obtained for each grid cell both for Tmax and Tmin using the classic aggregation (December–January–February for winter, March–April– May for spring, June–July–August for summer, and finally September–October–November for autumn). The sign and significance of the temporal trend were calculated by the Mann–Kendall test (Mann, 1945) using a significance level of α=.05, after discarding any autocorrelation (see Gonzalez-Hidalgo et al., 2015), and the rate of the trend was estimated following Sen (1968). Rates are shown as C per decade (C/dec). It is generally accepted that significance and rate in temperature trends depend on the chosen temporal window and length. However, there is no agreement on the minimum length required to robustly discriminate a trending signal from natural variability or noise. In our case, we set the minimum temporal span at 20 years, largely in excess of the recommendations of studies such as those of Knight et al. (2009), who suggested a minimum of 15 years, or Santer et al. (2011), who suggested a minimum of 17 years. More recently Huang (2013) noticed that ‘the most detectable secular trend signals appear in the CO 2 band and the time it takes to see these radiance changes is much less than 12 years’, with similar opinions expressed by Loehle (2009), Liebmann et al. (2010) and McKitrick (2014). Following the aforementioned research, temporal variations in trends were analysed by using increasing temporal windows spanning a minimum of 20 years (1916–1935), until covering the entire period 1916–2015. Additionally, decreasing temporal windows were also used, starting from the 1916–2015 window to the most recent 1996–2015 one (again, using a minimum length of 20 years). Increasing and decreasing temporal window trend analyses enabled us to find the effects of selected periods and temporal lengths on the resulting trends. In the first case, the effect of increasing the size of the period of analysis is similar to the process of updating a database year by year. The results, therefore, take into consideration the cumulative effect on the trendonbothmagnitudeandsignificance.Whenthe temporal windows decrease, the effect is equivalent to decreasing the age of the records, and it informs on the most recent period, mostly the second pause, and identifies its beginning, temporal length, and spatial variation, among other features. Finally, 30-years moving windows were also used to avoid the effect of the length of the windows on the confidence intervals of the trends (size effect). 3|RESULTS 3.1 |Seasonal temporal evolution of Tmax and Tmin on the Spanish mainland 1916–2015 The time series of mean seasonal Tmax and Tmin (as anomalies over the period 1916–2015) for the whole Spanish mainland are shown in Figure 1. The main result is that Tmin shows lower temporal variability than Tmax, except during winter. The series shows a generalized increase in temperature during the study period, both in Tmax and Tmin and in the four seasons. It was easy to identify the different periods detected at a global scale. Also, we noted the extremely low temperature recorded in the summer of 1977. The highest trend rate for the 1916–2015 period corresponds to Tmax in spring (0.16C/dec ±0.08) and summer (0.13C/dec ±0.06). For Tmin the highest rates occurred in summer (0.13C/dec ±0.04) and autumn (0.11C/ dec ±0.05). Table 1 shows seasonal rates for different periods of regional series. Global Tmax rates are not significant for winter since the 1965–2015 window, and since 1945–2015 for Tmin. For spring the not significant periods are 1985–2015 and 1992–2015, respectively. For summer they are 1981–2015 and 1987–2015. For autumn there is no clear behaviour SPANISH SECULAR SEASONAL TEMPERATURE 3 in Tmin because rates have changed to significant over the last few decades, while no significant trend were detected in Tmax. The highest seasonal trend rates occurred around the 1970s, and in recent decades the rates of both variables decreased for all seasons. Tmax trend rates were higher than Tmin in spring and winter when they were significant. In summer and autumn, Tmin rates were usually higher than winter Tmax, and higher in summer. In general, the significance of rates remained for longer periods for Tmin than Tmax. 3.2 |Spatial variation in trends using 30-years moving windows The evolution of the surface percentage according to the trend sign and significance at 30-years moving windows FIGURE 1 Regional seasonal mean series (Spanish mainland) for Tmax and Tmin. Values as anomalies over 1916–2015 [Colour figure can be viewed at wileyonlinelibrary.com] 4D. PEÑA-ANGULO ET AL. is presented in Figure 2. Two main rising periods are identified, and also intermediate and a final pause. The periods differ between Tmax and Tmin for season and annual values. The first rise in Tmax affected less than 50% of land in spring, summer, and autumn, and did not occur in winter. On the other hand, the second rise was more extended in winter and started earlier than spring and summer, while being practically absent in autumn. Also the first pause is clearly detected in Tmax, when significant negative trends affected extended areas (>50%) in spring, and to a lesser extent in summer (c. 25%) and autumn (c. 10%). Meanwhile, the second pause is detected in the last decades in the four seasons. No clear evidences of winter rise are detected in Tmin during the study period, while two rising periods are identified in spring, summer, and autumn Tmin. During the first rising period the area under significant trend was around 25% (higher values than Tmax), and during the second positive phase it extended to almost the entire area in spring and summer, but also in autumn during some decades around the 1970's. Extended areas with significant negative trend have been identified (>75%) during the first pause in spring and autumn and to a lesser extent in summer, and no significant negative trend areas have been detected in winter. No significant trends have also been detected during the last decades in Tmin, again showing the recent pause over extended areas. These results suggest that the trend observed in the mean annual Tmax depend mostly on spring, summer, and autumn during the first rising period; and winter, spring and summer during the second one. For Tmin, on the other hand, spring, summer, and autumn defined the two rising periods. The spatial evolution of trends according to 30-year moving windows is presented in Figure 3. In Tmax the first positive period until the mid-1950's mostly affects the south-eastern areas in spring, summer and autumn, while the second period begun around the 1960's and lasted until 2010 at most, being more extended. The Figure also shows that, for more than 30 years, no significant trend is detected for Tmax. With respect to the first pause, the negative significant trend is only detected in spring over extended areas. The spatial evolution of Tmin trend sign and significance also shows two different positive phases, lasting the second in spring and summer until the final window (1986–2015) over extended areas and particularly to the east. No significant trends are found in any part of the study area in winter and autumn in the period TABLE 1 Selected seasonal Tmax and Tmin trend rates in different temporal windows (10-year intervals) Sig. Winter (±) Sig. Spring (±) Sig. Summer (±) Sig. Autumn (±) Maximum 1916_2015_100 *** 0.11 0.06 *** 0.16 0.08 *** 0.13 0.06 * 0.09 0.07 1926_2015_90 ** 0.12 0.07 * 0.13 0.10 *** 0.13 0.07 0.05 0.09 1936_2015_80 ** 0.13 0.09 * 0.15 0.12 *** 0.16 0.09 0.07 0.11 1946_2015_70 ** 0.15 0.11 * 0.19 0.15 *** 0.18 0.11 0.03 0.14 1956_2015_60 ** 0.17 0.13 ** 0.31 0.18 *** 0.27 0.14 0.13 0.17 1966_2015_50 0.13 0.16 *** 0.52 0.21 *** 0.33 0.17 0.07 0.25 1976_2015_40 0.04 0.22 *** 0.57 0.28 ** 0.41 0.27 0.03 0.31 1986_2015_30 −0.16 0.42 0.36 0.44 0.27 0.48 0.18 0.43 1996_2015_20 −0.51 0.57 0.36 0.89 0.53 0.78 0.69 0.81 Sig Winter (±) Sig Spring (±) Sig Summer (±) Sig Autumn (±) Minimum 1916_2015_100 * 0.08 0.07 ** 0.09 0.05 *** 0.13 0.04 *** 0.11 0.05 1926_2015_90 * 0.09 0.09 * 0.08 0.06 *** 0.13 0.05 ** 0.09 0.06 1936_2015_80 0.09 0.11 * 0.08 0.07 *** 0.15 0.07 * 0.09 0.08 1946_2015_70 0.09 0.12 ** 0.12 0.09 *** 0.18 0.08 * 0.10 0.10 1956_2015_60 0.09 0.16 *** 0.21 0.13 *** 0.27 0.09 ** 0.19 0.13 1966_2015_50 0.08 0.23 *** 0.45 0.13 *** 0.38 0.12 *** 0.31 0.17 1976_2015_40 −0.01 0.33 *** 0.48 0.16 *** 0.44 0.18 * 0.30 0.25 1986_2015_30 −0.06 0.46 ** 0.33 0.26 * 0.30 0.25 0.13 0.42 1996_2015_20 −0.59 0.66 0.06 0.53 0.30 0.52 0.65 0.77 Note: Significance expressed by (***) p<.001, (**) p<.01, and (*) p<.05. Rate values in C/dec. SPANISH SECULAR SEASONAL TEMPERATURE 5 1980–2015. Significant trends tended to appear in the central-western areas during the first positive period, while during the second one the area affected was more extended and shifted to the east. Extended areas of negative and significant trends are found around the period 1945–1985, when around 100% of the study area was affected by significant cooling. This period was more extended in spring than in summer and autumn, and mostly affected central-western areas. The percentage of land affected by significant trends during the two positives periods and the first pause is higher in Tmin than Tmax in the four seasons. The above results seem to be an improved descriptor of warming than simple annual mean rates from a regional series. 3.3 |Spatial variations in trends using increasing temporal windows Figure 4 presents the variation of the percentage of land affected by positive and negative, significant and nonsignificant trends under increasing temporal windows. The first temporal window (20 years, on the left side) corresponds to 1916–1935, and the last one (on the right side) includes the entire 1916–2015 period, thus there exists a cumulative effect along the chart that must be considered. The total area affected according to the sign and significance of the trend differs by seasons throughout the temporal windows. Globally, winter Tmax trends were positive and significant and affected more than 50% of the area after the 1916–1982 window, increasing progressively to cover up to 94% of the territory in 1916–2015. Spring and summer Tmax also showed a generalized positive trend in the entire period, in accordance with previous studies (see Brunet et al., 2007). In spring, the evolution of the percentage of land displays two pulses. The first one started in the temporal windows from 1916–1944 to 1916–1974, where positive and significant trends affected more than 50% of the area. Subsequently, given the effect of negative and significant trends detected in the first pause, no significant trends were detected until 1916–1987, when the area affected by positive trend in Tmax starts again to increase to cover 93.4% of the area at the end of the study period. The spatial evolution of land affected by positive and significant trends in summer was similar to spring except that, during the first pulse, the affected area was smaller. In the second raising pulse, starting in 1916–1994 up to the present, the area affected by positive and significant trends covered the study area almost completely. The positive and significant trends in summer affected more than 20% of the area in the intermediate period, whereas it was close to 0% in spring. Evolution of the area affected by positive and significant trends in autumn differed: two pulses were detected in which the area under positive and significant trends increased (similar to spring and summer), but at no time covered more than 50% of the area (only one pulse of increase). The results suggest that the use of annual mean values of Tmax may mask important differences among FIGURE 2 The surface percentage under positive/negative trend and significance (p<.05); 30-years moving windows. Tmax (left), Tmin (right) [Colour figure can be viewed at wileyonlinelibrary.com] 6D. PEÑA-ANGULO ET AL. seasons and along time. Temperature rise until the 1970's, for instance, seems to be driven mostly by the evolution of spring and summer temperatures and, to a lesser extent, autumn. On the other hand, temperature evolution during the last few decades seems to be related with all seasons, albeit a lesser contribution of autumn. The temporal evolution of Tmin was quite similar in spring, summer and autumn. All of these seasons show Tmin increasing in two pulses throughout the 20th century, with a high percentage of land affected by positive and significant trends (>50% of total land in the first case and >75% in the second one). Also, the first pause is clearly represented with negative significant trends detected in spring and to a lesser extent in summer. Winter, however, showed a different behaviour again, and the percentage of land affected by positive and significant trends was less than 10% until 1990's, and the final maximum value reached 50%. These results agree with those previously presented and show that the first rise in Tmin was mostly driven by spring, summer, and autumn and the second one by winter, spring, summer, and autumn. Figure 5 shows a sequence of maps with time steps of 5 years, and the percentage of land affected by positivesignificant trends. Generally, the percentage of land affected by significant positive trends is higher for Tmin than Tmax in both periods of temperature rise, and particularly in summer and autumn. The figures show the importance of conducting detailed spatial analyses, because regional mean rates as those presented in Table 1 hide the rich spatial variability of temperature trends in the region. In Spanish mainland significant positive Tmin trends have been more extended and prolonged than Tmax in both rising periods. Two gradients of spatial variation dominate the development of trends along the temporal windows. The areas affected by positive-significant trends in Tmax expand FIGURE 3 Spatial variation in seasonal mean trends in Tmax and Tmin according to 30-years moving windows (5-year intervals). The figure includes the percentage under positive/negative predominant significant trend (Mann–Kendall test p<.05) [Colour figure can be viewed at wileyonlinelibrary.com] SPANISH SECULAR SEASONAL TEMPERATURE 7 FIGURE 4 The surface percentage under positive and negative trends and their significance (p<.05); increasing temporal windows. Tmax (left), Tmin (right) [Colour figure can be viewed at wileyonlinelibrary.com] 1916 1940 1916 1945 1916 1955 1916 1960 1916 1950 1916 1965 1916 1970 1935 1916 1916 1995 1916 2000 1916 2005 1916 2010 1916 2015 1916 1975 1916 1985 1916 1980 1916 1990 8% 48% 70% 54%58% 74% 45%1% 28% 42% 55% 75% 93%8% <1%0% 10% 0% 15% 29% 26%46% 29% 32%2% 52% 56% 77%82%89%29% 30%23% 39% 1% <1% 48% 16%62% 20% 21%<1% 41% 31% 28%32% 57%14% 44%8% 74% 2% 0% 0% 0%0% 0% <1%<1%61% 98% 99% 93% 94%<1% 38%11% 50% 11% 29% 65% 40%91% 46% 51%3% 65%80%92%97%100%35% 40%30% 57% Annual Winter Spring Summer Autumn Tmax <1% 0% 1% 4%<1% 7% 8% <1% 7% 22%41% 53% 58%1% 6%8% 10% 4% 48% 55% 64%61%75%54%1% <1%7% 7% 34% 60% 81%3%5% 19%<1%16% 11%<1% 8% 32% 58% 48%63% 59% 54%31% 58% 80% 92% 98% 99%12%1% <1%5%<1%6% 15% 22% 24% 69% 62%71% 71% 70%4% 63% 67% 82% 82% 94%11% 17% 1%1% 62% 18% 30% 63% 63%64% 69% 66% <1% 53% 76% 86% 93% 100% 12%6% 7%6%5%10% 42%<1% Annual Winter Spring Summer Autumn Tmin –0.05 +0.05– +–0.01 +0.01 FIGURE 5 Spatial variation in seasonal mean trends in Tmax and Tmin according to increasing temporal windows (5-year intervals). The figure includes the surface percentage under positive/negative significant predominant trend (Mann–Kendall test p<.05) [Colour figure can be viewed at wileyonlinelibrary.com] 8D. PEÑA-ANGULO ET AL. and contract from the east (Mediterranean coastland) to the west (inland), while the reverse is true for Tmin from west Atlantic-coastland to the east-inland. The two opposing gradients can be detected during the rising periods both in Tmax and Tmin in spring, summer, and autumn, while only the second one prevails in winter. The first pause is apparent in a wide area in spring Tmax, when positive significant trends were practically absent, whereas a small significant positive trend area remained in the eastern coastland in summer and autumn. For Tmin, spring and summer showed a significant cooling period mostly located to the north-east, longer-lasting in spring than in summer. 3.4 |Spatial variation in trends using decreasing temporal windows Figure 6 shows the temporal evolution of the percentage of land with significant positive and negative trends under decreasing temporal windows. From left to right, the plots confirm a global rise in temperature during the 20th century in Tmax at the annual and seasonal scales. The presence of the second pause is, however, very marked, since temperature ceased to rise significantly in all four seasons. Taking the 20% of total land affected by a significant trend as a threshold, the starting date of the second pause can be identified precisely for Tmax: in winter in 1965–2015; in spring and summer in 1985–2015 and 1983–2015. In autumn, the area affected by significant trends is less than 20% since 1933, but there is a brief recovery period around 1983. The same is true for Tmin. However, the winter pause started very early (1946–2015), while dates for the onset of the hiatus were 1992–2015 for spring, and 1988–2015 for summer; the autumn pause started in 1978–2015. 4|DISCUSSION 4.1 |General comments It has been assumed that global temperature rise has been controlled by the evolution of Tmax, mostly in winter (Jones et al., 1999). However, the negative trend of the diurnal temperature range since mid-century does not confirm this hypothesis (Rohde et al., 2013; Sun et al., 2018a,b). On the other hand, discrepancies arise in regional studies at the seasonal scales as those presented here. There is detailed information for mainland Spain on temperature evolution from 1950 onward, characterized by a global rise, mostly in summer and spring (Brunet et al., 2007; del Rio et al., 2011, 2012; Rios et al., 2012; Guijarro, 2013; Gonzalez-Hidalgo et al., 2015). However, there is no agreement on whether the increase was stronger for Tmax or Tmin. On the other hand, less information exists for the first decades of the 20th century. Our results show that, globally, seasonal mean values of Tmax and Tmin on the Spanish mainland rose between 1916 and 2015, in a similar sequence of periods as in the global data set and in previous research in the study area (see Brunet et al., 2007). Also, the results confirm that differences exist among Tmax and Tmin at the FIGURE 6 The surface percentage under positive/negative trend and significance (p<.05); decreasing temporal windows. Tmax (left), Tmin (right) [Colour figure can be viewed at wileyonlinelibrary.com] SPANISH SECULAR SEASONAL TEMPERATURE 9