pan european seasonal trends and recent changes of drought frequency and severity

From these series drought events were defined at seasonal scale and trends of frequency and severity of droughts and extreme droughts were analyzed for the periods 1950- 2015 and 1981-20

PII: S0921-8181(16)30180-1

Please cite this article as: Spinoni, Jonathan, Naumann, Gustavo, Vogt, J¨ urgen,

Pan-European seasonal trends and recent changes of drought frequency and severity, Global

and Planetary Change (2016), doi: 10.1016/j.gloplacha.2016.11.013

This is a PDF file of an unedited manuscript that has been accepted for publication.

As a service to our customers we are providing this early version of the manuscript The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

frequency and severity

Jonathan Spinoni1,2 , Gustavo Naumann1, Jürgen Vogt1

[1] European Commission, Joint Research Centre, Ispra, ITALY

[2] G.F.T Italy, Milano, ITALY

Correspondence to other authors

Jonathan Spinoni – jonathan.spinoni@gmail.com

Gustavo Naumann – gustavo.naumann@jrc.ec.europa.eu

ACCEPTED MANUSCRIPT

In the last decades drought has become one of the natural disasters with most relevant impacts in Europe and this not only in water scarce areas such as the Mediterranean that are inclined to such events As a complex natural phenomenon, drought is characterized by many hydro-meteorological aspects, a large variety of possible impacts and definitions This study focuses on meteorological drought, investigated by using indicators that include precipitation and potential evapotranspiration (PET), i.e the Standardized Precipitation Index (SPI) and the Standardized Precipitation-Evapotranspiration Index (SPEI) These indicators account for the lack of precipitation and the drying effects of hot temperatures and in this study have been computed for short-accumulation periods (3-month) to capture the seasonality of droughts The input variables, monthly precipitation and temperature for 1950-2015, stem from daily gridded E-OBS data and indicators were computed on regular grids spanning over the whole of Europe PET was calculated from minimum and maximum temperatures using the Hargreaves-Samani formulation Monthly precipitation and PET have then been used to compute the SPI-3 and the SPEI-3 time series From these series drought events were defined at seasonal scale and trends of frequency and severity of droughts and extreme droughts were analyzed for the periods 1950-

2015 and 1981-2015 According to the SPI (driven by precipitation), we found a statistically significant tendency towards less frequent and severe drought events over North-Eastern Europe, especially in winter and spring and a moderate opposite tendency over Southern Europe, especially in spring and summer According to the SPEI (driven by precipitation and temperature), Northern Europe shows similar wetting patterns, while Southern and Eastern Europe show a more remarkable drying tendency, especially in summer and autumn Both for frequency and severity, the evolution towards drier conditions is more relevant in the last three

During recent decades and as an implication of global warming (IPCC Fifth Assessment Report: Pachauri et al., 2014), a general tendency towards drier conditions (Trenberth et al., 2014) and higher frequencies of extreme climatic events (Trenberth, 2011; Trenberth et al., 2015)

ACCEPTED MANUSCRIPT

contributed to make drought a topic high on the scientific and political agendas (Mishra and Singh 2010) Following the criticism raised by Sheffield et al (2012) at previous drought incidence studies that reported a remarkable global drought increase in the last decades (e.g., Dai, 2011), the effective increase or decrease in the frequency and severity of drought events at global level was, however, debated without general consensus (Dai, 2011; Sheffield et al., 2012; Dai, 2013; Spinoni et al., 2014; Trenberth et al., 2014) The conclusions of Sheffield et al (2012) supported the results of other studies, showing declines in evaporation over an extended period

of time while temperatures were increasing (Fu et al., 2009; Jung et al., 2010), and were recently confirmed by other authors (Yin et al., 2014; Hobbins et al., 2016; McEvoy et al., 2016)

It is important to underline that this is also due to the variable spatial homogeneity and scale of drought events, which may be very limited, especially for very severe events (e.g., the Northern England drought in 1995; Buckland et al., 1997), or may involve very large areas such as was the case for Central Europe in 2003 (Rebetez et al., 2006), and for Russia in 2010 (Wegren, 2011) Spatially extended cases may even span over several years, as was the case for Australia in 2002-03 (Horridge et al., 2005) or the conterminous United States in the 1930s and in the 1950s (Schubert et al., 2004) Another reason behind the different opinions about drought trends at global scale is the great number of drought indicators commonly used to study drought patterns and evolution, depending also on the drought characteristics and sectorial impacts (Vogt and Somma, 2000; Heim, 2002; Hayes et al., 2011; Mishra and Singh, 2011; Zargar et al., 2011; Sepulcre et al., 2013; Staudinger et al., 2014)

With respect to the European continent, however, there is a well-established agreement that trends over the last decades show drying patterns in the Southern regions (Briffa et al., 2009; Vicente-Serrano et al., 2014), especially in the Mediterranean area (Sousa et al., 2011; Hoerling

ACCEPTED MANUSCRIPT

et al., 2012), mixed patterns in Central and Eastern Europe (Spinoni et al., 2013), and a trend towards wetter conditions in Northern and North-Eastern Europe (Bordi et al., 2009; Seneviratne, 2012; Damberg and AghaKouchak, 2014) Spinoni et al., (2015a) using a high spatial resolution dataset for the period 1950 to 2012 found that Scandinavia, Belarus, Ukraine, and Russia experienced a decrease in drought variables due to precipitation increase, while Central Europe and the Balkans experienced a moderate increase, mainly due to a temperature increase, and South-Western Europe, in particular the Mediterranean area, experienced a remarkable increase of drought frequency and severity driven by both a precipitation decrease and a temperature increase Such increase became even more prominent in the last decades, i.e from 1981 onwards

Because drought and climate processes are very complicated processes involving interactions among ocean processes (ocean teleconnections), land-based processes (the differences between precipitation, evaporation rates, and runoff), and several atmospheric processes such as atmospheric transparency (which affects atmospheric dimming and brightening and thereby evapotranspiration rates, humidity of the receiving air stream, and wind speed), it is desirable to have a maximum of information available for the analyses, possibly including all the mentioned forcing factors However, when looking at long term trends (66 years in our case), one can hardly reconstruct the details of every single drought episode, because many climatological variables are not easily retrievable at high spatial resolution for such long records Consequently,

in this study we focus on the meteorological aspects of drought, using drought indicators calculated from meteorological variables Since the E-OBS dataset (Haylock et al., 2008) includes only precipitation and air temperature amongst the cited variables, we selected the Standardized Precipitation Index (SPI; McKee et al., 1993, 1995) and the Standardized

The current study is a companion study to Spinoni et al (2015a), in which drought trends were investigated at the annual time scale, using SPI-12 and SPEI-12; here, the study is analyzing the seasonal 3-month scale using SPI-3 and SPEI-3 In Spinoni et al (2015a), we drew attention to the relative roles of precipitation and evapotranspiration for drought onset and evolution while in this study we separate the precipitation factor from the evapotranspiration factor by separately mapping results for SPI-3 and SPEI-3 The form of SPEI-3 presented here is driven by temperature inputs that are used to estimate solar radiation on one hand side and precipitation on the other hand side As input data, we used versions 11 and 12 of the E-OBS grids (Haylock et al., 2008), which have the same horizontal spatial resolution (0.25º) of the earlier versions, but extend the time interval to 1950-2014 (version 11) and 1950-2015 (version 12) and improve the quality and quantity of input data Moreover, besides analysing seasonal trends in drought

by investigating how seasonal drought trends have evolved over Europe from 1950 to 2015 during winter, spring, summer, and autumn These seasonal patterns are also very important when stakeholders and policy makers have to deal with drought impacts, which greatly vary depending on the season and on the sector involved, as discussed for example by Ciais et al (2005), Naumann et al (2015), and Stagge et al (2015a)

This paper is structured in four sections, including this introduction Section 2 describes the quality check and homogenization procedures that were applied to the input data, as well as the theoretical background to the computation of PET and the indicators SPI and SPEI, and details

on how we defined and computed the frequency and severity trends of drought and extreme drought events Section 3 presents the results in the form of maps and discusses drought trends separately for each season, with a particular focus on spatial patterns and statistical significance

of the trends In this section we also briefly compare our results with those reported in the literature Section 4 concludes the paper by summarizing the main outcomes and introducing

ACCEPTED MANUSCRIPT

possible improvements and planned research activities especially concerning projections of seasonal drought events until the end of this century using different emission scenarios (see, e.g., Spinoni et al., 2015c; Stagge et al., 2015b)

2 Data and Methods 2.1 Input data

As in the previous studies (Spinoni et al., 2015a, 2015b), we have used the E-OBS daily grids (Haylock et al., 2008) of temperature (minimum, TN; maximum, TX) and precipitation (P) as initial input data The spatial resolution is 0.25º x 0.25º, the area of interest spans the entire European continent (excluding Madeira, Azores, Canary Islands, and Arctic territories) and the time interval ranges from 1950 to spring 2015 Since E-OBS version 12 ranges from January

1950 to June 2015, our analyses end with the complete climatological spring 2015 (March-May) The E-OBS grids have been applied in a large number of scientific studies regarding, e.g., climate change (Hofstra et al., 2009), climate extremes (Kostopoulou et al., 2013), agro-meteorology (Harding et al., 2015), drought (Gudmundsson and Seneviratne, 2015; Roudier et al., 2015; Spinoni et al., 2015a, 2015b), though they have sometimes been criticized because they tend to smooth the extremes due to the interpolation model and the lack of input stations in some regions (Kysely and Plavcova, 2010)

To convert them into monthly inputs for the drought indicators, we transformed the daily values into monthly averages for temperature, and sums for precipitation, allowing for a maximum of three missing temperature values or one missing precipitation value for any given month Unfortunately, in the creation of the E-OBS grids, data which can be considered non-homogeneous might also enter the gridding routines This is necessary because gridding daily

ACCEPTED MANUSCRIPT

maps requires a very high density of stations and this might lead to calculating trends contaminated by station relocations, urban growth and similar problems that may not reflect a climate-only signal Thus, we performed quality checks on the monthly series, including a stepwise methodology to detect and eventually correct for outliers (see Caussinus and Mestre,

2004, for details) and a series of homogeneity tests using the Multiple Analysis of Series for Homogenization (MASH, version 3.03; see Szentimrey, 1999), being aware that this might not completely exclude spurious signals deriving from daily input series Only a minor fraction of monthly data did not pass all the tests (0.7%), mainly located in Turkey, Scandinavian Mountains, Russia, and Latvia When possible, older versions (v11, but also v10) of the E-OBS were used to substitute the suspect data, before re-performing the quality tests, but the total number of points that in the end failed the tests only marginally diminished to 0.6% All the points that failed the tests for temperature or precipitation have been left out from calculations However, there is a regional pattern that deserves some special attention: in Andalusia (Southern Spain) a few points of version 11 and 12 show statistical parameters very close to fail the homogeneity tests from the mid-2000s onwards, in particular regarding minimum temperatures The same points did not risk failing these tests in version 10 While this might be caused by several factors, it probably depends on the use of new stations to create version 11 of the E-OBS

in Southern Spain This hypothesis is supported by comparisons made by the EURO4M project: while there are no noticeable differences between v 11 and v 12 neither for precipitation, nor for temperature (http://cib.knmi.nl/mediawiki/index.php/Compare_E-OBS_v12.0_and_v11.0), both variables show differences between v10 and v11, i.e Andalusia is drier and hotter according to v11 (http://cib.knmi.nl/mediawiki/index.php/Compare_E-OBS_v11.0_and_v10.0), even though

in our tests the precipitation series in that region never showed a suspect behavior, just a slight

ACCEPTED MANUSCRIPT

decreasing trend in the last two-three decades However, it must be reported that Andalusia – in particular Almeria and its surroundings – might be experiencing a human-induced anomalous local climate effect caused by the large number of greenhouses, large to the point that this area is nicknamed ―sea of plastic‖ (Campra and Millstein, 2013)

2.2 Computation of Potential Evapotranspiration (PET)

The choice of the E-OBS grids is further justified by the fact that this study is a sequel of the previous study based on annual droughts (Spinoni et al., 2015a) This choice, however, also has a major drawback, as the E-OBS grids do not include variables – in particular solar radiation, vapor pressure, and wind speed – that would allow computing potential evapotranspiration (included in SPEI) with more sophisticated methods as the Penman-Monteith method (Allen et al., 1998) The latter is considered more complete and suitable for drought studies (Sheffield et al., 2012) than those based on temperature only Indeed, in this study monthly temperature data have been used to compute PET using the Hargreaves-Samani equation (Hargreaves and Samani, 1982; Hargreaves and Allen, 2003), which makes use of minimum and maximum temperature to account for solar radiation (see also Samani, 2000)

In previous works dealing with drought (e.g., Spinoni et al., 2015a, 2015b) we opted for the Thornthwaite equation (Thornthwaite, 1948) to compute PET and we reported its use in Europe and outside of extremely arid areas (van der Schrier et al., 2011), as well as the drawbacks intrinsically included in such formulation This time we opted for the Hargreaves-Samani equation because in Europe drought indicators based on minimum and maximum temperature proved to perform better than those based on mean temperature only (Vanderlinden et al., 2004; Weiss and Menzel, 2008; Shahidian et al., 2012) In Figure 1, we show the winter and summer

ACCEPTED MANUSCRIPT

PET values averaged over the period 1950-2014, and the corresponding linear trends, following three different formulations: Hargreaves-Samani, Thornthwaite using mean temperature as input, and modified Thornthwaite using maximum temperature as input Figure 1 demonstrates differences in the climatologies of PET derived from Hargreaves-Samani and from Thornthwaite A more detailed comparison can be found in Weiss and Menzel (2008), who compared global climatologies of PET according to four methods In Figure 1, the differences in the average PET over the period 1950-2014 are more evident in winter than summer, especially

in the British Islands and Russian territories On a spatial basis, the trends – which are more important than average values for our analysis – show slightly different absolute values, while the three models agree in sign for almost the whole of Europe

Taking as a reference the Penman-Monteith estimation of PET we evaluated the performance of the Hargreaves-Samani and Thornthwaite models for a subset of 48 European stations from the ECA&D dataset (http://www.ecad.eu) The scarce density of stations with at least 15 years of reporting of the meteorological variables that are needed to compute PET according to Penman-Monteith prevented us to perform a full continental analysis In Figure 2 we show PET computed

by Hargreaves-Samani and Thornthwaite as differences to PET computed by Penman-Monteith Figure 2 shows monthly series from 1981 to 2010 for six stations representing different climatic regions: for five stations out of six (the exception is La Coruña) the Hargreaves-Samani model estimates PET better than Thornthwaite, considering Penman-Monteith as reference method, and this is also valid for 43 stations out of the 48 being part of the subset analyzed From Figure 2, one can further notice that in Northern Europe the use of minimum and maximum temperature in the Hargreaves-Samani approach strongly reduces the seasonality (i.e., overestimation in hot months and underestimation in cold months) usually introduced by the Thornthwaite model

ACCEPTED MANUSCRIPT

However, such seasonality is still present, though less prominent, in the PET series for Dublin Another interesting feature is the more prominent climate-change signal dependency of the Thornthwaite approach in hot regions such as the Mediterranean, where the climate change signal has been strong in the last decades (e.g., Diffenbaugh et al., 2007)

Thus, though the use of the Hargreaves-Samani equation (Hargreaves and Allen, 2003) can be subject to similar criticisms as for the use of the Thronthwaite equation, we can conclude that for our study the Hargreaves-Samani equation – which makes use of minimum and maximum temperature to estimate solar radiation – is more suitable than the Thornthwaite equation to estimate PET While the Penman-Monteith methodology is widely considered the most realistic approach to estimate PET (Trenberth et al., 2014), one should also take into account that it is based on some assumptions that may not perfectly fit drought situations The Penman-Monteith equation holds under the assumption of a reference vegetation and ample supply of moisture (Allen et al., 1998, 2006) During drought events the latter assumption becomes questionable, since the relative humidity is rapidly reduced, the air temperature soars and both have the effect

of reducing the supply of moisture, which in turn will introduce a bias in the calculation of PET

2.3 Drought indicators and drought variables

Amongst all the indicators applied to meteorological droughts (Hayes et al., 2011), we selected two of the most used globally and in Europe, i.e the SPI (e.g., Lloyd-Hughes and Saunders, 2002) and the SPEI (e.g., Vicente-Serrano et al., 2012) We decided to separately study the drought events as depicted by the SPI (input: precipitation) and by the SPEI (input: precipitation minus PET) in order to account for the influence of rising temperatures, keeping in mind that the results derived from the SPEI should be considered the ―upper bound case‖, i.e depicting more

et al., 2015c; Vicente-Serrano et al., 2015), but according to our tests and taking into account that the best distribution strongly depends on the local climatic features, we assumed that the log-logistic option was more suitable for our scopes

For each grid point (0.25º x 0.25º) of the study area we computed monthly series of the SPI-3 and the SPEI-3 from March 1950 to May 2015 For each year, we considered the four climatological seasons: winter from December to February (DJF), spring from March to May (MAM), summer from June to August (JJA), and autumn from September to November (SON)

As suggested by McKee et al., (1993), a drought event is considered every time the SPI-3 (and, respectively, the SPEI-3) falls below -1 (one standard deviation from the mean) for at least two consecutive months, and it ends when the indicator returns above 0 If, during the event, the indicator reaches at least once a value below -2, it is considered as an extreme drought event (see Section 4)

A drought (or extreme drought) event may obviously last more than three months, so it can encompass two or more seasons In this case the event was assigned to the season in which it peaks, i.e shows the lowest values between the starting and the ending month Consequently, for each point and for each season, the drought (or extreme drought) events were recorded for every year from 1950 to 2015 for spring and from 1950 to 2014 for the other seasons The annual

ACCEPTED MANUSCRIPT

drought frequency values have been smoothed using a 5-year moving window weighted average, and subsequently trends in drought frequency have been computed by using a linear model and testing its statistical significance with the Student‘s T-test (Gosset, 1908; Willks, 2011) and a confidence level of 95% Differently from previous papers (Spinoni et al., 2014, 2015a), in which the entire period was divided in 5-year sub-periods, here the trends are based on annual values derived from a 5-year moving window in order to increase the number of points from which the trends are derived The second drought variable investigated is the severity of drought events: for every season with a drought event, we summed up (in absolute values) the values belonging to the event To give an example, if the event lasts from February to June and peaks in May, all the indicator values during the event are summed up and the resulting severity score is assigned to spring Similarly to drought frequency, linear trends have been computed over annual drought severity values obtained with a 5-year moving window smoothing approach Given that seasonal discrete intervals are only three months long, we did not compute drought duration trends

Drought frequency trend values (DRF) are expressed in number of events per decade, see

Figures 3-4, 6-7, and 9-10, while drought severity trend values (DRS) are expressed in a

severity score per decade in Figures 5 and 8 Drought severity should not be mistaken for drought intensity, defined as the lowest SPI (or SPEI) value recorded during a drought event (Spinoni et al., 2014) Separate analyses were performed for the SPI-3 and the SPEI-3 and trends computed for 1981-2014 (or 2015, for spring) Trend values for a grid point were considered valid only if they were based on at least sixty years of valid data for the period 1950-2015 and at least thirty years for the period 1980-2015

Both indicators produce similar spatial patterns in winter, i.e a remarkable decrease of drought

frequency over Northern and Eastern Europe (especially Russia) and a moderate increase over Southern Europe, except Sicily (Italy) and Western Spain For extreme drought events (Figure 4) the described spatial patterns are very similar in winter, though the values are smaller, with the exception of Spain, where mixed patterns emerge for extreme events depending solely on precipitation (SPI-3) Also winter drought severity trends (Figure 5) are visually comparable with drought frequency ones, but in this case, Spain shows a general increase, while the Benelux countries and France - north to Massif Central - exhibit a decrease in drought severity driven by

a combination of precipitation and temperature (SPEI-3) that is hardly noticeable for drought and extreme drought

ACCEPTED MANUSCRIPT

Differently from winter, the two indicators result in distinct patterns in spring According to the

SPI-3 (Figure 3), drought frequency shows a decreasing trend over Northern and Eastern Europe and opposite an increasing trend over Southern and Western Europe However, according to the SPEI-3, Eastern Europe and Ukraine, the Baltic Republics (excluding Lithuania), and also Central Europe (mixed for the SPI-3), excluding the Danube Delta, show an increase that is statistically significant Both the indicators indicate decreasing drought frequency over Bosnia-Herzegovina and Central Italy, while the SPEI-3 suggests an increase in Catalunya (Spain) and Serbia, opposite to what the SPI-3 suggests The frequency trends of extreme droughts (Figure 4) confirm that in spring temperature plays an important role especially over Central Europe In spring, drought severity trends are spatially more coherent than for drought frequency and Figure

4 evidences that over Central Europe the increase is statistically significant only if temperature is included in the analyses It is worth highlighting that both the indicators agree upon the decrease

of drought severity over Southern Italy

In summer, the effect of temperature increase is clearly evident for droughts in Europe (Figure

3) Considering only precipitation as driver, in the last sixty-five years the frequency of drought events increased only over the Iberian Peninsula, Southern France, Northern Italy, Northern Germany, the Carpathian region, the Balkans (excluding Bosnia), and Greece Considering also temperature, such an increase occurred over almost all of Europe, with few exceptions: Iceland, Scandinavia, and along the borders between Poland, Belarus, and Ukraine For the extreme drought frequency (Figure 4), the temperature impact is not relevant also over Germany, whilst for drought severity it is relevant over Germany but not for Belgium (Figure 5) Compared to other seasons, the increase in affected area as depicted by drought severity is larger both in space and absolute values, with the highest scores over the Mediterranean region, Turkey, and Poland

ACCEPTED MANUSCRIPT

As regards autumn, the SPI-3 leads to decreasing drought frequency and severity in most of

Europe, excluding localized areas in Ireland, France, Albania, Greece, Belarus and Latvia for frequency and even more localized areas for severity Similar to the summer, the analyses based

on both precipitation and temperature indicate more areas affected by increasing drought frequency However, in autumn the decreasing drought frequency over the Balkans and Romania

is confirmed by the SPEI-3 and in general the drought severity patterns are spatially more heterogeneous, showing mixed tendencies for almost all regions, excluding Iceland which in 1950-2015 experienced decreasing drought trends for every season, variable, and indicator considered in this study

The drought frequency trends, but also the drought severity trends in spring and summer, show positive values in a particular region in Norway, as opposed to the surroundings (see Figures 3 and 5) This is mainly due to the low station density in the 1950s and early 1960s in the ECA&D database used to create version 12 (and earlier ones), and in particular to having a long precipitation series near the coast (which is very wet) and which is used to fill-in higher elevation areas in Norway which lack data in the 1950s and early 1960s (Klein Tank et al., 2002) Although there is an elevation-dependent gridding procedure in E-OBS (Haylock et al., 2008), the station density is apparently not good enough to prevent this situation As shown in the next section, this region does not show the same issue when trends from 1981 onward are considered Another area behaving differently than the surroundings is the bordering region between France and Spain and also this is likely due to poor coverage of French stations used to obtain the E-OBS v12 and earlier

3.2 Drought trends in 1981-2015

ACCEPTED MANUSCRIPT

In the last decades, Europe is known to have been subjected to climate change effects, in particular a temperature rise and more frequent extreme meterorological events such as heat waves and extreme rainfall (Giorgi, 2006; Giorgi and Lionello, 2008; IPCC 5th report: Pachauri

et al., 2014) Since according to our analyses and published research, the turning point can be found in the late 1970s or at the beginning of the 1980s (Klein Tank and Konnen, 2003; Moberg and Jones, 2005; van den Besselaar et al., 2013), we repeated the study on drought trends for the period 1981-2015

Compared to the entire period (1950-2015), the importance of incorporating temperature (through PET) in a meteorological drought indicator is more evident in the period 1981-2015

According to Figure 6, the differences between drought frequency trends derived by using the

SPI-3 and the SPEI-3 are small for winter, while for spring there is spatial coherence in sign, but the values are larger for the SPEI-3, especially regarding the positive trends over Iceland, England, France and Ukraine In summer and in autumn the differences are even larger in absolute values for areas affected by an increase of drought events, particularly the Mediterranean area and Eastern Europe in summer and the Baltic Republics and Russia in autumn However, both the indicators agree in every region and in every season on the sign (increase or decrease) of the trend, with the only major exception of Ukraine in autumn

As for the frequency trends of all drought events over the entire period, the extreme drought

frequency trends show similar spatial patterns if computed using the SPI-3 or the SPEI-3 (Figure

7) Compared to 1950-2015, there are some relevant differences: Iceland shows no significant trends in 1981-2015, Russia shows a significant decrease in winter in 1950-2015 but no corresponding trends in the latest decades, and with the exception of spring Scandinavia behaves

as Russia, especially according to the SPI-3 During spring, the Balkans and Greece show a more

ACCEPTED MANUSCRIPT

relevant decrease of extreme drought frequency; in summer, the trends are coherent in sign (the positive ones in the entire period stays positive in the last decades and the same holds for the negative ones), but the absolute values of positive trends can be two times bigger in the last decades, as over Northern Italy for both the indicators and over Eastern Europe for the SPEI-3; the Mediterranean is characterized by an opposite behavior in autumn, that is a general tendency towards wetter conditions in the last decades following the SPI-3 (compared to the entire period analyzed in this study) and drier conditions following the SPEI-3, however the trends are significant only over some areas (Albania, Eastern Greece, Spain, and Central Italy)

With respect to drought severity and similarly to the entire period, the split between decreasing

trends in winter in Northern Europe and increasing trends in Southern Europe is not clear during the last decades due to mixed spatial patterns; over Russia the decrease is smaller, and the most relevant difference is related to the increasing drought severity in big parts of Central and Eastern Europe following the SPEI-3 (see Figure 8); on the other hand, in spring the results for 1981-2015 are similar to those found for 1950-2015, though the tendency of decreasing drought severity over the Balkans is more evident as well as the increase over France and Germany; oppositely to what occurs in winter, in summer the split between a tendency towards wetter Northern Europe and drier Southern Europe is much more clear in the last decades, and there is not a general evolution towards more severe drought events as depicted by the SPEI-3 if we consider the whole period 1950-2015; finally, in autumn both the indicators indicate increasing trends over Eastern Europe and Russia in the last decades that are not evident from the analysis

of the entire period

It is worth highlighting that similar seasonal drought trends, based on SPI only, have also been reported by Gudmundsson and Seneviratne (2015) Even though there are some differences in

ACCEPTED MANUSCRIPT

their approach, similar spatial patterns over the entire European continent are reported in particular in winter and summer, while the most remarkable differences can be found over Central Spain and Eastern Europe

3.3 Comparisons between 1950-1980 and 1981-2015

In previous sub-sections we discussed drought trends over the last six and a half and three and a half decades, respectively To better understand the differences in drought frequency between recent years (1981-2015) and the antecedent period (1950-1980), we now discuss the difference between the number of drought events (Figure 9; and extreme drought events, Figure 10) in the two periods Positive values mean that more droughts took place in the more recent years Values are shown in numbers of events per decade

The most relevant changes for droughts driven by a precipitation deficit only can be seen over Russia and Scandinavia in winter and spring and over Eastern Europe in spring In the first case,

a remarkable decrease of droughts (around -1.5 event/decade in most areas) is reported in the last decades: this is due to the frequent and severe droughts that have been reported to hit Northern Europe in the 1950s and 1960s (Veryard, 1956; Sheffield et al., 2009) and also Russia especially

in the 1950s and mostly in spring (Meshcherskaya and Blaznevich, 1997) Prolongued shortages

of rainfall have caused frequent spring-summer droughts over the Carpathian region, as it has also been reported by Spinoni et al (2013) If we include also temperature in the analyses, it becomes evident that Southern Europe has been hit by more frequent drougths in the last decades and the Mediterranean region shows such increase in every season, with the exception of the former Yugolavian countries in summer This is another finding that confirms that the Mediterranean area should be considered a hotspot for climate change, extreme events, and

ACCEPTED MANUSCRIPT

natural disasters as frequently reported in literature (Giorgi, 2006; Diffenbaugh et al., 2007), especially in spring and summer (Della-Marta et al., 2007; Vautard et al., 2007; Hirschi et al., 2011) The temperature increase effect is also evident over Eastern Europe, for which the SPEI-3 shows that not only spring and summer as for the SPI-3, but also autumn and winter have been affected by more frequent droughts in the last decades, (see for example the borders between Hungary and Romania) This is mainly due to the observed temperature increase over Eastern Europe (Bartholy and Pongracz, 2007; Spinoni et al., 2015d), which is reflected in more frequent and severe heat waves in every season, which in turn contribute to trigger more meteorological drought events (Spinoni et al., 2015e)

Regarding extreme drought events, the use of temperature as input data is important in particular for large-scale events such as the recurrent summer droughts over the Baltic Republics and Russia (in particular in 2010, see Trenberth and Fasullo, 2012) However, those events were often caused by both a precipitation deficit and excessive temperatures, thus they are not missed

by using the SPI-3 as indicator For extreme droughts, autumn seems to be the season where the increase or decrease is smaller in absolute values, while the maximum increase is found for both the indicastors over Latvia in spring and summer and the maximum decrease over Russia in winter, Scandinavia in spring, France and Germany in summer, and Iceland and U.K in autumn

3.4 Summary of the results

In order to highlight the most relevant findings, the main results are summarized in Tables 1-3, which group Europe in eleven macro-regions Trends in seasonal drought frequency are shown for each season and for the two periods in Table 1 and similarly the same for trends in drought

Combining the seasonal trends presented in this study with the database of exceptional meteorological droughts that occurred from 1950 to 2012 as described in Spinoni et al (2015b), some conclusions can be drawn Firstly, it seems that the frequent winter and spring droughts which hit Northern Europe and Russia in the 1950s (Briffa et al., 1994; Golubev and Dronin, 2004; Meshcherskaya et al., 2007) play a key role in the resulting decreasing tendency of winter and spring droughts in the period 1950-2015 Oppositely, the remarkably increased frequency and severity of summer droughts, in particular over Southern Europe, can be ascribed to big the summer droughts that occurred over large parts of Europe in 2003 (Ciais et al., 2005), over the Iberian Peninsula in 2004-05 (Garcia-Herrera et al., 2007), and in Russia in 2010 (e.g., Wegren, 2011) This tendency is mainly driven by recent heat waves, which hit Europe in 2003, 2007,

2010, and also in 2015 (van Lanen et al., 2016), and this supports the need to include indicators

ACCEPTED MANUSCRIPT

directly or indirectly based also on temperature when dealing with meteorological droughts Instead, some relevant single drought event like those that hit the British Islands and Central Europe in 1976 (Perry, 176; Parry et al., 2012) and Central and Eastern Europe in autumn 2011 (van den Besselaar et al., 2014) do not seem to influence the drought trends; the first event because it lies in the middle of the period, the second one because the autumn droughts are in general homogeneously distributed over the entire period 1950-2015 (Spinoni et al., 2015b) The tables highlight the well-known division between Northern and Southern Europe (e.g., Spinoni et al., 2015a) which are characterized by opposite tendencies especially in winter and spring, however the importance of seasonal results is more evident if we consider the impacts of droughts For example, the increased number and severity of drought events in spring and summer leads to the largest impacts in regions where the cultivation and harvesting of crops takes place between March and September, this is why the summer drought of 2003 has been so devastating (Fink et al., 2004; Ciais et al., 2005) Instead, regions prone to viticulture are more threatened by summer and autumn droughts (Bardaji and Iraizoz, 2015), while the regions where lemons and oranges are grown are threatened by winter droughts (Perez-Perez et al., 2009) Such seasonality does not involve only agriculture, but also many other fields which can be impacted

by droughts occurring in different seasons, see the collection of reported drought impacts in the European Drought Impact report Inventory (EDII; http://www.geo.uio.no/edc/droughtdb/) Consequently, we believe that our findings can be relevant not only for climatologists, but also for policy makers, stakeholders, land managers, and national authorities

4 Conclusions

ACCEPTED MANUSCRIPT

Starting from high-resolution gridded precipitation, and minimum and maximum temperature data, we computed two drought indicators (SPI and SPEI) at the 3-month accumulation scale for the whole of Europe from 1950 to 2015 From such indicators we derived trends of frequency and severity of meteorological drought and extreme drought events and we investigated the spatial and temporal patterns at the seasonal scale This study confirms that Europe is climatologically divided between general drying tendencies over the Southern regions and the Carpathians – which are larger if temperature is taken into account as driver – and opposite wetting tendencies over Northern Europe The important novelty of this paper is represented by the analyses of seasonal patterns derived from two distinct drought indicators, which are of particular relevance for drought impacts, mostly related to agriculture, but also to other economic sectors such as forest growth, energy consumption, hydropower generation, inland water transportation, etc., which are affected differently by winter droughts than summer droughts (Leuzinger et al., 2005; Vicente-Serrano, 2007; Falloon and Betts, 2010; Olesen et al., 2011; Naumann et al., 2015)

Though the findings described in this paper are based on input data that have been processed and homogenized, on widely used indicators, and on a robust theoretical approach that includes statistical tests on the significance of the trends, the following limitations should be noted Firstly, due to the use of one single input dataset, we could not investigate the variability

post-of drought trends introduced by different data sources Secondly, the approximation post-of PET by the Hargreaves-Samani model relies on temperature input data to estimate solar radiation, resulting in a simplification if compared to the more sophisticated Penman-Monteith model, which would require more input data Thirdly, the use of E-OBS did not allow computing more complex indicators such as the Palmer Drought Severity Index (PDSI; Palmer, 1965) and the

ACCEPTED MANUSCRIPT

self-calibrated PDSI (sc-PDSI; Wells et al 2004), though finding datasets of dense, high-quality, and long-term records of meteorological variables other than precipitation and temperature for entire Europe is a challenging task

However, considering the seasonal analysis, the continental scale (Europe), the long-time interval (66 years), and the medium to high spatial resolution (0.25°x0.25°), the results shown represent a novelty for the analysis of meteorological droughts It should further be noted that our approach focuses on the characteristics of drought events and not on the drought indicators and, for the time being, no comparable studies dealing with seasonal trends have been published for Europe Starting from the methodologies discussed here and in two other papers published by the same authors (Spinoni et al., 2015a and 2015b), we now use a similar approach to project meteorological drought trends until 2100 – considering the RCP4.5 and RCP8.5 scenarios – using an ensemble of models and scenarios from the bias-adjusted EURO-CORDEX simulations (Jacob et al., 2014; Nikulin et al., 2015) The first tests made over the Mediterranean area and the Carpathian region are promising (Spinoni et al., 2016), consequently the final results are expected to provide a complete picture of meteorological drought tendencies for the whole of Europe until the 2100

Acknowledgments

We would like to thank the two anonymous referees whose comments, recommendations and suggestions helped to improve the original manuscript

ACCEPTED MANUSCRIPT

References

Allen, R.G., Pereira, L.S., Raes, D., Smith M (1998) Crop Evapotranspiration - Guidelines for Computing Crop Water Requirements FAO Irrigation and drainage paper, n 56, Rome, 15 pp Allen, R G., Pruitt, W O., Wright, J L., Howell, T A., Ventura, F., Snyder, R., Itenfisu, D., Steduto, P., Berengena, J., Baselga-Yrisarry, J., Smith, M., Pereira, L S., Raes, D., Perrier, A., Alves, I., Walter, I., Elliott, R (2006) A recommendation on standardized surface resistance for hourly calculation of reference ET0 by the FAO56 Penman-Monteith method Agricultural Water Management, 81(1), 1-22

Asner, G P., & Alencar, A (2010) Drought impacts on the Amazon forest: the remote sensing perspective New Phytologist, 187(3), 569-578

Bardaji, I., & Iraizoz, B (2015) Uneven responses to climate and market influencing the geography of high-quality wine production in Europe Regional Environmental Change, 15(1), 79-92

Bartholy, J., & Pongrácz, R (2007) Regional analysis of extreme temperature and precipitation indices for the Carpathian Basin from 1946 to 2001 Global and Planetary Change, 57(1), 83-95 Begueria, S., Vicente-Serrano, S M., Reig, F., & Latorre, B (2014) Standardized precipitation evapotranspiration index (SPEI) revisited: parameter fitting, evapotranspiration models, tools, datasets and drought monitoring International Journal of Climatology, 34(10), 3001-3023 Bond, N R., Lake, P S., & Arthington, A H (2008) The impacts of drought on freshwater ecosystems: an Australian perspective Hydrobiologia, 600(1), 3-16

Bordi, I., Fraedrich, K., & Sutera, A (2009) Observed drought and wetness trends in Europe: an update Hydrology and Earth System Sciences, 13(8), 1519-1530

ACCEPTED MANUSCRIPT

Briffa, K.R., Jones, P.D., Hulme, M (1994) Summer moisture variability across Europe, 1892–1991: an analysis based on the Palmer drought severity index Int J Climatol 14 (5), 475–506 Briffa, K R., Van Der Schrier, G., & Jones, P D (2009) Wet and dry summers in Europe since 1750: evidence of increasing drought International Journal of Climatology, 29(13), 1894

Buckland, S M., Grime, J P., Hodgson, J G., & Thompson, K (1997) A comparison of plant responses to the extreme drought of 1995 in northern England Journal of Ecology, 875-882 Campra, P., & Millstein, D (2013) Mesoscale climatic simulation of surface air temperature cooling by highly reflective greenhouses in SE Spain Environmental science & technology, 47(21), 12284-12290

Caussinus, H., & Mestre, O (2004) Detection and correction of artificial shifts in climate series Journal of the Royal Statistical Society: Series C (Applied Statistics), 53(3), 405-425

Ciais, P., Reichstein, M., Viovy, N., Granier, A., Ogée, J., Allard, V., & Chevallier, F (2005) Europe-wide reduction in primary productivity caused by the heat and drought in 2003 Nature, 437(7058), 529-533

Cook, E R., Woodhouse, C A., Eakin, C M., Meko, D M., & Stahle, D W (2004) Long-term aridity changes in the western United States Science, 306(5698), 1015-1018

Dai, A (2011) Drought under global warming: a review Wiley Interdisciplinary Reviews: Climate Change, 2(1), 45-65

Dai, A (2013) Increasing drought under global warming in observations and models Nature Climate Change, 3(1), 52-58

Ding, Y., Hayes, M J., & Widhalm, M (2011) Measuring economic impacts of drought: a review and discussion Disaster Prevention and Management: An International Journal, 20(4), 434-446