Climatic ChangeDOI 10.1007/s10584-012-0559-5

Differential regional responses in drought length,intensity and timing to recent climate changesin a Mediterranean forested ecosystem

Julien Ruffault · Nicolas K. Martin-StPaul ·Serge Rambal · Florent Mouillot

Received: 20 January 2012 / Accepted: 28 July 2012© Springer Science+Business Media B.V. 2012

Abstract The Mediterranean area is one of the regions of the world where GCMsagree the most on precipitation changes due to climate change. In this study we aimto assess the impact of recent climate change on drought features of Mediterraneanecosystems in Southern France. Regional climatic trends for the 1971–2006 periodare compared to drought trends based on a water balance model accounting for soilproperties, vegetation structure and functioning. Drought, defined here as periodswhen soil water potentials drop below −0.5 MPa, is described in terms of intensity,duration and timing, which are integrative of both climate variability and siteconditions. Temporal trends in precipitation, temperature and solar radiation leadaltogether to drier and warmer conditions over the region but with a high spatialheterogeneity; for similar climatic trends, a significant increase in drought intensitywas detected in the wettest areas of the region, whereas drought intensity in thedriest areas did not change. Indeed, in the wettest areas, we observed an earlieronset of drought by about 1 month, but a constant end of drought. In the driestareas of the region, we observed the same earlier onset of drought but combinedwith an earlier end of drought, thus leading to a shift of the dry season withoutincreasing its duration. The definition of drought features both in terms of intensitybut also of seasonal timing appears relevant to capture historical or forecasted

Electronic supplementary material The online version of this article(doi:10.1007/s10584-012-0559-5) contains supplementary material, which is available toauthorized users.

J. Ruffault (B) · N. K. Martin-StPaul · S. RambalDREAM, CEFE-CNRS, 1919 Route de Mende,34293 Montpellier Cedex 5, Francee-mail: julien.ruffault@cefe.cnrs.fr

F. MouillotUMR CEFE, eq DREAM, IRD, 1919 Route de Mende,34293 Montpellier Cedex 5, Francee-mail: florent.mouillot@ird.fr

Climatic Change

changes in ecosystem functioning. Studies concerning climate change impacts onforested ecosystems should be interpreted with caution when using climate proxiesalone.

1 Introduction

There is global evidence that an increase in mean global temperature, along withchanges in the hydrological cycle, are locally leading to significant drying in thealready water-limited Mediterranean regions (Solomon et al. 2007). Furthermore,climate modeling studies for these regions converge towards a substantial dryingand warming in the forthcoming decades, especially during summer season (Somotet al. 2008). In Southern France, temperature increase during the last decades(0.49 ◦C/decade for the period 1979–2005; Lespinas et al. 2009) has been particularlyimportant compared to the global average on land (0.27 ◦C/decade for the period1979–2005; Brohan et al. 2006), and has been accompanied by a local decrease insummer precipitation and an increase in autumn precipitation (Lespinas et al. 2009;Chaouche et al. 2010). Understanding and predicting the consequences of theseclimatic changes on terrestrial ecosystems in these regions is emerging as one of themajor challenges for global change scientists.

Water deficit is increasingly debated as a trigger for decline of Mediterraneanforested ecosystems (Carnicer et al. 2011). Indeed, the Mediterranean climate ischaracterized by cool and wet winters, warm and dry summers and intense rainfallevents in spring and fall (Allen 2001). Under such a climate, soil water balance shapesthe local duration, intensity and annual timing of the water stress period (Rambalet al. 2003). These drought features control the functioning of Mediterranean forestecosystems and they strongly contribute to modify carbon sequestration (Allard et al.2008), species distribution (Peñuelas and Boada 2004), tree mortality (Martinez-Vilalta and Pinol 2002) and fire occurrence and spread (Dimitrakopoulos et al. 2011).In such a context, understanding and quantifying spatial heterogeneity of droughtand its temporal trends under on-going climate change is necessary to assess futureforest status.

However, drought quantification depends on drought definitions, which varyamong disciplines (meteorological, agricultural or hydrological) according to thetype of water deficit under consideration (Dai 2010). Drought quantification hasbeen widely based on rainfall indices such as the total rainfall amount, the number ofconsecutive days without significant rainfall or the standardized precipitation index(SPI; Hughes and Saunders 2002). More detailed indices account for the limitation insoil water holding capacity available for plant by combining climate indices with soilproperties such as the commonly used Palmer Drought Severity Index (PDSI; Daiet al. 2004) or the seasonal soil moisture (Sheffield and Wood 2008). These formerstudies considerably improved our ability to quantify drought by taking into accountthe complex links between climate dynamics and soil moisture. Nevertheless, inforested ecosystems, soil water content depends not only on abiotic factors (climate,soil), but also on the control of vegetation over water fluxes, mainly through canopydensity (i.e. leaf area index; LAI) and stomatal conductance (Hoff and Rambal2003). In addition, drought indices are generally computed for long time periods (i.e.month to year), making it difficult to assess putative changes in the precise timing

Climatic Change

of drought, which have proven to be crucial for forest functioning (e.g. Allard et al.2008; Misson et al. 2010). Considering the timing, intensity and duration of droughtmight be particularly important for defining drought indices in Mediterranean-typeregions. The study of such systems requires daily computation of soil water balance(Granier et al. 1999) to characterize drought as the period when soil water contentfalls below a threshold affecting plant functioning.

Yet, to our knowledge, few studies, if any, quantify recent climate changes interms of these drought characteristics for natural ecosystems and its variability at aregional scale by taking explicitly into account feedbacks between plant functioning,soil properties and climate. In this study, we investigated drought trends in forestedecosystems in a Mediterranean region in Southern France on the period 1971–2006 by using a daily water balance process-based model. A set of drought indiceswas specifically designed to capture yearly intensity, duration and timing of majordrought periods. The objectives were: (i) to assess the regional heterogeneity indrought conditions and its link with the regional climate gradient, (ii) to test whetherdrought characteristics have significantly varied in time within the last decades, and(iii) to discuss how these trends can be related to any drought proxies in the forcingdata sets and how they are locally affected by environmental parameters (LAI, soilcharacteristics).

2 Datasets and methods

2.1 Study area

The study area (27,800 km2) covers the forested area of administrative Languedoc-Roussillon (LR) region, located in Southern France (Fig. 1). LR is delimited bythe Pyrenees Mountains in the South, the Massif Central foothills in the North, theMediterranean coastline and the Rhone River in the east. Climate is Mediterraneanwith hot and dry summers, cool and wet winters. Forests are dominated by theevergreen broadleaf Quercus ilex (2,517 km2), the deciduous Quercus pubescens(799 km2), the evergreen needleleaf Pinus halepensis (140 km2) and the evergreensclerophyll mixed shrublands called “garrigues” (1,840 km2) (IFN forest inventory2006). We confined our study on Mediterranean evergreen vegetation type (MET),based on the presence of Quercus ilex forests, shrublands and Pinus halepensis. Theresulting study area covers 10,990 km2, including 4,357 km2 of vegetation (Fig. 1).Within this area, field measurements are regularly performed across the climategradient (Martin-StPaul et al. 2011) and a permanent forest research station has beenrunning since 1998 (Puechabon; see details in Rambal et al. 2003) thus providing thelocal datasets used in this study (see location and characteristics of study sites inFig. 1).

2.2 Water balance estimations and input data

Water stress indices for the period 1971–2006 were derived from daily water balancesimulations, spatially explicit at 1 km2 resolution over each grid cell. Variations insoil water content (SWC) are simulated using the water module from the SIERRA

Climatic Change

MontpellierPuechabon

Les Mages

Vic la Gardiole

4°E

4°E

3°E

3°E

2°E

2°E

44°N 44°N

43°N 43°N

Paris

0 5025 Km

Mediterranean Sea

Pyrenees Mountains

Massif Central Mountains

Roussillonfloodplain

Corbières Hills

Cevennes Mounts

GuarriguesShrubland

Languedoc floodplain

Elevation(m)

2900

-47

Fig. 1 Digital elevation model of Languedoc Roussillon (LR) Mediterranean region with majormountains (in blue) and minor topographic areas (in red). The locations of experimental sites inwhich field measurements were made for model validation are indicated (black dots). Validationsites are characterized by the following mean annual temperature and precipitation: Vic (14.5 ◦C;650 mm); Puechabon (13.4 ◦C; 907 mm); Les mages (12 ◦C; 1,150 mm). Small panel at the bottom-right indicates the geographical extent of Mediterranean type ecosystems considered in this study

process-based vegetation model (Mouillot et al. 2001) based on the water balancebetween precipitations (P) and water outputs:

�SWC = P − In − D − E − T (1)

Where the amount of precipitation intercepted by the canopy (In), the soil evapora-tion (E), the transpiration of vegetation (T) and the drainage (D), are all expressedin mm. Soil is represented by a 3-layer bucket model (depths 0–20 cm, 20 cm–1 m, 1 m–4.5 m) whose maximal depth is set at the average maximum rootingdepth recorded for Quercus ilex (Rambal et al. 2003). Potential evapotranspiration(PET) is computed using the Priestley–Taylor (PT) equation, suitable for large-scale simulations on Mediterranean biomes (Fisher et al. 2005). T depends onPET but is modulated by LAI (Granier et al. 1999) and by water stress (Mouillotet al. 2001). Water stress is simulated using soil water potential (ψsoil), related tosoil water content by the power function model for the retention curve (Campbell1974). For each grid-point, the model inputs consist of species functional parameters,soil features and daily climatic variables. A more detailed description of the waterbalance model and its input data is provided in Electronic Supplemental Material.

Soil features (Soil depth d; rock fragment content Rf c and texture) used inthe Campbell (1974) retention curve were derived for natural ecosystems fromthe regional DONESOL soil database (Gaultier et al. 1993). Daily precipitation,temperature and global radiation over the LR region during the 1971–2006 period

Climatic Change

were extracted from the SAFRAN spatially explicit database (8 × 8 km grid) (sourceCNRM, France; Habets et al. 2008). These daily climatic variables were then re-interpolated at a 1 km2 spatial resolution to match with the resolution of thevegetation model. To limit the sources of variations in our study, simulations wereperformed with a single “evergreen” vegetation type whose functional parameterswere based on Quercus ilex (Mouillot et al. 2001). Quercus ilex is a widespreadand thoroughly studied species in the region (Rambal et al. 2003; Martin-StPaulet al. 2011) thus allowing for a full validation of water balance simulations (seeSection 2.3). All functional parameters were kept constant across space and timeexcept the leaf area index (LAI) that varied spatially across the region. LAI wasestimated by using carbon and allocation subroutines of SIERRA (Mouillot et al.2001), based on the ecohydrological equilibrium hypothesis (sensu Eagleson 1982).This hypothesis states that in water limited environments, ecosystems develop astable canopy density, which both maximizes biomass and minimizes drought stress.This LAI map was kept constant throughout the simulations to prevent any yearlyadjustments that could hide the trends in drought indices we wanted to explore.

2.3 Model validation

Model validation was achieved following two steps. First, soil water content sim-ulations over the whole soil profile (0–4.5 m) were compared with neutron probemeasurements achieved on the experimental Quercus ilex site of Puechabon forthe 2001–2005 period (see details in Rambal et al. 2003). Secondly, simulated soilwater potentials (�soil) were compared with water potentials (�lpd) measured onthree forest stands covering the regional water availability gradient (Fig. 1). Theperformance of the model was evaluated by comparing predicted and observedvalues using Pearson correlations.

2.4 Stress indices

Under Mediterranean climate, vegetation has to face usually a major drought periodper year. Four indices were computed to account for different features of this majordrought period for Mediterranean forested ecosystems, namely its intensity, durationand timing. We stated that water deficit starts when soil water potential (�soil)falls below a critical value (�crit) of −0.5 MPa, matching the value beyond whicha decrease of leaf conductance is observed for Mediterranean evergreen species (e.g.Limousin et al. 2009). Based on this threshold, we defined the Drought duration(DD) as the number of consecutive days when �soil was below �crit (Myers 1988;Misson et al. 2010). The first and last day of this period, respectively called onsetdrought day (ODD) and end drought day (EDD) were also computed. When, insome rare cases, the potential did not fall under this specific threshold, ODD andEDD were considered as missing values and DD was fixed to 0 day. During thatdrought period, leaf stomatal closure affects water fluxes in such a way that actualtranspiration is less than potential transpiration. Thus, yearly drought intensity (DI)was calculated according as the sum of daily ratios between actual transpiration andpotential transpiration.

Climatic Change

2.5 Statistical analyses

We tested for temporal trends in climate variables and drought indices within eachgrid-point. Yearly drought indices were considered independent in time as soil isfully refilled every winter in the LR region (Hoff and Rambal 2003). Climatic timeseries were aggregated at the annual and seasonal (spring, summer, autumn andwinter) time step to capture annual trends and their intra-annual variability. Trendswere calculated using the non-parametric Mann–Kendall trend test (Kendall 1975),which is robust, distribution-independent and has been identified as an efficienttool for identifying trends in climatic and drought variables (e.g. Sheffield andWood 2008). We considered trends to be significant when p-values were lowerthan 0.1. When a significant trend was detected, we determined its magnitudeusing the non-parametric Theil-Sen approach (TSA; Sen 1968). TSA chooses themedian slope among all lines through pairs of two-dimensional sample points andcan be significantly more accurate than simple linear regression for skewed andheteroscedastic data (Sen 1968).

3 Results

3.1 Model evaluation on specific sites

Modeled and measured seasonal variations in relative water content (RWC) at thePuechabon site are presented in Fig. 2a along the 5-year period (Pearson correlation:r2 = 0.87). The comparison between observed and measured soil water potentials(�soil) shows that the model predicts accurately this variable across the three foreststands (Pearson correlation: r2 = 0.76; Fig. 2b). A single major discrepancy observedat the “Les Mages” stand might result from a very local precipitation event notcaptured by our climatic data.

a b

Fig. 2 a Comparison of daily relative water content (RWC) calculated with neutron probe in aQuercus ilex forest in Puechabon Stand with RWC predicted by the water balance model for the2001–2005 period (Pearson correlation: r2 = 0.87). b Comparison of measured soil water potential(�lpd) and soil water potential (�soil) predicted by the model along a precipitation gradient in3 Mediterranean Quercus ilex forests stands (Pearson correlation: r2 = 0.76). The location andcharacteristics of validation sites are indicated in Fig. 1.

Climatic Change

3.2 Regional heterogeneity and temporal trends in climate forcings

Annual mean temperatures varied between +6.0 ◦C at the highest altitudes(Cevennes mounts and Pyrenean mountains) to +15.9 ◦C in the Roussillon flood-plain, following the elevation gradient (Fig. 3a). There has been an overall significantincrease (Mann–Kendall (MK) test: P < 0.01) in mean annual temperature over theentire study area during the 1971–2006 period. However this increasing trend showedimportant spatial variability, ranging from +1.0 ◦C in the plains to more than +2.5 ◦Cin the northeastern part of the region (Fig. 4c). Moreover, this annual increase inmean temperature was unevenly distributed over the seasons, such that most of thewarming resulted from an increase in spring and summer mean temperatures (Fig. 4d;only shown for summer).

Mean annual precipitation ranged from 590 to 1,700 mm, and followed a gradientfrom the Mediterranean coast to the Eastern part of the Cevennes Mounts (Fig. 3b).Mean annual precipitation was highly variable from year to year but did notexhibit significant trends (MK test: P < 0.1). At the seasonal time-scale, significantdecreasing trends in summer precipitation and increasing trends in fall precipitation(Fig. 4a and b) were observed. Summer precipitation significantly decreased in theRoussillon floodplain (P < 0.01), the Southern part of Corbières Hills (P < 0.05)and the North-East of Cevennes mounts (P < 0.05) by −70 and −35 % (−80 to−25 mm) during the 36 years. Fall precipitations increased mostly in the Cevennesmounts (P < 0.05) and the Guarrigues shrubland (Fig. 4b) with a magnitude ofchange ranging from +30 to +70 % (+35 to +80 mm). Mean annual global radiationis quite uniform over the region with values ranging from 147 to 176 W.m−2 (Fig. 3c).Increasing trends in spring and decreasing trends in fall were observed (Fig. 4e andf). Spring global radiation increased by +7 to +21 % in the Cevennes mounts andNorth-East part of the Guarrigues shrubland (P < 0.05). By contrast, fall globalradiation decreased in the Cevennes Mounts (P < 0.05) by −6 to −19 %.

3.3 Spatial heterogeneity and temporal trends in drought indices

3.3.1 Spatial heterogeneity

The forested ecosystems in the region are characterized by a gradient of meandrought intensities (DI), with higher values along the coast and lower values in

a) b) c)

Mean annualprecipitation (mm)

1700

590

Mean annual temperature (°C)

16

6

Mean global radiation (W/m²)

176

147

Fig. 3 Annual average climatic conditions for the period 1971–2006

Climatic Change

0/0.50.5/11/1.51.5/22/2.52.5/3

<-70-70/-60-60/-50-50/-40-40/-30>-30

<1010/1212/1414/1616/18>18

<5555/6060/6565/7070/75>80

0.5/11/1.51.5/22/2.52.5/33/3.5

<-17-17/-15-15/-13-13/-11-11/-9>-3

a) Non-parametric trend in summer precipitation (%)

c) Non-parametric trend in mean annual temperature (°C)

e) Non-parametric trend in spring global radiation (%)

b) Non-parametric trend in fall precipitation (%)

d) Non-parametric trend in mean summer temperature (°C)

f) Non-parametric trend in fall global radiation (%)

Fig. 4 Estimated trends in climatic variables for the 1971–2006 period. Only trends with p-valueslower than 0.1 are shown. For each climatic variable, only major annual and seasonal trends areshown

the North (Fig. 5a) although a high variability can be locally observed. Annualmean daily DI values ranged from 0.01 to 0.25. The main drought period lasted onaverage between 2 and 169 days (Fig. 5b) and was highly correlated to DI (Pearsoncorrelation: R = 0.95). Drought periods started (ODD) between the 145th (May25th) and 239th (August 28th) day of the year, and on average on the 201th (July20th) across the region. Drought periods ended (EDD) between the 230th (August19th) and the 310th (November 7th) day of the year, and on average on the 268th(Sept 26th) across the region.

Spatially, the driest areas in the coastal floodplains were characterized by anaverage drought duration (DD) of 100 days, starting on average at day 190 (July9th) and ending at day 285 (October 11th). By contrast, the wettest areas located inthe North and in the South West were characterized by shorter DD (approximately25 days) and drought period started on average at day 210 (July 29th) and ended atday 235 (August 23rd).

3.3.2 Temporal trends

Significant temporal trends were observed for all drought indices, but in differentareas depending on the index. Trends in drought intensity (DI) indicated increasinglydry conditions all over the region but significant trends were found in only two areas(Fig. 6a) with DI increasing by +20 to +120 %: the largest area located in the North

Climatic Change

a) b)

c) d)

0 - 20

20-40

40-60

60-80

80-100

100-120

>120

<0.04

0,04 - 0,06

0,06 - 0,08

0,08 - 0,1

0,1 - 0,12

0,12 - 0,14

>0.14

<240

240-250

250-260

260-270

270-280

280-290

>290

<190

190-195

195-200

200-205

205-210

210-215

>215

Mean DI Mean DD (days)

Mean ODD Mean EDD

Fig. 5 Average stress indices of Mediterranean forested ecosystems for the 1971–2006 period.Drought intensity (DI), drought duration (DD), onset drought day (ODD) and end drought day(EDD) are represented

(MK test: P < 0.05) and the smallest in the South, between Roussillon floodplain andPyrenees Mountains (P < 0.1). Drought duration (DD) also tended to increase overthe entire region during the study period. However, trends in DD were significantonly over a smaller area (only 11 % of the study area; Fig. 6b), where significanttrends in DI were also found.

Temporal trends in both the onset (ODD) and the ending (EDD) of the droughtperiod reveal an overall shift towards an earlier occurrence and earlier ending ofdrought from 1971 to 2006 (Fig. 6c and d). Decreasing trends in ODD were significantover 66 % of the region, with drought starting 5–80 days earlier in 2006 than in1971. In the driest areas like the Roussillon floodplain and Guarrigue shrublands, thechanges were about −35 days. By contrast, decreasing trends in EDD were significantonly over a smaller area. Drought ended significantly earlier in the eastern part of theCevennes Mounts by 10–90 days (P < 0.05). In this area where an earlier ending of

Climatic Change

<30

30/40

40/50

50/60

60/70

70/80

>80

<60

60/70

70/80

80/90

90/100

100/110

>110

<-60

-60/-50

-50/-40

-40/-30

-30/-20

-20-10

>-10

<-70

-70/-60

-60/-50

-50/-40

-40-30

-30/-20

>-20

a) Non-parametric trend in DroughtIntensity (%)

b) Non-parametric trend in DroughtDuration (%)

c) Non-parametric trend in Onset DroughtDay (days)

d) Non-parametric trend in End DroughtDay (days)

Fig. 6 Estimated trends in drought indices of Mediterranean type ecosystems for the 1971–2006period. All trends are shown for the 36 years period. Trends in drought intensity (DI) and droughtduration (DD) are in %. Trends in onset drought day (ODD) and end drought day (EDD) are inJulian days. Only trend with p-values lower than 0.1 are shown

drought was associated with an earlier onset, no significant trends in DI or DD wereobserved.

3.4 Discussion

3.4.1 Linking regional climate and vegetation water def icit

Over the LR region, we observed a difference of 10 ◦C on the 500 m altitudinal gradi-ent, and a 1,100 mm difference of annual rainfall between Cevennes mounts and the

Climatic Change

nearby coastline (Fig. 3a and b). This pattern is common around the Mediterraneanbasin where topography plays a major role in shaping regional climatic features(Portalès et al. 2010). Drought duration (DD) followed the same regional gradientwith values ranging from 2 to 169 days (mean = 58 days). These estimations are lowerthan values simulated by Lavoir et al. (2011) in the same region (who accounted forthe total number of days with relative water content (RWC) < 0.7) but are consistentwith the few estimations of water stress intensity and duration locally reported forNorth Eastern Spain and Southern France (Martinez-Vilalta and Pinol 2002; Missonet al. 2010). At a finer scale, we observed discrepancies between rainfall amountand our drought indices. For instance, we detected lower drought intensities anddurations (<70 days) near riverbeds or accumulating sites where soils are deeper,confirming that soil properties can mitigate climate impacts under Mediterraneanclimate (Costa et al. 2010).

3.4.2 Recent climatic change: regional and seasonal pattern

The climate trends observed over the last three decades have lead to hotter anddrier conditions. The trends observed in mean annual temperature are in accordancewith the +0.5 ◦C per decade previously observed in this region (Lespinas et al.2009; Chaouche et al. 2010) but are chiefly due to spring and summer warming.Our spatially explicit approach highlights the spatial heterogeneity in temperatureincrease, which was closely related to altitude (Fig. 5; +5 % on the coast and +25 %in the Cevennes foothills and Pyrenees) thus confirming the altitudinal lapse rateof 20 % for 1,500 m hypothesized for global warming (Still et al. 1999). Regardingprecipitation, we face here a peculiar pattern: no change occurred in the annualrainfall amount, but we observed a significant decrease in summer precipitationcoupled with an increase in autumn precipitation, a pattern already observed withinthe region (Lespinas et al. 2009; Chaouche et al. 2010). As a result of increasing cloudcover during rainfall events, global radiation showed opposite seasonal trends withan increase in spring and a decrease in autumn.

3.4.3 Climate change impacts: longer and earlier droughts

We quantified the heterogeneous spatial patterns of increasing drought intensity(DI) and duration (DD) in Southern France (Fig. 6a and b). Significant increases inDI and DD were observed in only two specific areas: the North-Eastern part of theCevennes Mounts and the boundary between Pyreneans Mountains and Roussillonfloodplain. This spatial aggregation of drought trends implies that climate is the keydriver of the increase in drought intensity and duration. However, the high spatialheterogeneity in the magnitude of change observed within these areas suggests alsoan important role of soil characteristics and vegetation functioning in mitigating oramplifying trends in DI and DD. For instance, in these areas, the increase in DIvaried spatially and could range from 20 to 120 %, while the variations in summerprecipitation varied only by 30–70 %.

These results confirm earlier observations based on rainfall amounts (Altava-Ortiz et al. 2011) and drought severity indices (Sousa et al. 2011) indicating thatdrought episodes have become more frequent and persistent in the second half of the20th century over the Mediterranean area. Here, the most drought-increasing siteswere located at the wettest edge of the study area, corresponding to the Northern

Climatic Change

bioclimatic limits of the Mediterranean type ecosystems with low to middle elevationranges, while the already dry areas remained less affected. Such drought increase inthe most temperate areas may lead to unusual plant mortality (Sarris et al. 2011) andshifts in plant distribution (Peñuelas and Boada 2004).

Different kinds of changes in the timing of drought were revealed. In some areas,drought has been extended over the past decades mostly as a result of an earlierODD (e.g. Cevennes mounts; Fig. 6c). By contrast, in a significant fraction of theregion, a shift in both ODD and EDD resulted in a drought season of similar durationand intensity, but occurring 30 days earlier (e.g. Roussillon floodplain). Such driersprings have also been reported in North-Eastern Spain due to changes in the distri-bution of spring precipitation (Ramos 2001) and could have significant consequenceson ecosystem functioning. Numerous impact studies have reported the consequencesof earlier drought on evergreen Mediterranean forests through a significant decreaseof annual carbon balance (Allard et al. 2008), a shift in plant phenology (Misson et al.2010; Bernal et al. 2011) and subsequent plant growth and survival (Lindner et al.2010). Such observed changes actually match climate forecasts of a lengthening of thedry season by 1 week in Southern France, shifting towards spring (Giannakopouloset al. 2009). Thus, this earlier drought could strongly mitigate the positive impacts ofan earlier onset of the growing season due to increasing temperature (evaluated at24 days in the last 40 years; Gordo and Sanz 2010)

3.5 Uncertainties

We mainly investigated how changes in climate variables could modify droughtfeatures within the natural vegetation envelop in LR region. To limit the sourcesof variability in the simulated drought indices for this analysis, we modeled onlya single deep-rooted vegetation. However, the differential response of shallow vs.deep-rooted species to water deficit (Viola et al. 2008) should also be considered forclimate change impact assessments.

Previous studies have used remote sensing spatialized datasets to provide regionalassessments of LAI or FPAR (Maselli et al. 2004), thus capturing a regional het-erogeneity which accounts for both LAI adjustments and human impact on treedensity and landscape fragmentation. In this study, a self-fitted constant LAI wasestimated over the region, based on ecohydrological hypotheses (Eagleson 1982). Asensitivity analysis was performed by varying LAI values by ±20 % which generateda maximum 15 % uncertainty in the trends of final drought indices (not shown).Moreover, increasing trends in drought severity could be balanced by a year-to-yearadjustment in LAI. Such adjustments are difficult to integrate because interannualvariations in leaf production relies on fine scale allocation processes which are farfrom being understood and modeled (Litton et al. 2007). The decreasing trends inleaf density observed in the last decades in Southern Europe (Carnicer et al. 2011)could indicate this adjustment to increasing drought and points out the need formodeling LAI variations under climate projections as a mitigation process.

Soil characteristics were identified as the critical variables, driving field capacityand the relationship between soil water content and plant water extraction (Ostleet al. 2009). Usually, soil maps are represented by large polygons for soil types,and locally empirically modified in terms of soil depth or rock fragment content bycompound terrain indices (Gessler et al. 1995). We used here the most advanced soil

Climatic Change

maps over the region to decrease this uncertainty and our spatial analysis was ableto assess the model responses to different soil characteristics within homogeneousclimate areas.

Finally, we used a spatially interpolated daily climate database at 1 km resolution.Downscaling errors may induce regional biases due to local effects on climaticvariables, resulting in biases on the drought assessment, which, coupled with biasesin the modeling of water fluxes, may result in errors in reported trends. However, asthese biases are systematic and therefore uniform over time, it is likely they will notimpact the sign or strength of the calculated trends (Sheffield and Wood 2008).

4 Conclusion

Our study focused on understanding how climatic drivers may have interacted tomodify drought features in a Mediterranean type ecosystem over the past decades(1971–2006). The overall climatic trends observed in Southern France led to anincrease in drought intensity and duration mainly located in the North-Eastern partof LR region as well as a shift towards an earlier drought in a large part of this region.Our results highlight that soil properties and vegetation functioning must be takeninto account when assessing ecosystem vulnerability to climate change. Quantifyingspatial variation in the relative importance of different drivers of drought, suchas climate, soil parameters and vegetation structure, could improve predictions ofclimate change impacts on forest drought, especially in such areas as the NorthernMediterranean, where GCMs predict a decrease in summer precipitation along withan increase in temperature (Somot et al. 2008). In these areas, local changes in micro-climate, associated with particular soil properties, could serve as refugia for drought-sensitive species (Keppel et al. 2012), thus mitigating predictions on species survivaland conservation. Further efforts should also be developed to estimate droughtfrequency, in order to accurately assess the distribution and impact of multipleextreme drought events in semi-arid ecosystems (McAuliffe and Hamerlynck 2010).

Acknowledgements This work is a contribution to the UE 7th FP Env.1.3.1.1 FUME “Forest firesunder climate,social and economic changes in Europe, the Mediterranean and other fire-affectedareas of the world”, Grant agreement no. 243888 and to the French ANR VMC 2007 project“MESOEROS”. A doctoral research Grant was provided by the Languedoc-Roussillon (LR) regionand the Centre National de la Recherche Scientifique (CNRS) . The authors wish to acknowledgethe Institut National de la Recherche Agronomique (INRA) for providing the DONESOL soildatabase. We thank Cyril Bernard for help in GIS programming as well as Emmanuel Gritti andMaud Bernard-Verdier for help in improving this manuscript. We also thank three anonymousreviewers for their useful comments and suggestions.

References

Allard V, Ourcival JM, Rambal S, Joffre R, Rocheteau A (2008) Seasonal and annual variation ofcarbon exchange in an evergreen Mediterranean forest in southern France. Glob Chang Biol14(4):714–725

Allen HD (2001) Mediterranean ecogeography. Pearson EducationAltava-Ortiz V, Llasat MC, Ferrari E, Atencia A, Sirangelo B (2011) Monthly rainfall changes in

central and western mediterranean basins, at the end of the 20th and beginning of the 21stcenturies. Int J Climatol 31:1943–1958

Climatic Change

Bernal M, Estiarte M, Peñuelas J (2011) Drought advances spring growth phenology of theMediterranean shrub Erica multif lora. Plant Biol 13(2):252–257

Brohan P, Kennedy JJ, Harris I, Tett SFB, Jones PD (2006) Uncertainty estimates in regional andglobal observed temperature changes: a new dataset from 1850. J Geophys Res 111(D12):D12106

Campbell GS (1974) A simple method for determining unsaturated conductivity from moistureretention data. Soil Sci 117(6):311

Carnicer J, Coll M, Ninyerola M, Pons X, Sánchez G, Peñuelas J (2011) Widespread crown conditiondecline, food web disruption, and amplified tree mortality with increased climate change-typedrought. Proc Natl Acad Sci 108(4):1474

Chaouche K, Neppel L, Dieulin C, Pujol N, Ladouche B, Martin E, Salas D, Caballero Y (2010)Analyses of precipitation, temperature and evapotranspiration in a french mediterranean regionin the context of climate change. C R Geosci 342(3):234–243

Costa A, Pereira H, Madeira M (2010) Analysis of spatial patterns of oak decline in cork oakwoodlands in Mediterranean conditions. Ann For Sci 67(2):204

Dai A (2010) Drought under global warming: a review. Wiley Interdisciplinary Reviews: ClimateChange

Dai A, Trenberth KE, Qian T (2004) A global dataset of palmer drought severity index for 1870–2002: relationship with soil moisture and effects of surface warming. J Hydrometeorol 5(6):1117–1130

Dimitrakopoulos AP, Vlahou M, Anagnostopoulou CG, Mitsopoulos ID (2011) Impact of droughton wildland fires in Greece: implications of climatic change? Clim Change 109:331–347

Eagleson PS (1982) Ecological optimality in water-limited natural soil-vegetation systems 1. Theoryand hypothesis. Water Resour Res 18(2):325–340

Fisher JB, DeBiase TA, Qi Y, Xu M, Goldstein AH (2005) Evapotranspiration models compared ona Sierra Nevada forest ecosystem. Environ Model Softw 20(6):783–796

Gaultier JP, Legros JP, Bornand M, King D, Favrot JC, Hardy R (1993) L’organisation et la gestiondes donnees pedologiques spatialisees: le projet donesol. Rev Geomat 3:235–253

Gessler PE, Moore ID, McKenzie NJ, Ryan PJ (1995) Soil-landscape modeling and spatial predictionof soil attributes. Int J Geogr Inf Syst 9:421–432

Giannakopoulos C, Le Sager P, Bindi M, Moriondo M, Kostopoulou E, Goodess CM (2009) Climaticchanges and associated impacts in the Mediterranean resulting from a 2 ◦C global warming. GlobPlanet Change 68(3):209–224

Gordo O, Sanz JJ (2010) Impact of climate change on plant phenology in Mediterranean ecosystems.Glob Chang Biol 16(3):1082–1106

Granier A, Bréda N, Biron P, Villette S (1999) A lumped water balance model to evaluate durationand intensity of drought constraints in forest stands. Ecol Model 116:269–283

Habets F, Boone A, Champeaux JL, Etchevers P, Franchisteguy L, Leblois E, Ledoux E, Moigne PL,Martin E, Morel S (2008) The SAFRAN-ISBA-MODCOU hydrometeorological model appliedover France. J Geophys Res 113(D6):D06,113

Hoff C, Rambal S (2003) An examination of the interaction between climate, soil and leaf area indexin a Quercus ilex ecosystem. Ann For Sci 60(2):153–161

Hughes BL, Saunders MA (2002) A drought climatology for Europe. Int J Climatol 22(13):1571–1592Kendall MG (1975) Rank correlation measures. Charles Griffin, London 202Keppel G, Van Niel KP, Wardell-Johnson GW, Yates CJ, Byrne M, Mucina L, Schut AGT, Hopper

SD, Franklin SE (2012) Refugia: identifying and understanding safe havens for biodiversityunder climate change. Glob Ecol Biogeogr 21(4):393–404

Lavoir AV, Duffet C, Mouillot F, Rambal S, Ratte JP, Schnitzler JP, Staudt M (2011) Scaling-upleaf monoterpene emissions from a water limited Quercus ilex woodland. Atmos Environ 45:2888–2997

Lespinas F, Ludwig W, Heussner S (2009) Impact of recent climate change on the hydrology ofcoastal Mediterranean rivers in southern France. Clim Change 99(3–4):425–456

Limousin JM, Rambal S, Ourcival JM, Rocheteau A, Joffre R, Cortina RR (2009) Long termtranspiration change with rainfall decline in a Mediterranean Quercus ilex forest. Glob ChangBiol 15(9):2163–2175

Lindner M, Maroschek M, Netherer S, Kremer A, Barbati A, Garcia-Gonzalo J, Seidl R, Delzon S,Corona P, Kolström M (2010) Climate change impacts, adaptive capacity, and vulnerability ofEuropean forest ecosystems. For Ecol Manag 259(4):698–709

Litton CM, Raich JW, Ryan MG (2007) Carbon allocation in forest ecosystems. Glob Chang Biol13(10):2089–2109

Climatic Change

Martin-StPaul NK, Limousin JM, Rodriguez-Calcerrada J, Ruffault J, Rambal S, Letts MG, MissonL (2011) Photosynthetic sensitivity to drought varies among populations of Quercus ilex along arainfall gradient. Funct Plant Biol 39(1):25–37

Martinez-Vilalta J, Pinol J (2002) Drought-induced mortality and hydraulic architecture in pinepopulations of the NE Iberian Peninsula. For Ecol Manag 161(1–3):247–256

Maselli F, Chiesi M, Bindi M (2004) Multi-year simulation of Mediterranean forest transpiration bythe integration of NOAA-AVHRR and ancillary data. Int J Remote Sens 25(19):3929–3941

McAuliffe JR, Hamerlynck EP (2010) Perennial plant mortality in the Sonoran and Mojave desertsin response to severe, multi-year drought. J Arid Environ 74(8):885–896

Misson L, Rocheteau A, S Rambal S, J Ourcival J, J Limousin JM, Rodriguez R (2010) Functionalchanges in the control of carbon fluxes after 3 years of increased drought in a Mediterraneanevergreen forest? Glob Chang Biol 16(9):2461–2475

Mouillot F, Rambal S, Lavorel S (2001) A generic process-based SImulator for meditERRaneanlandscApes (SIERRA): design and validation exercises. For Ecol Manag 147(1):75–97

Myers BJ (1988) Water stress integral: a link between short-term stress and long-term growth. TreePhysiol 4(4):315

Ostle NJ, Smith P, Fisher R, Woodward FI, Fisher JB, Smith JU, Galbraith D, Levy P, Meir P,McNamara NP (2009) Integrating plant-soil interactions into global carbon cycle models. J Ecol97(5):851–863

Peñuelas J, Boada M (2004) A global change induced biome shift in the Montseny mountains (NESpain). Glob Chang Biol 9(2):131–140

Portalès C, Boronat N, Pascual JEP, Beser AB (2010) Seasonal precipitation interpolation at theValencia region with multivariate methods using geographic and topographic information. Int JClimatol 30(10):1547–1563

Rambal S, Ourcival JM, Joffre R, Mouillot F, Nouvellon Y, Reichstein M, Rocheteau A (2003)Drought controls over conductance and assimilation of a Mediterranean evergreen ecosystem:scaling from leaf to canopy. Glob Chang Biol 9(12):1813–1824

Ramos MC (2001) Rainfall distribution patterns and their change over time in a Mediterranean area.Theor Appl Climatol 69(3):163–170

Sarris D, Christodoulakis D, Körner C (2011) Impact of recent climatic change on growth of lowelevation eastern Mediterranean forest trees. Clim Change 106(2):203–223

Sen PK (1968) Estimates of the regression coefficient based on Kendall’s tau. J Am Stat Assoc63:1379–1389

Sheffield J, Wood EF (2008) Global trends and variability in soil moisture and drought characteris-tics, 1950–2000, from observation-driven simulations of the terrestrial hydrologic cycle. J Climate21(3):432–458

Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (2007)IPCC 2007: climate change 2007: the physical science basis. Contribution of working group Ito the fourth assessment report of the Intergovernmental Panel on Climate Change. CambridgeUniversity Press, New York

Somot S, Sevault F, Déqué M, Crépon M (2008) 21st century climate change scenario for theMediterranean using a coupled atmosphere–ocean regional climate model. Glob Planet Change63(2–3):112–126

Sousa PM, Trigo RM, Aizpurua P, Nieto R, Gimeno L, Garcia-Herrera R (2011) Trends andextremes of drought indices throughout the 20th century in the Mediterranean. Nat HazardsEarth Syst Sci 11:33–51

Still CJ, Foster PN, Schneider SH (1999) Simulating the effects of climate change on tropical montanecloud forests. Nature 398(6728):608–610

Viola F, Daly E, Vico G, Cannarozzo M, Porporato A (2008) Transient soil-moisture dynamics andclimate change in Mediterranean ecosystems. Water Resour Res 44(11):W11412