Environmental changes, climate and anthropogenic impact in southern-eastern Tunisia during the last 8 kyr

. Pollen and clay mineralogical analyses of a Holocene sequence from Sebkha Boujmel (southern Tunisia) traces the climatic and environmental dynamics in the lower arid bioclimatic zone over the last 8000 years. During the Mid-to Late Holocene transition, between ca. 8 and 3 ka, a succession of five wet/dry oscillations is recorded. An intense arid event 20 occurs between ca. 5.7 and 4.6 ka.

(wormwood) between ca. 1.1 and 0.8 ka (850 -1150 AD) is linked to intensive pastoral activity, associated with heightened interannual and/or seasonal climatic instability. A complete re-shaping of the landscape is recorded during the 20 th century.
The remarkable expansion of the olive tree, and the deterioration of regional ecosystems with the spread of desert species, is linked to recent local socio-economic changes in Tunisia.

Introduction 5
Subjected to climatic influences from both polar high latitudes and subtropical low latitudes, the Mediterranean Basin is considered to be one of the regions most exposed to climatic change (IPCC, 2014). Its ecosystems are particularly vulnerable to hydrological changes as well as to strong demographic pressure, particularly in coastal areas (Lionello et al., 2006;IPCC, 2014). In this context, the desert margins and arid ecosystems of North Africa constitute climatic and biogeographical transition zones that are among the most sensitive to variations in climatic parameters and human activities (Smykatz- Kloss 10 and Felix-Henningsen, 2004). Among these regions, south-eastern Tunisia is currently crippled by arid conditions and increasing human pressure, leading to a marked degradation of its natural environment and increased desertification (Floret and Pontanier, 1982;Genin et al., 2006). Moreover, prospective studies foresee a prolonging of drought periods and intensified desertification due to the effects of global warming (e.g. Gao and Giorgi, 2008;Giorgi and Lionello, 2008;Nasr et al., 2008;Giannakopoulos et al., 2009). 15 In the northern Mediterranean, numerous palaeoecological records reveal the climate dynamics for the Middle to Late Holocene (e.g. Roberts et al., 2011;Sadori et al., 2011;Magny et al., 2013;Azuara et al., 2015). For the Sahara, the scale of Holocene climate change, with the establishment of a hyper-arid desert, has been estimated using multi-proxy analyses (e.g. deMenocal et al., 2000;Swezey, 2001;Gasse and Roberts, 2004;Hoelzmann et al., 2004;Kröpelin et al., 2008;Lézine et al., 2011;McGee et al., 2013;Tierney and deMenocal, 2013;Cremaschi et al., 2014). However, palaeoecological and 20 palaeoclimatic records from the southern Mediterranean are still needed in order to complete the pattern of environmental and climatic changes along a latitudinal transect in the Mediterranean region. In fact, during the Early and Middle Holocene, the climatic and hydrological contrast between the northern and southern parts of the Mediterranean basin is already apparent (Magny et al., 2012Peyron et al., 2013). Regional-scale climate modelling simulations spanning the Holocene also reveal a climatic contrast between the east and west shores of the Mediterranean, along a line passing between 25 Italy and Tunisia (Brayshaw et al., 2011). Today, there is still a dearth of available data for the southern shore of the Mediterranean, and in particular for the arid regions of North Africa, thus providing only a fragmentary view of Holocene climate changes. Indeed, the scarcity of suitable deposits for pollen analyses in the arid regions has limited the acquisition of palaeoecological data (Horowitz, 1992;Carrión, 2002b). The lack of data for semi-arid and arid bioclimatic areas has been highlighted in Holocene climatic syntheses in the Mediterranean (Tzedakis, 2007;Roberts et al., 2011), leading to 30 geographical gaps in the outputs from climate modelling simulations (Brayshaw et al., 2011;Peyron et al., 2011).
Analysis of past environmental dynamics in pre-Saharan Tunisia is crucial and should allow the characterisation of: (1) the biogeographical history of the present landscape, (2) the resilience of vegetation in response to aridity and human disturbance and (3) the processes degradation and desertification of desert margins (Schaaf, 2008). Recent work undertaken on the continental Sebkha Mhabeul in southern Tunisia (Schulz et al., 2002;Marquer et al., 2008) and the Halk el Menjel 10 sebkha-lagoon in central Tunisia (Lebreton and Jaouadi, 2013;Lebreton et al., 2015) has already revealed the huge potential of sebkhas for pollen analysis. Sebkha Boujmel, on the southern coast of Tunisia, has yielded the first continuous geochemical and pollen record for the last 8000 years. The sequence reveals the vegetation history in the transition zone between the Mediterranean and the Sahara. In this study, vegetation dynamics and climatic changes are integrated in a transect of the central Mediterranean in order to identify their rhythmicity and to correlate them with regional and global 15 climate changes. The human impact is considered within the cultural framework of Tunisian archaeology and history.

Geographical settings, climate and current vegetation
Today, south-eastern Tunisia is dominated by an arid Mediterranean climate having a long dry season with occasional and highly irregular precipitation. Two hydro-climatic gradients are evident from north to south and from east to west, under the combined influence of proximity to the Mediterranean coast and upland topography, which captures humid air masses 20 (Berndtsson, 1989). Three climatic and phyto-geographical territories have been identified for south-eastern Tunisia: the Matmata mountain range, the desert, and the Jeffara pre-desert coastal plain (Le Houérou, 1959, 1995Floret and Pontanier, 1982;Genin et al., 2006) (Fig. 1b).
In the Matmata Mountains zone, to the west, increased humidity on relief up to 700m produces an altitudinal distribution of vegetation (botanical nomenclature is following Le Floc'h et al., 2010) (Fig. 1b). The northern summits of the Matmata 25 Mountains, endowed with a more humid climate (with average precipitation values of between 200 and 300 mm per annum) are part of the upper arid bioclimatic zone. On these summits, a low shrubland of Juniperus phoenicea and Rosmarinus officinalis develops locally. This vegetation gradually gives way to grassy and semi-desert steppe to the east and south due to increased aridity as one descends into the foothills (Fig. 1b).
A gradual transition towards the desert area is observed as soon as humidity, due to proximity to the Mediterranean and the 30 uplands, recedes. A desert bioclimate, with annual precipitation below 100mm, borders on the Matmata mountainous zone and the Jeffara coastal plain with the sand dunes of the Great Eastern Erg to the west and the Sahara to the south (Fig. 1b).
Further east, the Jeffara coastal plain, which forms an extensive pre-desert area delimited at the west by the Matmata Mountains, slopes gradually down to the coast (Fig. 1b). The coastal area features a large number of lagoons and coastal sebkhas (Fig. 1c). The Jeffara Plain forms part of the Mediterranean lower arid bioclimatic zone, with mild winters and 5 average annual precipitation of between 100 and 200 mm. It is occupied by semi-desert steppe, featuring Rhanterium suaveolens, Artemisia sp., Haloxylon scoparium, Gymnocarpos decander, and degraded grassland steppe with Stipa tenacissima on the foothills and plains with a calcareous crust. A halophytic crassulescent steppe, characterised by Amaranthaceae (Halocnemum, Salicornia, Salsola and Suaeda), develops in the saline soils around the sebkhas. On the Jeffara Plain, the anthropogenic impact is significant and includes ancestral pastoralism as well as recent conversion to 10 agriculture. The Jeffara was originally used as rangeland for nomadic tribes and their flocks. The recent anthropogenic impact takes the form of vast olive groves (Fig. 1b). In this area, the olive is outside its natural bioclimatic zone and its cultivation is made possible through dry-farming and traditional systems for controlling surface run-off in the mountains and foothills (Nasri et al., 2004).
The Sekhba Boujmel (33°16'N, 11°05'E, 2 m. asl) is situated on the coast of the Jeffara Plain and forms part of the paralic 15 complex of the Bibane Lagoon (Fig. 1c). This supratidal sekhba is irregularly flooded by rainwater and high tides via the Bibane Lagoon. The sekhba Boujmel and Bibane Lagoon correspond to valleys which were incised by the Wadi Fessi during the Upper Pleistocene (Medhioub and Perthuisot, 1981). The Sebkha Boujmel corresponds to the present-day and Holocene delta of the Wadi Fessi, which rises inland at Tatouine then drains the mountainous region and the coastal plain (Keer, 1976;Medhioub and Perthuisot, 1981) (Fig. 1b). The Holocene sedimentary filling of the Sebkha Boujmel is a typical lagoonal 20 sequence with biodetrital deposits at the base, followed by bioclastic and oolithic carbonate sands and, finally, by detrital facies which are principally of eolian origin (Fig. 2) (Lakhdar et al., 2006).

Core material and samples
The BJM2 core (33°18 '30.96"N, 11° 5'0.68"E, 160 cm depth) was extracted from the north-west of the sebkha in order to 25 carry out pollen and mineralogical analysis (Fig. 1c). The observed effect of core shortening (12.5 %), which commonly occurs in shallow unconsolidated coastal sediments, has been removed using field measurements following the method described in Morton and White (1997). 71 samples were selected for palynological and clay mineralogical analyses.

Chronology
The chronology of the BJM2 core is based on eleven AMS 14 C dates carried out on the organic fraction of the sediment 30 (Table 1). Study of the organic matter shows that it is of mixed origin, composed of marine planktonic/algal material and continental woody material (Lakhdar, 2009). Input of marine origin is also confirmed by the extensive presence along the core of shallow-dwelling benthic foraminifera of the genus Ammonia as well as Ostracoda and Mollusks. The 14 C dates were corrected for the reservoir effect, calculated at 400 yrs for the Mediterranean (Siani et al., 2001), and then calibrated using IntCal13 within the Calib 7.1 programme (Reimer et al., 2013) (Table. 1). The date at the top of the sequence, considered to be predominantly under continental influence, was not corrected for the reservoir effect. 5 The age-depth model (Fig. 2) was elaborated using the Clam software package (Blaauw, 2010;Blaauw and Heegaard, 2012).
Models with age reversals were rejected and best goodness-of-fit was obtained using 3 rd degree polynomial regression. The absence of sedimentary hiatus (Lakhdar, 2009), the number of AMS 14 C dates, and the coherence of the age-depth model, limit the influence of erroneous dates or dates impacted upon by a hard water effect.
The BJM2 core records 8000 years of infill history of the Sekhba Boujmel. The sampling resolution varies as a function of 10 the sedimentation rate. It is around 210 yrs/sample for the period between 8 and 3 ka and with an average of 60 yrs/sample for the rest of the sequence.

Pollen analysis
The pollen extraction follows the standard protocol (Traverse, 2007): 5g of sediment are treated by successive chemical attacks (HCl 18 %, HF 70 % and HCl 10 %), followed by ultrasonic filtration at 5µm. A tablet, calibrated for Lycopodium 15 spores, is added to each sample at the start of the treatment in order to estimate the pollen concentration (Stockmarr, 1971).
Taxonomic determinations are carried out using a Zeiss microscope under x1000 magnification. Identifications follow published pollen atlases (Reille, 1992, Ayyad andMoore, 1995;Beug, 2004) and pollen reference collections of the Prehistory Department of the Muséum National d'Histoire Naturelle and the Institut des Sciences de l'Evolution of Montpellier (ISEM). An average of 300 pollen grains are counted for each sample, and then completed by research of 20 entomogamous taxa that are common in arid regions and under-represented in pollen assemblages (Horowitz, 1992;Carrión, 2002b).
The pollen/vegetation relationship and the climatic affinities of the pollen flora are established on the basis of specialised flora (Le Houérou, 1959Houérou, , 1969Houérou, , 1980Houérou, , 1995Pottier-Alapetite, 1979, 1981Floret and Pontanier, 1982;Chaieb and Boukhris, 1998;Ozenda, 2004), completed by surface pollen spectra which are representative of the different regional environments 25 (Jaouadi et al., accepted). The Amaranthaceae are a significant floristic component of the desert margins of southern Tunisia.
Several botanic genera occupy either desert habitats or saline hollows along the coast. In order to separate the regional xerophytic from the local halophytic signal, the pollen morphology of the principal genera of desert Amaranthaceae in southern Tunisia (Anabasis, Cornulaca, Haloxylon and Traganum) has been studied with reference to the large ISEM reference collection and to previously published works (Nowicke, 1975;Nowicke and Skvarla, 1979;Salzmann and Schulz, 30 1995). Two pollen types have thus been distinguished: Amaranthaceae-type, which essentially groups together the halophytic taxa, and Cornulaca/Traganum-type which principally encompasses the xerophytic genera.
The pollen data are presented in the form of a simplified diagram of the most significant taxa grouped together in three ecological units, which are characteristic of the regional vegetation: Mediterranean xerophytes, steppic herbaceous plants, and desert herbaceous plants (Fig. 3). The percentages are calculated with respect to a basic sum that only includes these three groups. The percentages of aquatic-, Amaranthaceae-type-, long-range transported-, cultivated-, nitrophilous-and introduced taxa are calculated on the basis of a total sum of pollen grains identified within each pollen spectrum (Berglund 5 and Ralska-Jasiewiczowa, 1986). Olea was not included in the basic sum and is included in the cultivated taxa as soon as its record coincides with that of the latter ( Fig. 3 and Fig. 5). The Wet/Dry ratio (R/D) developed by Hooghiemstra (1996) is applied to the Sekhba Boujmel pollen data (Fig. 4n) in order to trace the changes between the xerophytic herbaceous vegetation of the steppe and desert (Dry amount = Asteraceae excluding Artemisia + Amaranthaceae Cornulaca/Traganumt.) and the steppic grasses that develop in more humid conditions (Wet amount = Poaceae + Cyperaceae) (Hooghiemstra, 10 1996;Mercuri, 2008;Giraudi et al., 2013;Cremaschi et al., 2014).
The coherence and correlation between the ecological groups, the taxa used for the W/D ratio, the pollen data and the clay mineralogical data, are provided by a Correlogram reordered based on the first axis of the Principal Component Analysis (Friendly, 2002). The processing of the pollen data is carried out using the "Rioja" package of the "R" software (R Core Team, 2013;Juggins, 2015). The Local Pollen Zones (LPZ) are determined by visual observation of the pollen diagram and 15 are validated by applying constrained hierarchical clustering by CONISS (Fig. 3) (Grimm, 1987). Long term changes in the vegetation are represented by applying a cubic smoothing spline method to the data (Fig. 4).

Clay mineralogy
A total of 68 samples were subjected to qualitative and semi-quantitative analysis of the clay fraction (< 2 µm) using x-ray diffraction. This procedure was carried out using a Bruker D4 Endeavor diffractometer coupled with a Lynxeye rapid 20 detector fitted with a copper anticathode. The samples were prepared using a standard protocol (Bout-Roumazeilles et al., 1999). The samples are broken down in distilled water, decarbonated using HCl N/5, and then deflocculated by repeated rinsing with distilled water. The suspensions obtained are placed in pill-boxes and the micro-aggregates eliminated by a microhomogeniser. The granulometric fraction below 2 µm is separated by collecting the upper part of the suspension after 1hr 15mins of decantation, and is then centrifuged. Diffractometric analysis is carried out on three preparations: untreated, 25 glycolated, and heated. Clay mineral identification is carried out on the three preparations (Brindley and Brown, 1980). The semi-quantitative analysis is based on the integration of the signals for the main peaks of the clay minerals using the MacDiff software programme (Petschick, 2001).

Pollen data
The pollen record for the Sebkha Boujmel features good pollen grain preservation, with an average pollen concentration of ca. 35000 grains per cm -3 , and wide taxonomic diversity with 114 pollen taxa being recorded.
For the Holocene, the Sebkha Boujmel, which was fed by the Wadi Fessi, has revealed fluvial and eolian pollen inputs from 5 local and regional vegetation (Fig. 1b). Pinus is never overrepresented, indicating a moderate maritime influence and pollen inputs from a catchment area that is limited to the Jeffara Plain, the Matmata highlands and the desert margins. However, a few non-native pollen grains from the temperate regions of North Africa and continental Europe (Alnus, Betula, Corylus, Carpinus/Ostrya, Cedrus and Myrica) have been identified throughout the sequence. These taxa, which are also recorded in other desert regions, demonstrate long-distance transportation of pollen by atmospheric circulation from the Mediterranean 10 to the desert regions (Cour and Duzer, 1980;Schulz, 1984). In fact, the diversity in the pollen assemblages from the Sebkha Boujmel reflects the local steppic zones, the mountainous hinterland and the desert region.

Clay composition and origins
Mineralogical analyses indicate that the clay fraction of the sediments is composed, on average, of 40 % kaolinite, associated with 25% palygorskite and 25% illite. Chlorite and smectite are less abundant and represent 7% and 2% of the clay 20 assemblage respectively. All clay minerals could have been transported to the Sebkha Boujmel via eolian processes or via the Wadi Fessi fluvial system. Both processes are rather irregular, and are seasonnaly modulated by the precipitation regime.
Illite mainly results from the process of physical erosion. It is often abundant in wind-blown particles and indicates a Saharan origin when it is associated with palygorskite (Foucault and Mélières, 2000; Goudie and Middleton, 2001).
Palygorskite is a mineral that is typical of arid and semi-arid Mediterranean environments, which are characterised by 25 alternating dry and humid periods. Palygorskite is a fibrous clay mineral abundant in wind-blown deposits and rare in river sediments since the elongated palygorskite particles are easily broken during fluviatile transportation. It is abundant in northern Saharan desert dusts and in Tunisian dunes and loess, particularly Matmata loess (e.g. Coudé-Gaussen and Rognon, 1988;Bout-Roumazeilles et al., 2007). In North Africa, kaolinite is more abundant in the central and southern Sahara, in Sahelian zones, where it is produced through intense hydrolysing processes, and in the eastern Sahara than in the western 30 Sahara (e.g. Caquineau et al., 1998). As a result, an increase in the proportion of kaolinite at the expense of illite in the eolian fraction suggests a dominant Sahelian origin for the clay particles (O'Hara et al., 2006;Hamann et al., 2009). The illite to kaolinite ratio (I/K) is, therefore, used to distinguish the respective contributions from Saharan and Sahelian sources in eolian deposits (Fig. 4p) (e.g. Caquineau et al., 1998;Formenti et al., 2011). Comparison of the Sebkha Boujmel samples with those from North African sands and sediments in terms of the I/K ratio and the percentage of palygorskite, indicates that the sediments are a mix of inputs of Tunisian loess (I/K= 1.5 and 45% palygorskite) and inputs from the Saharan zone (I/K = 0.4-2; 5-25% palygorskite). The rather high I/K ratio observed in the Sebkha Boujmel sediments, ranging between 0.5 5 and 0.9, rules out any major contribution from the Sahelian region (I/K = 0.1), but is consistent with mixed contributions from the south/central Sahara (I/K= 0.4) and the north/west Sahara (I/K= 0.5-0.7, up to 2) (e.g. Caquineau et al., 1998).

Biogeography and Holocene palaeoenvironment in southern Tunisia
Modern pollen rain from the vicinity of the sebkha and the mountainous hinterland tie in well with the regional 10 biogeogeography. In these arid areas, the modern pollen spectra display a very low presence of Mediterranean taxa that require more humid conditions (Jaouadi et al., accepted). Today, only degraded relics of Mediterranean vegetation still survive on the wetter northern peaks of the Matmata Mountains (Le Houérou, 1959Houérou, , 1969Houérou, , 1995Pottier-Alapetite, 1979, 1981Chaieb and Boukhris, 1998). However, Mediterranean vegetation, and particularly Pistacia, appears to have been well developed at the beginning of the Middle Holocene (Fig. 3). The uplands to the west of the Sekhba Boujmel probably 15 permitted the development of Mediterranean vegetation in southern Tunisia, alongside the Matmata Mountains at least as far as the latitude of Tataouine (Fig.1b), during wetter climatic episodes. However, the Mediterranean vegetation did not extend as far as the mountainous areas of Libya further south. In the Libyan part of the Jeffara, the woody species that developed at Jbel Gharbi between 9.4 and 5 ka constitute a desert-adapted shrubland formation (Capparis, Ficus, Salvadora persica and Tamarix) in which Mediterranean taxa remain poorly represented (Giraudi et al., 2013). Here the regional geographical  (Brun, 1992). The maximum extension of the Oleo-lentiscetum on the coast of the Gulf of Hammamet occurs at the 400 mm isohyet, which marks the limit of the semi-arid bioclimatic stage (Lebreton et al., 2015). At Sebkha Boujmel, Olea is recorded with low rates, reflecting a presence limited to the wettest biotopes. The Pistacia pollen taxon might correspond principally to P. atlantica. P. atlantica is more xerophytic than P. lentiscus and O. 30 europaea subsp. europaea var. sylvestris, with a wider distribution stretching from the sub-humid area to the Saharan area (Le Houérou, 1969). Moreover, its presence fits well with the structure of the pollen flora recorded at the Sebkha Boujmel.
Taking these biogeographical elements into account, over the course of the Middle Holocene, the base of the sequence indicates an open landscape with grass steppe in the coastal lowlands and woody steppe, featuring Pistacia atlantica, Juniperus phoenicea and Rhus tripartita, on the foothills and along the waterways. Towards the northwest, the wetter highlands of the Matmata Mountains (Fig.1b) would have been occupied by Pistacia lentiscus, Olea europaea subsp. europaea var. sylvestris and Cistus. It is also possible that Quercus ilex occurred towards the north on the more favourable 5 mountain slopes. The regional landscape of the plain would have remained relatively open with herbaceous steppes among which grass steppes were predominant. The arid phases are attested to in the development of xerophytic Amaranthaceae while Artemisia, although present, is never very abundant during the Middle Holocene.

Aridity trend and the Mid-to Late Holocene transition 10
Between ca. 5.7 and 4.6 ka, pollen data reveal a major shift towards the progressive establishment of steppic and pre-desert ecosystems, within which Mediterranean trees and shrubs play no role (Fig. 4). These environmental changes reflect the considerable impact of the climate during this period. A significant drop in humidity is recorded between ca. 5.7 and 4.6 ka, with two high-amplitude arid events (see Fig. 4, BJ6: 5.75 -5.5 ka and BJ5: 4.9 -4.6 ka). The abrupt decrease in pollen grains from aquatic plants reflects reduced precipitation associated with a reduction in permanent and/or seasonal water 15 bodies (Fig. 4). Such drastic change is accompanied by a drop in the W/D (wet/dry) and I/K (illite/kaolinite) ratios which attests to greater aridity and enhanced mobilization of fine particles over long distances, probably from the central Sahara (Fig. 4). These arid events are of major significance because they occur during a period that was globally more humid.
Subsequently, from about 4.5 ka onwards, pollen data and clay mineralogy indicate short-term, wet climatic events that are, in fact, phases where there is a slight tendency towards greater humidity during a period which, on the whole, is drier (Fig.  20 4). These observations could alternately be interpreted as episodic pulses in the Wadi Fessi fluvial system. Following a return to relatively wet conditions around 4.5 to 4 ka, aridity becomes progressively more pronounced from ca. 3.9 ka onwards, with a continuous and irreversible decrease in the percentages of pollen grains from aquatic plants and Mediterranean species as well a drop in the W/D and I/K ratios (Fig. 4). The co-eval decrease in the I/K ratio and palygorskite content (from 35% to 20%) suggests a decreased contribution of Tunisian loess and an enhanced contribution 25 from the central Sahara.
Several records, from the Mediterranean and from the Sahara, reveal similarities with those from the Sebkha Boujmel and indicate a significant trend towards aridification between ca. 5.7 and 4.6 ka (Fig. 4). This dynamic is also evidenced at Chott Rharsa where a definitive drop in the water table was accompanied by renewed sand dune formation after 5.6 ka (Swezey et al., 1999;Swezey, 2001). In the Libyan part of the Jeffara Plain, a return of eolian sedimentation occurs between 5.6 and 5.4 30 ka, followed by the end of the humid period at around 5 ka (Giraudi et al., 2013). On the northern margins of the Algerian Sahara, the definitive drying up of the Holocene palaeolakes, dated to ca. 5.2 ka (Callot and Fontugne, 2008), is followed by the formation of a calcareous crust between ca. 4.5 and 3ka at Hassi el Mejnah (Gasse, 2000(Gasse, , 2002. Further south, the establishment of hyper-arid climate is evident in the pollen data from Wadi Teshuinat and the Takarkori rock shelters in the Central Sahara between 5.7 and 4.6 ka (Mercuri, 2008;Cremaschi et al., 2014). Further north, in central Tunisia, significant changes in climatic conditions and ecosystems occur simultaneously with: 1) the end of the growth phase of a stalagmite at La Mine Cave at ca. 5.6 ka (Genty et al., 2006); 2) the transition from a permanent to a temporary lake at Sebkha Kalbiyya 5 after 6 ka (Boujelben, 2015); and 3) an arid episode between 5.5 and 5 ka associated with the development of Pistacia, based on pollen data from the Halk el Menjel sebkha-lagoon (Lebreton and Jaouadi, 2013). A transition to arid climate is also revealed by isotopic data from Gueldaman Cave in northern Algeria, with a bipartite arid phase identified between 5.7 and 5.2 ka (Ruan et al., 2016). In the south-western Mediterranean, pollen and limnological data from Lake Siles in southern Spain also show a major phase of dessication at around 5.4 -4.8 ka (Carrión, 2002a). At the same time, between 5.5 and 4.5 10 ka, a major arid phase is recorded in the sediments and marine cores from the Alboran Sea (Combourieu Nebout et al., 2009;Fletcher et al., 2012), the Adriatic Sea (Combourieu-   All of these contemporaneous events seem to indicate climatic mechanisms at the scale of the southern and eastern 15 the African monsoon never seems to have reached the Mediterranean coast (e.g. Arz et al., 2003;Tzedakis, 2007) and the stable isotope composition of groundwaters of the Great Eastern Erg suggest winter precipitation of northern origin, which is comparable to present-day winter precipitation resulting from the southward shifting of the mid-latitude westerlies (Gasse, 2002;Guendouz et al., 2003;Edmunds et al., 2004).
At the Sebkha Boujmel, the humid/arid transition, which is contemporary with the end of the AHP, may have resulted from 5 change in the configuration of global climatic drivers, particularly orbital parameters regulating insolation and Mediterranean atmospheric conditions. Changes in atmospheric and ocean circulation are mainly controlled by the Earth's orbital forcing (obliquity and precession) which acts on the climate system by controlling the seasonal and latitudinal distribution of insolation and temperature (e.g. Davis and Brewer, 2009;Brayshaw et al., 2010). Climate simulations indicate that enhanced seasonality of Early Holocene insolation, with minimum precession and maximum obliquity, generates i) an 10 enhancement of the African monsoon, and ii) increased storm track activity and winter precipitation over the Mediterranean (Brayshaw et al., 2011;Kutzbach et al., 2014;Bosmans et al., 2015). According to Kutzbach et al. (2014), the influence of these orbital parameters (summer insolation maxima and winter insolation minima) could have generated a Holocene humid period in the Mediterranean, between 30°N and 45°N, by increasing both winter and summer precipitation. Furthermore, pollen-inferred precipitation values from Lake Trifoglietti and Lake Pergusa in southern Italy also point to significant winter 15 and summer precipitation during the humid phases of the Early to Mid-Holocene at latitudes south of 40°N . During the changeover between the Middle and Late Holocene, the change in orbital parameters, and the weakening of the summer insolation, would, therefore, have led to a weakening of winter and summer storm tracks and more pronounced aridity over the southern Mediterranean, coeval with the end of the AHP.

Holocene Centennial-scale climate events in southern Tunisia since 8 ka 20
The pollen record of the Sebkha Boujmel highlights eight aridity pulses that punctuate the Middle and Late Holocene. These dry events, numbered BJ8 to BJ1, occur at ca. 8 -7.75, 7 -6.5, 5.75 -5.5, 4.9 -4.6, 3.7 -3.25, 3 -2, 1.4 -1.1 and 0.4 -0 ka respectively and alternate with seven more humid phases (Fig. 4). The magnitude of the BJ8 arid event, at the base of the sequence, between ca. 8 and 7.75 ka, could be associated with the 8.2 global event (Alley et al., 1997;Mayewski et al., 2004) and a significant instance of aridity recorded in the Mediterranean and in sites at the same latitude as the Sebkha 25 Boujmel, such as Sebkha Mellala (Gibert et al., 1990). Given the uncertainty regarding 14 C dates at the bottom of the sequence, BJ8 could be contemporaneous with the arid event recorded in various cores from the western (MD95-2043) and The four subsequent arid phases (BJ7 to BJ4) may be connected with available data from further south in the Libyan Jeffara 30 (Giraudi et al., 2013) and in the Sahara (Cremaschi et al., 2006;Mercuri, 2008;Cremaschi et al., 2014), particularly with the end of the AHP (Fig. 4). These events may also be linked to events recorded in the Mediterranean. In fact, numerous pollen records from the Mediterranean identify recurrent arid phases which are contemporaneaous with North Atlantic Cooling events (NAC) (Bond et al., 1997;Bond et al., 2001;Wanner et al., 2011). These dry episodes occur during the regression of Mediterranean plant groups and the increase in semi-desert associations in north-eastern Tunisia (Desprat et al., 2013), and are also evidenced by the decline of deciduous Quercus forest in Italy and Greece (Kotthoff et al., 2008a(Kotthoff et al., , 2008bSchmiedl et al., 2010; and the western Mediterranean (Combourieu Nebout et al., 2009;Fletcher et al., 2010;Fletcher et al., 2012;Fletcher and Zielhofer, 2013). Such vegetation dynamics may be linked to colder and/or more 5 arid climate (Combourieu Nebout et al., 2009;Desprat et al., 2013).
The 4.2 ka event, separating the Middle and Late Holocene (Walker et al., 2012), is seen to be associated with the NAC 3 event, which is evident on the northern side of the Mediterranean basin (Magny et al., 2009;Peyron et al., 2011), in the Siculo-Tunisian Strait (Desprat et al., 2013), in the Medjerda Valley Zielhofer et al., 2004) and in the speleothems of Gueldaman Cave (Ruan et al., 2016). However, evidence for this event is not very clear in the Sebkha 10 Boujmel record. Moreover, the 4.2 ka event is not recorded in core sites ODP 976 and MD95 2043 (Fig. 4), nor is it recorded in the eastern Mediterranean (Finné et al., 2011). The last three arid episodes recorded in the Sebkha Boujmel sediments (BJ3, BJ2 and BJ1) are well correlated with the Late Holocene NAC 2, NAC 1, and NAC 0 events, respectively (Fig. 4).
The centennial-scale climatic events recorded in southern Tunisia from 8 ka onwards, may indicate a climatic coupling between the southern Mediterranean and the Sahara during the Middle Holocene. Subsequently, in the Late Holocene, and 15 from 3 ka onwards, atmospheric coupling was established with the North Atlantic. Significant changes in orbital-, solar-, and ice-sheet climate forcing, as well as significant reorganisation of the global climate system and atmospheric circulation at the time of the Mid-to Late Holocene transition, might explain these modifications (Debret et al., 2009;Magny et al., 2013).

Late Holocene landscape evolution and human impact
Human impact has often been considered as overriding during the Holocene, shaping Mediterranean and Saharan landscapes 20 under the effects of agriculture, pastoralism and vegetation clearance (Pons and Quézel, 1998;Schulz et al., 2009). In southern Tunisia, therefore, the vegetation landscape would be the result of significant human impact during the historical period, characterized by the degradation of previously more diverse natural vegetation, consisting of Mediterranean shrubby forest accompanied by tropical steppe with Acacia tortilis subsp. raddiana (Le Houérou, 1959, 1969Frankenberg, 1986).
In arid environments, it is still difficult to distinguish between human and climatic factors solely on the basis of pollen 25 analysis because several of the marker taxa for anthropisation belong to the natural vegetation of arid regions (Horowitz, 1992). However, in central and southern Tunisia, studies of the dynamics of pre-desert vegetation relevant to human activity, particularly pastoralism, allows nitrophilous and unpalatable taxa to be distinguished within the regional vegetation (Le Houérou, 1959Houérou, , 1980Tarhouni et al., 2010;Gamoun, 2014). Crossing the nitrophilous and unpalatable taxa with cultivated and introduced taxa allows the completion of the picture and the interpretation of anthropogenic indicators (Fig. 5). 30

Landscape origin and human impact
The Epipalaeolithic Capsian culture developed in the Maghreb during the Holocene, between ca. 10 and 7 ka, and was replaced by the 'Neolithic of Capsian Tradition' (NCT) around 7 ka (Jackes and Lubell, 2008;Mulazzani, 2013). These cultural entities are largely distinguished on the basis of typological analysis of their lithic industries. However, in the Maghreb, the history of neolithisation and the emergence of a production economy are still not well understood. To date, 5 very few studies have provided information on the anthropological and economic aspects of Holocene societies in the Maghreb, and for human/environment interactions in this important period of cultural transition (Lubell et al., 1976;Roubet, 2003;Jackes and Lubell, 2008;Mulazzani, 2013). In Capéletti Cave (Aurès, Algeria), the NCT appeared around 7 ka, associated with a pastoral economy based on the rearing of sheep and goats (Roubet, 2003). Archaeological data from the Jeffara Plain and the Jebel Gharbi (Libya) attest to the development of the NCT from 8 ka, with pastoral encampments, 10 typical NCT lithic industries, and ceramics (Barich et al., 2006;Lucarini, 2013;Barich, 2014). Evidence suggests continuity in pastoral activities and transhumance in an area stretching from the foothills of Jebel Gharbi to the Jeffara Plain and on to the coast, up until the historical period (Lucarini, 2013). In the environs of the Sebkha Boujmel, rammadyats (open air sites), although damaged by recent human activity, are still visible and attest to Capsian and/or Neolithic occupation. The Holocene occupation of the region is also evident in the environs of the nearby Sebkha El Melah (Fig.1c) where sites attributed to the 15 NTC are dated to the 7 th milennium cal BP (Perthuisot, 1975). Therefore, it appears that ecological niches formed by the paralic sebkhas were occupied and exploited by prehistoric communities during the Holocene period in central Tunisia.
At the Sebkha Boujmel, human impact becomes evident after 3 ka with the significant and ongoing appearance of 20 nitrophilous taxa linked to pastoral practices (Fig. 5). No significant anthropogenic impact deriving from agricultural and/or pastoral activities has been recorded previously in the local ecosystems. At that time, the prehistoric societies had only a weak impact on the local environment. The emergence of present-day landscapes can, therefore, be correlated with important climate changes that occurred during the Mid-to Late Holocene transition. In fact, the presence of Mediterranean woodland vegetation in southern Tunisia is attested to at an early date, i.e. around 5 ka, on the island of Djerba (Damblon and Vanden 25 Berghen, 1993). Mediterranean woody species are well represented between 8 and 5.5 ka in the pollen record from the Sebkha Boujmel (Fig. 3). Such woodland formations are indicative of more humid climatic episodes up until 5.5 ka and the AHP termination, before the onset of increasing aridity. Therefore, in the Late Holocene and during the historical period, the vegetation landscape of southern Tunisia was already a steppic-and semi-desert formation in which woodland formations were limited. The same vegetation structure can be observed in pollen data spanning the last two millennia from the nearby 30 Sebkha Mhabeul (Salzmann and Schulz, 1995;Marquer et al., 2008). The dynamic of vegetation associations revealed at the Sebkha Boujmel before 3 ka, leaves no doubt regarding the primary role played by climate in the environmental changes that occurred during the Middle and Late Holocene. Nonetheless, an anthropogenic impact may have amplified the environmental changes of climatic origin during this period.

Landscape evolution during historical periods
Human impact increased over the last three centuries BC, during the historical period, with the continuous presence of nitrophilous taxa being recorded and testifying to the permanence of pastoral activities (Fig. 5). From this time onwards, the 5 impact of agricultural activity becomes evident in the record with the addition of cultivated taxa, primarily Olea (Fig. 5).
During the Libyo-Phoenician period, since at least the 5 th century BC, there is evidence for a high level of human occupation with several urban settlements in the immediate environs of the Sebkha Boujmel and the Bibane lagoon (Fig. 1b) (Trousset and Paskoff, 1991;Drine, 1993;Mattingly, 1995). While the Periplus of Pseudo-Scylax attests to the management of wild olive trees on the nearby island of Djerba as early as the 4 th century BC (Shipley, 2011), our data indicate actual cultivation 10 of Olea on a large scale from the 3 rd century BC onwards. Furthermore, the record for Vitis between the 2 nd century BC and the 2 nd century AD (Fig. 5) can be related to widespread cultivation of vines, the pollen of which does not disperse widely.
Notwithstanding the arid and pre-desert environment, Vitis could have played an important role in the regional economy, particularly in the mountainous regions and in oases where suitable irrigation techniques would have allowed its development. Seed remains from the Garamantes oases in Fezzan (Libyan Sahara) attest to the cultivation of vines from the 15 beginning of the 1 st millennium BC (van der Veen, 1992;Mattingly et al., 2003), well before the consumption and/or introduction of other Mediterranean fruit crops (Punica granatum and Olea europaea subsp. europaea var. europaea) towards the end of the 1 st millennium BC (Pelling, 2005).
In the surroundings of the Sebkha Boujmel, the cultivation of olive trees remained relatively important throughout the first two centuries AD. However, a gradual decline in its cultivation is recorded from the beginning of the Roman period (Fig. 5). 20 These changes may be related to the progressive decline in the economic and administrative importance of urban centres close to the Sebkha Boujmel during the Roman period (46 BC -429 AD), for example the cities of Zitha and Gergis (Mattingly, 1995). Furthermore, Roman oil presses have been discovered which display technical particularities that attest to low levels of local oil production and modest olive cultivation (Mrabet, 2011). From the 3 rd century AD onwards, during the Vandal, Byzantine and Early Medieval periods, including the Arab conquest (from 670 AD), low percentages for cultivated 25 plants indicate that agricultural activity was probably confined to the areas around Berber villages in the mountainous hinterland and that the lowlands were once again dominated by a pastoral economy.
The development of agricultural activities in south-eastern Tunisia between the 3 rd century BC and the 2 nd century AD coincides, in part, with a slightly more humid period which is recorded in the pollen and clay data between ca. 2 and 1.5 ka (50 BC -450 AD) (Fig. 4). This episode, the Roman Humid Period (RHP), is also recorded at other sites. At the Sebkha 30 Mhabeul, for example, the RHP is associated with a phase of hydrological stability which ends around 430 AD (Marquer et al., 2008). In the Medjerda Valley, the RHP coincides with a phase of stability and soil formation up until 1.7 ka . Further east, in the Dead Sea, the RHP is associated with an increase in precipitation (Neumann et al., 2010). A surge in agriculture associated with slightly more humid climate is also documented during the Roman period in other Mediterranean desert margins such as north-western Libya (Gilbertson et al., 1996), in Cyrenaica (Hunt et al., 2002), and at Wadi Faynan in southern Jordan (El-Rishi et al., 2007;Hunt et al., 2007).
The data from Sebkha Boujmel indicates a return to arid climate between around 1.4 and 1.1 ka (BJ2: 550 -850 AD), contemporary with the Dark Ages (DA) (Fig. 4). This phase is also marked by unstable climatic conditions, associated with a 5 rise in the frequency and intensity of flood events from 550 AD onwards and of arid conditions up to 950 AD at the Sebkha Mhabeul (Marquer et al., 2008) and by a renewal of river activity in the Medjerda Valley .
Between about 1.1 and 0.5 ka (850 -1450 AD), the W/D ratio, as well as a drop in herbaceous desert plants, indicate a humid period in southern Tunisia, which, while of modest magnitude, was more important than that recorded during the RHP (Fig. 4). Associated with these wetter conditions, a significant peak in Artemisia occurs between ca. 1.1 and 0.8 ka (850 10 -1150 AD) (Tab. 2, LPZ5 and Fig. 5), which essentially corresponds to the Medieval Climate Anomaly (MCA). However, it is still difficult to interpret these very high Artemisia values, which also feature in the pollen data for the Gulf of Gabes (Brun and Rouvillois-Brigol, 1985), as solely reflecting a change in climate. In fact, during the various humid/arid episodes identified between 8 and 4 ka, Mediterranean and desert taxa fluctuate significantly, but not wormwood ( Fig. 3 and Fig. 5).
In addition, Artemisia does not develop in the Sahara (Ritchie et al., 1985;Ritchie and Haynes, 1987;Lézine et al., 2011), 15 nor on the island of Djerba (Damblon and Vanden Berghen, 1993) during the Early and Middle Holocene. Therefore, Artemisia had only a limited presence in the vegetation groups of the southern Tunisian desert margins, which instead were dominated by grassy steppes and which were in direct contact with desert ecosystems (Amaranthaceae) over the course of the Early and Middle Holocene. In the same period, Artemisia was well developed further north in central Tunisia (Lebreton and Jaouadi, 2013). Its significant development, associated with that of nitrophilous taxa in southern Tunisia during the Late 20 Holocene, was probably, therefore, linked to human activity and, in particular, to pastoralism. In fact, with the exception of Artemisia herba-alba, the other species of Artemisia, such as A. campestris, are not particularly palatable and are avoided by grazing animals (Le Houérou, 1980;Tarhouni et al., 2010;Gamoun, 2014). In fact, studies of the dynamics of current steppic vegetation indicate that Artemisia campestris is a pioneer species in its occupation of Rhanterium suaveolens sandy steppes which have been degraded by vegetation clearance and over-grazing on the Jeffara Plain (Chaieb and Zaâfouri, 2000;25 Genin et al., 2006). Furthermore, historically this period (LPZ5, 850 -1150 AD) corresponds to a period of climate instability with a succession of droughts and famines inducing a decline in the urban and economic structure of the Maghreb.
A substantial shift towards pastoralism was also favoured by the migration of nomadic Banu Hilal tribes from southern Egypt to North Africa during the 11 th century AD (Allaoua, 2003). Thus, at the Sebkha Boujmel, the development of Artemisia during the Late Holocene, and particularly between ca. 1.1 and 0.8 ka, may have resulted from a combination of 30 several factors including deterioration in plant diversity under the pressure of intense pastoral activity and an unstable climate with significant seasonal and/or multi-year contrasts.
The humid period recorded at Sebkha Boujmel between ca. 1.1 and 0.5 ka (850 -1450 AD), and associated with the MCA, is followed between ca. 450 and 95 cal yr BP (BJ1, 1500 -1855 AD) by a drop in the W/D ratio and the progression of desert taxa (Fig. 4). Arid conditions are re-established in southern Tunisia and are contemporary with the Little Ice Age (LIA, ca. 1450(LIA, ca. -1850. This arid climatic episode during the LIA coincides with an increase in clastic deposits at the top of the sedimentary fill from the nearby Sebkha Mhabeul (Marquer et al., 2008), the renewal of river activity in the Medjerda Valley (Faust et al., 2004), and is also contemporary with the most recent Bond Event (NAC 0) at around 0.4 ka (Fig. 4).
This succession of Wet/Dry climatic conditions, which occurred during the MCA/LIA transition in southern Tunisia, is also 5 evident in other palaeoecological records from around the Mediterranean basin, particularly in sedimentary sequences from caves in Cyrenaica (Hunt et al., 2010;Hunt et al., 2011) and Wadi Faynan in Jordan (Hunt et al., 2007), in pollen data from the coast of northern Syria (Kaniewski et al., 2011) and from the Dead Sea (Neumann et al., 2010), and Siles Lake in southern Spain (Carrión, 2002a). However, in the western and central Mediterranean, several proxies reveal a reversed succession of climatic conditions, with an arid climate during the MCA and more humid conditions during the LIA. In the 10 central Mediterranean, this climate trend is recorded in the pollen data from Lake Pergusa in Sicily (Sadori et al., 2016) and at Lago di Venere on the island of Pantelleria (Calò et al., 2013). For the western Mediterranean, these climatic conditions (i.e. arid during the MCA and humid during the LIA) are evident in dendroclimatological data obtained from Cedrus atlantica in northern Morocco (Esper et al., 2007), in multi-proxy data from Lake Zoñar in southern Spain (Martín-Puertas et al., 2008), and in marine cores from the Alboran Sea (Nieto-Moreno et al., 2013). 15 In fact, over the course of the last millennium, the MCA (900 -1350 AD) and the LIA (1500 -1850 AD) climatic episodes exhibit significant anomalies in temperature and atmospheric circulation in the northern hemisphere which could have been triggered by changes in solar irradiance (Mann et al., 2009;Graham et al., 2011;Swingedouw et al., 2011). Data from Morocco and Spain indicate a strong correlation between arid/humid variability and positive/negative modes of the NAO (Esper et al., 2007;Martín-Puertas et al., 2008;Trouet et al., 2009;Nieto-Moreno et al., 2013;Wassenburg et al., 2013). 20 Over the course of the MCA/LIA, the marked regionalisation as well as the divergent responses between the central and western Mediterranean on the one hand, and the southern and eastern Mediterranean on the other, may reflect the greater impact of the NAO on storm tracks in the northern and western Mediterranean (Trigo et al., 2004;Trouet et al., 2009). The Mediterranean expression of the NAO, the Mediterranean Oscilliation (MO), could have driven the contrasts revealed in the palaeoecological records because it would have generated pressure and hydrological contrasts between north-western and 25 south-eastern Mediterranean areas (Dünkeloh and Jacobeit, 2003).

Current landscape degradation and desertification processes
The past century has witnessed profound change in the regional ecosystems of southern Tunisia (Fig. 5). Desert taxa are proliferating around the Sebkha Boujmel, as are other anthropogenic indicators which reflect the increasing impact of human activities on the environment (Fig. 5). Thus, for example, recently introduced xenophytic taxa, such as Casuarina, Acacia 30 cyanophylla and Eucalyptus are present in the Sebkha Boujmel pollen record. Olive growing is becoming more and more developed, and the rise in Olea is paralleled by the records for other cultivated taxa (Phoenix and Cerealia) (Fig. 5).
Pastoralism is indicated by ruderal and nitrophilous taxa (Peganum, Polygonum) and by the more common occurrence of the pollen taxa Tamarix, Thymelaea and those of the Zygophyllaceae family (Nitraria, Fagonia and Zygophillum) (Fig. 3 and 5).
The latter plants, which usually show a preference for desert conditions, are unpalatable and resistant to grazing (Le Houérou, 1980;Tarhouni et al., 2010;Gamoun, 2014). Their development, therefore, reflects deterioration in vegetation cover due to overgrazing, which actually favours their spread. Radical changes in the relationships between humans and the environment are thus recorded in south-eastern Tunisia, particularly in the pollen sequence from the Sebkha Boujmel. In the 5 Jeffara coastal plain, the traditional economy up until the end of the 19 th century was based on extensive pastoralism, ensuring the mobility of people and their herds. At this time, the anthroposystem was relatively balanced in the context of the limited, fragile resources (Talbi, 1997). After 1881 AD, under the French protectorate, these traditional social and production structures were supressed in favour of increased sedentism and the introduction of new systems of agricultural production based on individual ownership of land and intensive olive and cereal production (Abaab, 1986). These economic and social 10 changes resulted in increased impacts on the environment (Talbi et al., 2009). In fact, the disappearance of collective herd management led to a reduction in herd movements and caused locally intensive overgrazing with rapid deterioration of vegetation cover accompanied by an increase in unpalatable substitute species. All of these changes led to an accentuation of wind and water erosion and an intensification of the process of desertification (Sghaier et al., 2012).

Conclusions 15
Pollen and clay mineralogical data from the Sebkha Boujmel underline the significant potential of salt-flat sedimentary records as palaeoecological archives. Sebkhas are particularly conducive to the preservation of pollen in arid regions and, therefore, constitute a rare sedimentary resource favourable to palynology in sub-arid environments and allowing the reconstruction of the dynamics of vegetal environments. Pollen data from the Sebkha Boujmel provide the first, continuous, high-resolution record to allow the detailed reconstruction of the landscape dynamics of south-eastern Tunisia over the last 20 eight millennia. The palaeoecological records from the Sebkha Boujmel tie in with other regional and extra-regional proxies and the impact of global climate changes can be superimposed on local conditions with various degrees of sensitivity. These events highlight the complexity of climatic trajectories in the south-central Mediterranean, with the influence of several closely correlated parameters, such as solar forcing and NAO-type circulation, from ca. 8000 cal BP onwards. Furthermore, the vegetation dynamic suggests that human activity could have amplified the arid climatic event which began in the Late 25 Holocene. The anthropogenic impact became predominant when a sedentary way of life developed, i.e. during Classical Antiquity and, later, under the French protectorate. In the case of the latter period, spanning the last century, anthropogenic changes become evident and have led to profound alterations to the spatial organisation of the territory. The pronounced development of desert taxa and markers of agricultural and pastoral activity, attest to fundamental changes in the relationships between humans and the environment in southern Tunisia at the present time. The joint impact of climate 30 change and human pressure result in a drastic deterioration of the landscape by accentuating the process of desertification.