Introduction

1Since 1997, considerable archaeological and palaeoenvironmental research has taken place in the Ounjougou site-complex in the Yamé River Valley (Dogon country, Mali; Fig. 1), under the framework of the “Human populations and palaeoenvironment in West Africa” programme, coordinated by Eric Huysecom (University of Geneva). This research has enabled us to reconstruct the history of human settlement and its chronostratigraphic context for the entire Holocene (Huysecom, 2002; Huysecom et al., 2004; Ozainne, 2006; Rasse et al., 2006; Le Drézen, 2008; Lespez et al., 2008; Ozainne et al., 2009; Lespez et al., 2011; Ozainne, 2011), as well as the functioning of the fluvio-palustrine system (Rasse et al., 2004, 2006; Le Drézen, 2008; Lespez et al., 2008; Le Drézen et al., 2010; Lespez et al., 2011) and the evolution of the vegetational landscape (Eichhorn and Le Drézen 2006; Le Drézen, 2008; Garnier et al., 2013; Eichhorn and Neumann, in press). The position of the Bandiagara Plateau between the Sahelian and the Sudanese biogeographical realms is particularly interesting for the understanding of how savanna environments responded to climate variations and socio-economic changes (Fig. 1).

2In this semi-arid region, where palaeo-archive environments are rare, the reconstruction of landscape evolution is difficult. At the same time, most studies have focused on lacustrine or marine deposits (Maley, 1981; Lézine, 1989; Ballouche and Neumann, 1995; Salzmann and Waller, 1998; Salzmann et al., 2002; Lézine et al., 2005; Bouimetarhan et al., 2009). Nowadays, climatic variations during the Holocene are fairly well documented. On the contrary, the impact of human societies on the environment and particularly on vegetation is still poorly documented. In this context the biological indicators preserved in alluvial sediments are of great importance as they are very sensitive and reactive to human activities and hydro-climatic changes, and the river channels along with the waterlogged zones connected to them provide an excellent record of the functioning of these environments and these societies. In particular, fire signals are very well recorded in such environments, although very little research on past fires has been done untill now over a long temporal scale of 1000 years or more (Ballouche, 1998; Bird and Cali, 1998; Salzmann et al., 1998; Bird and Cali, 2002; Salzmann et al., 2002; Power et al., 2007). However, many geological, ecological and agroforestry studies consider the mechanisms of vegetation fires in contemporary West Africa (Monnier, 1968; César, 1990; Monnier, 1990; Schulz and Pomel, 1992; Bruzon, 1994; Pomel et al., 1994; Dolidon, 2005; Caillault, 2011).

3In Ounjougou we used various field and laboratory methods. Owing to the poor preservation of the organic remains in the alluvia, it was necessary to develop new indicators. Recent research on the early Holocene has demonstrated the potential of phytoliths for the reconstruction of vegetational environments based on alluvial records (Neumann et al., 2009) and this study was also undertaken for the rest of the Holocene (Garnier et al., 2013). In parallel, it has been possible to develop palaeo-fire research. We propose here a reconstruction of landscape evolution in the Sudano-Sahelian savanna based on these two methods of analysis undertaken on the Holocene sediment records at Ounjougou. At the same time, we will try to demonstrate the efficacy of these new methods for our understanding of the long-term relationship between nature and human societies in environments where the other bio-indicators are not preserved.

The hydro-geomorphological context of the Ounjougou site complex during the Holocene

4The current climate in Sudano-Sahelian Africa follows a zonal distribution, which is principally regulated by the movement of the Intertropical Convergence Zone (ITCZ) and responds to the seasonal migration of the insolation maximum. During the dry season (winter), the ITCZ migrates southwards; West Africa is thus subject to the influence of the Saharan high-pressure systems and to a continental influx from the east (the Harmattan). During the rainy season, the ITCZ migrates north, and that brings about the monsoon airflow. The river regime of the Yamé, which is the main river in the Dogon country, is typically tropical, with a low-water period in the end of the dry season and a high-water period from August to October. The erosion that is visible today on the sides of the valley (Fig. 2) is the result of an overbank flow of the Yamé after a capture in 1936 (Rasse et al., 2006). Regressive erosion has been very rapid in recent times (the surface area of badlands increased five fold between 1950 and 2006; Le Drézen, 2008). The fluvio-palustrine formations, visible in cross section, are very thick (10-20 m) and very rich in organic material, and they can thus provide a precise chronology for the hydro-geomorphological responses of the Yamé and its tributaries. These sedimentary formations rest upon Upper Precambrian sandstone formations, which are still visible in the interfluves of the Bandiagara plateau (Fig. 2).

5Geomorphological and micro-morphological analyses enabled us to reconstruct the evolution of the fluvial system during the Holocene (Le Drézen et al., 2010; Lespez et al., 2011). Four hydro-sedimentary phases have been recorded in the alluvial formations of the Yamé valley. They demonstrate both climate change and an increasingly significant human impact on the landscapes during the last millennia. These phases are in chronological order: the early Holocene (9785-6810 BC), the middle Holocene (6810-4080 BC), the late Holocene (1800-400 BC) and finally the terminal Holocene (after 400 BC). After the upper Pleistocene, which locally corresponds to the “Ogolian” period and is marked by a dry climate (Lespez et al., 2008), a humid optimum during the early Holocene transformed the landscapes completely. The watercourses of the Sahelian zone with their large hydrological capacity stimulated water-erosion and incision (Gasse, 2000), as has been demonstrated all along the Yamé River (Lespez et al., 2011). Gradually, a Sudanese lora developed on the plateaus and the edges of the permanently wet zones, and this limited further erosion. A phase of aridiication occurred 5000-4000 years ago (middle Holocene) with heavy consequences for the landscapes and societies as well as on the hydro-sedimentary functioning of the valleys. During this phase major soil erosion is recorded once more along with increased aggradation (Le Drézen et al., 2006, 2010). his change has been observed in many sites in West Africa (Servant, 1983; Ballouche and Neumann, 1995; de Menocal et al., 2000; Gasse, 2000; Salzmann et al., 2002; Marchant and Hooghiemstra, 2004; Shanahan et al., 2006). In Ounjougou, we observe alternating sand-silt laminae rich in charcoal and plant macro-remains, which constitute the deposits of the rainy seasons, and siltsand laminae rich in carbonaceous particles with positive grading, an oxidation horizon and desiccation cracks typical of dry season deposits. The results confirm, therefore, that during this major climate change seasonality increased and the length of the dry season also increased (Le Drézen, 2008; Le Drézen et al., 2010; Lespez et al., 2011).

Towards a landscape history

The role of phytolith studies in fluviopalustrine environments

6Phytoliths are opal particles formed by precipitation of amorphous silica (SiO₂ nH₂O) between and inside the cells of many plants. Composed of minerals, they are very resistant to chemical and physical attack. They also have the advantage of being preserved in supericial formations and soils, unlike pollen, which are preserved only in the oxygen-deicient sediments of the waterlogged zones. Phytoliths allow the extension of palaeoecological research to arid zones and fluvial environments. Nevertheless, palaeoenvironmentalists have generally favoured pollen or charcoal because their taxonomic signiicance is more precise. Indeed, the two major problems with phytoliths are multiplicity (a taxon can be represented by more than one morphotypes) and redundancy (one and the same morphotype can represent more than one taxon; Rovner, 1971). A phytolith can only occasionally be connected to a single plant species; however, its morphology is often characteristic of large plant families. Besides, the possibility of distinguishing the subfamilies of Poaceae and burning phytoliths can provide precise information on floral composition and on the impact of humans on it. In this study, we have been able to identify three groups of plant species: (i) the woody dicotyledonous plants; (ii) the Poaceae, among which we have been able to distinguish three subfamilies: the Panicoideae, tall grasses typical of the savannas and hydromorphic soils, but which also include domesticated cereals such as millet, sorghum, etc.; the Chloridoideae, the equivalent of the low xerophytic vegetation of the Sahelian savannah; the Bambusoideae, with the species Oxytenanthera abyssinica, which in general grows in the gallery forests of the Sudano-guinean zone; (iii) some specific families, among which the Arecaceae (palm trees that mostly grow along watercourses), the Cyperaceae and the Commelinaceae. These can either be plants of wet environments, or self-propagating. Finally, the Podostemaceae, which are weeds found on rocks in rapid watercourses, are also diagnostic of a specific morphotype.

7Unlike pollen, the phytoliths detected in the soil result from plant decay on a local scale (Fredlund and Tieszen, 1994; Piperno, 2006). Nevertheless, a taphonomic study of phytoliths in alluvial deposits shows that their origins differ according to the way they were transported (Garnier et al., 2013). Generally speaking, the origins of the phytoliths identified in channel deposits are extra-local, that is from the vegetation of the interfluves, while those recorded in the flood plains represent the local gallery forest. Therefore, it is important to take into account the sedimentary features of the deposits in order to determine the spatial scale of the vegetation reconstruction.

The contribution of phytoliths to the reconstruction to landscape evolution in Ounjougou during the Holocene

8We propose here a reconstruction of the valleybottom vegetation based on the study of 20 samples from Holocene floodplain deposits. At first, the middle Holocene phytolith assemblages (6810-4080 BC) indicate closed vegetation dominated by the woody dicotyledonous plants (45-80%) and the Bambusoideae (12-41%). The other Poaceae along with the Arecaceae are barely present and the Commelinaceae, Cyperaceae and Podostemaceae are rare. On the contrary, the rates of burned phytoliths are remarkable, from 6% to over 30%.

9At the beginning of the late Holocene (3370-2700 BC), the phytolith assemblages are more or less as before: the woody dicotyledonous plants are dominant and the presence of Bambusoideae indicates valley-bottom vegetation still as closed as before. During the next period (2700-1800 BC) a noteworthy change is observed in the composition of the phytolith assemblages combined with slight landscape opening: Panicoideae and also Cyperaceae and Commelinaceae phytoliths increase. On the other hand, we observe a significant decrease of burned phytoliths with their rates bordering on 10%.

10The end of the late Holocene (1800-400 BC) shows the recurrence of a more closed vegetation. The difference with the middle Holocene is that the woody dicotyledonous plants are dominant (67-77%) and so are the Arecaceae (16-25%). The most evident change is the strong decrease of the Bambusoideae that represent only 3% to 8% of the total assemblage. The end of this period (LH2C: 900-400 BC) the assemblage is characterised by a net increase of Poaceae phytoliths, which attain between 45% and 51% of the total assemblage. Likewise, the Cyperaceae and the Commelinaceae increase significantly, accounting for 2% to 5%. The woody dicotyledonous plants (30-48%), the Arecaceae (8-10%) as well as the burned phytoliths (4-7%) record lower values than during the previous period.

11Finally, the assemblages of the terminal Holocene (post 390 BC) confirm this tendency of landscape opening: we observed a marked increase of the Poaceae (60-64%), a disappearance of the Bambusoideae and a decrease of the woody dicotyledonous plants (18-19%). The phytolith assemblages identified in the sedimentary fluvial archives at Ounjougou indicate an evolution of the valley-bottom landscapes during the Holocene. The research results show the passage from a rich and closed gallery forest of the Sudano-Guinean type during the middle Holocene (7^th -5^th millennia BC) to a valley-bottom vegetation that is much more open, similar to today’s Sudano-Sahelian vegetation, established over the last three thousand years. Thus, besides the decrease in woody density within the valley-bottom landscapes, phytoliths show a taxonomic change, particularly with the disappearance of the Sudano-Guinean species Oxytenanthera abyssinica in favour to other species closer to northern climates.

The contribution of phytoliths to the definition of control variables affecting landscapes

12The landscape evolution recorded in Ounjougou accords with observations made on a regional scale: a general change of the vegetation connected to an increasingly dry climate due to the precipitation decrease 5000 years ago. However, the climate is not the only variable we need to take into account in relation to this change, because an increased percentage of burned phytoliths are observed between 6000 BC and 1500 BC. Such a record and, moreover, the fact that the phytoliths indicate mostly species of the gallery forest, can only be attributed to human actions. In addition, the phytolith assemblages from sediments dating between 2580-2350 BC are characterised by a sudden increase of self-propagating plants (Commelinaceae et Cyperaceae) and of the Panicoideae, a subfamily to which belongs the millet (Pennisetum glaucum). Due to the discovery of millet grains (Ozainne et al., 2009), we know that agriculture developed in Ounjougou as early as the 17^th -16^th c. BC: we can now suggest than humans practiced agriculture many centuries earlier, that is just after the Saharan populations of the Tilemsi valley, 250 km north from Ounjougou (Manning et al., 2011).

13Therefore, phytolith analysis can provide valuable information on the composition of vegetation, and on past climatic conditions and as well as on the anthropogenic processes on environments (agriculture, fires, etc.). In this way, phytoliths offer new perspectives for those environments where other biological indicators are not preserved.

A history of fires in Dogon country

Fire information from laminated deposits

14Fire signal analysis has to be preceded by precise chrono-stratigraphic and micro-morphological observations. It has been possible to proceed to high-resolution analysis of laminated deposits through different (complementary) methods of identification and quantification of the carbonaceous particles. This double approach is augmented with sieving data (granularity> 100 µm), thin section and pollen analyses.

15The examination of carbonaceous particles from “large format” (6x14 cm) thin section analyses was done under transmitted and reflected light (Scott, 1989; Jones and Chaloner, 1991; Vannière, 2001). The examination of carbonaceous particles during pollen analyses (Faegri and Iversen, 1966) under transmitted and reflected light (Clark and Royall, 1996; Noël, 2001; Vannière, 2001; Sebag, 2002) enable us to study the micro-charcoals under 100 µm and to determine reflectance. The study of the degrees of reflectance can define the frequency and intensity of fires with preciprecision: the particles with medium reflectance indicate medium intensity fires and have better floating capacity, while the particles with higher reflectance have been subjected to higher temperatures. Carbonaceous particles larger than 100 µm (Tinner et al., 1998; Vannière, 2001; Whitloch et al., 2007) are examined under a binocular magnifier after sieving and chemical treatment (hydrogen peroxide) per 10 g of sediments. The aim of this research was to reconstruct the history of fires, their mechanisms and their consequences on the ecosystems through the millennia.

A temporal pattern of fires

16A temporal pattern of fires for the last millennia needs to be based on a pattern taking into account the hydro-sedimentary cycles (Fig. 3), i. e. identifying the seasonal functioning of the environments from the fluvio-palustrine laminated deposits (Le Drézen et al., 2006; Le Drézen, 2008; Le Drézen et al., 2010). The annual recorded fire regime has three phases: (i) transportation during the wet season of the coarse sediments (mineral and organic) from the catchment and deposition in the form of sand-silt laminae; (ii) settling of the thinly bedded sediments and organic material (charcoal, leaves, etc., visible in the silt-sand laminae) from weak and slow flows at the end of the wet and the beginning of the dry season; (iii) possible local, and progressive, drying-up of water bodies and of sediments during the long and intense dry seasons characterised by the formation of polygonal soils and ferruginous areas (superficial oxidation and development of desiccation cracks).

The history of fires in the Dogon country

17At the beginning of the Holocene, the carbonaceous particles are rare, round and with very high reflectance. The first millennia of the Holocene are poor in carbonaceous particles (<100), indicating the rarity of fires and some of the particles almost certainly are derived from re-worked silt deposits of Pleistocene age.

18From the 6^th -5^th millennium BC, the carbonaceous particles increase (>1000/10 g sieved sediments), have an elongated shape and medium reflectance (Fig. 4). Gradually, from the 6^th millennium BC onwards, the fire signal analysis indicates an increase in fires during the transitional period between the wet and the dry season. The large volume and the number (>1000) of charcoal fragments in the deposits of the middle Holocene and the beginning of the late Holocene (Fig. 4) can be attributed to the thickness of the biomass within a vegetation of Sudano-Guinean similarities. However, such early fires could mean that humans are implicated in their appearance. The seasonal pattern established by micromorphology for the late Holocene, combined with the fire signal analyses, helps to provide a picture of the annual records of fires in the Yamé catchment (Fig. 3). Since the 4^th millennium BC, the deposits of the wet season contain an average of 600 carbonaceous particles (per 10 g sieved sediments) larger than 100 µm (Fig. 4). They have rounded and solid shapes with high reflectance. On top of these laminae, during the transitional period between the wet and the dry season, the carbonaceous particles are abundant, of elongated shape, small (40 µm) and with weak to medium reflectance. Gradually, inside the silt-sand lamina (during the dry season), the carbonaceous particles increase (900 to 3000), become smaller, elongated and are all in vertical position. The reflectance is medium to high and some fluctuations of the number of carbonaceous particles (Fig. 3) are perceptible during the two last millennia BC.

19From the middle of the late Holocene (2500 BC), fires are regularly recorded. They swept through a savanna vegetation of Sudanese type, which was pyrophilic, according to palaeobiological studies. Fire signal analysis over an annual cycle shows three phases:

20(i) During the first rains, carbonaceous particles from the catchment vegetation (500 km²) burned at the end of the last dry season, are transferred and deposited. The large size, surface and rounded shape of the observed particles corroborate this theory (Berger, 1996). During the wet season, the particles are transported along with the coarse mineral sediment by the strong Yamé flows. The very high reflectance is partly due to the high degree of combustion, characteristic of dry vegetation. Nevertheless, it also tallies with a reworking of older charcoal fragments during floods. The abundance of wood charcoal and the almost complete absence of inversion of radiocarbon dates for the late Holocene, confirm the regularity and the speed of the “turnover” of the carbonaceous particles in the coarsest laminae as well.

21(ii) At the end of the wet season and at the beginning of the dry season, we have observed a regular and progressive deposition by settling of the carbonaceous particles on the Yamé river banks and in the residual water bodies. These carbonised particles are transported either by fluvial or by aeolian processes (Le Drézen, 2008), as indicated by their elongated shape, their small size and their medium reflectance. The reflectance level can be attributed to low intensity fires in vegetation that was not completely dry. The abundance of the remains of leaves in this layer confirms that it should be attributed to the end of the wet season and the beginning of the dry season.

22(iii) On top of the thin section laminae, there is a rapid deposition of particles from the fires from the middle and the end of the wet season. The particles are abundant, of high reflectance, indicating strong fires that swept through very dry vegetation.

23During the two millennia AD (HT2: 4^th-5^th c. AD; HT4: 11^th-12^th c. AD), many carbonaceous particles have been observed (~1000/10 g sieved sediments) up untill the 15^th c., of very elongated shapes, fairly large size (>150 µm) and with rather homogenous high reflectance. In the sediments from the last centuries (HT6-HT7: 13^th-14^th c.), carbonaceous particles gradually decrease and are absent from the most recent sediments (Fig. 3). Thus, during the terminal Holocene (the last two thousand years), the fire signal remains regular till the 15^th c., with fires that appear every year during the dry season before their dramatic decrease. These fires were sweeping through the Sudano-Sahelian savanna (Fig. 4), in connection with a system of farming activities (baobab, Shea, Nitta Tree), which managed to remain locally untill the 20^th c. However, the fire signal diminishes and disappears in the most recent sediments, thus indicating a decline of the fires in the catchment. Nowadays, unlike many other areas in West Africa, there is practically no propagation of savanna fires in the Bandiagara Plateau (Dolidon, 2005). The result of this evolution is the formation of the contemporary landscapes, which are often linked to an intensive agro-sylvo-pastoral system with Faidherbia albida.

Research methods and results: advantages, crosschecking and the contribution of multi-proxies analyses

The marks of 4000 years of aridification on the landscapes

24The increasing aridity of the climate during the Holocene, combined with the prolongation and the distinct intensification of the dry season (Le Drézen, 2008; Le Drézen et al., 2010), can explain the regular fires in a Sudanese type savanna (Fig. 4). The fluctuating numbers of carbonaceous particles in the course of the last 6000 years indicates that fires are less intense during the wettest climate phases, probably in connection with a lower combustibility of the biomass. The principal climate factor, which is rainfall, influences the quantity of the biomass and therefore the material available for combustion, and consequently the expansion of fires. However, the regularity of fires on an annual basis is determined by the activities of human societies. The presence of early fires in the Yamé catchment, in a biomass that was not dry yet, also shows the anthropogenic origin of fires over almost 5000 years. The fairly recent disappearance of the fires from the Yamé catchment can be explained by the poor quantity of the biomass, which does not favour a regular fires and by important socioeconomic and environmental changes, among which are the increase increase in population density and the intensification of agriculture during the last centuries (Le Drézen, 2008). Furthermore, the number of vegetational species, trees or others, hygrophilous and of Sudano-Guinean character, clearly decreases during the last 4000 years confirming the aridification of the landscapes.

Anthropogenically altered landscapes

Fire: Affecting the savanna landscapes for more than 5000 years

25The recording of regular, recurrent fires in the seasonal deposits reveals an annual frequency of vegetation fires over more than 5000 years in the savannas of the Bandiagara Plateau and this provides information on the fire regime. The fact that fires are detected from the beginning to the end of the dry season, and even in the beginning of the wet season, is important for their interpretation. It means that there are different fires during the same seasonal cycle. If the fires of the dry season in the Sudanese savanna are due to natural causes, their annual frequency is, on the contrary, of human origin. The early fires are clearly recognisable at the end of the wet season and at the very beginning of the dry season and indicate an even more likely human origin. Fire could have been used very early by human as a means to control the vegetation for many reasons such as hunting, protection, clearances, maintenance of the savanna parklands, and opening of the riverside vegetation. The identification of millet domestication in Ounjougou during this period (Ozainne et al., 2009) helps to connect this fire regime with the emergence of Neolithic farming systems based on extensive agriculture (Fig. 5).

4000 years of diversified farming systems

26The comparison of different palaeobotanical analyses (phytoliths, palynology, palynofacies, charcoal analysis, etc.) gives a clear picture of landscape evolution during the last four thousand years (Le Drézen, 2008). At the beginning of the early Holocene, the dominating taxa are similar to the Sudano-guinean taxa (Lophira, Parinari, Uapaca, Alchornea or Syzygium). The Yamé valley landscapes create a patchwork (Fig. 4) that combines an open savanna and various forest types (woodland or dry forest) with gallery forests near the wetter zones. Around 4000 years ago the vegetation was thick savanna (Fig. 5) with Sudano-guinean features (Lophira, Parinari) and forest galleries (Uapaca, Alchornea, Syzygium). Phytoliths indicate an opening of the landscapes with a large increase of the Poaceae in favour of more and more open formations of the Sudanese type. Towards 1110 BC, in the course of a short humid phase (HR2B), the vegetational landscapes display a slight tendency to close-in again (Fig. 5). Indeed, phytolith, pollen and fire signal data all indicate an increase in tree species and a decrease in the carbonaceous particles. The end of the second sequence of the late Holocene (1800-400 BC) is decisive for the transformation of the savanna. The Sudano-Guinean species (such as the Aracaceae) disappear while the Sudanese species thrive along with the Sudano-Sahelian ones (Cyperaceae, Commelinaceae). The species that need wetter conditions decrease and concentrate more and more on the pond banks inside a gallery forest that is still taxonomically rich. As the ponds dry up every dry season, certain species such as the Uapaca, disappear.

Intensive agriculture since the 14^th c.

27For 700 years, the Yamé valley has been under constant and intense influence of human societies amidst a very unstable climate (circumstantial periods of strong flow causing major erosive phases). The erosion is so strong that the soil is removed from the sandstone interfluves and the valley is saturated with colluvium. The climate can explain this evolution, but only partly. Such instability should also be attributed to Man. The system of fallows, which are regularly set on fire every year, kept on modelling the environment. Certainly, the agricultural spaces cover larger and larger areas in the Yamé catchment, as the use of iron tools improves the clearance methods and increases the cultivated surfaces. Despite their great efficacy and their many beneficial effects, fires have probably much deteriorated the vegetation (disappearance of the Bambusoidae) and the pedological cover (development of bad lands) in sparsely populated areas such as the environment of the pre-Dogon and Dogon populations. The intensification of agriculture and the decrease of fallowing periods in combination with fires have undoubtedly favoured the formation of torrential floods because they accentuated the influence of hydro-climatic variability, as observations have proved particularly over these last years.

A history of pyrophytes in West Africa

28A remarkable proportion of pyrophytes, such as the families of Combretaceae and Sapotaceae (Vitellaria), Parkia, Pterocarpus, Prosopis and even the Alchornea (on the margins of the forest gallery), are detectable during the entire late Holocene. The impressive quantity of charcoal and burned phytoliths found in the sediments of the last millennia BC allows for a precise fire signal analysis. A better understanding of the savanna landscapes is thus possible as well as of the role played by natural and human forcing factors. For a long time, the history of pyrophytes in West Africa has been based on speculation. Henceforward, it is possible to demonstrate the role of human societies in the shaping and evolution of the savanna landscapes for over 5000 years, mainly through the analysis of the bioindicators from palaeofires. Fires occurred throughout the Holocene but they were rare during the first millennia. Towards the end of the middle Holocene, fires appeared in the humid zones (riparian forests, bamboo groves, etc.) and increased untill the end of late Holocene, 2000 years ago. The role of human societies in these early fires can be shown due to the wet biomass, which prevented the out-break and propagation of fires untill the middle of the late Holocene (3000 BC). The role of human societies has been demonstrated even more precisely for the end of the late Holocene through the examination of seasonal sediments: they indicate fires as early as the end of the wet season, precisely at a period when human intervention is essential for them to be triggered. These fires start really early in the history of the farming populations in this region and thus can have more than one initiators: nomad shepherds wishing to have access to water from the end of the middle Holocene; farmers clearancing farming plots or the enrichment of soils at the end of the late Holocene. The history of fires in West Africa during the last 5000 years informs us about: (i) the role of societies in landscape evolution and (ii) the fact that the key-period for studying the contemporary landscapes and their creation is the Late Holocene, i.e. the last 3000 years BC.

29During the final Holocene, fire frequency gradually decreased in the Dogon country and they disappeared towards the 17^th c. This can be explained by the progressive increase of the population in the region (one of the most densely populated Sudano-Sahelian regions in West Africa) and the advent of intensive agriculture.

Conclusions

30Geoarchaeological research in the Dogon country during the last few years has clarified the history of the landscapes of this Sudano-Sahelian part of Africa, in particular climate evolution and the history of human societies. The environmental changes recorded in the Yamé valley demonstrate the considerable evolution of the vegetational landscapes during the last five thousand years. From to phytolith studies, we have determined that during the Late Holocene the savanna vegetation comprised many pyrophytes that indicate regular fires. From the fire signal the origin of this vegetation was determined by means of the quantity of carbonaceous particles, their morphology and the circumstances of their deposition. Thus, the Ounjougou site complex offers a sedimentary context suitable for the definition of the fire regime and the biogeographical changes, based on the fluvio-palustrine sediments (rhythmites) from the late Holocene. For more than 5000 years, the Yamé catchment has been subject to widespread fires on a quasi-annual basis. This regularity has decreased only during the last centuries. The intensified seasonality of the hydrological activity during the late Holocene, with the increase in length and intensity of the dry season, is characteristic of a palaeoclimate favourable to vegetation fires. However, the annual return of fires around the Ounjougou site complex, in particular at the beginning of the dry season, strongly implicates humans as the principal cause of these fires. At the same time, human societies played an increasingly dominant role, wishing to control their environment in various ways: opening the vegetation along the watercourses, maintaining the savanna-parklands, growing millet, etc.

31The evolution of geoarchaeological field and laboratory research has enriched our knowledge of the history of landscapes, societies and climates in a region too often neglected because of the poor preservation of archaeological and organic remains. Within the geoarchaeological approach, the cross-comparison of the results from individual techniques permits a subtle analysis of environmental evolution during the Holocene and a better understanding of the role played by the ancient societies in the modelling of the landscapes of the West African savanna, as characterised by the progressive appearance of complex agrosylvo-pastoral systems.

Acknowledgements

32We wish to sincerely thank the “Stand and paleoenvironment in West Africa” programme manager, Eric Huysecom, and various French, German and Malian multidisciplinary colleagues. We also wish to thank the villagers of Dogon Country (Dimbal, Gologou) and financial partners in these programmes (ANR-DFG, CNRS, FNRS...) as well as our respective research laboratories and universities. Finally, we wish to sincerely thank the anonymous reviewers of this article.