A multi-proxy research on human-environment interactions between 8000 and 4400 years BP in Limburg, the Netherlands

A multi-proxy research on human-environment interactions

between 8000 and 4400 years BP in Limburg, the Netherlands

Waas Thissen

University of Amsterdam

Future Planet Studies

Bachelor thesis

2017

Supervised by dr. C.N.H. McMichael & dr. W.D. Gosling

Abstract

A great part of paleoecological research aims to understand the interactions between humans and the landscape after the first transition to farming in the so-called neolithic revolution. Research has

pointed out that from then on, humans increasingly influenced the vegetation composition, the climate, and fluvial systems across the globe. Humans affect their environment mainly through fire,

deforestation, propagation of beneficial species and the introduction of non-native crops and animals. However, regionally and even locally, human impact on the landscape can differ largely. This

multiproxy research aims to provide insight into how humans interacted with their environment in Limburg, the Netherlands, between 8000 and 4400 BP. Many archeological findings indicate a long period of human habitation in this region. This research builds on an existing paleoecological reconstruction of human-environment interactions where the composition of pollen, phytoliths, non-pollen palynomorphs, and charcoal were analysed. The aim is to extend this multiproxy dataset with additional proxies: macrofossils, loss on ignition (LOI) and particle size. Adding macrofossils can provide insight into the local vegetation surrounding the site. LOI and particle size can provide insight into hydrological influences. Moreover, temperature and precipitation data is added to infer the role of climate as a possible driver of important processes. By combining multiple proxies with a high temporal resolution and from diverse spatial scales with archeological findings, it can be possible to obtain a coherent view of human-environment interactions. The research has pointed out that over time humans caused regional and local deforestation, mainly visible by declines in Quercus pollen and synchronous declines in phytoliths and macrofossils. However, humans possibly also positively influenced the growth of Alnus and Corylus by maintaining hedges bordering fields and by woodland grazing. Moreover, it has shown that deforestation did not cause flooding at least not during the whole research period. Arguably, only in the late Neolithic in the most upper sequence deforestation would have this effect. Lastly, phytoliths, providing a local signal, could possibly be washed in during periods of high river flow, causing a distortion of the local vegetation signal.

Introduction

Human-environment interactions in the holocene

For most of human existence, our environment impacted us while our impact on the environment was minimal. This changed during the Holocene (11.700 BP - present) (Walker et al., 2009), a period of warming out of an ice-age, which gave way to the global transition to agriculture and accompanied population growth (Ruddiman, 2005). In the early Holocene, climate was the main driver of fire events (Zolitschka, Behre, & Schneider, 2003) as well as changes in vegetation composition (Huntley & Birks, 1983). But ever since the establishment of the first agricultural societies in the mid-Holocene

(the so called Neolithic Revolution), humans have started to transform Europe's landscapes (Kaplan,

Krumhardt, & Zimmermann, 2009; Kalis, Merkt, & Wunderlich, 2003). While climate remained relatively stable, humans became the foremost drivers of environmental change (Kalis, Merkt, & Wunderlich, 2003) primarily through deforestation (Kaplan et al., 2009). Humans became the foremost drivers of fire (Zolitschka et al., 2003). Humans increasingly affected the vegetation composition by deforestation (first with fire, later with tools (Ruddiman, 2005)), cultivation of beneficial species, introduction of ruderal species, and fragmentation of the landscape (Behre, 1988; Delacourt, 1987). Humans increasingly affected fluvial systems by deforestation causing higher peak discharges (Kalis, Merkt & Wunderlich, 2003; Toonen, 2013). Europeans possibly even started to change the climate when they started deforestation and crop cultivation around 8000 BP, which is earlier than commonly thought (Ruddiman, 2005; Kaplan et al., 2011).

Paleo-reconstructions that yielded us the aforementioned knowledge cannot be accomplished by direct measurement. Instead, proxies are used to infer the status or rate of a variable of interest. In case of plant proxies, this is not always straightforward, as a proxy present for one species is sometimes not present for another species. This can be because the plant does not produces the proxy or does the proxy is not well preserved (Piperno, 2006; Birks, 2007). Hence, it is helpful to perform a multiproxy analysis, so that gaps of knowledge left by one proxy are filled by others (Birks, 2007).

Proxies for regional, extra-local and local vegetation

Several proxies are found to represent different vegetation at different spatial scales. Janssen, (1966) divided pollen, a vegetation proxy, to derive from a local (in the range of several meters), extralocal (in the range of 100 meters) or regional (in the range of kilometers) area. Jacobsen & Bradshaw, (1981) set more specific definitions: local, extralocal and regional were within 20 m of the edge of the sampling basin, between 20 m of the edge and several hundred meters, and more than a hundred meters of the edge respectively (see figure 1). The latter definitions will be used in this research. Regional vegetation

Tree pollen are mostly wind-dispersed and travel over long distances, and can thus provide insight into the regional vegetation composition (Jacobson & Bradshaw, 1981). However, tree pollen are also local, from trees bordering the sampling basin, and extralocal, deriving from pollen blown between the trunk space between trees (sensu Tauber, 1965) (Janssen, 1966). Especially in closed-canopy forests containing small sites, most pollen is argued to be derived from a source area that is within a 20 - 30 meter range (Jacobson & Bradshaw, 1981). In this case, the regional signal is very small (see figure 1). As a prerequisite, local pollen, and to an extend extralocal pollen, ought to be excluded from the pollen sum when aiming to construct a regional signal (Janssen, 1966). In practice, this means that very high abundances of pollen should be considered to be excluded from the pollen sum as they are likely to be local or extralocal. The remaining signal is then regional and can be seen as a

Figure 1. This figure is adapted from Janssen (1966) with additions of Jacobson & Bradshaw, (1981). It depicts an idealized relation between the abundance of pollen (absolute, not relative) and the distance of the sample

location from the pollen source. It also shows that with small sites and/or in dense forests the regional signal is expected to shrink causing the local signal to be relatively stronger (lower line) Local vegetation To provide more insight into the local vegetation, phytoliths and/or macrofossils can be studied. Phytoliths are parts of silica in plant remains, and can be morphologically determined to fit certain plant species (Piperno, 2006). Macrofossils are plant remains (e.g. seeds or leaf fragments) that have been preserved well due to an environment low in oxygen (Birks, 2007). Both phytoliths and

macrofossils are heavy material relatively to pollen, and are found close to the source of dispersion given there is an environment with relatively few transportation pathways via wind or water (Piperno, 2006; Birks, 2007). Both then represent local and to an extend extralocal vegetation (Piperno, 2006; Birks, 2007).

Proxies for human activity

Obvious indicators of human activity are archeological findings. Mainly excavations of sites where people lived provide immediate information about the all aspects of daily life (Kalis, Merkt, &

Wunderlich, 2003). However, such archeological findings are fragmented spatially and temporally with limited possibility for interpolation. Paleoecological proxies can then add valuable information as they often have a higher temporal resolution and can represent larger spatial scales (Kalis, Merkt & Wunderlich, 2003). A good example is pollen, which production is continuous, and as it can represent regional vegetation.

Fire

Charcoal is a proxy that represents fire in the landscape (Scott, 2010). Most of this charcoal is produced by wildfires (e.g. Kuhlbusch & Crutzen, 1996; cited in Scott, 2010). However, climate data hints that in some regions, people increasingly become drivers of fire in the late Holocene (Power et al., 2008).Micro-charcoal (<180 µm) can be transported by wind and can thus shows a regional signal where wind is a possible transportation pathway (e.g. Smith et al., 1973; cited in Scott, 2010). Macro-charcoal (>180 µm) mainly shows a local signal unless transport via water is possible; as Macro-charcoal floats well, macro-charcoal then shows a regional signal (Nichols et al., 200; cited in Scott, 2010). Herbivores

Non-pollen palynomorphs (NPPs) are all micro-remains that are non-pollen. Among these are the spores of coprophilous fungal species. Abundance of these species can be used as a proxy for abundance of herbivores (van Geel, 2002; Baker, Bhagwat, & Willis, 2013). Research has found that of six researched types the Podospora-, Sordaria- and Sporormiella-type (sensu van Geel, 2002) have the most power in estimating herbivore abundance (Baker et al., 2013). As fungal spores are discharged up to a distance of few single meters (Ingold, 1971; Yafetto et al., 2008) they show local abundance of herbivores, which research has also indicated (van Geel, 2002). Most coprophilous fungi do not discriminate between herbivore species (i.e. they are generalists) so that it cannot be

distinguished of which herbivore species the abundance is inferred (Bell, 2005). Moreover, the question remains whether these herbivores would have been wild or domesticated ( Ralska-Jasiewiczowa & van Geel, 1992). In some studies, large increases in these spores have been ascribed to domestic animals in combination with archaeobotanical proxies, archeological findings and sometimes historical information (Gauthier et al., 2010; van Geel et al., 2003; López-Sáez & López-Merino, 2007).

Agricultural and agriculture-related crop species

Findings of pollen of crop- and related weed species are an indication that humans begin to adopt agriculture by introducing these antrochopores (sensu Behre, 1988) in the landscape (Kalis, Merkt & Wunderlich, 2003). In Neolithic Northern-Europe, major staple crops were mainly wheats (Triticum species), barley (Hordeum vulgare), legumes (Fabaceae species), and flax (Linum usitatissimum) (Price, 2000). Most of these pollen species are self-pollinating and so do not emit pollen into the atmosphere (Kalis, Merkt, & Wunderlich, 2003) but these pollen are released by people when harvesting the grains (Kalis, Merkt, & Wunderlich, 2003). Moreover, weed species related with agriculture such as Ribwort plantain (Plantago lanceolata) can be found, also indicating fallow agricultural fields (Behre, 1988).

Arboreal / Non arboreal ratios

Lastly, the ratio between arboreal pollen (AP) and non-arboreal pollen (NAP) is a proxy that indicates the amount of openness (forest or a park landscape) in the landscape (Kalis, Merkt & Wunderlich, 2003). Also the ratio between arboreal phytoliths (APh) and non-arboreal phytoliths (NAPh) can be constructed to detect local openness. Openness can be a sign of grazers, opening forest, or people clearing woodland for creating new cropland or pasture, or harvesting of building material (Kalis, Merkt, & Wunderlich, 2003).

Combining proxy data

By combining data on the aforementioned proxies, a coherent picture on the impact of humans on the landscape arises. For example Vink, (2016) finds a synchronous increases in cereal pollen, spores of coprophilous fungi, fire frequency, and a decrease in tree phytoliths. This was interpreted as an extensification of agriculture, where people were burning woodland to create new cropland and pasture (Vink, 2016).

Proxies for sediment properties

With loss on ignition (LOI) contents of water, organic and inorganic carbon, and silica can be measured (Heiri et al., 2001; Santisteban et al., 2004). Changes in these contents reflect environmental change (e.g. changes in biomass production, changes in sedimentation). Organic carbon could provide insight into the amount of vegetation, and silica could give insight into the amount of mineral material deposited at a certain time. Particle size distribution measures the amount of particles in their different sizes and changes in these amounts also reflect environmental change. High energy environments will appear in core samples as coarse grained samples, whilst low energy environments will appear as fine-grained samples.Together, particle size and LOI are proxies useful for e.g. reconstructing paelofloods (Gilli, Glur & Wirth, 2013; Toonen 2013). These flood events can be related to human activity. For example, humans can enhance erosion by deforestation. This could lead to an increased peak flow causing more severe flooding (Hollis, 1979). Also, flooding could force humans to relocate to dryer areas (Zuidhoff & Bos, 2017).

Research aim & objectives

This research will ask: what are the human-environment interactions between 8000 and 4400 BP in the region of Well-Aijen, Limburg, the Netherlands? The timeslot for this research was chosen in order to capture the whole range of development of human activity; at 8000 BP, early mesolithic hunter-gatherers were already present in the landscape as many archeological findings have pointed out (Louwe Kooijmans, van den Broeke, Fokkens & van Gijn, 2005;Tichelman, 2005; cited in Bos & Zuidhoff, 2012). 4400 BP is the upper limit of the sequence.

To achieve the goal of this research, an existing multi-proxy research from 2016 is extended with additional proxies (see figure 2) . New samples are taken from the existing soil sequence and

examined for three proxies; macrofossils for an indication of the local vegetation, and LOI and particle size for an indication of environmental changes. Also, proxy data that reflects air temperature and precipitation is obtained from literature. This is ratio of oxygen-16 and oxygen-18 isotopes (δ18O) from significant deposition sources. To what extend these proxies will provide this information was

unknown prior to research. Thus, a subquestion of this research is the following: to what extend do the additional proxies explain the interactions better? (see figure 2).

Figure 2. Human-environment

interactions. What are they? And to what extend do the additional proxies explain the

interactions better?

Relevance

Palaeoecological studies are ultimately ‘learning from the past’ that can lead to the formation of strategies for the future. Findings from these type of studies include pre-impact states, trajectories of recent change, causation, and more (Dearing et al., 2006). The potential outcomes of this particular research are high. The site can be regarded as a relatively good site for paleo-studies because it was deep, and relatively unaffiliated to other water bodies. Thus only ‘extreme’ events (high velocity floods) are likely to activate inflow of river sediment. These factors can be seen as requirements for the reconstruction of paleofloods (Gilli, Glur & Wirth, 2013) but are also applicable to the

reconstruction of a vegetation history.

Another strength of this research is also the combination of archeological findings and paleoecological data. Archeological data and paleoecological data are very complementary (Brown, 1997).

Archaeological findings provide a lot of very detailed information about humans, but they can be regarded as ‘point features’ in space and time; they arerather site specific (spatially fragmented) and not very continuous in time (temporally fragmented). On the other hand, using paleoecological proxies allows to investigate human-environment interactions on a larger spatial scale (Kalis, Merkt, &

Wunderlich, 2003) and more continuous through time. A proxy that best demonstrates this is pollen, which can provide a regional vegetation signal as well as a continuous signal through time. Lastly, the addition of proxies to the existing dataset was expected to lead to closure of gaps in this dataset. This research is the last possibility for reconstructing human-environment interactions on this location. Sadly, the Well-Aijen site has been destroyed because of an ecological and recreational undertaking in the Meuse floodplain.

Hypotheses

Three hypotheses were set up in order to test questions that remained unclear from former research. The hypotheses were set up so they could be tested. They were accompanied by a visualization of the correlations tested in the hypothesis and a conceptual model aiming to clarify the potential outcomes of the hypotheses.

Hypothesis I - humans are clearing the landscape on a local and a regional scale

In the 2016 research, Vink (2016), found that changes in tree pollen and three phytoliths occurred at the same time. The change was that tree cover declined locally as well as regionally, indicating deforestation (Vink, 2016). This was interpreted as a sign that agriculture was being practiced on a regional and local scale. One could argue that in order to construct an overview of the local

vegetation, if would suffice to study phytoliths. However, some European tree species, like Alnus do not produce phytoliths (McMichael, personal communication). Examining macrofossils of non-phytolith producing trees is to current knowledge the only method for estimating a local vegetation trend of these trees. However, Bos & Zuidhoff, (2015) assume Alnus pollen at the Well-Aijen site to be mostly derived from local vegetation instead of from regional vegetation due to its high presence in a

standard pollen count, and because of the geographical location of this research area. Indeed, in floodplain environments Alnus is often dominant, frequently reaching 90% of the total land pollen sum (Brown, 1997). This implies it could be necessary to exclude Alnus from the pollen sum in order to get a ‘clean’ regional signal. To know whether Alnus, Corylus and other species were local or not, it is worthwhile examining macrofossils to obtain an overview of the local presence of these species. Translating this into a null- and alternative hypothesis (see figure 3):

H0: There is no correlation between pollen abundance and macrofossil abundance of the same taxa. HA: There is a positive correlation between pollen abundance and macrofossil abundance of the same taxa.

Accepting H0 would mean that macrofossils do not follow the pollen signal, indicating a discrepancy between local and regional vegetation. Rejecting H0 in favor of HA would mean that macrofossils follow the pollen signal, indicating that local vegetation matches regional vegetation (see figure 3).

Figure 3. The null- and alternative hypothesis of hypothesis I, visualized as correlations between proxies (left) and as a conceptual model displaying signals of the proxies (right).

Hypothesis II - human clearing of the landscape causes flooding

It is expected that further into the holocene, human activity will become more and more a driver of LOI and particle size. For example, humans could influence the amount of vegetation in the direct

proximity, influencing the amount of carbon (LOI550) deposited in the abandoned channel. Moreover,

increasing deforestation and agriculture leads to erosion, bringing more sediment in the river (Ruddiman, 2005; Kalis, Merkt & Wunderlich, 2003) and increased peak flow resulting in floods (Hollis, 1979). Kalis, Merkt & Wunderlich, (2003) found that Early Neolithic humans had a neglectable impact on colluvia, floodplain deposits or lake sediments in Germany, but that from 6300 BP, humans did have a large impact on these sediments. Likely, these flood events can be retraced as increases in the sand fraction; with high flow velocities, what usually moves as bed-load during normal flow becomes suspended and can be trapped in depressions as water inundates the floodplain (Toonen et al., 2013). Likely these events will occur relatively similarly, or with a neglectable lag time, after humans clear the landscape, visible in a decline in tree pollen and an increase in charcoal. Another possibility is that precipitation, instead of humans appears the main driver of flood events throughout the researched time period. Increases in precipitation could obviously also lead to higher peak flow. Increases in the δ18O isotope can be seen as increases in precipitation. When the δ18O isotope concentrations in northern spain are low, this indicates higher rainfall totals in this area, as was measured in modern times (Smith et al., 2016; cited in Smith et al., 2016). These wetter conditions are present in times of a negative North Atlantic Oscillation (NOA) (Hurrell & Deser, 2009; cited in Smith et al., 2016). Oppositely, in times of a positive NOA, winter storms are forced to north-west Europe, causing wetter conditions here (Morley, Rosenthal, & DeMenocal, 2014; cited in Smith et al., 2016). It can thus be conlcuded that when north-west Europe receives more precipitation, this is reflected in an increasing δ18O isotope concentration in northern Spain.

Translating this into a null- and alternative hypotheses (see figure 4):

H0: the sand fraction does not correlate with charcoal, tree pollen, and precipitation.

HA: the sand fraction correlates positively with charcoal and negatively with tree pollen and not with precipitation.

HA2: the sand fraction correlates positively with precipitation and not with charcoal and tree pollen. Not rejecting H0 would indicate that neither humans or precipitation affect hydrological variability and so LOI and particle size would not follow the precipitation or human signal. Rejecting H0 in favor of HA would indicate that humans (when present in the landscape) are the main drivers of hydrological variability and so LOI and particle size follow a human signal. Rejecting H0 in favor of HA2 would indicate that precipitation is the main driver of hydrologic variability (seen in LOI and particle size) are mainly driven by precipitation (see figure 4).

Hypothesis III - microfossils can be washed in by flood events

As stated previously, phytoliths can provide an overview of the (extra)local vegetation and pollen of the regional vegetation. However, a prerequisite is that minimal transportation pathways are present (Piperno, 2006; Birks, 2007). Else, the scale of the source area from which microfossils were derived could be misinterpreted. Although the Well-Aijen site is cut-off from most hydrological features, it is known from prior research that flooding events occurred in this region (Bos & Zuidhoff, 2015). As the sand fraction is an indication for flood events (Toonen, 2013) it can be compared to the composition microfossils as an indication that floods might transport microfossils and deposit them in the

abandoned channel. Finding very different species abruptly (i.e. changes in the DCA scores) indicates changes in the composition of microfossils, possibly caused by transport by floods (peaks in sand). Translating this into a null- and alternative hypothesis (see figure 5):

H0: the sand fraction does not correlate with the microfossil composition (i.e. the DCA scores of pollen and phytoliths).

HA: the sand fraction has a positive correlation with the microfossil composition (i.e. the DCA scores of pollen and phytoliths).

Not rejecting H0 would indicate that a local and regional vegetation signal is not heavily influenced by the inwash of microfossils. Rejecting H0 in favor of HA would indicate that the river is washing in microfossils, possibly affecting the local and regional vegetation signal (see figure 5). These vegetation signals then have to be interpreted with caution.

Figure 5.The null- and alternative hypothesis of hypothesis III, visualized as correlations between proxies (left) and as a conceptual model displaying signals of the

Site description

Previous research

In 2011, a first core was taken from an abandoned river channel of the Meuse river, examining pollen, NPPs, charcoal, macrofossils and grain size (Bos & Zuidhoff, 2015). Findings were chronologically contextualized with palynological- and 14C dating. The abandoned river channel was located in the Meuse valley between the villages of Well and Aijen in the north of the province of Limburg, the Netherlands (Bos & Zuidhoff, 2015). The abandoned channel was created in the early Holocene (ca. 11,700 BP) as climate warmed, vegetation increased, and soils were stabilized (Zuidhoff & Bos, 2017). The braided channels of the Meuse river first deepened but were eventually abandoned as the Meuse concentrated in one channel (Zuidhoff & Bos, 2017). Subsequently, the abandoned river channels were filled with clastic and organic material (Bos & Zuidhoff, 2015). The core covered a large part of the holocene, approximately 8000 years (Bos & Zuidhoff, 2015) and contained material ranging from the Boreal (starting 9000 - 8000 BP) to the Subatlanticum (2500 BP - present)

(Mangerud et al., 1974; Janssen & Törnqvist, 1991). It thus contained the archeological periods of the (Dutch) Mesolithic (ca. 10800 - 6900 BP) to the beginning of Roman times (2000 BP) (Bos & Zuidhoff, 2015; Louwe Kooijmans et al., 2005). Moreover, the core and contained no hiatus proving excellent material for paleoecological research (Bos & Zuidhoff, 2015). In addition to this information, many archeological findings are situated near the Well-Aijen site (Louwe Kooijmans et al., 2005; Bos & Zuidhoff, 2015; Verhart, 2016). In 2016, a new research was executed at this site (Vink, 2016; van Teulingen, 2016; de Zwaan, McMichael & Gosling, 2016). Eight monolith tins with sediment (in total about 3 meters of sediment) were recovered from a slightly different location in the abandoned river channel (51°33'42.5"N

6°02'50.2"E) only about 10 meters away from the 2011 research (Bos, personal communication) (see figure 6). In this research, pollen, phytoliths, NPPs and charcoal were examined at an average resolution of approximately 50 years and contextualized with 7 14C samples (Vink, 2016; van Teulingen, 2016; de Zwaan, McMichael & Gosling, 2016). This broader analysis with the addition of phytoliths allowed for a reconstruction of the local vegetation. Also, more 14C samples allowed to construct a better

chronology. However, this core contains a gap in the sediment between 265.5 cm and 298 cm.

Figure 6. Site location (red star) of the 2016 research. Adapted from ADC archeoprojecten. Copyright 2015 by ADC archeoprojecten.

The agricultural history of Limburg and Well-Aijen

Archeological findings surrounding Well-Aijen have indicated that the site was already inhabited in the Mesolithic (Tichelman, 2005; cited in Bos, Zuidhoff,

van Kappel & Gerrets, 2012; Kimenai & Mooren; cited in Verhart, 2016) as well as the rest of northern Limburg (Verhart, 2016). Camps of these mesolithic people were small indicating they were highly mobile communities (e.g. Verhart, 2000; cited in Verhart, 2016). It is known that farming arrived in the loess regions of southern Limburg at ca 7300 BP where people created sedentary settlements (Louwe Kooijmans et al., 2005). These first farmers were part of the linearbandkeramik (LBK), a homogenous culture that spread fast across Europe’s loess regions (Louwe Kooijmans et al., 2005; Behre, 1988). The culture’s name derives from its distinctive pottery (Louwe Kooijmans et al., 2005). From research across many European sites it is known that the LBK’s most important crop was Emmer (Triticum diccocum) (Willerding, 1980; cited in Behre, 1988). It is also known that animal husbandry was practiced, and that cattle, sheep, goats, and pigs were left to graze in the woodland (Behre, 1988; Uerpmann, unpublished data; cited in Kalis, Merkt, & Wunderlich, 2003). Isolated artefacts in the northern regions of Limburg (near Well-Aijen) indicate that LBK people possibly used these grounds for hunting and herding of cattle (Louwe Kooijmans et al., 2005). These early farmers lived alongside mesolithic hunter-gatherers for a long period, but besides some possible trading of goods (Louwe Kooijmans et al., 2005; Luning et al., 1989; cited in Kalis, Merkt & Wunderlich; Verhart, 2016) very little interaction occurred (Louwe Kooijmans et al., 2005; Behre, 1988). Neolithisation (sensu Louwe Kooijmans et al., 2005) by mesolithic people was a very slow process (Behre, 1988) and research indicates that it took between 1000 (Out, 2008) and 1500 years (Price, 2000) before all aspects of agriculture expanded across it’s loess border. Around 6900 BP the LBK disappears from Limburg, marking the period of the Rössenculture, another agricultural society (Verhart, 2016). Archaeological research indicates that neolithisation in north-limburg occurred around 6400 BP (Verhart, 2016). Research from Germany also indicates increased human impact on the landscape around 6300 BP whereas before the impact was limited (Kalis, Merkt, & Wunderlich, 2003).

Methods

Chronology

Seven 14C samples were collected from bulk sediment for the establishment of an age-depth model. Raw data on the seven 14C dates (the uncalibrated radiocarbon dates BP, the one standard deviation error, the sample depth, and the corresponding field code) is provided in table 1 (appendix). The samples were taken at locations where shifts in telling proxies occurred. For example, one sample was taken at the first location where pollen of cereal appear, and another one where the abundance of cereal pollen sharply increase. Samples were also placed surrounding the gap in the sediment between 265.5 cm and 298 cm. The age-depth model was created using Bacon age-depth modelling (Blaauw & Christen, 2013)an open-source software package that runs with R (R core team, 2013). The model uses Bayesian statistics to reconstruct Bayesian accumulation histories based on radiocarbon dates or other types of dating (Blaauw & Christen, 2013). Prior information on the

accumulation rate has to be assumed for the whole model (Blaauw & Christen, 2013). The model was run using standard settings (acc.shape = 1.5, acc.mean = 20, mem.strength = 4, mem.mean = 0.7, res=5). More information on these settings can be found in the open manual of the Bacon model (Blaauw & Christen, 2013).The 14C dates were calibrated in Bacon with a modern calibration curve (IntCal13) to correct for differences in the abundance of the 14C isotope in the past. These corrected dates are referred to as cal yr BP (calibrated years before present). From the best-fit line, calibrated ages were extracted for every sample depth to obtain a chronology for the whole sequence.

Macrofossil determination

30 samples with an average volume of 18 cm3 were taken from the core. The resolution was approximately one sample at every 5 - 10 cm. No macrofossil samples were taken between the depths of 265.5 cm and 298 cm, as no sediment was present in this part of the monolith tin. The macrofossils were sampled at approximately evenly spaced distances. The samples were prepared in accordance with Mauquoy & van Geel, (2007). After cutting of the samples, the samples were put into a beaker with water and the volumetric displacement was used to determine the volume of the sample. The samples were soaked and heated in a 5% KOH solution for deflocculation or the organic material. The organic solids were sieved out with a 160 µm sieve and put to float in a petri dish with distilled water for examination. A stereomicroscope (magnification times 10 - 50) was used for examining large remains and a higher magnification microscope (magnification times 100 - 400) was used for examining small remains. Macrofossils were determined with catalogues and help of

macrofossil-expert Bas van Geel. The catalogues used were the Digital seed atlas of the Netherlands

(Vol. 4) (Cappers, Bekker, & Jans, 2012) and the Bestimmungsschlussel fur subfossile Gramineen-Fruchte (Korber-Grohe, 1991). Grouping of macrofossils into ecological groups and the interpretation of their presence was done with the Standaardlijst van de Nederlandse Flora 2003 (Tamis et al., 2004). Macrofossils were not always identified to their lowest taxonomic rank, rather, for most fossils an indication of the family or genus suffices regarding time efficiency (van Geel, personal

communication). Whether macrofossils were indicated as volume percentages or in absolute numbers depended on the type and abundance of macrofossils that were be found (cf. Mauquoy & van Geel, 2007). Species abundance was often calculated as number of species in a sediment volume of 15 cm3. Leaf fragments were estimated as a percentage of floating leaf area in a petri dish.

Determining loss on ignition

128 1cm3 samples were taken at every location where a microfossil sample was obtained from the sequence in the 2016 research. The resolution was approximately one sample at every 2 cm. No LOI samples were taken between the depths of 265.5 cm and 298 cm, as no sediment was present in this part of the monolith tin. LOI was be conducted via guidelines similar of Davies (1974) with addition of the improvements of the method by Heiri et al., (2001). First, platinum crucibles were weighed. Second, the samples were added and the crucible + sample were weighed again. Third, the crucibles were put in the oven for 24 hours at 105 °C. Fourth, the crucibles were left to cool in an exicator and weighed again. Fifth, the samples were put in the oven for 16 hours at 550 °C. Sixt, the crucibles were left to cool in an exicator and weighed again. Seventh, the samples were put in the oven for 3 hours at 950 °C. Eighth, the crucibles were left to cool in an exicator and weighed again. All weights after consecutive burning were then calculated as a percentage of the dry weight (DW105) (i.e. the weight of

sample after 24 hours at 105 °C). LOI550 was calculated as follows: LOI550=((DW105

-DW550)/DW105)*100. LOI950 was calculated as follows: LOI950=((DW550-DW950)/DW105)*100. The silica

content was calculated as follows: (DW950/DW105)*100.

It should be noted that limitations of the LOI method as described by Heiri et al., (2001) and Santisteban et al., (2004) were taken notice of but that not all possibly distorting factors could be taken into account. For example, at 550 °C other reactions than burning of organic carbon influencing weight loss, such as the loss of clay- or metaloxide-bound water, inorganic carbon (carbonates), and loss of volatile salts, already occur (Heiri et al., 2001; Santisteban et al., 2004). Moreover, the organic matter and organic carbon content of samples influences LOI outcomes Heiri et al., 2001; Santisteban et al., 2004).

Particle size determination

45 2 cm3 samples were taken at approximately evenly spaced distances. No particle size samples were taken between the depths of 265.5 cm and 298 cm, as no sediment was present in this part of the monolith tin. The resolution was approximately one sample at every 5 cm. Approximately 1 cm3 of material was prepared when the material was regarded to be relatively low in organic content. For materials that were deemed to have higher organic content (by visual examination), 1.5 to 2 cm3

samples were prepared. As these samples have a lower mineral fraction, more of this material is needed for analysis. The organic-poor samples received 5 ml of 30% hydrogen peroxide. The organic-rich samples received 7 - 10 ml of 30% hydrogen peroxide. Hydrogen peroxides reacts with the organic material and dissolves it. Most of the samples were regarded organic-rich and received 10 ml. The samples were put on a heater to induce the reaction. A small amount of demi-water was added in order to prevent boiling dry, but not too much as this would hamper the effectiveness of the hydrogen peroxide. When the reaction had taken place, visible by the lessening of grey fumes, the material was removed from the heater. The samples were sieved with a 200 micrometer sieve to remove any remaining large organics, and put on the heater to boil out the water that was used for sieving. When it was nearly boiled dry, 5 ml of 30% hydrogen peroxide was added to remove the last organic remains. All samples were topped to 100 ml with demi-water, and left to boil until 50 ml remained. Subsequently 5 ml of 10% hydrochloric acid was added to all samples to remove concretions, and for the mineral material to precipitate. The samples were heated until boiling point and then removed from the heater to react. The samples were then topped to 800 ml with demi-water and left overnight (about 16 hours). Then, the water was suctioned from the samples until 50 ml was left, and 300 mg of sodium pyrophosphate was added to neutralise the negative charge of clay particles. The samples were then analyzed with the Sympatec HELOS KR laser-diffraction particle sizer (VU, 2017). Output graphs are visible right after running the machine, and should approximately follow a gaussian curve. In case a bimodal curve is the output, this was regarded to be an error and the sample was run again. This was on recommendation of the analyst assisting at the VU.

Detrended correspondence analysis

An ordination method named detrended correspondence analysis (DCA) was performed in order to 1) identify two ecological gradients along DCA axes 1 and 2 on which the vegetation community

composition score of all samples are placed and 2) identify changes in the vegetation community composition that are not well visible by eye (the latter is also more prone to errors). DCA’s were performed for all the raw counts of pollen data, phytolith data, and macrofossil data separately. The analysis was also performed on the relative percentage data of pollen and phytoliths combined, and on pollen, phytoliths and macrofossils combined. DCA’s were run in R (R core team, 2013) with the free software packages Vegan (Oksanen et al., 2017). Interpretation of an ecological with help of the specie scores was done with the ‘Vegetation Mitteleuropas mit den Alpen’ (Ellenberg & Leuschner, 2010) and the ‘Standaardlijst van de Nederlandse Flora 2003’ (Tamis et al., 2004).

Visualization of stratigraphic diagrams

All data (the 2016 data and the new data) was plotted in stratigraphic diagrams using the program C2 (Juggins, 2014). Climate data (temperature and precipitation) was also added to one of the

stratigraphic diagrams. These proxies were the oxygen 18/ oxygen 16 isotope ratio in rain (δ18Op) from ostracods (Ostracoda) from a deep lake in southern Germany, used for reconstructing air temperature (von Grafenstein et al., 1999) and abundance of δ18O in speleothems in northern Spain used for reconstructing precipitation (Smit et al., 2016).

Correlations between proxies

Several correlations (pearson’s) were run in R (R Core Team, 2013) between the proxies, as discussed in the hypotheses.In order to correlated proxies measured on a different resolution, only pairwise measurements were used for correlation. Interpolation of data would have lead to data-creation which would have introduced too much error.

Results

Chronology

Figure 7 displays the age-depth model that was created for the sequence. The depth and estimated age of the seven 14C dated samples are located on the blue polygons that also show a spread of one standard deviation error. The red line indicates the best fit. From this line, calibrated ages were extracted for every sample depth. The grey/black area indicates the 95% confidence interval. The graph is relatively linear, indicating a relatively stable sedimentation rate throughout the infilling of the channel. Only between the fifth and sixth (seen from above to below) 14C dates it is visible that sedimentation accelerates compared to the rest of the graph. From 298 to 310 cm depth, the data was extrapolated, thus being more prone to errors than the interpolated data.

Figure 7. On the top of the figure: settings for the Bacon age-depth modelling. Settings were standard settings (see ‘Methods’ for more information). On the bottom of the figure: age-depth model for the Well-Aijen sequence. The seven 14C dated samples are indicated by the blue polygons that indicate the estimated age with a spread of one standard deviation error. The red line indicates the best fit.

Macrofossils

Figure 8 displays the abundance of all species of macrofossils found. An estimation of the percentage of leaf fragment surface cover was also added (first column). Species are grouped and color-coded to their ecological group. These groups are trees (green), water-fauna (pink), helophyte-like plants (light blue), aquatic plants (blue), and understory vegetation (light green). Species which ecological habitat is unknown are depicted white. A general trend that can be seen is that the abundance of all

macrofossils diminishes strongly from 6700 onwards. This decline is possibly due to the infilling of the channel. The channel would become more shallow and would possibly fall dry more often becoming more oxidized and bioturbated (Toonen, Kleihans & Cohen, 2012). These more aerobic conditions could lead to faster decomposition of organic material. Another trend is that water fauna appear to exist continuously throughout the sequence, indicating a wet environment was always present.

Particle size

Figure 9 displays the results of the relative percentages of clay- silt- and sand fraction, and the average particle size for the whole sequence. Clay and silt make up the bulk of the sediment. Note that this is percentage-wise and not necessarily volume-wise. The clay content shows the most variation (range of app. 20%) followed by sand (app. 15%) and silt (app. 10%). A first trend that can be spotted is that the sand content decreases over time. However, there are also peaks in sand, indicating high energy environments. These peaks are not large, indicating a relatively low flow velocity. A fining-upwards trend as is common for such abandoned channels (described in Toonen et al., (2012)) seems gradual but it is also visible that the clay content increases (and the sand content decreases) and stabilizes relatively fast around 7500 BP (see figure ..).

Loss on ignition

Figure 9 also displays the results of the amount of water (LOI105), organic carbon (LOI550), inorganic

carbon (LOI950) (mainly comprised of CaCO3), and the amount of silica in the bulk soil expressed as a

percentage of the dry weight. Water content (LOI105) decreases between 9400 and 8600 BP,

stabilizes and increases from 7300 to 4300 BP. The same trend is visible for the silica content. An opposite trend is true for the organic carbon content (LOI550). A general trend for inorganic carbon

(LOI950) is not visible.

Particle size and loss on ignition

Comparing trends and peaks between particle size and LOI variables also yields results. Peaks in sand content occur with peaks of organic carbon and inorganic carbon as is visible around 8500 and 7900 BP (Figure 9).

Detrended correspondence analysis

After all DCA’s were run, It was attempted to interpret ecological gradients running along the axes. However, this appeared not possible, at least within the time limit of the research. For example, when performing a DCA for tree pollen, both extremes of the vertical and horizontal axes had very similar preferences concerning light, temperature, geographic location, moisture, soil pH, and nitrogen (Ellenberg & Leuschner, 2010). For the other DCA’s similar attempts were made yet no clear ecological gradients could be extracted from the figure. Thus, the results were mainly used to plot against age to identify changes in community composition.

Discussion

In order to provide an answer to the hypotheses, the results of this research are combined with the results of the 2016 research. Hypothesis will be discussed with separate stratigraphic diagrams.

Hypothesis I - humans are clearing the landscape on the local and regional scale

Figure 10 depicts the abundance of tree and grass pollen compared to tree and grass macrofossils. It also depicts the APh/NAph ratio and the DCA axis 1 scores. By combining these proxies, the first hypothesis: “humans are clearing the landscape on the local and regional scale” can be examined. Visible examination of both figures hints that the local vegetation does not always matches the regional vegetation (see figure 10). To quantify this, pollen abundance (%) and macrofossil abundance (#/15 cm3) were correlated for the same tree taxa. All correlations coefficients are depicted in table 1 in the appendix.

Of all tree species, Quercus pollen and macrofossils were positively correlated (r = 0.407, p = 0.023), and Betula pollen and macrofossils were positively correlated (r = 0.541, p = 0.002) (for all

correlations see table 2 in the appendix). However, the correlation between Betula pollen and

macrofossils is heavily influenced by the presence of an outlier so this outcome is doubtful (see figure 1b in appendix). This means H0 is only rejected in favor of HA for Quercus only. Despite the

correlation being significant, it is not very strong. Moreover, peaks in leaf fragments, which were mainly determined to be Quercus leaf, visibly occur with peaks in Quercus pollen and peaks of Quercus macrofossils. Quantifying this yields a high correlation between leaf fragments and Quercus pollen (r = 0.694, p = 0.000) and a moderate correlation between leaf fragments and Quercus macrofossils (r = 0.392, p = 0.029). As H0 was rejected for Quercus, it is expected that Quercus was abundant strongly both regionally and locally. However, may anthers filled with Quercus pollen were found (see figure 2b in the appendix). During preparation of the pollen samples, these anthers could have been broken, releasing pollen and heavily influencing the pollen sum. This means that Quercus pollen that was assumed to be derived from the whole region could in fact be very local. It is important to note that it is thus very difficult to separate a regional and a local signal when anthers filled with pollen are found. Similarly, a large increase in Betula pollen between 6500 and 6750 corresponds with a large increase in Betula macrofossils. Although no anthers containing Betula pollen were found this does not exclude that anthers were present and released pollen, again making it difficult to distinguish a regional and local signal. For grass pollen and macrofossils no correlations were run as it was clear from visual inspection that these would yield very low results.

Another interesting feature is a synchronous increase of Alnus and Corylus pollen around 6500 BP. The species are moderately correlated throughout the whole sequence (r = 0.533, p = 0.000). Vink (2016) argues that these increases could possibly be caused by people. Alnus prefers wet soils (Tamis et al., 2004) and Corylus moist soils (Tamis et al., 2004) and would unlikely colonize

deciduous Quercus-dominated forest, as Quercus prefers dry soils (Tamis et al., 2004). So a Quercus dominated forest excludes Alnus and Corylus. However, it is argued that after clearing of the

landscape, Alnus and Corylus are more competitive than other trees as they are both early

successional species (Berglund, 1991; cited in Kalis, Merkt & Wunderlich, 2003; Queen Mary, 2001). By combining this increase in Alnus and Corylus pollen with an increase in fire frequency (de Zwaan et al., 2016) and a decrease in Quercus pollen these findings could indicate humans clearing woodland (Vink, 2016). This clearing of woodland would be to create fields for crops or pasture. The hedges believed to border these fields would exist mainly of these early successional species, and woodland grazing would also propagate early successional species (Kalis, Merkt & Wunderlich, 2003) and this would be reflected in increased pollen levels.

Alnus and Corylus pollen both seem to provide a regional signal as no macrofossils are found and the correlation between pollen and macrofossils is low (see table 2). This could give reason to think that Alnus and Corylus were present in the region on a large scale. However, to directly link this to human interference is doubtful. Arguably, around the low-lying abandoned river channel Alnus dominated naturally (Zuidhoff & Bos, 2017). Yet for Corylus an alternative explanation of the increase refrains.

Figure 10. Hypothesis I. The abundance of important tree pollen, the abundances of corresponding macrofossils, the Aph/NAph ratio, and the DCA axes 1 for the proxies separately and combined.

Moving upwards in the sequence, the abundance of macrofossils and tree phytoliths declines for all tree species. This could be due to local clearing of the landscape. However, what is also likely is that macrofossils and phytoliths in this part of the core are badly preserved, as almost no macrofossils except at one instance are found. Macrofossils of Alnus, like catkins and anthers filled with pollen could have decomposed, leaving only a pollen signal that was interpreted as regional.

Even though regional and local vegetation did not match based on the correlations between pollen and macrofossil abundance, drawing inferences on whether deforestation is happening regional or local based on these correlations seemed oversimplified as the total findings of macrofossils is quite low and do not represent real quantities (Birks, 2007) . Looking at the data in another way yields different results. First, it is visible that Quercus pollen declines whilst Tilia pollen increase from the moment that the LBK culture arrives in the south of Limburg in 7300 BP (Louwe Kooijmans et al., 2005), indicating regional human impact on the landscape. Second, the APh/NAPh ratio decreases greatly from around the time of LBK arrival indicating local clearance of trees. Third, macrofossil abundance of tree species decreases greatly short after the arrival of the LBK indicating local

clearance of trees. Fourth, the DCA of pollen and the DCA of pollen + phytoliths indicate that regional vegetation undergoes compositional changes at the same time as local vegetation as was also found by Vink, (2016). All this suggests that regional and local vegetation is likely to be cleared at the same time. However, it should be noted that the regionality of the pollen signal of tree species, namely of Quercus, is debatable as anthers filled with pollen distorted this signal. In the worst case, this regional decline is thus not necessarily regional but in fact very local, but this seems impossible to retrace. Moreover, it should be noted that in the upper part of the sequence, macrofossils could have been deposited yet not preserved due to an non-optimal preservation environment due to more oxygen and because of more bioturbation (Toonen et al., 2012).

Hypothesis II - human clearing of the landscape causes flooding

Figure 11 depicts LOI and particle size in relation to climate, pollen, and charcoal. By combining these proxies, the second hypothesis: “human clearing of the landscape causes flooding” can be examined. Visible examination of figure 11 hints that peaks in sand do not correlate with peaks in charcoal and declines in pollen. It also hints that peaks in sand do not correlate with peaks in precipitation. However, sand peaks do seem to correlate to a certain extend with temperature. When quantifying these visual examinations with correlations, it becomes clear that sand does not significantly correlate with tree pollen, charcoal, or precipitation (see table 3) However, it should be noted that because it is likely that not all tree pollen levels were declining due to deforestation, correlating pollen to sand did not appear a suitable method to infer flooding caused by deforestation. Correlating sand to Quercus pollen, only in the period after the LBK culture enter southern Limburg (Louwe Kooijmans et al., 2005) could have yielded different results. Due to time limits this could not be done in this research. Sand does have a significant correlation with temperature, although not very high (r = 0.312, p = 0.033). Possibly, the reason for this is that the temperature dataset (from Southern Germany) is

geographically located near Well-Aijen. Based on these findings it can be concluded that H0 is not rejected, and it is unlikely that people or precipitation (at least this dataset) alone are the cause of floodings in Well-Aijen. However, running correlations for part of the sequence could yield different results as people are expected to be drivers of floods mostly in the upper part of the sequence (Kalis, Merkt, & Wunderlich, 2003). What stands out is the high concurrence of the increase in clay and silica, and decrease of sand and organic carbon after arrival of the LBK culture in the south of Limburg in 7300 BP (Louwe Kooijmans et al., 2005). To the extent it was researched here, it appears that people do not cause indistinguishable flood events based on correlations of regional tree pollen and charcoal with the sand fraction. However, sand, clay, organic carbon and silica do correlate with Quercus pollen well (r = 0.689, p = 0.000; r = -0.628, p = 0.000; r = 0.830, p = 0.000; r = -0.828, p = 0.000 respectively). A negative correlation between Quercus pollen and clay could be indicating that deforestation increases the amount of mineral material in the sediment mainly consisting out of clay.

Possibly this is local clearance caused local erosion leading to sediment being deposited into the channel.

What stands out is that normally high energy flood events are characterized by finding low

carbon/high silica levels and low clay and silt/high sand fractions synchronously (Gilli, Glur & Wirth, 2013; Brown, 1997).However, here increases in the sand fraction occur with increases in organic carbon (see figure..). Quantifying this visual examination, it shows that the sand fraction and LOI550

(organic carbon) are highly correlated (r= 0.739, p = 0.000). Possibly this is caused by an inwash of organic material by flooding. It is expected that floodings occurred with a slow flow velocity, washing in organic material into the abandoned channel. Another interesting aspect is that the sand fraction is very negatively correlated with the silica fraction (r = -0.791, p = 0.000) while the clay fraction is very positively correlated with the silica fraction (r= 0.747, p = 0.000). Likely, this is due to the high clay content throughout the sediment, so that the majority of the silica in the sediment is derived from the clay fraction.

Hypothesis III - microfossils can be washed in by flood events

In figure 12 the sand fraction and the DCA of pollen and phytoliths are depicted. By combining these proxies, the third hypothesis: “microfossils can be washed in by flood events” can be examined. Visible examination hints that the sand fraction is not correlated with the pollen DCA’s but that it is correlated with the phytoliths DCA’s. When quantifying these visual examinations it becomes clear that the sand fraction correlates negatively with DCA 1 of pollen (r = -0.602, p = 0.000) and that sand correlates positively with DCA 1 of phytoliths (r = 0.621, p = 0.000) (see also table 4 in the appendix). Based on these findings it can be concluded that H0 is rejected in favor of HA for phytoliths, which could mean that phytoliths are possibly washed into the system by flood events. In hypothesis II it was argued that floods could have caused an inwash of organic material, which could be the same

material possibly delivering phytoliths to the abandoned river channel. The inwash of phytoliths could in turn could distort the local signal that phytoliths are normally assumed to display correctly. To what extent remains unclear.

The possible inwash of microfossils is very dependent on the river’s ability to deposit material in the channel. As the abandoned channel has only a slight angle with the current river channel, this could be indicating that the channel was abandoned due to avulsion (Toonen, 2013). In case the

abandoned channel was formed in such a manner, the formation of a plug bar (sensu Toonen et al., 2012) closing of one bifurcated river channel from the other, would be gradual. As the plug bar gradually grows, less coarse material is deposited in the abandoned channel because of a higher required flow velocity to deposit material in the abandoned channel (Toonen et al., 2012). After the transition from a transitional stage (active abandonment) (sensu Toonen et al., 2012) to a

disconnected stage (sensu Toonen et al., 2012) the river would deposit its material less easy. It can thus be expected that as a plug bar has not been formed yet, inwash of microfossils is more likely, however, this should be done in further research correlating these proxies after onset of the disconnected stage (i.e. starting in the upper part of the sequence).

Further discussion: the first farmers

Another interesting dissimilarity between this research and prior research is the first presence of farmers. When discussing farming, a clear definition is required. In the literature, the LBK culture can be regarded as the first farming culture as they were full time farmers (Kalis, Merkt, & Wunderlich, 2003). They were the first settled culture practicing crop cultivation, animal husbandry and, with less importance, pottery production (Louwe Kooijmans et al., 2005). In general, the first farmers can be identified in a stratigraphic diagram via the presence of cereal pollen in combination with the presence of species associated with farming, and via spores of coprophilous fungi. In Bos & Zuidhoff, (2015) the first signs of pollen that resemble cereals appear halfway in what the authors palynologically and radiocarbon-dated as the Mesolithicum (early and middle Atlanticum). These first signs of pollen correspond to an age of about 7000 - 7100 BP, known via a 14C sample that was taken at this location. However, because of the unlikeliness of finding cereal pollen as early as this, the pollen

found here are presumed to be wind-dispersed pollen from wild grasses; pollen of wild grass and cereal are virtual indistinguishable (Behre, 2007; cited in Bos & Zuidhoff, 2015). ‘Real’ cereal pollen are presumed to be found later than 6000 BP in a period that was palynologically and radiocarbon-dated as the Neolithicum (late Atlanticum) (Bos & Zuidhoff, 2015). In this period, cereal pollen concentrations rise to approximately 3-7 %, whereafter with a short lag, herbaceous species

associated with agriculture and trampling rise (Bos & Zuidhoff, 2015). In this research, the first signs of cereal pollen is around 8000 BP, confirmed by a 14C sample taken at this location (see figure 13). However, large increases of cereal pollen only occur from around 6600 BP, also confirmed by a 14C sample taken at this location (see figure 13). So similarly to Bos & Zuidhoff (2015), it is expected that the early cereal pollen signal is in fact pollen from wild grasses, and that findings true cereal pollen occur around 6600 BP, where concentrations of cereal pollen increase (see figure 13). This is supported by the fact that macrofossils of the Poaceae Glyceria species, a wild grass, are found to occur quite synchronously with the peaks of cereal pollen (see figure 13). These macrofossil findings indicate that many of these pollen deemed to be cereal are in fact likely originating from wild grasses. With the first signs of the ‘cereal’ pollen (see figure 13) Plantago major, a herbaceous plant

associated with farming, increases too, which hints to early agricultural practices. However, it is expected that this is due to a determination error. Peaks in Plantago major, starting at around 7900 BP occur similarly with increases in macrofossils of Alisma plantago-aquatica (see figure 13). Likely, pollen of Alisma plantago-aquatica was mistaken for Plantago major (van Geel, personal

communication). This is possible because Plantago species have periporate pollen (Weber, 1998) and Alismataceae also have periporate pollen (Argue, 1971).

Increased spores of the Sporormiella-type from 7300 BP, an increase in Poaceae pollen, and a decrease in the APh/NAPh ratio (figure 13) suggest that LBK people use the grounds around Well-Aijen as pasture for their animals, that maintain an open landscape by grazing (Vera, 2000; Kalis, Merkt & Wunderlich, 2003). The use of these grounds as pasture and for hunting is also argued by Louwe Kooijmans et al., (2015) based on findings of ‘remote’ artifacts (not near settlements) namely arrowheads dating from the Early Neolithic. Perhaps some crop cultivation already took place as cereal abundance fluctuates less in this period, as noted by Vink, (2016). However, it could also be pollen derived from imported cereal that was released when threshed, dehusked, and winnowed and instead of being from locally grown cereal (Verhart, 2016; Kalis, Merkt, & Wunderlich, 2003). It is argued that crop cultivation started later, around 6600 BP: first, cereal pollen increases greatly. Second, the charcoal volume increases greatly. Third, the APh/NAPh ratio decreases greatly (i.e. the grass phytoliths increase greatly whilst tree phytoliths decrease greatly). These three things point to local deforestation to clear land for crop cultivation. Fourth, Alnus and Corylus increase, possibly as a result of landscape clearing and its occurrence in hedges bordering fields (Kalis, Merkt, & Wunderlich, 2003). Fifth, it is suggested by archeology that neolithisation had only occurred in the north of Limburg around 6400 BP whereas before, people in the north relied more on a hunter-gatherer lifestyle

(Verhart, 2016). Sixth, research from Germany suggest human impact increases significantly around 6300 BP (Kalis, Merkt, & Wunderlich, 2003). Moreover, it is expected that agriculture intensifies around 5800 BP. This is based on findings of Plantago lanceolata (figure 13), a good indicator species of crop cultivation as it is a ruderal species that favors fallow land (Behre, 1988). Findings of this species occurs from around 5700 BP. Early neolithic farmers harvested only spikes of the cereals, not removing much biomass, and so a shifting culture is not assumed (Willerding, 1980). Thus no fallow phases were expected (Willerding, 1980; Knörzer, 1986; cited in Kalis, Merkt, & Wunderlich, 2003). Finding Plantago lanceolata could then indicate that a shifting culture was adopted as there was more fallow land leading to more Plantago lanceolata. Moreover, spores of the sporormiella-type increase around 5800 BP perhaps indicating extensification of livestock.

Shortcomings of the research

Interestingly, comparing macrofossils and pollen has provided some insight into differentiating a local and a regional signal (hypothesis I). However, a drawbacks of using macrofossils as a proxy for local vegetation is that when pollen anthers are present in the sediment, these can break during pollen sample preparation. This will highly influence the pollen sum, leading to an overestimation of the regional signal. Moreover, it is unknown whether all macrofossils are preserved equally well in the soil. It could be possible that Quercus material is preserved better because of its tannins having toxic effects on some microbial decomposers, slowing decomposition (Traw & Gift , 2010).Moreover, as a channel fills, it shallows, creating an oxidized and highly bioturbated environment at the top of the sequence (Toonen et al., 2012) which leads to increased decomposition of organic material. This could allow researchers to draw the conclusion that some species were present locally but others were not while these in fact were. Examining phytoliths is thus also recommended for examining a local vegetation signal. Lastly, there is a certain learning curve related to macrofossil determination. Macrofossils can remain unidentified when researchers do not recognise plant material as an interesting feature that requires inspection. There is also the possibility that researchers learn to recognize a certain macrofossil halfway through the research. Both lead to partially incorrect results and thus demonstrates the importance of expert knowledge.

Suggestions for further research

Several suggestion for further research can be given. A first suggestion for further research is to run correlations for separate zones. The zones can be constructed based on changes in human impact on the landscape, but also on hydrological impact. It could well be that in one zone, two proxies are not correlated whilst in another period, these proxies become strongly correlated. This could lead to totally different results. A second suggestion is to quantify impact that floods could have in washing in phytoliths, resulting in a distortion of the local signal. This would lead to a better comprehension of the gravity of distortion of the local signal. A third suggestion to extend this research is that the grain size analysis and age-depth modelling could be ameliorated with works by Toonen, (2013). This includes e.g. change point analysis (CPA) (Taylor, 2000; cited in Toonen, 2013) which could be used to confirm visually observed breaks in the sequence, and the use of other grain size descriptors and end-member modeling to extract otherwise overseen information (Toonen, 2013). This would also make zonation of hydrological impact easier. A fourth suggestion is that removing high-variation species in DCA’s could to lead to a better interpretation of an ecological gradient along the DCA axes, as this could drastically change the position of species scores on the axes. Lastly, the high

abundance of Quercus leaf fragments present in the sediment provide an excellent opportunity to perform an analysis of stomatal frequency to infer past carbon dioxide levels. (Wagner, Kouwenberg, van Hoof & Visscher, 2004).

Conclusions

This research has indicated that humans increasingly affect their environment through time. Mainly, human impact can be seen through deforestation. Deforestation is mainly visible in the decline of tree pollen of species like Quercus, indicating regional deforestation. However, the regionality of the pollen signal of Quercus is debatable as anthers filled with pollen distorted this signal. In the worst case, this regional decline is thus not necessarily regional but in fact very local, but this is hard to retrace. Increases of pollen from Alnus and Corylus are possibly due to the formation of hedges surrounding fields, or due to local prevalence although this is not visible in the macrofossils. However, the macrofossils could have been deposited yet not preserved due to an non-optimal preservation environment. Local vegetation is likely to be cleared as well, as the APh/NAPh ratio and macrofossil abundance decline from the point where Quercus pollen declines. Also the DCA of pollen and the DCA of pollen + phytoliths are similar indicating local and regional vegetation changing at the same time. It was found that people were not the main drivers of flood events throughout the researched time period. However, correlations could be run on only the upper part of the sequence in order to see whether humans could be important drivers here. Precipitation also did not seem the main driver of

flood events, yet temperature was more correlated. This could be due to the geographic locations of the proxy data used. Moreover, it was found that flood events, seen in high sand fraction, correlates with the organic carbon percentage, possibly because of an inwash of organic material. It was also found that phytoliths could have been washed in by these flood events, distorting the local vegetation signal. However, to what extent remains unclear.

This research also shed light on the advent of agriculture in the region. People presumably enter the landscape around 7300 BP accompanied by livestock, bow and arrow, and perhaps with some heads of cereal. Subsequently, in 6600 BP arable agriculture surely started as neolithisation completed in the region and in 5800 BP agriculture is argued to intensify. It should be noted that macrofossils appeared very useful for noticing errors in the dataset; pollen that was likely falsely identified as cereal instead of wild grasses, and pollen that was falsely identified as Plantago major instead of Alisma plantago-aquatica would have lead to false interpretation on the advent of farming. Instead these errors were found with the help of macrofossils.

Acknowledgements

First, I would like to thank Dante Follmi for the time we had working together on our theses. We had some long days, heated debates, insightful moments, and well-deserved beers. Cheers! Second, I would like to thank Crystal McMichael and William Gosling for offering us the possibility to join this project, the trust you put into our capabilities and our will to learn, and conveying your motivation and inspiration, which you emit so truthfully. Go paleo! Third, I would like to thank Bas (‘The Boss’) van Geel for the many hours of dedicated help, for keeping us sharp, your discipline, your sense of humor, and for your passion for what you do. Fourth, I would like to thank Hanneke Bos for taking the time to help us determining the macrofossils we and Bas were not able to determine ourselves. Fifth, I would like to thank Rutger van Hall for instructing us on the LOI procedure. Sixth, I would like to thank Martine Hagen, Maarten Prins, and Hans from the VU for helping us with operating the HELOS to determine the particle size of our samples. Seventh, I would like to thank last year's’ students for initiating this project and the work they’ve put into it. Eighth, I would like to thank my fellow bachelor students for the good vibes in the lab and the feeling we were the ‘paleo-crew’.