2.1 Assessing the degree of landscape change

Differences in landscape resilience, thresholds, triggers, feedbacks and impact of the landscape changes are evaluated and compared for the three regions, following definitions from the stable states theory of Scheffer et al. (Reference Scheffer, Carpenter, Foley, Folke and Walker2001). This theory describes how a system – in this case a geographical area with certain human–landscape interactions – can have multiple stable states. For example, a bog peat suited for reclamation can change into drowned tidal areas, which again can change into infilled tidal areas with dry places to live. Each state comprises its own set of distinct feedbacks, i.e. geomorphological processes and human–landscape interactions. Under varying environmental conditions, these areas can shift into different states (regime shift), but this does not happen via a linear response. Instead, these states are separated by a tipping point, essentially a point of no return, flipping the system into a contrasting self-stabilising state with its own distinct processes and feedbacks. Restoring earlier environmental conditions will not automatically recover the system’s previous state. The stability of such a state is known as resilience. A larger resilience will require a larger perturbation to cause a shift to an alternative stable state. The resilience may be affected by gradually changing conditions that may initially have little direct effect on the state of the landscape (e.g. slow peatland subsidence in response to reclamation). Over the course of time, however, such gradual conditions could make a landscape more fragile, so that it can more easily tip into an alternative state by stochastic events (e.g. by flood events). The concepts explained above will be used in this study to compare the size and impact of the landscape changes in the first millennium AD. These may range from almost no change (because of high resilience) to a full regime shift. They may also include significant changes, while still maintaining a comparable landscape with similar feedbacks and human–landscape interactions.

2.2 Data resolution and possible drawbacks

The quality of the reconstruction-based studies that underlie this review heavily depends on the resolution of the underlying landscape reconstructions and the degree to which the main driving factors could be assessed. In the coastal and fluvial area, new integrated GIS landscape reconstructions with a uniform spatial coverage presented in Pierik et al. (Reference Pierik, Cohen and Stouthamer2016, Reference Pierik, Cohen, Vos, van der Spek and Stouthamer2017a,b) show and analyse the main landscape changes. These studies mainly use existing maps in the coastal area, whereas in the fluvial area high-quality borehole data were used, providing high spatial and chronological resolution. In the Pleistocene sand area, a more site-based approach was applied or drift-sand reconstruction rather than a uniformly covering reconstruction (Pierik et al., Reference Pierik, Van Lanen, Gouw-Bouman, Groenewoudt, Wallinga and Hoek2018b), because of the patchy distribution of drift sands before the 17th century. Start and end ages of individual sites are generally dated within a 500-yr range. When optically stimulated luminescence (OSL) dates and/or historical references were used, the accuracy of driftsand activity can be several decades. By combining multi-proxy evidence of regional drift-sand activity from c.140 sites and 340 dates, the trends unravelled in drift-sand activity have a maximum age uncertainty of one to two centuries.

The factors that could have driven landscape changes either are natural or human-induced. Independent sedimentological records, for example, may reveal flood events as natural triggers. Individual severe floods generally were dated more precisely with 30–150 yrs precision, or on the resolution of a single year using dendrochronology (Jansma, Reference Jansma2020), whereas floods dated based on archaeology were dated at a 20–200 yr resolution (Toonen et al., Reference Toonen, Donders, Van der Meulen, Cohen and Prins2013; Cohen et al., Reference Cohen, Toonen and Weerts2016). Settlement dynamics from the archaeological record were used both as an indication for human pressure and as an indication of human response to landscape change. Often, archaeological finds are additionally used to date geomorphological changes. To avoid potential circular reasoning, independent additional proof of new absolute dates and stratigraphical correlations in the data-rich study area were used where possible. For example, archaeological finds in the coastal plain on top of peat covered by clay layers indicate reclamation activity. They were additionally used to date the peatland drowning indicated by the clay layers, assuming that peatland reclamation caused the drowning. The causal relation then had to be proven independently, for example by assessing additional evidence of the identified reclamation (rectangular in-filled tidal channels) and by using additional dates of associated materials (e.g. ¹⁴C of the peat or shells inside the tidal deposits). Another challenge in inferring geomorphological processes from reconstructions is that multiple processes can lead to similar outcomes (equifinality) and that local processes propagating to larger areas may be overlooked. This was partly circumvented by regional upscaling to allow comparison of the changes for multiple cases (rivers, tidal systems, sand areas), with well-known differing natural preconditions and forcings.

3.1 Coastal plain

In the first millennium AD, large tidal areas were present in the coastal plain with channels, shoals, supratidal saltmarshes, and coastal peat swamps (Figs 2 and 3). The coastal plain width ranged from c.60 km in the southwest and Holland to c.30 km in the northern Netherlands and includes the Almere lagoon in the middle part of the Netherlands. In the northern Netherlands the Wadden Sea was fringed by saltmarshes that became inhabited c.650 BC (Taayke, Reference Taayke1996). This area became one of the most densely populated areas in the Netherlands during the first millennium AD (Groenewoudt & Van Lanen, Reference Groenewoudt and Van Lanen2018). Settlements here were mainly confined to artificial dwelling mounds that protected the inhabitants from the highest water levels (Van Giffen, Reference Van Giffen1940; Gerrets, Reference Gerrets2010; Nieuwhof et al., Reference Nieuwhof, Bakker, Knol, de Langen, Nicolay, Postma, Schepers, Varwijk and Vos2019). Other densely populated areas in the coastal plain were tidal natural levees and abandoned tidal creeks around the estuaries of the Rhine, Meuse and Oer-IJ (Figs 2 and 3).

Fig. 3. (A) Three landscape states in the coastal plain (after Pierik et al., Reference Pierik, Cohen, Vos, van der Spek and Stouthamer2017a; De Haas et al., Reference De Haas, Pierik, Van der Spek, Cohen, Van Maanen and Kleinhans2018). From pristine peatlands to sea ingression dominant in the southwestern and northern Netherlands (discussed in this paper), to fully embanked and reclaimed in the Late Middle Ages. (B–F) GIS-derived coastal development since 600 BC (Pierik et al., Reference Pierik, Cohen and Stouthamer2016, Reference Pierik, Cohen, Vos, van der Spek and Stouthamer2017a). Pie charts demonstrate relative areal extent of the landscape units per coastal segment; colours correspond to the map legend units.

The tidal areas were fringed by large freshwater peat swamps. In the southwestern and northern coastal plain, these peatlands were reclaimed and cut for fuel on a large scale from the onset of the first millennium AD (Vos & Van Heeringen, Reference Vos and Van Heeringen1997; Vos & Knol, Reference Vos and Knol2015). These activities caused major landscape changes and eventually transformed a large part of the pristine coastal plain peatlands into tidal areas (Vos, Reference Vos2015; Pierik et al., Reference Pierik, Cohen, Vos, van der Spek and Stouthamer2017a – ‘sea ingression state’ in Fig. 3). These sea ingressions were mainly the effect of peatland subsidence caused by groundwater level lowering during reclamation. This was followed by considerable land losses, mainly during storm surges, as tidal channels initiated and grew by erosion of the tidal flow. The largest land losses occurred in the SW Netherlands, which was drowned for multiple centuries from AD 270 onwards; people abandoned large parts of this plain. In the northern Netherlands, ingression in peatlands probably started in the Roman age and expanded between AD 700 and 1000 (Vos & De Langen Reference Vos and de Langen2010; Vos & Knol, Reference Vos and Knol2015), most likely in response to fringing peatland reclamation from the 4th century BC onwards (Nieuwhof et al., Reference Nieuwhof, Bakker, Knol, de Langen, Nicolay, Postma, Schepers, Varwijk and Vos2019). The presence of supratidal saltmarsh levees fringing the Wadden Sea caused ingressions to be relatively smaller. Also in the NW coastal plain, barrier erosion combined with peat drainage caused tidal area expansion (Vos et al., Reference Vos, Van der Heijde and Stuurman2015b). Over the course of centuries, sediment import caused the ingressions to fill. The rate and scale of infilling depended on the extent of the ingressed area, sediment supply, flood dominance, and subsidence feedbacks (Pierik et al., Reference Pierik, Cohen, Vos, van der Spek and Stouthamer2017a). In the SW Netherlands, tidal sedimentation reached supratidal levels that became inhabited again around AD 700 (Vos & Van Heeringen, Reference Vos and Van Heeringen1997).

While most of the coastal plain experienced peatland loss in the first millennium AD, phases of peat expansion have also been observed. The natural closure of the Oer-IJ estuary around 200–150 BC (Vos et al., Reference Vos, De Koning and Van Eerden2015a) led to drainage problems generating peat expansion on the fringing saltmarshes at the onset of the first millennium. Other examples are found around the Meuse estuary (Vos, Reference Vos2015; Moree et al., Reference Moree, Van Trierum and Carmiggelt2018) and in the northern Netherlands (Layer of Tinga – Vos & De Langen, Reference Vos and de Langen2008). Both peat regrowth events have been linked to depopulation and subsequent discontinuation of maintenance of reclamation works, causing rewetting of the landscape. In the northern Netherlands, wetness of the landscape additionally increased as water routing towards the sea was hampered. This was the result of continued northward saltmarsh accretion combined with gradual sea-level rise (Vos & Knol, Reference Vos and Knol2015; Nieuwhof et al., Reference Nieuwhof, Bakker, Knol, de Langen, Nicolay, Postma, Schepers, Varwijk and Vos2019). These environmental factors may have contributed to the major population decline in the 3rd century in the northern Netherlands, besides being caused by cultural factors and large-scale migration (Nieuwhof, Reference Nieuwhof2016).

3.2 Rhine–Meuse delta

In the Rhine–Meuse delta, several river branches changed their courses around the first millennium AD (avulsion; Figs 2, 4 and 5 – Berendsen & Stouthamer, Reference Berendsen and Stouthamer2000, Reference Berendsen and Stouthamer2001; Van Dinter et al., Reference Van Dinter, Cohen, Hoek, Stouthamer, Jansma and Middelkoop2017). Stepwise, the main discharge was routed away from the Old Rhine river in the north (forming the Roman limes) to the Meuse estuary in the south of the delta. The formation of the Hollandse IJssel and Lek in the downstream delta between c.AD 1 and 300 is one of the first steps. For 3000 years, a large swamp had separated the Old Rhine and Old Meuse rivers here until these two new Rhine branches traversed it. Human-induced peat subsidence (in the current Rotterdam area) and wetland deforestation likely contributed to the avulsion success of these rivers (Pierik et al., Reference Pierik, Stouthamer, Schuring and Cohen2018a). Human-induced increased suspended sediment supply – related to deforestation in Germany – to the rivers and estuaries caused additional peat subsidence by loading processes in the lower delta, thereby probably accelerating the avulsion process. In the apex region, the Waal river gradually gained more discharge at the expense of the Nederrijn (Weerts & Berendsen Reference Weerts and Berendsen1995). The Gelderse IJssel formed, flowing outside the delta towards the central Netherlands lagoon (Almere), connected to the Wadden Sea in the north (Makaske et al., Reference Makaske, Maas and Van Smeerdijk2008; Cohen et al., Reference Cohen, Stouthamer, Hoek, Berendsen and Kempen2009, Reference Cohen, Pierik, Toonen, van der Meulen, Bomers, Gouw-Bouman, Van Popta, Jansma, Erkens, Busschers, Schuring, Volleberg, Stouthamer and Hoek2021). All these developments contributed to the gradual abandonment of the Old Rhine, which had been the main Rhine branch in the delta until the first millennium AD (Van Dinter et al., Reference Van Dinter, Cohen, Hoek, Stouthamer, Jansma and Middelkoop2017; De Haas et al., Reference De Haas, van der Valk, Cohen, Pierik, Weisscher, Hijma, van der Spek and Kleinhans2019).

Fig. 4. Geomorphological reconstructions for AD 100, 500 and 900, for the central and upper Rhine–Meuse delta (Pierik, Reference Pierik2017b; Pierik et al., Reference Pierik2017b). Cities: 1 = Utrecht, 2 = Nijmegen, 3 = Dorestad, 4 = Tiel.

Fig. 5. (A, B) River network around AD 100 and AD 900 (Jansma et al., Reference Jansma, Van Lanen and Pierik2017; Pierik et al., Reference Pierik2017b); (C) activity of major rivers in the research area after Cohen et al. (Reference Cohen, Stouthamer, Pierik and Geurts2012), Van Dinter et al. (Reference Van Dinter, Cohen, Hoek, Stouthamer, Jansma and Middelkoop2017), Pierik et al. (Reference Pierik, Stouthamer, Schuring and Cohen2018a); (D) Holocene recurrence time of floods from black to light grey: 250 yr (very severe flood), 100 yr (severe flood), 50 yr (moderately severe flood), after Toonen et al. (Reference Toonen, Donders, Van der Meulen, Cohen and Prins2013) and Cohen et al. (Reference Cohen, Toonen and Weerts2016); here the events are grouped per archaeological subperiod, and the ages of the three largest floods are indicated; orange bars indicate flood events with >50 and >100-yr recurrence intervals, derived from wood samples in the Netherlands (i.e. also outside the delta) (Jansma, Reference Jansma2020); (E) delta-wide average settlement elevation (Pierik & Van Lanen, Reference Pierik and Van Lanen2019); (F) elevation difference between persistent and abandoned settlements shows that settlements at higher locations persist more often (Pierik & Van Lanen, Reference Pierik and Van Lanen2019); (G) reconstructed number of inhabitants in the study area (after Van Lanen et al., Reference Van Lanen, De Kleijn, Gouw-Bouman and Pierik2018).

Habitation in the Rhine–Meuse delta mainly occurred on alluvial ridges, which formed the higher and drier parts of the delta (Modderman, Reference Modderman1948; Pierik & Van Lanen, Reference Pierik and Van Lanen2019). These alluvial ridges, consisting of crevasse splays and natural levees, were not only essential for delta inhabitants, they also largely controlled floodwater routing and sediment deposition (Pierik et al., Reference Pierik2017b). In the upper and central parts of the delta, avulsion paths were controlled by flood-basin and natural-levee topography. While the levees along young branches grew relatively high during the first millennium AD, the flood basins and fossil levees collected a substantial clay cover causing delta-topography levelling.

After AD 270, severe depopulation and relocation of settlements took place in the fluvial area (Fig. 5G; Hendrikx, Reference Hendrikx1983; Willems, Reference Willems1986; Van Lanen et al., Reference Van Lanen, De Kleijn, Gouw-Bouman and Pierik2018). The major part of the area was even abandoned during the 3rd century (Heeren, Reference Heeren2015). Besides cultural factors, this was possibly influenced by the varying environmental conditions. Between AD 270 and 850, preferred settlement locations on the alluvial ridges contracted and shifted towards the higher parts of the ridges, which is attributed to increased flooding intensity for this period (Fig. 5E and F; Pierik & Van Lanen, Reference Pierik and Van Lanen2019; Jansma, Reference Jansma2020). These settlement shifts hardly affected the route networks, since almost all alluvial ridges remained habitable and accessible. The strongest trends are observed in the lowest areas (Bommelerwaard, southern part of the delta). This area was initially dominated by the Meuse which had relatively low flooding activity between AD 250 and 950 (Peng et al., Reference Peng, Prins, Kasse, Cohen, Van der Putten, van der Lubbe, Toonen and van Balen2019). Around AD 450 the Waal (Rhine) branch became more dominant in this area, likely bringing higher and more frequent floods. Here, relatively many settlements were abandoned and the route networks were reorganised after the late Roman period (Pierik & Van Lanen, Reference Pierik and Van Lanen2019; Van Lanen & Pierik, Reference Van Lanen and Pierik2019).

3.3 Pleistocene sand area

The Pleistocene sand area consists of modestly elevated sandy uplands (several metres to tens of metres above sea level), with peat-infilled lower plains and brook valleys (Figs 1 and 2). Compared to the adjacent fluvial and coastal areas, this area was much less dynamic from a geomorphological perspective in the first millennium AD. Probably peat bogs were already cut for fuel locally, but large-scale peat cutting did not start before the Late Middle Ages (Spek, Reference Spek2004; Van Beek, Reference Van Beek2009). The onset of drift-sand formation – often related to deforestation or intense land use – was a second geomorphological change over the last few thousand years (Fig. 6; Koster et al., Reference Koster, Castel and Nap1993; Sevink et al., Reference Sevink, Koster, Van Geel and Wallinga2013). Drift-sand activity was mainly caused by human land-use practices (De Keyzer & Bateman, Reference De Keyzer and Bateman2018). This is evident from a peak in activity after c.AD 900 that coincided with higher population densities and larger-scale deforestation, especially in southern and middle sand areas where population increase was strongest. More frequent occurrence of drift-sand activity close to settlements and route corridors additionally indicates human pressure as an important cause behind sand drifting (Pierik et al., Reference Pierik, Van Lanen, Gouw-Bouman, Groenewoudt, Wallinga and Hoek2018b). Drift-sand activity and its impact on people in the first millennium AD were restricted to the local scale. The mid-Roman period with a high population pressure, however, counter-intuitively has low drift-sand activity. This may be linked to migration from sandy to slightly loamier areas, associated with abandonment of Celtic fields (Pierik et al., Reference Pierik, Van Lanen, Gouw-Bouman, Groenewoudt, Wallinga and Hoek2018b).

Fig. 6. Drift-sand locations in Roman Age (A) and Early Middle Ages (B). Data available via Pierik (Reference Pierik2017c).