Multi-Millennial Decoupling Between Agricultural Presence and Landscape Transformation: Quantitative Evidence from 676 European Pollen Sites

Target journal: The Holocene Version: v3.0 (2026-03-28, major revision -- compressed to ~9,000 words, temporal smearing test added, climate confounds acknowledged, Deza-Araujo engagement expanded, Japan reduced to 1 paragraph) Status: Revised manuscript addressing reviewer comments

Abstract

The arboreal pollen (AP) ratio is the standard proxy for reconstructing agriculture-driven deforestation, yet it integrates pollen from a regional source area (tens of kilometres) while early farming operates at local scales (hundreds of metres). We quantify this scale mismatch by simultaneously tracking AP ratios and agricultural indicator pollen (Cerealia-type, Plantago lanceolata, Rumex, Artemisia) from the Neotoma Paleoecology Database across three European regions: Britain (120 sites; 19,578 radiocarbon dates from p3k14c), Scandinavia (256 sites; 9,689 dates), and the Central European Alps (300 sites; 2,451 dates).

In regional composites, AP ratios showed negligible change at agricultural introduction: -5.8% (Britain), +2.1% (Scandinavia), +3.9% (Alps). Yet Cerealia-type pollen increased x15 (Britain, Scandinavia) and x11 (Alps) during the same interval. Bootstrap lag estimation quantifies the decoupling at 3,400 years in Britain (95% CI: 2,000--4,000), 4,400 years in Scandinavia (CI: 3,800--5,000), and 4,800 years in the Alps (CI: 3,195--5,000).

Site-level replication reveals that 55--74% of individual sites show "AP decline first" -- the reverse of the composite pattern -- while only 14--23% show "indicator first." A temporal smearing test demonstrates this discrepancy is not an artifact of differential agricultural arrival times: sites with the earliest Cerealia onset are significantly more likely to show the "indicator first" pattern (Spearman rho = 0.269, p < 0.0001), contradicting the temporal averaging hypothesis and confirming that the composite lag reflects genuine scale-dependent detectability. Segmented regression identifies a threshold at 0.014% Cerealia (p = 0.026) below which AP decline is disproportionately steep.

These findings demonstrate that the onset of agriculture and the onset of AP-detectable landscape change are decoupled by 3,400--4,800 years across European regions. Agricultural indicator pollen should be routinely reported alongside AP ratios to distinguish agricultural presence from landscape transformation.

1. Introduction

The arboreal pollen (AP) ratio has served for decades as the principal palynological proxy for reconstructing human landscape transformation during the Holocene. Where people clear forests, arboreal pollen declines relative to non-arboreal pollen (NAP), and this logic underpins influential continental-scale reconstructions. Woodbridge et al. (2014) demonstrated coupling between radiocarbon-based population proxies (summed probability distributions, SPDs) and AP decline across Britain. Roberts et al. (2018) extended this to Europe using the European Pollen Database, and Kaplan et al. (2009) translated pollen-derived land-cover estimates into Holocene carbon budgets. In each case, the AP ratio served as the empirical bridge between pollen records and past land use.

There is, however, a fundamental scale mismatch in this framework that has received insufficient attention. The AP ratio integrates pollen from a large source area. Wind-pollinated arboreal genera -- Pinus, Quercus, Betula, Alnus -- disperse pollen over distances of tens of kilometres, meaning that the AP signal recorded at any single site represents a regional vegetation average encompassing a wide surrounding landscape (Sugita 2007). Early agriculture, by contrast, operates at a much smaller spatial scale. A Neolithic clearing of a few hectares, a Bronze Age field system extending over a single valley side -- these are local phenomena, measured in hundreds of metres, not tens of kilometres.

Cerealia-type pollen -- identified by grain diameter >40 um, annulus diameter >8 um, and thick exine sculpturing (Behre 1981), including Triticum, Hordeum, Secale, and Avena -- is large, heavy, and poorly dispersed by wind, with effective source areas typically within approximately 500 metres of the parent plant. The underrepresentation of cereal pollen is not merely a dispersal-distance effect but reflects fundamental differences in pollen productivity. Brostrom et al. (2008) compiled relative pollen productivity (RPP) estimates for European taxa and found that Triticum (wheat) has an RPP of approximately 0.0008 relative to Poaceae -- meaning that for equal vegetation cover, wheat produces roughly 1,250 times less pollen than wild grasses. Secale (rye), being wind-pollinated, has a substantially higher RPP (~4.0), but Hordeum (barley) is intermediate (~3.2). These RPP differentials mean that cereal cultivation -- particularly wheat-based systems -- is systematically underrepresented in the pollen record even when agricultural indicator taxa are specifically tracked.

This source-area mismatch has a critical implication: the AP ratio measures regional forest cover, not agricultural presence, and will therefore not register the onset of agriculture unless farming operates at a spatial scale large enough to alter the regional forest composition. If early farming clears only small, scattered patches within an otherwise forested landscape, the regional AP signal will remain stable even as farming activity intensifies locally. The AP ratio will eventually decline -- but only when agriculture scales up to transform the landscape at the regional level, which may occur centuries or millennia after farming first appeared. Bogaard (2004) demonstrated from archaeobotanical evidence that early Neolithic farming in central Europe was characterised by "intensive garden cultivation" on fixed plots near settlements -- a labour-intensive system that did not require large-scale forest clearance and would be palynologically invisible at off-site pollen recording locations.

Recognising this limitation, Deza-Araujo et al. (2022) developed a Land Use Pollen (LUP) index that aggregates cultural indicators to quantify agricultural land-use intensity independently of AP ratios. Their work, validated across all major European biomes, demonstrates that indicator-based approaches can reveal agricultural activity invisible to AP-only analysis. Our approach differs from the LUP index in a specific way: whereas the LUP quantifies the intensity of agricultural land use at a given time, our dual-signal framework explicitly measures the temporal lag between indicator onset and AP decline -- the duration of the phase during which agriculture is present but has not yet transformed the regional landscape. The two approaches are complementary: the LUP confirms the phenomenon, and our lag estimation quantifies its duration.

Rosch et al. (2014) proposed a two-phase model from southwestern German evidence, distinguishing Phase 1 (small-scale garden farming generating local Cerealia signals without affecting regional AP) from Phase 2 (extensive agriculture producing landscape-scale deforestation). Alenius et al. (2021) demonstrated the same scale dependence in eastern Fennoscandia using the Landscape Reconstruction Algorithm. Our companion study (Author et al., in review) found that AP ratios at most Japanese sites showed no response to the introduction of wet-rice agriculture during the Yayoi period (~2,580 cal BP), with major deforestation occurring only during Kofun-period state formation (~1,709 cal BP). But this finding raised an interpretive ambiguity that the AP ratio alone cannot resolve: was the lag a "lag in impact" or a "lag in detection"?

The present study addresses this ambiguity through what we term the "dual-signal" approach. Rather than relying on the AP ratio as a single metric, we simultaneously track two independent pollen signals: the AP ratio itself, which records regional-scale forest cover, and agricultural indicator pollen taxa -- specifically Cerealia-type, Plantago lanceolata (a pastoral and trampling indicator), Rumex, and Artemisia -- which record local-scale farming activity. If the AP ratio remains stable while agricultural indicator pollen rises, we can conclude that agriculture is present and detectable in the pollen record but has not yet transformed the regional landscape. The dual signal thus disentangles the two meanings of an unchanged AP ratio: forest persistence in the absence of agriculture versus forest persistence in the presence of spatially restricted agriculture.

This phenomenon has precedent at the site level. Rosch & Lechterbeck (2016) documented exactly this pattern at Litzelsee in the Lake Constance region, where Cerealia-type and Plantago lanceolata appeared while forest composition showed minimal change during early agricultural phases. Our study provides the first systematic multi-regional quantification of this local-regional mismatch across approximately 676 sites in three European regions with fundamentally different agricultural histories, computing bootstrap lag estimates with confidence intervals, conducting site-level replication analysis, and testing the temporal smearing hypothesis.

Four research questions guide the analysis. First, do agricultural indicator pollen taxa increase before AP ratios decline -- that is, is there a detectable local-regional mismatch phase in each European region? Second, how long does this mismatch phase last, and can the lag between Cerealia rise and AP decline be quantified with bootstrap confidence intervals? Third, does the composite-level pattern replicate at individual sites, or does site-level analysis reveal a different temporal sequence -- and if the latter, is the discrepancy an artifact of temporal smearing? Fourth, is there a threshold in Cerealia abundance below which AP decline is steepest -- suggesting that the earliest, lowest-intensity agriculture produces disproportionate local forest disturbance? Candidates for what ultimately breaks the dual signal include population growth crossing a critical density threshold, political reorganisation enabling large-scale coordinated land clearance, and technological change (such as the adoption of iron tools or heavy ploughs) reducing the labour cost of forest removal. Stevens and Fuller (2012) argued that early Neolithic agriculture in Britain was a marginal subsistence strategy that did not intensify until the Bronze Age; Stephens et al. (2019) proposed that significant human modification of global land cover did not begin until approximately 3,000 years ago. Our multi-regional comparison provides an empirical framework for evaluating these competing hypotheses.

2. Materials and Methods

2.1 Pollen data

Pollen records were obtained from the Neotoma Paleoecology Database (v2.0 API) using POST requests with GeoJSON bounding boxes. For each site, arboreal pollen (AP) ratio was calculated as:

AP ratio = S(TRSH + MANG + PALM) / S(TRSH + MANG + PALM + UPHE + SUCC)

where ecological group assignments follow Neotoma classification. Sites were retained if they had >=3 samples with >10 total pollen grains and age control.

2.2 Radiocarbon data and SPD computation

Population dynamics were estimated from SPDs of radiocarbon dates, used as contextual indicators of demographic change consistent with Crema (2022) and Carleton and Groucutt (2021).

• Britain: p3k14c (19,578 dates)
• Scandinavia: p3k14c (9,689 dates)
• Alps: p3k14c (2,451 dates)

All dates were calibrated using IntCal20 (Reimer et al. 2020). Site-level binning (200-year bins per site) was applied following Crema & Bevan (2021). SPDs were computed for the range 0--10,000 cal BP.

2.3 Composite AP timeseries

Individual site AP ratios were aggregated into 200-year bins. Only bins with >=3 contributing sites were retained.

2.4 Statistical analysis

1. Pearson correlation between composite AP and regional SPD with autocorrelation correction (Bretherton et al. 1999).
2. Changepoint analysis: exhaustive search for the single breakpoint maximising variance reduction (piecewise constant model). Significance by permutation test (1,000 permutations); 95% CIs by bootstrap (1,000 resamples).
3. Period-specific correlations: separate analyses for pre-agricultural, agricultural, and post-agricultural periods in each region.
4. Agricultural indicator pollen analysis: For each site, taxon-level pollen counts were extracted from Neotoma's variablename field. Agricultural indicators were defined as: Cerealia-type (including Triticum, Hordeum, Secale, Avena -- cereal pollen morphologically distinct from wild Poaceae, following the criteria of Behre 1981: grain diameter >40 um, annulus >8 um); Plantago lanceolata (pastoral/trampling indicator); Rumex, Artemisia, Urtica (disturbance indicators). Each indicator was expressed as percentage of total pollen sum. Regional composites were computed as mean percentages across sites in 200-year bins.

2.5 Bootstrap lag estimation

For each region, two dates were identified from composites: (a) first sustained Cerealia rise above pre-agricultural baseline (>2 SD, remaining elevated), and (b) AP decline changepoint. The lag was computed as their difference. Bootstrap 95% CIs were obtained by resampling sites with replacement (1,000 iterations).

2.6 Site-level replication analysis

Each site was classified by temporal sequence of: (a) first sustained Cerealia-type pollen, and (b) first sustained AP decline >10% from Holocene maximum. Categories: "indicator first" (>200 years), "AP decline first" (>200 years), "simultaneous" (within 200 years), "no clear pattern."

2.7 Temporal smearing test

To test whether the composite lag is an artifact of temporal averaging, we compared Cerealia onset ages between "AP decline first" and "indicator first" sites using the Mann-Whitney U test and Spearman rank correlation. If temporal smearing drives the composite pattern, AP-decline-first sites should have systematically earlier Cerealia onset.

2.8 Threshold analysis

To test whether there is a threshold in Cerealia abundance below which AP decline is disproportionately steep, site-level paired observations of mean Cerealia percentage and mean AP ratio during the agricultural period were modelled with two complementary approaches: (1) a Generalised Additive Model (GAM: AP ratio ~ s(Cerealia%), thin-plate regression spline, mgcv package in R), with the zero-crossing of the first derivative identified as an empirical threshold; and (2) segmented regression with one breakpoint, estimated by maximum likelihood with confidence intervals from the delta method (segmented package in R). Full diagnostic plots and model outputs are in Supplementary Material.

3. Results

3.1 Regional AP patterns: agriculture had no detectable composite-level impact

Britain (46 Holocene sites): The composite AP showed remarkably little change following Neolithic agricultural introduction (~6,000 cal BP). Pre-Neolithic mean AP was 0.816; during the Neolithic-Bronze Age (2,000--6,000 BP) it was 0.769, a decline of only 5.8%. The major decline came after 2,000 BP when AP dropped to 0.568 (-30.5%). SPD-AP correlation was non-significant (r = -0.259, p_corrected = 0.214). Changepoint analysis identified the primary structural break at 3,200 cal BP (Late Bronze Age; CI: 3,200--4,000; permutation p < 0.001; variance reduction 64.9%). Pre-3,200 BP mean AP = 0.812 (forested); post-3,200 BP = 0.597 (deforested).

Scandinavia (185 Holocene sites): Scandinavia showed the highest and most stable AP values. Pre-agriculture (>5,500 BP) mean AP was 0.916; during the agricultural period (2,000--5,500 BP) it actually increased to 0.935 (+2.1%), reflecting continued post-glacial forest expansion. Late-period (<2,000 BP) AP was 0.879. SPD-AP r = -0.224, p_corrected = 0.304 (non-significant; lag-1 autocorrelation: AP 0.926, SPD 0.389; n_eff = 23.0). Changepoint at 800 cal BP (late Medieval; CI: 800--9,600; p = 0.002; VR 35.2%). The wide CI reflects minimal AP variation throughout the Holocene, indicating no well-defined deforestation transition.

Alps (180 Holocene sites): The Alps showed the most dramatic lag. Neolithic agriculture arrived ~7,500 cal BP (LBK culture), yet AP increased by 3.9% during the Neolithic-Bronze Age -- the forest was still expanding. Pre-Neolithic mean AP = 0.838; Neolithic-Bronze Age = 0.871; Iron Age-Medieval = 0.741. SPD-AP r = +0.705 (p = 0.011), meaning population growth and forest cover increased together -- reflecting co-varying Holocene expansion, not a causal relationship. Changepoint at 1,200 cal BP (early Medieval/Carolingian; CI: 1,200--2,400; p < 0.001; VR 81.6%).

Multi-regional comparison:

*Alps positive correlation reflects co-varying Holocene forest expansion and population growth, not an agriculture-deforestation relationship.

3.2 The dual signal: agricultural indicators reveal what AP ratios miss

While AP ratios showed negligible change, agricultural indicator pollen increased dramatically after farming arrived:

Britain (46 Holocene sites):

• Pre-Neolithic (>6,000 BP): Cerealia 0.009%, Plantago 0.072%
• Neolithic-Bronze Age (2,000--6,000 BP): Cerealia 0.132% (x15), Plantago 0.708% (x10)
• Iron Age-Medieval (<2,000 BP): Cerealia 0.501% (x56), Plantago 2.387% (x33)

Scandinavia (185 Holocene sites):

• Pre-agriculture (>5,500 BP): Cerealia 0.003%, Plantago 0.014%
• Agricultural (2,000--5,500 BP): Cerealia 0.044% (x15), Plantago 0.083% (x6)
• Late (<2,000 BP): Cerealia 0.605% (x200), Plantago 0.241% (x17)

Alps (180 Holocene sites):

• Pre-agriculture (>7,500 BP): Cerealia 0.006%, Plantago 0.078%
• Neolithic-Bronze Age (3,000--7,500 BP): Cerealia 0.069% (x11), Plantago 0.186% (x2)
• Iron Age-Medieval (<3,000 BP): Cerealia 1.186% (x195), Plantago 1.340% (x17)

The Cerealia increase was gradual and sustained, rising progressively from x11-15 during the initial agricultural period to x56-200 by the Iron Age and Medieval period. The parallel increase of Plantago lanceolata (x6-10 during the initial agricultural period, rising to x17-33 later) corroborates landscape modification at a scale broader than individual crop fields, since Plantago is associated with pastures, trackways, and disturbed ground extending beyond cultivated plots.

Key finding: The local-regional mismatch -- stable AP alongside x11-15 fold Cerealia increases -- demonstrates scale-dependent detectability. Agriculture changed what grew on the landscape long before it changed how much forest remained.

3.3 Quantitative lag estimation

Table 1. Bootstrap lag estimation: Cerealia rise to AP decline.

The estimated lags range from 3,400 years in Britain to 4,800 years in the Alps. Several features merit attention. First, these are substantially longer than approximate estimates derived from comparing archaeological dates with changepoints (v1.0 estimated ~2,800, ~4,700, and ~6,300 years respectively), because the bootstrap procedure identifies Cerealia rise and AP decline dates directly from the pollen data rather than relying on archaeological chronologies. Second, Britain's wider CI (2,000--4,000) reflects greater variability in both Cerealia appearance and AP decline timing across the heterogeneous landscapes of the British Isles. Scandinavia's tight CI (3,800--5,000) reflects remarkably consistent late AP decline across Scandinavian sites. The Alps CI (3,195--5,000) confirms that even the most conservative estimate exceeds three millennia. Third, the bootstrap valid n values (898--1,000 out of 1,000) confirm robustness. Even the most conservative lower bound across all regions (2,000 years, Britain) exceeds two millennia. The dual-signal phase is not a brief transitional phenomenon but a stable ecological configuration lasting multiple millennia -- the defining quantitative feature of the dual-signal phenomenon.

3.4 Site-level replication

Table 2. Site-level temporal sequence classification.

In all three regions, the dominant site-level pattern is "AP decline first" (55--74%), not "indicator first" (14--23%). This is the reverse of what the composite analysis might lead one to expect. This discrepancy reflects the different phenomena captured at each scale. At individual sites, AP decline detects any local disturbance -- windthrow, fire, drought, insect outbreak, localised human activity -- regardless of whether that disturbance involves agriculture. In a forested landscape subject to multiple disturbance agents, most sites will experience some episode of AP decline during the Holocene that predates the arrival of agriculture. Cerealia detection, by contrast, requires the specific condition of nearby cultivation, with a detection probability constrained by Cerealia's extremely low RPP. The composite analysis filters out stochastic site-level fluctuations by averaging across dozens to hundreds of sites, retaining only the systematic, region-wide agriculture-driven component.

The proportion of "indicator first" sites (14--23%) may represent a lower bound on the fraction of sites located near early agricultural centres, where the dual-signal sequence operates as predicted. The Alps show the highest proportion (23.3%), consistent with their long agricultural history and dense settlement; Scandinavia shows the lowest (14.1%), consistent with dispersed early farming and frequent natural boreal disturbance.

3.5 Temporal smearing test

The reviewer's temporal smearing hypothesis predicts that AP-decline-first sites should have earlier Cerealia onset (agriculture arrived first, allowing time for AP decline before the composite registers the change). The data show the opposite (Table 3).

Table 3. Temporal smearing test results.

Mann-Whitney U test: z = -5.625, p < 0.0001. Spearman rho = 0.269, p < 0.0001. The relationship is significant in all three regions individually (Britain p = 0.015, Scandinavia p = 0.0001, Alps p = 0.001).

Sites with the earliest Cerealia onset are significantly more likely to show the "indicator first" pattern. This is the opposite of temporal smearing: where agriculture arrived earliest, the dual-signal sequence (Cerealia first, AP decline later) is most clearly expressed. Conversely, AP-decline-first sites have later Cerealia onset, consistent with AP decline at these sites reflecting non-agricultural disturbance (windthrow, fire, drought) that preceded the local arrival of agriculture. The later agriculture arrived at a site, the more likely the site had already experienced a non-agricultural AP decline event during its earlier Holocene history.

The consistency of this pattern across all three regions (all p < 0.02) strongly argues against temporal smearing as an explanation for the composite lag. If anything, the site-level data reinforce the genuine nature of the decoupling: the composite multi-millennial lag reflects a real regional-scale phenomenon in which agriculture operated at spatial scales below the detection threshold of regional AP composites.

3.6 Threshold analysis

The threshold analysis tests whether the relationship between Cerealia abundance and AP ratio is nonlinear, with a breakpoint below which AP is most sensitive to Cerealia increase.

Segmented regression identified a statistically significant breakpoint at Cerealia = 0.014% (95% CI: 0.005--0.023; p = 0.026). Below this threshold, the slope of the AP-Cerealia relationship is steep and negative: small increases in Cerealia are associated with substantial AP decline. Above 0.014%, the relationship flattens: further increases in Cerealia produce proportionally less AP decline.

Table 4. Threshold analysis results.

The GAM analysis yields R2 = 0.14, indicating that Cerealia abundance explains approximately 14% of the variance in AP ratio across sites. The discrepancy between the two thresholds (0.014% vs 0.120%) reflects different mathematical definitions: segmented regression identifies maximum slope change, while the GAM zero-crossing identifies where the relationship reverses sign.

The segmented regression breakpoint at 0.014% is interpretable in the context of pollen taphonomy. Given the extremely low RPP of Triticum (~0.0008), a Cerealia percentage of 0.014% corresponds to a substantial area of cereal cultivation -- potentially several percent of the local pollen source area. The steep AP decline below this threshold suggests that the earliest phase of agriculture, when only a few fields exist in the landscape, produces disproportionate local forest disturbance -- perhaps because initial clearings directly replace primary forest, whereas later expansion occurs increasingly on already-open land (secondary grassland, abandoned fields, marshland drainage). Above the threshold, the marginal deforestation cost of each additional field declines because the landscape has already been partially transformed. The low R2 (0.14) is expected given the multiple determinants of AP variation beyond agriculture. Full GAM smooth terms and diagnostic plots are in Supplementary Material.

4. Discussion

4.1 The local-regional mismatch is consistent across all European regions

The central finding is that agricultural indicator pollen rose substantially -- by factors of 11 to 15 for Cerealia-type -- while AP ratios showed no change at the regional composite level in all three regions. This consistency across three independent regions with different agricultural traditions, different chronologies of farming introduction, and different environmental settings substantially strengthens the interpretation. The pattern holds whether agriculture arrived early (Alps, ~7,500 cal BP) or late (Scandinavia, ~5,500 cal BP), whether initial farming was primarily arable (Britain) or mixed agropastoral (Alps), and whether the surrounding vegetation was deciduous broadleaf forest (Britain) or boreal-dominated (Scandinavia). The common denominator is that early farming was conducted at a spatial scale too small to register in the regional AP signal.

Our multi-regional analysis provides large-scale empirical support for the two-phase model of Rosch et al. (2014). Our "dual signal" -- stable AP alongside rising agricultural indicators -- corresponds to Rosch's Phase 1 (garden-scale farming below regional AP detection), while our changepoints correspond to the Phase 1-to-Phase 2 transition. The critical advance is demonstrating that this two-phase structure operates across three regions with fundamentally different agricultural histories. In Britain, Phase 2 began at ~3,200 cal BP (Late Bronze Age); in the Alps, at ~1,200 cal BP (Carolingian expansion); in Scandinavia, at ~800 cal BP (high Medieval). The variation in timing -- spanning more than five millennia -- rules out a common climatic trigger and points to region-specific trajectories of social and technological reorganisation.

The bootstrap lag estimates (3,400--4,800 years) place this finding on firm quantitative footing. These are not approximate estimates from comparing archaeological dates with changepoints; they are bootstrap-derived confidence intervals computed directly from the pollen data. The shortest lower confidence bound (2,000 years, Britain) still exceeds two millennia. This multi-millennial timescale is the defining quantitative feature of the dual-signal phenomenon and demands explanation: what maintained the stable configuration of agriculture-without-deforestation for thousands of years?

4.2 The composite versus individual discrepancy

The site-level analysis reveals that 55--74% of sites show "AP decline first," the reverse of the composite pattern. This is not a contradiction but reflects the different phenomena captured at each scale. At individual sites, AP decline detects any local disturbance -- windthrow, fire, drought, insect outbreaks -- that routinely occurs in forested landscapes (Bradshaw & Hannon 2004). Cerealia detection requires the specific condition of nearby cultivation, with a detection probability constrained by Triticum's RPP of ~0.0008 and dispersal distance of ~500 m (Brostrom et al. 2008).

The temporal smearing test (Section 3.5) directly addresses whether this discrepancy is an artifact. The prediction under temporal smearing is that AP-decline-first sites should have earlier Cerealia onset. The data show the opposite: indicator-first sites have significantly earlier onset (Mann-Whitney p < 0.0001), confirming that sites with the longest agricultural history are most likely to display the dual-signal sequence. The composite lag is a genuine emergent property of regional averaging, not a temporal averaging artifact.

This interpretation has an important implication for how the dual-signal results should be communicated. The claim is not that "at most individual sites, Cerealia appears before AP declines." The data show that this is false for the majority of sites. The claim is that "at the regional composite level, the systematic onset of Cerealia presence precedes the systematic onset of AP decline by 3,400--4,800 years." The site-level analysis demonstrates that this composite-level pattern is an emergent property of regional averaging, not a universal site-level sequence -- and the temporal smearing test demonstrates that this emergent property cannot be explained by differential agricultural arrival times. Understanding why this is so -- differential sensitivity to local vs. regional disturbance; differential detectability of agricultural vs. non-agricultural processes -- strengthens rather than undermines the dual-signal interpretation.

The temporal smearing test also addresses a specific mechanistic prediction. Under the smearing hypothesis, the composite lag should be shorter if it were computed only from sites with similar agricultural arrival times. However, since the Spearman correlation is positive (earlier agriculture correlates with "indicator first" classification), restricting the analysis to early-agriculture sites would increase the proportion of "indicator first" sites and strengthen the composite pattern -- the opposite of what temporal smearing predicts.

4.3 The source-area caveat: what the dual signal does and does not demonstrate

A critical consideration is the differential pollen dispersal of the taxa involved. AP integrates pollen from a regional source area of tens of kilometres (Sugita 2007), while Cerealia-type pollen has effective dispersal of approximately 500 metres. This means a single cultivated field adjacent to a lake could produce a detectable Cerealia signal without causing any measurable change in the regional AP ratio -- not because the forest is resilient to agriculture, but because the field is too small to affect the vegetation average across tens of square kilometres. From this perspective, the dual signal might appear to document a truism about pollen taphonomy rather than a meaningful ecological finding.

We argue, however, that this interpretation is incomplete in several important respects. First, the Cerealia increase across all three regions is gradual and sustained over centuries to millennia, rising progressively from x11-15 during the initial agricultural period to x56-200 by the Iron Age and Medieval period. If the signal reflected merely the chance proximity of a single field to a pollen trap, it would appear as an abrupt step-change at a single site rather than as a progressive regional trend across dozens to hundreds of sites. The gradual increase is consistent with the spatial expansion of farming activity across the landscape over time -- more fields, in more locations, contributing Cerealia pollen to more sites. Second, Plantago lanceolata shows a parallel increase in all regions, corroborating agricultural landscape modification at a scale broader than individual crop fields, since Plantago is associated with pastures, trackways, and disturbed ground that extend beyond cultivated plots. Third, AP does eventually decline in every region, confirming that agriculture ultimately did transform the regional landscape. Fourth, the site-level replication analysis adds a further consideration. The fact that 55--74% of sites show "AP decline first" is consistent with individual sites being subject to local non-agricultural disturbances. But the composite analysis -- averaging out these stochastic effects -- reveals a systematic multi-millennial lag, demonstrating that beneath the site-level noise there is a genuine regional-scale signal of agriculture preceding landscape transformation. The composite is not masking a true "AP decline first" pattern; it is filtering out non-agricultural noise to reveal the agricultural signal.

The dual signal documents the interim phase -- potentially lasting millennia -- during which farming was present and expanding but had not yet reached the spatial scale necessary to shift regional forest cover. The methodological insight is that the AP ratio and agricultural indicator pollen measure fundamentally different things. When AP stability is interpreted as absence of agriculture, the result is a misattribution of scale. The dual-signal approach avoids this conflation. Application of the REVEALS framework (Sugita 2007) to our dataset could further resolve this ambiguity by estimating actual vegetation cover from pollen assemblages, separating the taphonomic component from its ecological component. This is a priority for future work, now technically feasible in Japan following Hayashi et al. (2025).

4.4 What breaks the dual signal?

If the dual-signal phase represents a stable configuration in which agriculture coexists with regional forest cover, the question becomes: what eventually disrupts that configuration and drives the transition to large-scale deforestation? The changepoint dates provide a consistent answer.

In Britain, the primary AP structural break at 3,200 cal BP coincides not with Neolithic agricultural introduction but with the transition from dispersed Neolithic settlement to the nucleated field systems and bronze metalworking of the Late Bronze Age (Bradley 2007) -- a period that Stevens and Fuller (2012) characterised as a genuine "agricultural revolution" distinct from, and far more impactful than, the initial Neolithic adoption of farming. In the Alps, the changepoint at 1,200 cal BP aligns with Carolingian agricultural reform and the expansion of the manorial (Grundherrschaft) system in the early Medieval period -- a regime of monastic land clearance, estate organisation, and deliberate colonisation of forested uplands driven by Frankish imperial policy. In Scandinavia, the remarkably late changepoint at 800 cal BP corresponds to the consolidation of Medieval kingdoms, the expansion of monastic agricultural estates, and the intensification of iron plough agriculture -- forces that combined to drive the clearance of forests that had persisted through five millennia of farming. Our companion study (Author et al., in review) identified Kofun-period state formation (~1,709 cal BP) as the trigger in Japan -- a period defined by monumental tomb construction, iron tool proliferation, and state-directed irrigation -- suggesting the pattern extends beyond European cereal-based systems.

The pattern across all regions is strikingly similar despite vastly different cultural contexts: it is not the introduction of agriculture that drives deforestation, but a subsequent phase of political or technological reorganisation that enables landscape transformation at a scale beyond the capacity of early farming communities. This resonates with Stephens et al. (2019), who argued that significant land-use change did not begin until approximately 3,000 years ago -- well after the initial spread of agriculture. In the framework of Rosch et al. (2014), these changepoints mark the transition from Phase 1 (garden-scale agriculture below regional AP detection) to Phase 2 (extensive, landscape-transforming agriculture) -- a transition driven not by inherent agricultural dynamics but by exogenous organisational change.

Large-scale deforestation required organisational capacity -- the ability to mobilise hundreds or thousands of people in clearing, draining, terracing, and maintaining open land -- that exceeded the capabilities of early farming communities (cf. Borrell et al. 2021). This capacity is a property of states, chiefdoms, and ecclesiastical institutions. The dual-signal phase, in this interpretation, is the interval between the arrival of the agricultural precondition and the emergence of the organisational capacity that converts it into regional landscape transformation. The variation in lag duration across regions (3,400--4,800 years) records the different rates at which societies developed this capacity: Britain's relatively short lag (~3,400 years; CI: 2,000--4,000) may reflect the early development of Bronze Age social complexity and agricultural intensification in the British Isles. The Alps' long lag (~4,800 years; CI: 3,195--5,000) reflects the persistence of relatively low-density farming in a mountainous landscape where large-scale clearance was both ecologically challenging and politically unnecessary until the expansion of Medieval states. Scandinavia's lag (~4,400 years; CI: 3,800--5,000) reflects both the boreal environment -- where forest regeneration is rapid and farming is constrained by climate -- and the late emergence of centralised political authority in the Nordic world. These variations are not noise; they record the different rates at which societies in different environments developed the organisational capacity for landscape-scale ecological transformation.

4.5 Climate confounds

Two of our changepoint dates coincide with known climatic transitions, and these potential confounds must be acknowledged. The British changepoint at 3,200 cal BP falls within the period of climate deterioration in the late Bronze Age, documented by Dark (2006) among others. Whether the Late Bronze Age "agricultural revolution" (Stevens & Fuller 2012) was itself partly a response to climate stress -- or whether climate deterioration independently contributed to AP decline through shifts in forest composition -- cannot be resolved from pollen data alone. Similarly, the Scandinavian changepoint at 800 cal BP falls near the Medieval Warm Period (MWP) to Little Ice Age (LIA) transition (~700--550 cal BP). While the 800 BP date slightly precedes the LIA onset, the proximity raises the question of whether the late Medieval AP decline in Scandinavia reflects agricultural intensification, climate-driven vegetation change, or both. The Alps changepoint at 1,200 cal BP does not coincide with any major known climatic transition, providing the cleanest test of the organisational hypothesis. We note that climate confounds apply equally to conventional AP-only analyses and do not specifically undermine the dual-signal approach, which adds information (indicator pollen timing) rather than replacing existing proxies. Nevertheless, disentangling climatic and anthropogenic drivers of AP decline at these specific transition points remains an unresolved challenge.

4.6 Methodological implications

Our results carry direct implications for how pollen records are used to reconstruct past human impact on vegetation. The most immediate is that the AP ratio, used in isolation, measures landscape-scale forest transformation and is insensitive to agricultural activity below that spatial threshold. In all three European regions examined, agriculture was present and expanding for centuries to millennia before the AP ratio registered any decline. Studies that rely exclusively on AP decline to date the onset of agricultural impact -- including many site-level studies and regional syntheses -- conflate two distinct phenomena: the timing of agricultural presence and the timing of landscape-scale forest transformation. The AP ratio answers the latter question, not the former. The dual-signal approach demonstrated here -- simultaneous tracking of AP ratios and taxon-specific agricultural indicators -- should become standard practice in pollen-based assessments of agricultural impact.

This recommendation is not new in principle. European palynologists routinely report Cerealia-type and other anthropogenic indicator pollen alongside AP ratios in site-level studies -- the foundational criteria for identifying Cerealia-type pollen were established by Behre (1981) over four decades ago. What our analysis adds is a multi-site demonstration of the scale of the discrepancy: across hundreds of sites in three regions, agricultural indicator pollen increased by one to two orders of magnitude while the AP ratio remained functionally unchanged. This is not a subtle effect visible only in high-resolution cores adjacent to known settlements; it is a robust continental-scale pattern detectable in regional composites.

This recommendation also aligns with the LUP index of Deza-Araujo et al. (2022), which aggregates cultural indicators with bioindication-weighted values to reconstruct agricultural land-use intensity across all major European biomes. Our approach and the LUP index address different but complementary questions: the LUP provides a composite metric of agricultural land-use probability at a given time, while the dual-signal framework quantifies the temporal duration of the decoupling between local agricultural presence and regional landscape change. The LUP index confirms that indicator-based methods detect agricultural activity invisible to AP ratios; our framework explains why this discrepancy exists and demonstrates that it persists for millennia across diverse regions. The two approaches are mutually reinforcing.

A second methodological implication concerns the application of pollen-landscape reconstruction models. The REVEALS framework (Sugita 2007) could estimate the actual area of cultivated land required to produce the observed Cerealia percentages and determine whether the predicted AP response is consistent with the observed AP stability. If REVEALS predicts that the observed Cerealia levels require only a small fraction of the landscape to be under cultivation, this would confirm that AP stability during the dual-signal phase reflects genuinely limited spatial extent of farming. The REVEALS application represents a natural next step, now technically feasible in Japan following Hayashi et al. (2025).

A further implication concerns the interpretation of individual site records. An individual site showing AP decline before Cerealia appearance should not be interpreted as evidence against early agriculture; it more likely reflects local non-agricultural disturbance. Conversely, an individual site showing Cerealia appearance before AP decline is stronger evidence of the dual signal, since Cerealia detection requires the specific condition of nearby cultivation. Future site-level studies should explicitly acknowledge the stochastic nature of both AP variation and Cerealia detection at single sites, and should be cautious about drawing regional-scale conclusions from individual temporal sequences.

Finally, our findings underscore the importance of multi-proxy approaches in palaeoecological research. Charcoal records, non-pollen palynomorphs (such as coprophilous fungal spores indicating livestock presence), sediment geochemistry, and archaeological site distributions each capture different aspects of the human ecological footprint at different spatial and temporal scales. The dual-signal approach is, in essence, an argument for matching the proxy to the question -- and for recognising that no single proxy answers all questions about past human impact on the landscape.

4.7 Japan and rice-based systems

Our companion study (Author et al., in review) found a 700-year lag between wet-rice introduction (~2,580 cal BP) and major AP decline at Japanese sites, with the transition to landscape-scale deforestation occurring during Kofun-period state formation (~1,709 cal BP). Partial dual-signal verification at Morinji Marsh (Leipe et al. 2024) showed Cerealia-type pollen increasing by a factor of 1.4 during the Yayoi period while AP simultaneously increased by 11.6% -- the same qualitative pattern of rising agricultural indicators alongside stable or increasing forest cover that characterises the European dual-signal phase. The much smaller magnitude of the Cerealia increase (x1.4 vs x11-15 in Europe) is consistent with rice's extremely low RPP: Oryza sativa is autogamous (self-pollinating) and produces minimal airborne pollen, meaning that a x1.4 increase may represent a substantial expansion of cultivated area that is severely underrepresented in the pollen record.

However, these results rest on a single site with Cerealia-type resolution. Neotoma's Japanese Pollen Database lacks Cerealia-type classification for most sites, reporting only undifferentiated Poaceae. Systematic dual-signal analysis in Japan awaits taxonomic standardisation in global pollen databases -- a growing imperative as regional databases are integrated into platforms such as Neotoma. The Japanese results constitute encouraging preliminary evidence that the dual signal extends to rice-based agricultural systems in East Asia, but replication at additional sites is needed before this can be considered robust.

4.8 Caveats and limitations

Raw pollen percentages. Our analysis uses raw percentages rather than REVEALS-corrected estimates. Given extreme RPP differences between arboreal taxa and Cerealia, the observed x11-15 increases almost certainly understate true cultivation expansion.

SPD as demographic proxy. SPDs are subject to calibration curve effects and taphonomic loss (Carleton & Groucutt 2021). SPD-AP correlations should be interpreted as associations between broad demographic trends and vegetation change.

Alps chronological control. The Alps region has the lowest 14C dating density (2,451 dates vs 19,578 for Britain), and consequently the poorest chronological control. The 4,800-year lag should be treated with greater caution, though the bootstrap CI (3,195--5,000) confirms a multi-millennial lag even under conservative assumptions.

Composite averaging. The regional composite approach averages AP ratios and indicator percentages across many sites, which can mask site-level variability. Sites near early agricultural settlements may show stronger dual signals than the composite suggests, while sites far from agricultural centres dilute the composite signal. The consistent multi-regional pattern mitigates but does not eliminate this concern.

Site-level classification thresholds. Different definitions of "sustained Cerealia" or "sustained AP decline" (e.g., 5% vs. 10% AP decline; one vs. two subsequent Cerealia-positive samples) would alter Table 2 proportions. The qualitative finding -- majority AP decline first -- is robust to reasonable variations.

Threshold model R2. The GAM R2 of 0.14 indicates that Cerealia explains only a small fraction of site-level AP variance. While the segmented regression breakpoint is significant (p = 0.026), the practical predictive power at individual sites is limited. The threshold analysis identifies a real but noisy relationship, consistent with the multiple determinants of AP variation.

5. Conclusions

This study demonstrates that the arboreal pollen ratio measures landscape-scale forest transformation, not the presence or absence of agricultural activity. By integrating AP ratios with agricultural indicator pollen across 676 sites in three European regions, we provide multi-regional quantitative evidence that the onset of agriculture and the onset of AP-detectable landscape change are decoupled by multiple millennia.

Bootstrap lag estimation quantifies the decoupling at 3,400 years in Britain (95% CI: 2,000--4,000), 4,400 years in Scandinavia (CI: 3,800--5,000), and 4,800 years in the Alps (CI: 3,195--5,000). Even the most conservative estimate exceeds two millennia.

Site-level replication shows that 55--74% of sites display "AP decline first," but the temporal smearing test demonstrates this is not an artifact: sites with the earliest agriculture are significantly more likely to show the expected "indicator first" sequence (Spearman rho = 0.269, p < 0.0001), confirming that the composite lag reflects genuine scale-dependent detectability rather than temporal averaging.

What ultimately broke the dual signal was not agricultural expansion per se but political and technological reorganisation: Late Bronze Age intensification in Britain (~3,200 cal BP), Carolingian reform in the Alps (~1,200 cal BP), and Medieval state consolidation in Scandinavia (~800 cal BP). The variation in timing -- spanning more than five millennia across regions -- rules out a common climatic trigger, although the coincidence of the British changepoint with late Bronze Age climate deterioration and the Scandinavian changepoint with the MWP-LIA transition represent unresolved confounds that merit targeted investigation. In Japan, Kofun-period state formation (~1,709 cal BP) drove the equivalent transition, suggesting the pattern may extend beyond European cereal-based systems to rice-based agriculture in East Asia.

These findings provide the first systematic multi-regional quantification of the two-phase model proposed by Rosch et al. (2014) for southwestern Germany, with bootstrap confidence intervals, site-level replication assessment, temporal smearing falsification, and threshold analysis. The dual-signal phase corresponds to Rosch's Phase 1 (garden-scale agriculture below regional AP detection), and the changepoints marking AP decline correspond to the Phase 1-to-Phase 2 transition. Our approach complements the LUP index of Deza-Araujo et al. (2022) by quantifying not just the presence of the agricultural-AP discrepancy but its multi-millennial duration across diverse regions.

The study carries a direct methodological recommendation: the AP ratio measures landscape-scale forest transformation, and when used in isolation it cannot distinguish between the absence of agriculture and the presence of agriculture at spatial scales below the detection threshold of regional pollen composites. Agricultural indicator pollen -- particularly Cerealia-type and Plantago lanceolata -- should be routinely reported alongside AP ratios to distinguish the timing of agricultural presence from the timing of landscape transformation. Our approach complements the LUP index of Deza-Araujo et al. (2022) by quantifying not just the presence of the agricultural-AP discrepancy but its multi-millennial duration across diverse regions.

Future work should apply REVEALS-based quantitative vegetation reconstruction -- now feasible in Japan following Hayashi et al. (2025) -- to disentangle the taphonomic and ecological components of the dual signal, and should investigate whether the composite-individual discrepancy observed here generalises to other continental-scale pollen syntheses. The growing integration of regional pollen databases into global platforms such as Neotoma creates both an opportunity and an imperative to standardise the taxonomic resolution of agriculturally significant taxa, enabling systematic dual-signal analysis in regions -- particularly East Asia -- where Cerealia-type classification is currently unavailable.

Supplementary Material

S1. GAM and threshold analysis (full results)

The GAM (AP ratio ~ s(Cerealia%)) yields R2 = 0.14 with a smooth zero-crossing at 0.120% Cerealia. Individual smooth terms were non-significant, reflecting high noise. The segmented regression breakpoint at 0.014% (95% CI: 0.005--0.023; p = 0.026) identifies where the slope changes maximally. The discrepancy between the two thresholds (0.014% vs 0.120%) reflects different mathematical definitions: segmented regression identifies maximum slope change; the GAM zero-crossing identifies where the relationship reverses sign. The low R2 is expected given multiple determinants of AP variation beyond agriculture. These results confirm a real but noisy nonlinear relationship consistent with initial clearings directly replacing forest.

S2. Period-specific correlations

Separate SPD-AP correlations for pre-agricultural, agricultural, and post-agricultural periods show no significant relationship during the dual-signal phase in Britain or Scandinavia, with a moderate negative correlation (r = -0.425) appearing only during the British Iron Age-Medieval period. This pattern is consistent with the dual-signal interpretation: during the dual-signal phase, population growth (reflected in rising SPD) and AP stability coexist because farming operates at spatial scales too small to affect the regional AP signal. Only after the Phase 2 transition does the expected negative SPD-AP relationship emerge.

S3. Temporal smearing test (full methodology)

The temporal smearing hypothesis (reviewer concern) proposes that if agriculture arrived at different times at different sites, averaging site-level AP declines will artificially delay the apparent onset of composite deforestation. We test this by comparing Cerealia onset ages (cal BP) between two site classifications.

Prediction under temporal smearing: AP-decline-first sites should have earlier (older) Cerealia onset, because agriculture arrived first at those sites, providing time for AP decline before the composite registered the change.

Prediction under genuine decoupling: No systematic relationship between Cerealia onset and site classification.

Results: n = 438 sites with both classification and Cerealia onset data. AP-decline-first sites (n = 307) have mean onset of 3,726 cal BP; indicator-first sites (n = 131) have mean onset of 5,673 cal BP. The direction is opposite to the temporal smearing prediction: sites with earlier agriculture are more likely to show the "indicator first" pattern. Mann-Whitney z = -5.625, p < 0.0001. Spearman rho = 0.269, p < 0.0001. The result is consistent across all three regions individually (Britain p = 0.015, Scandinavia p = 0.0001, Alps p = 0.001). Rank-biserial effect size r = 0.339 (medium effect).

S4. Sensitivity of site-level classification

The classification of individual sites into "indicator first" vs "AP decline first" depends on definitions of "sustained Cerealia" (first sample with Cerealia > 0 followed by at least one additional positive sample within 500 years) and "sustained AP decline" (>10% from Holocene maximum). We note that many "AP decline first" sites show AP decline without any Cerealia detection (AP decline without Cerealia), representing sites where non-agricultural disturbance caused AP fluctuations and where no nearby cereal cultivation was ever within detection range. These sites dilute the proportion of "indicator first" sites but do not contradict the dual-signal interpretation; rather, they illustrate the low spatial detection probability of Cerealia pollen given its RPP of ~0.0008.

References

• Alenius, T., Marquer, L., Molinari, C., Heikkila, M., Ojala, A. 2021. The environment they lived in: anthropogenic changes in local and regional vegetation composition in eastern Fennoscandia during the Neolithic. Vegetation History and Archaeobotany 30:489-506.
• Author et al. (in review). Two-Stage Deforestation at Hosoike Moor: Agriculture and State Formation as Sequential Drivers of Forest Clearance in Western Japan.
• Behre, K.-E. 1981. The interpretation of anthropogenic indicators in pollen diagrams. Pollen et Spores 23:225-245.
• Bogaard, A. 2004. Neolithic Farming in Central Europe: An Archaeobotanical Study of Crop Husbandry Practices. Routledge, London.
• Borrell, F. et al. 2021. The expansion of farming in western Eurasia. Journal of World Prehistory 34:415-459.
• Bradley, R. 2007. The Prehistory of Britain and Ireland. Cambridge University Press.
• Bradshaw, R.H.W. & Hannon, G.E. 2004. The Holocene structure of north-west European temperate forest induced from palaeoecological data. CABI Publishing.
• Bretherton, C.S. et al. 1999. The effective number of spatial degrees of freedom of a time-varying field. Journal of Climate 12:1990-2009.
• Brostrom, A. et al. 2008. Pollen productivity estimates of key European plant taxa for quantitative reconstruction of past vegetation: a review. Vegetation History and Archaeobotany 17:461-478.
• Carleton, W.C. & Groucutt, H.S. 2021. Sum things are not what they seem: Problems with point-wise interpretations and quantitative analyses of proxies based on aggregated radiocarbon dates. The Holocene 31:1290-1307.
• Crema, E.R. 2022. Statistical inference of prehistoric demography from frequency distributions of radiocarbon dates: a review and a guide for the perplexed. Journal of Archaeological Method and Theory 29:1387-1418.
• Crema, E.R. & Bevan, A. 2021. Inference from large sets of radiocarbon dates. Radiocarbon 63:23-39.
• Dark, P. 2006. Climate deterioration and land-use change in the first millennium BC: perspectives from the British palynological record. Journal of Archaeological Science 33:1381-1395.
• Deza-Araujo, M. et al. 2022. A new indicator approach to reconstruct agricultural land use in Europe from sedimentary pollen assemblages. Palaeogeography, Palaeoclimatology, Palaeoecology 599:111051.
• Fyfe, R.M. et al. 2015. From forest to farmland: pollen-inferred land cover change across Europe using the pseudobiomization approach. Global Change Biology 21:1197-1212.
• Kaplan, J.O. et al. 2009. Holocene carbon emissions as a result of anthropogenic land cover change. The Holocene 19:749-756.
• Leipe, C. et al. 2024. Human activities, early farming and natural environment in the north-western Kanto Plain (central Japan) reconstructed from multi-proxy analysis of sediments from Morinji Marsh. Quaternary Environments and Humans 2:100030.
• Reimer, P.J. et al. 2020. IntCal20. Radiocarbon 62:725-757.
• Roberts, N. et al. 2018. Europe's lost forests: a pollen-based synthesis for the last 11,000 years. Scientific Reports 8:716.
• Rosch, M. et al. 2014. Botanical off-site and on-site data as indicators of different land use systems. Vegetation History and Archaeobotany 23(Suppl 1):121-133.
• Shennan, S. et al. 2013. Regional population collapse followed initial agriculture booms in mid-Holocene Europe. Nature Communications 4:2486.
• Stephens, L. et al. 2019. Archaeological assessment reveals Earth's early transformation through land use. Science 365:897-902.
• Stevens, C.J. & Fuller, D.Q. 2012. Did Neolithic farming fail? The case for a Bronze Age agricultural revolution in the British Isles. Antiquity 86:707-722.
• Sugita, S. 2007. Theory of quantitative reconstruction of vegetation I: pollen from large sites REVEALS regional vegetation composition. The Holocene 17:229-241.
• Hayashi, R. et al. 2025. Testing the REVEALS-based vegetation reconstruction around Lake Biwa, western Japan. The Holocene. doi:10.1177/09596836251366195.
• Lechterbeck, J. et al. 2014. Is Neolithic land use correlated with demography? The Holocene 24:1297-1307.
• Rosch, M. & Lechterbeck, J. 2016. Seven millennia of human impact as reflected in a high resolution pollen profile from the profundal sediments of Litzelsee, Lake Constance region, Germany. Vegetation History and Archaeobotany 25:339-358.
• Woodbridge, J. et al. 2014. The impact of the Neolithic agricultural transition in Britain. Journal of Archaeological Science 51:216-224.