A critical assessment of human-impact indices based on anthropogenic pollen indicators

Anthropogenic pollen indicators in pollen records are an established tool for reconstructing the history of human impacts on vegetation and landscapes. They are also used to disentangle the in ﬂ uence of human activities and climatic variability on ecosystems. The comprehensive anthropogenic pollen-indicator approach developed by Behre (1981) has been widely used, including beyond its original geographical scope of Central and Western Europe. Uncritical adoption of this approach for other areas is risky because adventives (plants introduced with agriculture) in Central Europe can be apophytes (native plants favoured by human disturbances) in other regions. This problem can be addressed by identifying region- speci ﬁ c, anthropogenic-indicator pollen types and/or developing region-speci ﬁ c, human-impact indices from pollen assemblages. However, understanding of regional variation in the timing and in- tensity of human impacts is limited by the lack of standardization, validation and intercomparison of such regional approaches. Here we review the most common European anthropogenic pollen-indicator approaches to assess their performance at six sites spanning a continental gradient over the boreal, temperate and Mediterranean biomes. Speci ﬁ cally, we evaluate the human-indicator approaches by using independent archaeological evidence and models. We present new insights into how these methodologies can assist in the interpretation of pollen records as well as into how a careful selection of pollen types and/or indices according to the speci ﬁ c geographical scope of each study is key to obtain meaningful reconstructions of anthropogenic activity through time. The evaluated approaches generally perform better in the regions for which they were developed. However, we ﬁ nd marked differences in their capacity to identify human impact, while some approaches do not perform well even in the regions for which they were developed, others might be used, with due caution, outside their original areas or biomes. We conclude that alongside the increasing wealth of pollen datasets a need to develop novel tools may assist numeric human impact reconstructions.


Introduction
Palaeoecology provides valuable records of past ecological change and its drivers over centennial to millennial timescales at decadal to centennial resolution. Environmental drivers of ecosystem change (e.g. climate, human activities, natural disturbances) may operate simultaneously and thus it may be difficult to disentangle their specific role (Nelson et al., 2006). In Europe, human activities became a major driver for landscape dynamics, land-cover change, species distributions and disturbance regimes as early as ca. 9000-7000 years ago with the onset of farming at the beginning of the Neolithic. As a result, the transformation of forests into heathlands (e.g. Calluna vulgaris), shrublands (e.g. Corylus avellana, Alnus viridis), maquis, garrigue, grasslands or meadows may have resulted from humaninduced deforestation including excessive fire disturbance and/or browsing (Gobet et al., 2000;Tinner et al., 2005;Carrión et al., 2010;Rey et al., 2019). These examples illustrate the role of discrete disturbance events and highlight the need for high-resolution reconstructions of past land use and environmental change to disentangle anthropogenic and natural forcing. The contribution of the various forcing factors may be assessed using multi-proxy palaeoecological studies providing independent lines of evidence (Birks and Birks, 2006;Colombaroli et al., 2007), where the results are ideally validated with local archaeological evidence (Hjelle et al., 2012).
Vegetation changes inferred from palynological sequences have traditionally been linked to climate change when occurring more or less synchronously over broad areas (Jalut et al., 2009), but this assumption might result in overlooking the role of broad-scale concurrent human activities (Tinner et al., 2013;Walsh et al., 2019). Previous research has also shown that climate change may exert a strong influence on land use, leading to synchronous patterns over wide areas (Gobet et al., 2003;Tinner et al., 2003;Oliver and Morecroft, 2014). For instance, widespread forest opening reconstructed from several Mediterranean pollen records for the past ca. 7000 years has been attributed to either a continental-scale decrease in moisture availability ("aridification" hypothesis; see Jalut et al., 2009;Sadori et al., 2011), increasing human activity, including burning (e.g. Tinner et al., 2009;Bisculm et al., 2012) or a combination of both factors (e.g. Carrión et al., 2010). This controversial Mediterranean example illustrates challenges in unambiguously inferring anthropogenic impacts.
Reconstructing human impacts on the environment using palynology has largely relied on the presence and abundance of anthropogenic pollen indicators. Pollen of adventives (i.e. plant species not native to a specific area) track intentional or unintentional (i.e. cultivated crops vs. weeds) introductions by humans and are therefore considered to be reliable indicators for past human activities (Behre, 1981(Behre, , 1988Huntley and Webb, 1988). Although with less diagnostic capacity than adventives, the pollen of apophytes (i.e. native plant species favoured by human activities) also provide information regarding anthropogenic impacts on the landscape (Behre, 1981;Lang, 1994).
Off-site and on-site palaeorecords (e.g. pollen, macrofossils, megafossils, aDNA), in combination with archaeology, provide the only unambiguous evidence for determining the native ranges of cultivated plants and weeds. Such evidence has shown that depending on the region of interest, many plant species may be regarded either as adventives or apophytes (di Castri et al., 1990;Lang, 1994;Conedera et al., 2004;van Leeuwen et al., 2008;Krebs et al., 2019). In the case of palynology, data interpretation requires a strong background knowledge of the processes controlling pollen, spore and other microfossil production, dispersal and preservation (Webb and Goodenough, 2018). This condition adds further complexity to the inference of human impacts from palynological data (Behre, 1981), especially in a quantitative manner, and limits the use of pollen-inferred reconstructions of land-use history by a broader community.
Early applications of the pollen-indicator approach used pollen from plants particularly sensitive to winter frost for palaeoclimatic reconstructions (Iversen, 1944). This methodology has later been extensively applied to land-use reconstruction. Behre (1981Behre ( , 1988) assembled lists of reliable anthropogenic pollen indicators for Central and Western Europe (north of the Alps). Behre's pioneering work was later extended to other areas (e.g. the Middle East, Behre, 1990;China, Li et al., 2015;Mexico, Franco-Gaviria et al., 2018) and refined in Western Europe (Mazier et al., 2006;Brun, 2011). Moreover, Behre's comprehensive account of anthropogenic pollen indicators has been widely used in European areas outside of its original calibration area (e.g. Novenko et al., 2017;Cartier et al., 2018;Fredh et al., 2018;López-Sáez et al., 2018), although the chosen indicator taxa may not be necessarily suitable in these regions (apophyte vs. adventive problem; Moore et al., 1991;Lang, 1994).

6
The presence and abundance of anthropogenic pollen indicators provide valuable evidence for the occurrence and intensity of past land use (Behre, 1981). The sum of the percentages of these pollen types is often plotted separately in pollen diagrams as curves of Principal and/or Secondary Indicators ('PI', 'SI';e.g. Lang, 1994). In contrast to other fields of palynology (e.g. treeline studies), absolute values such as influx are less often used for human-impact reconstructions (Koff and Punning, 2002).
Complementarily, the 'AP/NAP' ratio between the percentages of arboreal (sum of trees and shrubs) and non-arboreal (sum of upland herbs) pollen has long been used to reconstruct changes in forest cover quantitatively, including anthropogenic clearance of forests (Aario, 1944;Berglund et al., 1991).
Conventionally, the interpretation of AP/NAP is straightforward; very high (>10, corresponding to ca. 91% AP), intermediate (>4, 80% AP) and low (<1, 50% AP) values, represent very closed forests, semi-closed forests and open vegetation, respectively (Mitchell, 2005;Favre et al., 2008;Zanon et al., 2018). However, detailed interpretation of intermediate values of this ratio remains unclear (Favre et al., 2008). When AP/NAP is combined with Behre's cultural indicators, it is possible to infer the beginning of farming-induced forest opening in Europe (e.g. Lang, 1994;Rey et al., 2019;. In summary, although the potential of anthropogenic pollen indicators to reconstruct the impact of human activities on past vegetation dynamics has long been recognized and broadly used, no comparative evaluation of their performance is available yet. The emphasis of previous studies (e.g. Steckhan, 1961;Turner, 1964;Kramm, 1978;Riezebos and Slotboom, 1978;Behre, 1981;Lang, 1994;Koff and Punning, 2002;Tinner et al., 2003;Mercuri et al., 2013 a, b;Kouli, 2015;Berger et al. 2019) on local to regional approaches is due to the different vegetation (e.g. biomes) and landuse (e.g. farming) conditions in space and through time. The aim of this study is to provide an overview of the existing methodologies and to understand their relative advantages and limitations. Here we apply the most commonly used human-indicator species approaches to six study sites distributed over a wide latitudinal gradient across Europe, spanning from boreal to mediterranean ecosystems. To evaluate their performance we compare the palynologically-inferred human impacts with archaeological evidence. A specific aim of this study is to identify the performance of the chosen approaches inside and outside their region or biome of origin. Finally, we briefly address potential future avenues in the field, including validation using archaeological data, the value of taxonomically highly resolved records and the difficulty to produce generalized and standardized approaches that may identify human impact at the continental scale.

2.1.Study sites and anthropogenic pollen indicator indices
We applied eight previously developed anthropogenic pollen-indicator approaches (PI, SI, CI, API, OJC, OJCV, PDI and AP/NAP ratio, see Table 1 for details), to six post-glacial lake pollen sequences with high time-resolution and spanning a latitudinal and ecological gradient from Scandinavia to Sicily (Table 2, Figure 1). The pollen datasets were obtained from the European Pollen Database (EPD; Fyfe et al., 2009;Giesecke et al., 2014) and the Alpine Pollen Database (ALPADABA) via Neotoma (Williams et al., 2018). We harmonized pollen nomenclature for consistency amongst the different sequences (Giesecke et al, 2019), and calculated pollen percentages with respect to the terrestrial pollen sum, i.e. excluding pollen from aquatic and wetland plants and spores. We processed the data with the R packages 'Neotoma' version 1.7.4 (Goring et al., 2015) and 'Rioja' version 0.9-15.1 (Juggins, 2017) running in R environment (R Core Team, 2018). We then plotted the obtained values for the various pollen indices at each site as stratigraphic-pollen diagrams against age in calibrated years, using the age-depth models published by the authors in the original papers and stored in the databases (except for Lago di Origlio, where the calibration curve has been updated).

2.2.Archaeological validation framework and its limits
We validated the capacity of the indices to express quantitatively the human impact by comparing the results with the regional archaeological record at each study site. To convert the archaeological evidence into a quantitative scale, we grouped the main archaeological and historical periods according to five stages of human impacts on the European environment: (1) very low/non detectable To delimit the timeframe of each human-impact stage at each site, we synchronized the reference chronologies of the main, well-established archaeological and historical periods in Europe in a regional scheme (Figure 3). While the definition of historical periods is established by events that are documented in historical sources (i.e. precisely dated), archaeological epochs are based on regionally different typo-chronological changes in material culture (i.e. classification of objects and architecture; Besserman, 1996;Shackley, 2001;Carson, 2016). When absolute dates (e.g. radiocarbon, dendrochronology) were available in the literature, we used them for the proposed chronologies (e.g. Knutsson and Knutsson, 2003;Bietti Sestieri, 2013a, 2013bCapuzzo et al., 2014;Lo Vetro and Martini, 2016;Pacciarelli et al., 2016;Natali and Forgia, 2018;Stöckli, 2016;Radi and Petrinelli Pannocchia, 2018;Alessandri, 2019). Conversely, we referred to relative dating where no absolute chronologies were available. As the limits of the different archaeological periods are not always supported by radiocarbon or dendrochronological dating, the presented supra-regional synchronization of periods is tentative and some period boundaries are uncertain. Furthermore, as Neolithisation is not a one-time event but rather a long and complex process, it is often difficult to assign precise dates to the boundary between the Mesolithic and the Neolithic in each region (Dolukhanov et al., 2005;Gronenborn, 2005;Lemmen et al., 2011). Moreover, Mesolithic and Neolithic lifestyles may have coexisted simultaneously for several centuries in some regions (as e.g. proposed for Poland; Nowak, 2013). Nevertheless, the temporal resolution of our approach refers to long prehistorical and historical periods, which partially offsets possible chronological errors.

2.3.Performance of the indices
To evaluate the performance of the indices, we compared the results at the different study sites according to the technological stages ( Figure 3). To make the results comparable, we first rescaled the values of the different indices between 0 and 1 using a minimax transformation: where Z i is the minimax-transformed value of the index V for the -th sample of a given record (V i ), and V max and V min are the maximum and minimum values of V in the entire sequence. Secondly, we averaged the values of every index for each chronological human-impact stage and plotted boxplots to visually assess the trends in the values of the indices. We hypothesize that the magnitude of human impact on the environment will monotonically increase towards present across the stages of population and technological development of European societies during the Holocene as inferred from archaeological and historical evidence ( Figure 2). To test this hypothesis, we ran pairwise comparisons between the five considered human-impact stages on the environment (Figures 2, 3) using the nonparametric Mann-Whitney U test (Wilcoxon rank-sum test), whose null hypothesis is that the two samples of the pairwise comparisons come from the same population.

Results
In northern and north-central Europe, the primary anthropogenic indicators index (PI) only increases during the last 500-1000 years at Holtjärnen (southern Sweden) and Lake Gosciaz ( only present during the last millennium. In contrast, the CI record at Lake Gosciaz starts during the Neolithic and is continuous until the present, with conspicuous increases at the end of the Bronze Age, and substantially increases during the Iron Age and the last millennium. PDI mirrors the main trends of CI despite its nearly continuous record throughout the Lake Gosciaz sequence (Figure 4).
In Central Europe, the records of PI and CI from Burgäschisee (Swiss Plateau) suggest that the first noticeable human impact occurred during the Neolithic (ca. 4500 cal BC), with remarkable stepwise increases from the Iron Age onwards and particularly during the past millennium ( Figure 4c).

4.1.Overall suitability of the anthropogenic indicators
In this study we tested the suitability of widely used palynological indices based on anthropogenic pollen indicators for different European biomes. We compared the results obtained by the indices with human-impact stages, as derived from human population density growth and the related land-use caused by the increasing carrying capacity of agriculture, throughout different historical epochs (Kremer, 1993;Lemmen et al., 2011;US Census Bureau, 2018). The analysed indices do not always show a general increase since the Neolithic and they also show disagreements among them on the extent and timing of land-use-related human impacts. In contrast, the human population experienced linear, exponential or logarithmic growth through the millennia with only episodic interruptions or reversed trends during environmentally-caused production crises such as volcanic eruptions, mass migrations, and climatic reversals (e.g. Little Ice Age; Bentley, 2013). In this context, we identify indices as the lowest performing if they show trends that are opposite to those expected from the prehistorical and historical evidence used to model human population dynamics ( Figure 5). For instance, such dissimilarities occur in the case of high SI and API values before the Neolithic (i.e., stage 1) at several sites and in AP/NAP at all sites. Likewise, unexpected decreases in OJC and OJCV at Gorgo Basso between stages 3, 4, and 5 and in PDI, between stages 2 and 3 are seemingly inconsistent with agricultural intensification ( Figure 5). Best performance in terms of monotonic increasing trends across the human-impact stages are generally provided by indices using few, specific pollen indicators such as PI and CI. Although both PI and CI performed best among all indices at Gorgo Basso, they did not fully match population growth expectations. Similarly, SI has a good performance from the second human-impact stage onwards (except in the southern sites; Figure 5).
Changes in PI at Holtjärnen are limited to human-impact stage 5, which prevents an assessment of increasing trends ( Figure 5), but likely reflects the remoteness of the northernmost site in regard to arable farming.
In general, pollen indicator approaches perform best in the regions in which they were developed (Figures 4 and 5). Specifically, most of the considered indicator species approaches show increasing human impacts between sequential human-impact stages outside the Mediterranean realm (e.g. PI and CI at Burgäschisee and Lago di Origlio; PI at Gosciaz), where this technique was originally developed (Behre, 1981), plant diversity is lower (Mutke et al., 2010), and wild relatives of (south-west Asian Neolithic) crops and weeds are rare (Zohary et al., 2012). Nevertheless, certain indices may perform well in regions different to the one where they were conceived. For instance, the PDI pastoral index developed in northern Greece performed reasonably well in tracking anthropogenic vegetation change in Holtjärnen (southern Sweden), particularly from the Iron Age onwards ( Figure   4). Such episodic good performances may however be coincidental. Indeed, the Holtjärnen record also provides reason for caution. An early incidence of forest opening at Holtjärnen dated around 3700 cal 14 BC, which caused PDI to increase, probably resulted from a shift in atmospheric circulation that naturally affected forests, and not from human impacts (Hammarlund et al., 2002;Giesecke, 2005).
A given index may vary in performance among different human-impact stages when applied along the entire Holocene (e.g. Behre, 1981). In most of our cases, the AP/NAP and SI indices have a high potential in detecting deforestation at the Neolithic/Bronze Age transition. This transition is easy to infer because the development of more advanced farming techniques during the Bronze Age likely resulted in a permanent ecosystem shift, in contrast to earlier transient forest clearances that were followed by forest regeneration (Lang, 1994;Poska et al., 2004;Rey et al., 2019). Conversely, some indices may fail to identify human impacts at a later, higher technological stage (i.e. stages 3-4, Figure   5), specifically at southern sites such as Accesa and Origlio. The cause might be a reduction of deforestation rates and an increased efficiency in farming practices during the Roman Imperial Period, for instance (Howatson, 2011). In addition, several centuries-long crises (e.g. migration period, Little Ice Age) are documented in archaeological and historical records (e.g. Maise, 1998) and by temporally well resolved palaeoecological records (Lotter, 1999;Tinner et al., 1999;Rey et al., 2017). These crises may not be evident in our comparisons among human-impact periods ( Figure 5), which was designed to assess the impacts of long-term trends in human population growth.

4.2.Direct indicators
The indices based on direct indicators (i.e. crops and strict adventives, CI and PI;Lang, 1994) showed an overall good performance at the temperate sites (Figure 5a-d). In temperate environments PI and CI were sensitive enough to detect initial stages of human impacts on the landscape when they were used in the appropriate setting, such as the early Neolithic farming (ca. 3800 BC; Cortaillod typochronological unit) at Burgäschisee (Rey et al., 2017). Similarly, CI was used to trace the major milestones in the history of human occupation around Lago di Origlio, such as Neolithisation and the establishment of permanent settlements during the Bronze Age (Tinner et al., 1999). Likewise, CI tracks the main economic changes at Lake Gosciaz since the Neolithic (Ralska-Jasiewiczowa and van Geel, 1992), whereas PI shows inconclusive evidence until the Iron Age (Figure 4b). In boreal environments, CI performed better than PI (Holtjärnen), while PI was superior at Mediterranean sites (Lago dell'Accesa and Gorgo Basso). The reduced performance of PI in boreal environments might be connected to the prevalence of pastoral activities (Morris et al., 2014), as also revealed by the good agreement with PDI ( Figure 4). Conversely, the reduced performance of CI at the Mediterranean probably relates to the supposed natural occurrence of Plantago lanceolata-type. Indeed, Tinner et al. (2009) concluded that intense agriculture around Gorgo Basso prior to the early Neolithic (ca. 6000 BC) would be unrealistic. To overcome this issue, the authors relied on the combined evidence of crops (e.g. Cerealia-t., Ficus carica) and weeds to track the onset of Neolithic farming (Tinner et al., 2009). Here, we cannot assess to what extent the natural occurrence of cereal species may affect the interpretation of Mediterranean pollen records (Roberts et al., 2011). A way to overcome this difficulty and to identify unambiguously arable farming is to associate Cerealia-type pollen with that of other crops (e.g. figs), adventives and/or apophytes (Tinner et al. 2009). Our results emphasize that the indicative power of single taxa should not be considered in absolute terms but rather within the ecological context (reflected in the pollen assemblages) in which it was growing.
Other important direct indicators according to Behre (1981) such as Fagopyrum and Linum usitatissimun were completely absent or recorded just in modern samples at our study sites and did thus in general not contribute to the index values. The only exception is a single pollen grain of Linum usitatissimun found at the end of the Mesolithic (ca. 3700 BC) in the Burgäschisee pollen sequence.
An additional issue related to the use of indices based on direct pollen indicators such as the PI as quantitative proxies for human impact is the possible bias introduced by agro-industrial practices. This is the case for instance in Origlio during the stage of very high human impact (i.e. stage 5, last 1000 years) when Cannabis-type pollen became strikingly abundant in the sediment samples as a result of water-retting of hemp for fiber extraction (Bradshaw et al., 1981). These practices caused a marked rise in Cannabis-type pollen abundance (from ca. 2 to 40% of the terrestrial pollen sum) and an overrepresentation of this pollen type in the PI index values (up to 98.6%).

4.3.Indirect indicators 16
Our results show that the environmental context is determinant for the interpretation of indices based on secondary indicators such as SI, API and PDI. At all the study sites, but more strongly in the Mediterranean realm (i.e. Lago dell'Accesa and Gorgo Basso), these indices, particularly SI and API, suggest strong human impact during the Palaeolithic and the Mesolithic (before 8000-9000 BC; Figures 4, 5), which is inconsistent with archaeological evidence (Bietti Sestieri, 2013a, 2013b. The underlying reason is that several of the pollen types included in SI, API and PDI (Table 1)  Cheddadi et al., 2019). As a consequence API clearly fails to recognize low or absent human impact before the mid-Holocene, while subsequently it follows the more reliable direct indicators such as PI and CI at sites with temperate (i.e. Burgäschisee and Lago di Origlio) and meso-mediterranean vegetation (i.e. Lago dell'Accesa, Figures 4c-e, 5c-e). In general, SI and API show similar performances (Figures 4, 5). Last but not least, these indices would probably benefit from more detailed taxonomic resolution in pollen identification to enhance the value of indicator species. For instance, the API index merges several well characterized pollen types such as P. lanceolata-type, P. major-type, P. media-type, P. maritima, P. tenuiflora, and P. coronopus-type (according to Moore et al., 1991) into the genus Plantago. Such coarse taxonomic resolution causes a large loss in ecological information, directly affecting human-impact reconstructions (e.g. P. lanceolata is likely adventive north of the meso-mediterranean and thermo-mediterranean vegetation types).
Other indirect indicators of human impact (Behre, 1981;Lang, 1994) displayed a very low contribution to the SI sum (e.g. Polygonum aviculare-type in all cases except for Höltjarnen and Accesa, where it was absent) or occurred regularly throughout the core (e.g. Rumex acetosa-type in all cases, with smaller percentages for Accesa and Gorgo Basso), even at stages of high human impact.
These results suggest that some taxa may be used as qualitative indicators (presenceabsence) more than quantitatively (i.e. with abundance values).

4.4.Woody crops
The performance of the indices based on woody crops, i.e. OJC and OJCV, was generally low across the selected study sites (Figures 4,5), with only one exception in the meso-mediterranean vegetation (Accesa), where these indices were originally conceived (Figures 4e-f, 5e-f). The native status in the Mediterranean region of the constituent species (i.e. Olea europaea, Juglans regia, Castanea sativa), along with their relatively recent and often massive cultivation, potentially implies the coexistence of wild and domesticated trees in certain periods (Conedera et al., 2004;Pollegioni et al., 2017;Langgut et al., 2019), in addition to biased indication power  when applying these methodologies. For instance, strikingly high values of OJC in thermo-mediterranean Sicily (Figure 4f) during the Neolithic are very likely due to abundant Olea europaea pollen from wild populations (Olea europaea var. oleaster, see discussion in Tinner et al. 2009). Indeed the archaeobotanical evidence places the origins of olive tree domestication in the Mediterranean Levant during the Chalcolithic at ca. 4000 BC and tree cultivation likely arrived in Sicily at the beginning of the Bronze Age at ca. 2000 BC (Besnard et al., 2018;Langgut et al., 2019). Furthermore, an OJC drop in Sicily during Roman Times (Figure 4f) is inconsistent with the agricultural intensification inferred from archaeological evidence during that human-impact stage. OJC values are largely driven by Olea pollen percentages throughout the record and primarily reflect the demise of Mediterranean mixed evergreen broadleaved woodlands related to land-use intensification (Tinner et al., 2009), rather than the collapse of olive plantations (Figure 5f). Thus, only the later medieval increase in OJC, driven by Olea pollen abundance, should be attributed to broad-scale olive cultivation in the area (Figure 4f; Tinner et al., 2009). Palaeobotanical evidence also supports the native status of Castanea sativa to several Mediterranean areas including the sub-mediterranean Italian Peninsula, where it may have survived the harshest periods of the last glaciation (Krebs et al., 2004(Krebs et al., , 2019. Indeed, when the OJC index was proposed, Mercuri et al. (2013b) warned about the need for independent archaeological information and the use of other pollen indicators to support inferences based on this index. The incorporation of Vitis to the index (i.e. OJCV) faces the same issues because Vitis is also native to the Mediterranean Basin (Morales and Ocete, 2015), and Vitis pollen shows regular occurrences in many Mediterranean and sub-Mediterranean palynological sequences throughout the Holocene, probably related to Vitis vinifera subsp. sylvestris (e.g. at Lago di Origlio; Tinner et al., 1999). Despite these issues, these indices may help to corroborate intensified land-use where any of the included taxa constituted a relevant food source: e.g. Juglans regia on the Swiss Plateau, Castanea sativa in the Southern Alps, and Olea europaea under sub-humid meso-mediterranean conditions (Tinner et al., 1999;Colombaroli et al., 2008;Rey et al., 2017). In summary, the suitability of tree crop taxa as anthropogenic indicators is highly context-dependent (both geographical and historical), as clearly highlighted by the large dominance of single pollen types in the composition of the OJC/OJCV depending on the site and historical period considered (see above). In this regard, similar to API and SI, OJC/OJCV may be best interpreted as a summary of pollen types of indirect value than a human impact index sensu strictu (Behre, 1981).

Conclusions
Disentangling anthropogenic and natural drivers of vegetation change is of paramount importance in palaeoecology. The most widely used method continues to be the long-standing species indicator approach, although alternative methodologies and proxies have been developed more recently (e.g. Sugita, 2007;Fyfe et al., 2010). In particular, systematic approaches and standardized tools assisting pollen-based reconstructions of land use are still lacking. In this context, detailed and region-or site-specific assessments of the native plant range and the definition of apophytic and adventive anthropogenic pollen indicators (e.g. following a probabilistic approach as proposed by Krebs et al., 2019) is crucial to improve the performance of the existing indices to track changes in land-use intensity. Although the effects of the taxonomic resolution in the identification of human indicators has not been addressed in detail so far, enhanced taxonomic resolution allowing stricter selection of the anthropogenic pollen indicators may also play a role in improving estimates of human impact through long timescales, as the indices with low taxonomic resolution may not perform well even in their scope area. Future research on this topic should therefore aim to develop more detailed and articulated algorithms for assessing human impacts based on multi-proxy palaeoecological data. In particular, we stress the importance of developing a generalized context-dependent approach that considers the geographic area of reference, and analyses the accompanying taxa in the corresponding stratigraphic levels of the pollen assemblage when assessing the indicative power. Finally, for a more precise and accurate independent validation of long-term vegetation dynamics using archaeology, it will be crucial to synthesize the available archaeological evidence (e.g. radiocarbon dates and their material culture context) to better infer major technological innovations, economic changes, land use, and population densities.  (Behre, 1981;Lang, 1994); CI: Cultural indicators ; API: Anthropogenic pollen indicators (Mercuri et al., 2013a); OJC: Olea, Juglans, Castanea (Mercuri et al., 2013b); OJCV: Olea, Juglans, Castanea, Vitis (Bevan et al., 2019;Roberts et al., 2019); PDI: Pollen Disturbance Index (Kouli, 2015); AP/NAP: Arboreal pollen/ non arboreal pollen ratio (Berglund et al., 1991;Favre et al., 2008).   stages that according to our hypothesis, will increase its magnitude towards present.