Climate variability and human impact on the environment in South America during the last 2000 years: synthesis and perspectives

Introduction Conclusions References

pollen, charcoal and tephra from lake sediments and peat bogs) have been particularly underutilized in this regard.
Increasingly studies have demonstrated the integration of multiple proxies (Li et al., 2010) in a climate reconstruction, with a special focus on the two millennia (2 k) before present (BP, present defined as AD 1950). This period could be considered a baseline 10 to current conditions as climate has been very similar to the present. This integration is still in its infancy in South America (SA), especially in the tropics. Since 2009, regional climate reconstructions from this region have gained momentum by compiling multiple datasets and fine-tuning reconstruction methods (Villalba et al., 2009). An adequate spatial distribution of proxy data sets is one of the key necessities identified (Villalba 15 et al., 2009;Flantua et al., 2015a). Fortunately, tree rings studies have expanded their geographical coverage. These constitute a widely distributed and frequently used highresolution climate archive (Boninsegna et al., 2009;Villalba et al., 2009). However, the temporal range of tree rings records is limited compared to the expanse of spatial and temporal coverage provided by pollen records. The newly updated inventory of paly-20 nological research in SA (Flantua et al., 2015a) documents the extensive spatial and temporal coverage of pollen-based research available throughout the continent. Additionally, developing alternative recalibrated age models and evaluation of chronologies is another step forward in multi-proxy integration in SA (Flantua et al., 2015b). However, multi-proxy climate reconstructions from the last 2 ka have hitherto been focused fluence of these modes on South American climate. The spatial influence of climate modes is assessed by documenting their role in driving interannual precipitation and temperature variability. Following this climatic framework, we explain the procedure for selecting robust pollen records in seven regions and assess the regional significance of climatic and human drivers of vegetation change over the last 2 ka. Records along 10 coastlines influenced by sea level changes were not included.

Continental overview climate zones and modes
Continental SA extends from the tropics (12 • N) to mid-latitudes (55 • S). Three major noticeable climate zones can be distinguished; tropical South America, subtropical South America and austral South America. Atmospheric circulation and climate in all 15 three zones is highly modulated and constrained by the orography of the Andes, the shape of the continent and interactions with the underlying land-surface, vegetation and soil moisture (Wang and Fu, 2002;Li and Fu, 2006). The climate of tropical SA is dominated by the seasonal migration of the Intertropical Convergence Zone (ITCZ) over the Atlantic and Pacific, and the seasonal development 20 of convective activity associated with the South American Summer Monsoon (SASM) over the interior of the continent (Fig. 1). The seasonal migration of the ITCZ affects primarily coastal areas and northernmost SA as it is characterized by a fairly well constrained narrow band of low level wind convergence over the equatorial oceans. The SASM is a seasonal phenomenon that develops between September and April and af- 25 fects primarily the Southern Hemisphere tropics and subtropics (Garreaud et al., 2009). During the austral spring-summer (December to February, DJF) transition, moisture influx from the ITCZ contributes to the development of this monsoon system (Zhou 3480 not result from spurious common trends. More information on the methodology can be found in the Supplement. In all correlation maps (Figs. 2 and 4) we show correlations in excess of ±0.2 only, which approximately corresponds to the 95 % significance level. For the regression maps (Figs. 3 and 5) we used thresholds of ±0.12 • C and ±50 mm, respectively. 10 The correlation maps can help inform whether a certain temperature or precipitation anomaly in the regression map is statistically significant. In our discussion we focus primarily on the impact of the positive phase from each of these modes, as these are the fingerprints presented in Figs. 2-5. Since this is a linear analysis the negative phase of these modes would lead to the same changes in temperature and precipita- 15 tion, but with the sign reversed. In general these outcomes are consistent with earlier analyses reported by Garreaud et al. (2009). However, some differences are apparent and most likely related to different time periods analyzed, our choice of using the hydrologic year as opposed to the calendar year, and different definitions of the indices used (see Supplement for more details). For example, Garreaud et al. (2009) used the along the west (W) coast of SA, however, with a somewhat stronger influence on temperature in N-C Chile. It is noteworthy that the IPO impact over SA is almost identical to the influence of the Pacific Decadal Oscillation as described in Garreaud et al. (2009). The N Atlantic modes, Atlantic Multidecadal Oscillation (AMO) and Tropical North Atlantic SST (TNA) are also quite similar, both featuring warming over tropical SA dur- 15 ing periods when sea surface temperature (SST) in the N Atlantic domain are above average, most notably so over the southern C Amazon Basin (Figs. 2 and 3). In fact the warming associated with a unit variation in the AMO or TNA index is larger over most of the Amazon Basin than the warming associated with ENSO. The region of largest warming is co-located with an area of strong precipitation reduction during the warm 20 phase of the TNA and the AMO (Figs. 4 and 5). This suggests that much of the warming is caused by cloud cover and soil moisture feedbacks associated with reductions in precipitation (reduced cloud cover leading to enhanced solar radiation and reduced soil moisture limiting evaporative cooling).
The S Atlantic counterpart, the TSA, is associated with a temperature dipole over 25 subtropical SA, characterized by warming along a zonal band extending from the S-C Brazilian coast westward to Bolivia, while C Argentina contemporaneously experiences cooling (Figs. 2 and 3). The warming in the subtropical region coincides with a region of reduced precipitation during the TSA positive phase (Fig. 4) warming is at least in part caused by changes in the hydrologic cycle (cloud cover and/or soil moisture feedbacks). The SAM is positively correlated with temperature over Patagonia (Fig. 2) and also shows a weak negative temperature departure over western tropical SA during its positive phase (Fig. 3). The warming over Patagonia is strongest during austral summer 5 (Garreaud et al., 2009; not shown) and results from enhanced heat advection, combined with higher solar radiation receipts due to cloud free conditions (Gupta and England, 2006).

Precipitation
Given that ENSO is the source of the strongest interannual variability on Earth, it is 10 not surprising that it also leads to the strongest modern precipitation anomalies over SA (Fig. 5). In general in the tropics, El Niño events lead to significant precipitation reductions over much of tropical SA, with the strongest signal seen in N Brazil along the Atlantic coast and in the Andes of Colombia. Over NE Brazil the precipitation reduction is the result of El Niño events inducing a delayed anomalous warming of the 15 tropical N Atlantic in boreal spring (e.g. Curtis and Hastenrath, 1995;Giannini et al., 2001). Hence the ENSO influence in this region strongly projects onto the TNA pattern (Fig. 4). Over the N Amazon Basin the precipitation reduction is the result of a shifted Walker circulation, enhanced subsidence and reduced convective activity (e.g. Liebmann and Marengo, 2001;Ronchail et al., 2002). In the subtropics on the other hand 20 precipitation is enhanced during El Niño events, in particular over southeastern SA (see also Grimm et al., 2000). The only tropical location that sees an increase in precipitation during El Niño is along the Pacific coast of Ecuador and northern Peru, where flooding is a common occurrence during these events (e.g. Takahashi, 2004). During La Niña events these precipitation anomalies are essentially reversed. The correla- 25 tions are weaker in our annual analysis over some regions where the ENSO influence is highly seasonal, such as the precipitation reduction over the "Altiplano" (high plain) Introduction  (Vuille et al., 2000) or the enhanced precipitation during El Niño in C Chile in June to August (JJA; Montecinos and Aceituno, 2003). The largest change in the IPO in the period analyzed is related to the Pacific climate shift of 1976-1977, when the tropical Pacific switched from its cold to its warm phase. Since El Niño events also became more frequent and stronger over this period (includ-5 ing the two extreme events of 1982-1983 and 1997-1998), it is no surprise that the observed changes in precipitation associated with the IPO are similar to the ENSO footprint, albeit somewhat weaker. Indeed the low-frequency modulation by the IPO may strengthen El Niño events during its positive phase and weaken La Niña events, while the opposite is the case during the IPO negative phase, a phenomenon known as "con-10 structive interference" (e.g. Andreoli and Kayano, 2005). Espinoza Villar et al. (2009) documented the influence of Pacific interdecadal variability on precipitation over the Amazon Basin and showed that its positive phase is related to a decrease in precipitation over the basin since 1975, consistent with our results.
Precipitation is reduced in the southernmost part of SA during the positive phase of 15 the SAM (Fig. 4). This reduction extends N into the subtropics along both the Atlantic and Pacific coast to approximately 30 • S (Silvestri and Vera, 2003;Gillett et al., 2006). Most of this precipitation reduction is associated with reduced westerly moisture flux and moisture convergence from the Pacific (Garreaud et al., 2013). The correlation (Fig. 4) and regression (Fig. 5) maps also suggest a significant influence of the SAM 20 on precipitation in parts of the tropics. This signal, however, is not well documented and its physical mechanism is unclear. It may to some extent be related to teleconnections and an anticorrelation between ENSO and the SAM (e.g. Carvalho et al., 2005), which is supported by the fact that the Niño3.4 index and the SAM correlation maps are almost mirror images of one another (Fig. 4). 25 The AMO and the TNA have a similar fingerprint on the hydrologic cycle of SA (Fig. 5). Both modes are characterized by a significant reduction in precipitation over much of the Amazon Basin during their positive phase, with the amplitude of the changes being slightly larger associated with TNA forcing. This negative precipitation Introduction anomaly is associated with the northward displacement of convective activity in the ITCZ region due to warmer SST in the tropical North Atlantic and Caribbean during the positive phase of the TNA (and to a lesser extent also the AMO). This directly affects precipitation amounts over NE Brazil (e.g. Hastenrath and Greischar, 1993;Nobre and Shukla, 1996), while the northward shift in the core region of convection also leads 5 to anomalous subsidence, located over the Amazon basin. In fact the recent droughts in 2005 and 2010 in the Amazon Basin were both associated with such anomalously warm SST in the tropical N Atlantic (Marengo et al., 2008;Lewis et al., 2011). The only region where precipitation is enhanced is in the NW part of the Amazon belonging to Venezuela, Colombia and Peru (Fig. 4). 10 An anomalously warm tropical S Atlantic (positive phase of the TSA) leads to the exact opposite conditions, with the ITCZ displaced anomalously far south, causing copious rainfall over NE Brazil, with weaker positive anomalies extending inland as far as the Peruvian border (Fig. 5). Another region of enhanced precipitation is located in S Brazil, associated with a southerly movement of the SACZ ( Fig. 1; e.g. Doyle and 15 Barros, 2002).

LAPD overview and selection of pollen records covering 2 ka
From the newly updated Latin American Pollen Database (LAPD, Flantua et al., 2015a) we selected the records that cover the last 2 ka. Good chronological control is required for PAGES-2 k, but the youngest ages in pollen records are typically constrained by 20 geochronological data. An assessment of the pollen records by the authors with expertise in each sub-region of SA has revealed 585 records with pollen samples within the 2 ka-range (Fig. 6), of which 337 and 182 records, respectively, contain one or more geochronological date within that time period. Thus, 182 studies were considered suitable for paleoclimate reconstruction as outlined by the PAGES-2 k criteria. Introduction Through the identification of temperature sensitive proxies, the development of climate reconstructions has advanced thanks to regional efforts (e.g. LOTRED-SA) to compile a proxy database (PAGES-2 k Consortium, 2013). Both temperature and moisture records are currently being collected, requiring a set of criteria that define the suitability of individual records.

5
As a global initiative, a defined set of criteria guarantees the quality of the proxies used for climate reconstructions and therefore focuses on chronological accuracy and record resolution (Table 2). Within this paper we regarded criteria A (peer-reviewed publication) and B (minimum duration of the record of 500 years) as the base line criteria. All criteria followed those stated by PAGES-2 k except for the criterion on resolution. 10 Implementing the criteria of a maximum resolution of 50 yr sample −1 , would leave only a handful of pollen records to discuss. The sparsity of samples that meet the stringent PAGES-2 k resolution criterion occurs because palaeorecords with long time spans (> 10 000 years) are typically sampled at coarser temporal resolution. Furthermore, many lowland sites have very low sedimentation rates, which preclude high-resolution 15 sampling. Therefore we propose a more flexible temporal resolution, depending on the identified relevance of the case study. Records with a resolution of 200 to 300 years are included in our discussion. Within the regional assessments, only records that fulfil more than three criteria are discussed, unless the records are considered particularly valuable for regional climate assessments. Introduction

Climate-vegetation interaction in the Venezuelan Guayana highlands and uplands
The study area, known as the Gran Sabana (GS), is located in SE Venezuela between the Orinoco and Amazon basins ( Fig. 6a; Huber and Febres, 2000). Huber (1995) recognized three main elevational levels on the Venezuelan Guayana: lowlands (0-5 500 m a.s.l.), uplands (500-1500 m a.s.l.) and highlands (1500-3000 m a.s.l.). Lowlands are absent in the GS, which is mainly characterized by a continuous upland peneplain spiked with isolated highlands (table-mountains, "tepuis"). The GS highlands are part of the so-called Pantepui phytogeographical province, which is characterized by unique biodiversity and endemism patterns, encompassing all the tepui summits above 10 1500 m a.s.l. (Huber, 1994;Berry et al., 1995). The tepuian vegetation is characterized by a mosaic of bare rock, pioneer vegetation, tepuian forests, herbaceous formations and shrublands (Huber, 1995b). Additional background information is provided in the Supplement.
In the GS, 22 pollen records cover the last 2 ka. There are 4 records with a chronol-15 ogy based on one control point and an additional 10 records from which most, or all, control points lie outside 2 ka. Three potentially suitable records originate from the highlands, Eruoda PATAM6-A07, Churí Chim-2 and Apakará PATAM9-A07, and only 1 is found in the uplands, Laguna Encantada PATAM4-D07 peatland ( Fig. 7a; Table 3). Of the 3 records of the highlands, just Eruoda provides sufficiently high 20 resolution to explore the objectives proposed here. However, only Churi Chim-2 and Apakará contains several age control points within the last 2 ka, and Laguna Encantada presents a relatively low sampling resolution of 200 to 300 years. The criteria for chronological control has excluded some of the most relevant work for the research questions posed by this paper. For example, the vegetation at the Eruoda Introduction the vegetation dynamics observed in the fossil records are fully climate driven and therefore a record valuable for LOTRED-SA. Equally of high importance is the Urué record in the uplands, which does not meet the dating control constraints but the sampling resolution is high enough to provide important insights into the vegetation-climate dynamics during the last 2 ka, and will be therefore be presented here.

5
The Eruoda summit represents an important reference to which almost all the tepuian summits vegetation dynamics can be compared (Fig. 7b). In general, these summits are insensitive to temperature change (for 2 ka), whereas moisture variations potentially may cause small internal reorganisations of plant associations although these shifts are considered to be of minor ecological significance. Shifting river courses are considered to influence local vegetation patterns through the lateral movement of gallery forests in landscapes dominated by broad-leaved meadows (Rull, 2005a, b).
The Urué sequence spans the last 1.6 ka and records the vegetation dynamics after an important fire event dated ∼ 1.6-1.8 ka. Three main vegetation stages were reported coeval with high charcoal abundances at the bottom of the sequence, corresponding to 15 plant communities' transitions from open secondary forest to fern-dominated associations transitional to savanna. Savannas were fully established around 0.9 ka, coinciding with the beginning of a phase of lower charcoal values, and continued as the dominant plant association until present-day. Savannas were accompanied by Mauritia flexuosa palm swamps ("morichales") that established a phase that was likely more humid. 20 These palm swamps greatly varied in extent through time, showing a parallel between the lowest palm abundance and two drought intervals' occurrence. These two drought intervals were centred during the 0.65-0.55 and 0.15-0.5 ka coeval to the Little Ice Age (LIA) signal observed in the Venezuelan Andes (Rull et al., 1987;Rull and Schubert, 1989;Polissar et al., 2006). Generally, the vegetation dynamics recorded so far in the 25 Venezuelan Guayana uplands have shown a higher sensitivity to changes in the available moisture than to potential shifts in the average temperatures. The last 2 ka have been mainly characterised by vegetation change at a local scale. Introduction

Climate-vegetation interaction in the Northern Andes
The region of the N Andes consists in political terms of Colombia, Ecuador and Venezuela and includes a wide range of different ecoregions (Fig. 6b). Sharing both the Caribbean and the Pacific coastline and various climate influences, Colombia has a unique pattern of different ecosystems shared with neighbouring countries. Pollen 5 records are found throughout a wide range of biomes and elevations (Flantua et al., 2015a), from the tropical rainforest and mangroves along the coast to the high Andean "páramos". The complex formation of the Andes with the three mountain ridges characterizes this region with numerous valleys and watersheds. A total of 64 records are available that present pollen data within the last 2 ka. Of this 10 number, 21 fulfilled four of the PAGES-2 k criteria, another 24 fulfilled three criteria, and the remaining records presented at least two dating control points within the last 2 ka. Unfortunately, 14 were presented in publications without a peer-review procedure or presented only as a summary diagram (7 records with four positive criteria). An additional 5 records which fulfilled all criteria suggested human presence during most part 15 of the last 2 ka, and were therefore excluded for climate reconstructions. From the remaining records, only lakes Pallcacocha and Papallacta PA1-08 in Ecuador lack human interference during the last 2 ka. The others describe human indicators over limited periods of time and are considered valuable for PAGES-2 k purposes. Most of the records complying with three criteria (n = 24), most of them identify human presence 20 in the near surrounding of the record during a reasonable period of time, leaving only ECSF Refugio potentially suitable for 2 ka climate reconstructions. Beginning at the far N of the region ( Fig. 6b and Fig. 8), Lake Valencia is represented by three cores with varying quality in chronology and resolution. In spite of the low resolution sampling (10-14 samples for the last 2 ka), some general information 25 can be derived from the joint interpretation of these three cores. The last 2 ka are characterised by a decline of forest cover, attaining the lowest values of the Holocene, at the expense of savannas. Aquatic proxies indicate declining lake levels and increasing nu-Introduction Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | trient input, a trend that accelerated during the last 0.5 ka, when human activities were more intense around the lake. Considering the entire Lateglacial-Holocene record, the Lake Valencia catchment has shown to be more sensitive to moisture variations than to temperature, as known from tropical lowlands. In the Andean region, changes of the altitudinal position of the upper forest line 5 (UFL) are instrumental in reconstructing temperature changes. This ecotone is defined as the highest elevation contour of continuous forest and marks the boundary between the forest and high Andean páramo biome (Moscol-Olivera and Hooghiemstra, 2010; Groot et al., 2013 samples indicates significantly depressed UFL in comparison to today. Along the transitional zone between savanna and tropical rainforest in the eastern Colombian savannas, three pollen records fulfil at least three criteria. This climatesensitive transition zone is thought to reflect precession-forced changes in seasonality, latitudinal migration of the ITCZ, and changes in the ENSO (Figs. 3 and 4;Wille et al., 20 2003). Since 2 ka gradual increase in savanna vegetation is observed, suggesting a period of progressively drier conditions, e.g. Loma Linda and Las Margaritas). However, expanding Mauritia palm forest during this period was observed in several records considered to reflect increased local water availability and precipitation (Fig. 8b), and/or human impact (Rull and Montoya, 2014 cated in the coastal plain receiving signals from shifting mangrove forests. These shifts were considered not to be climate related but explained by tectonic events in the region and/or dynamic shifts of the river deposition patterns. Frequent erosion events, various seismic shifts and disturbance indicators from mixed origin during the last 2 ka hinder consistent conclusions for the region. Changes in vegetation composition around 5 0.65 ka were assigned in El Caimito to reduced flooding and possible human intervention, while similar changes at Jotaordó were ascribed to endogenous dynamics. Only the multi-proxy approach of El Caimito suggests a possible relationship between periods of higher riverine dynamics and the frequency of long term ENSO variability. Interestingly, within this region Cecropia is used as natural disturbance indicator due to fluvial-marine dynamics, while in the other Colombian regions this fast-growing species is considered characteristic of human interference: both settings have disturbance as a common factor. In the Colombian Andes there are no undisturbed pollen records during the last 2 ka suitable for climate reconstructions. Before the human disturbances, the La Cocha-1 15 record in the far S of Colombia (Fig. 8b) indicated generally wetter conditions similar to the N Ecuadorian pollen records of Guandera-G15 and Guandera-G8. Andean records can display dissimilar timing and trends behaviours due to differences in precipitation along the eastern Andean flank and specific regional landscapes (Moscol Olivera and Hooghiemstra, 2010;Marchant et al., 2001). A different kind of index to 20 highlight vegetation-climate interaction was used in the eastern Ecuadorian Andes at Papallacta PA1-08. Established to characterize the SASM and ENSO, the index interprets cloud transported forest pollen taxa and Poaceae as a proxy for upslope cloud convection. Supported by a high resolution (∼ 15 year), a high frequency of dry and humid episodes is detected during the last 1.1 ka. In this alternation of convective 25 activity, the MCA, LIA and current warm period are considered detectable.
In S Ecuador 4 pollen records suitable for PAGES-2 k purposes are found within a relatively small sub-region. Tres Lagunas suggests a cold phase, possibly the LIA, as one of several warm and cold phases detected during the last 2 ka (Fig. 8b) Cold and moist conditions are related to high abundances of Poaceae, Isoëtes and Gentianella. At Laguna Zurita, the decrease of Isoëtes was considered an indication of increased precipitation after ∼ 1.2 ka, observed similarly in other fossil pollen records in the C Peruvian Andes. On the other hand, chemical analyses from the same core suggested drier conditions during the last millennium, confirmed by a different set 5 of palaeoclimatic records. Unknown human interference in the last millennium could be related to these divergent patterns, as the nearby ECSF Refugio and Laguna Daniel Álvarez detected Zea mays around 1.4 and 0.8 ka respectively. Climate was considered to be drier overall before 1.2 ka.

10
The C Andes includes the high elevation plateau of the Altiplano, above 3000 m a.s.l., in S Peru, Bolivia and N Chile (Fig. 6c). The Altiplano is an area of internal drainage within the Andes that contains multiple peaks over 5000 m a.s.l. The vegetation of the Altiplano is characterized by different grassland types, collectively known as "puna" (Kuentz et al., 2007). Within the grassland matrix are patches of woodland dominated 15 by trees of the genus Polylepis (Fjeldså and Kessler, 1996). To the E and W of the Altiplano are the steep flanks of the Andes. In total 57 pollen records covering the last 2 ka were identified from the Altiplano in Peru and Bolivia. Only 4 of the Altiplano records met all PAGES-2 k criteria: (i) Cerro Llamoca, (ii) Marcacocha, (iii) Chicha Soras, and (iv) Pacucha ( Fig. 9a; Table 3). 20 From the surrounding regions 2 additional records are also considered here because of their importance and fit to the PAGES-2 k criteria: (i) Consuelo on the eastern Andean flank, at mid-elevation (1370 m a.s.l.) within the cloud forest fulfils two of the criteria, while (ii) Urpi Cocha on the Pacific coast at sea-level, ∼ 1 km inland within the archaeological site of Pachacmac (near Lima), satisfies four of the criteria. Of the Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | humans from 1.5 ka onwards, and the other 4 (Marcacocha, Pacucha, Nevado Coropuna, and Urpi Cocha) indicate human impact throughout the last 2 ka. Discerning a climate signal from the pollen records of the last 2 ka in the C Andes is a challenge due to the long legacy of human occupation and landscape modification (Bennett, 1946;Dillehay et al., 2005;Silverman, 2008). However, some idea of 5 vegetation-climate relationships can be gained from modern pollen studies within the Puna, e.g. Kuentz et al. (2007) use the ratio of Poaceae:Asteraceae (Coropuna), or Schittek et al. (2015) focus on the abundance of Poaceae (Cerro Llamoca) as an indicator of moisture availability. In the other records, where there is no direct relationship between vegetation and climate discernible, some authors look at the relationship 10 between the pollen records and other indicators to disentangle climate and human induced vegetation change; such as independent evidence of farming activity (e.g. oribatid mites), or association with archaeological evidence for abandonment/occupation (Chepstow-Lusty, 2011).
The two records considered here that are purported to have no local human impact 15 (Cerro Llamoca and Consuelo) provide the best opportunity of extracting a clear insight into past climatic change in the C Andes during the last 2 ka. The record from Cerro Llamoca (4450 m a.s.l.) indicates a succession of dry and moist episodes (Fig. 9b). After 0.5 ka sediments at Cerro Llamoca are composed of re-deposited and eroded material and consequently interpretation of the latter half of the record is 20 difficult. In contrast little compositional change is evident in the Consuelo record, with the most significant variance during the last 2 ka being a rise in Cecropia sp. pollen after 1 ka. Cecropia pollen is typically interpreted as an indicator of disturbance (Bush and Rivera, 2001) and therefore, in the absence of humans signal, the rise in Cecropia could be interpreted as an elevated level of natural disturbance. Archaeological evidence from Chicha-Soras (∼ 3500 m a.s.l.) does not show any evidence of human occupation of the valley between ∼ 1.9 ka and ∼ 1.4 ka. Between 1.4 and 1 ka and between 1 and 0.65 ka, high abundance of Chenopodiaceae/Amaranthaceae (Cheno/Am) could be interpreted as either indicating arid conditions or expansion of quinoa crops (Ledru et al., 2013). However, a drop in charcoal 5 fragments (fire activity) coupled with the absence of archaeological evidence (∼ 1.9-1.4 ka), suggests that people abandon the valley during 1.5-0.5 ka and, consequently, that the aridity signal from the pollen could be interpreted as a climatic one.
Some climate information has been inferred from the four remaining sites (Marcacocha, Pacucha, Nevado Coropuna and Urpi Cocha) despite the 10 strong human influence over the vegetation. At Nevado Coropuna humid conditions persisted until a short dry episode occurred 0.97-0.82 ka (Fig. 9b). During the last 2 ka at Marcacocha (3300 m a.s.l.) successive peaks in Cyperaceae pollen have been interpreted as indicative of three periods of elevated aridity while elevated Plantago at ∼ 1.9 ka is suggested to indicate cooler conditions, and Alnus at ∼ 1-0.5 ka could in- 15 dicate warmer and drier conditions (Chepstow-Lusty et al., 1996); although discerning the climate signal related to Alnus is difficult due to its utilisation in agro-forestry practices (Chepstow-Lusty and Jonsson, 2000). At Pacucha and Urpi Kocha significant changes to the pollen assemblage in the last 2 ka are attributed to human activity rather than climate. 20 Generally the pollen records from the Altiplano tend to show a greater sensitivity to precipitation, rather than temperature. The greater sensitivity to precipitation is because moisture availability is in most areas the limiting factor for both vegetation communities and human populations. However, two records infer significant changes in temperature related to vegetation/human occupation: (1) Marcacocha when the sudden 25 stop in agricultural activities is attributed to colder temperatures, and (2) at Coropuna when the increase of human occupation (expansion of Inca culture) at higher elevation shows that there was no glacier and warmer temperatures. On the Altiplano variation in the SASM has been attributed as a major driver of changes in moisture balance Introduction at Llamoca, Coropuna, Pacucha through altering the summer precipitation. SASM is also thought to be responsible for precipitation variation on the E Andean flank (Consuelo), while on the western Andean flank (Urpi Cocha) precipitation variation is attributed to the ENSO through tsunamis and the abrupt floodings on the Pacific coast. Although the pollen records of the Altiplano are likely to be somewhat obscured 5 by the agricultural activities and irrigation of the crops all the records point towards a dry event occurring roughly between 1.2 and 0.75 ka.

Climate-vegetation interaction in the Lowland Amazon Basin
For the purpose of this review, the Lowland Amazon Basin constitutes those regions of the Amazon drainage < 500 m a.s.l. and extends to the lowland Guianas (Fig. 6d). This encompasses the evergreen rainforest, which covers most of Amazonia, as well as the S transitional/seasonally-dry tropical forests located in NE Bolivia and S Rondônia, N Mato Grosso and N Para State, Brazil. It also includes the Llanos de Moxos savannas of NE Bolivia, the ecotonal rainforest-savanna areas of N Roraima State, Brazil, and extends to the coastal swamps/grasslands of N Brazil and French Guiana. 15 In total 42 published pollen records that cover the last 2 ka were identified from the Lowland Amazon Basin. Only 5 records complied with all four of the criteria and 11 records met with three criteria ( Table 3). Most of the remaining records span a period ≥ 0.5 ka, but do not meet with any of the other criteria. One of these records, lake La Gaiba, is situated just outside the Amazon Basin, in the Pantanal region of central 20 Brazil/SE Bolivia. However, the record and its hydrological catchment reflect Holocene precipitation in the S Amazon Basin (Whitney et al., 2011), and therefore was included as part of this review.
By applying the dating constraints of the PAGES-2 k criteria, the majority of pollen records from the Amazon Basin are discounted from any analysis of climate-vegetation Introduction Lake Quistococha in the NE Peruvian Amazon is an infilled river channel surrounded by Mauritia flexuosa-dominated palm swamp. Vegetation has undergone several significant species compositional changes over the past 2 ka. The broad pattern of vegetation change was from Cecropia-dominated riverine forest at ∼ 2.2 ka, to abundant Cyperaceae and floating grasses/ferns and the commencement of peat forma-5 tion ∼ 2.1 ka, then to seasonally-inundated riverine forest, with abundant Moraceae and Myrtaceae from ∼ 1.9 ka, and finally, the development of closed-canopy, Mauritiadominated swamp from ∼ 1 ka until present. Superimposed on this broad pattern of change were rapid, centennial-scale shifts in forest composition and degree of openness. However, these rapid shifts were attributed by the authors to hydrological dynamics, rather than climate change or human impact.
Lake Werth belongs to a collection of sites (also Gentry, Vargas and Parker) in the "Madre de Díos" region of the SE Peruvian Amazon. Site Werth is surrounded by humid evergreen rainforest. The lake formed ∼ 3.4 ka and records continuous evergreen rainforest throughout, with little evidence of burning. The records from the sur- 15 rounding three lakes concur, suggesting that, regionally, rainforest (and climate) has been stable over the last 2 ka.
Laguna Granja is located on the edge of the Pre-Cambrian Shield in NE Bolivia. Its location is at the margin of the modern Madeira-Tapajós rainforest ecoregion, which extends southwards from the main Amazon River to the southern margin of the Amazon 20 Basin in Bolivia (Olson et al., 2010). The record has a maximum age of 6 ka and indicates that savanna characterised the landscape around Granja from 6 ka. This is in agreement with a regional scale reconstruction from the much larger Lake Orícore (not shown, Carson et al., 2014), which is located < 20 km away from Granja, and shows climate-driven expansion of evergreen rainforest in this region between ∼ 2 and 25 1.7 ka. However, forest expansion does not occur on the Granja site until 0.5 ka. The distribution of forest vs. savanna around Granja was shown to be heavily influenced by human land use between 2.5 and 0.5 ka (Carson et al., 2014(Carson et al., , 2015, therefore, it is not suitable for analysis of naturally-driven vegetation dynamics. Introduction The Fazenda Cigana record is derived from a palm swamp in the savanna-gallery forest mosaic landscape of N Roraima State, in the N Brazilian Amazon. The core was taken as one of a pair, along with the Terra Indígena Aningal record, which was cored from the same Mauritia swamp. The pollen records are dominated by Mauritia throughout, which the authors attribute to continuously wet climate in this region 5 in the late Holocene. There are however centennial-scale periods of gallery forest reduction and grassland expansion, accompanied by increased charcoal concentrations. Da Silva Meneses et al. (2013) infer that these periods of high burning were anthropogenic in origin, and compare them to modern day prescribed burning practices used by indigenous people in the northern Amazon to maintain an open savanna landscape.

Conclusions
Despite the potential human interference, these records demonstrate natural stability of the forest-savanna ecotone over the last 1.5 ka in this particular part of the N Amazon.
The French Guiana K-VIII record is situated in from the coastal wetland savanna of French Guiana. The record was taken within a landscape of pre-Columbian mounded agricultural fields, with the principal aim of investigating ancient human land use asso-15 ciated with these earthworks on a local scale. From this earliest part of the record, the fossil pollen spectra indicate seasonally-inundated savanna, dominated by Cyperaceae and Marantaceae until 0.8 ka when human inference is detected. In the post-European period after ∼ 0.5 ka, however, charcoal abundance increases, probably reflecting more intensive use of fire by colonial populations. This again is a record that reflects sub-20 stantial anthropogenic impact on the landscape, and is therefore not suitable as an independent proxy record of climate-induced vegetation change; at least not within the last ∼ 0.8 ka.

Climate-vegetation interaction in Southern and Southeastern Brazil
The vegetation in S-SE Brazil includes forest ecosystems such as the tropical Atlantic 25 rainforest, Araucaria forest, semi-deciduous forest, "Cerrado" (savanna woodland) and different grassland ecosystems such as "Campos" and "Campos de Altitude" (high elevation grassland) (Fig. 6e) to 200 km narrow zone in the coastal lowlands along the Atlantic Ocean, and on the coastal eastern slopes of the mountain ranges. The tropical semi-deciduous forest occurs further inland in SE Brazil. The Cerrado is found primarily in C Brazil, but also in the N part of SE Brazil. The subtropical grasslands are found in highland S Brazil and lowlands of the southernmost region of S Brazil. Additional background information is 5 provided in the Supplement.
There are approximately 50 pollen records known from S-SE Brazil, but many sites have not been published in peer-reviewed journals and were therefore not considered. Unfortunately, the 2 records that agree with all criteria, show human interference (Table 3). Therefore a general overview of climate-vegetation interaction from the region 10 are presented, considering the 7 records sites that fulfil some of the criteria (Table 3, Fig. 11a).
Sites located in the mountains of S-SE Brazil and from the transition area between the subtropics and tropics are sensitive to both temperature and precipitation, e.g. the length of the dry season is considered to play an important role. In S Brazil pollen 15 records indicate vegetational changes that reflect a change from relatively dry climate during early and mid Holocene to wetter conditions after about 4.3 ka, and in particular after 1.1 ka (Fig. 11b). Increasing moisture is clearly indicated on the S Brazilian highlands by the expansion of Araucaria forests in form of gallery forests along rivers and a pronounced expansion of Araucaria forest into the Campos after about 1.1 ka (e.g. 20 Cambara do Sul and Rincâo das Cabritas). The expansion of gallery forests at similar time periods (5.2 and 1.6 ka, respectively) is also recorded in the southernmost lowland in S Brazil by the São Francisco de Assis record. Study sites that reflect changes in the Atlantic rainforest area indicate an expansion during the Holocene where overall wetter conditions prevailed compared to highland and southernmost low-25 land areas, e.g. Ciama 2 (Fig. 11b).
In contrast to other sites and regions, a relative humid and warm phase during the LIA is interfered from the high resolution Cambara do Sul record as an expansion of Weinmannia in the Araucaria forest is observed. In SE Brazil the Lago do Pires and Introduction Lagoa Nova record indicate that a dense and closed semi-deciduous forest existed in the region only in the late Holocene period under the current climatic conditions with a ∼ 3 month dry seasons. In the mountains of SE Brazil (e.g. Serra dos Orgâos record) a reduction of Campos de Altitude occurred 0.9 ka indicating a change to wetter conditions that is broadly coeval with a similar trend in the Lago do Pires record 5 (Fig. 11b).

Climate-vegetation interaction in Pampa plains
This region extends E of the Andes, between 30 and 40 • S (Fig. 6f) and is characterized by dominant aeolian landforms marking the structural characteristics of the subsurface geology and climatic gradient of the landscape (Zárate and Tripaldi, 2012). The natural 10 vegetation of the Pampa is a tree-less grassland. Potential vegetation units can be characterized as: E Pampa, inland Pampa and S Pampa, based on phytosociological characteristics, historical observations on land use, and climatic and geomorphological differences (Tonello and Prieto, 2008; Supplement). The region has a relatively short farming history, since most of the area remained as native grassland until the end of 15 the 19th and the beginning of the 20th century (Viglizzo and Frank, 2006). Today, only around 30 % of the region is covered by natural or semi-natural grassland.
In total 9 pollen records were assessed for the last 2 ka. All four dating criteria were met in one record only (Lonkoy) and three criteria were matched at Sauce Grande (Table 3). The pollen record of site Hinojales-San Leoncio does not fulfil the four 20 dating criteria, however the record shows important hydrological signals for the last 2 ka and is therefore briefly discussed (Fig. 12b).
Aquatic ecosystems are considered sensitive to climatic and/or hydrological variations, and exhibit frequent fluctuations in their water level and extension, leaving flooded or exposed plains. The multi-proxy approach allows the identification of re-25 sponses to natural and/or anthropogenic forcing factors. Pollen together with non-pollen palynomorphs and plant macrofossil analysis present similar trends in SE Pampa that support climate to be a regional trigger of change (Stutz et al., 2015). From 2 to 0.5 ka 3500 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | an instable regional environment with drier climatic conditions than present are inferred from the region, based on halophyte plant communities (Chenopodiaceae) surrounding the lakes whereas Chara and other aquatic plants (e.g. Myriophyllum, Potamogeton) characterized the water bodies. Towards ∼ 0.5 ka vegetation switched to Cyperaceae dominance and aquatic plants similar to modern community. Thus, turbid conditions 5 with higher water level and/or extension of surface lakes under more stable environmental conditions are inferred. These support humid conditions similar to present with a noticeable increase of precipitation after 0.3 ka, indicated by high Cyperaceae abundances. However, a integrative multi-proxy approach allow inferring stable conditions and higher salinity values between 1.9 and 0.9 ka and periods of water level fluctua-10 tions after 0.9 ka, with high water levels between 0.66 and 0.27 ka. These changes may have been caused by fluctuations in precipitation (Fontana, 2005).

Climate-vegetation interaction in the Southern Andes and Patagonia
The study area comprises the S Andes, which includes subtropical and temperate regions (22-56 • S) on both sides of the Andes, including Patagonia (40-56 • S) which ex- 15 tends from the Andes eastwards to the Atlantic Ocean (Fig. 6g). The region has different geomorphological settings associated with glacial, volcanic and tectonic activities. Vegetation associations reflect the W-E precipitation gradient from the wet Nothofagus forest to the dry grass and shrub steppe towards the Atlantic coast. The S-N gradient along the Andes ranges from the Nothofagus temperate forest in the aus-20 tral region to the Nothofagus-Astrocedrus forest, sclerophyllous forest and xerophytic woodland in the C region. In the northernmost end of the latitudinal gradient, the vegetation is adapted to extremely arid conditions characterized by small and dwarf shrubs and scarce cover (Supplement). Anthropogenic activities during the last century have caused a range of disturbances (e.g. fire, forest clearance, grazing, agriculture) and 25 major vegetation changes in forest and steppe areas have occurred. In this region, there are 48 pollen records that cover the last 2 ka with at least one chronological control point during this period. Of these, the 19 records that fulfil 3501 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | PAGES-2 k criteria are mostly concentrated in the temperate forests, while only few originate from xerophytic shrub steppe (1 record), subtropical forest -sclerophyllous forest (2 records) and grass steppe (4 records) (Table 3; Fig. 13a). At the most southerly sites, the "Tierra del Fuego"s Onamonte mire (54 • S) located at the Nothofagus forest-grass steppe ecotone shows a gradual Nothofagus for-5 est development between 1.5-0.5 ka followed by a major forest development up to the present, reflecting increased precipitation (Fig. 13b) In C Patagonia (47-44 • S) pollen records are located at the E of Andes (Fig. 13a). At Parque Nacional Perito Moreno (47 • S) a shrub-steppe expansion (Asteraceae and Embothrium dominance) suggests lower precipitation values between 1.2 and 0.25 ka compared to previous values, after which an increase in grass-steppe occurs due to higher moisture availability (Fig. 13b) shows dry-warm phases which were associated with the MCA period. Cold and wet conditions, inferred by the relation between Nothofagus and Poaceae, and changes in the depositional time, prevailed during the LIA, possibly related to El Niño and La Niña influencing these wet and dry phases respectively (Fig. 5).
To the N (westward Andes), the Lago Aculeo record (34 • S) shows dominance 5 of Poaceae dominance suggesting relatively steady conditions during the last 2 ka with expection of last 0.1 ka, when a trend towards warmer conditions or human disturbance is reflected by increase in Chenopodiaceae (Fig. 13b). Interestingly, the sedimentary record shows a series of turbidite layers associated with major ENSO frequency between 1. steppe, shows a shrubland community between 1.2-0.7 ka, associated with drier conditions than at present. An increase in moisture after ∼ 0.7 ka is indicated by Poaceae and Juncaginaceae pollen. Cabo Vírgenes CV22 shows a similar trend, with dry grass-shrub steppe between 1.05-0.6 ka, followed by a grass-dominated steppe suggesting higher moisture availability.

Indicators of human land use in 2 ka pollen records
In general, indicators of human activities in pollen records are decrease in forest taxa (degraded forest) and/or forest representation ( (Fig. 14). Manihot esculenta and other crops such as Zea mays are 5 considered direct indicators of human influence and provide clear evidence of land use. Indirect indicators such as change in forest composition (e.g. due to deforestation) or species known as disturbance indicators (Cecropia and Mauritia) need additional proxies to derive conclusive findings. Only by looking at pollen changes in context with other evidence -e.g. charcoal, limnology, sedimentology, archaeology-can the correct origin 10 of these changes be identified.
In any palaeo-reconstruction concerning the past 2 ka, human land use must be considered as a potentially important agent of environmental change. However, where there is no direct evidence of human land use, such as cultigen pollen, distinguishing natural from anthropogenically induced burning and vegetation change can be diffi- 15 cult. Similar to human indicators in pollen records, complementary paleo-proxies can support more confident interpretation. This further highlights the difficulty of inferring climate-induced vegetation changes, without reference to independent climate proxy data, such as geochemical records from speleothems.
To date, major human impact in the Venezuelan Guayana uplands has been sug-20 gested for the last 2 ka and inferred from the charcoal record, without any evidence of crops. Compared to the highlands (1500-3000 m a.s.l.), the situation in the uplands (500-1500 m a.s.l.) differs substantially as fire is maximally responsible for vegetation change during the last 2 ka. The Urué record shows the consequence of repeated burning upon the vegetation, preventing the recovery of pre-existing forests and allow- 25 ing the appearance of a "helechal" (fern-dominated vegetation; Huber and Riina, 1997), and finally the establishment of the savanna. The occurrence of high fire regimes during the last 2 ka is a common feature of mostly all the upland records analysed so far, regardless the plant association present at each location. Synchronous with this in- Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | crease in fire regime, those records that nowadays are characterised by Mauritia palm swamps, showed parallel a sudden appearance and establishment of Mauritia. Human activities have been proposed as the likely cause of this high abundance of fires, and thereby of the consequences that produced upon the landscape. In this sense, the repeated use of fires would have promoted the reduction of forests and expansion of the 5 savanna, favouring the establishment of Mauritia swamps after clearing. Two records are particularly relevant regarding the human influence on the Venezuelan Guayana uplands. Lake Chonita sequence (Table 3) registered among the earliest Mauritia establishment coeval with a significant increase in the fire regime during a likely local wet period around 2 ka. In the southernmost boundary of the Venezuelan Guayana, 10 El Paují (Table 3) was interpreted as potentially reflecting human activities since the mid Holocene. This location is characterised today by treeless savanna surrounded by dense rainforests that established ∼ 1.4 ka as shown by the highest abundance of algal remains (local wet conditions) and charcoal particles (fire regime). The establishment of the present-day landscape was interpreted as mainly anthropogenically driven, with 15 the arrival to the study area of the current inhabitants. The occurrence of a previous secondary dry forest was interpreted as the result of climate-human interplay, linking land abandonment and likely drier climate as the main responsible favouring the vegetation shift. From the Colombian savannas, human occupation is attested since the mid Holocene (Berrio et al., 2002). At site Loma Linda a plausible signal of human 20 interference in the last 2 ka is shown by increased savanna, although precipitation increase during the same period (Marchant et al., 2001(Marchant et al., , 2002 could be interfering with that signal. The human history in the N Andean region goes back to the Lateglacial (Van der Hammen and Correal Urrego, 1978). The high plains of the Colombian Cordilleras 25 provided suitable conditions for human settlements since the start of the Holocene. Increasing human occupation became evident in pollen records after ∼ 3 ka, such as Fúquene-2 (Van Geel and Van der Hammen, 1973) and Pantano de Genagra (Behling et al., 1998b). In several Andean diagrams, Rumex marked the onset of more  Bellwood, 2004). Before these dates, indigenous populations were scarce and their practices negligible in terms of impact, especially at high elevations sites such as Piedras Blancas in Venezuela.
In the C Andes a high level of human activity, spatially variable in intensity, has been shaping the landscape for the last 2 ka. Cheno/Am and Zea mays generally ap-5 pear in all the records in the central Andes after 4 ka, e.g. Pacucha, Marcacocha, Chicha-Soras and Urpi Kotcha. After 2 ka, Alnus and agroforestry practices are observed (Marcacocha, Pacucha). When irrigation started to be developed in sites without a nearby lake as for instance ∼ 1 ka at Coropuna, Ambrosia may be used as a terrace consolidator. 10 Evidence of afforestation in two sites with high human influence (Marcacocha and Pacucha) are observed. Indeed Alnus acuminata is a tree planted by the Inca to stabilise landscapes (Chepstow-Lusty, 2011). At lower elevation, in the Andean forest, the last 2 ka pollen data indicate little change in woodland cover which remains high on the eastern Andean flank (Consuelo), and low in the west (Urpi Kocha; 52 m a.s.l.). 15 In the tropical lowlands along the Pacific coast, increases in the presence of palms (mainly Euterpe/Geonoma), are commonly interpreted as a result from more intensive forest use, e.g. Lake Piusbi (Behling et al., 1998a). Pollen grains from crops like Zea mays, Phaseolus and Ipomoea are found in Piagua (Vélez et al., 2001). Human disturbance to the forest is considered indicated by high percentages of abun-20 dance of Cecropia, ferns and palms. Decreases in human impact during the last 2 ka has been described by sites like Pitaliton, Timbio, La Genagra, Quilichao and La Teta, as grassy vegetation (Poaceae) and Zea mays disappeared and forest started to recover. This vegetation change could be related to the first arrival of the Spanish "conquistadors" (González-Carranza et al., 2012), or a set of different causes (Wille and Hooghiemstra, 2000).
Of the 42 pollen records identified from the Lowland Amazon Basin, 15 show evidence of pre-and post-European land use within the last millennia. Human land use is inferred from these records from cultigen pollen grains, charcoal and forest clearance Introduction  (Table 3). In some cases there is also archaeological and archaeobotanical evidence for human land use. At many of the sites occupied by native Amazonians, evidence of land use as a decline in burning by or before 0.5 ka, probably in relation to first European contact. However, some sites, such as French Guiana VII and Granja show evidence of continued post-European land use.

5
Pampa vegetation does not show evidence of human impact prior to European settlement at 0.4 ka. Europeans introduced several tree species (e.g. Eucalyptus, Pinus), as well as cattle (cow and horse) and crops (wheat, sunflower), but the intensive agricultural activities only began 0.05 ka (Ghersa and León, 2001). The paleoenvironmental history of shallow lakes shows a change to more productive systems (higher mass of 10 phytoplankton and organic matter content) during the last 0.1-0.08 ka probably due to agricultural activities. On the other hand, pollen records show an increase of pollen types associated with overgrazing (Plantago and/or Asteraceae Asteroideae) and exotic trees during the last 0.1 ka.
In S Andes and Patagonia, there is not conclusive evidence of native human activi- 15 ties in the pollen records and native-fire disturbance has been long discussed. Charcoal records from E Andes have not revealed fire activity associated with native populations. A probable explanation for this lack of evidence is a low density of populations associated with sporadic forest impact (Iglesias and Whitlock, 2014 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 5 Discussion of the regional assessments

General observations for 2 ka pollen compilations
This review reveals that those records with better dating resolution in the late Holocene (e.g. Granja, French Guiana K-VIII in the Amazon Basin) are often from cores that span a shorter time period, while longer temporal records have less well resolved 5 Holocene chronologies. This likely reflects (i) the need to spread limited numbers of radiocarbon dates in order to provide robust age models for these deeper time records, (ii) the greater interest of previous researchers in potential large-scale palaeovegetation changes, driven by glacial-interglacial climate cycles, and other significant periods of climatic change, such as the early-to-mid Holocene drought, (iii) the low sedimentation 10 rate during the last millennia in certain regions, e.g. lowland Amazonia. Furthermore strong anthropogenic interference during the last 2 ka complicates the interpretation of many records from a palaeoclimate perspective. Additional difficulties arise from the one topic focus of many studies and authors do not often present the full range of data in their publications that are required for a 2 ka reconstruction.

Venezuelan Guayana highlands and uplands
In this region, highlands and uplands are discussed separately due to their outstanding disparities in physiographical, climatic and ecological features, as well as in the intensity of human pressure on their respective ecosystems. Highlands are virtually pristine and, according to the paleoecological records, they 20 have remained in this state at least since the early Holocene. Therefore, climate has been the main driver of change. Paleoecological records for the last 2 ka are scarce and generally of low resolution but a common trait is the ecological stability as expressed in the vegetation constancy. The following hypotheses have been suggested to explain these observations: (i) environmental changes were insufficient to affect the highland Introduction  Briceño et al., 1990) have buffered climatic changes, and (iii) the study sites are unsuitable for recording significant vegetation changes because there are no vegetation ecotones nearby (Rull, 2015). Further work is needed focused on these proposals. So far, paleoecological fieldwork atop the tepuis has been carried out in an exploratory, non-systematic manner due to the remoteness of the tepuis, and the logistic and ad-5 ministrative constraints (Rull et al., 2008). In the LOTRED-SA framework, the issue of vegetation constancy emerges as a priority and should be addressed properly by finding suitable coring sites to be analysed with high-resolution multiproxy tools. The use of physical-chemical proxies independent from pollen and spores is essential to record climatic shifts, thus avoiding circularity. Lake sediments would be excellent for this purpose but, unfortunately, lakes are absent on tepui summits, the only permanent lake known so far is lake Gladys atop the Roraima tepui, of which age and origin remain unknown (Safont et al., 2014). At present, the analysis of the Apakará PATAM9-A07 core, which meet the PAGES-2 k criteria, is in progress. The preliminary study of this core showed the main Holocene vegetation trends at millennial resolution (Rull et al.,15 2011), and the current analysis is being performed at multidecadal resolution. A new core obtained in the Uei summit (PATAM8-A07; not included in the Chimantá massif) containing a decadal record for the last 2 ka is also being currently analysed (V. Rull, personal communication, 2015).
In the GS uplands, the situation is very different and the main driver of ecological 20 change is fire. This does not mean that eventual climatic shifts have been absent or that they have not affected the vegetation but the action of anthropogenic fires overwhelms and obscures the action of climate (Montoya and Rull, 2011). So far, regional paleoclimatic trends based on the Cariaco records have been used as a reference for the GS uplands (Rull et al., 2013) but a more local paleoclimatic record for this area 25 is still lacking and urgently needed, not only for the last 2 ka but also for the entire Holocene. Another limitation is that most paleoecological records available for the GS uplands are from its southern sector, which is the lowermost part of the peneplains, and has a different climate and vegetation regime as compared to the northern sec- tor. Some records from the northern sector are available that fit with the chronological PAGES-2 k requirements (Leal et al., 2011) but only summary diagrams are provided in peer-review publications and cannot be used in this reconstruction. The decadal to multidecadal analysis of a new core obtained in Kamoirán (PATAM10-A07), in the northern GS uplands, is in progress (V. Rull, personal communication, 2015).
It should be stressed that the last 2 ka seem to have been critical for the ecological history of the GS uplands and its detailed knowledge may be crucial to understand the origin of the present-day landscape. The reason is intimately linked to the temporal patterns of human impact using fire. The date of arrival of the current indigenous culture (Pemón) at GS is still unknown. Based mainly on historical documents, it has been postulated that this culture settled in GS ∼ 0.6 to 0.3 ka, coming from Guyana or Brazil (Thomas, 1982;Colson, 1985;Huber, 1995a). But these could be considered minimal ages, as recent palaeoecological studies suggest that human groups with landscape management practices similar to the Pemón people would have been present in the GS since ∼ 2 ka (Montoya and Rull, 2011;Montoya et al., 2011a). Before that time, the GS landscape was different from the present, including larger extents of forested areas since the Lateglacial and the absence of Mauritia palm swamps until ∼ 2 ka. The same time period seems to have been a landmark in neotropical history for similar reasons as Rull and Montoya (2014) showed a generalized increase of Mauritia pollen abundances in northern South America during the last 2 ka. 20 Given the northern position of the area, the vegetation responses studied have been normally related to ENSO and ITCZ movements and their consequences in the oceanatmosphere circulation patterns. These two main drivers are included in Niño 3.4, AMO, IPO and TNA modes, which are those acting in the area as shown in the Figs. 2-5 (especially regarding temperature). Previously the effect of SAM has hardly been con-25 sidered and the lack of AMO in the region regarding precipitation is surprising. It is noteworthy to compare the climatic inferences made through fossil pollen records and the climate modes effect on the area. Fossil pollen records have suggested available moisture (or precipitation/evapotranspiration ratio: P/E) as the main climatic driver to Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | take into account for vegetation responses. However, these inferences are based on very local spatial scale proxies (e.g. algal remains) and P/E is a complex process that relies in a wide range of factors, including both temperature and precipitation. Its interpretation in the fossil record is therefore complex and sometimes ambiguous. On the other hand, climate modes have showed the potential large effect of both temperature 5 and precipitation trends in the region. Such findings suggest that the variations of P/E inferred from the fossil record could be caused by either of these two factors, or by both. Additional higher resolution multi-proxy analyses should shed a light on previous undetected modes in the region as well as disentangling the combined effect of several forcing factors. Nevertheless, upland records have been interpreted as primarily 10 human-driven vegetation responses, so for the last 2 ka the climatic conclusions are constrained. Highland records have been described as an example of constancy, even insensitive to temperature change during the last 2 ka, which could confirm that the temperature driven modes in this region have been of a lesser magnitude than those required to cross the vegetation tolerance ranges or the intrinsic characteristics of the 15 sites studied so far inhibit detecting any change.

Northern Andes
Pristine regions have been not identified with certainty within this region, inhibiting a clear signal of climate tendencies in the last 2 ka. Drier conditions prevailed in Colombian savanna lowlands although the increased presence of Mauritia suggests either in-20 creased humidity or human influence. Along the Pacific coast, general wetter conditions prevailed ( Fig. 8b)  the N Andes until 0.45 ka and interdecadal variability during the last 0.5 ka, respectively. Also Pallcacocha in S Ecuador shows a close match with ENSO recording its strength during the last 15 ka. Similarly associated with ENSO are the changes in the plant assemblages detected in the high resolution record of El Junco on the Galápagos islands.

5
Comparing vegetation-climate signals between the Colombian lowlands and E Venezuela and NE Brazil has shown opposite climate conditions. Dry conditions identified in the Colombian savannas (suggesting an ENSO -La Niña), concur with similar conditions in the Bolivian pollen records. During an El Niño setting, when Bolivian savannas indicated wet conditions, the signal from Lake Valencia in Venezuela 10 reflected dry conditions (Wille et al., 2003). Lowland sites generally show similar patterns of climate change during the last 2 ka and apparent synchronous events are observed over a larger spatial scale. The sites in the Andean region are much more influenced by local geographical variability, causing a more variable response mechanism.

Central Andes
The records of the Altiplano suggest an oscillation in moisture availability (precipitation) on a multi-centennial timescale during the last 2 ka (Fig. 9). These oscillations are probably due to the strength of the summer precipitation. The timing of wet and dry events is not uniform between sites probably due to local micro-climates and differ-20 ences in vegetation sensitivity to climate change (the high elevation grassland (Puna) vs. mid elevation Andean forest). The archaeological site of Cerro Llamoca provides the most robustly dated record (33 radiocarbon ages); however, the strong local human influence means that climatic interpretations of palaecological evidence should be done with caution. Records of glacial advance and retreat from the Altiplano record 25 variations associated with the LIA, but in all records, apart from Llamoca, any climatic impact of the LIA on vegetation is masked by the European conquest effects (abandonment of the sites, and/or changes in agricultural practices). Interpretation of the climate signal from the C Andes fossil pollen records suggests that during the last 2 ka precipitation, rather than temperature, was the key natural driver of vegetation change. However, modelling suggests that on an annual scale temperature (Figs. 2 and 3), rather than precipitation (Figs. 4 and 5) is more likely to have altered due to switch in climate mode; particularly in the western part of the C An-5 des. The increase in temperature observed at Coropuna during the Inca period, after 0.85 ka, could correspond to El Niño or IPO forcing and the decrease in temperature observed at Marcacocha between 1.85 and 0.85 ka could be related to La Niña. Similarly these modes, show a high influence on the coast which is in agreement with the results of the coastal pollen record (Urpi Cocha) where ENSO is considered responsible for extreme flooding events. Also the TSA mode shows a strong influence on the Bolivian Altiplano. The greater sensitivity to precipitation seen in the pollen records is probably because moisture availability is in most areas the limiting factor for both vegetation communities and human populations. The increase of temperature induced by Niño and IPO on the Figures 2 and 3 show no link with the precipitation. On the Altiplano 15 variation in the SASM has been attributed as a major driver of changes in moisture balance at Llamoca, Coropuna, Pacucha, Consuelo through altering the summer precipitation. Precipitation patterns (Figs. 4 and 5) are less pronounced as it occurs here during a short seasonal period of 2-3 months, hence the used month average weakens the correlation. Notwithstanding, ENSO has been shown to have significant 20 influence in the region (both temperature and precipitation) in numerous studies.

Lowland Amazon Basin
Figures 2-5 reinforce the importance of the SAM in Amazonia, a pattern that has been identified and discussed extensively in the palaeoecology literature of lowland Amazonia. The SAM is recognised as a key driver of precipitation and attributed to the 25 expansion of rainforest during the late Holocene, which is observed in the large lake records from S Amazonia (Chaplin, Bella Vista, Orícore, Carajás). However, these sites do not meet the PAGES-2 k dating criteria. As for its effect on tempera-3514 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ture, in this region vegetation records are probably not sensitive to small temperature changes (e.g. 0.5 • C) that might relate to SAM strength. The sensitivity of the records could be related to the temperature and precipitation ranges of tropical plant families (Punyasena, 2008;Punyasena et al., 2008). Figures 2-5 also show considerable spatial complexity in climate over lowland Amazonia, in terms of the impact of different 5 climate modes, especially ENSO. However, if modes such as ENSO do have a long term drying effect over the past 2 ka in lowland Amazonia, it does not appear to affect vegetation in a way that is visible in the pollen records that we have reviewed. Given the size of the region and the relative sparseness of sites, it is perhaps unlikely that such spatial complexity would be captured.

10
The better-resolved late Holocene records tend to come from small lake basins (e.g. oxbows), which have small pollen catchment areas. This means that they reflect predominantly local-scale changes and are, therefore, more susceptible to having their palaeo-record dominated by signals of ancient human land use and local hydrology (e.g. savanna gallery forest), rather than regional climate. Many of these smaller 15 records were specifically selected in the original study to investigate local-scale human impacts around known occupation records (Iriarte et al., 2012;Whitney et al., 2014;Carson et al., 2014Carson et al., , 2015. In order to address these complicating factors of pollen catchment area and the anthropogenic signal, any future effort to obtain better-resolved Holocene pollen records 20 in the lowland Amazon should make careful consideration of the sampling methodology employed. Carson et al. (2014) demonstrated that sampling a combination of small and large lake basins from within the same catchment allows a distinction to be made between local-scale, anthropogenic impact and regional-scale, climate-induced vegetation changes. In regions such as the C Amazon, where lakes are predominantly 25 limited to small oxbows, an sampling approach might be to analyse cores from multiple records within the same locality, and to compare those records, in order to identify any regionally significant pattern of palaeovegetation change (Cohen et al., 2012;Whitney et al., 2014). Oxbow lakes are dynamic features, and so require careful interpretation. Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | However, their higher sedimentation rate means that they have the potential to provide the high temporal resolution palaeovegetation records of the late Holocene, which currently are largely absent from the Amazon lowlands.
Considering the large area of the Amazon Basin, the number of pollen records is very small, and by applying the PAGES-2 k criteria, those numbers are further reduced. Furthermore, the records which are excluded from the analysis by these criteria include some of the most important records of climate-driven vegetation change in the Amazon basin, e.g. Lakes Orícore (Carson et al., 2014), Carajás (Hermanowski et al., 2012), and lakes Bella Vista and Chaplin (Mayle et al., 2000).
In order to avoid a "black hole" situation over the Amazon lowlands in any regional 10 synthesis, one approach may be to apply a lower threshold of dating criteria. If the selection criteria are relaxed to allow for those records that are > 500 years old and have at least two chronological control points within the last 2000 years, a further 14 records are added to the list of qualifying records. Also, if the criteria are stretched further to allow records with a lower date which is older than, but close to 2 ka, the Lake Chaplin 15 and Gentry records would also be included. Including these records would provide coverage from the central Amazon river region, the N Brazilian Amazon, the E and NE coastal Amazon and the SE and SW basin. However, even with these relaxed criteria, a number of key records would still be excluded, e.g. Pata (Bush et al., 2004;D'Apolito et al., 2013), La Gaiba (Whitney et al., 2011) and Bella Vista (Mayle et al., 2000). 20 Any future investigation of late-Holocene climate-vegetation interaction may require new dating efforts to improve the age models of these key records. A Holocene aged record from Laguna La Gaiba produced by McGlue et al. (2012) has produced a better-resolved age model than the longer record from Whitney et al. (2011), which would meet the PAGES-2 k criteria. However, McGlue et al. (2012) analysed the geo-25 chemical properties of sediments from a new core taken after the Whitney et al., (2011) study, and did not include any pollen data. No attempt has been made subsequently to correlate the chronologies of the two records. Although the dating resolution in the late Holocene is poor in many lowland Amazonian pollen records, it should be noted that the majority also show little variation in vegetation over the past ∼ 1 or 2 ka. Whether this reflects genuine ecosystem (and climate) stability over the late Holocene, or is a product of low sampling resolution within these long records is unclear. Most of these deep temporal pollen records, as they 5 are published now, likely have sub-sample intervals of insufficient resolution to be able to discern high-frequency events, such as vegetation changes associated with ENSO variability. However, in some cases, such as Bella Vista and Orícore, the potential for such fine temporal reconstructions may be limited by the low sedimentation rate of the basins. Often these records come from short sediment cores, in which the 10 Holocene time interval is contained within a short depth range (i.e. < 1 m). A number of shorter records, spanning Holocene time periods, exist in the eastern coastal Amazon, and could potentially provide high temporal-resolution reconstruction over the last 2 k. However, most do not currently meet the PAGES-2 k dating criteria.

15
The limited number of pollen records from S/SE Brazil for LOTRED-SA-2 k has several reasons besides the insufficiently dated cores: (1) many archives, in particular peat bogs have a very low sedimentation rates. Often 100 cm of peat deposits cover already the complete Holocene and the last 2000 years have relative small amount of deposits, (2) the upper part of peat archives actively growing roots and it is difficult to date. 20 However, general vegetation changes in S/SE Brazil can be explained by a change to wetter conditions, in particular due the reduction of the dry season length. This is generally reflected in SE Brazil between 6 to 4 ka, and in particular strong in S/SE Brazil during the ∼ 1 ka. Several new records for S/SE Brazil are in process or to be studied in the next years. These records will most likely present improved chronology for the last CPD 11,2015  of the continent display a relatively simple pattern within this region as the degree of overlap is minimal.

Pampa plains
There are several pollen records in Pampa plains that span Holocene times, but few of them have well resolved chronologies for the last 2 ka. Just one site fulfils all PAGES-5 2 k criteria. Conventionally, pollen analyses in the region were carried out in alluvial sequences or archaeological sites which usually contain sedimentological discontinuities that impede a good chronological control. These pollen records show regional vegetation changes and climate inferences related to precipitation changes (humid/dry/arid conditions) or sea level fluctuations, mainly at millennial or centennial scale. Until today, few studies has been focused on elucidating palaeoenvironmental changes at high temporal resolution during the last 2 ka. Furthermore, the Pampa plains has a high number of potential sites; shallow lakes characterized by a continuous sedimentation that would provide robust age models and high quality pollen records. Conversely, the current pollen records do not have the necessary resolution to identify vegetation- 15 human interaction during the last 0.3 ka and therefore improved chronological control higher resolution is necessary. General climatic tendencies in the region can be inferred although the few accurate pollen records are available. While individual palaeoecological studies reveal local developments, general patterns emerge when information from several sites is combined 20 together, such as Lonkoy and Hinojales-San Leoncio (Fig. 12b). A multiproxy approach, including pollen analyses, shows synchronic changes in these shallow lakes from SE Pampa that are mainly a response to precipitation variations. Thus, between 2 and 0.5 ka drier conditions than present are inferred, then a transition phase towards more humid conditions is observed which stabilizes between ∼ 0.3 and 0.1 ka, with 25 values close to modern (Stutz et al., 2015). These climatic inferences are valid for the SE region but do not extend to the entire Pampa plains. At S Pampa plains, multiproxy interpretation at Sauce Grande shows a similar change to more humid conditions 3518 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | at 0.66 ka, and similar conditions to present day after 0.27 ka, but pollen composition shows low responsiveness to change (Fig. 12a). As seen in Figs. 2-5, these plains falls outside the areas influenced by strong climate modes expressions. New paleoenvironmental reconstructions based on pollen records are needed to disentangle the intrinsic ecosystem variability from climate, and to elucidate if climatic events as MCA or LIA 5 had different expressions in the Pampa region.

Southern Andes and Patagonia
Even though a high number of pollen records are available in the region, just 19 (between 32-54 • S) fulfil the PAGES-2 k criteria. In Patagonia most of pollen studies have been carried out on long temporal records (in many cases until end of Last Glacial Max-10 imum) focusing on the Pleistocene-Holocene transition or the entire Holocene vegetation and climate dynamics. Moreover, appropriate records in areas of northern S Andes and Patagonia, are scarce because of the absence of depositional sites and archives or the lack of palaeo-research.
The pollen records are considered to mainly reflect the SWWB. Southern records 15 receive precipitation related to the SWWB, whereas those located to the north (40-32 • S) are also influenced by the Subtropical Pacific Anticyclone (SPA) that blocks winter precipitation in a latitudinal gradient (decreasing precipitation during JJA in the S part to scarce precipitation during DJF in the N part). southernmost Patagonia arises as a key area to study climate-vegetation variability associated to SAM (e.g. Lago Cipreces). Sites in N Patagonia and C Chile reflect synchronicity with ENSO activity (e.g. Lagos San Pedro and Aculeo) given the relationship between high precipitation/El Niño phase and low precipitation/La Niña phase (Montecinos and Aceituno, 2003). LIA and MCA chronozones are well recorded both in 5 southern and northern Patagonia (e.g. Lagos Cipreses, Peninsula Avellaneda Bajo, San Pedro), however not in central Chile. The region cannot be explained by a single climate unit due to the presence of different climatic modes and forcing. Different patterns are distinguished (Fig. 13b), due to their geographical position, latitude and E/W side of the Andes, and intrinsic sensitivity of each record to climatic variability.
For example, wet conditions are inferred in N Patagonia between ∼ 1.5-0.3 ka whilst in S Patagonia dry conditions dominated with increased variability.

Synthesis and conclusions
Although the number of pollen records suitable for a 2 ka climate reconstruction is still relatively small compared to the total potential number of records available from South

15
America (Flantua et al., 2015a), an increasing number of records are showing potential for inclusion in PAGES-2 k. From reviewing c. 180 pollen records from South America we conclude that: the lack of South American records that fulfil PAGES-2 k criteria is an issue of short vs. long records. Most of the more important records have been ruled out 20 because they turned out to be "old" and based on the sedimentation rates of the sedimentary archives, radiocarbon samples within the last two millennia are mostly missing.
-Pollen records detect long-distance synchronicity (differences and similarities) in vegetation changes as an indication of regional precipitation and temperature vari- 25 ability, but they also detect the local-scale change/variability. Diverse patterns of 3520 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | vegetation response to climate change are observed, with more similar patterns of change in the lowlands and varying intensity and direction of responses in the highlands. Hence pollen records can serve as integrating proxies over long distances and allow assessing changes in the large-scale atmospheric circulation. Therefore they can serve as powerful archives, which can help better understand 5 past changes in the strength and area of influence of climate modes such as ENSO or the AMO.
-Throughout South America a number of overlapping climate modes operate, meaning that every single pollen record most likely captures the signal of various modes (Figs. 2-5), although they do not all operate in the same frequency 10 bands and modes interact with one another through constructive interference. The causes of ambiguous climate-vegetation responses observed in pollen records can therefor probably be best ascribed to the degree of climate mode interaction at a location.
-Depending on the geographical and altitudinal location a record may be more 15 sensitive to temperature-or precipitation-related forcing (Figs. 7-13). The baseline for understanding climate-driven changes in vegetation is related to either of these variables, but interpreting pollen records in terms of a response to largescale climatic forcing may yield further insights as it allows for an attribution of, temperature-and/precipitation-driven changes to forcing from climate modes 20 originating in either from Atlantic or Pacific.
-For human impact studies, it is important to consider a set of different proxies to make more confident interpretations in terms of climate vs. human drivers of vegetation changes.
-There is a need to increase efforts in high-resolution studies with accurate 25 chronology for the last 2 ka. -The PAGES-2 k criteria should be adjusted for pollen records, especially by applying a lower threshold of dating criteria. A region such as the lowland Amazon is notorious known for its paucity of records with good dating (e.g. Ledru et al., 1998). Therefore the few valuable sites available should be considered for the overall purpose of understanding vegetation-climate linkages.

7 Recommendations
Below we list a few specific recommendations for future engagements between climate and pollen related studies: 1. quantitative translation from pollen metrics to climate variables: assembling a meaningful multi-site and multi-proxy dataset is hampered by the current gap between the palynological and the climate dynamics and modelling community, both in terms of interpretation and quantitative translation of pollen data into climate indicators. This gap can be narrowed when pollen studies provide -if the data is suitable for that purpose -their own temperature or precipitation approximations. There are only a few pollen studies that provide a quantitative interpreta- Multi-proxy based research should become a mandatory goal for all further investigations. Caution should be exercised when interpreting apparently contradictory records provided by different groups for the same region; the interpretation of climatic and anthropogenic signals in each record may be based on very different (indirect) proxies. Hence the apparent asynchronies or contradictory interpreta-5 tions could simply occur as a result of methodological artefacts (e.g. by not including charcoal records, non-pollen palynomorphs, geochemical analyses, etc.). On the other hand, this is especially relevant for those areas where human impact has been found for the last 2 ka and any climatic interpretation is aimed.

For the stated purposes of the current and future PAGES initiatives, researchers
should be motivated to further improve chronologies for existing sites. Further advances in understanding climate-human relationships are also likely to be made by the integration of palaeoecological and archaeological data (e.g. Mayle and Iriarte (2014) through conceptual modelling, which can provide a framework for identifying patterns and trajectories of change (e.g. Gosling and Williams, 2013).  5. All Andean zones are quite active from tectonic and volcanic points of view, and those drivers will have had huge impacts on the vegetation and maybe in the fossil pollen records signal. However, this feature has been only included in the southern region of Andes. A chronology database focused on tephra control points could support current chronology constraints and improve comparison between 25 records. The recent geochronological database of the LAPD can support multiproxy approach for paleoecological integration (Flantua et al., 2015b).

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version