To obtain an overview of the control points and age modelling methods used in pollen records throughout the region, we performed a thorough review of the LAPD and corresponding literature database (Flantua et al., 2015). A total of 1245 publications were checked regarding their chronological information covering 1369 sites. For 270 sites only biostratigraphic dates were mentioned, no chronological details were provided, or the original publications with specifications were not found. These sites originate primarily from the 1970s and the 1980s, although even some recent publications lack details on the chronology. All other sites consisting of at least one chronological reference point enter the geochronological database at this stage (Fig. 1). The following chronology metadata were collected for each site: Site Name, Year of Data Preparation, Age Model, Calibration Method, Software, Material Dated, Depth (min, max, mean), Thickness, Laboratory number, pMC (error), 13C adjusted (±standard deviation), 14C date (min, max, errors), Reservoir correction, Calibrated age (min, max, best age, errors), Additional relevant comments from authors. Furthermore, all additional parameters needed to correctly reconstruct the chronologies, such as presence of hiatus, slumps, contaminated control points, and other outliers identified by authors, were included. As a result, the Neotropical Geochronological Database (Neotrop-ChronDB) currently contains a total of 5116 chronological dates from 1097 sites throughout the study area.

From the Neotrop-ChronDB, all sites present in Venezuela, Colombia, Ecuador, Peru, and Bolivia were extracted (Fig. 1, countries in grey). Over 300 publications were consulted to recalibrate control points and rebuild age models of 234 pollen records (Table 1). When more than one chronological date was available, new chronologies were generated with the updated calibration curves for the northern and the southern hemispheres, and maintained as closely as possible to the authors' interpretation of the age model. New chronologies were generated with updated calibration curves to (a) be able to implement the temporal uncertainty analysis (the “star classification system”); (b) to provide age models to studies without chronologies; (c) to provide alternative age models for records based on older calibration curves or Southern Hemisphere records using the northern hemispheric calibration curves; (d) to estimate the temporal resolution of pollen records in general and at specific time windows of interest in NW-SA.

The most common control points are radiocarbon dates. For the age model generation we included the reported uncertainty of a date regardless of its origin (conventional or Accelerator Mass Spectrometry, AMS). Additional important control points in constructing chronologies are ages derived from tephra layers from volcanic material, radioactive lead isotopes (210Pb) and fission track dates.

For the generation of the recalibrated age models, stratigraphic dates were not used. Use of these layers would ignore the possibility that for example the palynologically detectable onset of the Holocene was asynchronous throughout northern South America. Therefore any further inferences on spatial leads, lags, or synchroneity would become flawed. Only in very few cases were very recent time markers used like the introduction of Pinus.

The non-“decapitated” top of the sediment sequence can be assigned to the year of sampling, if explicitly mentioned by the authors as the result of being the youngest sample in an undisturbed way. Frequently, however, assigning depths to core tops adds a factor of uncertainty because the uppermost sediments have not been consolidated and can be lost during coring. We did not use most of the estimated core tops as additional ages, but as with the bottom ages, let the recalibrated age model produce the new ages of the core tops. In case of considerable extrapolation or heavy overshooting of the age model (very young top ages), we produced alternative age models including the estimated top age. We decided to use the uncertainty range of ±50 years considering that this standard deviation results in c. 300 years of total uncertainty. We consider this value an appropriate estimate of uncertainty of core top ages. As the R-code of the procedures here presented is made available, researchers may adjust this value accordingly. Extrapolations from the new chronologies that went beyond -50 cal yr BP (years before AD 1950) were not used for the estimates on resolution.

The South American continent covers the Northern Hemisphere (NH) as well as the Southern Hemisphere (SH). The previous SH calibration curve (SHCal04) only extended to 11 thousand calibrated years before present (here abbreviated as kcal BP). In age model tools like CLAM (Blaauw, 2010), options were provided to “glue” the NH calibration curve to the SH curve to extend back to 50 kcal BP. However, recently the SH calibration curve was extended to 50 kcal BP (Hogg et al., 2013) and now obviates the need to use the NH curve for older dates in the SH. This provides new opportunities to recalibrate age models with updated calibration information and produce additional sample ages for reevaluation. Nevertheless, tropical regions still face an uncertainty factor open to discussion, namely the southern limit of the Intertropical Convergence Zone (ITZC). McCormac et al. (2004) defined this limit to be the boundary between the NH and the SH, but models need additional data to better determine its exact location through time (McGee et al., 2014). For internal consistency we assigned the curve according to the general delimitation by Hogg et al. (2013) and Hua et al. (2013), or used the preferred calibration curve by the authors for the creation of the chronology. Mayle et al. (2000) for example, explicitly explain why their site in the Bolivian Amazonia experiences NH influences. Finally, a total of 22 sites include post-bomb dating for which five different regional curves options exist (Hua et al., 2013). Post-bomb calibration curves were as used by original authors or assigned according to Hua et al. (2013).

Depending on the number of available control points, two age–depth models were created per site. All age–depth relationships were reconstructed using the R-code CLAM version 2.2 (Blaauw, 2010; R Development Core Team, 2014), which is an R code for “classic age-modelling” (Blaauw and Heegaard, 2012). The simplest age model, namely the linear interpolation method, produces a straightforward interpolation. It connects individual control points with straight lines which is in most cases unrealistic as it assumes abrupt changes in sedimentation rates at, and only at, the dated depths in the sediment core. The second age model method we used is the smoothing spline, with a default smoothing factor of 0.3. This interpolation method produces a curve between points that is also influenced by more distant control points. This method provides a smoother outline of age model and is considered to produce a more realistic model of the sedimentation process compared to the linear interpolation method. However, smoothing splines can only be modelled at sites that present four or more control points. Furthermore, age models were not run on cores that were problematic from the start. Examples are: cores where a hiatus/slump disrupts the age model in a way that no linear interpolation is possible; cores with many age reversals (when an older date lies above a younger date with limited dates collected); and cores with many nearly identical radiocarbon dates regardless of depth. Studies using tuning methods to establish their age models were not included.

The sample depths were derived from either the raw data set provided by the authors from the original paper or from the specifications and figures in the original publication. In a few cases, neither were available, so a 10 cm sample interval was assigned based on our assessments of the most likely depths for such dates. The sample age is obtained as the highest-probability age based on the distribution of estimated ages from 1000 Monte Carlo runs and the uncertainties are provided as 95 % confidence intervals.

For each site, the newly produced models were evaluated and if necessary adjustments were made to deal with obvious outliers, “overshooting” of the age model towards the top, and degree of “smoothness” of the smooth spline model. Outliers were identified visually when control points deviated excessively from the general depth–age tendency. To solve over-extrapolation at the top (future dates), additional age models were created that included estimated surface dates. In some cases the default smoothing level of 0.3 was adjusted to “touch” more of the available dates or to avoid an age reversal in the model. The most appropriate age model was selected in accordance to the authors' description, with a general preference for the smoothing spline model. With this model, we calculated the multi-site summary values, such as overall resolution and star classification system.

The original data, the R-codes and the recalibrated age models from this paper are available through: https://figshare.com/s/0e9afb8fe758a0e6e8c8. We provide a manual that explains step by step the setup of the data and the use of the codes. For each individual pollen record, the corresponding folder contains the description of the original age model (copyright prevented the inclusion of pictures/figures), details on the recalibrated age models and the outcomes of the star classification system at sample level.

We followed the age model evaluation proposed by Giesecke et al. (2014) to define the temporal quality and uncertainty of the chronologies and individual samples. An uncertainty classification based on assigning semi-quantitative “stars” focuses on the density of control point. The classification is additive and samples are assigned to the lowest class (a single star) where the estimated sample age is within 2000 years of the nearest control point. Additive stars are given at 1000-year and 500-year proximity to the nearest control point (Table 2). In addition to the three stars that characterize proximity to the nearest control point, an extra star is given to samples that are situated in a straight section of the sequence. The “straightness” star is given to a sample where, within the nearest four control points, the modelled sediment accumulation rate changes less than 20 %. Only sequences with at least four control points can obtain such an additional star. The evaluation is based on the position of the sample relative to the control points and is independent of the interpolation procedure. Therefore stars are assigned to the smooth spline output unless insufficient control points are available. The outcome of this classification produces a text file with the assigned number of stars for each sample along the core that is based on the depth file. The star classification is visualized along the vertical axis of the age model with coloured symbols (Fig. 2).

Rapid events of climate change occurred during the Dansgaard–Oescher (D–O) cycles spanning the last glacial cycle and during the Holocene. Recently published pollen records, like at Lake Titicaca, Bolivia (Fritz et al., 2010) and Lake Fúquene, Colombia (Groot et al., 2011) show clear evidence of millennial climate variability of large amplitude during Marine Isotope Stage (MIS) 4 to 2. As an example of the implementation of the star classification system, we select a series of consecutive time windows relevant for paleoclimate reconstructions at millennial timescale. These time windows are: MIS 5 (c. 130–70 kcal BP), MIS 3 (c. 60–27 kcal BP; Van Meerbeeck et al., 2009), Heinrich event 1 (H1; c. 18–15 kcal BP; Álvarez-Solas et al., 2011), and the Younger Dryas (YD)/Holocene transition (c. 12.86–11.65 kcal BP; Rasmussen et al., 2006). For these time windows we summarize and discuss the temporal resolution and control-point density (the star classification system).

The number of available pollen records in this region has increased considerably in the last 20 years (Flantua et al., 2015). During recent years, the number of control points used for stratigraphic age models has trended upwards; since 2010, the mean and median number of control points per published pollen site has been five and three, respectively (Flantua et al., 2015). Here we provide more detail on the available chronologies, describing the most commonly used control points for dating, age modelling, and calibration methods.

The Neotrop-ChronDB stores a total of 5116 dates of which the most common control points are radiocarbon (14C) dates. Radiocarbon dating has been used to date pollen records for more than five decades now. The first dated records in South America came from the Orinoco delta of Venezuela (Muller, 1959), and from Colombian sites such as Ciudad Universitaria, Laguna de la América, and Páramo de Palacio (Van der Hammen and González, 1960) and Laguna de Petenxil in Guatemala (Tsudaka, 1967). In the early stages of 14C measurement, this technique required a minimal sample size of 0.5 g carbon (Povinec et al., 2009), while sample sizes differed greatly among materials (Bowman, 1990). In paleoecological research, this has always been a limiting factor as natural samples generally present a small 14C / C ratio. As a consequence, material to obtain a 14C date sometimes originated from a wide depth interval of the sediment core. Consequently, conventional radiocarbon dating based on bulk samples of lake sediments is often a high-risk undertaking as it can result in a substantial uncertainty and puzzling date estimates.

The great breakthrough came from the development of AMS dating in 1977 that consisted of direct counting of the 14C atoms present in a sample (Bowman, 1990; Povinec et al., 2009). This technique reduced the requirements for sample size and therefore improved the accuracy of samples. Furthermore, the required time to obtain dates was reduced from months to minutes. It took some time for AMS dating to appear in the Neotropics. It was not until the early 1990s that AMS dating was used in sites as Lake Miragoane, Haiti (Brenner and Binford, 1988), Laguna de Genovesa, Ecuador (Steinitz-Kannan et al., 1998) and Lake Quexil, Guatemala (Leyden et al., 1993). Ever since, an increasing number of sites report AMS dates to support their chronologies with higher precision. Nevertheless, even in a recent record with AMS ages, authors have been struggling to compile a consistent age model due to low carbon content of the samples (Groot et al., 2014). The advantages of using 14C as a dating method, having broad applicability on many different sample materials and covering the most prevalent time range (50 kcal BP), mean that it surpasses other methods and therefore remains to be the most commonly applied scientific dating method.

Currently c. 68% of the geochronological dates in the LAPD fall within the last 10 kcal BP, 20% within 20–10 kcal BP, and 4% within 30–20 kcal BP. A wide range of materials is used for dating: cellulose-containing materials (woods, seeds, achenes, plant remains, insect chitin; n= 1732); charcoal and charred material (n= 191); carbonates (shells and calcite; n= 118), collagen-containing materials (bones and coprolites; n= 48); and bulk sediments from different materials (n= 1074).

The terminology Tephrochronology means “use of tephra layers as isochrons (time-parallel marker beds) to connect and synchronize sequences and to transfer relative or numerical ages to them using stratigraphy and other tools” (Lowe, 2011). The process of obtaining a numerical age or date for a tephra layer deposited after a volcanic eruption either directly or indirectly is called Tephrochronometry (Lowe, 2011). Primary minerals, such as zircon, K-feldspar, and quartz, can be used to date tephras directly. Indirect methods include different applications such as radiometric dating (radiocarbon dating, fission-track dating, argon isotopes K/Ar, Ar/Ar, luminescence dating, U-series, 238U/238Th zircon dating) and incremental dating (annually banded found in the layering of ice cores; Lowe, 2011). This field of advanced chronology is of essential importance in the search for precise dates for high-resolution paleoenvironmental records and research (Davies et al., 2012). Tephrochronology has become increasingly popular across a range of disciplines in the Quaternary field (Bronk Ramsey et al., 2015; Lowe, 2015), especially for linking and synchronizing paleorecords accurately along longer timescales. Several uncertainties in tephrochronology are similar to those known from radiocarbon dating such as methodological and dating errors, and reworking of dated layers. The specific challenges for this dating technique lay in that different tephras may display similar major element composition, or the same tephra may have a temporal and spatial compositional heterogeneity (for a review and examples see Lowe, 2008, 2011). International initiatives such as INTIMATE (http://intimate.nbi.ku.dk/) and INTREPID (Lowe, 2010) have aimed at improving uncertainties from tephrochronologies, supported by an expanding global database on tephra layers (http://www.tephrabase.org/). Although not extensive, we provide here an overview of studies that welcomed this technology to improve the chronologies of their pollen records.

From Mexico down to Patagonia, there are regions of elevated volcanic activities where frequent tephra layers can be found. Mexico's active seismic zones have numerous active volcanoes in the so-called “Mexico's Volcanic Axis” or “Trans-Mexican Volcanic Belt” (Eje Volcánico Transversal). Ortega-Guerrero and Newton (1998) collected tephra layers in southern Mexico specifically aimed to produce stratigraphic markers for palaeoenvironmental research. Tephra layers called Tlácuac, Tlapacoya, and Toluco can be found in different pollen records such as Lake Texcoco (Lozano-Garcìa and Ortega-Guerrero, 1998) and Lake Chalco (Lozano-Garcìa et al., 1993). Additional tephra layers played an important role in the chronology of Lake Peten-Itza PI6, Guatemala (Hodell et al., 2008) and Laguna Llano del Espino and Laguna Verde, El Salvador (Dull, 2004a, b).

The northern Andes forms part of the “Northern Volcanic Zone” (Stern, 2004; Rodríguez-Vargas et al., 2005) and is shared by Colombia and Ecuador. In the Ruiz-Tolima region (Central Cordillera of Colombia), Herd (1982) identified 28 eruptive events during the last 14 000 years. Sites like Puente Largo and Llano Grande (Velásquez et al., 1999) make use of these events in their chronologies. Even sites along the Eastern Cordillera capture these volcanic ashes, like Funza (Andriessen et al., 1994; Torres et al., 2005) and El Abra (Kuhry et al., 1993), while the ridge itself lacks volcanic activities (Rodríguez-Vargas et al., 2005). Otoño-Manizales Enea (Cleef et al., 1995) reports five events between 44 and 28.5 thousand calendar years (kyr) BP and Fúquene another six events between 30 kyr and 21 kyr BP (Van Geel and Van der Hammen, 1973). Fission-track ages on sparse zircons were obtained for the long cores from Funza 1, Funza 2, Rio Frío, and Facatativá (Andriessen et al., 1994; Wijninga, 1996).

Ecuador is also well known for its very active volcanic region. Two eruptions of the Guagua Pichincha and one of the Quilotoa were seen at pollen site Papallacta (Ledru et al., 2013b). Thanks to four radiometric 40Ar–39Ar dates from tephra deposits, the chronology of the Erazo pollen record was placed within the middle Pleistocene period (Cardenas et al., 2011). An important overview of tephrochronology in southern Ecuador was provided by Rodbell et al. (2002).

The central Andes forms part of the “Central Volcanic Zone” (Stern, 2004; Rodríguez-Vargas et al., 2005) and is shared by Peru and Bolivia. Several ice cores from the Sajama Ice Cap in Bolivia use ash layers from Volcán Huaynaputina in Peru as dating control (Reese, 2003). To support the chronology of the long core of Lake Titicaca, nine aragonite-rich layers for U/Th supported correlation with the last interglacial period (MIS5e; Fritz et al., 2007).

Finally, towards the south, the “Southern Volcanic Zone” covers Chile and Argentina (Stern, 2004). An overview of the Holocene tephrochronology of this volcanic zone is presented in Naranjo and Stern (2004). The Pleistocene–Holocene transition has shown similarity in timing with an increase in volcanic activity in southern Chile (Abarzúa and Moreno, 2008). Jara and Moreno (2014) assessed the potential of volcanic events as being a driver of vegetation changes at a (sub-) millennial timescale based on 30 tephra layers since 13.5 kcal BP. Other sites with tephras to support their chronology are at Puerto del Hambre in Chile (Clapperton et al., 1995) and Rio Rubens in Argentina (Markgraf and Huber, 2010), among others.

Before dating by 14C became available and more affordable, many records relied on the identification of biostratigraphic zones. Biostratigraphy is a branch of stratigraphy based on the study of fossils (Traverse, 1988; Bardossy and Fodor, 2013). Delimitated zones were interpreted as sequences of rocks that are characterized by a specific assemblage of fossil remains (Gladenkov, 2010). Each zone is a reflection of changing paleoecological settings different from the previous zone, identified by a set of characteristics such as taxon composition or abundance, or phylogenetic lines (Gladenkov, 2010). In general, stratigraphic schemes are still subject to constant adjustments, being updated by new records, improved dating, and taxonomic revision. Difficulties arise in the accurate delimitation of the boundaries of biostratigraphic zones. Furthermore, older records relied heavily on zonal matching without accurate chronological background and assuming synchronicity. Additionally, the zonation and biostratigraphy may depend on localized stratigraphic nomenclature and is sometimes not even directly applicable to adjacent areas. Finally, a biostratigraphic layer may have been defined using a sparse data set while depending heavily on correct taxonomy identification. Challenges of biostratigraphic correlation techniques are further explored in Punyasena et al. (2012) and Barossy and Fodor (2013).

Several biochronological schemes are used or under discussion in South America and describing their development (e.g. Van der Hammen, 1994; Van der Hammen and Hooghiemstra, 1995a) goes beyond the scope of this paper. Here we mention briefly some zones for NW-SA. Older records used presumably synchronous onsets of the Lateglacial as a reference point in time, such as numerous pollen records from the Valle de Lagunillas (González et al., 1966), Sierra Nevada (Van der Hammen, 1984), and Central Cordillera (Melief, 1985; Salomons, 1986). The transition of the Pleistocene/Holocene is often mentioned in diagrams, as is the YD. The onset of the Bølling/Allerød is less frequently used, whereas referring to and correlating regionally defined stadials and interstadials is more popular. For example, the “Guantiva interstadial” (Van der Hammen and González, 1965a; Van Geel and Van der Hammen, 1973) and “El Abra stadial” (Kuhry et al., 1993; Van der Hammen and Hooghiemstra, 1995b) are commonly used biostratigraphic dates within Colombia. These periods are considered to be an equivalent to the North Atlantic Allerød Interstadial and the Younger Dryas sequence, respectively (van der Hammen and Hooghiemstra, 1995b). Similarly in the tropical Venezuelan Andes, the “Anteojos” cold phase was proposed as equivalent to the cold reversal of the YD and as in some aspects comparable to El Abra (Rull et al., 2010).

An exceptional dating method was used at Ciama 2 in Brazil, through Optically Stimulated Luminescence (OSL) encompassing the period between the MIS3 (MIS5 ages were discarded) and the last millennium (de Oliveira et al., 2012). The same technique was used at the Potrok Aike lake in Patagonia. A 65 kyr long sediment core was recovered by the Potrok Aike Maar Lake Sediment Archive Drilling Project (PASADO; Recasens et al., 2012), where a combination of OSL, tephra, and 14C was used to establish its chronology (Buylaert et al., 2013; Recasens et al., 2015). The pollen record from this multi-proxy study is to be published soon and will be an important comparison to other long cores from South America regarding late Quaternary climate variability.

There are two important records that serve in South America as a key reference for regional chronology testing, which are Fúquene-9C (Groot et al., 2014) and the MD03-2622 marine core from the Cariaco Basin (González et al., 2008). Both cores were analysed at high resolution (Fq-9C: 60 years; Cariaco: 350 years) and cover c. 284–27 and 68–28 kcal BP, respectively. Both sites, however, implement different kinds of age models, namely frequency analyses of arboreal pollen % and orbital tuning (Fq-9C) and tuning to reflectance curve of another marine core (Cariaco, which itself has been tuned to Hulu Cave in China). Long records, such as also from lake Titicaca (LT01-2B and LT01-3A; Hanselman et al., 2005, 2011; Fritz et al., 2007; Gosling et al., 2008, 2009), rely on advanced methods of orbital tuning for the older sections and are therefore not considered in this study for the recalibrated age model or star classification.

Through the years the radiocarbon community has presented a series of papers indicating the proper way of reporting 14C data (Stuiver and Polach, 1977; Mook and Van der Plicht, 1999; Reimer et al., 2004a). In the early days, the world's laboratories reported all of their produced radiocarbon dates in the journal Radiocarbon, a journal then dedicated to compiling these overviews. Probably the earliest radiocarbon dates from the Neotropics can be found in Vogel and Lerman (1969), describing in detail dates produced from Cuba, Jamaica, Colombia, Guyana, Surinam, Peru, and Argentina. However, this system could not keep up with the increasing number of both laboratories and studies reporting radiocarbon dates. Since then the correct reporting of 14C dates has relied completely on the experience and willingness of the researchers.

Measured radiocarbon concentrations require an additional correction due to mass fractionation of 14C atoms during natural bio-geochemical processes (e.g. photosynthesis; Drake, 2014), and sample preparation and measurement (Wigley and Muller, 1981). This is a δ13C-based correction which has a default value of -25 ‰ based on wood (Stuiver and Polach, 1977). In the Neotrop-ChronDB 1283 14C dates have reported fractionation corrections ranging from -42 to 30.2 ‰, but it is not always clear whether the authors implement any correction. This number represents a quarter of the total number of radiocarbon dates in the database, meaning that over 600 studies do not report this fractionation correction.

Studies specifying additional corrections such as the possible reservoir age are rare. Although organic material potentially presents this 14C offset, it is rarely identified in terrestrial pollen records in the area of interest. For the marine reservoir correction, the marine calibration curves incorporate a global ocean reservoir correction of c. 400 years. Nevertheless, regional differences in reservoir values should be applied according to the Marine Calibration data set (http://www.calib.qub.ac.uk/marine). Some marine studies in the region implemented a fixed reservoir effect of 400 years (according to Bard, 1988) for marine dates, while others only mentioned the used version of the CALIB program. A handful of marine cores in Chile (MD07-3104; MD07-3107; MD07-3088) estimate different local reservoir ages on calibrated ages from the IntCal calibration curve.

While Stuiver and Polach (1977) were the first to establish the conventions for reporting radiocarbon data, Reimer et al. (2004b) dealt with the growing use of postbomb 14C and a corresponding new symbol in 14C reporting. Correct postbomb 14C reporting is problematic in the Neotropics. Negative 14C ages are treated highly variably, from being totally discharged, titled “modern” or “too young” without specified 14C value, or considered valid as the subtracted age from 1950 AD (resulting in any age estimate between 2014 and 1950). Also postbomb dates as percentage modern carbon values (% pMC, normalized to 100 %) or “fraction of modern” (F14C, normalized to 1) sometimes mislead uninformed authors to be acceptable 14C ages. At this moment, only one pollen record is known to report the F14C value with the corresponding postbomb curve as proposed by Reimer et al. (2004b), namely Quistococha in Peru (Roucoux et al., 2013). Laboratory sample or identification numbers (ID), which are given to the samples by the radiocarbon dating laboratory, enable the laboratory to be identified and should always be published alongside the 14C measurements (Grimm et al., 2014; See the long version of the workshop report published at http://www.pages-igbp.org/calendar/127-pages/826-age-models-chronologies-and-databases).

The relatively recent development of freely available computing packages has as a consequence that there is a large bulk in the Neotrop-ChronDB without any age model (n= 457), where most radiocarbon dates are simply plotted along the pollen record without an explicit age-model. The most common age model (n= 298) is based on the simplest design, namely the linear interpolation between the dated levels, even though this is hardly a realistic reflection of the occurred sedimentation history (Bennett, 1994; Blaauw and Heegaard, 2012). Polynomial regression methods (n= 31) and the smooth spline (n= 12) are becoming increasingly popular but mostly in international peer-reviewed journals compared to national publications. In the latter linear interpolation is more persistent. In six cases, age models and calibrated ages were created by the authors without further explanation. In a significant number of cases, age–depth modelling was performed with uncalibrated 14C ages, which does not produce valid results due to the non-linear relationship between radiocarbon years and calendar years.

The unclear geographical boundary between the NH and SH calibration curve has led to finding pollen records from the same region using curves from either side of the hemisphere. This is seen in the highland of Peru and Bolivia where the boundary between the IntCal13 (NH-curve) and SHCal13 (SH-curve) realms is still unclear and even causing the use of different calibration curves for the same lake. Several Bolivian lowland studies explain the influence of the southern range of the ITZC migration and therefore justify the use of the northern calibration curve (Mayle et al., 2000; Maezumi et al., 2015). The existence of a 14C age difference of up to a few decades between the NH and SH has been discussed in the literature, e.g. McCormac et al. (1998), Turney and Palmer (2007), and Hogg et al. (2013). This temporal uncertainty should be taken into account and it would be useful if authors address the choice of calibration curve in the publications.

Statistical approaches to chronological modelling have expanded dramatically over the last two decades. Advances in computer processing power and methodology have now enabled Bayesian age models which require millions of data calculations – a method which would not have been possible before. The development of such freely available Bayesian age-modelling packages as “OxCal” (Bronk Ramsey, 1995), “BCal” (Buck et al., 1999), “Bchron” (Parnell et al., 2008), “BPeat” (Blaauw and Christen, 2005) and “Bacon” (Blaauw and Christen, 2011), has greatly advanced the science. To our knowledge, however, so far there has been only a single application of Bayesian methods for age modelling in South America, namely at Papallacta 1-08 (Ledru et al., 2013b). The authors included a priori information on sedimentation rates and tephra layers to construct the age model and consequently derive the best age for an uncertain tephra deposition. The use of the sedimentation conditions is a highly relevant component for age model development but rarely seen to be taken into account. Plotting the sediment record next to the age model would complement greatly the interpretation of the chronology (as shown as an example in Fig. 2).

Combining prior information from the sequences with the geochronological data is the basis of a Bayesian approach to construct an age–depth model (Blaauw and Heegaard, 2012). The current lack of Bayesian-based age models in the Neotropics could be due to classic age–depth models (based on linear interpolation, smooth splines or polynomial regressions) being regarded as the most realistic models, or to the usefulness of Bayesian methods not yet having been explored. Each model comes inherent with errors and uncertainties (Telford et al., 2004), and each method consists of different approaches to address them. Linear interpolation for example provides reasonable estimates for ages and the gradients between adjacent pairs of points, but only includes the errors at the individual age-determinations and does not consider uncertainties and additional measurements (Blaauw and Heegaard, 2012). A wider range of possible errors can be included in “mixed-effect models”, while Bayesian age–depth modelling produces more realistic estimates of ages and uncertainties. Although we did not engage in Bayesian modelling in this study, even if researchers find themselves without much prior knowledge of regional accumulation rates, Bayesian methods could well provide more realistic estimates of chronological uncertainties than classical methods (Blaauw et al., 2016). Researchers are encouraged to make use of the freely available character of the Bayesian software packages to test multiple age–depth models, compare models that best approximate their knowledge of the sediment conditions, and address these comparisons in their studies.

From a total of 292 pollen records revised, 242 preliminary age models were regenerated based on the provided dates. The other 50 pollen records either presented a lack of multiple geochronological dates or had too many chronological problems. During the process of adjustments of the age models for hiatus, outliers, and slumps, another nine pollen records were rejected as no reliable models could be produced. In 125 cases both linear interpolation and spline could be implemented, requiring at least four valid geochronological dates for the latter. The median number of stars for recalibrated chronologies of NW-SA is 3, which we consider surprisingly high.

Based on the 233 checked and recalibrated age models from NW-SA (Table 1), the sample resolution (maximum, minimum, median, and mean value) was estimated per pollen site and for the entire NW-SA. The resolution was calculated as the time between two consecutive depths with proxy information (sample depths). Minimum resolutions range from 10 years to 1 kyr, compared to the maximum value between 5 and 36 kyr (mostly due to extrapolations). The overall sample resolution estimates indicate that the average temporal resolution of this multi-site synthesis is c. 240 years, a resolution that allows analyses of ecological responses to sub-millennial-scale climate change. From a synoptic perspective, the NW-SA pollen records do not show spatial clustering based on the assigned stars (Fig. 3a). In other words, chronologies with good and poor control point density (number of control points per unit time) can be found along all the different elevational and latitudinal ranges. The best context to the star classification system can be given in conjunction with the sample resolution estimates as chronologies might present high sample resolution but poor chronological backup, and vice versa. What is evident as a result of the recalibrated age models is the high number of pollen records within the 0–500 years resolution with relatively high temporal quality (Fig. 3b).

Within this study, this time window is represented by only 4 pollen records from two lakes, namely from Lake Titicaca LT01-2B and LT01-3A (Hanselman et al., 2005, 2011; Fritz et al., 2007; Gosling et al., 2008, 2009), and Fúquene 3 and 7 (Mommersteeg, 1998; Van der Hammen and Hooghiemstra, 2003; Vélez et al., 2003; Bogotá-Angel et al., 2011). Research into millennial-scale climate variability is difficult during this time window, as sample resolution varies greatly from a few centuries to several millennia. For periods older than 65 kcal BP, mean resolution shifts around 2000 years per sample with a star classification of mostly 0–1. Temporal uncertainty is high due to extrapolation of age models through a limited number of control points and additional hiatus difficulties.

MIS 3 is better represented in samples (Fig. 4a) and sites (Fig. 4b), and shows a wider variation in the star classification. The median number of 1 star still indicates a relatively poor control point density in the chronologies and therefore high temporal uncertainty. This time window is characterized by relatively older sites with reduced chronological quality even though overall resolution is at centennial timescale (430 years).

The vast majority of chronologies cover the Holocene and Lateglacial time intervals because they have been established from lakes formed after the last glaciation. Consistent with the large number of pollen records that reflect the Holocene (Flantua et al., 2015), the highest density of palynological sampling covers the last 10 kcal (Fig. 4c). Most samples fall within the category of presenting “good” control point density, namely either 3 or 4, just as the individual sites evaluated (Fig. 4d). There is an overall good point density in the NW-SA sites that cover the YD/Holocene transition but the Last Glacial Maximum (LGM) and H1 are represented by far fewer records with varying temporal quality.

The integration of the recalibrated chronologies and the estimated sample resolutions indicate the essential value of the existing radiocarbon calibration curves: there is a clear threshold at c. 55 kcal BP (beyond the extent of the current 14C calibration curves) from where the control point density and resolution currently do not support research on millennial timescales, as sample resolutions are on average 1300 years and temporal uncertainty high (Fig. 5).

The relevance of publishing details on the sample, laboratory and reference numbers, provenance and reservoir correction details seems underestimated by authors in many cases. Studies with insufficient chronology reporting undermine the consistency and credibility of the results presented, and weaken the value of the radiocarbon dates. Furthermore, considering the expanding palynological research (Flantua et al., 2015), papers with deviations in chronology reporting will most likely not be used within the context of multi-proxy comparisons or more expanded regional synthesis efforts. Additionally, paleo-vegetation records with proper chronology details are frequently scanned by the archaeological community to correlate human and environmental dynamics (Aceituno et al., 2013; Delgado et al., 2015). Equally relevant are paleoecological records with solid chronologies for late Pleistocene understanding of megafaunal extinctions (Barnosky et al., 2004). Missing out on the chronology description is without doubt an unnecessary way to affect the credibility and citation rate of any study. A top-down approach to improve radiocarbon reporting initiates at the journals demanding complete and correct chronology information. Not less important are the reviewers in critically evaluating the presented age models. Sources to remain updated on the requirements of dating reporting are numerous (e.g. see Millard, 2014), but specific details can be online accessed through http://www.c14dating.com/publication.html. Additional recommendations can be found in Blaauw and Heegaard (2012) and from the “Neotoma Age models, chronologies, and databases workshop” in Grimm et al. (2014).

The importance of high-resolution records but especially temporal quality has been illustrated through the development of updated age models and control point density assessments. Compared to the implementation of the method in the EPD (Giesecke et al., 2014), there is a higher proportion of samples and sites in the last 5 kcal BP in NW-SA. The most common sample resolution in the EPD is between 50 and 250 years, while the NW-SA has a mean resolution of 235 years. This resolution is actually higher than we expected and this could be due to several reasons. First of all, during the age modelling procedure, chronologies with too many disturbing features were not used, implementing a first selection towards the best possible age models. Secondly, to assign 10 cm sample intervals for older pollen records to unknown sample depths could be an overestimation for sample resolution (many older records were sampled at > 20 cm). Thirdly, there are several very high-resolution sites that cover significant time periods overpassing greatly in sample numbers the sites with relatively low temporal resolution. Any calculation based on multi-site information should use a median value instead of the mean value (Fig. 3), which is less sensitive to extremes. Nonetheless, the general tendency is that pollen records in NW-SA are improving chronological settings with high sample resolution on centennial timescales.

Until now, differences in resolution and chronological quality between older and newer sites have hampered the ongoing discussion on the rapid climatic shifts such as the YD. A synchronous similar climate reversal at the YD is not evident throughout South America. Differences in magnitude have been observed between Venezuela and Colombia (Rull et al., 2010), while pollen records at relatively close distances in Peru/Bolivia are considered both different in timing and expression (Hansen, 1995; Paduano et al., 2003; Bush et al., 2005). This points again to the danger of using assumed synchronous events to align archives across a region – e.g. Israde-Alcántara et al. (2012a), who align several poorly dated sites in Latin America to circularly argue for a YD comet impact (Blaauw et al., 2012; Israde-Alcántara et al., 2012b). New studies on correlating biostratigraphic patterns with improved chronology are important as they can identify possible long-distance synchronicity of climate signals, but at the same time display their own local signature when supported by high-resolution data. Therefore, additional well-dated records have a high potential of contributing to this current discussion (e.g. Rull et al., 2010; Montoya et al., 2011a). However, advanced tools to assess leads, lags, and synchronicity in paleorecords are still urgently needed (Blockley et al., 2012; Seddon et al., 2014) while only few case studies have yet explored the available tools (Blaauw et al., 2007, 2010; Parnell et al., 2008). As long as the discussion consists of correlating poorly dated events, new hypotheses based on assumed synchronous events fail to provide additional insights to current questions.