Long-term ecological legacies in western Amazonia

1. Modifications of Amazonian forests by pre-Columbian peoples are thought to have left ecological legacies that have persisted to the modern day. Most Amazonian palaeoecological records do not, however, provide the required temporal resolu tion to document the nuanced changes of pre-Columbian disturbance or post-disturbance succession and recovery, making it difficult to detect any direct, or indirect, ecological legacies on tree species. 2. Here, we investigate the fossil pollen, phytolith and charcoal history of Lake Kumpak a , Ecuador,

The ecological legacies left by pre-modern human activities can either be direct, that is, the intended product of manipulation of the environment, or indirect, that is, the unintended product of manipulation ( Figure 1). Direct legacies would include the intentional enrichment of preferred species, or the intentional removal of non-preferred species (Clement et al., 2015). The direct enrichments or depletions caused by past human activities can affect the successional trajectory, or ecological legacy, in two ways. One possibility is that the populations that were enriched or depleted quickly rebound to pre-disturbance abundances (non-persistent enrichment/depletion, Figure 1). The other possibility is that enriched or depleted abundances of species persist through time, where enriched species establish reproductive populations that continue to the present day and depleted species do not recover to pre-disturbance abundances ( Figure 1a). For example, palms with fruits eaten by people, such as peach palm Bactris gasipaes (Clement, 1988;Clement, Rival, & Cole, 2009;Hernández-Ugalde, Mora-Urpí, & Rocha, 2011), and Moriche Mauritia flexuosa (Rull & Montoya, 2014), along with Brazil nut Bertholettia excelsa (Levis et al., 2012;Scoles & Gribel, 2011; F I G U R E 1 Schematic diagram showing the rise and fall of species through time in response to an episode of sustained disturbance (shaded grey). Panel (a) direct enrichment of preferred species (dark green line) which either maintain/increase population levels, or, quickly rebound to pre-disturbance abundances after the disturbance ends. Indirect enrichment of species (light green line) that increase during or after disturbance, but which are not intentionally enriched or altered. Species that show no population change during disturbance are no effect species (purple line). Panel (b) direct depletion of non-preferred species (dark blue line) which either maintain/decrease population levels, or, quickly rebound to pre-disturbance abundances after the disturbance ends. Indirect depletion of species (light blue line) that decrease during or after disturbance, but which are not intentionally depleted or altered. Species that show no population change during disturbance are no effect species (purple line)  Shepard & Ramirez, 2011), are cited as evidence of persistent forest enrichment legacies.
Indirect legacies include the responses of weedy or successional species that were not intentionally altered (Figure 1). Some populations can exhibit a short-term increase or decrease associated with disturbance that we term non-persistent, for example, Ambrosia or Cecropia would often be examples of non-persistent enrichments during disturbances (Figure 1).
Pre-Columbian people in Amazonia may have also depleted the relative abundances of some species (Clark, Clark, Sandoval, & Castro, 1995;Ferreira et al., 2019;Figure 1b). The depletions are likely to be among species that were cut to provide timber for construction or where consumption was a one-time disturbance, for example, harvesting for heart of palm (Haynes & McLaughlin, 2000). In Costa Rica, the absence of Iriartea deltoidea from forests near rivers in La Selva Biological Reserve was identified as a legacy effect of colonial era logging (Clark et al., 1995). In Amazonia, pollen percentages of I. deltoidea and phytolith percentages of taxa including Iriartea were lower when direct indicators of human activity (i.e. charcoal or crop pollen) were at their highest Heijink et al., 2020). Depletions, like enrichments, can be persistent, non-persistent or indirect ( Figure 1b).
The intensity, manner, scale and duration of past disturbance all influence post-abandonment successional trajectories (Bodin, Molino, Odonne, & Bremond, 2020). In some areas, the land use was a polyculture featuring mixed crop field systems and orchards (Denevan, 2001). If abandoned, succession would have quickly displaced crop plants and disturbed-ground weeds with fast growing, light-loving trees, as a stage before a dense forest established (Bush & Colinvaux, 1988, 1994Uhl, Clark, Clark, & Maquirino, 1982).
Dwelling sites would have been surrounded by useful species (sensu Clement et al., 2015) ranging from dye plants to food plants. As many of these useful species would have been trees of mature forest, the successional trajectory would have followed a different path compared with the open ground. Consequently, the modern forest, perhaps no more than 400 years old, could still bear different ecological legacies, especially in non-natural enrichments of large longlived trees (Bodin et al., 2020;Figure 1a).
Past human occupation of Amazonian forest systems is readily identified in palaeoecological records by an increase in the pollen and phytolith abundances of grasses, crops and taxa indicative of forest openings (gaps), or the presence of charcoal (Bush, 2002;McMichael Correa-Metrio & Bush, 2012;Whitney et al., 2019). Addressing questions of past forest disturbance and recovery is best accomplished through a multiproxy approach that capitalises on the different strengths of the evidentiary lines. For example, though phytoliths cannot distinguish many of the tree taxa that can be identified with pollen, pairing the two proxies: (a) increases certainty in identifying cultivation and forest opening and (b) provides a more comprehensive view of the change in taxa associated with disturbance and recovery processes. Unlike pollen, phytoliths can also distinguish grasses to sub-family or sometimes even genera, making them particularly useful for disentangling grasses typical of disturbed or open settings, grasses that grow under the forest canopy, and aquatic grasses (Morcote-Ríos, Bernal, & Raz, 2016;Piperno, 2006b). Palms are also abundant producers of phytoliths and many phytolith morphotypes can be identified to genus (Piperno, 2006b).
The study of past disturbance in lowland rainforests also has to be conducted at an appropriate temporal scale. Most palaeoecological records are temporally coarse-grained, generally offering centennial resolution, and cannot assess the nuanced changes of pre-Columbian disturbance, post-disturbance recovery and various types of ecological legacies of tree species. Here, we provide an almost unprecedented temporal resolution for a multiproxy Amazonian palaeoecological record. Using fossil pollen, phytolith, and charcoal from the 2,415-year Lake Kumpak a record, Ecuador, at 3-50 year time intervals, we assess the types, duration and variation of ecological legacies resulting from pre-Columbian fire, agriculture and forest disturbance.

| S TUDY ARE A
Lake Kumpak a (2°50ʹ11″S, 77°57ʹ41″W; 333 m elevation) is located in lowland Amazonia, Ecuador ( Figure 2a). The lake has a diameter of 420 m, a maximum water depth of 19.5 m and is thought to be volcanic in origin (Colinvaux, Millerf, & Liu, 1985; Figure 2b). Lake Kumpak a is permanent and has no riverine influence, making it a rarity in Amazonian systems. The region receives approximately 2,000-3,000 mm of precipitation per year and lacks a defined dry season (https://www.clima te-data.org; Liu & Colinvaux, 1988). Mean annual temperature is 24°C. Small streams deliver sediment-rich water to all sides of the mesotrophic lake ( Figure 2b). A lack of outlets suggests that water level changes in response to precipitation. We found the lake to be anoxic below 2 m water depth (measured in 2014).
The lake was first cored in 1983 by Paul Colinvaux's team and a low-resolution 5,860-year history of its pollen and sediment stratigraphy was published (Liu & Colinvaux, 1988;Figure 2c,d). That study revealed a largely stable rainforest system, with more apparent ecological changes in the last 1,500 years than in the earlier portion of the record. The possibility that there was a human history was largely ignored in this initial study.
Following establishment of a Salesian mission station in the 1960s, most of the modern landscape surrounding Lake Kumpak a has been modified for small-scale agriculture and grazing. Today, a small community of the Shuar nation inhabits the areas around the lake.

| MATERIAL S AND ME THODS
In July 2014, an 18.0-m sediment core was retrieved from 19.5 m water depth. The core was wrapped in the field in sections of 1 m, transported to Florida Institute of Technology and stored at 4°C. The sediment core represents the limit of coring ability as bedrock was not reached. Terrestrial macrofossil remains (n = 15) were isolated and radiocarbon dated to establish the Kumpak a chronology Subsamples were taken for pollen and phytoliths at every 5 cm from 0 to 530 cm core depth and at every 10-20 cm for the intervals from 530 to 760 cm. Charcoal was subsampled at 1-5 cm resolution throughout the 760 cm section. Pollen samples (n = 126) were treated according to standard methodology (Supporting Information: Methods; Faegri, Kaland, & Krzywinski, 1989;Stockmarr, 1971) and a minimum of 300 pollen grains was counted in each sample.
Identification of pollen grains was made using appropriate keys (Colinvaux, De Oliveira, & Moreno, 1999;Roubik & Moreno, 1991), the Neotropical Pollen Database (Bush & Weng, 2007)  were prepared according to standard methodology (Piperno, 2006b) and a minimum of 200 phytoliths were counted in each sample (Piperno, 2006b;Supporting Information: Methods). Phytolith keys (Morcote-Ríos et al., 2016;Pearsall, 1978;Piperno, 2006b) and the reference collection at University of Amsterdam were used for identification. Pollen and phytolith samples were analysed using Zeiss F I G U R E 2 Map of the study area of Lake Kumpak a , Ecuador. (a) Map of South America with a focus on Ecuador and Peru showing study site Lake Kumpak a and sites Lake Ayauch i (Bush & Colinvaux, 1988), Lake Sauce  and Lake Gentry . (b) Bathymetry map of Lake Kumpak a showing coring location (red circle) and inlet streams (blue lines; modified from Liu & Colinvaux, 1988).
Detrended correspondence analysis (DCA) was performed using R (R Development Core Team, 2019) and the vegan package (Oksanen et al., 2013) on the pollen and phytolith data, separately, to quantify trends in (dis)similarity between samples (slices of time within the stratigraphic sequence), and to determine which taxa within pollen or phytolith assemblages were most responsible for driving those differences (also see Supporting Information: Methods). We also used the DCA results to determine whether successional trajectories or ecological legacies could be seen in the data.

| RE SULTS
The 18.0-m core spans the last 5,400 years, and all 15 radiocarbon ages were accepted to provide the chronology (Table S3; Figure S1).
This paper focused on the last 2,415 years (760 cm), where sediment accumulation rate averaged 3.24 years per centimetre and remained consistent almost throughout the entire core ( Figure S1).
This 760 cm of sediment was dominated by alternating pale and dark organic clay laminations of very thin thickness (<1 mm) interspersed with medium (2-5 mm) to very thick (>10 mm) laminations (Supporting Information: Results; Figure S2).
The temporal resolution of sampling provides a c. 3-to 15-year resolution for charcoal occurrence and a c. 15-to 50-year periodicity for pollen and phytolith data. Old-growth arboreal pollen and phytoliths were prominent features throughout the Kumpak a record, Ecuador, with values fluctuating between 30% and 90% for pollen and 35%-95% for phytoliths ( Figure 3). In the pollen record, decreases in old-growth arboreal forest and palm elements (e.g.

| D ISCUSS I ON
The Lake Kumpak a record is a rarity in Amazonian systems as it comes from a permanent lake without riverine influence, offering the potential for an unusually detailed sedimentology. The high deposition rate allowed us to reconstruct charcoal at an almost F I G U R E 4 Detrended Correspondence Analyses biplots of fossil pollen data (a and b), eigenvalues 0.2008 and 0.07919 on axes 1 and 2, and phytolith data (c and d), eigenvalues 0.1422 and 0.02908 on axes 1 and 2, from Lake Kumpak a , Ecuador, showing sample and species scores separated in six categories: pre-disturbance (2415-   continuous resolution of 3.24 years, confirming the true rarity of fire in this system, and pollen and phytoliths at a c. 15-to 50-year resolution. The laminated sediments and 14 C ages provided reassurance that there has been little post-depositional disturbance, allowing an almost linear age model to be constructed ( Figure S1). Our study broadly supported the original pollen reconstruction from Lake Kumpak a , which had samples every 100-300 years, but had no corresponding charcoal or phytolith analysis (Liu & Colinvaux, 1988).
Our study, however, documented the ability of multi-proxy palaeoecological datasets analysed on decadal time scales to address questions regarding the long-term disturbance and recovery patterns of Amazonian forests.

| The intensity of past disturbances at Lake Kumpak a
Based on our aerial imagery, the modern landscape around Lake forest cover contain about 93% forest pollen, of which <5% were from palms (Figures 3 and 5). The arboreal elements found in the pollen assemblages included: Moraceae, Urticaceae, Combretaceae, Melastomataceae, Alchornea, Didymopanax, Celtis and Cecropia ( Figure 3).

In the modern botanical inventory surveys, species of Didymopanax,
Moraceae, Combretaceae and Melastomataceae were observed in remnant forest around the lake, providing a confirmation between the modern vegetation and corresponding pollen assemblages. Phytolith assemblages during the last 30 years included c. 60% forest taxa, 35% palm taxa and 5% grass taxa (Figures 3 and 5). Clearly, the common palms around the lake today, primarily Wettinia, Mauritia and Bactris, were more strongly represented in phytolith than pollen sequences.
Maize was also being grown around the lake when it was cored in 2014, which was detected in the phytolith samples (Figure 3). The period from c. 1070 to 680 cal. year BP was probably the most intensely disturbed of the entire record (Figures 3-5), based on the amounts of Cecropia, Poaceae, Z. mays, and herb taxa in the pollen and phytolith data. In disturbed Amazonian forests, as much as 50% of the pollen signal comes from Cecropia, a pioneer secondary forest species that produces enormous quantities of pollen (Bush & Rivera, 1998).
Percentages of Cecropia are similar between the modern system and the period from c. 1070 to 680 cal. year BP, though the amount of landscape opening may have varied. Palm abundances in both the pollen and phytolith record are low during this period compared with recovery periods and the modern period (Figures 3-5). The decline in arboreal elements was also significantly larger from 1070 to 680 cal.
year BP than during the modern disturbance period (Figures 3-5).
The reduction in forest pollen percentage is probably not linearly related to the area of forest clearing (Whitney et al., 2019), but it may provide a guide to estimating the extent of disturbance. Arboreal elements comprise 90% of the modern pollen assemblages of the last 30 years, where land clearance is known to affect 20% of the landscape (Figure 2c-f). When arboreal percentages decrease to 75% of the pollen assemblage, it suggests land clearance may have increased a similar amount, or up to 35% of the area around the lake. While this is a very speculative number, it is safe to say that the combined pollen and phytolith data suggest that the peak of clearance at c. 950 cal.
year BP was substantially larger than that of today, and palms were likely disproportionately removed (Figures 3 and 5).
Herbaceous and grasses in the pollen and phytolith record suggest the landscape around Lake Kumpak a in the earlier period of distur-  (Figures 3 and 5). Maize cultivation was found in both the pollen and phytolith records, indicating local cultivation (Jarosz et al., 2003;Piperno, 2006a;Raynor, Ogden, & Hayes, 1972), and palms were significantly less abundant than in today's landscape, suggesting the indigenous inhabitants exploited them.
At Kumpak a , despite the evidence of multiple centuries of maize cultivation, remarkably little charcoal was recovered from the sediment (Figures 3 and 5). Between 1070 and 680 cal. year BP charcoal fragments were only found in one sample, occurring at the very end of the disturbance period. Fire was also rare in the earlier period of disturbance, from 2150 to 1430 cal. year BP, with similar amounts of forest taxa as the modern forests ( Figure 5). Charcoal abundances at Lake Kumpak a were orders of magnitude less than in other Amazonian lakes containing similar evidence of maize cultivation, for example, Lakes Ayauch i , Sauce and Gentry (Supporting Information: Discussion; Bush, Silman, & Listopad, 2007; McMichael, McMichael, Correa-Metrio, et al., 2012). Lake Ayauch i , just 25 km away from Kumpak a , was occupied for over 6,000 years with maize cultivation and long-term fire use, suggesting that slash-and-burn agriculture was used to manage the landscape (Bush & Colinvaux, 1988;McMichael, Correa-Metrio, et al., 2012;Piperno, 1990). Natural fires are almost non-existent in western Amazonia, as the wet conditions of the forests limit the spread of fire, and findings of charcoal fragments in palaeoecological records have repeatedly been linked with human land use (Aragao et al., 2008;Cochrane, 2009;Cochrane & Schulze, 1999;Kelly et al., 2018;McMichael, Correa-Metrio, et al., 2012;Nepstad et al., 2004;Urrego et al., 2013).
Fire may not have been an important management tool at Lake Kumpak a . Charcoal was almost uniquely associated with maize ( Figures 3 and 5), but the paucity of samples containing charcoal, and the low amounts of charcoal found in those samples, suggest that the cultivation around Lake Kumpak a was not reliant on fire.
One possibility was that rather than using slash-and-burn cultivation, the inhabitants of Kumpak a may have been using slash-andmulch (Joslin et al., 2011). This technique helps to retain nutrients volatilised during burning, and limits soil compaction, which reduces erosion from a denuded forest surface (Hölscher, Möller, Denich, & Fölster, 1996;Jordan, 1985). Mulching reduces weed growth while providing a slower, more sustained nutrient release than burning, and runs less risk of carbon ignition and soil degradation than slashand-burn (Arroyo-Kalin, 2012;Joslin et al., 2011;van Vliet, Adams, Vieira, & Mertz, 2013). The fertility of the soils around the lake is not known, but if the origin of the lake is a volcanic maar, the soils may be more fertile than typical of Amazonia (Rostain, 2012). The Lake Kumpak a data appear to be the first empirical evidence for the pre-Hispanic use of slash and mulch in western Amazonia.

| Disturbance and recovery
The composition of pollen from the modern disturbance period falls within the variability of past disturbances, and shares characteristics with both the disturbance ending at 1430 cal. year BP and 680 cal.
year BP ( Figure 5). Our pollen and phytolith data, however, suggest that modern human disturbance between 10 and −10 cal. year BP (AD 1940(AD -1960, and the subsequent modern occupation, created ecologically different conditions from those that occurred in the past (Figures 4 and 5). As with past cultivation periods, Iriartea and most other arboreal forest elements decline abruptly while disturbance taxa such as Cecropia, Trema, Solanaceae and Poaceae increase in abundance. During the modern disturbance, however, none of the herb taxa (e.g. Asteraceae, Piperaceae and Acalypha) increased in percentages as they did during and after past disturbances. The differences between the modern disturbance period and past disturbances are probably caused by the increased reliance on fire in modern times compared with the past (Figures 3 and 4), and are consistent with expectations of slash-and-burn agriculture (Arroyo-Kalin, 2012;van Vliet et al., 2013).

| Ecological legacies
The predominant idea among archaeologists regarding ecological legacies is that useful species were deliberately enriched or depleted in the pre-Columbian era, and that those altered abundances have persisted until the modern era (Ferreira et al., 2019;Levis et al., 2012Levis et al., , 2017; Figure 1). Under this scenario, useful or domesticated species in the Lake Kumpak a sediment record should have flourished during the peak of maize cultivation and show persistently elevated abundances even after the cessation of cultivation (direct and persistent enrichment; Figure 1a). Our record does not support this contention as no useful taxa (i.e. Lecythidaceae, Theobroma, Mauritia, etc.) showed such a permanent pattern of enrichment. We also saw no evidence of persistent depletion, as the taxa with depleted populations, that is, Iriartea, recovered ( Figure 6). Instead, our data show that while pre-Columbian activities can alter recovery trajectories for hundreds of years following the cessation of a disturbance, the trajectories do not include evidence of persistent enrichment or depletion ( Figure 6).  Figure 6 for representative responses of Acalypha and Cecropia). Among these taxa Malvaceae, Tontelea and Zea (Clement et al., 2015) could be useful species and therefore the product of direct enrichment, whereas the others are most likely to have been indirect enrichments. Some of these are known to be early successional taxa (e.g. Cecropia, Trema, Acalypha, Asteraceae, Piperaceae) that are non-persistent and decrease shortly after cultivation ends (Figures 3 and 6). Ficus and some mid-successional F I G U R E 6 Detailed views of the last 100 years of disturbance and postdisturbance successional trajectories (recovery periods) of representative taxa at Lake Kumpak a , Ecuador ( Figure  S4). Panel ( taxa showed post-occupation increases, followed by declines about 100 years into the succession (Figures 3 and 6; Figure S4). Here, we classify these as indirect, non-persistent ecological legacies (Figures 1a and 6; Figure S4).
Iriartea deltoidea is one of the commonest trees of western Amazonia (ter Steege et al., 2013, and has consistently shown a negative response to human activity in palaeoecological records (Figure 4; Heijink et al., 2020).
At Lake Kumpak a , the periods of maize cultivation coincided with the loss of Iriartea, which is to be expected, as it would have been harvested for timber (Supporting Information: Discussion;Bernal et al., 2011;Clark et al., 1995;Zambrana et al., 2007). A century or more after the depletions of Iriartea populations ended, they recovered at Lake Kumpak a . The populations also recovered to levels that exceeded those in the pre-cultivation period (Figures 3   and 6) and its abundance prior of the last 5,200 years (Liu & Colinvaux, 1988). We classify this population expansion as an indirect persistent enrichment that terminates with another direct depletion in near-modern times.
Our observations fit with the known autecology of Iriartea in that it is known to be a mid-successional species that can grow at high stem densities in floodplain forests subject to long-term disturbance by rivers (Balslev, Luteyn, Ollgaard, & Holm-Nielsen, 1987;Pitman et al., 2001). It seems unlikely that the increases seen in the Lake Kumpak a record resulted from Iriartea being cultivated for remote usage, as transporting trunks the 3 km to the nearest river would have been problematic. Far more likely is that Iriartea regrew naturally as an early-mid successional species, and that its surge in abundance c. 200 years after cultivation ended was the synergistic effect of regional increased precipitation levels (van Breukelen, Vonhof, Hellstrom, Wester, & Kroon, 2008) and the release from human exploitation. These surges in abundance may have also been facilitated by changes in the soil, drainage or soil erosion caused by occasional fire and cultivation. Iriartea maintained high populations for >500 years after abandonment, that is, several tree generations (Pinard, 1993 The successional dynamics and lack of persistent ecological legacies at Lake Kumpak a show that modern abundances of useful species should not be used to infer past human activity. High abundances of useful species are found in forests that contain no evidence of past human impact (Bodin et al., 2020). Natural floodplain and gap succession are all that are required to account for the great majority of Iriartea occurrences (Pitman et al., 2001). The ability of seeds to disperse also plays a large role in the relative abundances of many tropical forest plants. For example, local palm abundances may decrease if seed dispersers are lost, but increase if, despite poor seed dispersal, seed predators are also lost (Wright et al., 2000). Hunting and deforestation that occurred from the time of the early Jesuit missions in the 1600s, through the Amazonian rubber boom of AD 1850-1920, to the modern era, are all likely to have shaped the relative abundances of many Amazonian plants. Indeed, these more recent activities are more likely to have legacy effects on the forest than those occurring before European arrival.

| CON CLUS IONS
We were able to document successional dynamics and potential for ecological legacies for hundreds of years following the cessation of a disturbance from Lake Kumpak a in lowland Amazonian Ecuador, at time intervals of 5-50 years. Our data contained three different disturbance periods over the last 2,415 years, including the period of modern disturbance (Figure 1c-f). The disturbances all varied in intensities and durations. The modern disturbance period, which contained the highest fire frequency, was compositionally different from the two past disturbance periods, which contained little to no fire. The Lake Kumpak a record suggests that the intensity of disturbance, the duration of the disturbance and the presence or absence of fire in the disturbance regime affect the subsequent successional trajectories and potential ecological legacies.
We found evidence of non-persistent enrichment and depletions of taxa, but those legacies were not present in the modern-day forest. It takes hundreds of years for a complex tropical forest to regain its pre-disturbance composition, in which time early and mid-successional species become rarer. Many taxa considered to be useful species are best suited to early successional settings and would be outcompeted during succession. We found no evidence of persistent enrichment or depletion of taxa (i.e. direct legacy effects) in a setting that contained pre-Columbian cultivation until c. 680 cal.
BP. These findings underscore the inaccuracy and uncertainty of using modern plant population densities to infer past human management and ecological legacies without corresponding direct empirical data. Our data also reinforce the potential power of drawing upon highly resolved multiproxy data palaeoecological datasets to address important questions in neoecology.