Climatic control of the biomass-burning decline in the Americas after ad 1500

The significance and cause of the decline in biomass burning across the Americas after ad 1500 is a topic of considerable debate. We synthesized charcoal records (a proxy for biomass burning) from the Americas and from the remainder of the globe over the past 2000 years, and compared these with paleoclimatic records and population reconstructions. A distinct post-ad 1500 decrease in biomass burning is evident, not only in the Americas, but also globally, and both are similar in duration and timing to ‘Little Ice Age’ climate change. There is temporal and spatial variability in the expression of the biomass-burning decline across the Americas but, at a regional–continental scale, ‘Little Ice Age’ climate change was likely more important than indigenous population collapse in driving this decline.


Introduction
Recent syntheses of sedimentary charcoal records suggest that biomass burning in the tropical Americas declined between ad 1500 and 1600 (Bush et al., 2008;Dull et al., 2010;Nevle and Bird, 2008). However, the geographic extent, magnitude, and cause of this decline are uncertain because of the limited geographic coverage and temporal resolution of available data, as well as incomplete evaluation of all potential controls in these syntheses. Proponents of the populationcollapse hypothesis (PCH) (Dull et al., 2010;Nevle and Bird, 2008) argue that this decline reflects a reduction in anthropogenic biomass burning due to decimation of indigenous populations by diseases, such as smallpox, following European contact. If the PCH is correct, this implies that the pre-Columbian (pre-ad 1492) fire regime of tropical ecosystems was largely anthropogenic, consistent with the 'cultural parkland' hypothesis (Erickson, 2000;Heckenberger et al., 2003), and that post-Contact (post-ad1492) indigenous population collapse resulted in 'reforestation' owing to a reduction in anthropogenic ignitions. This scenario has been used to explain declines in atmospheric δ 13 CH 4 (Ferretti et al., 2005;Mischler et al., 2009), CO (Wang et al., 2010) and CO 2 concentrations (Frank et al., 2010;Ruddiman et al., 2011) between ad 1600 and 1700 (Dull et al., 2010;Faust et al., 2006). An alternative, but not mutually exclusive, explanation (the climate change hypothesis, CCH) proposes that climatic variations were largely responsible for this decline in tropical biomass burning Pechony and Shindell, 2010).
Here, we examine the PCH and CCH for the post-ad 1500 decline in biomass burning in the Americas by synthesizing charcoal series spanning the last 2000 years (2 kyr) from an updated version of the Global Charcoal Database (GCD) (Power et al., 2008) (Figure 1), and comparing these with reconstructions of climate and population (Goldwijk et al., 2010;Mann et al., 2009). To examine the implicit prediction of the PCH that the post-ad 1500 decline in biomass burning in the Americas was globally unique, we compared the composite 2-kyr charcoal curve for the Americas with that of the rest of the world ('non Americas').

Charcoal data sources
We obtained 498 sedimentary charcoal records covering part or all of the last two millennia from the Global Charcoal Database (GCD version 1), co-author contributions, and from published syntheses Vannière et al., 2011). The GCD contains charcoal records from lacustrine, fen, bog, marine, archaeological, and other sites. We intentionally excluded nonterrestrial charcoal records, including marine sediments, and records from alluvial fans and soils (Carcaillet, 2001), because of the challenges of temporally constraining these types of deposits.
In an attempt to improve our geographic coverage of late-Holocene fire activity across the Americas, an additional 84 charcoal records (Table 1, Figure 1) were acquired from contributors to the GCD or by digitizing records from the published literature. Uncalibrated radiocarbon dates from new (additional) records were calibrated according to the protocol established for inputting data (Power et al., 2008) for consistent treatment to all radiocarbon-dated chronologies.

Statistical/numerical analyses of charcoal data
To make charcoal data comparable from a wide range of sampling, processing, and quantification techniques from hundreds of sites, we used a protocol that normalizes individual records in order to stabilize the variance and standardize the results (Power et al., 2008(Power et al., , 2010. Composite records were created by smoothing the standardized records using a two-stage smoothing process (Power et al., 2010). First, individual records were sampled and smoothed by selecting all data points within each 20-year interval in a record and calculating a locally weighted mean ('lowess') based on a tricube weight function, which weights the middle 10 years of the window most heavily (Cleveland and Devlin, 1988). If no data points existed within a given 20-year window, no interpolation was performed, which is important because it avoids creating data through interpolation rather than observation. In the second smoothing step, samples from all records with data for a given 100-year interval were combined and used to calculate the lowess regression. In this latter step, a 'robustness iteration' was performed that minimizes the influence of extreme charcoal values, providing an estimate of the sample mean that is robust to outliers. This methodology emphasizes long-term trends in the charcoal data that are characteristic of multiple records. Confidence intervals for each composite curve were generated by a bootstrap re-sampling (with replacement) of individual sites (not samples) over 1000 replications. Bootstrap confidence intervals for each target point were taken as the 2.5th and 97.5th percentiles of the 1000 fitted values for that target point. Trends in population estimates were constructed by summarizing the HYDE 3.1 data set (Goldwijk et al., 2010) by geographic region (as shown in Figure 2) Population time series were generated by areally averaging gridded HYDE data over the same region used for compositing and summarizing charcoal series. Regional reconstructions for temperature and precipitation data were obtained from a variety of published sources (see Figure 4). These climate time series were smoothed using the same approach taken with the charcoal data (i.e. lowess smoothing based on a 150-year window width).

Archaeological site locations
Locations of large (~ >1 km 2 ) archaeological sites were obtained from the United Nations Educational, Scientific and Cultural Organization (UNESCO) and from available literature, regional and local park services, and data available through the World Wide Web. UNESCO sites are considered 'of outstanding universal value' by the World Heritage Committee and are available through their web page (http://whc.unesco.org/en/criteria). All sites were separated into two temporal categories: (1) prehistoric sites (i.e. sites that were occupied between ad 1250 and 1500), and (2) sites occupied prior to ad 1250. Site names and locations  (Power et al., 2010) and new records (blue circles) incorporated into this study. Charcoal records from the Americas not included in GCD v1 were obtained: (1) as raw data from data contributors (included here as co-authors), (2) by digitizing previously published charcoal records, and (3) from data contributed to previously published charcoal syntheses (Table 1). shown in Figure 2 were not intended to be comprehensive, but depict the broad geographic distribution of important and well documented archaeological site locations.

Results and discussion
Linkages among biomass burning in the Americas, the rest of the world, and global biogeochemical cycling and climate change The composite curve for the Americas (Figure 3a) shows a decrease in charcoal influx (biomass burning) between AD 1500 and 1650, with a minimum between AD 1650 and 1700, the lowest level for the past 6 kyr (see Figure 5). There is also a prominent local minimum in the non-Americas composite curve ( Figure  3b) of similar duration to that in the Americas, but its onset and maximum expression occurs about 100 years earlier. Although the timing of this minimum in the Americas is consistent with the general chronology of population (Figure 3e,f) associated with the arrival and spread of European diseases across the hemisphere (Denevan, 1992), the slightly earlier parallel decline outside the Americas is impossible to explain through population collapse in the Americas. A climate-driven reduction in biomass burning can, however, explain both declines, as well as the difference in timing between them inasmuch as changes in atmospheric circulation associated with the 'Little Ice Age' (LIA), defined by maximum global cooling AD 1400-1700 (Mann et al., 2009), were not globally uniform or synchronous.
The low levels of biomass burning between ~ AD 1500 and 1750 have been used to explain some striking features in ice-core records of global biogeochemical cycles. In particular, atmospheric CO 2 concentrations (Figure 3i) decreased by around 5 ppm between AD 1500 and 1700 (Frank et al., 2010;Joos and Spahni, 2008), a pattern attributed by others (Dull et al., 2010;Faust et al., 2006;Ruddiman et al., 2011) to increased carbon sequestration in the Americas related to population collapse. Similarly, the record of δ 13 CH 4 (Figure 3g), which includes biomass burning as one of the source terms in its budget, also shows a steep decline between ad 1400 and 1750, and this too has been ascribed to a decrease in biomass burning in the Americas following population collapse (Ferretti et al., 2005;Mischler et al., 2009). The ice-core record of CO from combustion sources also shows a broad decrease from ad 1350 to 1650, followed by an increase (Wang et al., 2010), attributed to variations in Southern Hemisphere biomass burning. It is clear, however, that a global reduction in biomass burning (Figure 3a,b) prevailed between ad 1500 and 1750, and, crucially, the declines in CO 2 and δ 13 CH 4 were already underway by ad 1200 and ad 1400, respectively. The proposal that carbon sequestration in the Americas (and elsewhere) may have contributed to the sharpness of the decreases in CO 2 between ad 1500 and 1700 (e.g. from 280 to 275 ppm) is not supported by simulations with coupled climate-carbon cycle models (Pongratz et al., 2011), even if exaggerated estimates of the potential land-cover change are used. Similarly, land-cover changes, either 'best estimates' or larger, cannot explain the longer-term (Holocene) trends (Stocker et al., 2011), and similar arguments could be made for the trends in δ 13 CH 4 and CO. Moreover, the global expression of the biomass-burning reduction in the charcoal evidence implies that there is no particular reason to ascribe the trends in greenhouse gas concentrations in the ice-core records to biomass-burning changes either in the Americas or in the Southern Hemisphere alone, as opposed to ascribing them to global changes. The potential controls of the global decrease in biomass burning between ad 1500 and 1750 can be inferred both by direct comparison with climatic reconstructions (Mann et al., 2009) and through biomass-burning simulation studies (Pechony and Shindell, 2010). Annual mean temperatures were generally decreasing over the 1000 years prior to the interval of low biomass burning, both in the Americas and elsewhere (Figure 3), with temperatures in the Americas reaching a local minimum after those elsewhere (i.e. ad 1650 versus 1500), parallel in timing with the minima in the charcoal curves. Temperature, through its influence on vegetation productivity, and hence fuel, is the primary control of biomass burning on long timescales and large spatial scales (Harrison et al., 2010). The 'non-Americas' biomass-burning curve increases from ad 1000 to 1400, in opposition to the temperature curve, and is driven largely by an increase in biomass burning in the southern extra-tropics, in particular Australia Mooney et al., 2011). A recent simulation study (Pechony and Shindell, 2010) explains this upturn via increasing precipitation in the southern extra-tropics over this interval. In temperate Australia, where biomass burning is limited by moisture-controlled fuel availability (Mooney et al., 2011), an increase in effective moisture increases burning. Other aspects of these simulations (Pechony and Shindell, 2010) bear on the issues here -the interval of globally lower biomass burning observed in the charcoal data is well expressed both globally and regionally in simulations that either include or exclude a parameterization of the effects of population density on ignition and suppression of fire. The similarity in the trends in biomass burning in the Americas and elsewhere, the particular timing of the local minima, and the straightforward explanation for these in climate terms, suggest   (Power et al., 2008)) (b), plotted against temperature anomalies (Mann et al., 2009) for the Americas (c) and 'non-Americas' (d), and for the Hyde 3.1 human population estimates (Goldwijk et al., 2010) for the Americas (f) and 'non-Americas' (e). The period of the 'Little Ice Age', shown as a blue bar, is defined as ad 1400-1700 (Mann et al., 2009). Values for methane isotopes, δ 13 C of CH 4 (Ferretti et al., 2005;Mischler et al., 2009), are expressed as a 150-yr lowess smoothed (g). The magenta bar, labeled T3, is the interval of ad 1589-1730, argued by Ferretti et al. (2005) and Mischler et al. (2009) to be a period when population collapse led to decreasing biomass burning. Also shown are ice-core derived carbon monoxide (h) (Wang et al., 2010) and CO 2 values (i) for 'edml_CO2' (Siegenthaler et al., 2005), 'Law_CO2' (Etheridge et al., 2001), 'esp_CO2' (Siegenthaler et al., 2005), 'mm_CO2' (MacFarling et al., 2006), and 'EDC_CO2' (Monnin et al., 2004). The black line in (i) is a 150-yr lowess-smoothing of all CO 2 records (allowing comparison with the smoothed charcoal time series); the gray lines show 50-and 150-yr smoothing-spline curves as used in Frank et al. (2010), which do not down-weight outliers as in the Lowess smoothing. Similar to Frank et al. (2010), CO 2 values are plotted only up to 290 ppm. Modeled global fire (j) shows simulations with (red line) and without (gray line) direct anthropogenic influence (ignition and suppression) (Pechony and Shindell, 2010). The vertical blue gradient, beginning at ad 1500, provides a visual guide. See Figure 2 and Table 1 for locations of charcoal records, and Power et al. (2008Power et al. ( , 2010 for discussion of methods. that the interval of lower biomass burning in the Americas between ~ ad 1500 and 1750 is parsimoniously explained by the CCH.

Regional patterns of biomass burning, climate change, and population in the Americas
Analysis of regional 2-kyr charcoal series (Figures 4 and 6) across the Americas allows further consideration of the relative merits of the two hypotheses. If the PCH is correct, one would predict that the highest amplitude, and earliest, post-contact charcoal decline would have occurred in tropical Middle America (MA), where indigenous populations were highest (Denevan, 1992;Goldwijk et al., 2010) (Figures 2 and 4) and where their collapse through introduction of European diseases (Denevan, 1992) first began. Conversely, middle-to-high latitude regions would be expected to have experienced relatively more subtle, and later, post-contact declines in biomass burning because of their relatively lower population densities (Figures 2 and 4) and later arrival of European diseases. However, our results do not show this. Instead we find that sparsely populated (Denevan, 1992;Goldwijk et al., 2010) southern South America (SSA) experienced a major decline in biomass burning, commencing sharply at ad 1550, and of comparable magnitude to that of MA. Furthermore, the charcoal decline in MA began at ~ ad 1350, preceding European contact by 150 years (Figures 4d and 6). Similarly, in western North America (WNA), the onset of the charcoal decline also preceded European contact, beginning at ~ad 1450 (Figures 4c and 6, see also Marlon et al., 2012). The charcoal curve for tropical South America (TSA) also does not conform to the predictions of the PCH.
Instead of a major downturn, there was only a subtle decline in biomass burning in TSA, which did not begin until ad 1700, c. 200 years after European contact (Figures 4e and 6).
Despite this inter-regional variability in the initial onset, and magnitude, of charcoal declines, our data show a consistent postad 1500 charcoal minimum across the Americas, centered between AD 1600 and 1750, that correlates with the height of the LIA (Figures 3 and 6), suggesting a regional-to hemisphericscale climatic explanation for this charcoal minimum. The climatic expression of the LIA across the Americas (Figure 4), however, is heterogeneous, not only in terms of cooling intensity (relatively greater in the mid-high latitudes; Mann et al., 2009;Mooney et al., 2011), but also precipitation, which, for example, decreases in MA (Figure 4d) but increases in Patagonia (Mayr et al., 2005;Meyer and Wagner, 2009;Moy et al., 2008).
The spatial-temporal variability in LIA climate change, together with large-scale differences in vegetation type, fuel load and flammability, and indigenous population density (Figure 2), undoubtedly caused regional variations in biomass burning across the Americas. The low pre-Columbian population densities in high latitudes, including the northeast deciduous/boreal forest (NDB) and Patagonia (SSA) (Denevan, 1992;Goldwijk et al., 2010), is inconsistent with the PCH as an explanation for the high-amplitude steep charcoal declines there (Figures 4a,f and 6). Correlation between temperature and charcoal minima for NDB between ad 1600 and 1800 (Figure 4a) suggests that cooler temperatures reduced fuel flammability, and possibly also lowered the prevalence of lightning (an important ignition source in boreal forests today (Food and Agriculture Organization (FAO), 2007), and thereby decreased biomass burning then. In SSA the charcoal (Continued) minimum between ad 1600 and 1800 correlates more closely with peak precipitation than lowered temperature (Figure 4f; Mayr et al., 2005;Meyer and Wagner, 2009;Moy et al., 2008), which also would have reduced fuel flammability, and thereby suppressed biomass burning. The similarity among decreasing Greenland LIA temperatures and western US tree-ring-inferred temperatures and decreasing biomass burning across temperate North America (WNA and the Plains/SE USA (PSE)) (Figure 4b,c) also supports the LIA CCH.
The differing patterns of charcoal decline between tropical MA versus TSA are intriguing. In MA the charcoal decline began well before European contact and correlates well with the LIA declines in both temperature (Mann et al., 2009) and precipitation (Figure 4d), providing stronger support for the CCH than the PCH. The precipitation decline in MA (Cariaco record, Haug et al., 2001) correlates with a decrease in biomass burning, suggesting that a prolonged reduction in rainfall would have limited fuel availability in drier tropical ecosystems (Archibald et al., 2009). The Caribbean Cariaco record may not adequately capture the spatial complexity of climate across MA or LIA cooling, considering that fire is most common in tropical systems with intermediate levels of fuel loads and precipitation. A more detailed study of MA records, at higher spatio-temporal resolution, encompassing differences in vegetation, topography, and climate, is needed to explore the relationship between LIA climate change and biomass burning in this region of the Americas.
The relative importance of the LIA versus the PCH in explaining the charcoal decline in TSA is less clear than in MA because there is only a minor post-ad 1500 charcoal decline in TSA, beginning c. ad 1700 (Figures 4e and 6), which significantly postdates both the onset of the LIA in the region as well as the postcontact population collapse (Denevan, 1992) (Figure 4e). Therefore, neither climate change nor demographic collapse had a strong influence on post-ad 1500 biomass burning in TSA. The absence of a clear driver for the TSA biomass-burning signal may reflect the large seasonal and interannual variations in climate, vegetation and topography across this vast region -encompassing humid rainforest, cloud forest, seasonally dry forest, savannas, and paramo, and climatic regimes from ever-wet to arid and lowland tropical to alpine. Today, climatic phenomena such as the El The upper and lower 95% confidence intervals are shown in gray and the number of charcoal samples contributing to each overlapping 100-year window in the bootstrap analysis is shown below each regional composite charcoal time series. Each regional biomassburning composite is plotted against 100-year smoothed, regionally averaged, climate and human population data. Regionally summarized Hyde 3.1 (Goldwijk et al., 2010) population data, providing an aerially averaged estimate of population per region, are shown at the bottom of each regional panel. Climate reconstructions are aerially averaged paleotemperature reconstructions used in Mann et al. (2009). Proximal paleoclimate proxies of temperature and moisture variability from previously published studies are also shown for each region: Greenland Ice Sheet Project 2 (GISP2: Stuiver et al., 1997), Greenland summit ice-core δ 18 O (NGRIP, 2006), western North America tree-ring based temperatures (Jones and Mann, 2004), Cariaco basin titanium concentrations (as a proxy for precipitation) (Haug et al., 2001), and Quelccaya ice-core δ 18 O (Thompson et al., 1985).   (Figure 2) of the Americas. The LIA, as defined by (Mann et al., 2009), is shown by the vertical blue bar; the biomass-burning minima for the entire Americas, centered ~ ad 1600-1750, is shown as a gray box.
Niño Southern Oscillation (ENSO) have markedly spatially heterogeneous impacts in terms of precipitation across TSA (drought in some areas, increased rainfall in others) (Garreaud et al., 2009;Nepstad et al., 2004), in which case the LIA may well have also had different climatic impacts across the region. Furthermore, even under the same directional shift in precipitation, different vegetation types across TSA may have had opposite responses in terms of the amount of biomass burnt -e.g. an increase in dry season precipitation causing an increase in biomass burnt in moisture-limited savanna and caatinga (cactus thorn-scrub) as fuel load increases, but decreases in biomass burnt in seasonally dry forest as flammability decreases (Archibald et al., 2009;Nepstad et al., 2004).
Even at the local scale, Bush et al. (2007) found that neighboring sites, within a few kilometers of each other, can yield markedly different charcoal time-series. However, given that these lakes in the Peruvian Amazon have the same climatic regime, soils, and vegetation type (humid rainforest), variability in biomass burning was interpreted, not as climate-driven, but reflecting highly localized patterns of pre-Columbian anthropogenic rainforest burning (Bush et al., 2007).

Conclusions
We conclude that, although there are regional differences in the timing and pattern of biomass burning trends, there is a broadly consistent post-ad 1500 decrease in biomass burning across the Americas, which is most consistent with LIA climate change as the predominant driver or first order control at a regional to hemispheric scale. This implies that pre-Columbian indigenous peoples did not exert as strong an influence upon large-scale biomass burning as previously supposed. However, our findings do not preclude the possibility that post-contact population collapse may have been a more important control at smaller/finer spatial scales (e.g. evident from tight clusters of lakes in Peruvian Amazonia, Bush et al., 2007), or that the magnitude of the post-ad 1500 charcoal decline, especially in the densely populated tropics (MA) (Denevan, 1992), may have been amplified by the impact of postcontact population collapse. Our findings also show that the 16thcentury downturn in biomass burning was not unique to the Americas, but was a global phenomenon that was underway well before AD 1500. Consequently, the minima in ice-core trace-gas indicators of biomass burning, and in atmospheric carbon dioxide concentrations, after AD 1500 should not be attributed to postcontact decreases in anthropogenic biomass burning and consequent increases in carbon sequestration in the Americas.