Fire has long been recognized as an important ecological process because of its influence on species distributions and role in shaping other key ecosystem properties (Bond and Keeley, 2005). Fire also affects regional and global biogeochemical and hydrologic cycles (Shakesby and Doerr, 2006; van der Werf et al., 2006), geophysical processes (Morris and Moses, 1987; DeBano, 2000), and the climate system (Randerson et al., 2006; Ward et al., 2012; Saleh et al., 2014). Nevertheless, large gaps remain in our understanding of the interactions between fire and climate, despite an increasing need to manage fire and its emissions (Keywood et al., 2013).

Fire activity has been characterized at a wide range of spatial and temporal scales using field observations and historical data (e.g., Mouillot and Field, 2005; Gavin et al., 2007), dendrochronological data (e.g., Falk et al., 2011), satellites (e.g., Mouillot et al., 2014), ice cores (e.g., McConnell et al., 2007), and charcoal deposits in sediments, peat bogs, swamps, soils, and other environments (e.g., Whitlock and Bartlein, 2004). Sedimentary records are unique among these data sources because of the broad temporal and spatial coverage they provide, which includes reconstructions of fire history at local to global spatial scales and decadal to millennial temporal scales (e.g., Carcaillet et al., 2002; Brown, 2005; Marlon et al., 2008; Iglesias and Whitlock, 2014).

Results from paleofire research have helped lay a foundation for understanding the linkages among fire, climate, vegetation change, and human activities across a broad range of temporal and spatial scales. Fire-history data from sediment records highlight the importance of fire as a force of long-term global environmental change. Syntheses of data in the Global Charcoal Database (GCD), for example, reveal important variations in biomass burning during the last glacial period (Daniau et al., 2010), the last 21 000 years (Power et al., 2008; Daniau et al., 2012), and the last 2000 years (Marlon et al., 2008). With the increasing number of sites in the GCD, regional syntheses became possible, including long-term analyses of climate and human influences on burning in Australasia (Mooney et al., 2011; Williams et al., 2015), the Mediterranean (Colombaroli et al., 2009; Vanniere et al., 2011), the western USA (Marlon et al., 2012), and the Americas more broadly (Whitlock et al., 2007; Power et al., 2012).

Here we briefly review the history of biomass burning reconstructions based on charcoal data, and we introduce version 3 of the GCD (GCDv3, n= 736), which improves on GCDv1 (Power et al., 2008) and GCDv2 (Daniau et al., 2012) by adding 57 records. We also present the GCDv3 in a new globally gridded format along with several broad-scale syntheses created using the open-source paleofire R package (Blarquez et al., 2014). The new gridded maps illustrate the spatial and temporal variability in fire activity over the past 22 000 years, highlighting recent departures from the long-term trends. The maps should be useful for modelers as well as others in the Earth sciences, particularly given the wide-ranging impacts of fire. Finally, we review several important limitations to charcoal-based records and identify promising future directions for the field.

Fire-history research based on sediment-charcoal data has advanced rapidly in recent decades. Early analyses of sedimentary charcoal were typically conducted to support studies focused primarily on reconstructing past vegetation changes (Heusser, 1995; Fuller et al., 1998; Haberle, 1998; Behling, 2001). A few early studies focused more directly on fire (Swain, 1973; Burney, 1987; Delcourt et al., 1998). In many cases, microscopic (< 100 µm) charcoal particles were tallied alongside pollen grains. Pollen and charcoal particles were converted to concentrations using the abundance of exotic markers of a known quantity added to each sample, and charcoal data were presented as ratios of the relative abundance of charcoal to pollen. Records were usually sampled at low temporal resolution due to the intensive labor and time required to analyze pollen. Samples represented broad spatial areas because microscopic charcoal can travel hundreds of kilometers (Clark, 1988; Conedera and Tinner, 2010). Variations in both pollen and charcoal abundances can influence the ratios, however, and so changes in pollen productivity could produce apparent changes in fire activity when none occurred. The differential production of charcoal from grass versus wood species could also alter charcoal / pollen ratios. Thus, early reconstructions based on microscopic charcoal-to-pollen ratios provided new and often useful insights, but the information was relatively coarse and potentially unreliable for inferring past regional fire activity.

Currently, most paleofire researchers analyze macroscopic charcoal particles (> 100 µm) sampled contiguously from sediment cores to produce fire-history records that are more spatially and temporally precise (e.g., “local” histories at decadal timescales) compared to earlier methods. Macroscopic charcoal is typically quantified by simple particle counts or area measurements made using image analysis (Carcaillet et al., 2001b). However, particles can also be characterized using morphotypes. Two primary particle forms for non-arboreal charcoal exist: (1) cellular “graminoid” (thin rectangular pieces; one cell layer thick with pores and visible vessels and cell wall separations) and (2) fibrous (collections or bundles of this filamentous charcoal clumped together). Arboreal charcoal can be characterized by three morphotypes: (1) dark (opaque, thick, solid, geometric in shape, some luster, and straight edges), (2) lattice (cross-hatched forming rectangular ladder-like structure with spaces between), and (3) branched (dendroidal, generally cylindrical with successively smaller jutting arms; Jensen et al., 2007; Tweiten et al., 2009).

The analysis of high-resolution macroscopic charcoal records focuses on decomposing temporal variations in particle measurements into low- and high-frequency signals. The low-frequency signals were originally termed “background” (Clark and Patterson, 1997) and were thought to primarily reflect non-fire processes within and around a site unrelated to fire occurrence, largely due to sediment redeposition. The background component was therefore explicitly filtered out and disregarded in analyses. Subsequent research, however, demonstrated that background charcoal contains important information about the relative amount of biomass burned through time (Haberle and Ledru, 2001; Carcaillet et al., 2002), particularly when combined across multiple records (Clark and Royall, 1996; Carcaillet et al., 2002; Marlon et al., 2006). Thus, the collection of many records into a single repository, including those of insufficient resolution for local fire-history reconstructions, became an important prerequisite for reconstructing variations in biomass burning at regional to global spatial scales.

Recent studies demonstrate that background charcoal corresponds well with independent evidence of area and/or biomass burned both at landscape scales (Higuera et al., 2011; Kelly et al., 2013) and regionally (Marlon et al., 2012). However, it is not possible to quantify absolute area burned in the absence of a calibration data set, and the influences of non-fire-related processes such as erosion or vegetation change on biomass burning reconstructions remain poorly understood (Aleman et al., 2013). These limitations highlight a need for more calibration studies to understand how charcoal production and taphonomy relates to the area and amount of biomass burned across a range of vegetation types and climate conditions.

Another recent advance in fire research is the reconstruction of fire frequency based on peaks in sedimentary charcoal records. Fire frequency is an important component of the fire regime, but such analyses require data sets with decadal resolution that are relatively uncommon in the GCD. In order to reconstruct fire frequency, records must be sampled contiguously, have high temporal resolution relative to the expected mean fire return intervals, and have sufficient particle counts in each sample to separate peaks from “background” (Higuera et al., 2007, 2010). In addition, relatively stable sediment accumulation rates are ideal because peak frequencies will vary with changes in sedimentation rates (Carcaillet et al., 2001a; Higuera et al., 2010). For these reasons, our analyses of the GCDv3, which are focused on broad-scale changes in fire, are limited to the reconstruction of fire activity or biomass burning rather than to changes in fire frequencies.

Many other methodological approaches to long-term fire-history reconstruction are developing from a variety of combustion products in ice cores (Kehrwald et al., 2013), including the analysis of ammonium (NH4+; Savarino and Legrand, 1998), methane (Fischer et al., 2008), carbon monoxide (CO; Wang et al., 2010), black carbon (Han et al., 2012; Lehndorff et al., 2015), vanillic acid (McConnell et al., 2007), and levoglucosan (Zennaro et al., 2014), as indicators of past fire activity. Laboratory and analytical methods are also advancing through the use of image analysis for counting charcoal and charcoal morphotypes (Enache and Cumming, 2006; Jensen et al., 2007; Thevenon and Anselmetti, 2007; Gu et al., 2008; Moos and Cumming, 2012).

Overall, the wealth of methods and approaches to fire research are providing a broad range of insights into fire, both as an ecological process and as an integrated component of the Earth system. However, much work remains to understand the impact of wildfires and biomass burning emissions on climate, and vice versa (Keywood et al., 2013). Research on human–fire interactions using paleorecords is developing rapidly (Colombaroli et al., 2008; Perry et al., 2012; McLauchlan et al., 2014; Munoz et al., 2014), but applying insights from paleofire research to fire management and emissions reduction plans remains comparatively limited (Whitlock et al., 2003; Cyr et al., 2009; Coughlan and Petty, 2012; Munoz et al., 2014). By compiling diverse types of paleofire data in a central location and developing open-source analysis tools to explore those data, research can advance more quickly on these topics.

The structure and contents of earlier versions of the GCD are outlined in Power et al. (2010). Here we review the database design and focus primarily on detailing new entries in GCDv3. Version 3 extends the total number of sites in the GCD to 736. It includes 679 sites from version 2 (Daniau et al., 2012) as well as new sites from recent regional syntheses from Australasia (Mooney et al., 2011), the Americas (Marlon et al., 2009, 2012; Power et al., 2012), and Europe (Vanniere et al., 2011).

To efficiently visualize GCDv3 and facilitate comparisons with model output, we present a spatially gridded version of GCDv3 using dot maps (Figs. 5, 6) alongside composite time-series curves (Figs. 5, 6). Vertical gray bars on the composite graphs indicate the time periods reflected in the maps. Each dot on the map represents a composite charcoal series constructed from all records within a fixed distance of the dot, such that the area represented by each dot is the same. However, the dots are positioned on a regular latitude/longitude grid, and the area of each grid cell varies by latitude (i.e., cells near the Equator cover larger areas than those near the poles); spacing dots in this way maximizes the compatibility of the gridded charcoal data set with other global data products. On such a grid, the absolute distance between dots (or nodes) decreases with distance from the Equator. We defined the radius used to identify sites contributing to a dot as half the distance between diagonally adjacent dots at the Equator (e.g., ∼ 395 km for a 5∘ × 5∘ grid). This radius ensures that all GCD sites contribute to at least one dot but also causes sites to influence multiple dots, especially at high latitudes where dots are relatively close together in terms of absolute distance (Fig. 7). Finally, our gridding approach prevents interpolation into areas that are not represented in the GCD, which is desirable given the great spatial heterogeneity of fire regimes.

Anomaly maps illustrate the gridding approach at six discrete intervals during the past 1000 years (Fig. 5, left panel) and 22 000 years (Figs. 6, left panel). Maps from each 100-year period during the past millennium and each 1000-year interval since the Last Glacial Maximum (LGM) are provided in the Supplement. The charcoal values are plotted on a 5∘ grid, and the dots are colored and sized to reflect the value and statistical significance, respectively, of the biomass burning anomalies (Fig. 5 and 6, right panels). The maps include data from three 100-year intervals (Fig. 5) and three 1000-year intervals (Fig. 6). Red dots on the maps indicate positive mean z scores for sites in that location relative to their own long-term mean, which was calculated using a base period between 1000–200 years (Fig. 5) and 21 000–200 cal yr BP (Fig. 6). Blue dots on the map indicate negative mean z scores. Because each dot shows changes in biomass burning relative to its own long-term average for that location, comparisons among dot colors on a single map (i.e., for a specific time) cannot be used to infer geographic patterns in biomass burning. For example, it is possible (or very likely, in fact) that for a given time period, a blue dot in Africa represents more biomass burning than a red dot in the Arctic. By contrast, changes in the color of a dot over time indicate meaningful temporal variability in the relative rate of biomass burning. A red dot in one time period that changes to a blue dot in the same location at another time period, for example, reflects an actual decrease in biomass burning over time at that location. One point of note is that it is possible in some cases for a recent time period to have less data than an older time period because samples from sediment cores are not regularly spaced in time, and core sections or tops are sometimes lost or destroyed in the field or during extraction. Most lake sediments provide continuous records, but soil and bog profiles often have hiatuses when sites dry out or peat is burned, and occasionally this happens in lake and marine sediments as well. Another reason that a site may have less data closer to present than in the distant past is when sedimentation rates decline over time. In this case, a section of the core the represents the most recent past may only have one or two samples, whereas sections of the same size further down core may contain many samples.

A diagnostic map of the gridded charcoal data shows the effects of summarizing all data within a constant specified distance from each dot (Fig. 7). Effectively, the gridding approach allows each site to influence an equivalent spatial area on the map. However, it is helpful to keep in mind that given the same number of sites at high latitudes and at the Equator, the high-latitude sites will be more smoothed relative to those at the Equator, which is evident in the diagnostic maps from different time periods. Another effect of using equal-area circles to construct the dot maps is that a circle can be centered quite far from shore but still encompass a site on land. Thus dots may represent terrestrial sites despite being plotted in the ocean on our maps (although in some cases they represent charcoal data actually collected from marine cores; see Figs. 1 and 7 for a comparison between location of sites and dots). Large (small) dots indicate biomass burning anomalies that are (not) significantly different from 0.

Global biomass burning during the past millennium (Fig. 5) shows a gradual long-term decline until the 17th century during the Little Ice Age (LIA; Mann et al., 2009), as observed in previous reconstructions (Marlon et al., 2008). This decline is more pronounced in the Northern than Southern Hemisphere (Fig. 5, top and bottom panels). After the LIA, global biomass burning increases gradually until the 19th century, then rapidly until the 20th century. Maximum levels of biomass burning in the Northern Hemisphere occur prior to maximum levels in the Southern Hemisphere, and both hemispheres experience sharp declines in biomass burning during the second half of the 20th century. The maps of biomass burning show the spatial heterogeneity underlying the composite curves. Biomass burning in central and eastern North America is highest from 1850 to 1950 CE, for example, whereas burning in western North America is highest during the most recent period (1950–2010 CE). In contrast, burning in western and southern Europe is generally higher 1000 years ago than it is in the past two centuries. Burning in southeast Asia is very high from 1850 to 1950 CE and remains high in several locations for the period 1950–2010 CE where data are available.

The most recent upturn in fire activity globally, but particularly in the Northern Hemisphere reconstruction, is supported by a larger data set than GCDv1. Marlon et al. (2008) used GCDv1 to document the large decrease in biomass burning in the 20th century, but the reconstruction had large uncertainties in the trend over the last few decades. The addition of new records to versions 2 and 3 of the GCD, along with a finer-scale temporal focus now reveals the most recent increases in fire activity observed not only in the charcoal data but also in several lines of independent evidence, including satellite and observational data (Giglio et al., 2013; Dennison et al., 2014).

Global biomass burning since the LGM, 21 000 years ago, shows a long-term increase (Fig. 6) consistent with increasing temperatures, atmospheric CO2 concentrations, and burnable biomass (Daniau et al., 2012; Martin Calvo et al., 2014). The reconstructions from GCDv3 (red lines) are very similar to those from GCDv2 (thin gray lines) for the globe, northern extratropics (> 30∘ N latitude), tropics (> 30∘ N latitude and < 30∘ S latitude), and southern extratropics (< 30∘ S latitude), with the exception of burning in the northern extratropics during the LGM, which registers as very low with the additional records in GCDv3 as compared with GCDv2 (Fig. 6). However, the Northern and Southern hemispheres show somewhat inverse patterns of burning during the Holocene, with fire increasing steadily in the northern extratropics during the Holocene, but declining in the early to mid-Holocene in the tropics and southern extratropics, before increasing in the late Holocene.

The gridded maps provide insight into the spatial variations in biomass burning since the LGM.

Burning is generally higher in the past millennium than at any time since the LGM with the exception of central-western South America (Fig. 6), where some locations had higher than average burning during the mid-Holocene and below average burning in the past millennium. Levels of burning during the LGM in turn were generally lower than at later periods, with a few localized exceptions. Particularly high levels of biomass burning in the past millennium are observed in many locations in the Southern Hemisphere (e.g., New Zealand, central Africa, the Amazon) as well as in parts of the Northern Hemisphere (e.g., northeastern North America, southern California, and the southern Iberian Peninsula). The maps also reveal spatial coherence in regional biomass burning since the LGM, which likely reflects climate controls on fire in some cases and human controls on fire in others – the degree of coherence alone cannot distinguish causal mechanisms at this scale.

The development of the GCD is motivated by the need to understand the history of fire on Earth and the linkages among fire, climate, vegetation, and human activities. As the GCD continues to expand, the expectation is that knowledge of fire histories will become more detailed. Analyzing charcoal-based fire-history records with modern data from satellites (e.g., van der Werf et al., 2010; Giglio et al., 2013), fire scars (e.g., Girardin and Sauchyn, 2008; Marlon et al., 2012), or historical records (e.g., Mouillot et al., 2006; Lamarque et al., 2010) is necessary to connect relative or qualitative variations in biomass burning from charcoal records (Aleman et al., 2013) to quantitative estimates of burned area or carbon emissions. To test hypotheses related to drivers of fire activity over longer timescales, however, research needs to integrate paleofire data with modeling approaches. As the spatial network of charcoal records become denser, there is increasing opportunity to identify locations where varying types of fire records overlap and thus more opportunities to study changes in fire regimes that span multiple spatial and temporal scales.

Fire modeling efforts have advanced rapidly in the last decade (Arora and Boer, 2005; Kloster et al., 2010; Kelley et al., 2014; Lasslop et al., 2014; Yue et al., 2014; Le Page et al., 2015), providing a better understanding of the varied impacts that fires have on humans, the biosphere, and the atmosphere (Harrison et al., 2010), as well as the mechanisms through which climate changes and human activities affect fire regimes. Simulations of fire activity using physically based empirical relationships between flammability and its controlling variables, such as temperature and soil moisture, have helped identify the global drivers of modern burning (Arora and Boer, 2005; Kloster et al., 2010; Pechony and Shindell, 2010; Thonicke et al., 2010; Li et al., 2013; Pfeiffer et al., 2013). Fire modeling studies have also qualitatively compared paleofire trends with simulated global fire activity (Pechony and Shindell, 2010; Kloster et al., 2012; Li et al., 2013), but quantitative testing of the physically based relationships that drive fire models – the mechanics of the models themselves – has only focused on modern climate conditions thus far. As a result, large gaps in knowledge exist about how fire, climate, vegetation, and humans interact under different climate conditions and over long timescales. Despite the fact that mechanistic global fire models remain largely untested outside modern climate parameters, these models are being used to predict the response of fires to ongoing climate change (Pechony and Shindell, 2010; Kloster et al., 2012).

The fire modeling studies that have explicitly considered paleofire data provide examples of the challenges in comparing data and models. A study by Pechony and Shindell (2010) tested a global fire model scheme within a Global Climate Model simulation of the past millennium, for example, and found that at coarse spatial scales precipitation was the most important factor driving multi-centennial variations in fire activity in the model. However, the spatial patterns underlying these trends, and the extent to which finer-scale variations match paleofire evidence are unknown. Moreover, the finding that precipitation is more important than temperature in driving trends in fire activity globally contradicts analyses of paleodata (Daniau et al., 2012; Marlon et al., 2012; Power et al., 2012; Marlon et al., 2013), as well as satellite remote-sensing data (Bistinas et al., 2013), raising key questions about how temperature, precipitation, and their interactions affect variations in global biomass burning. Another fire modeling study (Brücher et al., 2014) compared model output to paleofire data from the GCD at regional scales from the mid-Holocene until the pre-industrial era in the 18th century. Kloster et al. (2015) go one step further to test the sensitivity of the same model to variations in fuel availability, fuel moisture, and wind speed, as well as their synergy for the same regions and time period.

The new approach to gridding GCD data presented here (and included in the paleofire R package) should help further paleofire data–model comparison studies. Whereas modeling studies to date have focused on global or regional trends, the growing number of records in the GCD allows for evaluation of model performance at finer spatial scales. However, site-specific variability is often high among charcoal records, and driver data sets for many global fire models may be of relatively coarse resolution. As a result it is ill-advised to compare model output to individual charcoal records. The gridded approach offers a flexible compromise that can be tuned in terms of spatial resolution depending on data availability, model driver data sets, and other factors. As an example, we present here a global map of simulated area burned using the CLIMBA model (Brücher et al., 2014), overlaid with gridded composite charcoal anomalies from the GCD. CLIMBA consists of the EMIC CLIMBER-2 (CLIMate and BiosphERe; Petoukhov et al., 2000; Ganopolski et al., 2001) and JSBACH (Raddatz et al., 2007; Brovkin et al., 2009; Reick et al., 2013; Schneck et al., 2013), which is the land component of the Max Planck Institute Earth System Model (MPI-ESM, Giorgetta et al., 2013). Simulated area burned throughout the Holocene was treated analogously to GCD data to produce a gridded map of area-burned anomalies at 6000 BP relative to present (i.e., 6000 BP z scores minus 0 BP z scores).

Overall, data–model agreement is weak, with many grid cells disagreeing in terms of the sign of the anomaly estimated from the CLIMBA model versus GCD data (Fig. 8a). However, the exercise shows promise for some regions. In eastern North America, for example, site-level GCD data are difficult to reconcile with model output (Fig. 8b), but the gridded data product shows that both data and model generally agree that 6000 BP was a period of lower fire activity than present for the region (Fig. 8c). It is beyond the scope of this paper to evaluate the importance of this agreement, or the causes of data–model mismatch in other regions throughout the globe. Instead, we present the example as a proof-of-concept to motivate future studies. Important basic research topics to pursue include evaluation of spatiotemporal patterns in data–model comparisons, and a critical assessment of how uncertainties in both GCD data and fire model output contribute to the comparisons.

Using fire-history data from the GCD to constrain fire model simulations, or conversely, using fire model simulations to understand variability in the fire-history data from the GCD, requires careful consideration of the uncertainties associated with both data types. For paleofire records, quantifying and accounting for age uncertainties is a major concern, but progress is occurring on this front through the development of Bayesian age-modeling methods (Blaauw and Christen, 2011; Goring et al., 2012). Uncertainties in charcoal records also come from the many natural processes related to charcoal production, transportation, and deposition, which interact to produce variability in charcoal accumulation over time. These processes are being studied through field experiments and calibration studies that will enable the development of higher quality fire-history reconstructions and a better understanding of uncertainties (Tinner et al., 2006; Higuera et al., 2011; Aleman et al., 2013). An important source of uncertainty in global fire models is the parameterization of the processes most directly controlling fire activity (e.g., human influence, climate influence; e.g., Pechony and Shindell, 2009; Pfeiffer et al., 2013). The sensitivity of simulated fire activity to such parameterizations needs to be tested to understand model uncertainty. Uncertainty in modern fire records arises from any extrapolation or interpretation beyond the available fire records (Mouillot et al., 2006) or to limits in the satellite data itself (Giglio et al., 2013). With detailed considerations of both the limits and uncertainties of all data sources and model parameterizations, connecting GCD to fire models represents the natural evolution in the effort to understand fires in the Earth system.

There are several research areas that, with further development, would facilitate rapid integration of fire data and a more comprehensive understanding of fire across spatiotemporal scales. Here we identify particular areas that would help address specific barriers to progress in paleofire research.

Charcoal calibration studies in diverse environments. A major limitation of biomass burning reconstructions is that they can only represent relative changes in burning from an arbitrary baseline. Calibration studies that relate variability in charcoal accumulation to fire regime characteristics from historical, fire-scar, satellite, and other recent data could allow additional information to be obtained from charcoal records. Given the complexities of charcoal production, transportation and deposition, it is unlikely that the absolute amount of biomass burning from a single paleofire time series can be known; however, with a better understanding of how charcoal abundances relate quantitatively to area burned or other fire-regime metrics, constraints on paleofire reconstructions can be established and integrated into models that can then provide quantitative estimates of variables like area burned and carbon emissions.

In addition to the research needs above, several practices could aid in the development of high quality charcoal-based fire-history reconstructions and facilitate data integration across labs and therefore across different environmental contexts. The practices may be more useful to new researchers entering the field or establishing new labs.

Continuous sampling of macroscopic charcoal data. Although many researchers now sample lacustrine sediment continuously and quantify macroscopic charcoal, many continue to tally microscopic particles, or to sample discontinuously. Taking the latter approach may be necessary due to methodological, funding, or other constraints, but when it is possible, the former approach is more desirable. Research on charcoal particle size classes supports macroscopic particles (> 100 µm) as a reliable indicator of local (within 1–10+ kilometers of a study site) fire activity (Whitlock and Bartlein, 2004), whereas smaller particles integrate biomass burning from a larger spatial domain (Conedera et al., 2009). If both macroscopic and microscopic particles can be tallied, they may provide complimentary evidence of past fire regime change. However, if only one particle size is collected, analysis of macroscopic charcoal usually provides a better signal for local fire reconstruction. While continuous sampling is more time and cost intensive, it facilitates reconstructing event frequency, aligning multiple cores, detecting unique events, and examining rates of change.

The GCDv3 incorporates 736 charcoal records and can now be gridded globally for the modeling community to ease future data–model comparisons. Fire-history reconstructions from the GCDv3 demonstrate that increases in biomass burning since the last glacial period were widespread, as are unusually high levels of burning over the past several decades. Present-day burning inferred from the charcoal data is particularly high in western North America and southeastern Australasia. Detailed reconstructions of temporal variations in biomass burning during the past 1000 years reveal that a global biomass burning decline from 1000 to the LIA was more pronounced in the Northern than Southern Hemisphere. In addition, variations in fire activity during the past 200 years show very different spatial patterns. In general, data–model comparisons with paleofire data provide a powerful method for testing hypotheses about interactions between climate and fire outside the range of modern climate conditions. Results from such data–model comparisons will highlight gaps and weaknesses in both data and models, allowing targeted refinements to be identified and prioritized. We identify five areas of focus to promote future progress in paleofire research, including (1) charcoal calibration studies in diverse environments, (2) multiproxy studies of paleofire history, (3) paleofire data–model comparisons, (4) filling gaps in paleofire data, (5) comparisons between charcoal data and other large data sets, and (6) enhanced data extraction from existing cores, like continuous sampling and herbaceous charcoal identification.