Late 19Th Century Pioneer Agriculture Revolution in Northern Hemisphere Ice Core Data and Its Atmospheric Interpretation G

The University of Maine DigitalCommons@UMaine

Earth Science Faculty Scholarship Earth Sciences

10-27-1996 Historical Biomass Burning: Late 19th Century Pioneer Agriculture Revolution in Northern Hemisphere Ice Core Data and its Atmospheric Interpretation G. Holdsworth

K. Hiuchi

G. A. Zielinski

Paul Andrew Mayewski University of Maine, [email protected]

M. Wahlen

See next page for additional authors

Follow this and additional works at: https://digitalcommons.library.umaine.edu/ers_facpub Part of the Atmospheric Sciences Commons, Climate Commons, Geochemistry Commons, Glaciology Commons, and the Hydrology Commons

Repository Citation Holdsworth, G.; Hiuchi, K.; Zielinski, G. A.; Mayewski, Paul Andrew; Wahlen, M.; Deck, B.; Chylek, P.; Johnson, B.; and Damiano, P., "Historical Biomass Burning: Late 19th Century Pioneer Agriculture Revolution in Northern Hemisphere Ice Core Data and its Atmospheric Interpretation" (1996). Earth Science Faculty Scholarship. 241. https://digitalcommons.library.umaine.edu/ers_facpub/241

This Article is brought to you for free and open access by DigitalCommons@UMaine. It has been accepted for inclusion in Earth Science Faculty Scholarship by an authorized administrator of DigitalCommons@UMaine. For more information, please contact [email protected]. Authors G. Holdsworth, K. Hiuchi, G. A. Zielinski, Paul Andrew Mayewski, M. Wahlen, B. Deck, P. Chylek, B. Johnson, and P. Damiano

This article is available at DigitalCommons@UMaine: https://digitalcommons.library.umaine.edu/ers_facpub/241 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 101, NO. D18, PAGES 23,317-23,334, OCTOBER 27, 1996

Historical biomass burning- Late 19th century pioneer agriculture revolution in northern hemisphere ice core data and its atmospheric interpretation G. Holdsworth,•'2 K. Higuchi,a G.A. Zielinski,2 P.A. Mayewski,2 M. Wahlen,4 B. Deck,4 P. Chylek,s B.Johnson,s and P. Damianos

Abstract. Ice core data from Yukon and Greenland spanning from •01750 to 1950 indicate that between •01850 and <1910 a clear atmospheric signal exists of an episodic biomass burning event that is referred to as the Pioneer AgricultureRevolution. This is bestseen in NH4+ ionand particulate concentrations but also in some limited black carbon concentration data, where for all three quantities maximum levels reach about 3 times the prerevolution background concentrations.Tree cellulose513C data and someearly, controversial, French, air CO2 data, occurring within the same time interval, are interpreted as providing other independent evidencefor the same, mainly North American, late 19th century biomassburning event. Some hitherto problematic northern hemisphereice core derived CO2 concentrationdata may now be interpreted as containing a biomass burn signal, and these data are compared,especially as to the time of occurrence, with all the other results. A global carbon cycle model simulation of atmospheric CO2 mixing ratios usinga maximuminput of 3 Gt(C)/yr at northernmidlatitudes produces "anomalous" CO2 levels close to some of the ice core carbon dioxide values. However, other values in this data set do not reasonably represent fully mixed atmospheric values. This suggeststhat these values might be transients but still "tracers" for biomass burning. Nevertheless, it appears possible that interhemispheric CO2 gradients of similar magnitude to the present one could have existed briefly late last century.

Introduction stageof the eventwas referredto by Wilson[1978] as the "PioneerAgriculture Explosion,"a term which de- The term "PioneerAgriculture Revolution"(PIA- scribesthe rapid rise of biomassclearing and burning GREV) usedthroughout this paper was first formulated associatedwith agrarian activities that occurred, ac- by Leavitt[1991], who modifiedthe originalterminology cordingto the tree core data, from about 1860 to 1890. of Wilson[1978]. Here its useis basedon evidencefor Wilson[1978] deconvoluted 513C data from a singlebut sustained,anthropogenically triggered biomass burning free standing Californian Bristleconepine core to com- and related emissionsthat attained maximum intensity pute the changein atmosphericCO2 concentrationthat during the latter half of the 19th century. The main the data implied. He estimated that the preindustrial revolution CO2 levels were close to 270 parts per mil- lion by volume(ppmv) and that over three decadesit 1Arctic Institute of North America, Universityof Calgary, climbed to about 300 ppmv. Alberta, and National Hydrology Research Centre, Saskatoon, The biosphere-atmosphereprocess model that Wilson Saskatchewan, Canada. used is now considered to be oversimplified. Whereas 9'GlacierResearch Group, Institute for the Study of Earth, Wilson assumedthat the apparent magnitude of the Oceans and Space, University of New Hampshire, Durham. event that he identified in one tree core was global in 3AtmosphericEnvironment Service, Department of the Envi- extent, our data are more indicative of the probably ronment, Downsview, Ontario, CanaAa. dominant North American component. This is also evidently the casefor Wilson's data. It is now known that 4Scripps Institution of Oceanography,University of Cahfornia, La Jolla. preindustrialCO2 levelswere closerto averagingabout 280 ppmv [Friedli et al., 1986; Wahlenet al., 1991]and sAtmospheric Science Program, Dalhousie University, Halifax, that Wilson's "explosionstep" in carbon dioxide was Nova Scotia, Canada. neither everywhereas large as he calculatednor as sud- den, nor in one single step. Furthermore, many tree Copyright1996 by the AmericanGeophysical Union. cores examined subsequentlysimply do not show any Papernumber 96JD01158. significantperturbation in 5•3C duringthe PIAGREV. 0148-0227/96/96JD-01158509.00 Such a result is due, evidently, to the complexitiesin the

23,317 23,318 HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA

processeswhereby the different isotopesof carbon are of atmosphericCO2 was active during the last quarter sequesteredinto tree cellulose,as well as to the length of the nineteenth century." Even so, it was evident that of the growing seasonand the position of the tree with these data were highly controversial.However, seen in respect to the major smoke plumes. Here it should be the light of the presentice core data, someof the French noted that the main burning seasonduring the PIA- direct-air CO2 concentration data now take on a sup- GREV must have been several months in duration; in portive role in defining part of the PIAGREV. comparison,the growing seasonfor many of the trees Some previously unpublished "anomalous"ice core sampled by dendroclimatologistscan be as short as 6 CO2 resultsfrom Mount Logan and somerecently pub- weeks(as in the caseof the Bristleconepine). The lisheddata for Greenland[Wahlen et al., 1991]are now tree core data as a whole representsa curiousenigma indirectly and qualitatively supported by the chemical in terms of defining the PIAGREV, yet it still forms ice core evidence presented here. Finally, to determine an important part of our evidence. Nevertheless,there if the CO2 data are realistic, we used an estimated car- is overwhelmingglaciochemical and historicalevidence bon input function (derivedfrom publishedestimates that the PIAGREV was a real event. The fact that the and from new data presentedhere) in a globalcarbon signature of the PIAGREV has been left in several es- cycle model. The complexity of carbon sink functions sentially independent domains strengthensthe overall (involvingthe well-known"missing sink") make this ex- interpretation of the event. ercise somewhat qualitative, but results are presented The main activities associatedwith the agriculture which show a small perturbation in atmosphericcarbon revolution were the cutting of forest stands and the dioxidein the high latitudes of the northern hemisphere clearing of vast tracts of land for agricultural purposes during the PIAGREV. Whereas the global atmospheric and the burning of wood, both as a means of removal carbon cycle model computes CO2 concentrationsin a and for domesticheating. Contemporaneouswith these fully mixed atmosphere after each time step, we inter- activities were the tilling and repeatedploughing of the pret some of the ice core CO2 concentration data as soil. This activity introduced mineral and organic soil not being representativeof a fully mixed atmosphere. particles into the atmosphere. it has been estimated This doesnot necessarilydetract from the value of those [Won#, 1978] that for the contiguousUnited States, data, which can still be used qualitatively as a "tracer" 25% of the forest area was cleared during the 19th cen- for the most intense phasesof the PIAGREV. tury. In addition to natural forest fires, many large Severalsources of information suggestthat the PIA- "controlled" and uncontrolled fires occurred in mid- GREV event consistedof two main episodes. Peaks in latitude North America, particularly during the latter activity evidently occurred between about 1850 to 1860 quarter of last century. Whereas woodenbuildings ini- and betweenabout 1890to 1900,the secondepisode (or tially sequesteredcarbon, large property fires, some- emissionpulse) being the largerone. In the last decades times involvingwhole towns (as in the caseof Seattle, of last century the major railways of North America Spokane,Ellensburg, and Vancouverin 1889 [Gould, were being established,and the secondpulse of the PI- 1938])suddenly injected large amountsof particulate AGREV may be directly and indirectly associatedwith matter, carbon dioxide, and other gasesinto the at- that event. mosphere. Enhanced levels of black carbon would also The evidencepresented here strongly suggeststhat have been introduced into the atmosphereduring this the Antarctic ice core sites are not very "sensitive" to time. The opening up of vast tracts of interior North short-term transient events in the middle to high lati- America with the building of the railway system during tudes of the northern hemisphere, yet the Antarctic ice the latter half of the last century must have played an core gas data have usually been used to describethe important role in the PIAGREV. state of the global atmosphereprior to about the mid- Here, usingice coredata, we are able to trace, in more dle of the 20th century. Most importantly, the northern detail than before, the courseof the PIAGREV in the hemisphereice corescontain information on trace gases northern hemisphere. Principal evidence for the PIA- that is not immediately apparent in the high-latitude GREV comes from ice core sites in central and southern ice cores of the southern hemisphere. This is some- Greenlandand on Mount Logan(Yukon, Canada). Af- what surprisingconsidering that there were concurrent ter reviewing the nature of the ice core sites and their pioneeragricultural activities taking place in the south- positions with respect to emission sources,previously ern hemisphere[Wilson, 1978] and that someexposed published and new data are presented to show that trees in coastal South America show a 513C decrease at there were sustained biospheric emissionsduring the the time of the PIAGREV. The observation that other period from about 1850 to 1900. New detail for the southern hemispheretrees do not show this underscores course of the PIAGREV comes from ice core chemical the importantfact that tree 513Cdata are oftenvery data, from existing but selectedtree ring data, and from difficult to interpret. This is apparent from the very rich direct air measurements made late last century. These literature on the subject. Recent developmentsin mod- latter measurementswere discussedby several authors eling forest fire plume transport on a hemisphericscale in somedetail between 1982 and 1985 in the journal Cli- [e.g., iacobeiliset ai., 1994] seemto providea means matic Change.$tanhill [1982]stated that oneparticular of interpreting some of the large variability in the ice seriesof the French data was "compatible with the hy- core and the tree ring data, and particularly between pothesisthat a major and variable nonfossilfuel source differencesin the publishedtree 513Cdata. HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA 23,319

The philosophy of this paper is based on an inter- boundariesmay generate"excess" CO2 [Delmas,1993], disciplinary approach to problem solving and on the but we do not have convincingin situ ice core chemical principle that multiple piecesof independent evidence, evidence to support this model. If this chemical pro- which on their own would hardly be sufficient to sup- cesshas occurredin our samples,then, becauseof its port a thesis, accumulatea higher degreeof probability position in the core, a necessaryincrease in carbonate when considered in combination. Thus a "cumulative dust influx would have been required about a century probability" results that must be significantly greater earlier than when the PIAGREV-era gas was trapped. than any one of the individual probabilitiesattached to Then,the CO2 anomalies(as well as the carbonatemin- each singlepiece of evidence. This conceptis given a eral dust) wouldbe unrelatedto the PIAGREV. There formal but not a rigorousbasis at the end of the section are, however, other possibilitiesfor the anomalousCO2 synthesizingthe evidencefor the PIAGREV. concentrations in the ice core. When only trace gas data with the lowest error bars Ice Core Sites and Data Overview are selected from the Mount Logan ice core data set, the "excess"CO2 anomalies(above the Antarctic CO2 The namesand locationsof the principal ice coresites referencecurve) have a striking correspondencein time are asfollows: "GISP2" (72.2øN;37.8øW; 3200 m eleva- with the PIAGREV emission history to be described. tion abovesea level (asl)) is in centralGreenland, 20D This is alsothe casefor the GISP2 data. Riming (solid (64.99øN;44.44øW; 2625 m elevationasl) is in south- condensate from a supercooled, supersaturated atmo- ern Greenland,and the Mount Logan site (60.62øN; sphere)onto the surfacesof sastrugi(snow dunes), par- 140.52øW;5347 m elevationasl) is in the St. Elias ticularly on Mount Logan, may also have affected trace Mountains, Yukon, Canada. Published data from the gas (and chemicalspecies) enrichment factors, but we Cr•te (Greenland)ice core site (110 km southof GISP2; are not currently able to quantify this process. 3170 m elevationasl) is alsoused. GISP2 and Mount Major chemical time series for the GISP2 core have Logan are the coldestsites available in their respective been presentedby Mayewskiet al. [1993]. The diag- regions,with mean annual temperatures of-30 4- IøC. nosticpulses of thePIAGREV may be seenin theNH4 + The two sites,one in centralGreenland [Wahlen et al., series.Just after AD 1900,when NH4 + concentrations 1991](site temperature-31øC 4- 0.5ø) and the otherin declineto pre-PIAGREV levels,both (non-seasalt)SO• the St. EliasMountains [Holdsworth et al., 1992](-29øC and NOa+ beginclimbing, indicating the switchfrom 4- 0.5ø), lie essentiallyin the melt-free "polar snow" mainly biomassfuel burning to predominantly fossilfuel zone [Benson,1962]. This thermal regimeis important combustion. for the preservation of chemical speciesin the snow. It Core from the southernGreenland site (20D) (mean is especiallyimportant when studying past CO2 con- annual temperatureabout -20øC) has providedinter- centrations in cores where the "ice" formed from dry pretable time seriesof the soluble ions over the last two snow. The formation of ice at the surfacewill sequester centuries[Mayewski et al., 1990; Dibb et al., unpub- CO2 which will subsequently contaminate the core in lisheddata]. One of thesetime series(NH4 +) is used the regionat depth wherethe intersticesbetween grains here, with the GISP2 data, as primary evidence for the are sealingoff a sample of the atmosphericair. Careful biomassburning component of the PIAGREV. Trace core sampling will reveal the local contamination, en- CH4 and N20 data from this site and from Mount Lo- abling the "correct" atmosphericsample to be obtained gan and relevant to the PIAGREV have been presented [Staufferet al., 1985]. elsewhere[Dibb et al., 1993]. We now presentthe pri- Carbon dioxide data for the GISP2 site are already mary data used in building up a self-consistentdescrip- published[Wahlen et al., 1991] but were not fully in- tion of the PIAGREV. terpreted at that time. There, the formation of ice at We have reported the data as the original ionic (or the surfaceis rare, and strict samplingprocedures were other component)concentrations in the ice core melt- used which discriminated against surface-formed"ice" water, rather than convertingto fluxes. Although,in in the core. At the site on Mount Logan, transient principle, this step could be taken, sincethe timescales meltwater ice lenses have not been seen in the recent and annual incrementsare known very well, we have not snowpack.These would affect not only gas adsorption done so because there are too many uncertainties asso- but also soluble ion concentrations at the surface. Ice ciated with atmosphericand surfacesnow processesoc- lensingis unlikely to have occurredsignificantly during curringat the ice coresites involved [Dibb et al., 1992]. the last, generallycolder, three centuries,which the core The carbon input function used in the global carbon spans.However, hard (highdensity) timing [Alfordand cyclemodel is a flux, but it wasgenerated using (as a Keeler, 1970; Labelk, 1974; G. Holdsworth, unpublished guide)published fluxes derived from tree core 5•aC data data, 1986, 1988]occurs infrequently but is one possi- together with form functions derived from the ice core ble causefor sporadicanomalies seen in ~9% of part of data. the Mount Logan CO2 concentrationdata obtained in 1984 [Claud, 1985]. These data will not be presented Data and Results here becauseof the generaldifficulty in interpretingall the anomalies that seem to be present in those data. We first present data that are strongly associated It has been suggestedthat acid-basereactions at grain with knownagricultural activities which occurred (pri- 23,320 HOLDSWORTH ET AL' BIOMASS BURNING HISTORY IN ICE CORE DATA marily in North America) duringthe latter half of the Black Carbon (Soot) Data 19thcentury. Several ions (including NH4 +) associated It is reasonable to associatethe presenceof elemen- with biomass burn compoundshave been identified in tal carbon(and other carbonaceousmaterials) in ice Greenlandcores [Legrand et at., 1992; Whittowet at., cores with mainly cornbusted, or sometimes incom- •4]. pletely combusted,terrestrial biomassor fossil hydro- carbons. The method used for analyzing carbon in the Ammonium Data ice coresis basedon measuringthe light transmittance through a particle-trapping quartz filter before and af- Figure1 showsthe NH4+ ion concentrationsin the ter combustionat 700øC[Chytek et at., 1992]. All car- GISP2 and 20D cores. Enhanced levels occur between bonaceousmaterial is thus convertedto gaseousform ~1850 and ~1920, with a prominent minima occurring and expelled. There is thus no discrimination made at AD 1875+5. These are the highest sustainedlevels between graphitic carbon and partly, or even wholly, reached in the complete 20D core as well as the GISP2 uncombustedsubstances (such as lignins) on the fil- core from AD 1650 to present. At Mount Logan, there ter. Becausethe origins of all such atmospheric car- are spikescorresponding to the Greenlandspikes [Whit- bonaceousmaterials during the PIAGREV were prob- tow et al., 1994],but they are of the sameorder of mag- ably closelyrelated, the data presentedare assumedto nitude as other spikes in the core. It is possible that broadly representthat event in a qualitative, integrated the Mount Logan core site is regularly recordinga rela- way. Thus the referenceto "graphitic carbon," black tively high level of forest fire chemistryfrom naturally carbon(or soot)is usedhere in a rather loosesense. In occurringwildfires in the Yukon and Alaska (and also Figure 2, we present the results of the carbon analyses probablySiberia) and that theNH3 (NH• +) signalfrom carried out on all remaining but discontinuoussamples more southern(PIAGREV) sourceswould have atten- of Mount Logan core at Dalhousie University. Three uated to low levels after air associated with those emis- overlappingcores (A, C, and D) wereused. CoreD has sions completed a circuit around the northern hemi- been dated accurately[Hotdswor•h et at., 1992]. The sphere(usual transit time of the order of ~10 days). other, shorter, coresare keyed to it by cross-correlation The residence time of ammonia in the troposphere is using isotopic and chemicaltime series. Core A is 100 probably of the order of a few days. On the other hand, m south of the site where the contiguouscores C and the Greenland Ice Sheet is very well situated for record- D were taken on the NW Col of Mount Logan. We do ing direct anthropogenicemissions from central North not yet have high-resolutiondata for Greenland within America[Mayewski e! aL, 1990].The transittime here the same time interval, but one should expect that car- is short and probably of the order of the residencetime bon concentrations would show similar characteristics of gaseousammonia in the lower troposphere. to the form shown in Figure 2, becausethis nonreac-

I I I I I I I , • I , • I I [ I I I I [ I I I I

3o 3o

•., 25

20 - 20

-!-.•t. 15 [ 15 lO 10

5 5

I I , I I i I , , I ' ' i I I I , i I I I I , , 1750 1800 1850 1900 1950 2000 Year AD

Figure 1. Plotof NH4+ ionconcentrations fortwo Greenland ice cores: 20D in southernGreen- land (thick curvebased on filteredsubannual data) and GreenlandIce SheetProject 2 (GSP) centralsummit (thin curvebased on filteredbiannual data). The dashedline is the meanvalue curve. Enhancedvalues from ~1850 to 1910 are indicative of North American biomassburning during the Pioneer Agriculture Revolution. The existenceof the two prominent peaksis assumed to be mainly related to biomassburning history and is usedin estimatingthe shapeof the source functionin the carbon-cycle model (see Fig. 5). HOLDSWORTH ET AL.' BIOMASS BURNING HISTORY IN ICE CORE DATA 23,321

4 i i i i i i i ! i i i i i i i i i i i i 4

31-

,750' is'o , ,oo 15o ooo YEAR AD

Figure 2. Black carbon concentration(/•g C/L meltwater) in the Mount Logan ice cores [Holdsworthet al., 1992]. Mean valuesare shownby blocks;otherwise, single-year values are shownby verticalline (subannualvariations shown by adjacentshort and longlines). Gapsin record are due to limited availability of remaining core.

tive tracer appears to be showing the same variations and NH4+ time series[Whitlow et al., 1994]from the as the GreenlandNH4 + data, whichsuggests a hemi- Mount Logan core do not look very similar. Because spheric "soot" signal of the PIAGREV. In support of the GreenlandNH4 + signalis of similarform to several this, Junge[1963] found that insolubleaerosols have a other time series of PIAGREV indicators at different fairly uniform concentration for a few kilometers above locations(as describedin this paper),it is thoughtthat about 5 km altitude. The Mount Logan ice core site during the PIAGREV the North American contribution is located in this part of the troposphere. It thus ap- to atmospheric loading probably tended to dominate. pears that temporary mixing occurs at these altitudes Microparticle Data and that residencetimes of the solid particles(such as soot) will be muchlonger than that of gaseous(fire) Insoluble microparticle concentrations for the GISP2 derivatives, such as ammonia. core are presented in Figure 3. The particles are as- Referring to Figure 2, it must be realized, however, sumed to be mineral matter and soil as well as burned that after about 1910, carbon sourcesfrom increasing or unburned biomass material. These data further sup- fossil fuel combustion started to become important and port the influence of the PIAGREV on the remote at- that these early 20th century sourceswere not subjected mosphere in Greenland. Peaks in microparticle con- to emissioncontrols. Thus the maintenanceof high lev- centrations over the last 250 years occur around 1850 els of nonforest fire carbon after about 1910 is to be to 1870, 1885 to almost 1900, and again between 1910 expected. In this respect, the Mount Logan site would and 1920, time periods that closely coincide with max- seem to be a good receptor for soot transported long imum biomass burning indicated by increasesin am- distances in the northern hemisphere. The amounts of moniumlevels (Fig. 1). The increasesin particlemass carbon contributed from local fires is not known. Also, occurringduring thesesame general time periods(Fig. becausethe chemistriesand transports of ammonia and 3) reflectan increasein the abundanceof largerparti- soot are quite different, it is not surprising that the C cles. Higher particle concentrationsimply greater at-

150 500

•concentration _

...... mass - 4O0 •o 100 - - == ._o x 300 • • E -

o E 50 200•

lOO

175o 18oo 185o 19oo 195o 2000 Year A.D.

Figure 3. Plot of insoluble microparticle concentrationsand mass in the 0.65- to 13-/•m size range from biyearly samplesin the GISP2 ice core. Particle numbers were determined with an ElzoneTM particlecounter. Mass values were calculated assuming a particledensity of 2.65 g/cm3. 23,322 HOLDSWORTH ET AL' BIOMASS BURNING HISTORY IN ICE CORE DATA

mosphericturbidity during PIAGREV, probablyfrom wetake the offsetas 100 years as this produces the best the knownclear-cutting, biomass burning, and tilling of matchwith the Greenlandice sheetCO2 data,to be top soil at that time. Such actions would increase the presented. The gas/ice age offsetof the GISP2 data amount of material availablefor transport by wind to [Wahlenet al., 1991]is -,•220years. The ageerror bars Greenland. We next look at other sources of data that on the Greenlanddata [Staffelbachet al., 1991; Wahlen support the existence of the PIAGREV. et al., 1991] must be of the order of a decadein order Ice Core Carbon Dioxide Data for the data to connect with the atmosphericdata set [Keelinget al., 1989]. There may still be an unresolved Ice cores from glacier sites where the mean annual constantoffset error (_<10 yrs) betweenthe differentice temperature is less than about-25øC are generally core data sets that are presented later. thought to trap atmospherictrace gasesat about the Close-off at the tim-ice transition even in a sum- same mixing ratio as they had in the free atmosphere mer/fallseason (when smoke levels tend to be greatest) [Raynaudand Barnold,1985]. The oneproviso is that is quite rapid (comparedwith many high polar glacier there should be no melt-ice formation at the surface or sites),as the verticalflow rate of the ice at 65-m depth near surface;otherwise, excess CO2 will be sequestered is -,•0.25 m/yr. This may be comparedwith the ice into the snowpack.However, suitable sampling can usu- core sampling interval. Gas subsamplingwas made at ally be made to avoid suchice layers[Stauffer et al., 3- to 4-cm intervals(Table i footnote),so that one in 1985]. Anotherprocess that couldsequester CO2 in the situ ice subsamplewould have had its air bubbles sealed surfacesnowpack is harA •i•i• (•o,ivod from super- off in -,•1.7 months, which is well within the duration cooledwater dropletsoccurring in cloudsat low tem- of a fire season. Because the site is situated on a sad- peratures). This processmay occurmore readily at dle (col), where winds may reachjet streamspeeds of high-altitudesites such as on Mount Logan. 160 km/hr, the possibilityfor (summer)seasonal trace In Table i we presentthe resultsof the most reliable, gas "contamination" in some air bubbles is evident. In previouslyunpublished, analyses of the Mount Logan contrast, the winter air would be more mixed. At this ice coremade using the "Frenchtechnique" [Raynaud latitude the current seasonalCO2 variation is about 15 andBarnola, 1985]. For the newMount Logandata the ppmv [Conwayet al., 1988; Tanakaet al., 1988;Keeling erroron theestimated age [$chwander, 1989] of thebub- et al., 1989]. Transientpeaks in air trace gasconcentra- ble air is typically 4-6 years. Becausethe combinedice tions would tend to be damped by diffusionprocesses in coredata set mergeswith the "instrumental"data set, the tim, but we knowof no reliablemodel (that applies a degreeof confidencein the ice core timescalesis im- to the Mount LoganNW Col site) to accountfor this. plied. It shouldbe notedthat the Mount Logancarbon The data presentedin Table i is compatiblewith the dioxidedata presentedhere is basedon analysescarried Greenland ice core data. Significantly,the combinedice out in 1981and 1988at the Laboratoirede Glaciologie core CO2 data show temporal features that associate et Geephysiquede L'Environnement,Grenoble. It is it with the PIAGREV. This important point will be thus necessaryto discussthese particular data in much demonstratedin a later sectionof the paper. Thus any more detail here than the Greenland data which have attempted explanation of the high CO2 values based been published and documented elsewhere. solelyon in situ ice core chemicalreactions [Delmas, 1993] would have to provide a reasonwhy thesereac- Note on the Time Offset Between the Age of tions are concentrated in just those places. Because the Air in the Bubbles and the Age of the Ice the chemical data available for the Mount Logan core Surrounding Them do not seem to favor the processproposed by Delmas [1993],we do not attribute the anomalousCO2 val- The gas/iceage offset for the Mount Logancore was ues to in situ postdepositional processes. This would taken by Dibb et al. [1993]as 103 q- 5.5 years. Here specificallyrequire excess acid and/or excesscarbonate

Table 1. Mean Concentrationsof CO2 in Mount Logan "D" Ice Core Depth Interval, cm Gas Age (AD) a [CO•],• ppmv 9993-10053 1805q-7 275q-8 9920-9952 1819q-6 2854-7;286q-5 9590-9688 1839q-6 286q-8 9339-9366 1860q-6 301q-6c 8663-8696 1897q-5 308q-6;309q-7 aGas/iceage offset used here is 100 years. The absoluteage error barsfor the Mount Logan data may be larger than those calculated. bMeanof subsamples cutfrom « core;subsample size: 3 to4 cmvertical cuts. CfN__uuc •,,un,•,uu•I ...... v•ucl__ rcmuvcu.J _•l,•,y•e•& by J.-M Barnold. o I • data [Claud,1985] (seven points) is excludedbecause of higherror bars (>8 ppmv)and the existenceof severallarge spikesof unknownorigin. However,some of those data are consistentwith the data givenabove. Interpretion of thosedata shouldbe investigated. HOLDSWORTH ET AL.' BIOMASS BURNING HISTORY IN ICE CORE DATA 23,323 saltsoccurring in closeproximity. (Althoughthere is been plotted over the same time interval. To avoid an a higherthan averageCa ++ concentrationfrom about overconcentrationof lines in Figure 4, we have omitted 1760 to 1780, the acid concentrations are lower than similar data that are also available from other refere'nces average. Ice in this interval is contiguouswith bubble [e.g., Tans and Mook, 1980; Stuiver,1986; Stuiver and air a century youngerin age. In addition, the middle of Braziunas,1987]. The combinedseries typically show this last interval(where the gasage is ~1870) wasnot a prominent reversal of gradient beginning somewhat marked by particularly high valuesof the principal PI- before 1850. This increase in negative gradient corre- AGREV chemical "tracers," as our collected data tend spondsto a decreasein 5•3C.Such a 513Cstep, or ramp, to show). Furthermore,there is no quantitativesup- occurringsoon after --•1850,may be foundin many (free port for in situ chemical reactions for the CO2 data standing)tree core recordsand has been interpreted of Wahlenet al. [1991].Concentrations of Ca++ and [Freyer,1978; Wilson,1978; Peng et al., 1983; $tuiver, Mg++ in the GISP2 corefor the last millenniumdo not 1986] to signal a rapid increasein atmosphericCO2. positively correlate with high values of CO2; in fact, However, there are other environmental influencessuch almostthe reverse(negative) relationship is observed. as climatic factors [Peng et al., 1983; Leavitt, 1993]or biomass-drivenisotope fractionation chemistry [Francey andFarquhar, 1982] that e[ffect5•3C in wood,and some Presentation and Discussion of northernhemisphere trees studied [Leavitt, 1993] do not Supporting Data appearto carry a clear (•1850+) signalnear the start of the PIAGREV.

Tree Core Data Another possibilityfor explaining part of the overall largevariability in northernhemisphere tree core513C Because there are certain similarities among Fig- valuesduring the PIAGREV may be from different de- ures1, 2, and 3 (within the contextof the PIAGREV), greesof exposureof sampledtree standsto CO2 plumes we have searchedfor and compileddata (from various emanating from forest fires or other centers of massive publishedsources) which might also act as proxy for CO2 generation. This may be deducedfrom the plume biomassburning activity. In Figure 4, we show a se- transport modelingresults of Iacobelliset al. [1994], lectionof time seriesof 5•aCgradients (dSi/dt, where where CO2 concentrationsof several ppmv above back- 5i arespecific tree 5•aC data and t is time). Thesese- ground may occur over large, continental-scaleareas on riesare derivedfrom various5•aC data [Freyer,1978; a timescale of weeks. Wilson, 1978; Freyer and Belacy, 1983; Stuiver et al., Also, the 5•3C PIAGREV signalwas not identified 1984; Freyer, 1986] determinedfrom isotopic analysis in Tasmantan wood sampled more than a decade ago of northern hemispheretree ring samples. They have [Francey,1981]. However,these data might morenearly

0.01 ! I I I

SLOPE 5'

I ß © 000600 I , 74,0oo 0.0 -.•--'n• o• [• 0.0

.,' L___•/ t **•' ß

O.01 0.01

0.02 0.02 time I 1750 1800 1850 1900 1950 2(: •oo YEAR AD

Figure 4. Processed513C data from various sources. Tree cellulose: curve 1 (solidline, 1760- 1940), Wilson[1978]; curve 2 (opentriangle), Freyer[1978]; curve 3 (solid circle), Freyer and Belacy[1983]; curve 4 (opencircle), Stuiver et al. [1984];curves 5 and 6 (opensquare, dashed), Freyer[1986]. Ice coreair bubbles:curve 7 (solidtriangle), Friedli et al. [1986].Symbols (points) tracehistograms of 20-yearaveraged time derivativesof deviations(5i) of the 5•3C valuesfrom a designatedmean [Freyer, 1986; Stuiver, 1986]. In somecases, a histogram"block" is missing because of large scatter of data in a 20-year interval. Spans A, B, and C define the three main stagesof the PioneerAgriculture Revolution (PIAGREV): A, explosion;B, postexplosion minimum;C, final maximum(see Fig. 5). 23,324 HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA

reflect complexmultivariate atmosphericand local tree Direct Air CO2 Data chemistryvariations [Francey and Farquhar,1982] as A number of different measurements of air CO2 con- they are not similar to other coastal southern hemispheretree core5•3C data [$tuiveret al., 1984]from tent were made in the late 19th century. It is not possible here to present the complete data sets and dis- specially selected trees or to Antarctic ice core data cussthem. From and Keeling[1986] have discussed and [Friedli et al., 1986],which clearlyshow the declinein apparently validated one particular data set obtained 5•3C(consistent with an increasein thebiospheric com- from coastal France by J. Reiset in 1872-1880. The re- ponentof CO2) around 1850. The error on the dating mainder have been under considerablesuspicion. One of of the ice core air is given as 4-11 years, and there was thosedata sets(from the MontsourisObservatory) was no stated correction made for possiblegravitational and consideredreliable by $tanhill [1982],who further de- extraction isotopic fractionation effects. However, these fendedtheir integrity [$tanhill, 1983,1984] against mul- are the only such data available at this time. Despite tiple criticisms[Waterman, 1983; Wigley,1983; $iegen- these limitations of the data, this situation should not thaler, 1984]. The Montsourisdata (to be presented significantlyaffect grossrelative changesin the later) showa prominentpulse in CO2 centeredon about data and hence the time gradients which are shown AD 1900. Whatever is the correct judgement on these (dataset 7), togetherwith the treecellulose 5•aC data, data, we note that the relative variations within this in Figure 4. data set are qualitatively consistentwith some of the Considerabledoubt still existsregarding the suitabil- ice core data and the tree core data. The absolute val- ity of many tree 5•aC seriesas reliableatmospheric in- ues, however, do not appear to be those of a fully mixed dicators[Francey, 1981; Leavitt, 1993; Peng et al., 1983] atmosphere.(This conditionalso applies to the ice core emphasizingthe fact that propertree selection(as with and the tree ring data). ice coresite selection)is very important [$tuiveret al., 1984]. Thus in the light of theseconsiderations we ac- knowledgethat the (~1850) tree 5•aC anomalycould Synthesis of Evidence for Deforestation, correspond,in part, to a changein the isotopiccompo- sition of atmosphericCO2 as a result of biomassburning BiomassBurning, and Soil Cultivation and the or to a changein someclimate variables (temperature, Probability that IndependentData SetsIndicate moisturecontent) or to changesin forestregime. This the Same Event anomalyis hardlyaffected by absoluteerrors in 5•3Cde- terminations,which are suggestedby a recentinterlabo- ratory calibrationexercise (C. S. Wongand M. Stuiver, In summary, there are six lines of evidence suggest- personalcommunications, 1993). Regionaldifferences ing intensified deforestation,cultivation, and biomass in atmospheric variables as well as possibledifferences burning, particularly in North America, during the late in reliability of data have made it difficult to identify 19th and early 20th centuries.These are (1) ice core a consistent(1850+) signalin either hemisphere,but NH4+ data,(2) icecore carbon data, (3)ice coremi- the majority of available time series contain it, partic- croparticleconcentrations, (4) ice coreC02 data (two ularly in the northern hemisphere,as one would expect independentsets), (5) treecellulose 5•aC data (multiple [e.g., Freyer, 1978; Wilson, 1978; Freyer, 1979; Freyer sets),and (6) direct air CO2 data. The peak biomass and Belacy, 1983; $tuiver et al., 1984; Peng and Freyer, burningactivity seemsto have occurredwithin the in- 1986; $tuiver and Braziunas,1987]. tervalAD 1889-1910(see also Wong[1978] and Penget Some of the inconsistenciesamong different tree core al. [1983]). Althoughnot entirelyindependent of the data sets may be due to the particular geometriesof above evidence,another form of support is providedby smoke(and gas) plume trajectoriesthat evolvefrom the atmosphericmodeling results, which provide a ver- the fire source. This is suggestedby the CO2 trans- ification of the CO2 data. port modelingby Iacobelliset al. [1994]. Sometrees The main signalfor biomassburning comesfrom the have a very short wood-forminginterval. For Bristle- NH4+ ion concentrationdata for the Greenlandsites. conepine it is only about 6 weeks[Wilson, 1978]. The The precursorgas for ammonium compoundsin snow trees that were situated in a preferred plume path would (e.g., [NH412SO4)isammonia gas, which is a con- show more depressedvalues of 5•aC than trees well stituentof forestfire smoke[Laursen et al., 1992;Levine outside the plume path. Furthermore, in casesof ex- et al., 1992]. NH3 may also be emitted from soilsaf- treme contamination by fire emissions,after cessation ter forestburning. Legrandet al. [1992]have already of those emissionsor changesin the burn patterns, the pointed out the strong likelihoodthat ammoniumspikes 5•aC values in the wood should recover toward a value in pre-AD 1700, central Greenlandsnow originated from more nearly representative of the value that pertains North Americanforest fires. Junge[1963] reported that to a more fully mixed atmosphere. Some time series atmosphericNH3 wasmeasured in the secondhalf of the seem to show post-PIAGREV features that might be 19th century at variouslocations in westernEurope, in- attributed to this effect[e.g., Freyer, 1986; Leavitt and cluding the 2800-m-highsite on Pic du Midi in France. Long,1992], although climatic effects in this time inter- The values were "rather high comparedwith more re- val maybe just as importantin affectingthe tree cent data" [Junge,1963, p. 87]. This informationmay values. be usedas supportingevidence for the PIAGREV. Re- HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA 23,325

cent forest fire plume transport simulations(S. F. Ia- bined data sets that indicate the PIAGREV as being cobellis,unpublished, 1994) indicatethat North Amer- the common denominator. ican emissionscan impact both Greenland and Europe. Probability Estimate for Combined Data Depending on the prevailing climatology,it is possible that anomaliesof severalppmv above far-field mixed air Given that each data set described above has an indi- existed on timescales of weeks. The levels of biomass vidual probability of being related to a common event, burningmay alsohave beenhigh in westernEurope late the system of data sets provides a means of estimat- last century. ing the combined probability that the PIAGREV was The carbonconcentration data for Mount Loganwas a "fundamental" event. If, for simplicity, we initially reported as soot, but we caution that it may include take only three of the eventsand assignindividual prob- non-combusted carbonaceous matter as well. Before abilities[P(Ei); (i = 1,2,3)] that they are all related about 1900 the ice core carbonaceous material is as- to a single main event, E, we can estimate this com- sumed to be mainly from forest wood and other biomass bined probability P(E) usingthe generaladdition law destruction. The only data available for this study are of probabilities[Chatfield, 1983]. The combinedprob- from Mount Logan, Yukon. The data are sparse, but ability that the three "subevents" are related to the they show a variation with time which has similarities parent event E is thus given by to the NH4+ datafrom Greenland up to the beginning

of this century. This correlation(except for the most 3 recentpart of the data set) suggeststhat the Mount Lo- gan site was a good receptor and indicator for mainly - - + i=1 North American carbon emissions.With more complex emissionsfor the 20th century the conditionof a "good P(E•E2) - P(E2Es) - P(E•Es) + P(E•E2Es). indicator" may no longer apply. Additionally, the almost ubiquitous burning of fossil fuels with inefficient Taking a relatively low probabilityof P(Ei) - 0.50 combustionbefore 1950 must be responsiblefor the high for all three events, we calculate a combinedprobability carbonlevels during the first part of this century(Fig. of P(E): 0.875that the threeevents are related to the 2). parent event. Next, to accommodate the six events, we The ice core microparticle concentrations for the repeat this procedurefor two combined"events" (each GISP2 core showsignificantly higher levelsand greater composedof three subevents)with a probabilityof 0.875 variance over the last half of the 19th century and the of being related to a commonevent (again E). The first two decades of the 20th century compared with required equation is the levels a century before and the seven decades after. The link with the PIAGREV in time suggeststhat P'(E) = P(Ea + Eb) : P(Ea) + P(Eb) - P(EaE•) the particulates were predominantly produced as a result of agricultural activities, such as tilling of newly where E• and Eb are now each taken as equal to the exposed soils. The particulate inventory must also in- previousP(E). clude various particles derived from burned or incom- We thusget a newprobability, P•(E) = 0.984,which pletely burned biomass material, such as is recorded in is close to the probability level acceptable in most sta- the data of Figure 2. tistical studies. Even using a less optimistic value of The northern hemisphere ice core CO2 data for the P(Ei) = 0.40 (40% probability),the equationsyield a late 19th centuryhave represented (with respectto the valueof P•(E) : 0.954,which is still within acceptable Antarctic data) an anomalythat was not easilyinter- statistical limits. If we take a more confident value of preted until now. Seen in the context of the PIAGREV, P(Ei) = 0.60 (and this would not be unrealisticfor these data appear to take on a particular significance, the combined ice core CO2 data, in which each data set especiallyafter globalcarbon cycle modeling (to be de- would, of course, have a much lower individual proba- scribed)shows that elevatednorthern hemispheric car- bility), then we find that P•(E) = 0.996,which defines bon dioxidelevels (over Antarctic values), in response a very high statistical probability. to concentratednorthern midlatitude biomassburning, could have occurred. In addition, as we have seen, some of the controversial French direct air CO2 data from late Atmospheric Carbon Input Fluxes From last century can be claimed to qualitatively support the Biomass Sources ice core data. Data from tree cores, usually in the form of time se- In 1860, fossil fuel burning was taking place, but at riesof 513C,have been confusing and oftencontradic- a very low level compared to extant biomass burning tory. However, it is evident that in many free-standing [e.g., Keeling, 1973; Marland and Rotty, 1984; Mar- trees there is a "signal"that correspondsin time to the land, 1989; Marland et al., 1989]. Furthermore, car- PIAGREV. Wherethe 51sCstep occurs, it is usuallyof bon dioxide input to the atmosphere from this source the right order of magnitude to be explained predomi- was to remain very low until the start of the 20th cen- nantly by an increasein atmosphericCO2 [e.g.,$tuiver tury. Figure 5 shows various biomass carbon source et al., 1984]. Next, we look at the strengthof the com- functions(input into the atmosphere)which are possi- 23,326 HOLDSWORTH ET AL.' BIOMASS BURNING HISTORY IN ICE CORE DATA

' • • I F ATMOSPHERIC CARBON FLUX. • z/'""• '"-•'--'"--"•'"•'-.,',--'"-I ,y ....o<,,^•,,v>/, .o.... o...

\ 0 \ \ kS ii2 I I I I 17•0 1800 1850 1900 1950 2000 YEAR AD

Figure 5. Estimatedcarbon input (Gt[C]/yr) to the atmospherefrom biomass burning as derived from data shownin Figure4 by differentauthors: curve 1, Freyer and Belacy[1983]; curves 2 and 3, Peng and Freyer [1986]. For clarity of presentation,not all publishedcurves for this function are shown here. Curve 4 definesthe estimated input function used in the global carbon cycle model to best producc an output showingthe main features of the available CO2 data. This function is basedon earlier, near-maximumestimates [Peng and Freyer, 1986; $tuiver, 1986] and modified accordingto information shownin Figures 1 and 2. Sourceis applied at northern hemispheremidlatitudes until 1950;thereafter it becomesa sink (dashedcurve S) T, tropical sourcefunction. Curve F is the fossilfuel curve [Stuiver,1986]. PIAGREV, PioneerAgriculture Revolution. ble in the light of several independent carbon budget show a fairly smooth increase, but there are weak in- studies[Higuchi, 1983a; Peng and Freyer, 1986; $tu- creasesevident at ~1850 and 1900 that now have special iver, 1986; Siegenthaler and Oeschger,1987; Tans et significancebecause of the timing. An alternative ex- al., 1990],and whichshow the probableminimum and planation, that these variations merely representmea- maximum limits of carbon input from natural and an- surement errors in the data, is hardly supported by the thropogenicallytriggered sources. For a more complete observation that the "excess" values occur at the two coverageof the large differencesin carbon input func- significanttimes of the PIAGREV. As we mentioned tions, see Peng and Freyer [1986]. The twin-peaked earlier,there is alsothe hint of two stepsin the 513C biomass source function that we constructed for our data of Freidli et al. [1994] at the same two times. model study (to be describedin a later section)was These resultsimply that (1) somemixed PIAGREV structuredto reflectthe variationsin the NH4+, car- CO2 may have reachedhigh southernlatitudes but that bon and particle concentrations(up to ~1910) shown (2) sinksbetween middle northernlatitudes and the in Figures 1, 2 and 3. The initial, trial function was Antarctic ice core sites largely absorbedthe transient set with a low amplitude at the level of the function excessCO2. Next, we present the combinedevidence derivedby Peng and Freyer[1986]. The amplitudewas for elevatedCO2 levels(over Antarctic values) at spe- successivelyincreased until the model output produced cific locations in the northern hemisphere during the CO2 levelsthat approximately fitted the combinedCO2 PIAGREV. data (to be presentedin the next section)as closelyas possible.It may be mentionedhere that the 513Cdata from the Antarctic ice core [Friedli et al., 1986]shows Discussion of CO2 Data two possibledrops, one about mid-19th century, the other (significantlylarger) near 1900 (althoughthese In Figure 6, we showice core CO2 data (symbols data are subject to absolute errors mentioned earlier, with estimatederror bars) obtained from the GISP2site the relative errors should hardly be responsiblefor the [ Wahlenet al., 1991],the Cr&teice core [$taffelbachet most prominentvariations seen in thosedata). Thus al., 1991],and the 1980Mount Logan"D" core.The lat- these observationsare consistentwith our twin-peaked ter, previously unpublished data are taken from Table atmospheric carbon input function. For this reason we 1. It is thought that the time error betweendata setsis have not attempted to interpret the variations in chem- probably within about 10 years. Becauseof subsample ical speciesat the ice core sites in terms of major atmo- averaging,the original atmospherictrace gasshort-term sphericcirculation changes,for which reliable evidence (transient)variations are further reduced. However, the throughoutthat time span is not presentlyavailable to persistenceof peaksin the data of Figure 6 seemsto im- us. We ...... examine, in qllClJitll•-'-:' 1 "-bill2 •l,V•ll•l, "'' Dle (lal•8•' ' for ply-"•'ma• signlnca•' '" '' leVelS.... '- of enrlcnmenl;'' ' occurred over past levelsof carbon dioxide in the atmosphere. decadal timescales,which is consistentwith the histor- In contrast to the northern hemisphereCO2 data, the ical evidence as well as with the 19th century French Antarctic CO2 data [Friedli et al., 1986;Enting, 1992] air CO• data that will be discussedlater. Curve a in HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA 23,327

I I I i I i I i I i i i i I I I I I I I i I i i I CARBONDIOXIDE CONCENTRATION DATA ** 1350 [+MT. LOOAN(YUKON, ICECORE "](b)

o •o

• I I AT MT LOGAN• •'•/vV •' _.• ANTARCTICICEc0 .• •'[ , •• •to . 500

• (a)• • 6 I PIONEER I 1750 1800 1850 1900 1950 2000 YEAR AD

Figure 6. Two centuriesof atmosphericcarbon dioxide variations as suggestedby variousdata sourcesand by modelresults. The double-dashedcurve a is a splinefit to the Antarctic (Siple Station)ice core data, here shifted 2 yearsto the rightso that it lies(observationally) ~1 ppmv belowthe Mauna Loa data [Keelinget al., 1989]starting in 1958 (curveb) (rangeindicated by solidtriangle and invertedsolid triangle). Someindividual Antarctic data points are shown eitherby smallhorizontal dashes (averages) or by smallopen squares [see Neffel et al., 1994]. (In addition,we haverecently added the newdata of Etheridacet al. [1996].These data are shownby the solidcircles connected by thin lines.) The two verticalarrows shown below curve a indicate the location of possibleperturbations that may be related to the PIAGREV. The symbols(open triangle, open inverted triangle, open circle, asterisk) define annual ranges at severalcoastal stations [Conway et al., 1988;Tanaka et al., 1988;Keeling et al., 1989]near 60øN, adjustedto the altitude of Mount Logan. The cappedarrows above and belowthe earliestMauna Loa data define the estimated annual range at 60.5øN at the altitude of the Mount Logan site (similarfor Greenland).CO2 data derived from northern hemisphere ice cores are shown as single points:(solid square) Mount Logan; (open square) GISP2, Greenland[Wahlen et al., 1991];(solid diamond)Cr•te, Greenland[Staffelbach et al., 1991]. Error barsare estimatedabsolute values. Curve c (joinedopen circles from 1890to 1910)shows the measuredvariation of CO2 in the air at the MontsourisObservatory, near Paris, France [Stanhill,1982]. The coastalFrench data of J. A. Reiset [From and Keeling,1986] is shownin the block marked R. Curve d: Model simulationfor high-latitudenorthern hemisphere sites (60o-80G) basedon curve4 and curve F input functionsshown in Figure 5 and employingexperimentally constructed sink functionsto duplicate the Antarctic curve for high southern latitudes.

Figure6 definesthe Antarcticice coredata set [Friedli mbar) is indicatedby the vertical arrowsat the start et al., 1986]spline fitted by Siegenthalerand Oeschger of the instrumental data. Northern hemisphere ice core [1987]and here shifted 2 yearsto the rightto producean CO2 data are shownas singlepoints (see captionfor (observational)~1 ppmv underfitwith the Mauna Loa explanation).Error barsare estimatedabsolute values. data [Keelinget al., 1989](shown as curveb). Sucha For each set, the relative errors are less than the er- shift is still within the age error bar given for these data. ror barsshown. Also shownin Figure6 (curvec) are Some of the original Antarctic ice core data values are the direct(chemically determined) air CO2data (1890- shown at two significant times during the PIAGREV, 1910,annual averages) obtained at the MontsourisOb- marked by the vertical arrows below curve a. servatory,near Paris, France. The absolutevalues of Direct air measurements(data points) shownafter the air sampledata (as with someof the ice coredata) 1958, and defining seasonal ranges, are taken from do not appear to be those of a fully mixed atmosphere Bischoff[1981], Conwayet al. [1988], Tanaka et al. (see Discussionfor resultsof forest fire plume trans- [1988],Keeling et al. [1989],and from air samplestaken port modelingstudies). Anotherset of chemicaldata on Mount Loganin July 1986(top solidcircle). (Anal- from the north coast of France spanning 1872-1880 is yses of these samples are courtesy of C.S. Wong, In- alsoincluded in Figure6 (data blockR) becauseit is stituteof OceanSciences, Sidney, B.C., Canada.) The consideredreliable [From and Keeling,1986] within the expectedannual range at Mount Logan (60.5øN, 500 plotted error range. (These data were obtainedwhen 23,328 HOLDSWORTH ET AL.' BIOMASS BURNING HISTORY IN ICE CORE DATA

there was an apparent recessionin the PIAGREV, but two-dimensional, zonally averaged, and vertically lay- becauseof the location, the samplesalso have a much ered; within this atmosphere,the air massestransport higher probability of representinga fully mixed atmo- trace gasesby advectionand diffusion;(2) an oceanic sphere). model, with three explicitly resolvedbut independent In addition,$tuiver et al. [1984]provide a setof 5•aC water masses,all coupledto the atmosphere;(3) a ter- data and an atmosphericCO2 reconstructiontogether restrial biosphere,which is treated as a lower boundary with a biospheric carbon flux curve which indicates a forcingfunction to the model atmosphere. pulse injection of CO2 into the atmosphere for about a decade through AD 1900. The CO2 reconstruction Model Atmosphere is qualitatively realistic from ~1700 to early 20th cen- The two-dimensionalatmospheric model architecture tury. Thereafter, the assumptionsmade in the $tuiver is identical to the one used by Pearman and Hyson e! al. model appear to break down. Fortuitously, the [1980] except that the presentmodel is composedof time of collectionof the most reliableMontsouris (air) 162 boxes(18 vertical columns,each column spanning CO2 data occurredin this interval. The correspondence a 10o latitude band and composedof nine layers from between these data and the other data presentedear- the 1000-mbar surface to a 10-mbar upper level. The lier suggestsstrong coherence between the data setsand massof the modelatmosphere is 5.26 x 102• g. The support for the secondpulse in the PIAGREV signal. advectivefluxes, Fi (i - y, z wherey is meridionaldi- Lastly, becausetrace gas enrichmentduring the PIA- rection,z is verticaldirection), of CO• (in unitsof con- GREV is seenin othertra. ce ga.s data [Dibbet al., 1993], centration/unittime) are givenby those data shouldbe interpreted as supportingthe PI- AGREV. In particular, there are stepsin methane and FyAr>V - (pVOA•I)/M, nitrous oxide in the Mount Logan and 20D cores in the interval 1850-1900. Both thesegases are associated with biomassburning directly or with postburnsoil gas F•ADV - (pWOAv)/M, emissions. where p is air density, V and W are the monthly aver- Next, we present a strategy involving a coupledatmos- aged meridional and vertical wind speeds,respectively, phere-oceanmodel, which was used in an effort to sim- C is the CO2 concentration, M is the mass of the at- ulate the atmospheric CO• variations over the last two mospheric box under consideration,and AH and Av centuries. This exerciserequires a knowledgeof the biomass "source" function and a certain freedom to are the horizontal and vertical areas of the box, respec- construct carbon "sink" functions. An experimental, tively, through which the gasflux takesplace. Diffusion empirically determined, carbon input function was es- of the input CO• is governedby the followingdiffusion tablished after a series of iterations were carried out equations: to produce a simulated CO• curve that comparedwith the available data shown in Figure 6. Each iteration required a modification to the sink functions. While FyDIFF - p(Kyy OC/g9y + Kyz OC/g9z) A•t/M, this procedurerequires a significantdegree of operator control(which involves manipulation of the model),the important point is that the atmosphericmixing ratios FzDIFF - p(Kzy OC/g9y + KzzOC/g9z) Av /M, for CO• in the high northern hemispherelatitude bands (wherethe datawere obtained) is still dominatedby the carboninput function rather than by the "inferred"sink whereKij are diffusioncoefficients. The eddy diffu- functions. At the very least these sink functions could sioncoefficients (i = j) are alwayspositive. The coeffi- be used to guide experimental investigationof carbon cientsdefined by i • j are thoseintroduced by Reedand sinks, which are still not properly understood. German[1965] to accountfor the countergradientflux in the stratosphere. The seasonally stratified meridional winds were modified after Hyson e! al. [1980] Global Carbon Cycle Model by Higuchi [1983a]. Vertical winds in a columnwere calculated using the mass conservationequation. The The time-dependent global carbon cycle model diffusion coefficientsused were derived from krypton 85 (GCCM) usedin this studyis a modifiedversion of the data [Morimoto,1994]. modeldeveloped by Higuchi[1983a, b]. It describesthe The rate of changeof CO• concentrationin an atmo- routing of CO2 injected into the atmosphere. As the sphericbox is given by overall structure of the model is essentiallythe same, only a short descriptionis given here. The model is formulated to include, as much as is reasonablypossible, the essentialand fundamental processesof the global carbon cyciethat are necessaryto reproducesome of the 'The time step At -- i day for the atmosphericmodel. globally consistent and basic features of the observed The "box" model atmosphere is in contact with and CO2 time series. The model structure is made up of coupled to the "box" model ocean, which will now be three components:(1) an atmosphericmodel, which is described. HOLDSWORTH ET AL.' BIOMASS BURNING HISTORY IN ICE CORE DATA 23,329

Model Oceans shown.The resultantlatitudinal profile of Pm (partial pressureof carbondioxide in the mixedlayer) is guided The oceansconstitute one of the two major internal by the resultsgiven by Broeckeret aL [1980].The fol- reservoirs(the other beingthe land biosphereand, to a lowingvalues for the total borateconcentration (• B), lesserbut unknowndegree, the oceanbiosphere) which carbonate-boratealkalinity constant(A), and chlorin- have any significantfeedback to the anthropogenicCO2 ity (Cl), are usedfor all the oceansin calculating in the atmosphere.The amount and the rate of absorp- values'• B - 0.40 per mil, A - 2.43 per mil, and C! tion of excessatmospheric CO2 by the oceansdepend - 19.24 per mil. on physicaland chemicalprocesses operating acrossthe The initial total carboncontent in the surfacelayers air-sea interface and at depth within the oceans. is specifiedas 830x 10•5 g C, whichis about1.4 times The oceanic reservoir is represented by a model com- that of the initial atmosphericcarbon mass of 610 x 1015 posed of three basins: the Atlantic, the Pacific, and the g C (correspondingto a mixingratio for CO2 of about Indian. Each basin is treated as independent and is 280 ppm). Here we are only concernedwith carbonin representedby a zonally averagedtwo-dimensional ad- the form of carbon dioxide. vective multibox model with a highly simplified merid- In the present model, the role of the marine biomass ional circulation pattern, which, however, resemblesthe is not explicitly accountedfor becauseof the large un- grosswater movementobserved in nature [Broeckeret certainties in the processesthat are involved. al., 1980]. Carbon dioxideis transportedin the direc- In the couplingto the atmospherethe net flux, zlF, tion of net water flow within a circulation cell. In each of acrossthe air-seainterface for wind speedsw > 3 m/s the major oceansthe surfacelayer (0- to 100-mdepth), [seeTans e! aL, 1990]is givenby is dividedinto 11 zonally averagedboxes. (A single box representsthe equatorial surface layer from 20øN zxF - k(a - , to 20øS except for the Indian Ocean, where it is 0ø to 20øS).Five more boxesare usedto representthe in- where Pa and P,• are the pCO2 of the atmosphereand termediate layer water column down to 1000 m. Thus the mixed layer of the ocean, respectively.The param- there are a total of 16 boxes. Outside the equatorial eter k is given(again following Tans et al. [1990])by regions box, the intermediate layer is not modeled ex- the equation plicitly. The deepocean (>1000-m depth) is alsonot treated explicitly. Since the turnover time of the deep k - 0.016(w- 3) , water is of the order 500-1000 years, it is treated as an infinitely large sink for recent century-scaleemissions of where w is the horizontal wind speed at 10 m above anthropogenic CO2. the sea surface, in meters per second. Under steady Water fluxes for each surface box are determined from state conditions,zlFj of box j is made equal to the specified current speeds and from the vertical area of amount of carbon made deficient (or surplus) by the each box. Once the surfacemeridional fluxes are speci- advective current in the surface layer. In addition, 5 11 fied, all other fluxesare determinedusing the equations •-2•j=lAFj d- • AFj -- AF6 for eachof theocean for conservationof mass. To produce downwellingand basins,where th• index i representsthe mostnorthern prevent upwelling of water into the extratropical surface (60ø-70øN)box, and the index 11 representsthe most layer, the poleward flowing meridional surfacecurrent in southernbox (60ø-70øS).The equatorialand largest box eachof the model oceansis suitablyspecified [Higuchi, is represented by the index 6. 1983a,b]. As an exampleof one of the parameterscal- The model of Nit and Lewis [1975] is employedin culated from the model, the deep water flux upwelling transporting carbon dioxide from the extratropical sur- in the equatorialregion is 5.2 x 106m3/s. face layer into the intermediate layer and finally into The initial total carbondioxide concentration •C0 the equatorial column. The time step used in the ocean is assignedto the surface layer of each ocean so that it model is i year. The residence time in the intermedi- increasespoleward in accordancewith a decreasingsea ate layer is defined here as the time which CO2 spends surface temperature schemeoutlined in Table 2. In the in transit between the surface layer and the equatorial sametable the correspondingvalues of •C0 are also column. The residence times of excess CO2 in the in-

Table 2. Temperature and Carbon Data Used in Model Oceans

Latitude Zone, Sea Surface Total Initial Carbon Concentration deg Temperature, 0C •Co, mol/cm3

0- 20 28 2.00 20- 30 23 2.02 30 - 40 18 2.04 40- 50 12 2.07 50- 60 7 2.14 60 - 70 3 2.17 23,330 HOLDSWORTH ET AL' BIOMASS BURNING HISTORY IN ICE CORE DATA termediatelayer rangefrom 30 yearsfor downwellingin input to the GCCM over a prescribednorthern hemi- low latitudes to 100 years for downwellingin high lati- spheremidlatitude band (40ø to 60øN), and the out- tudes. Carbon advected into the deep water is treated put (atmosphericCO•. concentrationsat high northern in this modelingexercise as carbonlost from the cycle. hemispherelatitudes) was examined. It wasnecessary, The ocean model is calibrated with observed vertical at this stage to modify the sink functionsso that the profilesof prebombx4C data [Bien et al., 1960;Broecker Antarctic ice core CO•. data could be matched. These et al., 1960] and comparedwith resultsobtained by data are assumed to closely represent the concentra- Bjorkstrom[1979]. The atmosphericproduction rate tions of CO•. in the remote atmosphereof the southern of x4C atomsby cosmicrays is assumedto be constant hemisphere.By progressivelyincreasing the amplitude at 2 atoms/cm•'/s.Given the observedsurface values of the initial carbon input function, it was possible to of xqC/X•'C,the oceanmodel is ableto simulateAxqC observethe correspondingincrease in atmosphericCO•. valuesand henceC TM ages, relative to the surfacelayer, in the northern hemisphere. Each step required an ad- of over 200 yearsfor the deepwater upwellinginto the justment of the sink function structure in order to du- Atlantic equatorialcolumn, and over 1000years for the plicate the high-latitude southern hemisphereice core deepwater upwellinginto the Pacificequatorial column. data. Details of this function and a sensitivity analysis of the model will be describedin a paper in preparation Model Carbon Budget and Comparison Between by K. Higuchi. This studyof the time variationof atmo- This and Other GCCMs sphericCO•. over the last two centuriesis quite limited The overall carbonbudget in the model is balanced by deficienciesin knowledgeabout the physicsand dis- for eachtime step. For example,in the 1985 simulation tributionof carbonsinks [Tans et at., 1990;Quay et al., the followingidealized input (in Gt/yr) was applied' 1992]. Here, as in any otherlong-term (century-scale) fossilfuel input - 5.5; tropicalbiomass input - 2.0 (• carbon cyclestudies, estimated empirical sink functions = +7.5 Gt/yr). The followingsinks were calculated: at- have to b e used. mosphericreservoir - 3.0; oceansink - 2.0. To balance The biosphericcarbon function (curve 4) shownin the budget,a northernhemisphere midlatitude (land) Figure 5 was the last of a seriesof trial functions. It sink- 2.5 wasintroduced (•-•. - -7.5 Gt/yr). This pro- still producesless total cumulativecarbon flux into the duces a "model" northern hemisphere-southernhemi- atmospherethan curvei (in Fig. 5), whichset the up- spherelatitudinal gradient of about5 ppm,whereas the per limit of our trials. Even so, curve 4 producesa 5-10 observedvalue is about 4 ppm. Comparingthese results ppmv increaseabove the Antarctic curvebetween about with Tables 1.1 and 1.3 in IntergovernmentalPanel on 1850 and 1920, as can be seenin Figure 6. The output ClimateChange [1994], it may be seenthat the GCCM curve d is consideredto possesssimilarities to the avail- describedhere is broadly consistentwith other carbon able data in a qualitative sense. Further manipulation cyclemodels. The main difficultyfor all thesemodels is of the carboninput function and the sink functionwas to account for all the carbon sinks. It has been recog- not pursuedas this wasnot consideredto be in the best nized that both northern and southern hemispheresink interestsof this exercise.Rather, the main point is that fluxescan changequite significantlyon an interannual it is possibleto generate,in this model,elevated levels timescale(especially the southernhemisphere, where a of CO•.mixing ratio in the high-latitudenorthern hemi- 300% subdecadalchange has been postulatedin order spherethat are relativelyinsensitive to the structureof to explainvariations in the data) [Conwayet al., 1994]. the moresoutherly sinks, whose architecture is necessar- The mechanisms of these variations are not properly ily of an empiricalnature, when the only "calibration" understood. of the model output for the high southernlatitudes is In the model simulation to examine the atmospheric the Antarctic ice core data. effectsof the PIAGREV, the model was started in AD The simplifiedphysical ocean sink mechanism in the 1800. The biomasscarbon input function(described originalmodel [Higuchi,1983b] was found to be in- in the next section)was applied. Empiricalsink func- adequate;that is, not enoughgas is absorbedby the tions were constructedin order to replicate,as nearly as modelocean. Subsequently, it hasbeen found that there possible,the atmosphericCO•. curvefor highsouthern areother processes [e.g., Farmer et at., 1993]operating latitudes obtained from Antarctic ice cores. at the air-ocean interface that enhance the absorption. Also, we are ignoringthe possibilitythat the oceans Biomass Carbon Input Function haveundergone decadal-scale circulation changes. Such The biosphericcarbon function shownin Figure 5 changescould have affected atmospheric CO•. levelsin (curve4) wascreated from an initial estimated function a similarway that oceanicE1 Nifio'saffect atmospheric usinga successivetrial procedure.We manually created CO•.. There is somepreliminary observational evidence the trial function with an amplitude equal to the func- to suggestthat deepwater formation in the North At- tion derivedby Pengand Freyer [1986] and with a shape lantic(Greenland Sea) has slowed down in recentyears. that reflectedthe two peaksin activity of the PIAGREV A recentglobal carbon model [Kwon and $chnoor, 1994] as seen in our own ice core data. Because the GCCM is similarlylimited with respectto sink structure. is zonallyaveraged, the carboninput functiondoes not In Figure6, it is to be notedthat peaksin the simu- discriminate between emissions that came from North lated CO•. curve d showa lag with respectto peaksin America, Europe, or Eurasia. This initial functionwas the forcingfunction of the modelnorthern hemisphere HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA 23,331

atmosphere. Without a suitably prescribed additional transient CO2 values seen in northern hemisphere ice sink to the south, Antarctic atmosphericCO2 concen- coresand to the hitherto unexplainedirregularities ex- trations would closelyfollow the northern hemisphere, istingbetween different tree 51aC time series. Using the with a lag of between1 and 2 years[see also Keeling biomass CO• emissionsfor 1900 derived in this paper ½tal., 1989]. Carbonforcing functions in the northern (input functionin Fig. 5) and applyingthem to mid- hemispheremidlatitudes and tropics after ~1950 are latitude North America during the PIAGREV, recent designedto replicate, in the model, the high-latitude simulationsby S. Iacobellis(unpublished data, 1995) northern hemisphereinstrumental data as well as to be indicate that "excess"CO• gas concentrationsof 2-4 consistentwith variousestimates of biomassburning ppmv above backgroundcould have reachedparts of [e.g.,Peng and Freyer,1986]. The importantfeatures the high Canadian Arctic, Greenland,and Europe and of curved are the twin maxima between1850 and 1950, persistedon the timescaleof 1 month. The simulations whichclearly reflect the assumedbiomass forcing func- made so far use only the 1979 global wind data set. For tion. The difference between curves a and d defines a extreme meridional airflow, significantlyelevated levels possibletransient interhemispheric gradient that might of CO2 could have impacted distant tree stand and ice have been similar to the present one, on a timescaleof coresites to producethe anomaliesthat are seenin the decades,during the late 19th century. data. The currentmaximum interhemispheric (approximate An example of a visible smoke layer in close prox- poleto pole) differencein CO2 mixingratio is 3-4 ppmv imity to the ice coresite in the St. Elias Mountainsis [Conwaye! al., 1988; Tanakae! al., 1988;Keeling e! al., describedby Benjey [1974]. The smokepall was seen 1989],whereas its value in 1957 was closeto ~1 ppmv in late June 1969, and it was estimated by pilots to [Keeling½! al., 1989](see Fig. 6). The valuebefore the have a ceiling of around 5000 m against Mount Logan. start of instrumental measurements is not known with This smoke originated from forest fires burning along any certaintybut hasgenerally been assumedto be neg- the Alaska-Yukon border. In August 1986 the first au- ligible (other model studiesincluded a much smaller thor saw a similar brown smokelayer (approximately biomassinput functionthan we have used). However, 1.5 km thick), which was attributed to forest fires in in the light of the data presentedhere, this does not Alaska. The ceiling then was at least 5000 m altitude. appear to have been the case. For future, more precise, Although of a regionalscale, these observationsserve to model simulations it is of interest to know whether the reinforce the conjecture that unmixed forest fire emis- past differencebetween the CO2 mixing ratios of the sions may impact ice core sites in the northern hemi- high-latitude northern and southernhemisphere atmosphere. spherescould have been similar to or greater than the The collectednorthern hemisphereCO• data suggest present value. that concentrationsof this trace gas may have signifi- cantlyexceeded southern hemisphere values (by several Discussion of Results and Conclusions ppmv) at certain times and for brief seasonalperiods Given that the PIAGREV signature in the ice cores at the ice core sites. These northern hemispherehigh- is a result of the "fallout" of biomass burn products latitude CO2 levels(elevated by severalppmv) could transported over long distances,we can make use of a have been seasonally sustained on the timescale of a large literature base upon which to develop a discus- decade, which is the timescale of the PIAGREV. The sion of some of the dynamics of the processesinvolved possibility was already considered by $iegenthaler et [e.g., Levine, 1991]. From the purely meteorological al. [1988] that atmosphericCO2 levels may have in- standpoint the recently developed concept of "tropo- creasedsignificantly, albeit temporarily, before the ex- spheric rivers," or filaments of water vapor in the tro- tensive use of fossil fuels. However, these data, which posphere[Newell e! al., 1992;Newell and Zhu, 1994], came from a south pole core, have not been substanti- appearsto offer a useful avenuefor discussingthe dy- ated from additional Antarctic cores or from Greenland namics of long-rangetransport of surfaceemissions as- core studied so far. sociated with agrarian activities. It is possible that The GCCM modeling suggeststhat the higher CO2 smokeplumes from major fires in North America may values plotted in Figure 6 do not representfully mixed have formed similar "filaments" of aerosols in the atmo- northern hemisphere atmospheric annual values, but sphere, even in association with the filaments of water neverthelessthey can be consideredto be acting as vapor, which appears to have the capacity to form an tracers for biomass burning. Similarly, certain trees aerosolwith CO•. This has a bearing on some of the throughout the northern hemisphere would have re- anomalouslyhigh ice core CO• data presentedearlier. spondedto regional(and seasonal)climatic variations (There is somebasis for supposingthat solidrime ice and to smokeassimilation according to atmosphericcir- derived from supercooledwater vapor droplets is en- culation at the time. Certain trees only form new wood riched in CO•. Rime ice has been found infrequently at for very brief periods, with timing determined by local high altitudeson Mount Logan.) climatic conditions. This could account for much of the The time-dependent evolution of forest fire smoke nonuniformity seen in the late 19th century tree ring plumes that has been simulated in a numerical at- data. mospherictransport model by lacobelliset al. [1994] The atmospheric modeling results reported here in- also provides possible answers to the question of the dicate that the maximum short-term interhemispheric 23,332 HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA

CO2 difference(the north-southgradient) might have References equaledor slightlyexceeded the presentvalue, although this modelingis not highly constrained.Nevertheless, Alford, D., and C. Keeler, Stratigraphic studies of the win- the modelingresults do supportthe atmosphericCO2 ter snow layer, Mount Logan, St. Elias Range, in Icefield data in several important aspects. The structure and Ranges Research Project, Scientific Results, vol. 2, edited results of the GCCM used here also suggestthat car- by V. Bushnell and R. H. Ragle, pp. 37-42, Am. Geogr. bon cyclemodels should be improvedby incorporating Soc., New York, and Arctic Inst. N. Am., Montreal, 1970. Benjey, W. G., The effects of forest-fire smoke on insolation better physicalmechanisms into the carbonsinks [e.g., in the St. Ehas Mountains, in Icefield Ranges Research Robertsonand Watson,1992; Farmer et al., 1993] so Project, Scientific Results, vol. 4, edited by V. Bushnell that current semi-empiricalor quasi-arbitrarymethods and M. G. Marcus, pp. 129-131, Am. Geogr. Soc., New of model tuning may be reducedin the future. Also, York, and Arctic Inst. N. Am., Montreal, 1974. in modelingthe carboncycle through the PIAGREV, Benson, C. S., Stratigraphic studies in the snow and firn the effects of possibleatmospheric and oceaniccooling of the Greenland Ice Sheet, Research Report 70, 93 pp., should be accounted for. Snow, Ice, Permafrost Res. Establ., Wilmette, II1., 1962. Bien, G. S., N. W. Rakestraw, and H. E. Suess,Radiocarbon There are other trace gasesthat may be used as in- concentration in Pacific Ocean water, Tellus, 1œ,436-443, dicators of biomass burning. Methane and nitrous ox- 1960. ide concentrationdata for Mount Logan [Dibb et al., Bischof, W., The CO2 content of the upper polar tropo- 1993]are smoothlyvarying and marginallysuggest (as spherebetween 1963 and 1979, in Carbon Cycle Modelling, doesthe CO2 data) small increases~1860. The more Scope,vol. 16, edited by B. Bohn, pp. 113-116, John Wi- anthropogenicallysensitive southern Greenland (20D) ley, New York, 1981. Bjorkstrom, A., A model of CO2 interaction between at- core[Dibb et al., 1993]shows consistently elevated lev- mosphere, oceans and land biota, in The Global Carbon els of methane during the late 19th century. These Cycle, Scope,vol. 13, edited by B. Bolin, E. T. Degens, collected data offer qualitative supporting evidencefor S. Kempe, and P. Ketner, pp. 403-457, John WHey, New biomassfire activity [Laursenet al., 1992;Levine et al., York, 1979. 1992]in North America. Furthermore,the implications Broecker, W. S., R. Gerard, M. Ewing, and B. Heezen, Nat- of biomassburning for global climate variations are now ural radiocarbon in the Atlantic Ocean, J. Geophys.Res., well known. 65, 2903-2931, 1960. Broecker, W. S., T.-H. Peng, and R. Engh, Modelhng the It appears that associated factors, such as surface carbon system, in Carbon Dioxide Effects Research and albedochanges and atmosphericdust and smoke[Sagan AssessmentProgram, edited by L. E. Schmitt, Proc. Car- et al., 1979],causing local cooling (at leastin the north- bon Dioxide Clim. Res. Program Conf., 11, 43-99, 1980. ern hemisphere)temporarily negatedany tendencyfor Chatfield, C., Statistics for Technology,3rd ed., Chapman radiative gas greenhousewarming. As a result, the PI- and Hall, New York, 1983. AGREV, through the influenceof smokeplumes [Pit- Chylek, P., B. Johnson,and H. Wu, Black carbon concentra- rock et al., 1988; Penner et al., 1991] and associated tion in a Greenland Dye-3 ice core, Geophys.Res. Lett., 19, 1951-1953, 1992. agricultural dust, may have helped deepenand prolong Claud, C., Analyse du CO2 conrenu dans les bulles d'au the endingof the Little Ice Age [Grove,1988], which piegeesdans la glacedu Mont Logan (Yukon, Canada), is recognizedto have occurred in many glaciated alpine Mere. Lab. Glaciol. Geophys. Environ., 67 pp., CNRS, regions. Documentation of the northern hemispherere- Grenoble, France, 1985. gional climatic regimesthroughout the PIAGREV will Conway, T. J., P. P. Tans, L. S. Waterman, K. W. Thoning, be the subject of a paper in preparation. K. A. Masarie, and R. H. Gammon, Atmospheric carbon dioxide measurements in the remote global troposphere, Acknowledgments. We acknowledgesupport from En- 1981-1984, Tellus, dOB, 81-115, 1988. vironment Canada (National Hydrology ResearchInstitute Conway, T. J., P. P. Tans, L. S. Waterman, K. W. Thoning, and the AtmosphericEnvironment Service), the Electric D. R. Kitzis, K. A. Masarie, and N. Zhang, Evidence for Power Research Institute of California, the U.S. National interannual variability of the carbon cycle from the Na- ScienceFoundation, Natural Sciencesand Engineering Re- tional Oceanicand AtmosphericAdministration/Climate search Council of Canada, and the Arctic Institute of North Monitoring and Diagnostics Laboratory Global Air Sam- America, University of Calgary. We alsothank J.-M. Barnola pling Network, J. Geophys.Res., 99, 22,831-22,855, 1994. for providing the Mount Logan ice core CO2 data and D. Delmas, R. J., A natural artifact in Greenland ice core CO2 Rayhand and C. Lorius (CNRS, Grenoble)for supporting measurements, Tellus, •5B, 391-396, 1993. the collaboration. J.E. Dibb, M.J. Spencer, and S. Whitlow Dibb, J. E., J. L. Jaffrezo, and M. Legrand, Initial findings helped in data interpretation and A.T. Wilson, C.S. Wong, of recent investigations of air-snow relationships in the J.-M. Barnola, M. Stuiver, and S.W. Leavitt read earlier summit region of the Greenland Ice Sheet, J. Atrn. Chern., versions of the manuscript and offered many helpful sug- ld, 167-180, 1992. gestions. L.J. Toolin and C.J. Eastoe kindly read the final Dibb, J. E., R. A. Rasmussen, P. A. Mayewski, and G. manuscriptand noted that their unpublished/;•aC record Holdsworth, Northern hemisphere concentrations of me- for C4 grassesfrom the American Southwest appeared to thane and nitrous oxide since 1800: Results from the corroborateat least the NH4+ recordin Figure1 for the Mount Logan and 20D ice cores, Chernosphere,œ7, 2413- late 19th century. We thank the journal reviewers and the 2423, 1993. editor (J. Penner) for helpful adviceand detectionof ob- Enting, I. G. The incompatibility of ice core CO2 data with scure statements and for overall improvement to the paper. reconstructions of biotic CO2 sources,II, The influence of We also acknowledgeF.O. Smith for editing and document CO• fertilized growth, Tellus, JJB, 23-32, 1992. preparation. Acknowledgement of individuals does not im- Etheridge, D. M., L. P. Steele, R. L. Langenfelds, R. J. ply that they necessarily agree with all our interpretations Francey, J.-M. Barnola, and V. I. Morgan, Natural and of the data. anthropogenic changesin atmospheric COa over the last HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA 23,333

1000 years from air in Antarctic ice and firn, J. Geophys. model: The atmosphere-terrestriM biosphere-oceaninter- Res., 101, 4115-4128, 1996. action, GlobalBiol. Cycles,8(3), 295-306,1994. Farmer, D. M., C. L. McNeil, and B. D. Johnson, Evidence Labelle, J., Snow studies at high elevations, Mount Logan, for the importance of bubbles in increasing Mr-sea gas Yukon, in Icefield Ranges Research Project, Scientific Re- flux, Nature, 361, 620-623, 1993. sults, vol. 4, edited by V. Bushnell and M. G. Marcus, pp. Francey,Ri J., TasmanJan treerings belie suggested anthro- 211-218, Am. Geogr. Soc. and Arct. Inst. N. Am., 1974. pogenic3C/•2C trends,Nature, œ90, 232-235, 1981. Laursen, K. K., P. V. Hobbs, L. F. Radke, and R. A. Ras- Francey, R. J., and G. D. Farquhar, An explanation of mussen, Some trace gas emissionsfrom North American •JC/X2Cvariations in treerings, Nature, œ97, 28-31, 1982. biomass fires with an assessmentof regional and global Freyer, H. D., Preliminary evaluation of past CO2 increase fluxes from biomassburning, J. Geophys.Res., 97, 20,687- asderived from •aC measurementsin treerings, in Carbon 20,701, 1992. Dioxide, Climate and Society, edited by J. Williams, pp. Leavitt, S. W., The Great Lakes forest fires of 8-10 October, 69-77, Pergamon, New York, 1978. 1871, pp. 519-522, in Global Biomass Burning, edited by Freyer,H. D., On the •3C recordin tree rings,I, xJCvari- J. S. Levine, MIT Press, Cambridge, Mass., 1991. ations in northern hemispheric trees during the last 150 Leavitt, S. W., Environmentalinformation from •3C/•2C years, Tellus, 31, 124-137, 1979. ratios of wood, in Climate Change in Continental Iso- Freyer, H. D., Interpretation of the northern hemisphere topic Records, Geophys. Monogr. Set., vol. 78, pp. 325- recordof x3C/x•C trendsof atmosphericCO2 in tree 331, AGU, Washington, D.C., 1993. rings, in The Changing Carbon Cycle: A Global Analysis, Leavitt, S. W., and A. Long,Altitudinal differencesin 5•3C edited by J. R. Trabalka and D. E. Reichle, pp. 125-150, of Bristlecone Pine tree rings, Naturwissenschaften, 79, Springer-Verlag, New York, 1986. 178-180, 1992. Freyer,H. D., and N. Belacy,X3C/xeC records in northern Legrand, M., M. DeAngelis, T. Staffelbach,A. Neftel, and B. hemispheric trees during the past 500 years-Anthropogen- Stauffer, Large perturbations of ammonium and organic ic impact and climatic superpositions, J. Geophys.Res., acids content in the summit-Greenland ice core, Geophys. 88, 6844-6852, 1983. Res. Left., 19(5), 473-475, 1992. Friedli, H., H. LStscher, H. Oeschger,U. Siegenthaler, and Levine, J. S. (Ed.), Global BiomassBurning, MIT Press, B. Stauffer,Ice corerecord of the 13C/12Cratio of atmo- Cambridge, Mass., 1991. spheric CO2 in the past two centuries, Nature, 32-4, 237- Levine, J. S., W. S. Cofer III, D. I. Sebacher, R. P. Rhine- 238, 1986. hart, E. L. Winstead, S. Sebacher, C. R. Hinkle, P. A. From, E., and G. D. Keeling, Reassessmentof late 19th cen- Schmaltzer, and A.M. Koller Jr., The effects of fire on tury atmospheric carbon dioxide variations in the air of biogenic emissionsof methane and nitric oxide from wet- western Europe and the British Isles based on an unpub- lands, J. Geophys. Res., 95, 1853-1864, 1992. lished analysis of contemporary air massesby G. S. Cal- Marland, G., Fossil fuels COe emissions, CDIAC Communi- lendar, Tellus, 38B, 87-105, 1986. cations, Winter 1989, pp. 1-3, Carbon Dioxide Inf. Anal. Gould, D. F., Beyond the Shining Mountains, Binsford and Cent., Oak Ridge Natl. Lab., Oak Ridge, Tenn., 1989. Mort, Portland, Or., 1938. Marland, G., and R. M. Rotty, Carbon dioxide emissions Grove, J. M., The Little Ice Age, Methuen, New York, 1988. from fossil fuels: A procedure for estimation and results Higuchi, K., A global carbon cycle model, Ph.D. disserta- for 1950-1982, Tellus, 36B, 232-261, 1984. tion, Univ. of Toronto, Ontario, Canada, 1983a. Marland, G., T. A. Boden, R. C. Griffin, S. F. Huang, P. Higuchi, K., Effect of two different atmospheric models on Kanciruk, and T.R. Nelson, Estimates of COe emissions the absorptive rate of excess atmospheric carbon by the from fossil fuel burning and cement manufacturing, based sea, Geophys.Res. Left., I0(9), 869-872,1983b. on the United Nations Energy Statistics and U.S. Bu- Holdsworth, G., H. R. Krouse, and M. Nosal, Ice core cli- reau of Mines cement manufacturing data, NDP-030, Oak mate signalsfrom Mount Logan, Yukon, A.D. 1700-1987, Ridge Natl. Lab., Oak Ridge, Tenn., 1989. in Climate Since AD 1500, edited by R. S. Bradley and P. D. Jones, pp. 483-504, Routledge, New York, 1992. Mayewski, P. A., W. B. Lyons, M. J. Spencer, M. Twick- let, C. F. Buck, and S. I. Whitlow, An ice core record Hyson, P., P. J. Fraser, and G. I. Pearman, A two-dimensional transport simulation model for trace atmospheric con- of atmospheric response to anthropogenic sulphate and nitrate, Nature, 3d6, 554- 556, 1990. stituents, J. Geophys. Res., 85, 4443-4455, 1980. Iacobellis, S. F., R. Frouin, H. Razafimpanilo, and R. C. Mayewski, P. A., L. D. Meeker, M. C. Morrison, M. S. Twickler, S. I. Whitlow, K. K. Ferland, D. A. Meese, M. J. Somerville, Biomass burning in the savannasof North Africa and atmospheric carbon dioxide, J. Geophys.Res., R. Legrand, and J.P. Steffensen,Greenland ice core "signal" characteristics: An expanded view of climate change, 99, 8321-8334, 1994. J. Geophys. Res., 98, 12,839- 12,847, 1993. Intergovernmental Panel on Climate Change, Climate change1994: Radiative forcing of climate changeand an Morimoto, S., A study of the carbon cycle on the earth's evaluation of the IPGG 1992 emission scenarios, edited by surface from measurements of the carbon isotope ratio of J. T. Houghton, L. G. Meira Filho, J. Bruce, H. Lee, B. atmospheric CO2, Ph.D. thesis, Tohoku Univ., Sendal, A. Gallander, E. Haites, N. Harris, and K. Maskell, 339 Japan, 1994. pp., Cambridge Univ. Press, Cambridge, New York, 1994. Neftel, A., H. Fried]i, E. Moor, H. LStscher, H. Oeschger, Junge, G. E. Air Chemistry and Radioactivity, Academic, U. Siegenthaler, and B. Stauffer, Historical CO2 record San Diego, Calif., 1963. from the Siple Station ice core, edited by T. A. Boden, D. Keeling, C. D., Industrial productionof carbon dioxide from P. Kaiser, R. J. Sepanski, and F. W. Stoss, pp. 11-14, in fossil fuel and limestone, Tellus, 25, 1174-1198, 1973. Trends '93: A Compendium of Data on Global Change, Keeling, C. D., R. B. Bacastow,A. F. Carter, S.C. Piper, T. ONRL/CDIA C-65, Carbon Dioxide Inf. Anal. Cent., Oak P. Whorl, M. Heimann, W. G. Mook, and H. Roeloffzen, Ridge Natl. Lab., Oak Ridge, Tenn., 1994. A three-dimensional model of atmospheric CO2 transport Newell, R. E., and Y. Zhu, Tropospheric rivers: A one year based on observed winds, 1, Analysis of observational record and a possible application to ice core data, Geo- data, in Aspectsof Climate Variability in the Pacific and phys. Res. Lett., 21, 113-116, 1994. the Western Americas, edited by D. H. Peterson, pp. 165- Newell, R. E., N. E. Newell, Y. Zhu, and C. Scott, Tropo- 236, AGU, Washington, D.C., 1989. sphericrivers?-A pilot study, Geophys.Res. Lett., 1œ(œ•), Kwon, O.-Y., and J. L. Schnoor, Simple global carbon 2401-2404, 1992. 23,334 HOLDSWORTH ET AL.: BIOMASS BURNING HISTORY IN ICE CORE DATA

Nir, A., and S. Lewis, On tracer theory in geophysicalsys- Stauffer, B., A. Neftel, H. Oeschger,and J. Schwander,CO2 tems in the steady and nonsteady state, I, Tellus, œ7, concentration in air extracted from Greenland ice sam- 372-383, 1975. ples, in Greenland Ice Core: Geophysics, Geochemistry, Pearman, G. I., and P. Hyson, Activities of the global bio- and the Environment, edited by C. C. Langway Jr., H. sphere as reflected in atmospheric CO2 records, J. Geo- Oeschger,and W. Dansgaard, Geophys.Monogr. Set., vol. phys. Res., 85, 4468-4474, 1980. 33, pp. 85-89, AGU, Washington, D.C., 1985. Peng, T.-H., and H. Freyer, Revised estimates of atmo- Stuiver, M., Ancient carbon cycle changesderived from tree- sphericCO2 variationsbased on the tree-ring13C record, ring •aC and •4C, in The ChangingCarbon Cycle: A in The Changing Carbon Cycle: A Global Analysis, edited Global Analysis, edited by J. R. Trabalka and D. E. Re- by J. R. Trabalka and D. E. Reichle,pp. 151-159, Springer- ichle, pp. 109-124, Springer-Verlag,New York, 1986. Verlag, New York, 1986. Stuiver,M., andT.F. Braziunas,Tree cellulose 1•C/12C iso- Peng, T-H., W. S. Broecker, H. D. Freyer, and S. Trumbore, tope ratios and climatic change, Nature, 328, 58-60, 1987. A deconvolutionof the tree ringsbased 5XaC record, J. Stuiver,M., R. L. Burk, and P. D. Quay, 1•C/12C ratio Geophys.Res., 88, 3609-3620, 1983. in tree rings and the transfer of biospheric carbon to the Penner, J. E., S. J. Ghan, and J. J. Walton, The role of atmosphere, J. Geophys.Res., 89, 11,731-11,748, 1984. biomassburning in the budget and cycle of carbonaceous Tanaka, M., T. Nakazawa, S. Aoki, and H. Ohshima, Air- soot aerosols and their climate impact, in Global Biomass craft measurements of tropospheric carbon dioxide over Burning, edited by J. S. Levine, MIT Press, Cambridge, the JapaneseIslands, Tellus, JOB, 16-22, 1988. Mass., 1991. Tans, P. P., and W. G. Mook, Past atmospheric CO2 levels Pittock, A. B., J. S. Frederiksen, J. R. Garratt, and K. andthe 13C/12Cratios in treerings, Tellus, 3œ, 268-283, Walsh, Climatic effects of smoke and dust produced from_ 1980. nuclear conflagrations, in Aerosols and Climate, edited Tans, P. P., I. Y. Fung, and T. Takahashi, Observational by P. V. Hobbs and M.P. McCormick, pp. 395-410, A. constraints on the global atmospheric CO2 budget, Sci- Deepak, Hampton, Va., 1988. ence, œ4{7,1431-38, 1990. Quay, P. D., B. Tilbrook, and C. S. Wong, Oceanicuptake of Wahlen, M., D. Allen, B. Deck, and A. Herchenroder, Ini- fossil fuel CO2:Carbon-13 evidence, Science, œ56,74-79, tial measurementsof CO2 concentrations(1530 to 1940 1992. AD) in air occludedin the GISP2 ice corefrom central Raynaud, D., and J. M. Barnola, An Antarctic ice core re- Greenland,Geophys. Res. Left., 18(8), 1457-60,1991. veals atmospheric CO2 variations over the past few cen- Waterman, L. S., Comments on 'The Montsouris series of turies, Nature, 315, 309-311, 1985. carbon dioxide concentration measurements, 1877-1910', Reed, R., and K. German, A contribution to the prob- by G. Stanhill, Clim. Change, 5, 413-415, 1983. lem of stratosphericdiffusion by large scalemixing, Mon. Whitlow, S. I., P. A. Mayewski, G. Holdsworth, M. S. Twick- Weather Rev., 93, 313-321, 1965. ler, and J. E. Dibb, An ice core based record of biomass Robertson, J. E., and A. J. Watson, Thermal skin effect of burning in the arctic and subarctic, 1750-1980, Tellus, the surface ocean and its implications for CO2 uptake, •6B, 234-242, 1994. Nature, 358, 738-740, 1992. Wigley, T. M. L., The pre-industrial carbon dioxide level, Sagan, C., O. B. Toon, and J. B. Pollack, Anthropogenic Clim. Change, 5, 315-320, 1983. albedo changesand earth's climate, Science, œ06, 1363- Wilson, A. T., Pioneer agriculture explosion and CO2 levels 1368, 1979. in the atmosphere, Nature, 273, 40-41, 1978. Schwander, J., The transformation of snow to ice and the oc- Wong, C. S., Atmospheric input of carbon dioxide from clusion of gases,in The Environmental Record in Glaciers burning wood, Science, 200, 197-200, 1978. and Ice Sheets,edited by H. Oeschgerand C. C. Langway, pp. 53-67, John Wiley, New York, 1989. G. Holdsworth, The Arctic Institute of North America, Siegenthaler,U., 19th century measurementsof atmospheric Universityof Calgary, Alberta, Canada T2N 1N4 (email: CO2-A comment, Clim. Change, 6, 409-411, 1984. [email protected]). Siegenthaler,U., and H. Oeschger,Biospheric CO2 emissions K. Higuchi, Atmospheric Environment Service, Depart- during the past 200 years reconstructedby deconvolution ment of the Environment, Downsview, Ontario, Canada of ice core data, Tellus, 39B, 140-154, 1987. M3H 5T4 (email: [email protected]). Siegenthaler, U., H. Friedh, H. Loetscher, E. Moor, A. Nef- P.A. Mayewski and G.A. Zielinski, Glacier Research tel, H. Oeschger, and B. Stauffer, Stable-isotope ratios Group, Institute for the Study of Earth, Oceans and Space, and concentrationof CO2 in air from polar ice cores,Ann. Universityof New Hampshire,Durham, N.H. 03824 (email: Glaciol., 10, 151-156, 1988. [email protected];[email protected]). Staffelbach,T., B. Stauffer, A. Sigg, and H. Oeschger,CO2 B. Deck and M. Wahlen, Scripps Institution of Oceanog- measurements from polar ice cores: More data from dif- raphy, University of California, La Jolla, CA 92093-0220 ferent sites, Tellus, d3B, 91-96, 1991. (email: [email protected]; mwahlen@ssurf. ucsd.edu). Stanhill, G. The Montsouris seriesof carbon dioxide concen- P. Chylek, P. Damiano, and B. Johnson, Atmospheric Sci- tration measurements 1877-1910, Clim. Change, d, 221- ence Program, Dalhousie University, Halifax, N.S., Canada 237, 1982. B3H 3J5 (email: [email protected]). Stanhill, G., Reply with some additional details on 'The Montsouris series of carbon dioxide concentration measurements, 1877-1910', Clim. Change, 5, 417-419, 1983. Stanhill, G., A further reply concerningthe accuracyof 'The Montsouris series of carbon dioxide concentration mea- (ReceivedApril 26, 1995;revised March 11, 1996; surements, 1877-1910', Clim. Change, 6, 413-415, 1984. acceptedMarch 12, 1996.)