R EPORTS

layer dominated by horizontal flow and shorter wavelengths in the upper and lowermost isotropy has been shown to adequately describe the emphasize the unique character of the two mantle (fig. S3), but not in the bulk of the lower data in many of the regional studies (5). mantle. On this basis, we limited our inversion to 21. A. K. McNamara, P. E. van Keken, S.-I. Karato, Nature superplume regions, for which we bring degree 8 in ␰. Statistical tests also demonstrate 416, 310 (2002). additional evidence of large-scale up- improvementin fitsignificantata 95% confidence 22. L. Stixrude, in The Core-Mantle Boundary Region,M. level for the CMB-sensitive portion of the data set Gurnis, M. E. Wysession, E. Knittle, B. A. Buffett, Eds. welling. Although measurements of defor- ␰ mation at the pressures and temperatures for the model with structure in the lowermost (American Geophysical Union, Washington, DC, 400 km of the mantle compared to an isotropic 1998), pp. 83–96. corresponding to the CMB region are not model (SOM text). 23. E. Knittle, R. Jeanloz, Science 251, 1438 (1991). yet available, our results suggest that sim- 14. J.-P. Montagner, T. Tanimoto, J. Geophys. Res. 96, 24. V. Babuska, M. Cara, Seismic Anisotropy in the Earth ilar relationships between anisotropic sig- 20337 (1991). (Kluwer Academic, Boston, 1991). 15. G. Ekstro¨m, A. M. Dziewonski, Nature 394, 168 (1997). nature and flow prevail in the uppermost 25. J.-P. Montagner, D. L. Anderson, Phys. Earth Planet. 16. Y. Gung, M. Panning, B. Romanowicz, Nature 422, Inter. 54, 82 (1989). and lowermost mantle. 707 (2003). 26. M. E. Wysession, K. M. Fischer, G. I. Al-eqabi, P. J. 17. A. M. Dziewonski, D. L. Anderson, Phys. Earth Planet. Shore, I. Gurari, Geophys. Res. Lett. 28, 867 (2001). References and Notes Inter. 25, 297 (1981). 27. This is contribution 03-10 of the Berkeley Seismological 1. T. Lay, Q. Williams, E. J. Garnero, Nature 392, 461 (1998). 18. G. Masters, S. Johnson, G. Laske, B. Bolton, Philos. Laboratory. Supported by NSF grant EAR-9902777. 2. S.-I. Karato, Earth Planets Space 50, 1019 (1998). Trans. R. Soc. London Ser. A 354, 1385 (1996). 3. J.-M. Kendall, P. G. Silver, Nature 381, 409 (1996). 19. Y. J. Gu, A. M. Dziewonski, W. Su, G. Ekstro¨m, J. Supporting Online Material 4. L. P. Vinnik, V. Farra, B. Romanowicz, Geophys. Res. Geophys. Res. 106, 11169 (2001). www.sciencemag.org/cgi/content/full/303/5656/351/DC1 Lett. 16, 519 (1989). 20. As shown by resolution tests, only the long-wavelength SOM Text 5. T. Lay, Q. Williams, E. J. Garnero, L. Kellogg, M. features of our model are well constrained (13). More- Figs. S1 to S7 Wysession, in The Core-Mantle Boundary Region,M. over, we have ignored azimuthal anisotropy, which may Gurnis, M. E. Wysession, E. Knittle, B. A. Buffett, Eds. influence the details of our model, although radial an- 15 September 2003; accepted 3 December 2003 (American Geophysical Union, Washington, DC, 1998), pp. 219–318. 6. S. A. Russell, T. Lay, E. J. Garnero, J. Geophys. Res. 104, 13183 (1999). 7. C. Thomas, J.-M. Kendall, Geophys. J. Int. 151, 296 (2002). Siberian Peatlands a Net 8. M. S. Fouch, K. M. Fischer, M. E. Wysession, Earth Planet. Sci. Lett. 190, 167 (2001). 9. X. D. Li, B. Romanowicz, Geophys. J. Int. 121, 695 (1995). Sink and Global Source 10. C. Me´gnin, B. Romanowicz, Geophys. J. Int. 143, 709 (2000). 11. Our data set consists of three-component long- Since the Early Holocene period seismograms (minimum period of 60 s for surface waves and 32 s for body waves) inverted in L. C. Smith,1,2* G. M. MacDonald,1,3* A. A. Velichko,4 the time domain in the framework of nonlinear 1 4 1 1,4 asymptotic coupling theory (9). In this normal- D. W. Beilman, O. K. Borisova, K. E. Frey, K. V. Kremenetski, mode perturbation-based approach, we calculate Y. Sheng1 two-dimensional great-circle sensitivity kernels that account for the distribution of sensitivity along and in the vicinity of the ray when using Interpolar methane gradient (IPG) data from ice cores suggest the “switching finite-frequency waveform data. Additionally, this on” of a major Northern Hemisphere methane source in the early Holocene. approach allows us to use diffracted waves as well Extensive data from ’s West Siberian Lowland show (i) explosive, as simultaneously arriving phases, which is not widespread peatland establishment between 11.5 and 9 thousand years ago, possible when using ray theory– and travel time– based approaches (fig. S1 and SOM text). predating comparable development in North America and synchronous with 12. Our model is parameterized in terms of radial increased concentrations and IPGs, (ii) larger carbon anisotropy, which is usually described by density ␳, ϳ ϭ␳ 2 ϭ␳ 2 stocks than previously thought (70.2 Petagrams, up to 26% of all ter- and five elastic coefficients (A VPH, C VPV, ϭ␳ 2 ϭ␳ 2 restrial carbon accumulated since the Last Glacial Maximum), and (iii) little L VSV, N VSH, and F). We chose to describe this equivalently in terms of the isotropic P and S evidence for catastrophic oxidation, suggesting the region represents a velocities [V2 ϭ 1/5(V2 ϩ 4V2 ) and V2 ϭ 1/3(V2 ϩ Piso PV PH Siso SH 2 ␰ϭ long-term sink and global methane source since the early 2VSV)] and three anisotropic parameters [ N/L, ␾ϭC/A, and ␩ϭF/(A Ϫ 2L)]. The isotropic velocities are Holocene. derived from the Voight average isotropic elastic proper- ␳ 2 ϭ ϩ ties (24). These can be expressed as Vp (1/15)[3C Ice-core records of atmospheric methane mates of carbon storage in high-latitude ϩ ␩ ϩ Ϫ␩ ␳ 2 ϭ ϩ Ϫ (8 4 )A 8(1 )L] and Vs (1/15)[C (1 concentration show dramatic peaks in the peatlands are poorly constrained but large 2␩)A ϩ (6 ϩ 4␩)L ϩ 5N]. Because in this approach we assume anisotropy to be small, we can simplify the early Holocene, drawing considerable de- (180 to 455 Pg C) (6), representing up to kernel calculation by using ␩Ϸ1 and A Ϸ C to reduce bate as to their source (1). Expansion of ϳ1/3 of the global carbon pool (1395 the Voight average expressions to the ones used above. tropical has emerged as a popular Pg C) (7). Most are known to have formed To reduce the number of free parameters in the inver- sion, we assume the following scaling relationships: ␦ hypothesis (2–5), in part because high-lat- since the Last Glacial Maximum and thus ln(V ) ϭ 0.5 ␦ ln(V ), ␦ ln(␩) ϭϪ2.5␦ ln(␰), ␦ itude peatlands were not well developed in represent a major terrestrial carbon sink Piso Siso ln(␾) ϭϪ1.5␦ ln(␰), ␦ (␳) ϭ 0.3␦ ln(V ). Scaling ϳ and ln Siso North America by 11,000 calendar years during the Holocene (8). However, the true relations for the anisotropic parameters have been de- termined in the laboratory only for conditions down to ago (11 ka), a period of peak methane magnitude and timing of this sink is poorly 500 km (25), so we also performed a lower resolution concentration. However, the timing and known because of insufficient data on - inversion where ␰, ␾, and ␩ were allowed to vary volume of peatland growth in Russia, land distribution through time (9), depth, ␰ independently. The structure in changed very little, which contains perhaps half of the world’s area, age, and carbon content (6, 10). These and analysis of the resolution matrix indicated that ␰ was indeed the best resolved of the three parameters, peat, is virtually unknown. Published esti- uncertainties make it difficult to infer the and tradeoffs were not too strong (fig. S2). The model influence of northern peatlands on Holocene was parameterized with 16 splines radially, and by 1Department of Geography, 2Department of Earth concentrations and to predict spherical harmonics horizontally up to degree 16 for and Space Sciences, 3Department of Organismic Biol- ␰ the amount of sequestered carbon that could VS and degree 8 for . ogy, Ecology and Evolution, University of California, iso potentially be mobilized under a warmer Arc- 13. We performed several tests using the resolution Los Angeles, CA 90095–1524, USA. 4Russian Academy matrix calculated for an inversion up to degree 16 tic through water table lowering, peat ␰ of Sciences, Moscow 109017, Russia. in both VS and . Tests with a random input model oxidation, and CO outgassing (6, 11, 12); with a white wavelength spectrum indicate that *To whom correspondence should be addressed. E- 2 we are capable of correctly resolving the long- mail: [email protected] (L.C.S.); macdonal@geog. biosphere uptake (11, 12); or increased dis- wavelength ␰ structure with some resolution at ucla.edu (G.M.M.) solved organic carbon efflux to rivers (13).

www.sciencemag.org SCIENCE VOL 303 16 JANUARY 2004 353 R EPORTS During summer field campaigns in 1999, 2000, and 2001 (14), we collected 87 peat cores from Russia’s West Siberian Lowland (WSL), the world’s largest peat- land complex (6) (Fig. 1). These campaigns were directed at previously unstudied peat- lands of the WSL, particularly in perma- frost (15). Radiocarbon dating of peat ma- terial at the base of each core establishes the age of peatland initiation (table S1). These radiocarbon dates, together with a compilation of 139 additional dates gleaned from a variety of published and unpub- lished sources (16), provide a comprehen- sive database of peatland initiation for the entire WSL (Fig. 2). Figures 1 and 2 show that WSL peatlands expanded broadly and rapidly in the early Holocene (11.5 to 9 ka), a period previously thought to be unfavor- able for northern peatland development (1, 3, 4, 6, 9, 10, 17, 18). The rapidity of this expansion directly contradicts a “steady- state” peatland growth model previously theorized for the region (18), and the timing of maximum expansion is coincident with peak values of atmospheric methane con- centration as recorded in Greenland Project 2 (GISP2) and Taylor Dome ice cores (Fig. 2). There are several reasons to believe that Siberian peatlands were a contributing source to the high atmospheric methane concentrations of the early Holocene. First, field experiments confirm that the WSL is a global source of methane today (19, 20). Second, box-model studies of the interpolar methane gradient (an indicator of the lati- Fig. 1. Location of peat cores drilled in this study. Circle diameters are scaled by basal radiocarbon tudinal distribution of methane sources, age. Broad spatial distribution of early Holocene ages [11,500 to 9000 calendar years before computed from the difference between present (11.5 to 9 cal ky BP)] confirms that West Siberian peatlands were widely established during this time of high atmospheric methane concentration. Inset shows figure location and geographic Greenland and Antarctic ice core methane extent of our GIS-based peatland inventory. concentrations) strongly suggest existence of a Northern Hemisphere methane source Ϫ1 (17, 21, 22). Figure 2 shows a sharp in- an estimated 0.3 to 14 Tg year CH4 widespread peatland growth there from crease (88 to 164%, or 30 to 41 Tg year–1) release from the region. Actual emissions ϳ11.5 to 9 ka. Closer inspection of Fig. 2 in Northern Hemisphere emissions during could have been ϳsix times larger (1.8 to finds a robust increase in interpolar meth- the early Holocene Preboreal period (9.5 to 84 Tg yearϪ1), because, unlike today, mac- ane gradient during the Preboreal period 11.5 ka) as compared with the Last Glacial rofossil assemblages in our cores show (9.5 to 11.5 ka) as compared with the Maximum (17, 21) (tropical sources also domination by eutrophic fen species in the Younger Dryas (11.5 to 12.5 ka), requiring rose by 79 to 89%, or 53 to 58 Tg year–1). early Holocene (25, 26). Furthermore, this the “switching on” of ϳ40 to 65% higher This new Northern Hemisphere methane calculation ignores completely the possible (18 to 25 Tg year–1) Northern Hemisphere source was not present during either the development of peatlands in central and methane source in the Preboreal. We be- Younger Dryas (11.5 to 12.5 ka) or eastern Siberia, inclusion of which would lieve we have found one such source in Bølling-Allerød (13.5 to 14.8 ka) periods. more than triple the total peatland area to Siberia. Third, plausible area-based calculations for 3.7 ϫ 106 km2 today (27). Our GIS-based peatland inventory (23) possible early Holocene methane fluxes High-latitude peatlands have been also finds that the WSL peat carbon pool yield large values. Through comprehensive, doubted as an early Holocene methane is substantially larger than previously geographic information system (GIS)– source because of presumed aridity (4), thought, primarily owing to new data from based data inventory (23), we found that terrestrial glaciation (1–3), and the belief ongoing Russian field survey programs and WSL peatlands currently occupy ϳ600,000 that northern peatlands did not expand sub- our extension of coverage beyond that of km2. Inclusion of thin (Ͻ50 cm) stantially until the mid-Holocene (6, 10, 17, earlier inventories. Previous estimates roughly doubles this extent (24). Applying 18). These reasons remain largely true for range from about 40 to 55 Pg C (27, 24). a contemporary flux range (4.2 to 195.3 mg North America. However, it now appears Our compilation, digitization, and spatial Ϫ2 Ϫ1 Ϫ1 ϳ CH4 m day for 120 day year )(19)to improbable that the WSL was ice-covered analysis of 30,000 unpublished Russian an early Holocene WSL landscape contain- during the Last Glacial Maximum (28), and measurements of peatland depth, area, bulk ing just one-half of today’s peatlands yields in any event our radiocarbon dates establish density, and ash content, combined with

354 16 JANUARY 2004 VOL 303 SCIENCE www.sciencemag.org R EPORTS Fig. 2. Timing of WestSiberi- an peatland establishment and atmospheric methane concentrations (top), and northern hemisphere meth- ane emissions derived from the interpolar methane gradi- ent(IPG) ( bottom). Occur- rence frequency (leftvertical axis) of basal peatradiocarbon age ranges (95% confidence level) is plotted with the use of the cumulative probability histogram method (36). At- mospheric methane concen- trations (right vertical axis) are from the GISP2 (dark solid line) and Taylor Dome (light solid line) ice cores (21). Northern Hemisphere contri- butions to the IPG are from Chappellaz et al.(17) (solid symbols), Brook et al.(21) (open symbols), and Dallen- bach et al.(22) (gray sym- bols). Low Northern Hemi- sphere during the Younger Dryas (11.5 to 12.5 ka) and Bølling- Allerød (13.5 to 14.8 ka) indi- cate atmospheric methane concentrations were driven by tropical sources. However, during the Preboreal (shaded area, 9.5 to 11.5 ka), increased methane concentration and Northern Hemisphere emis- sions also coincide with rapid and widespread establishment of Siberian peatlands. our own core, depth, ground-penetrating C yearϪ1 to the , boosting the that West Siberian peatlands have behaved radar, and visible/near-infrared satellite im- present-day rate of atmospheric CO2 in- primarily as a long-term sink of atmospher- agery data, yields a WSL peat carbon pool crease by 0.07 parts per million by volume ic CO2 and net source of CH4 since their estimate of 70.2 Pg C. This value is con- yearϪ1 (ϳ4% faster than the current rise). rapid development in the early Holocene. servative because, like previous investiga- In terms of net greenhouse forcing, the tors (18, 24), we do not consider thin peats warming effect of such a release would References and Notes (Ͻ50 cm) in our inventory. We also assume likely be offset by reduced methane emis- 1. J. P. Kennett, K. G. Cannariato, I. L. Hendy, R. J. Behl, Methane Hydrates in Quaternary : The only 52% peat organic carbon content (after sion (29) but, even in the unlikely event of (American Geophysical loss on ignition), a more conservative value a total shutdown of methane emission, Union, Washington, DC, 2003). than is sometimes used (56 to 57% of total would still attain at least ϳ80% of its 2. J. Chappellaz, J. M. Barnola, D. Raynaud, Y. S. Korot- kevich, C. Lorius, Nature 345, 127 (1990). peat mass) (10). With a minimum stock of greenhouse warming potential (30). Evi- 3. J. Chappellaz et al., Nature 366, 443 (1993). 70.2 Pg C, the WSL represents a substantial dence for a recent slowdown or stoppage in 4. T. Blunier, J. Chappellaz, J. Schwander, B. Stauffer, D. Raynaud, Nature 374, 46 (1995). Holocene sink for atmospheric CO2, aver- WSL peat accumulation does exist (31, 32), Ϫ1 5. J. P. Severinghaus, E. J. Brook, Science 286, 930 aging at least 6.1 Tg C year for the past and detailed studies of contemporary peat (1999). ϳ11.5 ka and storing 7 to 26% of all ter- accumulation rates are needed to confirm or 6. E. Gorham, Ecol. Appl. 1, 182 (1991). restrial carbon accumulated since the Last disprove this possibility. In the present 7. W. M. Post, W. R. Emanuel, P. J. Zinke, A. G. Stangen- berger, Nature 298, 156 (1982). Glacial Maximum (270 to 1050 Pg C) (12). study, we measured fine debris fraction 8. J. W. Harden, E. T. Sundquist, R. F. Stallard, R. K. Mark, Predicting how northern peat carbon (Ͻ150 ␮m) (33) throughout our cores to Science 258, 1921 (1992). stocks may respond to a warming Arctic address past decomposition history. We 9. K. Gajewski, A. Viau, M. Sawada, D. Atkinson, S.Wil- climate is a complex problem that remains found spatially and temporally varying de- son, Global Biogeochem. Cycles 15, 297 (2001). 10. M. S. Botch, K. I. Kobak, T. S. Vinson, T. P. Kolchugina, intractable to date. One scenario is that bris fractions (ϳ25 to 70%) throughout the Global Biogeochem. Cycles 9, 37 (1995). 11. W. C. Oechel, G. L. Vourlitis, Trends Ecol. Evol. 9, 324 warming will fuel an appreciable new CO2 region but little evidence for a sustained source, should currently frozen or water- interval of synchronous oxidation (34). (1994). 12. G. M. Woodwell et al., Clim. Change 40, 494 logged peats experience warmer tempera- Therefore, we doubt occurrence of a past (1998). tures, permafrost degradation, decreased event of massive, region-wide peat oxida- 13. C. Freeman, C. D. Evans, D. T. Monteith, B. Reynolds, water table elevation, and enhanced aerobic tion and CO outgassing. However, our N. Fenner, Nature 412, 785 (2001). 2 14. L. C. Smith et al., Eos 497, 497 (2000). decomposition (6, 11, 12). Such fluxes are evidence for low to moderate decomposi- 15. A total of 58 cores were drilled from frozen peat with potentially large: Assuming no enhanced tion at all depths and latitudes does suggest the use of a gasoline-powered CRREL (Cold Regions carbon uptake by the biosphere or , that most WSL peats have been subjected Research and Engineering Laboratory) permafrost corer. A total of 29 cores from thawed terrain were complete oxidation of WSL peatlands over to varying decomposition, even in current taken with a rotating-sleeve Russian peat corer. Basal the next 500 years would release ϳ140 Tg permafrost regions. We therefore suggest ages were determined by accelerator mass spectro-

www.sciencemag.org SCIENCE VOL 303 16 JANUARY 2004 355 R EPORTS

metry and conventional radiocarbon dating of bulk extreme (and unlikely) scenario, assuming (i) com- species assemblage on f 150, we visually assessed peatsamples from thelowestvisually apparentpeat plete oxidation of all WSL peat carbon stocks in the macrofossil taxonomies (as percentage of total) with

horizon in each core. Substantially older radiocarbon next 500 years, (ii) a complete shutdown of all WSL the use of light microscopy. Regression of (1 – f 150) ages from organic-rich gytjja (mineral substrate) methane emission during the same 500 years, (iii) a with core depth and macrofossil abundances within

were excluded from new data presented in this paper. 2.5-fold greater infrared absorptivity of CH4 relative five functional or taxonomic groups (Sphagnum, Samples were processed by Beta Analytic, Incorpo- toCO2 [because of the shorter lifetime of CH4 in the brown moss, sedge, herbacious, and woody plants) rated (Miami, FL) and calibrated with the use of atmosphere, the greenhouse of CH4 shows that less than 12% of the linear variance in (1 ϳ 2 ϭ CALIB 4.3 (35) and the Intcal98 data set. is only 2.5 times that of CO2 when integrated over – f 150) is associated with these factors (r 0.114; 16. X. Kremenetski et al., Quat. Sci. Rev. 22, 703 (2003). a 500-year time scale (29)], and (iv) high WSL meth- P Ͻ 0.0001), allowing the assumption of external ϳ Ϫ1 17. J. Chappellaz et al., J. Geophys. Res. 102, 15987 (1997). ane flux atpresent( 14 Tg CH4 year , calculated influence (i.e., changing acrotelm temperatures and 2 ϫ Ϫ1 ϫ 18. M. I. Neustadt, Seria Geogr. 1, 21 (1971). as 600,000 km 120 day year 195.3 mg CH4 water table position) on f 150. Ϫ2 Ϫ1 ϳ ϫ 12 19. J. Heyer, U. Berger, I. L. Kuzin, Tellus 54B, 231 m day ), the resulting release of 11.7 10 35. M. Stuiver et al., Radiocarbon 40, 1041 (1998); avail- 1 (2002). mol CO2 year (from oxidation) would be effectively able at http://radiocarbon.pa.qub.ac.uk/calib. ϫ 12 Ϫ1 20. E. A. Oberlander et al., J. Geophys. Res. 107(D14) , reduced by 2.2 10 mol CO2 year (the radiative 36. I. D. Campbell, C. Campbell, Z. Yu, D. H. Vitt, M. J. ϫ 11 Ϫ1 article no. 4206 (2002). equivalentof theloss of 8.7 10 mol CH4 year Apps, Quat. Res. 54, 155 (2001). emission), or only ϳ18.8%. 21. E. J. Brook, S. Harder, J. Severinghaus, E. J. Steig, 37. We thank F. S. Chapin III (University of Alaska, Fair- C. M. Sucher, Global Biogeochem. Cycles 14, 559 31. D. Peteet, A. Andreev, W. Bardeen, F. Mistretta, banks), S. E. Trumbore (University of California, Ir- (2000). Boreas 27, 115 (1998). vine), and two anonymous readers for constructive 22. A. Dallenbach et al., Geophys. Res. Lett. 27, 1005 32. J. Turunen, T. Tahvanainen, K. Tolonen, A. Pitkanen, reviews. Research funding was provided by NSF (2000). Global Biogeochem. Cycles 15, 285 (2001). through the Russian-American Initiative on Shelf- 23. Our estimates of WSL peatland extent (592,440 km2) 33. T. J. Malterer, E. S. Verry, J. Erjavec, Soil Sci. Soc. Land Environments of the Arctic (RAISE) of the Arctic and pool (70.2 Pg C) have their basis in Am. J. 56, 1200 (1992). System Science Program (ARCSS). The research pre- a comprehensive GIS-based inventory of all peatlands 34. Fine debris fraction (percentage of peat fragments sented here was developed jointly by L.C.S. and throughout the region. Data sources assimilated are Յ150 ␮m) was measured with the use of the follow- G.M.M. Numerous Russian scientists, graduate stu- (i) an archive of printed data reports based on ϳ40 ing variation of the ASTM (American Society for dents, and government officials are thanked for their years of field surveys by Geoltorfrazvedka, Moscow, Testing Materials) Fiber Weight Method (33). Two- invaluable logistical and scientific support during the beginning in the 1950s, with updates by the Russian mL peatsubsamples were takenevery 10 cm Siberian field campaigns. Ministry of Natural Resources in the 1990s. These throughout all cores, soaked for 12 hours in a dis- reports contain detailed maps with associated field persing agent (5% sodium hexametaphosphate) and Supporting Online Material and laboratory measurements of peatland depth, washed through a 150-␮m sieve. Coarse debris con- www.sciencemag.org/cgi/content/full/303/5656/353/ DC1 area, bulk density, and ash content for 9691 peat- tent ( f 150) was computed as the ratio of oven-dried lands throughout the WSL, for a total of 29,350 mass (100°C) of this fraction relative to that of a Table S1 measurements digitized. (ii) Our own field data col- complete sample. Percent fine fraction was comput- lected from 1999, 2000, and 2001, including peat ed as (1 – f 150). To identify the possible influence of 18 August2003; accepted1 December 2003 physical properties from 87 cores, 78 additional mea- surements of peatland depth, and 16 ground-penetrat- ing radar transects. (iii) A Russian typological wet- land map extending coverage beyond that of the Geoltorfrazvedka reports. (iv) Visible/near-infrared Community Assembly Through MSU-SK satellite images from the Russian RESURS- 01 satellite. (v) 62 additional depth measurements gleaned from the published literature. Because of Adaptive Radiation in the flat terrain, WSL peatlands are broadly expan- sive with little depth variation. Ordinary kriging of all available depth data therefore allowed reason- Hawaiian Spiders able interpolation of missing values for 3904 of 9691 peatlands inventoried. All data were digitized Rosemary Gillespie and incorporated into a ARC 8.2 GIS database and together cover the entire WSL. Total carbon pool Communities arising through adaptive radiation are generally regarded as (CP) was computed as n unique, with speciation and adaptation being quite different from immigration ϭ ͩ͸ ⅐ ⅐ ⅐ ⅐ ͪ and ecological assortment. Here, I use the chronological arrangement of the CP Ai Di ri [(100-ai)/100] c i ϭ 1 Hawaiian Islands to visualize snapshots of evolutionary history and stages of where n is the number of digitized peat polygons community assembly. Analysis of an adaptive radiation of habitat-associated, (9691) and Ai, Di, ri, ai, and c representpeatland area, mean depth, depth-averaged bulk density, polychromatic spiders shows that (i) species assembly is not random; (ii) within depth-averaged ash content, and organic carbon any community, similar sets of ecomorphs arise through both dispersal and fraction (52%), respectively. evolution; and (iii) species assembly is dynamic with maximum species numbers 24. S. P. Yefremov, T. T. Yefremova, in West Siberian Peat- lands and : Past and Present, S. V. Vasiliev, in communities of intermediate age. The similar patterns of species accumu- A. A. Titlyanova, A. A. Velichko, eds. (Agenstvo Sibprint, lation through evolutionary and ecological processes suggest universal princi- Novosibirsk, Russia, 2001), pp. 148–151. ples underlie community assembly. 25. The lower (oldest) sections of our cores are dominated by fen species Scorpidium sp., Calliergon sp., and Drepanocladus sp., with later succession to species Community assembly has intrigued biologists comprising closely related organisms—a Sphagnum fuscum and S. angustifolium that dominate for decades (1), leading to a sophisticated feature that has produced insights into evo- the WSL ecology today. Modern transfer function–de- understanding of the ecological parameters lutionary patterns of species composition rived methane fluxes for these fen species average that dictate community membership (2). The and number (5). In particular, two patterns about six times higher than for these bog species (26), suggesting substantially higher WSL methane emission role of evolution in shaping communities is have emerged: (i) A predictable number of in the early Holocene. also well appreciated (3, 4), although the species can exist for a given area on remote 26. J. L. Bubier, T. R. Moore, S. Juggins, Ecology 76, 677 steps through which communities are islands, apparently driven by higher rates (1995). 27. S. E. Vompersky et al., Pochvovedenie 12, 7 (1994). assembled as a result of evolutionary pro- of speciation on larger islands (6); and (ii) 28. S. L. Forman, O. Ingolfsson, V. Gataullin, W. F. Manley, cesses have been enigmatic. Adaptive radi- the end-product of adaptive radiation is H. Lokrantz, Geology 27, 807 (1999). ations on remote islands provide opportu- often a nonrandom set of species, with 29. G. J. Whiting, J. P. Chanton, Tellus 53B, 521 (2001). nities to study multiple communities, each lineages diversified in such a way that the 30. Desiccation of WSL peat carbon stocks would elevate same set of ecomorph types occur on each atmospheric CO2 concentrations through peat oxida- tion but reduce CH4 emissions through associated Division of InsectBiology, Universityof California, island (7, 8). These findings suggest that drying. Because of high temporal and spatial variabil- 201 Wellman Hall, Berkeley, CA 94720–3112, USA. evolution on more remote islands can act in ity in contemporary rates of CH4 emission, itis difficult to estimate the potential net radiative bal- *To whom correspondence should be addressed. E- a manner analogous to immigration on less ance of these opposing effects. However, even in an mail: [email protected] remote islands, giving rise to similar spe-

356 16 JANUARY 2004 VOL 303 SCIENCE www.sciencemag.org