Atmosphere (Academic Press, New York,1977).

6. WMO/ICSU, Report of the Joint OrganizingCommittee Study Conference on Climate Mod-els: Performance, Intercomparison and Sensi-tivity Studies [GARP (Global Atmospheric Re-search Program) Publication Series 22, WorldMeteorological Organization, Geneva, 1979],vol. 1.

time.

A number of mass extinctions have"'reset" major parts of the evolutionarysystem during the Phanerozoic. Howev-er, the precise timing and magnitude ofthese events has been difficult to mea-

sure because data from the fossil recordare fragmentary. Comprehensive and ac-curate data on extinct species have al-ways been unobtainable, and thereforemost workers have been forced to inves-tigate extinctions at the level of genera,families, and orders, with family-leveldata generally preferred as the best com-promise between sampling limitationsand taxonomic uncertainty (1). Histori-cally, the three best summaries of famil-ial data from the fossil record have beenthose of Newell (2), Cutbill and Funnell(3), and Valentine (4). But even withthese data sets, identification of specificmass extinctions has been difficult andoften subjective because of taxonomicproblems and especially stratigraphic im-precision. Many macroevolutionary phe-nomena including mass extinctions havecharacteristic time scales that are geo-logically rather short (less than severaltens of millions years) and can becomelost or grossly distorted when analyzedwithout adequate stratigraphic control.We now present a new analysis of

extinctions based on a more comprehen-sive and accurate data set for marineanimal families. Marine vertebrates aswell as invertebrates and protozoans are

included, and the data benefit from com-pilation of taxonomic and stratigraphicinvestigations far beyond traditionalsources (5). The compilation encom-

passes approximately 3300 fossil marinefamilies, ofwhich about 2400 are extinct.Times of extinction for 87 percent of thefamilies have been resolved to the levelof the stratigraphic stage (mean duration,7.4 x 106 years), and most of the re-

maining data has been resolved to strati-

SCIENCE, VOL. 215, 19 MARCH 1982

7. 1. Shukla, D. Randall, D. Straus, Y. Sud, L.Marx, Winter and Summer Simulations with theGLAS Climate Model (National Aeronauticsand Space Administration/Goddard Space FlightCenter, Greenbelt, Md., in press).

8. J. Charney, W. J. Quirk, S. H. Chow, J. Korn-field, J. Atmos. Sci. 34, 1366 (1977).

2 December 1981

graphic series (mean duration, 20 x 106years).The rates of extinction calculated from

the familial data plotted against geologictime are illustrated in Fig. 1. Each pointwas calculated as follows: the number offamilies that became extinct in each ofthe 76 post-Tommotian (early LowerCambrian) stages (6) was divided by theestimated duration of the stage (7); theseinitial rates were then modified by add-ing extinction rates calculated from thelower resolution series-level data to theappropriate stages. Calculations weremade separately for "shelly" taxa andfor rarely preserved taxa (8). The effectof this segregation was negligible in mostcases so that the data for rarely pre-served animals are not included withmost points in Fig. 1. For four stages,however, addition of rarely preservedfamilies increased calculated extinctionrates by more than 0.5 family per million

years. These are the stages that containthe four major Lagerstatten of the Phan-erozoic marine record: Burgess Shale(Cambrian, Templetonian), HunsruckShale (Devonian, Siegenian), MazonCreek concretions (Carboniferous, Mos-covian), and Solnhofen Limestone (Ju-rassic, "Tithonian"). The combinedrates for shelly and rarely preserved fam-ilies for these four stages are indicated inFig. 1 by X's with the rates for shellyfamilies shown below. Only the BurgessShale (Templetonian) stands out on theplot.The distribution of the 76 points for

shelly animals in Fig. 1 suggests that tworates of extinction have been operativethrough the Phanerozoic. (i) Normal, orbackground, extinction: the majority ofpoints fall in a rather tight cluster atextinction rates less than 8.0 extinctionsper million years. (ii) Mass extinction:several points stand out as being consid-erably higher than the background andshow a maximum of 19.3 familial extinc-tions per million years.The problem of determining rigorously

which points in Fig. 1 should be consid-ered mass extinctions can be approachedas a simple data analysis problem ofidentifying trends and outliers. As aninitial step, we computed a linear regres-sion (not shown) for all 76 extinctionpoints as a function of geologic time andthen searched for significant departuresfrom this line. Four points (or 5 percentof the data) fell above the one-sided 99percent confidence interval. Thesepoints, which are circled in Fig. 1, are(per million years) the Ashgillian (19.3fm), Guadalupian (14.0 fm), Dzhulfian(15.7 fm), and Maestrichtian (16.3 fm). Afifth point, the Norian (10.8 fm), fell

20- ASHG Fig. 1. Total extinction rate(extinctions per million years)

DZHULF MAEST through time for families of

marine invertebrates and ver-

q 15 -UAD' tebrates. The plot shows sta-U~ ~ ~ ~ ~ ~ GAC tistically significant mass ex-

TEM NOR tinctions late in the Ordovician10FRAS (ASHG), Permian (GUAD-GIV .AFAME l TITHDZHULF), Triassic (NOR),L~~~r-___EIE-x- | _ l LI xand Cretaceous (MAEST). An

o * . aG; - -- - - - - a extinction event in the late.De-5s - _ . * .. . - vonian (GIV-FRAS-FAME) is

noticeable but not statisticallysignificant. Circled points are

those where the departureo - Xe I eX s XD I CI P Ili | I K T - from the main cluster is highly600 400 200 ° significant (P < .01); X's indi-

Geological time (10's years) cate those cases where inclu-sion of rarely preserved animal groups substantially increases the calculated extinction rate (thepoint directly below the X is the rate calculated without the rarely preserved groups). The figurealso shows a general decline in background extinction rate through time. The regression line isfit to the 67 points having extinction rates less than eight families per 106 years, and the dashedlines define the 95 percent confidence band for the regression. Abbreviations: TEM, Temple-tonian; ASHG, Ashgillian; SIEG, Siegenian; GIV, Givetian; FRAS, Frasnian; FAME, Famen-nian; MOSC, Moscovian; GUAD, Guadalupian; DZHULF, Dzhulfian; NOR, Norian; TITH,Tithonian; MAEST, Maestrichtian.

0036-8075/82/0319-1501$01.00/0 Copyright 1982 AAAS

Mass Extinctions in the Marine Fossil RecordAbstract. A new compilation offossil data on invertebrate and vertebrate families

indicates thatfour mass extinctions in the marine realm are statistically distinctfrombackground extinction levels. These four occurred late in the Ordovician, Permian,Triassic, and Cretaceous periods. A fifth extinction event in the Devonian stands outfrom the background but is not statistically significant in these data. Backgroundextinction rates appear to have declined since Cambrian time, which is consistentwith the prediction that optimization offitness should increase through evolutionary

1501

on

Oct

ober

8, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Oct

ober

8, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Oct

ober

8, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

900

a0

E

0

z

400 200Geological time (106 years)

above the one-sided 95 percent confi-dence interval. Some or all of these massextinctions have been recognized previ-ously but without consistency and with-out statistical testing.The outlying points identified above

also can be recognized as major pertur-bations in marine diversity. The Pha-nerozoic diversity curve compiled fromthe familial data is shown in Fig. 2. Fiveextinction events are seen as sharp dropsin standing diversity. Four of these(counting the Guadalupian and Dzhulfianas a single event) match the statisticallysignificant outliers in Fig. 1. The fifth,labeled "2" in Fig. 2, is a late Devonianextinction that has been recognized byprevious workers. This extinction doesnot appear as a statistically significantevent in Fig. 1 because the family extinc-tions are distributed over two stages, theFrasnian and the preceding Givetian,which have a combined duration ofabout 15 million years (9). This smearingof extinctions may represent samplingerror in that failure to identify the actualtime of extinction will almost alwayspush apparent extinctions backward intime. Alternatively, the smearing mayreflect a real phenomenon-an extinc-tion "event" that took place over mil-lions of years. The continuation of highextinction rates into the Famennian isconsistent with this hypothesis. Howev-er, it should be noted that, on the basis ofother information, McLaren (10) sug-gested a meteorite impact as one possi-ble explanation for the Frasnian extinc-tions.

In summary, five mass extinctions are

clearly defined iri the familial data. Theseextinctions occurred in the Late Ordovi-cian (Ashgillian), Late Devonian (Give-tian-Frasnian), Late Permian (Guadalu-pian-Dzhulfian), Late Triassic (Norian),and Late Cretaceous (Maestrichtian).The occurrence of these major extinc-tions near the ends of geologic periodssimply reflects the fact that the strati-

1502

Fig. 2. Standing diversitythrough time for families ofmarine vertebrates and inver-tebrates. Rarely preservedgroups are not included. Fivemass extinctions, indicated bynumerals, are recognizable by

/5 abrupt drops in the diversitycurve. The relative magni-tudes of these drops (mea-sured from the stage before tothe stage after the extinctionevent) are given in parenthesesin the upper left. All mass ex-tinctions but No. 2 (Devonian)

I| EIy T are statistically significant inFig. I and three (Nos. 1, 3, and

0 5) are highly significant(P < .01).

graphers who established the geologictime scale in the first half of the 19thcentury chose major faunal breaks asboundaries for the principal subdivi-sions.With the major Phanerozoic events

isolated in Fig. 1, a more accurate as-sessment of the nature of backgroundextinction can be made. Although somesmaller but well-known extinctionevents may remain hidden in Fig. 1 (11),the residual cluster of points suggeststhat background rates have been declin-ing since the early Paleozoic. The solidline in Fig. 1 is a linear regression fittedto the 67 extinction rates for shelly ani-mals after removal of the major extinc-tion events; the dashed lines, which en-velop nearly all these points, representthe 95 percent confidence band for theregression. The correlation coefficientfor the regression is .47, which can beconsidered statistically significant ifproblems of time series and data selec-tion are ignored. The slope of the regres-sion line is nontrivial and indicates thatthe total rate of background extinctionhas decreased from about 4.6 to 2.0 fmper million years since the Early Cambri-an. This is surprising in view of the factthat the rates are not normalized forstanding diversity, which has increasedsubstantially since the Cambrian (Fig. 2).The decline in extinction rates could bejust an artifact of the "pull of the Re-cent" (12). In contrast, a decrease inextinction rate is predictable from firstprinciples if one argues that general opti-mization of fitness through evolutionarytime should lead to prolonged survival.This is speculative but it is worthy offurther consideration because broad pre-dictions of progressive change in evolu-tionary dynamics are so rarely realizedwhen tested with data.The decline in background extinction

rate from the Early Cambrian to theRecent means that approximately 710family extinctions did not occur that

would have if the Cambrian rate hadbeen sustained. This number is essential-ly identical to the amount by which fa-milial diversity increased over that inter-val (680 families) (Fig. 2). This suggeststhat the net increase in standing diversitythrough the Phanerozoic may have beenmore an effect of decrease in extinctionthan increase in origination.

In conclusion, our analysis shows thatmajor mass extinctions are far more dis-tinct from background extinction thanhas been indicated by previous analysesof other data sets. Four mass extinctionsare statistically significant events and arelikely to represent phenomena qualita-tively different from the background.The data do not tell us, of course, whatstresses caused the mass extinctions.The extinctions were short-lived eventsin geological time, but the data do nothave the resolving power to show wheth-er the events were also short-lived inhuman or ecological time.

DAVID M. RAuPField Museum ofNatural History,Chicago, Illinois 60605

J. JOHN SEPKOSKI, JR.Department of Geophysical Sciences,University of Chicago,Chicago 60637

References and Notes

1. J. W. Valentine, J. Paleontol. 48, 549 (1974).2. N. D. Newell, Sci. Am. 208, 77 (1963); Geol.

Soc. Am. Spec. Pap. 89, 63 (1967).3. J. L. Cutbill and B. M. Funnel, in The Fossil

Record, W. B. Harland et al., Eds. (GeologicalSociety of London, London, 1967), p. 791.

4. J. W. Valentine, Palaeontology 12, 684 (1969).5. The data were compiled from R. C. Moore et

al., Eds. Treatise on Invertebrate Paleontology(Geological Society of America and Univ. ofKansas Press, Lawrence, 1953-1979); W. B.Harland et al., Eds., The Fossil Record (Geolog-ical Society of London, London, 1967); A. S.Romer, Vertebrate Paleontology (Univ. of Chi-cago Press, Chicago, 1966); also, 380 additionalpapers and monographs. A complete listing ofthese data is scheduled to appear in J. J. Sep-koski, Jr., Milwaukee Pub. Mus. Contrib. Biol.Geol., in press.

6. Vendian and Tommotian points were excludedfrom the analysis because these intervals haveexceptionally low diversities and therefore ex-ceptionally low extinction rates.

7. The geologic time scale used is a compositebased on a number of recently published stage-level time scales and differs only slightly fromthat used by J. J. Sepkoski, Jr., Paleobiology 5,222 (1979).

8. Four general kinds of animals were consideredto have low fossilization potential: (i) soft-bod-ied animals without mineralized skeletons (forexample: Nemertina, Priapulida, and Sipuncu-lida as well as many Hydrozoa, Scyphozoa, andPolychaeta); (ii) animals with lightly sclerotizedskeletons (such as many Crustacea); (iii) rarelyreported fossil animals with multielement skele-tons that dissociate rapidly after death (such asOctocorallia and Holothuroidea as well as someAsterozoa and Osteichthyes); and (iv) deep-seaanimals with extremely poor fossil records (suchas some Crinoidea, Chondrichthyes, and Os-teichthyes).

9. The Frasnian might still be considered statisti-cally distinct in Fig. 1. If each point in that figureis considered an independent event, then theprobability that three of the nine highest pointswould be clustered about the Frasnian point isquite low (P = .002).

10. D. J. McLaren, J. Paleontol. 44, 801 (1970).11. "Minor" mass extinctions, which do not appear

as noticeable perturbations in Figs. I and 2,

SCIENCE, VOL. 215

include at least three to five events in theCambrian [A. R. Palmer, J. Paleontol. 39, 149(1965); M. E. Taylor, ibid. 42, 1319 (1968); J. H.Stitt, ibid. 45, 178 (1971)], an Early Jurassic(Toarcian) event [A. Hallam, Paleobiology 3, 58(1977)], a terminal Eocene event [H. Tappan andA. R. Loeblich, Geol. Soc. Am. Spec. Pap. 127,247 (1971); A. G. Fischer and M. A. Arthur,Soc. Econ. Paleontol. Mineral 25, 19 (1977)],

major oil spill could expose significantcontamination.

The California population of the seaotter [Enhydra lutris nereis (Merriam)]was listed as "threatened" in 1977 (1)pursuant to the Endangered Species Actof 1973 (2). The listing was based on thepossibility that a major oil spill couldoccur within the sea otter range andcould kill a significant portion of thepopulation, placing it in danger of extinc-tion. This concern arises from the knownsensitivity of the species to oil contami-nation (3). A spill of gasoline and dieseloil nearshore in the Kurile Islands,U.S.S.R., spread through 40 km ofcoastline and killed over 100 sea otters(4). Concern for the status of the Califor-nia sea otter is heightened by the lackof evidence of significant populationgrowth since 1973 (5).

It is difficult to project the critical day-to-day movements of floating oil near thesea otter range on the basis of existingoceanographic data. Surface current pat-terns off central California (San Francis-co to Point Conception) have been exam-ined with several techniques (6-9). Theprincipal result is the description ofmean flow patterns on a seasonal scale.However, studies of drogues and remoteimagery have shown that short-term de-partures from mean seasonal drift maybe frequent in the California current sys-tem (6, 7, 10, 11). Such departures in-volve tidal oscillations and mesoscalemeanders and eddies (6, 7, 10, 11). Theprediction of the direction of drift offloating oil is further complicated by thedominant role of wind stress at the air-sea interface (12). As far as we know,there are no records of major oil spills offcentral California on which to base pre-dictions of oil drift. We know of nopublished studies of day-to-day move-ments of other floating materials off cen-tral California over an appropriatelysmall time scale.

SCIENCE, VOL. 215, 19 MARCH 1982

and possible events in the latest Jurassic (Tith-onian) and early Late Cretaceous (Cenomanian)(see A. G. Fisher and M. A. Arthur, above).

12. D. M. Raup, Carnegie Mus. Nat. Hist. Bull. 13,85 (1979).

13. We thank R. K. Bambach for critical commentson this work.

28 September 1981; revised 31 November 1981

numbers of California sea otters to oil

In this report we describe the move-ments and beaching of a large volume oflumber spilled off central California inthe winter of 1978. Floating materialssuch as drift cards and plastic sheetshave been used successfully by others in

P

Santa Cruz

Monterey'acific Ocean Bay a Moss

16 FOb..; Landing16 Feb 4

* ontereyYiJ 14 Feb Point Lobos

Spill site Garrapata Creek12 Feb. Point Sur

17 Feb. n Torre Canyon

Lopez Point23 Mar...

Cape San Martin19 Feb.Salmon Creek

i' 21 feb .3Point Piedras2t .-. ~~~Blancas

o Cambria24 Feb* Point Estero

Estero Bay. Morro_Bay11 Mar. e slay Creek

Diablo CanyonPort San Luis Pismo Beach

vpOceano

T oint Sal25 Feb. / California

(Pit Arguello

[Ioint Conception

Santa Barbara Channel SantaxwBarbara

San Miguel 0

Island

30 km

a Major floating patch* Other floating patches

Beached lumber8*e otter rangoe

modeling the movements of oil on thesea surface (13). Our data provide a firstapproximation of the disposition of thefloating component of a large oil spilloccurring under similar conditions ofweather and sea. Information of this kindis needed if we are to understand thepotential impacts of oil spills on theCalifornia sea otter population and todevelop management plans for improv-ing the status of the population, nownumbering about 1800 animals (5).A cargo of 2 x 106 board feet of fin-

ished lumber (volume equivalent to2.9 x 104 barrels of oil) spilled from abarge under tow in heavy weather 40 kmwest of Point Sur on 12 February 1978(14). The spilled lumber was hazardousto navigation (15) and was thereforemonitored by aircraft and merchant ves-sels in subsequent weeks (16). We com-piled a record of observations of thefloating lumber at sea along with sight-ings of lumber washed ashore after thespill.Much of the spilled lumber remained

in a single large patch that moved firsttoward the coast and then southeast-ward, parallel to the shoreline, duringthe first 10 days after the spill (Fig. 1).By 24 February the major patch waswithin 7 km of shore near Point Esteroand remained relatively close to shorefor the balance of the observation period(through March). Other patches of float-ing lumber were seen off Monterey,Point Lobos, Cape San Martin, andPoint Arguello during the survey period.Beached lumber was found throughouttwo sections of coastline within the seaotter range, a northern section of about

Fig. 1. Sightings of the major patch andsmaller patches of floating lumber and areasof significant lumber beaching in and near therange of the sea otter population in California.Observations were made after the spillage of2 x 106 board feet of lumber off Point Sur on12 February 1978. All sightings of beachedlumber were recorded between 12 Februaryand 31 March 1978. The times and the posi-tions corresponding to each sighting of themajor patch of floating lumber are as follows:12 February, 0650 GMT 36°15'N, 122°25'W(spill site); 14 February, 1930 GMT, 36°19'N,122°07'W; 17 February, 0130 GMT, 35°55'N.121°51'W; 19 February, 2030 GMT, 35°40'N,121°37'W; 21 February, 0136 GMT, 35°29'N,121°29'W; 24 February, 2100 GMT, 35°27'N,121°05'W; 8 March, 1816 GMT, 35°30'N,121°16'W; 11 March, 2006 GMT, 35°18'N,120°58'W. Mean wind speed (meters per sec-ond) and the direction (relative to true north)between sightings of the major floating patch,during the period when the patch was closestto shore, were as follows: 21 to 24 February,3.57, 317°; 24 February to 8 March, 0.90, 154°;8 to 11 March, 4.30, 3080 [wind data are fromthe shore station at Point Piedras Blancas (18,19)].

0036-8075/82/0319-1503$01.00/0 Copyright ©) 1982 AAAS

Lumber Spill in Central California Waters:Implications for Oil Spills and Sea Otters

Abstract. A large quantity oflumber was spilled in the ocean offcentral Californiaduring the winter of 1978, and it spread through most of the range of the threatenedCalifornia sea otter population within 4 weeks. The movement rates of lumber weresimilar to those of oil slicks observed elsewhere. These observations indicate that a

1503