NEST FAILURES IN THE FULMAR: THE EFFECT OF OBSERVERS BY JANETC. OLLASONAND G. M. DUNNET INTRODUCTION Numerous seabirdstudies have dealt with the factorsaffecting breed- ing success(e.g., Coulson,1900; Mills, 1973;Davis, 1970; Brooke, 1978; Ollasonand Dunnet, 1978). Complementaryto thesestudies is the in- vestigationof causesand timing of breeding failures. Unfortunately it is impossiblein practiceto examine natural lossesof eggsor chicksin isolationfrom unnatural ones becausethe observer'spresence is likely to increasethe losses,i.e., causedisturbance. However, the stagein the breeding cycle (egg or chick) at which lossesare more likely to occur and their timing in relation to date can be determined. In the Fulmar (Fulmarusglacialis) Dunnet et al. (1903) suggestthat mostfailures occurat the egg stage:of all egg lossesin the first 9 days after laying,71% occurwithin the first 3 days.Mougin (1907) showstwo peaksin egg losses,one shortlyafter laying,the other around hatching. Ollasonand Dunnet (1978) demonstratethat late-laid eggsare more likely to fail. Breeding successin the Fulmar can vary widely from year to year, e.g., on the islandof Eynhallowin Orkney, Dunnet et al. (1979) quote a range of 16% to 52% over 28 years,although no known changeoc- curred in conditions on the island. This does not account for variations betweenyears in observereffort, which will now be consideredin detail. The aim of this paper is to determine for the Fulmar the most vul- nerablestage of the breeding cycle,to describethe pattern of egg losses in relation to date, and to examine the effects of observersstudying breeding biology on the responseof breeding adults and their subse- quent breeding success.Causes of failure for a generalizedseabird will be describedtheoretically, and the Fulmar fitted into this framework. METHODS In 1950, R. Carrick and G.M.D. began color-bandingbreeding Ful- mars caught from their nestson the small uninhabited island of Eyn- hallow, Orkney. Observationsduring the breeding seasonhave contin- ued annually to the present time. General methodshave already been described(Dunnet and Ollason, 1978). The initial aims of the study were to collectdata on breeding and survival.Only recentlywas it re- alized that we have useful data on the effects of disturbance. The data analyzedhere were collectedover 21 years from 1958 to 1978. Three annual vistswere taken to Eynhallow:in May (to count eggs),in July (to determine hatchingsuccess), and in August (to deter- mine fledgingsuccess). 40] j. c. Ollasonand G. M. Dunnet J.Field Ornithol. Winter 1980 70-- 60-- E 30-- z 20-- 10-- I 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 May Date June FIGURE1. Average number of eggslaid in 1960, 1961, 1962, and 1978. Visit in May In 1960, 1961, 1962, and 1978 the May visit lasted for about three weeksduring laying.Nests were checkedas often as possible(usually daily) to obtain laying datesand the number of eggslaid (apart from any eggslaid and lost betweenconsecutive checks of nests).The mean laying datesin theseyears ranged from 21.6 May in 1961 to 23.5 May in 1962 (Variance Ratio F = 7.61, df = 3.498, P < 0.001). However, it was felt that combiningthe data from all four yearsshould produce a distribution of laying dates typical of an average year (Fig. 1). In the remaining 17 years (1958, 1959, and 1963 to 1977), Eynhallow was visited during the fourth quarter of the laying distribution for 2 to 6 days beginning between 26 May and 8 June. The number of eggs ob- servedin eachof theseyears is therefore lessthan the total laid (because some egg losseswould have occurred before the first annual visit); it consistsonly of the number of eggspresent on the first day of the visit, plus any laid subsequently.A correction therefore needs to be applied to the observedtotals of these 17 years to allow for egg lossesbefore the first day of the May visit. Using the data from 1960, 1961, 1962, and 1978, the pattern of eggslaid and eggslost during laying has been determined for an averageyear: Figure 2 shows(a) the number of eggsto be laid after a particular day, expressedas a pro- Vol.5l, •4o.1 ObserverEffect on Nesting Fulmars [41 A 1'01 0000 0 0'8-- 0 0 B 0'4-- ß o 0'2• 0000 0 0• 0000 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 1 May Date June F•GURE2. Patternof egg-layingand egg-lossduring the laying period of an average year: (A) eggslaid after day x as a proportionof eggspresent on day x pluseggs laid after day x; (B) eggslost by day x as a proportionof eggspresent on day x. Data from 1960, 1961, 1962, and 1978. portion of those present on that day plus those laid subsequently,and (b) the number of eggslost by a particular day, expressedas a propor- tion of those present on that day. The totals were adjustedas follows: the observedtotal was multiplied by the proportion (a) referring to the first day of the May visitto give an estimateof the number of eggslaid after the first day of the visit.Subtracting this number from the observed total givesan estimateof the number of eggspresent on the first day, which is then multiplied by the relevant proportion (b) referring to the first day of the May visit to give the estimatednumber of eggslost by that day. Adding the three estimatesgives an adjustedtotal for each of the 17 yearswhich can then be comparedwith eachother and with the observedtotals of 1960, 1961, 1962, and 1978. No significantdifference was found betweenthe distributionsof the observedand adjustedtotals 42] J. c. Ollasonand G. M. Dunnet j. FieldOrnithol. Winter 1980 (Kolmogorov-Smirnovtest X22 = 5.29, NS) and the overallbreeding suc- cess(percentage eggs that fledge) calculatedfrom the observedtotals is highly correlated (% = 0.961, df = 17, P < 0.001) with that calculated from the adjustedtotals. The observednumber of eggspresent on the firstday is not significantlydifferent from the estimatednumber of eggs present on the first day (•: = 144.9 and 148.8; SD = 36.28 and 39.39, n = 16) and they are highly correlated(r = 0.988, P < 0.001). There- fore in all subsequentanalyses in this paper, the adjustedtotal number of eggshas been used for 1958, 1959, and 1963 to 1977. Visitin July In all years of the study, a secondvisit to Eynhallowof three or four daysoccurred in July, coveringpart of the hatchingperiod. By the end of'this visit most eggs had hatched.However, some eggs remained, a few of whichmay havehatched but failed beforefledging. Hatching success will therefore tend to be slightlyunderestimated. Visitin August The final annualvisit, of one day, occurredduring August,by which time almostall adults had left the area. All nestswith fledglingswere recorded,and therefore a good estimateof overall breeding success couldbe determined.Since the estimateof hatchingsuccess will be less accuratethan that of overall breeding success,all analysesinvolving hatchingsuccess have also been carriedout on overallbreeding success. In 1978, the responseof Fulmarson an egg or a chick to the stan- dardizedapproach of one particularobserver (J.C.O.) wascategorized as follows:"on"--bird remainedon the egg or chickall the time while the observerwas visible; "at"--bird left its egg or chickbut did not fly away and usuallystayed within a few inchesof its nest site; and "off"-- bird left its egg or chick and flew away. For each nest observedmore than five times,an index of remaining"on" the nestwas determined by observingthe number of times that the parent(s) remained "on" the nest and expressingthis as a proportion of the total observationsof parentsat that nest, while it had an egg or a chick. Nestswith fewer than six observations (55% of the total 215 nests) were eliminated to avoidthe biasesof smallsamples. This procedurewill alsohave removed nestswith birds very sensitiveto disturbance,i.e., neststhat failed after five or fewer disturbances.Thus the index of remaining "on" the nest cannot include the most sensitiveparents. The data werecoded and storedon magneticdisc in a database (using the AberdeenUniversity Data BaseManagement System). Most analyses were carried out using a Honeywell Level 66 computer. RESULTS TheLaying Period: Eggs Laid and EggsLost The combinedlaying distributionsfor 1960, 1961, 1962, and 1978 are assumedto be typicalof an averageyear (• -- 22.7 May, SD = 3.40, Vol. 51, No. 1 ObserverEffect on Nesting Fulmars [43 T•,BI, F• 1. Mean percentagefailures before hatchingfor varioussets of years. Mean SD n d P (a) Years with long May visit = 66.90 3.289 4 1960 to 1962 and 1978 NS (b) Years with short May visit = 65.78 8.624 17 1958 and 1959, 1963 to 1977 (c) Yearswith large group in 71.79 7.007 8 July = 1971 to 1978 3.119 <0.01 (d) Years with small group in 62.42 6.115 13 July = 1958 to 1970 (e) Years with large group in 71.79 7.568 7 July excluding years with long May visit = 1971 to 1977 2.854 <0.02 (f) Yearswith smallgroup in 61.57 6.804 10 July excluding years with long May visit = 1958 and 1959, 1963 to 1970 NS (g) Years with long May visit and 65.27 0.473 3 small group in July = 1960 to 1962 n = 502; Fig. 1). The proportion of eggslaid each day that fail to reach hatchingincreased significantly with laying date (rs = 0.759, df = 17, P < 0.001). No similar change occurred with laying date in the propor- tion of chickshatching that eventuallyfledged. During the layingperiods of 1960, 1961, 1962, and 1978, the day on which each egg was lost was known. The proportion of eggslost each day was not constant(Kolmogorov-Smirnov test X.,2 = 13.52, P • 0.01) but varied in an inconsistentway with date.