Realm: Terrestrial
Climate: Temperate
Biome: Temperate broadleaf and mixed forests
Central latitude: 51.698000
Central longitude: -5.277000
Duration: 47 years, from 1928 to 1979

528 records

29 distinct species

Across the time series Haematopus ostralegus is the most frequently occurring species

Methods

Skokholm was established as a bird observatory by R.M. Lockley. He took a lease on the island in October1927. and kept records of breeding birds from 1928 to1940 (Lockley 1938. 1947). After the war theobservatory was run by the West Wales Field Societyand the Council for the Promotion of Field Studies (nowthe Field Studies Council). and later by the EdwardGrey Institute of Field Ornithology of the University ofOxford. Records of breeding and. for most species inmost years. counts or estimates of the numbers ofbreeding pairs are available from 1928 to 1979. with abreak from 1941 to 1945. making 47 years of records.Those up to 1967 have been published by Lack (1969)and analysed by Williamson (1981). Those since 1967have been extracted from the Annual Reports of theObservatory in the Alexander Library of the EdwardGrey Institute. The whole record for land and waterbirds is given in Tab. 1.The criterion for inclusion in Tab. 1 is breeding. butas is now well recognised. there are different degrees ofcertainty in such records. For a study of immigrationand extinction the important question is whether a pairof birds was attempting to breed. rather than whether itbred successfully. For instance the raven Corvus coraxhad one pair every year (two in 1961). but in 1978built a nest in Steep Bay. but it remained empty. Inthe British Trust for Ornithology code this is B. Nestbuilding. and counts only as Probable breeding in thisspecies. though it is regarded as Confirmed breedingin most others (Sharrock 1976). For this study it countsas a population of one pair. In most cases the annualreport gives a single figure for the number of pairs. Forneighbouring integers. e.g. 12 or 13. I have taken thehigher figure. for ranges. e.g. 35 to 40. a figure at or just slightly from Lack (1969). and includes figures thatabove the mid-point (38 in this case). Lack records as question marks. and Lockley also givesFor the pre-war figures Lockley (1947) differs the figures for 1940. I have therefore taken histhroughout. rather than Lack's. The only importantdifference in relation to the MacArthur-Wilson theoryis that Lockley records one pair of sky lark Alauda arvensisin 1938. which Lack (and Williamson 1981) recordas an extinction followed by an immigration in1939. Lockley also notes that he shot or deported littleowls Athene noctua when he could. because they atestorm petrels Hydrobates pelagicus. so I have excludedlittle owls from Tab. 1.It is evident that many of the estimates of the commonbirds in Tab. 1 are round figures. In some yearsestimates are lacking. though breeding is recorded. andas a full set of figures is needed for the calculations thatfollow I have interpolated. approximately linearly. insuch cases. The interpolated figures are in italics in Tab.1.Skokholm is a most important station for sea-birdpopulations. Throughout the years there have beenthousands of pairs of Manx shearwater Puffinus puffinus.storm petrel Hydrobates pelagicus. and puffinsFratercula arctica. and these are the largest colonies.along with those on the neighbouring island of Skomer.in the Celtic Sea (Cramp et al. 1974). Herring gullsLarus argentatus and lesser black-backed gulls Larusfuscus have increased from hundreds of pairs pre-war tothousands now. There are also hundreds of pairs ofrazorbills Alca torda and guillemots Uria aalge. Thegreat black-backed gull Larus marinus population iscontrolled at around ten pairs. The fulmar Fulmarusglacialis first bred in 1967 and even now occupies lessthan twenty sites. There have been casual records ofbreeding of shag Phalacrocorax aristotelis and kittiwakeRissa tridactyla (Lockley 1969). There have also beenthirty to forty pairs of rock pipits Anthus spinolettabreeding on the cliffs each year. Lack (1969) did notregard them as land-birds. and I have. somewhat reluctantly.followed him. partly because the estimates ofpopulation size are even rougher than those of themeadow pipit A. pratensis.MethodsMany features of the data can be brought out by simplemethods (Williamson 1981). However. to produce asatisfactory ordination of both species and years is notso simple. and requires a consideration both of differentordination methods. and of whether the data should betransformed.For most biological populations a transformation tothe logarithm of the numbers is desirable (Williamson1972a). When there are many zeros. as in Tab. 1. log(n+1) or better the arc-sinh transformation. which islog (n+(n2+ 1)12). can be used (Williamson 1981). butthese small numbers show features of a Poisson distribution.s o a square root transformation may be desirable(Pollard 1977). Tukey (1977) gives methods fordetermining an appropriate power transformation.which he refers to as re-expression. (with the logarithmbeing equivalent to a zero power). It is well-known thata plot of the logarithm of the mean against thelogarithm of the variance may indicate that a transformation is needed. When the variance is proportional tothe mean. a square root transformation is indicated; thestandard deviation proportional to the mean indicates alogarithmic transformation. Taylor et al. (1978) givemany examples of such plots showing clearly. to mymind. that most biological data needs transformationand usually something. on Tukey's scale. bounded bythe square root and the logarithm.These tests on the data of years in Tab. 1. both includingand excluding zeros. point to a square roottransformation. By species. again with and withoutzeros. the results are more variable. which is not surprising.The raven population is almost constant at onepair. The stock dove Columba oenas and jackdaw Corvusmonedula populations vary from zero to sixty pairs.In view of this scatter.although a logarithmic transformationwould be better for the common species. asquare root transformation is justifiable for all. and sohas been used.A standard Principal Component Analysis of datawith many zeros usually produces a curved result. thehorseshoe effect. from artificial data designed to belinear (Williamson 1978). Various methods have beenproposed to prevent this including multi-dimensionalscaling. MDS (Carroll and Arabie 1980). reciprocal averaging(or contingency analysis). and Hill's developmentof it. Decorana (Gauch 1982). and the step-acrossmethod (Williamson 1978). Reciprocal averaging andstep-across are direct; multi-dimensional scaling isiterative. and requires a variety of starting configurations.Decorana assumes the second Principal axis to bean artefact. and reduces it by a slicing technique.Step-across applies only to incidence matrices. whereall the entries are one of zero. Much of the variation inTab. 1 and similar data occurs in the number of entitiespresent. The similarity index used in step-across is theJaccard one. the number of joint occurrences. Thenatural extension of this to continuous data is illustratedin Fig. 2. It is the total set of joint occurrences. the areaof the intersection of the species graphs. Mathematically.this is Kendall's (1971) circle product. I min (a.b). The effect of using the circle product on species withmoderate overlap is to make them more similar than ifthe Jaccard index is used. so tending to re-create ahorseshoe. This can be corrected by a transformation;the square root of the circle product has been used here.This development of step-across can be called stepalongfor convenience.English (1982) compared MDS. Decorana andstep-along analyses of both the Eastern Wood (Williamson1981) data and a sub-set of the Skokholm data.using years in which complete records exist. Not allvariants of MDS are available at York. but he foundthose that were very expensive in computer time. requiring arbitrary choices of starting point. and the resultsdisappointing. Decorana produced the straightesttime axis. but. as will be seen. the time axis at Skokholmshould be curved. So the method has had the unfortunateeffect of removing. as an artefact. a real feature.The species ordination by Decorana was not readily interpretable.Consequently. the remainder of this paperdeals only with step-along results. The transformationsused here. of square roots for both data and circle product.are not critical. Runs with other transformationsproduce recognisably the same result Unit of abundance = Count, Unit of biomass = NA

Citation(s)

Williamson, M. (1983) The Land-Bird Community of Skokholm: Ordination and Turnover. Oikos, 41, 378–384.