How to go extinct: lessons from the lost plants of Krakatau

Aim Few data sets exist that allow measurement of long-term extinction and turnover rates for islands of the size of the three main islands of the Kra katau group. We test the reliability of previous estimates of plant species extinction and examine structure within the extinction data. Location The data analysed are for the three older Krakatau islands: Rakata , Sertung and Panjang in the Sunda Strait, Indonesia. Methods Our analysis is based on a comprehensive database incorporating all species records for each island since recolonization began after the 1883 sterilization, plus attributes such as distribution, phylogeny, population status and dispersal mechanism for each species. We employ a combination of univariate and multi-term analyses in analysing structure, and derive Mini mal Adequate Models using binary logistic analyses of variance and covarian ce. We compare the 1883-1934 data set with the contemporary nora as represe nted by (1) 1979-83 records las used in previous analyses) and (2) 1979-94 data (original). Results The improved data for the contemporary flora reduces the number of missing species by one-third. We show that a variety of estimates of extinc tion rate can be produced depending on what assumptions are made concerning the status of particular species groups. Structural features in the extinc tion data persist despite the reduction in overall numbers of losses. Losse s relate to: (Ij the number of islands on which a species originally occurr ed, (2) the primary dispersal mode, and (3) the original abundance of a spe cies (e.g. whether it was known to have established a successful resident p opulation, and whether it was in decline or increasing in c. 1930). The 'be st' descriptive model employs the variables denoted under (3). A high propo rtion of losses comprised species introduced by people and rare or ephemera l species. Losses of 'residents' that had colonized naturally could largely be accounted for by reference to (1) successional loss of habitat and, to a lesser degree, (2) other habitat disturbance or loss. Main conclusions Previous analyses, based on a more limited data set, have significantly over-estimated extinction from the Krakatau flora. Few natura lly colonizing and established species have become extinct. The findings in dicate that caution is necessary in interpreting 'headline' island ecologic al rates, and in analysing and modelling such data. Examination of structur al features of the data appear to be valuable both in providing ecological insights in their own right, and in enabling refinements to estimates of ex tinction and thus turnover.