Journal of Volcanology and Geothermal Research 92 Ž1999. 31–38 www.elsevier.


Statistical analysis of the frequency of eruptions at Furnas Volcano, Sao ˜ Miguel, Azores
G. Jones a , D.K. Chester b, F. Shooshtarian

c, )

Department of Computing and Mathematics, London Guildhall UniÕersity, London, EC3N 1JY, UK b Department of Geography, UniÕersity of LiÕerpool, LiÕerpool, L69 3BX, UK c Department of Electronics and Mathematics, UniÕersity of Luton, Luton, LU1 3JU, UK Accepted 20 April 1999


¸ores. Both Furnas and Fogo have Furnas Volcano is one of three major volcanic centres on the island of Sao ˜ Miguel, Ac displayed violent explosive activity since the island was first occupied in the early 15th century AD. There is concern that future volcanic activity will not only cause major economic losses, but will also result in widespread mortality, and it is for these reasons that a major programme of hazard assessment has been undertaken on Furnas. The present study is part of this programme and involves both the general statistical modelling of the record of historic eruptions and, more specifically, develops a technique for determining the rate of volcanic eruptions Ž l., an important parameter in the Poisson probability model. q 1999 Elsevier Science B.V. All rights reserved.
Keywords: frequency of eruptions; Furnas Volcano; Sao ˜ Miguel, Azores

1. Introduction Furnas volcano is one of three major active volcanic centres on the island of Sao ˜ Miguel in the Ac ¸ores. The last eruption of Furnas in 1630 AD killed ; 100 people ŽCole et al., 1995.. Queiroz et al. Ž1995. have demonstrated that there was also an historic eruption ; 1440 AD, during the period of the first settlement of the island by the Portuguese. The eruptions of Furnas over the last 5000 years are well documented ŽBooth et al., 1978, 1983; Cole et al., 1995, 1999-this issue; Guest et al., 1999-this


Corresponding author. E-mail:

issue. and a chronology is provided in Table 1. These eruptions range from sub-plinianrphreatomagmatic events, such as 1630 AD, to larger plinian episodes — for example Furnas C. Even the smaller eruptions, if repeated, would lead to total devastation within the caldera ŽFig. 1.. In an assessment of the volcanic hazard of Furnas, it is important to consider the probability of a future eruption and Moore Ž1990., on the basis of the number of eruptions over the last 5000 years, showed that eruptions had taken place on average once every 300 years and, with the last eruption occurring almost 400 years ago, he considered that there was at present a high probability of eruption. Eruptive mechanisms of volcanoes are not, however, sufficiently well understood to allow determin-

0377-0273r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 7 - 0 2 7 3 Ž 9 9 . 0 0 0 6 5 - 7

r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 Table 1 Eruption chronology. Ac . Sao ¸ores Žbased on Queiroz et al. Jones et al. and other sources.32 G. ˜ Miguel.

1.G.. . Furnas Volcano: general location map Žfrom Chester et al. 1995.. r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 33 Fig. Jones et al.

types. c. 1985. 1966a. 3. 2. has introduced the concept of a Nonhomogeneous Poisson model. The data Data are presented in Table 1 and dates of eruption are referred to two time axes. Various statistical methods for calculating l are examined to show how the record of volcanism on Furnas may be used to estimate values of l and so demonstrate the limitations involved in calculating l. it is our contention that estimates made from our model are meaningful. 4. A contagious distribution. according to . Newhall.. The Poisson process is used to describe a wide variety of stochastic phenomena that share certain characteristics and in which some ‘happening’ — or event — takes place sporadically over time. labelled RADIOCARBON. A random distribution. Ho et al. 1991. l is a critical parameter in the probability calculation. Using Table 1.’ If events in a Poisson process occur at a mean rate of l per unit time Ž1 year. Jones et al.. 1991. A random distribution Žnot to be confused with the term ‘random variable’.. 4. hence. Before doing so it should be noted that Ho et al. Ž1991. Scandone. spatial patterns and probabilities of eruptions. see Ho et al. Both scales use intervals of 200 years. Such volcanoes are ‘without a memory’ of previous events and. a Poisson model will be fitted to the eruptions of Furnas. of occurrences in an interval of time in t units is l t Že.d. but Babbington and Lai Ž1996.e. a volcanic eruption is defined as such an event. 105 year. who discussed the applicability of various Poisson models to volcanic events.. Such models have allowed many important features of future eruptions to be predicted. e. all of which have been incorporated into risk assessments ŽWickman. argue that such a model can possess certain undesirable features.. in a manner which is commonly believed to be ‘random. 1976. 1991. Choice of model Application of statistical methods to volcanic eruptions began with the pioneering research of Wickman Ž1966a. 1984.. etc. It should be noted that events can have: 1. Wickman observed that for this class of volcano age specific eruption rates are not time dependent. 1982. that is a model in which the occurrence of an event is dependent on the elapsed time since the previous event. is scaled from 5000 years BP to present. Condit et al. r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 istic predictions of future activity to be made with confidence.. In this paper we propose a statistical approach which is applicable equally to Furnas and to other volcanoes with similarly deficient data sets. 1989. 1982. The first. but note that radiocarbon dates have an arbitrary reference point of 0 BP s 1950 AD. 2.b. Taking into account the limited availability of precise data on the past eruptions of Furnas. A regular distribution. 3... A regular distribution suggests that events occur at fixed. b. therefore. The rate of occurrence of volcanic eruptions.g.. A Poisson distribution provides a suitable model for such behaviour and it is the hypothesis of randomness that is usually the first to be tested.c. regular intervals and a contagious distribution implies that there is a clustering of events. 1983.. both statistically and geologically. Klein.34 G. implying that events occur ‘at random’. On Furnas and many other volcanoes prediction must rely on statistical modelling using eruption records. then the expected number Žlong-run average. In particular. The second scale is chronological... d. His research involved the calculation of eruption recurrence rates for a number of volcanoes with different styles of activity. having a range from 3000 BC to 2000 AD. These have included: the frequencies. we estimate the unknown parameters which are required to calculate the probability of eruptions. The nature of these data are such that values are estimated when they do not coincide with either of the axes ŽTable 1. Mulargia et al. Methodology By electing to employ a Poisson model means that volcanic eruptions occur at random. Cruz-Reyna. are termed Simple Poissonian Volcanoes. In this paper. 1984.

l. 1. Then R follows a Poisson distribution with parameter l.’’. . Then the number. . . That is. . is given by: f Ž t . . To obtain an estimator for l the method of Maximum Likelihood is used. The observed sample values. repeating a random experiment several times to obtain information about the unknown parameterŽs. is negative when evaluated at x. . Let X denote the number of volcanic eruptions for a 10 5-year period for the NTS ŽNevada Test Site. This technique is called maximum likelihood estimation. the likelihood function is: n is 1 n is 1 L Ž l. and the time between eruptions has a negative exponential distribution whose probability density function is f Ž t . . 50. t d t .. s yn l q Ý x i ln l y Ł x i ! is 1 is 1 n ž / . the probability of two or more events occurring in a small interval of time Ž t . . X . r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 35 the Poisson process. 2. is oŽ d t . . x 2 . but are so small that they can be ignored. it follows that x 1 . . is l t q oŽ d t . One meaning of ‘best’ is to select the parameter values that maximise the likelihood function. x 2 . independence exists. . s yn q Ý is 1 xi l s 0. of occurrences per unit time has a probability P Ž X s x . Also. The distribution from which the sample arises is the population. s y Ý 2 xi l2 . Mathematically. Let this estimation technique be demonstrated. t q d t . s leyl t . will often be used. and the log-likelihood for a random sample from a Poisson distribution is: n n ln L Ž l . d t of the time interval is such that at most one event can occur. 3. denoted x 1 . where l is the rate of volcanic eruptions. x 2 . .2 . t . . argue that. p. . The maximum likelihood equation is therefore: d dl ln L Ž l . This is inn deed a maximum because the second derivative: ln L Ž l . s Ł n eyll x i x i! Ý xi L Ž l . Note that any value of l that maximises MLE. . . will also maximises that log-likelihood. l is assumed to remain constant throughout. are used to determine information about the unknown population Žor distribution. xi ˆ s Ýn which has the solution l is 1 d2 dl n is 1 s x. Assuming that x 1 . x n is a sample from the associated distribution. s eyll rrr ! for r s 0. s eyll xrx !. occurrences in non-overlapping intervals are independent of one another. . x n represent a random sample from a Poisson population with parameter l. since this is the rate of occurrence of volcanic eruptions. . Of course. Ž1991.1. the assumptions for a Poisson distribution are as follows: 1. The term oŽ d t . Ho et al. s Ł f Ž x i . This means that the probability of R s r eruptions in unit time is given by: P Ž R s r . The model indicates that the length. The analysis Let R denote ‘the number of eruptions per unit time’. also denoted ˆ . .. x s 0. If this type of Many statistical procedures employ values for the population parameters that ‘best’ explain the observed data. Observe that l is a parameter to be estimated from the data. the alternate form of the maximum likelihood equation. . ‘‘in dealing with distributions. or l LŽ l. s eyn ll is 1 n is 1 Ł x i! 5. Thus. t ) 0. and the maximising parameter values are called maximum likelihood estimates. . The collection of resulting observations.. l . x n . x 1 . x 2 .G. s leyl t . Jones et al. is useful. l ) 0. the probability of an event occurring in a small interval of time Ž t . denotes that terms such as d t 2 and higher powers are present. 2. . for computational convenience. ln LŽ l. x n are observations of a random sample of size n.. Often these observations are collected so that they are independent of each other. the time between eruptions follows an exponential distribution whose probability density function Žpdf. one observation must not influence the others.

l q q y 1195 q where E s total number of eruption during the observation period. 10 T s 0 20 T1 s 300 30 WHILE ABSŽT1 y T. using Program 1 shown below. then: Ža. these 16 values represent a sample of size 16 from a Poisson random variable with average recurrence rate m.. r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 region from this assumed Poisson process. .TABŽ20.. . t n . and T s observation period.36 G.1rT1 70 NEXT 80 END .5rŽEXPŽ1870)1rT.. Based on this ˆ can be defined as: estimation technique. dl s0 Equate this to zero and solve numerically. . Ž 1 y ey 730 l . to be 1000 BC. x 2 . t 2 . However. One event takes place after 360 years ŽG. t1 . our method in estimating l is different from Ho et al. If we wish to estimate l for the Quaternary using the Poisson count data for the NTS region.75 y 1402. s Ž 1 y ey 870 l . . Three events take place over a 730 year period ŽD q E q F. . 60 PRINT TABŽ1. the natural logarithm of the equation is taken and then differentiated with respect to l. the successive number of eruptions from the last 16 consecutive intervals of length 105 years Ž16 = 10 5 s 1. although we also use MLE.. . 4 s dl 3ey8 70 l Ž 870 . The above description indicates that Eq. We. with m s l t s 10 5l. x 16 . the distribution of x i ’s can provide information for model selection. Že. . . is set up. Thus. 3 3 ˆs l m ˆ 10 5 16 s Ý 6 is 1 Ž 16 = 10 . Then X follows a Poisson distribution with average recurrence rate m. must be estimated. . is the joint probability of the observations we may write: L Ž l . .. for checking the adequacy of the model and for parameter estimation in general. One event occurs after 35 years ŽI. The number of observed eruptions per interval are denoted as x 1 . t 2 .. t n . No events have happened for a period of 365 years.. q 3ln Ž 1 y ey7 30 l . Estimating the mean of the Poisson variable from these count data gives: 16 msxs and Ý is 1 xi 16 . One event occurs after 265 years ŽH.. . xi =ley3 60 lley2 65 lley3 5 lley1 70 l ey3 65 l . . . l the observation period Žin years. 3ey7 10 Ž 710 .1rT. Jones et al. . Three events occur during a period of 870 years ŽA q B q C. t 1 . Žb. Žg. to give: l s 0. 1ye 5610 y7 10 l e1870 l y 1 q e710 l y 1 d Ž ln L . 1 y ey8 70 l s y 1195 l 2130 4 This shows that the estimated annual recurrence ˆ . y 532. Turning once more to the data relevant to the Furnas research problem and taking the origin on the chronological scale ŽTable 1. Note that for the estimation of l in this model. ) 0. Žd. Ž1991. Ž1. y 1. is the average number of eruptions during rate. Žc. One event occurs after 170 years ŽJ. The resulting expression is made equal to zero and the equation is solved using an appropriate method. and we believe it provides a better estimate of the parameter l. . Taking natural logarithms we have: ln L s 4ln l y 1195 l q 3ln Ž 1 y ey8 70 l . the likelihood function LŽ l. . t n . estimate l as follows: First.00403075. is the ‘best’ estimator arising out of the application of method of MLE. y 1.5rŽEXPŽ710)1rT.0000001 40 T s T1 50 T1 s 298. Differentiating and rearranging terms we have: d Ž ln L . l ˆs l E T Ž 1. t 1 . However.. therefore. an individual observation x i is not required. Žf.6 = 10 6 s Quaternary period. . Given that LŽ l.

J. 1983. P. J. 39–53... A quantitative study of five thousand years of volcanism on Sao ˜ Miguel. Queiroz. that is. Soc.. 54.I. R. G.. Walker...B.00148467 . London 288A. s 0.P. Hawaii.D.L. 1999. J. also follows a Poisson distribution. A recently published analysis of global volcanic activity argues ŽHo et al. there is a 2% chance of an eruption occurring.. Res. 1991. a 95% interval estimate for l has been estimated and this may be used to provide interval estimates of probabilities of eruption over a range of times from 5 to 100 years.. the lower 2.. Forecasting Volcanic Events.817 P Ž T ) 75 . Gaspar. Wallenstein.C. 271–319. Styles of volcanism and volcanic hazards on Furnas volcano. has a negative exponential distribution with parameter l then 2 nx l is distributed as a x 2 random variable on 2 n degrees of freedom.W.5% point is 34. Moore.. there is a 3% chance of an eruption occurring..H. Further. A. Amsterdam.603 P Ž T ) 100 . r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 37 Program 1 From this we can see that: P Ž T ) 5 .. We suggest that estimates based on a 95% confidence interval for l would be more useful. 94. Sao ˜ Miguel. Cole.E.. Bull. 69.S. A.E. Coutinho.591 and the upper 2.980. s 0. Klein. J-M. s 0. Volcanol. Geotherm. s 0. the Poisson probability model. Klein.. Res. Pacheco.. ŽEds. Res. USA: statistical estimation of recurrence rates. s 0. J. Gaspar. Guest. s 0. Sao ˜ Miguel. Volcanol. s 0.W. Azores. L.. 1082–1975. It is well known that if a random variable..G. 1991. Elsevier.471 Thus. The method is based on the number of eruptions occurring over a defined period of observation.947 P Ž T ) 25 . Volcanol. Azores. that is. the revised probability that no eruption occurring within the next 5 years is 0. Naumann. 6.. 1978. 54. Volcanol. 602–614. Wallenstein. Wallenstein. Azores. Geotherm.. the probability that no eruption will occur within the next 5 years is 0. Elstow. J. F. Feuerbach. Guest. J.M.. R. Guest. Jones et al.668 Thus. J. C. 92.. N.. 1989. Patterns of historical eruptions at Hawaiian volcanoes. Trans. Volcanic geology and eruption frequency. J. Sao ˜ Miguel.. 1995. Geotherm. Dibben.. an estimated value for l has been derived. G. 1–29. Sabroux. Smith. G. Eruptive probability calculation for the Yucca Mountain site.l 0. T. Duncan.868 P Ž T ) 50 ..D. W. Aubele.904 P Ž T ) 50 .. J. G. s 0. Booth.. L. 1990...De la.5% point is 9. Res... 89. Eruption forecasting at Kilauea volcano.736 P Ž T ) 75 . Philos. s 0. Volcanol.739 P Ž T ) 100 . Croasdale. Azores. Patterns of volcanism along the southern margin of the Colorado plateau: the Springeville Field. s 0...L. Thus. Geotherm... L. 1984. H. Res.D.K. Queiroz. E. R. an important parameter in . Crumpler. J. 57–67. Duncan. Gaspar. Open File Report 4. pp. B. R. T. 3059–3073. s 0. Volcanol. that the number of eruptions occurring per unit time. Croasdale.170. 117–135. Ho. with x s 320 and n s 10 we can set up a 95% confidence interval for l. By using the method of maximum likelihood. Volcanic hazard on Sao ˜ Miguel. Report on the evacuation of the Furnas District. Cole.-C. Sao ˜ Miguel.L. Bull. 1–35. N.M. The estimates above of the probability of no eruption occurring prior to a given elapsed time are based on the point estimate of l. C.973. Geophys.. Condit. An historic subplinian r phreatomagmatic eruption: the 1630 AD eruption of Furnas Volcano.. University College LondonrCommission for the European Union. Poisson-distributed patterns of explosive eruptive activity. A. R. Ferreira. Duncan. P. Chester. Conclusions The problem discussed in this paper involves the derivation of a technique for estimating the rate of eruptions by means of Ž l. 50–56. 12.00403075. D. 1999. Cole. The volcanic geology of Furnas volcano. These values lead to the 95% interval 0.D. 1982. 99– 109. Pacheco. 52.980 P Ž T ) 10 . F. In: Tazieff. namely 0. S. References Booth. Azores.00528974. N.M. G.. Queiroz. J. J. 1991.R. T. Bull. Res.960 P Ž T ) 25 .L... P...E. D. J..E.973 P Ž T ) 10 .. Ferreira. Azores.P.. C. Azores. Walker.. s 0. Geophys. This enables us to refine our estimates to give: P Ž T ) 5 . J. Cruz-Reyna. B. Sao ˜ Miguel. 1995. J. Volcanol. With 20 degrees of freedom. X .

4. In: Tazieff.. 337–350...E.. F.E.. Pacheo. 4. Sabroux..E.. C. 1982.. 1. na primera metade do Secob XV. A. Washington. 319–335. J. 1985. N. Duncan. ´ Scandone. Geotherm..C. pp. J. A method for estimating intermediate and long-term risks from volcanic activity. pp. C. Mineral. D. United States Geological Survey. r Journal of Volcanology and Geothermal Research 92 (1999) 31–38 Wickman. Amsterdam. J. R. Semiquantitative assessment of changing volcanic risk at Mount St. Jones et al. Markov models of repose-period patterns of volcanoes. 4.. Eruption histories of some selected volcanoes. In: Merriam... ŽEd. 135–161. United States Geological Survey. Wickman. 84–272. Eruption histories of some East Indian volcanoes. Wickman. F. 82–396. G.E. S. Wickman... Ac ¸oreana 8. Guest.F. Mineral. F. Tinti. F. Ac ¸ores. Elsevier.. F. Open File Report. 1966a. Springer.L. Repose period patterns of volcanoes: V..G. General discussion and a tentative stochastic model. Cole. 1984.G. Queiroz. Berlin. Mineral. Random Processes in Geology. Repose period patterns of volcanoes: IV. Forecasting Volcanic Events.. Newhall. Repose period patterns of volcanoes: I. Washington. Helens. 351–367. A statistical analysis of flank eruptions on Etna Volcano. Ark. Washington.M. 1966d. DC. Mineral. 1976. Boschi. Volcanol..E. Geol. J. Mulargia.38 G. Wickman. Mineral. ŽEds.. E. Wallenstein. Ark. Newhall. Geol. F. Ark. Erupc ¸oes no Vale ˜ vulcanicas ˆ das Furnas ŽInla de S. Miguel. 291–301.. Geol. 303–317. H. 1983. . 4. Repose period patterns of volcanoes: II. J. Open File Report. Helens.D. Problems related with the evaluation of volcanic risk. P. 1995.. Res. 1966b.E. 4. Geol. with an example from Mount St. Eruption histories of some Japanese volcanoes. F. Wickman. 263–272. Geol. 1966e.E. Developments in Volcanology. DC. 23. Volcanic eruptions regarded as random phenomena. Repose period patterns of volcanoes: III. Ark. 57–69. 159–165. Ark. Gaspar.. 1966c..

Sign up to vote on this title
UsefulNot useful