Murdoch University Research Repository

Welcome to the Murdoch University Research Repository

The Murdoch University Research Repository is an open access digital collection of research
created by Murdoch University staff, researchers and postgraduate students.

Learn more

The development of a model of the spread of the pilchard fish kill events in southern Australian waters

Murray, A.G., O'Callaghan, M. and Jones, B.ORCID: 0000-0002-0773-2007 (2000) The development of a model of the spread of the pilchard fish kill events in southern Australian waters. Australian Government. Fisheries Research and Development Corporation


The pilchard mass mortalities of 1995 and 1998/9 were unprecedented in their rate and geographical scale of spread. Waves of mortality spread from South Australia to Western Australia and to Queensland at a rate of 10 - 40 km d-1. In many cases, stocks were reduced by over 60%. The cause of this mortality was certainly a herpesvirus, although as it has proved difficult to infect fish with this virus Koch’s postulate remains unfulfilled.

We have developed a range of models looking at disease transmission from the school to the national level. These models enable us to determine which parameters control the transmission of disease. At the school level we conclude that small - scale fish mixing patterns do not play a dominant part in the local development of disease. Hence we are able to model the larger scale transmission without considering lower level population details.

At the larger scale we produce models that generate realistic epidemic waves. The model we have produced differs from standard forms in that it uses fixed length latent and infectious periods, rather than continuous turnover between these phases. Using analytical methods we find that three parameters control the epidemic wave’s geographical spread: the rate of disease transmission, the length of the latent period, and diffusion coefficient.

We also use the observed local pattern of mortality to constrain the model. Initially, in South Australia, there is recurrent mortality over days or weeks. Later, when the epidemic is matured, mortality occurs over a few days at any given location.

The epidemic wave’s speed is least sensitive to the rate of disease transmission, however this parameter could vary by orders of magnitude, so weak sensitivity does not necessarily mean low importance in explaining variation in the wave’s speed. A large decline in the number of virus - containing lesions in the gills of sick fish was observed between 1995 and 1998. This would indicate reduced viral transmission. At large values of the viral transmission rate the wave’s speed becomes increasingly less sensitive, so there is a value beyond which wave speed becomes independent of this parameter.

The transmission rate is multiplied by population density in our standard model, however, population density has not varied by orders of magnitude (although it has off Japan) and so the epidemic wave speed is only very weakly sensitive to changes in population density. Alternatively, because schooling effectively keeps population density constant, the viral transmission may be population density independent, in which case population density has no effect on the epidemic’s speed.

Large values are inconsistent with the large wave speeds observed and small values produce unrealistic initial epidemic behaviour, so values of around 4 days give the best results. If this parameter could be experimentally evaluated the model would be very strongly constrained.

The diffusion coefficient is the parameter that reflects the large - scale spatial transmission of the virus. The diffusion coefficients generated by fish swimming patterns appear to be quite sufficient to explain the observed rate of spread of the epidemic. Indeed diffusion coefficients larger than those which fish can generate (perhaps as a result of bird transmission) result in mortality patterns that are inconsistent with the observations: several days of similar levels of mortality. This inadmissibility of very large diffusion coefficients does not rule out vector transmission, but it does make it far less likely that vectors are involved. Change in diffusion is the most likely process to explain the large difference in speed between east and west bound waves in a single epidemic.

The model generates very high levels of infection, in excess of 90% for realistic mortality distribution patterns. This means the critical parameter for determining the epidemic’s longer - term impact is the proportion of those infected fish which survive infection. A model of post - epidemic population recovery indicates that this should be fairly rapid, even with high levels of epidemic mortality. However, the same model shows that persistently elevated mortality, even to a small degree, leads to serious decline in fish stocks.

Because of weak sensitivity to adult population levels, fisheries management strategies based on the manipulation of populations are very unlikely to succeed. Control of vectors is also unlikely to be effective. Juvenile pilchards appear to be confined to embayments their populations do not mix easily. This makes the preservation of as many nursery sites as possible the best means of protecting stocks from epidemics. As yet, the origin of the virus is unknown. In the longer term, exploitation of the adult population's strong degree of mixing may make it possible to inoculate the population with low mortality virus.

Item Type: Report
Series Name: FRDC Project No. 99/225
Publisher: Australian Government. Fisheries Research and Development Corporation
Publishers Website:
Item Control Page Item Control Page