Physics of the Earth and Planetary Interiors, 21(1980) 267—281 267 © Elsevier Scientific Publishing Company, Amsterdam — Printed in The Netherlands APPLICATION OF KEILIS-BOROK AND MCNALLY PREDICTION ALGORITHMS TO EARTHQUAKES IN THE LAKE JOCASSEE AREA, SOUTH CAROLINA JEANNE SAUBER and PRADEEP TALWANI Geology Department, University of South Carolina, Columbia, SC 29208 (U.S.A.) (Received December 12, 1978; accepted for publication April 2, 1979) Sauber, J. and Taiwani, P., 1980. Application of Keilis-Borok and McNally prediction algorithms to earthquakes in the Lake Jocassee area, South Carolina. Phys. Earth Planet. Inter., 21: 267—281. Following aML = 3.2 earthquake in November 1975 in the vicinity of Lake Jocassee in the northwestern part of South Carolina, microearthquake activity has been monitored on portable seismographs, with events as small as ML = —0.6 being well located. Using modified forms of prediction algorithms developed by Keilis-Borok et a!. (1977) and McNally (1978a, b) systematic patterns in temporal and spatial distribution of the smaller (—0.6 ~ M~, < 2.0) events were found to precede the larger felt events (ML ~ 2.0). Using the Keilis-Borok algorithm, “swarms” were recognized before five of the seven large earthquakes, while with the McNaliy method three of the larger earthquakes were preceded by large clusters (>three events per cluster), and one was preceded by several smaller clusters (two or three events per cluster). The series of events were also tested for a Poisson random process. For the earlier, more active time frame, November 1975—June 1976 (mean rate of occurrence between locatable events —‘20 hours), the Poisson hypothesis was rejected at a highly significant level. However, for the less active time period, October 1976—December 1977 (—40 hours for mean rate), the Poisson process may be a reasonable approximation to the earthquake occurrences. In comparing the two prediction algorithms, the modified form of the Keilis-Borok algorithm was found to be more successful in the recognition of precursory activity with a smaller number of false alarms (three). When the data begin to approach a Poisson process, however, the algorithm is less accurate and physical constraints are relied on to decrease the number of false alarms. The McNally method was found to be less biased in application and easy to use for future examination of seismic activity. A major weakness, however, was the need for an additional con- straint to differentiate between precursory activity before the larger events (ML ~ 2.0) and smaller events (1.0 ~ ML< 2.0). 1. Introduction seismic quiescence, which in turn is followed by the main shock. Recent studies of rockbursts (Brady, 1977), reser- The second stage in the above sequence, when voir-induced microearthquakes (Taiwani, 1978) and there is a significant increase in seismic activity in the large earthquakes (Evison, 1977a, b;Haberman and area of a future event, could, if recognized, serve as a Wyss, 1977;Ohtake et al., 1977;McNally, 1978a, b) precursor to a large event. Instead of being scattered have reported one or more of the following stages of in distribution like background seismicity, the events precursory seismicity. Firstly, there is a small initial that occur in the second stage are usually clustered. increase in seismicity outside the future rupture zone, One problem that arises in the recognition of this pre- while within it there is little or no seismic activity (this cursory activity is that of differentiating between a stage is not always present). This is followed by a sig- slight increase in seismic activity, and a significant nificant increase in seismic activity in the future flip- increase in the area of a future large earthquake. If ture zone. Following this increase, there is a period of the existing observations of seismicity patterns could