D. Vere-Jones (Institute of Statistics and Operations Research, Victoria University of Wellington, Wellington, New Zealand) Ma Li (Analysis and Forecasting Center of China Earthquake Administration, Beijing 100036) M. Matthews (Department of Mathematics, MIT.Cambridge, MA02139-

4307, USA) Abstract This paper applies the M8 algorithm of Ma Li and Vere-Jones (1997) to the New Zealand local earthquake catalog, together with Keilis-Borok, Kossobokov and Rinehart (1986), Keilis-Borok, Knopoff, Kossobokov and Rotvain (1990)

, Matthews and Switzer (1992) compared early applications of the M8 to the California local earthquake catalog.

The New Zealand local earthquake catalog contains many smaller, especially medium-deep earthquakes not included in the NEIC catalog, and this is the first time M8 has been used to study these events.

Substituting local magnitude ML for Ms or mb given in the NEIC catalog, and applying the complete catalog including medium and deep source as well as shallow earthquakes.

M8 generated 4 TIPS for earthquakes with target magnitude ML≥7, one was successful, one failed, and two are in progress and have not yet ended.

Two areas included in the ongoing TIPS have experienced large shallow earthquakes since the beginning of TIPS, one less than ML=7 and the other ML=7.

These results are very similar to the M8 application results of California NEIC data performed by Keilis-Borok et al. (1990) and later confirmed by Matthews and Switzer (1992).

In both cases, they believe the algorithm results in a probability gain of about an order of magnitude.

In both cases, the results seem to depend on some surprising details.

Matthews and Switzer (1992) pointed out that the California results were strongly dependent on a set of small earthquakes in the Geysers geothermal area, which may have been caused in part by anthropogenic activity.

In New Zealand, the results depend on whether the catalog contains medium-deep source earthquakes, and if these events are removed, the same results cannot be replicated.

It is possible that at least part of the observed effect is due to improved network performance and directory reporting locating procedures.

1 Introduction This article summarizes the main results of the M8 algorithm applied to New Zealand data by Ma Li and Vere-Jones [5], and compares some of the salient features with those given by Keilis-Borok et al. [2] and Matthews and Switzer [6]

The California data of the M8 application was compared.

Compared to the initial global large earthquake studies citing M8, the New Zealand and California studies both involved regional studies of moderate seismicity.

Although both are located on the border of the Pacific Ocean, the tectonic environments of the two regions are very different.

The New Zealand region includes two subduction zones with different properties, each with a depth of several hundred kilometers. One is a volcanic zone and the other is a transition zone.

The California region is controlled by the San Andreas Translational Fault and its associated fault systems.

Despite these differences, the results described by Ma Li and Vere-Jones [5] and Keilis-Borok et al. [4] are very similar, at least superficially.

An algorithm is used to find increases in the probability of earthquakes with magnitude 7 or greater ("TIPS") for each 5-year interval.

The two regions contain a series of overlapping circular zones in the middle (7 in New Zealand and 8 in California), which the algorithm calculates over a period of about 20 years starting in 1975.

Four TIPS were obtained in both areas, including in California all three earthquakes with MS≥7 that occurred in the area during the observation period.

Also in New Zealand, TIPS includes two earthquakes with ML≥7 that occurred in the area during the observation period.

This is an impressive record, given that TIPS in each case covers only a small fraction (about 1/7) of the entire space-time period covered by the study.

It represents a probability gain of about 5 to 10 in each case, and may only happen by chance. Assuming that the research period TIPS in each circle area is randomly assigned, then the probability of each area is about 1/50

Or the two districts combined are about 1/2000.

Regardless, a closer exploration of the factors that produce TIPS in both cases reveals some troubling features.

Matthews and Switzer [6] pointed out that results from California data are sensitive to a set of small seismic events in the Geysers geothermal area, which are thought to be caused by water injection to increase steam production.

The results for the New Zealand data are very dependent on whether the catalog includes medium-deep source earthquakes that occur in subtracting plates.