Enhancement of the EEPAS model for long-range earthquake forecasting

Technical Abstract

The EEPAS model is a long-range earthquake forecasting method, which is showing promise in forecasting major earthquakes in several regions, including New Zealand, California and Japan. This model uses all the previous earthquakes in the catalogue to forecast the probability of future earthquakes occurring at future times, magnitudes and locations. It relies on a precursory increase in the rate and magnitude of minor earthquakes, which precedes most major earthquakes in the long term.
In some regions located on the boundary between two tectonic plates, including New Zealand and Japan, the edge of one tectonic plate is forced under the edge of another. Then it is possible to distinguish three types of earthquake: slab earthquakes occurring within the lower plate, interface earthquakes occurring close to the plate boundary, and crustal earthquakes occurring within the upper plate. The objective of this study is to improve the EEPAS model by distinguishing these three types of earthquake.

The earthquake catalogue of Japan with magnitude M ≥ 4 was used and each earthquake was classified as slab, interface or crustal. The period from 1966-1995 was used as a learning period for fitting models, and 1996-2005 as a testing period for confirmation of the results, with the target earthquakes being those with magnitude M ≥ 6. There are 105 target earthquakes in the learning set (42 slab, 21 interface and 42 crustal) and 37 in the testing set (13 slab, 7 interface and 17 crustal).
Fitting of the EEPAS model to the learning period has shown that:

1. Major slab earthquakes tend to occur near to where major slab or interface earthquakes have occurred previously, and minor slab and interface earthquakes tend to occur as precursors to major slab earthquakes.
2. Major interface earthquakes tend to occur where major interface or shallow slab earthquakes have occurred previously, and minor interface earthquakes tend to occur as precursors to major interface earthquakes.
3. Major crustal earthquakes tend to occur near to where major crustal or interface earthquakes have occurred previously, and minor crustal earthquakes tend to occur as precursors to major crustal earthquakes.

The fit of the EEPAS model to the learning set is improved by using only slab and interface events to forecast major slab earthquakes, only interface events to forecast major interface events, and only crustal events to forecast major crustal events. This new model fits the slab and interface earthquakes better than it does the crustal earthquakes. The improvement in forecasting performance was confirmed by applying the new model to the testing period.

The Japan catalogue was used because it is large and the earthquake types are relatively easy to assign. The results are relevant to other similar regions, such as New Zealand. Therefore, an improvement in forecasting performance could probably be obtained by applying the same methods to the New Zealand catalogue.