On the validation of earthquake-forecasting models: The case of pattern recognition algorithms

Earthquake forecasting is one of the geophysical issues with a potentially large social and political impact. Besides the purely scientific interest, the loss of lives and the huge damage caused by seismic events in many regions of the world have led many research groups to work in this field. Until now, however, the results obtained have not been convincing and they often have been a matter of intense debates. In part, these debates are due to the ambiguous definition of key concepts, such as precursor and forecast/prediction, as well as to the lack of a clear strategy to set up and check an earthquake-forecasting model. In this article, we provide insights that might contribute to better formally defining the earthquake-forecasting problem, both in setting up and in testing the validity of the forecasting model. As an illustration, we have applied these insights to forecasting models M8 and CN based on a pattern recognition approach. We found that the forecasting capability of these algorithms is very likely significantly overestimated.