Inference and Assumption in Historical Seismology

The principal aim in studies of historical earthquakes is usually to be able to derive parameters for past earthquakes from macroseismic or other data and thus extend back in time parametric earthquake catalogues, often with improved seismic hazard studies as the ultimate goal. In cases of relatively recent historical earthquakes, for example, those of the 18th and 19th centuries, it is often the case that there is such an abundance of available macroseismic data that estimating earthquake parameters is relatively straightforward. For earlier historical periods, especiallymedieval and earlier, and also for areas where settlement or documentation are sparse, the situation is much harder. The seismologist often finds that he has only a few data points (or even one) for an earthquake that nevertheless appears to be regionally significant. In such cases, it is natural that the investigator will attempt to make the most of the available data, expanding it by making working assumptions, and from these deriving conclusions by inference (i.e. the process of proceeding logically from some premise). This can be seen in a number of existing studies; in some cases extremely slight data are so magnified by the use of inference that one must regard the results as tentative in the extreme. Two main types of inference can be distinguished. The first type is inference from documentation. This is where assumptions are made such as: “the absence of a report of the earthquake from this monastic chronicle indicates that at this locality the earthquake was not felt”. The second type is inference from seismicity. Here one deals with arguments such as “all recent earthquakes felt at town X are events occurring in seismic zone Y, therefore this ancient earthquake which is only reported at town X probably also occurred in this zone”. While in many cases such assumptions may very well be correct, they are usually not testable – or at least untested. Furthermore, it is possible to produce numerous contrary examples. It is concluded that the use of inference to amplify poor data must be made very transparent to the end user of the results, to avoid misleading appearances of accuracy. In many cases it may be best to abandon the quest for parameters altogether and admit that the data are inadequate.