Automatic Clustering of Macroseismic Intensity Data Points from Internet Questionnaires: Efficiency of the Partitioning around Medoids (PAM)

Abstract

Online Material: Tables of earthquakes and clustering results, maps of questionnaire results, ananimation showing the evolution of the Barcelonnette event questionnaire clustering with time, and figures showing clustering comparisons (zipped archive). In the early history of scientific seismology, the systematic collection of macroseismic information was performed through standardized paper questionnaires. The Internet is changing the way earthquakes are monitored by citizens (Wald et al. , 1999; Bossu et al. , 2011, 2014). With the phenomenal growth of the Internet, many Internet‐based macroseismic survey systems have been implemented in the last decade in many parts of the world (see Wald et al. , 2011, for a survey of these systems). These kinds of systems include automatic procedures for the assignments of intensity values. The two main processing stages in assigning intensity consist of grouping data by place then determining the intensity degree that best fits the descriptions. Grouping macroseismic observations is an elaborate process: groups should be large enough to allow a general assignment of the intensity (here, “general” means that the intensity value is not assigned to an individual questionnaire) but small enough to fairly represent local geographical conditions. In well‐recognized organizations dealing with Internet macroseismic questionnaires, some ad hoc partitioning techniques are currently being used. Among these techniques, the most commonly used in macroseismology makes regular tessellation with grid squares, perhaps because it is the easiest to create and deal with. Results from grid partitioning are expected to be influenced by grid geometries. Grid geometries are probably not significantly biasing results when raw data are already artificially discretized. Artificial distributions may be expected when city center point coordinates or postal ZIP codes are used to …