GNGTS 2013 - Atti del 32° Convegno Nazionale

Recognition of nodes prone to earthquakes M5+. MZ analysis delineated 102 nodes within the study region: 60 of them are located in lowland environments in interiors of the Po plain, the remaining 42 nodes sit on the first rank lineaments that divide the Po plain from the surrounding mountain chains. The considered parameters (Tab. 1), particularly the morphometric ones, differ significantly for nodes located within the Po plain and for nodes related with the first rank lineaments that delimit the plain. Therefore the recognition has been performed separately for these two groups of nodes, applying the CORA-3 recognition algorithm to each of them. Accordingly, at the learning stage all the nodes have been a priori divided into two sets, so as to define the training sets for the CORA-3 algorithm. Specifically, to assemble the set D 0 , the nodes situated most closely to the epicenters of past M ≥5.0 earthquakes are considered; the remaining nodes have been assigned to the set N 0 . Recognition of nodes located within the Po plain. For the group of nodes located in the Po plain, four variants of classification of nodes into classes D and N have been performed, in order to identify the most stable one. The results of recognition are summarized in Tab. 2. Three variants of recognition (I-III) have been obtained using a different learning set D 0 . In the variant I the set D 0 was formed using the nodes located most closely to the epicenters of all earthquakes shown in Fig. 1. In the variant II the nodes 28 and 80 (Fig. 2), hosting only ancient earthquakes whose location is poorly defined, were excluded from D 0 . Because the catalog UCI is representative for M5+ since 1500, in the variant III we excluded from D 0 all the nodes hosting only events occurred before 1500 (specifically nodes 28, 33, 37, 80, 82, and 89). An additional classification of the nodes (IV) was obtained using the characteristic traits defined by CORA-3 for D and N nodes located in lowland areas of Iberia (Gorshkov et al. , 2010). Tab. 2 – Results of recognition for nodes located in the Po plain. The total number of nodes, both D and N , is 60.The total number of large past earthquakes occurred within the nodes is 18. Classification variant I All events from UCI catalog are used to define D 0 II Nodes 28 (Milano) and 80 (Venice) are excluded from D 0 III Only nodes hosting events after 1500 are used to define D 0 IV D - and N -traits as defined for Iberia Total number of identified D nodes 32 (53%) 20 (33%) 20 (33%) 37 (62%) Total number of target earthquakes within the identified D nodes 18 (100%) 16 (89%) 13 (72%) 8 (44%) The variants I and IV are discarded because too many nodes are assigned to class D ; moreover the variant IV also misses several nodes hosting target events. The variants II and III are practically equal, in spite of the difference in the learning sets D 0 . Although the variant III misses three nodes (namely 28, 80, 82), all hosting events occurred before 1500, it is considered the main variant, because to select D 0 only nodes hosting events M5+ that occurred after 1500 are used, when the UCI catalog is reasonably complete. Nevertheless, the importance of the nodes 28, 80 and 82 (Fig. 2), which experienced events occurred before 1500 (when the reliability of the epicentral locations in the catalogue is severely conditioned by the naturally limited amount of available information), calls for a special investigation that will be performed in a forthcoming study. 91 GNGTS 2013 S essione 2.1