Title: Pattern recognition methodologies or the spacetime identification of strong earthquakes: the case of
1Pattern recognition methodologies or the
space-time identification of strong earthquakes
the case of Italy
A. Peresan G.F. Panza
.
Dipartimento di Scienze della Terra Università di
Trieste
2Outline
- Pattern recognition of earthquake prone-areas
- Algorithm M8 in Italy
- Algorithm CN in Italy
- CN algorithm and long lasting changes in reported
magnitudes - Stability of CN predictions with respect to
random errors in reported magnitudes
3Pattern-Recognition of Earthquake Prone Areas
4Pattern Recognition of Earthquake Prone areas
- Pattern recognition technique is used to
identify, independently from seismicity
information, the sites where strong earthquakes
are likely to occur - Assumption strong events nucleate at the nodes,
specific structures that are formed around
intersections of fault zones.
5Pattern Recognition of Earthquake Prone areas
- The nodes are defined by the Morphostructural
Zonation Method - delineates a hierarchical block structure of
the studied region, based on
- topography
- tectonic data
- geological data
6Pattern Recognition of Earthquake Prone areas
- The Earthquake Prone Areas are identified
evaluating the following topographic
characteristics - Elevation and its variations in mountain belts
and watershed areas - Orientation and density of linear topographic
features - Type and density of drainage pattern.
- These features indicate higher intensity in the
neotectonic movements and increased fragmentation
of the crust at the nodes.
7Pattern Recognition of Earthquake Prone areas
GELFAND I., Guberman Sh., Izvekova M.,
KEILIS-BOROK V F. RANTSMAN E. (1972) - Criteria
of high seismicity determined by pattern
recognition. Tectonophysics, 13 (1/4),
415-422. Gvishiani A., and Soloviev A. (1980). On
the concentration of the major earthquakes around
the intersections of morphostructural lineaments
in South America. In V.I.Keilis-Borok and
A.L.Levshuin (eds), Methods and Algorithms for
Interpretation of Seismological Data. Nauka,
Moscow, 46-50 (Computational Seismology, Iss.13,
in Russian).
- The fact that earthquakes are nucleated at the
nodes was first established from observations in
the Pamirs and Tien Shan (Gelfand et al., 1972). - The non-random nature of this phenomenon has been
proven by an especially designed statistical test
(Gvishiani Soloviev, 1980)
8Pattern Recognition of Earthquake Prone areas
GELFAND I., Guberman Sh., Izvekova M.,
KEILIS-BOROK V F. RANTSMAN E. (1972) - Criteria
of high seismicity determined by pattern
recognition. Tectonophysics, 13 (1/4),
415-422. Gvishiani A., and Soloviev A. (1980). On
the concentration of the major earthquakes around
the intersections of morphostructural lineaments
in South America. In V.I.Keilis-Borok and
A.L.Levshuin (eds), Methods and Algorithms for
Interpretation of Seismological Data. Nauka,
Moscow, 46-50 (Computational Seismology, Iss.13,
in Russian).
This approach has been applied to many regions of
the world. The predictions made in the last 3
decades have been followed by many events in the
analysed areas, 84 of which occurred in some of
the nodes previously recognized to be the
potential sites for the occurrence of strong
events.
9Morphostructural zonation of the Italian area
- The morphostructural map of peninsular Italy,
Alps and Dinarides (at a scale 11,000,000) has
been compiled based on - topographic maps
- tectonic maps
- satellite photos
- relevant publications
- Special attention was paid to the present-day
topography.
10GORSHKOV A.I., Panza G.F., Soloviev A.A. Aoudia
A. (2002) - Morphostructural zoning and
preliminary recognition of seismogenic nodes
around the Adria margin in peninsular Italy and
Sicily. JSEE Spring 2002, 4, No.1, 1-24.
Lineaments and Nodes in peninsular Italy and
Sicily
- Lineaments
- First rank
- Second rank
- Third rank
Gorshkov et al. (2002)
11Observed epicentres and Nodes Prone to
Earthquakes with M?6.0 in peninsular Italy and
Sicily
Published in Gorshkov, Panza, Soloviev, Aoudia ,
JSEE, 7 (2002) http//www.iiees.ac.ir/English/Publ
ication/abs/jsee7_1.html
First presented at IUGG 1999 - Birmingham, UK
12Observed epicentres and Nodes Prone to
Earthquakes with M?6.5 in peninsular Italy and
Sicily
Published in Gorshkov, Panza, Soloviev, Aoudia ,
JSEE, 7 (2002) http//www.iiees.ac.ir/English/Publ
ication/abs/jsee7_1.html
First presented at IUGG 1999 - Birmingham, UK
13GORSHKOV A.I., Panza G.F., Soloviev A.A. Aoudia
A. (2003) - Identification of seismogenic nodes
around the Adria margin case study for the Alps
and Dinarides
Morphostructural zonation of the Alps
- Lineaments
- First rank
- Second rank
- Third rank
Gorshkov et al. (2003)
14Observed epicentres and Nodes Prone to
Earthquakes with M?6.0 in the Alps
Nodes circles with 25 Km radius The five
parameters that compose the decision rule are
essential for the recognition. They are the
distance, L, between the points with Hmax
(maximum topogaphic altitude) and Hmin (minimum
topographic altitude), the relief gradient
(DH/L), the minimum value of the Bouguer anomaly
(Bmin), the distance to the nearest node (Dn),
the difference between maximum and minimum values
of the Bouguer anomaly (DB), observed inside the
node
Gorshkov et al. (2003)
15Algorithm M8
16Algorithm M8 in Italy
Algorithm M8 was designed for the prediction of
the strongest earthquakes worldwide, with
magnitude 8.0 and above (Keilis-Borok and
Kossobokov, 1987), and was later adapted for the
prediction of earthquakes with smaller magnitudes
(Keilis-Borok and Kossobokov, 1990).
17Algorithm M8 in Italy
The M8 algorithm analyses the seismic activity
inside a set of Circles of Investigation, CIs,
with radius normalized by the linear size of the
events to be predicted, i.e. proportional to
magnitude threshold M0. A hierarchy of
predictions is usually delivered for different
magnitude ranges M0, considering values of M0
with an increment of 0.5 (i.e. M0 indicates the
magnitude range M0 ? M ? M00.5).
18Algorithm M8 in Italy
For the monitoring of seismicity in the Italian
territory, a new spatially stabilized variant of
the algorithm M8, namely M8s algorithm, is used,
where the seismicity is analysed within a dense
set of overlapping circles covering the monitored
area (Kossobokov et al., JSEE 2002). In Italy
predictions are performed in the three different
magnitude ranges defined by M6.5, M6.0 and
M5.5.
19Algorithm M8s in Italy steps of the analysis
- The territory is scanned with a set of small
circles distributed over a fine grid, with the
radius of the small circles approximately equal
the grid spacing and to the linear dimensions of
the source of target events. - The seismically active grid points are then
selected by the condition that the average annual
rate of seismic activity, within the small
circle, is above a given threshold. - The grid points where data are insufficient for
the application of M8 algorithm and isolated grid
points are excluded.
20Algorithm M8s in Italy steps of the analysis
- The M8 algorithm is then applied with the circles
of investigations, CIs, centred at each of the
selected grid points. - An alarm is declared for a CI only if the
overwhelming majority (more than 75) of the CIs
centred at the neighbouring grid points are also
in state of alarm.
21M8s algorithm in Italy
- Magnitude
- M? 6.5
- Radius of CI
- 192 Km
- Monitored region
- Alerted region
Predictions as on 1-7-2003 (subject to update
on 1-1-2004)
22M8s algorithm in Italy
- Magnitude
- M? 6.0
- Radius of CI
- 138 Km
- Monitored region
- Alerted region
Predictions as on 1-7-2003 (subject to update
on 1-1-2004)
23M8s algorithm in Italy
- Magnitude
- M? 5.5
- Radius of CI
- 106 Km
14/09/2003 Mmax5.6
- Monitored region
- Alerted region
Predictions as on 1-7-2003 (subject to update
on 1-1-2004)
24- All together there are 22 strong events (sum over
the three M ranges) 14 of them are predicted
(about 64) - The space time volume occupied by alarms is
around 40, in each of the three magnitude ranges
Space-time volume of alarm in M8 application in
Italy
Algorithm M8 predicted 64 of the events occurred
in the monitored zones in Italy, i.e. 14 out of
18 events occurred within the area alerted for
the corresponding magnitude range
25Algorithm CN
26Seismotectonic zoning of Italy
- Seismotectonic model defined by GNDT (Gruppo
Nazionale per la Difesa dai Terremoti) - (Meletti et al., 2000)
27Rules for the definition of CN regions according
to the seismotectonic model
A single region includes 1. adjacent zones
with the same seismogenic characteristics (e.g.
only compressive or only extensive) 2. zones
with transitional properties. A transitional
zone is included in a region if 1. it is
between zones of the same kind 2. it is at the
edges of the region and the space distribution of
the aftershocks reveals a possible
connection. (Peresan, Costa Panza., 1999)
28NorthernRegion
14/09/2003 M5.5
- Alerted area
- Alarm Periods
TIPs
Strong Earthquakes predicted
- Prediction of the events with M?5.4
- Updated to January 2004
- (next update March 2004)
- TIP 29.4 of total time
29Central Region
- Alerted area
- Alarm Periods
TIPs
Strong Earthquakes predicted
- Prediction of the events with M?5.6
- Updated to January 2004
- (next update March 2004)
- TIP 23.6 of total time
30SouthernRegion
- Alerted area
- Alarm Periods
TIPs
Strong Earthquakes predicted
Failure to predict
- Prediction of the events with M?5.6
- Updated to January 2004
- (next update March 2004)
- TIP 29.0 of total time
31The regionalization based on the seismotectonic
model
- The regionalization of the Italian territory
strictly based on the seismotectonic zoning
permits to reduce the spatial uncertainty of
predictions. A general reduction of the
percentage of total TIPs has been observed as
well. - The hypothesis, involved by the regionalization
based on the seismotectonic model, that
precursors can be found inside seismogenically
homogeneous areas associated to a dominating
geodynamic process, seems supported by the
corresponding improvement of results.
32The regionalization based on the seismotectonic
model
- The seismotectonic model, supported by kinematic
arguments, can be viewed as a useful tool that
permits to optimise the selection of the fault
systems involved in the generation of strong
earthquakes.
33AdriaticRegion
Alarm Periods
Alerted Area
TIPs
Strong Earthquakes predicted
Failure to predict
- Prediction of the events with M?5.4
- Updated to September 2003
- (next update November 2003)
34Space-time volume of alarm in CN application in
Italy
Algorithm CN predicted all of the events
occurred in the monitored zones in Italy, i.e.
each event was preceded by alarms in at least one
region.
35Updated predictions, at national and global
scale, are available at the following web
site http//www.ictp.trieste.it/www_users/sand/pr
ediction/prediction.htm To avoid misuse, the
access to current predictions is restricted to
experts (password required).
36 CN algorithm and long lasting changes in
reported magnitudes the case of Italy
37CN algorithm and long lasting changes in
reported magnitudes
38CN algorithm and long lasting changes in
reported magnitudes
Time diagrams of the standard CN functions
obtained for the Central region(Peresan et al.,
1999) Functions Sigma, Smax and Zmax and are
evaluated for 4.2?M?4.6, functions K, G, N3, q
for M?4.5 and functions N2 for M ?5.0
39Catalogues comparison
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
CCI1996
Common events ?T?1min ?Lat,
?Lon?1(Storchak, Bird and Adams, 1998)
40Magnitude comparison Central Region
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Duration Magnitude
Md3.0
Yearly Average differencesNEIC-ING
ML3.0
Local Magnitude
41Magnitude comparison Central Region
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Local MagnitudeDifferences
42CN algorithm and long lasting changes in
reported magnitudes
Time diagrams of the standard CN functions
obtained for the Central region(Peresan et al.,
1999) MML(ING)0.5
43Magnitude comparison Northern Region
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Yearly Average differencesNEIC-ING
Local Magnitude ?MLML(NEIC)- ML(ING)
Average difference
44Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Events used for Md analysisYearly number of
common events used for the comparison between ING
and NEIC catalogues Events used for ML
analysis
45Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
?Md
Frequency scatter-plots of ?Md and ?ML versus the
corresponding NEIC magnitude
?ML
46Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
?Md
?ML
47Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
?ML0
All
48Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Duration MagnitudeNEIC - ING
LocalMagnitudeNEIC - ING
49Magnitude comparison Italian territory
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
LDG - ING
ML3.0
Local MagnitudeYearly Average differences
ML3.0
NEIC - LDG
50CN algorithm and long lasting changes in
reported magnitudes
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, reported
by ING and NEIC, indicates, since 1987, an
average underestimation of about 0.5 in the Local
Magnitude provided by ING.
51CN algorithm and long lasting changes in
reported magnitudes
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
- The presence of a general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. -
- (Peresan, Panza Costa, GJI 2000)
-
- (Gasperini, Vannucci Orlanducci
Rivalutazione della magnitudo per i terremoti
italiani nel periodo post 1980. In Catalogo
strumentale dei terremoti Italiani dal 1981 al
1996 2001)
52Compilation of a homogeneous updated catalogue
for CN monitoring in Italy
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
Databases available to us CCI1996 PFG
revisedING bulletins (Italian catalogue,
available up to July 1997) Priority ML, Md,
MI NEIC PDE Preliminary Determinations of
Epicenters from NEIC (global catalogue). Priority
to be defined (available M mb, MS, M1,
M2) ALPOR Catalogo delle Alpi Orientali (local
catalogue for eastern Alps) Priority ML, MI
53Compilation of a homogeneous updated catalogue
for CN monitoring in Italy
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
- Procedure
- Study of the completeness of PDE catalogue
- Study of the relations between different kind of
magnitudes reported in the CCI1996 and PDE
catalogues - Formulation of a rule for the choice of magnitude
priority in PDE, similar to the priority used for
CCI1996 - Construction of the Updated catalogue,
integrating CCI1996, ALPOR and NEIC data
(compatibly with the completeness of NEIC).
54Operating magnitude selection
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
M(CCI1996) mb(PDE)
M(CCI1996) M1(PDE)
55Operating magnitude selection
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
M(CCI1996) MS(PDE)
M(CCI1996) M2(PDE)
56Operating magnitude selection
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
57Operating magnitude selection
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
Central Italy
M(CCI1996) M2(PDE)
Southern Italy
58The Updated Catalogue of ItalyUCI2001
59The Updated Catalogue of Italy
60Stability of intermediate-term earthquake
predictions with respect to random errors in
magnitude the case of Central Italy
61CN is essentially based on the information
contained in the earthquake catalogues. It
analyses the seismic flow using origin time,
epicentral coordinates and magnitudes of
earthquakes. All catalogues are inevitably
affected by errors. Magnitudes are characterised
by the most significant errors, conditioning both
aftershocks removal and seismic flow.
Systematic Errors
Random Errors
How random errors on reported magnitudes affect
CN prediction results?
62Randomised Magnitude
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
Mc operating magnitude ?MI measurement error
?MD error of discretisation
Measurement random errors
P(?MI) FTR(?MI) Truncated normal probability
distribution with FTR(?MI) F(?MI)/2F(?Mmax)-1
Mmax maximum assumed error on magnitudes
Errors of magnitude discretisation
P(?MD) Uniform probability distribution
Interval -d/2 d/2) ddiscretization
step0.1
63Magnitude randomisation
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Discretization step
Discretisation step d0.1
Distribution of the number of events versus
magnitude for the original and the randomised
catalogues
64Results obtained with the original catalogue
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
TIPs obtained with the original catalogue with a
different length of the thresholds setting
period (Learning)
- Prediction of the events with M?5.6
65Results obtained with the randomised catalogue
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
- ? Original catalogue
- ? All randomised catalogues
- Average of randomised
Long Learning
Short Learning
66Results obtained with the randomised catalogue
- N is the number of strong earthquakes occurred
during the time period T covered by predictions - The alarms cover altogether the time t and they
have missed n strong events
?n/N the rate of failures-to-predict ?t/T
the rate of time of alarms
67Results obtained with the randomised
cataloguevariation of target events
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
List of earthquakes with Central Italy
1950-1999
68Results obtained with the randomised catalogue
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Average differences in prediction errors
OC Original Catalogue RCs Randomised Catalogues
69Stability of TIPs diagnosis
- DST
- In order to perform the magnitude comparison, the
events common to the different catalogues are
identified according to the following rules a)
time difference 1 minute b) epicentral
distance 1 for the comparison with the global
catalogue (Storchak, Bird and Adams, 1998). No
limitation is imposed to magnitude or depth
differences.
Central Italy (Long threshold setting period)
? percentage of tests for which the recognition
of the time t does not change with respect to its
average value 0???50
?(t) percentage of tests for which the time t
belongs to a TIP
70Stability of CN predictions with respect to
random errors in magnitude
- DST
- The analysis of CN functions in Central Italy
allowed us to detect a relevant long lasting
change in the reported magnitudes. - The comparison of individual magnitudes, and ,
reported by ING and NEIC catalogues indicates,
since 1987, an average underestimation of about
0.5 in the provided by ING. - The hypothesis of general local magnitude
underestimation in the Italian ING bulletins is
substantiated by the cross-comparison performed
between ING, LDG and NEIC catalogues. - The presence of the evidenced magnitude change
prevents the use of ING bulletins for CN
algorithm application and makes necessary to use
global data like NEIC.
- The results of prediction remain stable for
?Mmaxlt0.3. - To guarantee the stability of the results, the
thresholds setting period must be long enough to
include a significant sample of dangerous and non
dangerous intervals of time. - The quality of predictions is mainly controlled
by the percentage of failures to predict, which
depends on the changes in the number of strong
earthquakes. - The identification of TIPs is very stable during
most of the time and the randomisation does not
introduce spurious alarming patterns associated
with the occasionally strong events.