XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

Retrospective short-term forecasting experiment in Italy based on strong foreshocks

P. Gasperini1,2, B. Lolli2, E. Biondini1, A. Petruccelli1 e G. Vannucci2

1Dipartimento di Fisica e Astronomia, Università di Bologna2Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Bologna

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

In Gasperini et al. (2016, GJI, 207, 150-159) we computed the relativefrequencies with which strong shocks (4.0≤Mw

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

We use the HOmogenized instRUmental Seismic catalog (HORUS) of Italianseismicity from 1960 to present :

from 1960 to 1980, coincides with the dataset by Lolli et al. (2018, BSSA, 108,481-492), which can be downloaded from the electronic supplement.

from 1981 to present, merges various data sources with magnitudeshomogenized to Mw according to Gasperini et al. (2013, BSSA, 103, 2227–2246).

The catalog used here is updated up to Apr 2017, but we are working to a near-realtime procedure to automatically download new data and convertmagnitudes to Mw (Lolli et al., in preparation).

The final catalog will be made available on a public web site.

Dataset

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

The overall magnitude completeness threshold for the period 1960-2014 havebeen assessed by Lolli et al. (2018) and Gasperini et al. (2013) to be about 4.0

However, the completeness threshold might be definitely higher in sea-coveredareas and abroad owing to the large distances from the closest seismic stations(forcedly located on land).

Then for the present study we only consider earthquakes occurred on landand within Italian national borders.

As our interest is to forcast earthquakes that potentially threaten life and goodswe also limit the analysis to shocks shallower than 50 km.

Dataset

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

We issue a forecast allarm of duration Δt within a circle of radius R, starting atthe time (ta) at which a strong shock with Mmin≤M

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

As source areas we consider a tessellation of the Italian territory made of circleswith radius R.

Starting from the first circle centered at latitude 47 and longitude 7 we computethe centers of other circles by moving with steps D=R√2 first in longitude (from 7to 19) and then in latitude (from 47 to 36) so covering the entire definitionrectangle with partial overlap among circles.

Based on the results of our previous analysis (Gasperini et al., 2016, GJI, 207,150-159), we choose a radius R=30 km as a good compromise between theopposing demands of spatial resolution and of sufficient numbers ofearthquakes within the circles so obtaining in all 695 partially overlapping circlescovering the definition rectangle.

We only consider circles including at least one earthquake in HORUS cataloglimited to inland (in all 245)

Setting the forecasting hypothesis

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

In case of seismic sequences it is definitely more useful to predict theoccurrence of the first main shock rather than the later ones

because after a first main shock, cautionary measure issued by civil protectionservices, usually prevent further casualties and injuries in the population.

Hence we also evaluate the ability of our hypothesis to forecast only the firstmain shock of each sequence

eliminating from target events the main shocks (above the threshold Mm)occurred within a time window Δt after and within a spatial distance R from thefirst and any subsequent main shocks of the sequence.

As any shock occurring in such time windows cannot be successfully predicted,we do not issue any alarm during the time intervals Δt after each main shockwithin the circle.

Setting the forecasting hypothesis

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

We retrospectively test the goodness-of-fit of our forecasting hypothesis on theHORUS catalog from 1 Jan 1960 to 27 Apr 2017 by comparing the numbers ofpredicted target shocks (successes) with the total number of target shocksoccurred

We also compare the number of successfull allarms with the total number ofallarms (in different circles)

and the sum of allarm times (normalized for the number of circles) with thetotal duration of the retrospective forecasting experiment (57.3 years)

We compute these counts for both the full set of target main shocks (notdeclustered) and the set of first main shocks (declustered) only.

This test cannot be considered as an evaluation of the efficiency of theforecasting hypothesis (which could only be assessed prospectively on futuredata) but rather a sort of sanity test to check the goodness of fit with past data.

Retrospective testing

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

We establish the same Mw thresholds of Gasperini et al. (2016) for target events: ≥5.0, ≥5.5 and ≥6.0.After a few trials we also choose:

Δt=3 months (or more exactly 0.25 years)

and 4.25≤M

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

Using Δt = 3 months and 4.25 ≤ M < 4.75, we get for Mw≥5.5 (declustered)

Retrospective testing

Year

Month

Day

Lat

Lon

Mw

Δ t (days)

Epicentral area

1962

8

21

41.233

14.933

5.7

0.09255

2.22 h

Irpinia

1968

1

15

37.700

13.100

5.7

0.42544

10.2 h

Belice

1976

5

6

46.250

13.250

6.4

0.00078

67 s

Friuli

1976

9

11

46.300

13.317

5.6

Miss

Friuli

1979

9

19

42.717

12.950

5.8

Miss

Valnerina

1980

11

23

40.800

15.367

6.8

Miss

Irpinia

1984

4

29

43.204

12.585

5.6

Miss

Umbria settentrionale

1984

5

7

41.666

13.820

5.9

Miss

Monti della Meta

1990

5

5

40.650

15.882

5.8

0.00015

13 s

Potentino

1997

9

26

43.023

12.891

5.7

22.10119

Umbria-Marche

1998

9

9

40.060

15.949

5.5

Miss

Appennino lucano

2002

10

31

41.717

14.893

5.7

Miss

Molise

2009

4

6

42.342

13.380

6.3

6.49585

Aquilano

2012

5

20

44.896

11.264

6.1

0.11834

2.8 h

Pianura Emiliana

2016

8

24

42.698

13.234

6.2

Miss

Amatrice

2017

1

18

42.531

13.284

5.7

49.75008

Montereale-Campotosto

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

Molchan diagram

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

ROC diagram

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

ROC diagram

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

We set up a testable forecasting hypothesis for the future occurrence on theItalian mainland of potentially damaging earthquakes (Mw>=5.0, 5.5, 6.0) basedon the previous occurrence of strong shocks (Mw

Thank you for your attention

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

XXXVII Convegno G.N.G.T.S, Bologna 19-21 Novembre 2018

Diapositiva numero 1Diapositiva numero 2Diapositiva numero 3Diapositiva numero 4Diapositiva numero 5Diapositiva numero 6Diapositiva numero 7Diapositiva numero 8Diapositiva numero 9Diapositiva numero 10Diapositiva numero 11Diapositiva numero 12Diapositiva numero 13Diapositiva numero 14Diapositiva numero 15