radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio ....

8
GNGTS 2015 SESSIONE 2.1 35 RADON TIME SERIES IN DIFFERENT SITE: FURTHER INSIGHTS A. Riggio 1 , A.Tamaro 2 , M.Santulin 3 , F. Gentile 1 1 Istituto Nazionale di Oceanografia e Geofisica Sperimentale – OGS, Trieste, Italy 2 Dottorando dell’Università di Udine presso l’OGS, Trieste, Italy 3 Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Milano, Italy Introduction. In the framework of the UR. 2 of the project DPC-INGV S3 “Short term earthquake prediction and preparation“ (Albarello, 2013), a database, containing radon data collected and provided by scientific institutions or volunteers, accompanied, when available, by meteorological data recorded by the radon instruments, or from meteorological stations located near the site of detection of radon, was created. The database has 3,961 sites and more than one million records (Martinelli et al., 2013; Riggio et al., 2013). The collected data are related to radon acquired in the air, soil and water, for different purposes (health, environmental, seismological). Consequently, also the types of instruments are different and can be used in active or passive mode, in the first case recording the amount of radon present in the environment, and in the second one aspiring radon through an incorporated pump. They consist of commercially available monitors, monitor prototypes from scientific institutions or amateurs, or by commercially available dosimeters. Another important information to evaluate the data, reported in the database, is given by the characteristics of the site, that is, if the instrument is located in a well, in a basement, in homes, or on the ground, with a probe that aspires in the soil. The sampling time is between ten minutes and twelve hours for the measures considered continuous, and between fifteen days and one year for the discretized ones. When the measurements were made in water, the temperature, conductivity, pH and EH values, if available, are listed as well. For all types of measurement, if available, also the meteorological data are presented. The key points of the data analysis were the definition of anomaly and its characteristics (Precursor Time, Amplitude, Duration), the estimation of the influence of environmental parameters and the way to select predictable earthquakes. Literature on the possible connection between the variations of radon and the crustal deformation is considerable, because radon was identified as a possible precursor already in 1920 (Petrini et al., 2012; Riggio et al., 2013; Riggio and Santulin, 2015; Martinelli et al., 2015). Once established a definition of anomaly and of how to obtain it from the experimental data, the first approach was that of identifying univocal criteria for the determination of anomalies (Riggio et al., 2014). When anomalies are systematically studied, it is evident that the process of preparation and occurrence of large earthquakes influences the characteristic of the anomaly, in particular the Duration and the Precursor time of the observed anomalies, that can range, according to earthquakes characteristics, from one to several years in the phase of long-medium term,

Transcript of radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio ....

Page 1: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

GNGTS 2015 sessione 2.1

35

radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio1, A.Tamaro2, M.Santulin3, F. Gentile1

1 Istituto Nazionale di Oceanografia e Geofisica Sperimentale – OGS, Trieste, Italy2 Dottorando dell’Università di Udine presso l’OGS, Trieste, Italy3 Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Milano, Italy

Introduction. In the framework of the UR. 2 of the project DPC-INGV S3 “Short term earthquake prediction and preparation“ (Albarello, 2013), a database, containing radon data collected and provided by scientific institutions or volunteers, accompanied, when available, by meteorological data recorded by the radon instruments, or from meteorological stations located near the site of detection of radon, was created. The database has 3,961 sites and more than one million records (Martinelli et al., 2013; Riggio et al., 2013). The collected data are related to radon acquired in the air, soil and water, for different purposes (health, environmental, seismological). Consequently, also the types of instruments are different and can be used in active or passive mode, in the first case recording the amount of radon present in the environment, and in the second one aspiring radon through an incorporated pump. They consist of commercially available monitors, monitor prototypes from scientific institutions or amateurs, or by commercially available dosimeters. Another important information to evaluate the data, reported in the database, is given by the characteristics of the site, that is, if the instrument is located in a well, in a basement, in homes, or on the ground, with a probe that aspires in the soil. The sampling time is between ten minutes and twelve hours for the measures considered continuous, and between fifteen days and one year for the discretized ones.

When the measurements were made in water, the temperature, conductivity, pH and EH values, if available, are listed as well. For all types of measurement, if available, also the meteorological data are presented. The key points of the data analysis were the definition of anomaly and its characteristics (Precursor Time, Amplitude, Duration), the estimation of the influence of environmental parameters and the way to select predictable earthquakes.

Literature on the possible connection between the variations of radon and the crustal deformation is considerable, because radon was identified as a possible precursor already in 1920 (Petrini et al., 2012; Riggio et al., 2013; Riggio and Santulin, 2015; Martinelli et al., 2015).

Once established a definition of anomaly and of how to obtain it from the experimental data, the first approach was that of identifying univocal criteria for the determination of anomalies (Riggio et al., 2014).

When anomalies are systematically studied, it is evident that the process of preparation and occurrence of large earthquakes influences the characteristic of the anomaly, in particular the Duration and the Precursor time of the observed anomalies, that can range, according to earthquakes characteristics, from one to several years in the phase of long-medium term,

Page 2: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

36

GNGTS 2015 sessione 2.1

before a strong earthquake occurs, and from one day to months, usually in the short term and impending phase (Zhang et al., 1996). In this paper, we show the methods used to identify the effects of meteorological phenomena on the radon values recorded in various time series, and the subsequent removal of these effects.

Data analysis. The aim of the project was to verify if the observables multiparametric monitoring could represent a tool to improve the knowledge about the seismic hazard (Albarello, 2015). The priority areas, object of study, are the Po Plain and the southern Apennines. The data provided by the various institutions were homogenized according to previously defined metadata sheet. Among all the data contained in the database, those with the longest period of acquisition and the smallest sampling intervals were taken into consideration. Are at least six institutions have provided data acquired continuously for a time period greater than or equal to a year, in one or more sites, for a total of fifteen time series. Ten of these are located within a maximum radius of 300 km from the Project S3 priority area of the Po Valley (Italiano et al., 2012). The series here considered were acquired in the soil, but with different instruments and sampling intervals.

The characteristics of the sites which have produced the longest time series are reported In Tab. 1.

The Cazzaso survey site, managed by the Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS), is equipped with a continuous radon recording, Lucas scintillation cell type. The operating modalities are both continuous and grab sampling. The air is inhaled by pumping, from a 40.5 m deep well with a 9 cm diameter, at a 7 m depth. The groundwater level is about 17 m deep from the top of the well. The well is closed by a gully-hole (Riggio and Sancin, 2005). The site is located between the Alpine and Dinaric structures.

The sites in the Novara zone, managed by the Geophysical Observatory of Novara, are located on both sides of the Cremosina fault. The sensors are placed in wells, that in the case of Pogno and Gozzano were placed in basements well closed and automated.

Torre Pellice site, managed by Seismic Precursor Study Center (SPSC), and Rocca di Papa site, managed by the Istituto Nazionale di Geofisica e Vulcanologia (INGV), are located in basements whereas Prato site, managed by Prato Research and CAI - Section of Prato, is located in a cave.

In the first phase of the project, the longest time series with the most continuous acquisition time were analysed, each with their time sampling (between ten minutes to twelve hours), identifying as anomalies those values that exceed the limit of 2-sigma (Wakita, 1980). In Tab. 2, an example related to the site of Cazzaso is shown. Earthquakes in the table have been recorded

Tab. 1 - Characteristics of continuous monitoring Radon sites, in central-northern Italy.

SITE NAME MEASUR.UNIT ACQUIS.MOD. SITE TYPE INSTRUM.

BORGOLAVEZ. (NO) counts/h soil well Prototype

CASALINO (NO) counts/h soil well Prototype

CAVAGLIA (NO) counts/h soil well Prototype

GOZZANO (NO) counts/h soil well Prototype

NOVARA counts/h soil well Prototype

POGNO (NO) counts/h soil well Prototype

CAZZASO (UD) Bq/m3 soil well PrassiSilena

TORRE PELICE (TO) Bq/m3 air basement Geoex 1027

ROCCA DI PAPA (RM) Kbq/m3 soil soil AlphaGuard

PRATO Bq/m3 air cave AlphaGuard

Page 3: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

GNGTS 2015 sessione 2.1

37

by the seismic network of Friuli Venezia Giulia – OGS, Trieste (OGS, 2002 – 2014) and selected according to Hauksson and Goddard (1981). They formulated the experimental relationship between the magnitude of earthquakes and the maximum distance so that transient phenomena may occur in the fluids circulating in the underground (Anomalies). Shallow earthquakes were used to formulate the relationship.

The selected events have to satisfy the following condition:M ≥ 2.4 log10 D - 0.43

where M is the minimum magnitude required to obtain a radon anomaly at distance D (km).For the anomalies detected before earthquakes, the characteristic parameters were extracted:

Amplitude, Precursor Time, and Duration. From the experimental data, theoretical anomalies have been determined in order to verify the possibility to find a correspondence with occurred earthquakes that, in case, could be considered as potentially predictable. For a more accurate evaluation of the data, it was decided to remove from the time series the effects caused by meteorological parameters, such as, in particular, temperature and pressure changes.

Since the final objective is a joined interpretation, first among radon data from different sites, and then, among all the observable analysed within the project, homogenization has been carried both of radon data, as of meteorological one, through a daily sampling.

Various methods to remove any spurious effects are reported in the literature (Zmazek et al., 2010; Gregoric et al., 2012; Gualadini, 2014; Riggio et al., 2014); the difficulty of finding a standard procedure also derives from the variety of data types and different characteristics of the sites.

The first approach to analyse some considered time series was based on the possible occurrence a mathematical correlation between radon and the temperature and pressure values, measured by the instrument itself or in meteorological stations located a few kilometres away. Correlation coefficients never exceeded 0.2, indicating a very poor influence of meteorological parameters.

Only in the case of the site of Prato, a strong seasonal component was evident. To this case a methodology was applied which consisted of a cross-correlation computation between the original function and the function calculated by averaging the data daily, over one year, for all the years. The correlation coefficient was found to be greater than 0.8. The original function was then corrected by removing the seasonal component by the following operation:

r = a – b * M.S.D. (a)/ M.S.D. (b) * c.c.

where a is the original function, b the corrected function, M.S.D. is the mean square deviation and c.c. the correlation coefficient.

In the corrected function the same anomalies calculated for the original function have been found.

This approach, however, is not applicable to all types of data: in particular, neither in the lack of a seasonal variation nor in case of series shorter than one year.

The following step was to use other multiparametric statistical tools, with specific attention to the Principal Component Analysis (PCA), in order to remove the meteorological effects by the time series of each site and to get series which consist of geodynamic interest information. The correct series have been, then, correlated to each other, and the joint analysis with the other observables included in the project DPC-INGV S3 is in progress. Results are expected to allow the formulation of large scale models and to give information about the investigating scale in space and time. To this purpose, the program ORIGIN 9.0 (Microcal Software, 2009) was used, which allowed to apply the PCA methodology (Joliffe, 2002; Shlens, 2009; Bailey, 2012) to the radon series and to the time series of temperature and pressure. Only those periods were considered in which the temperature and pressure values were present along with the radon value.

Page 4: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

38

GNGTS 2015 sessione 2.1

Tab. 2 - Anomalies detected in the Cazzaso time series, along with earthquakes recorded by the Friuli Venezia Giulia seismometric network and localized by OGS, selected according to Hauksson and Goddard (1981).

ANOMALY Start Time End Time Earthq. Mag LAT LON Epicentral zone Date (MD)

2002-2004

1 13/11/02 20/11/02 27/11/02 3.3 46.42 12.67 Forni di Sotto (Friuli)

2 30/06/03 14/07/03 30/08/03 3.7 46.30 12.77 Tramonti di Sotto (Friuli)

3 24/12/03 02/07/04 12/07/04 5.1 46.30 13.63 Kobarid (SLO)

4 24/08/04 25/08/04 28/08/04 3.3 46.39 12.85 Villa Santina (Friuli)

29/08/04 3.8 46.35 12.70 Fornidi Sotto (Friuli)

07/10/04 3.4 46.41 13.12 MonteSernio (Friuli)

24/11/04 5.1 45.68 10.60 Gargnano (Lomb.)

2004-2005 14/01/05 4.1 46.20 14.03 Podbrdo (SLO)

5 04/02/05 04/03/05 23/03/05 3.3 46.20 12.73 Maniago (Friuli)

6 01/07/05 12/07/05 12/12/05 3.4 46.50 13.37 Malborghet. (Friuli)

2006-2007

7 11/08/06 31/03/09 11/08/06 3.1 46.31 13.14 Gemona ( Friuli)

26/02/07 3.9 46.25 12.52 Claut (Friuli)

26/02/07 3.7 46.25 12.51 Claut (Friuli)

2008-2009 29/02/08 3.7 46.31 13.00 Trasaghis (Friuli)

23/12/08 5.5 44.52 10.23 Nevi. D. Ard. (Emilia)

12/03/09 3.1 46.33 13.28 Chiusaforte (Friuli)

06/04/09 6.1(1) L’Aquila (Abruzzo)

2011

8 11/03/11 18/03/11 04/07/11 3.1 46.39 12.90 Villa Santina (Friuli)

17/07/11 5.0 44.92 11.23 S. Mrt. In Spi. (Emilia)

9 07/09/11 07/09/11 29/10/11 4.4 45.70 10.95 M. Lessini (Trentino)

2012

10 15/05/12 16/05/12 20/05/12 5.8(1) 44.76 11.17 Camposanto (Emilia)

20/05/12 5.1 44.78 11.42 S. Agost. (Romagna)

29/05/12 5.6(1) 44.77 11.06 S. Prospero (Emilia)

29/05/12 5.3(1) 44.79 10.88 Carpi (Emilia)

09/06/12 4.4 46.19 12.46 Barcis (Friuli)

11 23/07/12 24/07/12

2013

12 14/01/13 20/01/13 25/01/13 4.8 44.16 10.45 Pianura Veneta

13 11/02/13 13/02/13 12/02/13 3.8 46.30 12.55 Claut(Friuli)

(1) The reported magnitude is MW (ISIDe, 2009, 2012).

The time series values were normalized and standardized. The PCA analysis has highlighteda low contribution of temperature and pressure to the original functions. The correct function was obtained by summing the products of the Principal Components multiplied by the coefficients related to radon. The original function corrected has been rebuilt, with reverse process, multiplying the correct function by the standard deviation and adding the media.

The difference between the original function and the correct one is very low. For example,

Page 5: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

GNGTS 2015 sessione 2.1

39

in the case of the Cazzaso site, the difference is in the order of 5%, showing no dependence of radon values from atmospheric influence. The temperature data have been recorded by the Tolmezzo meteorological station (5 km away), managed by Friuli-Venezia Giulia Region (www.meteo.fvg.it), and the atmospheric pressure by the amateurs site MeteoMin of Udine (www.meteomin.it). The radon values over the 2-sigma value are the same in the original series and in the corrected series (Fig. 1). The anomalous period between 2006 and 2008 was confirmed also in the corrected time series of Cazzaso.

Fig. 1 – Radon in soil time series acquired at Cazzaso -Friuli- (blue line), corrected by meteorological parameters by PCA method (brown line), Temperature (heavenly line) and Pressure (pink line). The 2 - sigma over different periods are marked as reported in Legenda.

Tab. 3 - Characteristics of the anomalies detected in the Cazzaso time series.

ANOMALY PERIOD DURATION AMPLITUDE MAGNITUDE PRECURSOR TIME Distance (MD) (km)

14-11-2002/18-11-2002 5 days 78.43% 3.3+3.7 4 months < 20

30-06-2003/10-07-2003 10 days 42.94% 3.7 61 days < 30

24-12-2003/02-07-2004 6 months + 8 days 755.5% 5.1 + 5.1 6,5 months / < 250 11 months

24-08-2004/25-08-2004 1day 25.9% 3.3 + 3.8 5 days < 30

04-02-2005/04-03-2005 1 month 109% 3.3 1 month + 19 days < 40

02-07-2005/13-07-2005 11 days 64.72% 3.4 5 months + 10 days < 35

11-08-2006/31-03-2009 2 years + 7 months 328.39% 6.1(1) 2 years + 7 months + < 500 26 days

17-02-2007/18-02-2007 2 days 201.46% 3.9 + 3.7 9 days < 50

16-02-2008/18-02-2008 3 days 129.03% 3.7 13 days < 25

15-07-2008/17-07-2008 2 days 167% 5.5 5 months + 8 days < 350

11-03-2011/18-03-2011 7 days 298.65% 3.1 + 5.0 4 months + 6 days < 350

07-09-2011/07-09-2011 1 day 182% 4.4 1 month + 22 days < 350

15-05-2012/16-05-2012 2 days 167.55% 5.8(1) 5 days < 250

14-01-2013/20-01-2013 6 days 493% 4.8 11 days < 400

11-02-2013/13-02-2013 3 days 684% 3.8 1 day (cosismic) < 40

(1) The reported magnitude is MW (ISIDe, 2009; 2012).

Page 6: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

40

GNGTS 2015 sessione 2.1

The characteristic parameters of each anomaly were obtained by this first analysis (Tab. 3). The considered anomalies are those in which radon values exceed the value of 2-sigma. The Duration is the time interval, expressed in days, months, years, between the start and the end of the anomaly. The Amplitude is the maximum variation, per cent, of the radon value over the 2-sigma value. The Precursor Time is the period between the start time of the anomaly and the occurrence of the linkable earthquake. Distance is the longest distance between the radon monitoring site and the epicentre of the linkable earthquakes.

In the other radon time series, an influence of the temperature was detected, but it does not affect the determination of the anomalous periods. For example, the PCA analysis applied to the radon series acquired at Torre Pellice site did not eliminate, in the correct function, the anomalies detected in the original series, but rather made them more evident.

Although the results obtained from the analysis of PCA seem to suggest that, although at first sight, temperature and pressure may influence the radon behaviour, there is not an objective correspondence between the meteorological parameters and radon values.

As the radon feature, present in most of the data, is a seasonal variation, and considering that the work was also aimed at finding a method, applicable to all the acquired series to clean up them from the influence of meteorological parameters, we tried to identify the seasonal component for all the studied series and to find the most suitable method to eliminate it from them. We chose the Program AutoSignal V1.5 (Systat Software, 2004) for Windows. It lets you automate spectral analysis, time-domain analysis, and signal processing without programming. To use the application, you have to select the analysis techniques, the algorithm and the options from a menu or toolbar . The software then provides immediate feedback via 2- and 3-D graphs and numeric summaries. Built-in procedures include FFT, autoregressive moving average, complex exponential modelling, minimum-variance methods, eigenanalysis frequency estimation, and wavelets. A library of six Fourier-spectrum methods lets you see a complete picture of the frequency space. A set of 30 data-tapering window functions allows you to solve the leakage problem of standard FFTs. Lomb-Scargle Fourier-domain analysis techniques permit you to handle unevenly sampled data.

For the considered time series, the program also highlighted an important content at frequencies corresponding to less than one year periods, but only the annual component, attributable to a seasonal variation, was eliminated. It has been avoided removing true anomalies attributable to geodynamic causes. With this method as well, the corrected series shows a variation with respect to original series that does not exceed 2%. Once again, the anomalies detected on the series consisting of the original data without any type of correction are present as well in the corrected series (Fig. 2).

Fig. 2 – Daily averages of radon time series acquired at Cazzaso -Friuli-: original data (blue line), data corrected for the seasonal component (red line): The 2-sigma over different periods are marked as reported in legenda.

Page 7: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

GNGTS 2015 sessione 2.1

41

The corrected daily mean radon series will be published in the database of the S3 project. For this purpose, a dedicated metadata sheet, in which the time series is described in its content and format, was compiled.

The time series depurated from such component have been treated again with the PCA method, to verify possible correlations between the various sites. The Prato series was corrected by both the methods of cross-correlation and spectral analysis.

Fig. 3 shows the diagram of the various sites referred to the PCA1 and PCA3. The sites that seem to have common characteristics are Gozzano, Pogno, Cazzaso and Borgolavezzaro. On the other hand, these are the sites where the sensors are installed in the basement and at least at 1 m depth, or in deep wells.

Conclusions. The anomalies that greatly exceed the limit of 2-sigma are not cancelled by the procedures used to clean meteorological phenomena. This result, as well as the presence of anomalies at the same time at different sites, supports the possibility of a correlation between anomalous geochemical transients and deformations preceding earthquakes.

Acknowledgements. This research was developed in the frame of the Seismological Project S3 (2014-2015), This study has benefited from funding provided by the Italian Presidenza del Consiglio dei Ministri – Dipartimento della Protezione Civile (DPC). This paper does not necessarily represent DPC official opinion and policies.

ReferencesAlbarello D.; 2013: Project S3 - Short-term earthquake prediction and preparation. Final Report, https://sites.google.

com/site/ingvdpc2012progettos3/home. pp. 1 - 31.Albarello D.; 2015: Project S3 – Short-term earthquake prediction and preparation. Final Scientific Report. pp. 1

– 31.Bailey S.; 2012: Principal Component Analysis with Noisy and/or Missing Data. Publications of the Astronomical

Society of the Pacific 124.919 (2012), pp. 1015 -1023.Gregoric A., Zmazek B., Dzeroski S., Torkar D., Vaupotic J.; 2012: Radon as an Earthquake Precursor – Methods

for Detecting Anomalies. Earthquake Research and Analysis – Statistical Studies, Observations and Planning, Sebastiano D’Amico (ED), (460), pp. 179 – 196.

Gualadini A.; 2014: Analisi delle Componenti Principali: cenni storici. Seminario all’interno del progetto S3, Bologna, 10/07/2014, pp. 1 – 18.

Hauksson E. and Goddard J.G.; 1981: Radon Earthquake Precursor studies in Iceland. J. Geophys. Res. 86, pp. 7037 – 7054.

Fig. 3 – Analysis of PCA for all the considered time series.

Page 8: radon tIme serIes In dIFFerent sIte: Further InsIghts A. Riggio , … · 2017-03-14 · (Riggio . et al., 2014). When anomalies are systematically studied, it is evident that the

42

GNGTS 2015 sessione 2.1

Italiano F., Liotta M., Martelli M., Martinelli G., Petrini R., Riggio A., Rizzo A. L., Slejko F., Stenni B.; 2012: Geochemical features and effects on deep-seated fluids during the May-June 2012 southern Po Valley seismic sequence. Annals of Geophysics, Vol.55, n. 4, pp. 815-821; doi: 10.4401/ag-6151.

Joliffe I.T.; 2002: Principal Component Analysys, Second Edition, (405) Springer, ISBN 0-387-95442-2.Martinelli G., Riggio A., Dadomo A., Italiano F., Petrini R., Pierotti L., Santulin M., Slejko F., Tamaro A.; 2013: D1.2

Database of time series relative to hydrogeochemical and radon observations. DPC-INGV-S3 Project short term earthquake prediction and preparation. Sites.google.com/site/ingvdpc2012progettoS3/documents.Sites.google.com/site/ingvdpc2012progettoS3/documents.

Martinelli G., Dadomo A., Heinicke J., Italiano F., Petrini R., Pierotti L., Riggio A., Santulin M., Slejko F.F., Tamaro A.; 2015: Boll. Geof. Teor. Appl., Vol. 56, n. 2, pp. 115-128. DOI 10.4430/bgta0147.

Meteo.fvg – Osservatorio meteorologico regionale del FVG; 2002 – 2014: Dati di Temperatura. www.meteo.fvg.it.MeteoMin – Stazione meteorologica di Udine; 2004-2013: Dati di Pressione. www. meteoMin.it/Dati_Meteo.Microcal Software, Inc., One Rounhouse Plaza, Northhampton, MA 01060; 2004; OriginPro 9.0OGS; 2002 – 2014: Bollettino della Rete Sismometrica del Friuli – Venezia Giulia. OGS, Trieste.Petrini R., Italiano F., Riggio A., Slejko F.F., Santulin M., Buccianti A., Bonfanti P. and Slejko D.; 2012: Coupling

geochemical and geophysical signatures to constrain strain changes along thrust faults. Boll. Geof. Teor. Appl., 53, pp. 113-134, doi: 10.4430/bgta0017.

Riggio A., Sancin S.; 2005: Radon measurements in Friuli (N.E. Italy) and earthquakes: first results. Boll. Geof. Teor. Appl., Vol. 46, n. 1, pp. 47 – 58.

Riggio A., Santulin M., Tamaro A.; 2013: Radon as seismic precursor in the framework of the S3 Project. In: Atti del 32° Convegno GNGTS, 19-21 novembre 2013, Trieste, Vol. 2, pp. 123-130.

Riggio A., Santulin M., Tamaro A.; 2014: Prima analisi dei dati di radon inseriti nella Banca Dati del Progetto S3. In: Atti del 33° Convegno GNGTS, 25 – 27 novembre 2014, Bologna, Vol.2, pp. 58 – 64.

Riggio A., Santulin M.; 2015: Earthquake forecasting: a review of radon as seismic precursor. Boll. Geof. Teor. Appl., Vol. 56, n. 2, pp. 95-114.

Shlens, J. (2009). A tutorial on principal component analysis. Centre for Neural Science, Salk Institute for Biological Studies. New York University, version 3, pp 1-12.

Systat Software; 2004: AutoSignal V 1.5. Wakita H., Nakamura Y., Notsu K., Noguchi M., Asada T.; 1980: Radon anomaly: a possibile precursor of the 1978

Izu-Oshimakinkai earthquake. Science 207. pp. 882-883.Zhang G., LI X., LI L.; 1996: Research on Earthquake Prediction in China Since the 1980s, in The Selected Papers of

Earthquake Prediction in China. State Seismological Bureau. Seismological Press. Editor in Chief, GE Zhizhou, Beijing, pp. 9 – 18. ISBN 7-5028-1331-4/P.836 (1768).

Zmazek B., Dzeroski S., Torkar D., Vaupotic J., Kobal I.; 2010: Identification of radon anomalies in soil gas using decision trees and neural networks. Nukleonica 2010, 55(4), pp. 501-505.