application of detection probabilities in the idc global
TRANSCRIPT
Application of Detection Probabilities in the IDC Global Phase Association ProcessTormod Kværna(1),Frode Ringdal(1) and Jeffrey Given(2)
(1) NORSAR, Kjeller, Norway ([email protected]), (2) CTBTO Prep Com, International Data Centre, Vienna, Austria
Abstract Examples of Application to SEL3 Event ListsDetection CapabilitiesThe Global Association (GA) process at the IDC is an automated procedure that associ-ates detections by stations in the International Monitoring System (IMS) in order to formevent hypotheses. These hypotheses will later be reviewed by analysts before theReviewed Event Bulletin is issued. We have begun investigating ways to improve theGA process for seismic data, in particular by incorporating amplitude data and stationdetection probabilities in the automatic process. We build on a previous study which hasprovided regional detection capability estimates for individual primary and auxiliaryIMS stations, and use these estimates to develop and test various consistency measures.The purpose of these measures is to provide a means to assess the validity of automati-cally defined seismic events and the consistency of individual phases associated withsuch events. By feeding the results of such assessments back to the GA procedure, weanticipate that the results of the global association can be iteratively improved. Animportant contribution to these assessments will be provided by incorporating the resultsof the continuous IDC Threshold Monitoring process. The initial results of this study arepromising, and this paper will present some case studies illustrating our approach.
We will illustrate the procedure by considering an example: We consider one sta-tion (ARCES) and a specific source area (1.5 degrees within 32N, 104E). Our pur-pose is to estimate the station detection threshold for events from this limitedsource area. From the REB, we obtain a large number of events, some detected byARCES, some undetected (right figure above). We can the estimate the thresholddirectly by maximum likelihood (MLE), using the detection/nondetection infor-mation (Method 1).
We would like to make use of the SNR information to obtain estimates when thereare only few data. For each event detected by ARCES, we scale down the mb by itslog(SNR), to arrive at an instantaneous 'noise magnitude'. We then add a signal-to-noise ratio of 3 (0.5 mb units) to obtain an instantaneous estimate of the ARCESdetection threshold. This way we can obtain an average threshold based on thedetected events. (See the figures to the right above). We denote this Method 2.
Note that the undetected events will cause a bias in the application of Method 2. Forthis reason, we have developed Method 3, as indicated below.
In Method 3, we compensate for the non-detections by applying maximum likelihoodestimation to the scaled-down thresholds. This is is our preferred method and will beapplied where possible (i.e. where information on non-detections is available).
1.5 2 2.5 3 3.5 4 4.5 5 5.5 60
100
200
300
Num
ber o
f eve
nts
Scaled Network mb1MLE
- Detections- Non-detections
Ndet : 722Nnon : 697MLEscale : 3.814Std : 0.311
3 3.5 4 4.5 5 5.5 6
Network mb1MLE
Sign
al-t
o-no
ise
ratio
103
102
101
100
1.5 2 2.5 3 3.5 4 4.5 5 5.5 62.5
3
3.5
4
4.5
5
Network mb1MLE
Scal
ed N
etw
ork
mb1
MLE
MLEscale : 3.814AVGscale : 3.690
0.21
1.07
4.40
10.86
ARCES P-beam, 2.5 - 4.5 Hz
10 15 200 5
SNR
10 15 200 5
Networkmbmx
3.405
3.951
4.444
5.168
SNRmax 4.92
Amp
SNRmax 21.37
SNRmax 45.77
SNRmax 131.34
mbmx - log10(SNRmax)
2.713
2.620
2.780
3.050
1.5 2 2.5 3 3.5 4 4.5 5 5.5 60
0.2
0.4
0.6
0.8
1
1.5 2 2.5 3 3.5 4 4.5 5 5.5 60
50
100
150
200
Magnitude
ARCES Lat: 32oN Lon: 104oE Radius:1.5o Delta: 56.5o
Network mb1MLE
Prob
abili
tyN
umbe
r of e
vent
s
Network mb1MLE
- Detections- Non-detections
Ndet : 722Nnon : 697MLEdet : 3.745Std : 0.271
0˚
0˚
20˚
20˚
40˚
40˚
60˚
60˚
80˚
80˚
100˚
100˚
120˚
120˚
20˚ 20˚
30˚ 30˚
40˚ 40˚
50˚ 50˚
60˚ 60˚
70˚ 70˚ARCES
Map showing the estimated regional event detection thresholds for IMS primarystation AKASG. For each of the 2x2 degree bins, a minimum of 5 events arerequired for calculation of the detection threshold. Distances of 20, 95 and 144degrees from AKASG are illustrated by red dashed lines.
0 20 40 60 80 100 120 140 160 1800
1
2
3
4
5
6
Distance (degrees)
Det
ectio
n Th
resh
old
(mb1
)
AKASG bin size 2.0
115°����°������������������� 20°�����°��������������������� 0°�����°���������������������
The black dots show the estimatedregional event detection thresholdsfor IMS primary station AKASGplotted versus epicentral distance.For each of the distance ranges 0-20degrees, 20-90 degrees, and 115-180degrees, the mb1 amplitude-distancecurve is fitted to the data. The num-bers for each of the distance ranges,given in the lower right box, are:1) the constant offset level of the mb1 amplitude-distance curve2) the standard deviation, and3) the number of bins.In general, the lower the constantoffset level, the better is the overallstation detection capability.
Estimating detection thresholds
The left figure above shows how to compensate for the average bias in theestimate by Method 2. We will need to use Method 2 for estimating thethresholds in those cases when non-detection information is not available.This means outside of the 20 -9 0 degrees distance range for primary stations and at all distance ranges for auxiliary stations
The right hand figure above shows that Method 1 (MLEdet) and Method 3(MLEscale) are consistent when there is sufficient data to obtain reliableestimates by both methods.
The figure above illustrates, as an example, the number of IDC REB events with P-phases reported at IMS primary station AKASG. Only events with depth less than 50 km are considered. The upper map is a global projection where the colors correspond to the number of events in 2x2 degree bins. Distances of 20, 95 and 144 degrees from AKASG are illustrated by red dashed lines. The lower map is an azimuthal projection where the colors correspond to the number of events in 1x1 degree bins. The red dashed circle shows a distance of 20 degrees from AKASG. Ranking of primary stations (top) and auxiliary stations (bottom) in the
distance range 20-90 degrees.
We have developed a scheme to automatically process the individual candi-date events in either SEL1, SEL2 or SEL3. For each event a set of consis-tency measures is calculated, and some of them are shown in the examples.The consistency measures can be used individually or in combination. Themost effective scheme is still to be developed.
In the examples presented here, we have, in addition to calculating consis-tency measures based exclusively on SEL3 bulletin data, also correlated thecandidate events with the events actually accepted in the REB (or LEB).This provides a possibility to evaluate the consistency measures against thedecision of the analyst.
In practical operation, it is intended to apply this processing scheme auto-matically, using the information from the SEL bulletins, the regionalizeddetection thresholds of the IMS stations, and the Threshold Monitoringresults which are automatically computed at the IDC. Candidate eventscould be either accepted or rejected, or in some cases returned to the GAprocess for reanalysis, with one or more phases excluded from consider-ation.
Besides producing improved SEL lists, this processing scheme could also beapplied in an interactive mode to provide the analyst with an assessment ofthe quality of the candidate event with respect to the pattern of detections/non-detections in view of the individual station capabilities.
First we show an example of one of manyexcellent automatic solutions in the SEL3.This is a candidate SEL3 event in theAfghanistan-Tajikistan border region,which has been accepted in the LEB andREB. It satisfies all our main indicators,and the REB location is only 16 km awayfrom the automatic solution. This eventwould be acceptable for analyst reviewwithout any further processing.
Non-detecting operational IMS primary stations
Station DeltaEvent
magnitudeDetectionthreshold
SigmaDetectionprobability
TORD 9.70 3.5363 2.9086 0.3000 0.981800
MKAR 83.73 3.5363 3.7313 0.3300 0.277324
FINES 59.18 3.5363 3.8318 0.3370 0.190282
BRTR 48.14 3.5363 3.8190 0.3000 0.173042
AKASG 52.37 3.5363 3.8287 0.3000 0.164861
GERES 44.78 3.5363 3.8528 0.3070 0.151312
YKA 92.09 3.5363 4.0508 0.4260 0.113574
ARCES 65.59 3.5363 4.0040 0.3740 0.105570
ZALV 84.80 3.5363 3.9853 0.3360 0.090752
LPAZ 65.64 3.5363 4.0100 0.3410 0.082410
ESDC 32.42 3.5363 3.9990 0.3120 0.069051
NOA 55.29 3.5363 4.0030 0.3000 0.059908
KBZ 56.13 3.5363 4.0293 0.3160 0.059384
WRA 139.62 3.5363 4.1460 0.3800 0.054315
ASAR 138.29 3.5363 4.1330 0.3620 0.049652
BOSA 46.69 3.5363 4.0929 0.3150 0.038626
GEYT 65.34 3.5363 4.0750 0.3000 0.036284
KEST 31.76 3.5363 4.2390 0.3460 0.021141
CPUP 59.80 3.5363 4.2120 0.3200 0.017361
BDFB 47.30 3.5363 4.1760 0.3000 0.016499
KMBO 44.03 3.5363 4.3623 0.3360 0.006981
SCHQ 67.56 3.5363 4.3396 0.3150 0.005388
MAW 88.74 3.5363 4.6437 0.3850 0.002011
ROSC 67.83 3.5363 4.5593 0.3340 0.001096
ILAR 102.56 3.5363 4.6168 0.3480 0.000952
KSRS 118.26 3.5363 4.8590 0.3970 0.000432
SONM 99.45 3.5363 4.6297 0.3160 0.000270
MJAR 125.17 3.5363 4.9410 0.3550 0.000038
CMAR 101.86 3.5363 4.9074 0.3420 0.000030
PETK 118.40 3.5363 5.0220 0.3650 0.000023
VNDA 109.40 3.5363 5.3630 0.4290 0.000010
NVAR 102.58 3.5363 4.9104 0.3000 0.000002
USRK 116.42 3.5363 5.1440 0.3370 0.000001
PPT 142.92 3.5363 5.5010 0.3000 0.000000
Number of primary non-detections: 34
Number of stations exceeding third detection probability 22
Number of stations exceeding lowest detection probability 22
Northwest Africa Event only in SEL3
Event id Year Mo Dy Hr Min Sec Lat Lon DepthREBLat
REBLon
REB-SEL3loc. diff
SEL3mb
NewmbMLE
6828087 2010 11 10 03 24 24.400 7.17 -6.10 0.0f - - - 4.23 3.54
2)
Detecting stations
Station Phase DeltaEvent
magnitudeDetectionthreshold
SigmaDetectionprobability
Stationmagnitude
DBIC Pg 1.33 3.5363 1.1059 0.4320 1.000000 -
PLCA P 75.84 3.5363 4.3974 0.3000 0.002051 4.30
ULM P 84.40 3.5363 4.4208 0.3090 0.002102 4.50
TXAR P 93.08 3.5363 4.0911 0.3220 0.042467 3.90
Third highest detection probability 0.002102
Lowest detection probability 0.002051
This shows a candidate SEL3 event in Northwest Africa which has not been accepted in the REB. Three of the four detecting stations have a detection probability near zero. One non-detecting station (TORD) has a detection probability as high as 0.98. As many as 22 stations have better detection probability than the third best detecting stations. This event is clearly false.
This shows a candidate SEL3 event in North Indian Ocean, which has been accepted in the REB, although with an epicentre moved by 586 km. We note that out of the three detecting stations, one (DBIC) has a detection probability as low as 0.000042. In this case, DBIC would be deleted from consideration, and the remaining detections would be returned to GA for reprocessing.
North Indian Ocean Event in SEL, REB and LEB
Event id Year Mo Dy Hr Min Sec Lat Lon DepthREBLat
REBLon
REB-SEL3loc. diff
SEL3mb
NewmbMLE
6828625 2010 11 10 5 20 18.000 0.21 91.70 0.0f 3.87 95.52 586 km 5.10 3.49
3)
Detecting stations
Station Phase DeltaEvent
magnitudeDetectionthreshold
SigmaDetection
probabilityStation
magnitude
CMAR P 19.48 3.4880 3.6318 0.4020 0.360322 -
ASAR P 47.34 3.4880 3.4294 0.3510 0.566320 -
DBIC P 96.48 3.4880 4.7119 0.3110 0.000042 5.10
Third highest detection probability 0.000042
Lowest detection probability 0.000042
Non-detecting operational IMS primary stations
Station DeltaEvent
magnitudeDetectionthreshold
SigmaDetectionprobability
WRA 46.31 3.4880 3.4691 0.3640 0.520715
MKAR 47.12 3.4880 3.4952 0.3300 0.491359
TXAR 147.16 3.4880 3.6420 0.3670 0.337386
SONM 49.12 3.4880 3.7132 0.3160 0.238007
ZALV 53.85 3.4880 3.7700 0.3360 0.200658
TORD 89.95 3.4880 3.8572 0.3390 0.138082
FINES 78.37 3.4880 3.9500 0.3370 0.085202
BRTR 65.77 3.4880 3.9190 0.3000 0.075408
AKASG 72.74 3.4880 3.9524 0.3000 0.060826
GEYT 48.62 3.4880 3.9750 0.3000 0.052261
ARCES 81.64 3.4880 4.1104 0.3740 0.048049
KSRS 49.86 3.4880 4.0816 0.3270 0.034737
KBZ 61.31 3.4880 4.0680 0.3160 0.033221
GERES 81.94 3.4880 4.0844 0.3070 0.026026
VNDA 85.97 3.4880 4.3682 0.3950 0.012930
USRK 56.55 3.4880 4.2385 0.3330 0.012105
YKA 114.39 3.4880 4.4480 0.4170 0.010663
BOSA 69.53 3.4880 4.2313 0.3150 0.009142
NOA 85.20 3.4880 4.2189 0.3000 0.007416
MAW 70.59 3.4880 4.4299 0.3850 0.007212
LPAZ 154.49 3.4880 4.4103 0.3680 0.006099
MJAR 56.21 3.4880 4.3241 0.3290 0.005522
PETK 75.63 3.4880 4.4923 0.3850 0.004546
NVAR 132.68 3.4880 4.4140 0.3370 0.003000
KMBO 54.46 3.4880 4.4266 0.3360 0.002608
KEST 83.66 3.4880 4.4626 0.3460 0.002426
ESDC 94.23 3.4880 4.4686 0.3120 0.000836
ILAR 102.71 3.4880 4.6138 0.3480 0.000608
PLCA 136.56 3.4880 4.9210 0.3980 0.000159
SCHQ 122.37 3.4880 4.9540 0.4000 0.000124
CPUP 140.45 3.4880 4.9090 0.3820 0.000100
ULM 129.31 3.4880 4.8710 0.3700 0.000093
BDFB 137.38 3.4880 5.0540 0.3730 0.000013
ROSC 165.16 3.4880 5.1427 0.3630 0.000003
PPT 117.36 3.4880 6.3010 0.3000 0.000000
Number of primary non-detections 35
Number of stations exceeding third detection probability 32
Number of stations exceeding lowest detection probability 32
Afghanistan-Tadjikistan border region Event in SEL, REB and LEB
Event id Year Mo Dy Hr Min Sec Lat Lon DepthREBLat
REBLon
REB-SEL3loc. diff
SEL3mb
NewmbMLE
6830385 2010 11 10 22 21 36.250 36.34 71.40 157.5 36.20 71.37 16 km 3.52 3.16
1)
Detecting stations
Station Phase DeltaEvent
magnitudeDetectionthreshold
SigmaDetectionprobability
Stationmagnitude
AAK P 6.87 3.1557 2.7047 0.4260 0.855138 -
GEYT P 10.73 3.1557 3.2519 0.4180 0.408932 -
MKAR P 13.36 3.1557 2.6928 0.4470 0.849779 -
KURK P 15.31 3.1557 2.8657 0.3910 0.770830 -
BVAR P 16.84 3.1557 2.7482 0.3360 0.887391 -
AKTO P 17.19 3.1557 2.8662 0.3030 0.830295 -
ZALV P 20.07 3.1557 2.7491 0.3360 0.886870 3.40
BRTR P 29.81 3.1557 3.4075 0.3000 0.200592 3.20
FINES P 37.90 3.1557 3.2897 0.3370 0.345431 3.60
ARCES P 41.53 3.1557 3.4163 0.3740 0.242926 3.70
HFS P 43.48 3.1557 3.4440 0.4010 0.236025 3.60
TORD P 66.01 3.1557 3.1298 0.3390 0.530415 3.60
Third highest detection probability 0.855138
Lowest detection probability 0.200592
Non-detecting operational IMS primary stations
Station DeltaEvent
magnitudeDetectionthreshold
SigmaDetectionprobability
ASAR 83.84 3.1557 3.2145 0.3510 0.433491
WRA 81.58 3.1557 3.2310 0.3640 0.418019
KBZ 23.00 3.1557 3.2372 0.3160 0.398225
ILAR 74.99 3.1557 3.2713 0.3480 0.369896
SONM 28.26 3.1557 3.2928 0.3160 0.332172
YKA 81.53 3.1557 3.3805 0.4260 0.298848
AKASG 33.36 3.1557 3.3591 0.3000 0.248844
CMAR 30.05 3.1557 3.4142 0.3420 0.224810
GERES 43.34 3.1557 3.3915 0.3070 0.221233
NOA 44.80 3.1557 3.4917 0.3000 0.131319
KSRS 44.67 3.1557 3.6005 0.3270 0.086881
ESDC 57.86 3.1557 3.6144 0.3120 0.070742
PLCA 150.00 3.1557 3.8126 0.3980 0.049416
USRK 46.06 3.1557 3.7229 0.3330 0.044235
KEST 49.41 3.1557 3.7904 0.3460 0.033292
PETK 59.98 3.1557 3.9567 0.3850 0.018741
DBIC 74.98 3.1557 3.8082 0.3110 0.017938
MJAR 52.82 3.1557 3.8620 0.3290 0.015901
BOSA 77.71 3.1557 3.8413 0.3150 0.014758
KMBO 48.85 3.1557 3.9416 0.3360 0.009665
TXAR 114.65 3.1557 4.0336 0.3670 0.008375
SCHQ 82.49 3.1557 3.9831 0.3150 0.004309
VNDA 125.13 3.1557 4.4546 0.4290 0.001232
LPAZ 138.99 3.1557 4.3666 0.3680 0.000500
CPUP 135.50 3.1557 4.5006 0.3820 0.000215
ULM 93.19 3.1557 4.3342 0.3090 0.000068
NVAR 105.20 3.1557 4.5056 0.3370 0.000031
BDFB 122.68 3.1557 4.6956 0.3730 0.000018
ROSC 128.05 3.1557 4.7096 0.3630 0.000009
MAW 103.69 3.1557 4.9377 0.3850 0.000002
PPT 139.41 3.1557 5.7926 0.3000 0.000000
Number of primary non/detections 31
Number of stations exceeding third detection probability 0
Number of stations exceeding lowest detection probability 9
3 3.5 4 4.5 5−0.4
−0.2
0
0.2
0.4
Mean : 0.117St.dev : 0.069Nobs : 984
AVGscale mb1
MLE
scal
e −
AVG
scal
e
P1 = 0.103, P0 = −0.278
3 3.5 4 4.5 5−0.4
−0.2
0
0.2
0.4
St.dev : 0.064
AVGscale mb1
MLE
scal
e −
AVG
scal
e
ARCES
3 3.5 4 4.5 53
3.5
4
4.5
5
MLEdet mb1
MLE
scal
e m
b1
ARCES
Mean difference: 0.039St. dev : 0.047Nobs : 580
−1
−0.5
0
0.5
1
1.5
ASA
RTO
RDWRA
MKA
RILAR YKA TXAR
BGCA
SONM
PDAR
ZALV
FINES
AKA
SGBR
TRCM
AR
NVA
RGER
ESARC
ESLP
AZ
NOA
KBZ
ESDC
STKA
KSRS
DBIC
VNDA
BOSA
CPUP
BDFB
ULM USR
KSC
HQ
PLCA
KEST THR
MJAR
MAW
PETK KM
BORO
SCPP
T
Stat
ion
dete
ctio
n le
vel (
mb1
uni
ts)
Primary stations Teleseismic distances (20 - 90 degrees)
−1
−0.5
0
0.5
1
1.5
KURK BV
AR
QSP
ASIV SN
AA
AKT
OINK HFS MATP
RES EKA
AAK
CFAA
ELK
LSZ
FITZ MSK
USD
VMLR
ANMO
APG LBTB
TSUM
MDT
FRB
SPITS
DAVO
XLV
CNEW DLB
CTK
LYB
HVR
AC
ASF IDI
KDAK
DAV
MMAI
PSI
CTA
ATA
H
Stat
ion
dete
ctio
n le
vel (
mb1
uni
ts)
Auxiliary stations Teleseismic distances (20 - 90 degrees)
Database