EARTHQUAKE CLOUDS AND
SHORT TERM PREDICTION
I and Darrell Harrington submitted “A study of earthquake
prediction by atmosphere precursors” (1) to Susan Hough, Chief Editor of Seismological Research Letter (SRL), in Oct. 2002. She sent it to
Alan Jones, retired International Business Machines Corporation (IBM)
employee and adjunct professor of the
calculated the Normalized Score x= 0.854 or the Likelihood
(Integrative Probability) p= 27% to reject our paper (6). Unfortunately, there were many artificial
in his calculation and If those errors had been corrected, the
likelihood would have gone down far below 5%. We appealed the review
on this basis, but the appeal was rejected without
further review and instead the journal might publish it with “Apart from prediction” (7).
We declined since we believed that the successful predictions were core
evidence for the validity of the prediction method. On
In May 2007,
Ara, Japanese predictor, wrote me that Roger Hunter, former employee of the
USGS, had been attacking me with Jones’ permission (11).
Hunter asked me why I had not showed Jones’ review by the permission (12).
I asked Jones if he knew who had told Hunter the permission. He admitted,
“I did” on
I requested Jones to show his scores by excel for detail, but he rejected it (17). Thus, I try to make the Excel Correction (18), whose Column A~C and F~I are as the same as Column ‘No.’, ‘Shou Prob.’, ‘Hit?’, ‘Jones Prob. A-S’, ‘Jones Hist. Prob’, ‘Jones total Prob’ and ‘Jones hit?’ of Jones’ Spreadsheet respectively (19). Column B, F, G and H depict Shou’s probability and Jones’ aftershock probability, general probability and total probability respectively. Sign ‘1’ and ‘0’ show hit and miss in Column C, D and I, independence and dependence in Column E and J, and Jones’ not aftershock and aftershock in Column K respectively.
Column L and M reveal Shou’s score s=(b-c)In(b(1-b)) and variation v=b(1-b)(In b(1-b))2 here b: Shou’s probability in Column B and c: Shou’s hit and miss in Column C. The total score is S=Σs=19.22 (L56 of 18), the total variation is V=√Σv=4.87 (M57) and the Normalized Score z (Jones’ x) =S/V=3.94 (L59), whose Integrative Probability p= 0.000044 (L60) according to the Normal Distribution Table (20). All above formulas are from (4). This example shows how to calculate the total probability by the score method laconically.
In contrast, Jones calculated x= 0.854, or p= 0.27 (if x= 0.854, then p= 0.196 according to 20). Let’s focus on the calculation in Column P ‘Jones sco H~K’ that adopts all his rules such as “Peer on”, dependence, aftershock probability, his calculated probabilities and yields for z= 0 (P59). It is a puzzle of how he gets x(i.e. z)= 0.854. Since he doesn’t show further details, my correction has to begin from Column P and I will now show where he went wrong.
Five highlighted misses
Jones highlighted 5 misses of No. 19, 23, 26,
32 and 45 of my predictions by the clouds with red color maybe
to expose my “guilt” on
Jones highlighted the 5 misses, but only “changed 23 and 32 to misses”. I wondered about this puzzle. He replied, “I gave you hits on the three events you mention above” (22), so his ‘fair review’ exaggerated the miss number 2 to 5 to lure attention. The following table shows data by his largest magnitude rule and utopia “Peer on”
Note: LT: local time of the west
coast. Mag. the largest magnitude according to
Jones. Lat. Latitude. Lon. longitude. No. 19 and 26 are clear hits. No. 32.1
is classified as a miss by longitude error of 0.01
times more precise than 0.1
degree of the minimum error of the
USGS. However, I tolerate it for
his utopia “Peer
on”. On the other hand, both the
Jones wrote, “Shou calls a 5.9 a hit but it is not in his mag window. However, I find a 6.1 so I give him a hit” (19). He noticed our average magnitude (2), but replaced his largest magnitude by our average to make No. 26 a miss. He found the 6.1 and even claimed, “I give him a hit”, but marked No. 26 a miss in his Spreadsheet. Column R corrects this mistake (18), so its score increases 1.42(R56) from -0.58 (P29) to 0.84 (R29), the Normalized Score z (Jones’ x) increases 0.33(R59) and the Integrative Probability p reduces to 37.07% (R60).
Jones wrote, “Shou claims hit but only 5.8”. He replaced his largest magnitude by our average to make No. 19 a miss, too. Column S corrects this mistake. Its score increases 1.91 from -0.35 (P22) to 1.56 (S22), the Normalized Score z increases to 0.78(S59) and the Integrative Probability p reduces to 21.77% (S60).
wrote, “Nos 31 and 32 are not independent. No hit. Slightly out of region by 0.01 deg”. First, No. 31
hit a quake on Dec.12, 1998, while No. 32 began on
gave No. 45 a hit on
don’t know why
Jones claimed 8 dependences: No.20~22 & No.30~34. They are all hits except No. 32 for ‘Slightly out of region by 0.01 deg.’ Then, he wrote, ‘Take out dependent events, aftershocks, and change 23 and 32 to misses’ (6, 19 ). It is interesting to withdraw No. 32 for a miss after ‘Taking out dependent events’. He can either give both the 7 ‘dependent hits’ and the ‘dependent miss’ the same score ‘0’ by ‘dependence’, or give the 7 ‘dependent hits’ 7 plus scores, and the ‘dependent miss’ a minus by hit, but he cannot give the 7 ‘dependent hits’ 0 by ‘dependence’, while gives the ‘dependent miss’ a big minus by miss, which is a clear bias. Since No. 32 is already corrected, I just discuss the 7 ‘dependent hits.
Seven ‘dependent’ hits
In fact, the 8 ‘dependent’ predictions are all independent. No.30
is obviously independent for its time window in "
Eight dependent events
Jones did not claim his independent rule even to the USGS until Feb.14, 2003. As a result, the USGS had signed some overlapping predictions. Thus, it is not my fault, but his. A reasonable way to solve this problem is to delete all dependent predictions from our paper. I already did it. Column E depicts 8 dependent predictions: No. 3, 4, 41, 43, 44, 47, 49 and 50 with green ‘0’. Column W solves this problem. The Normalized Score z decreases to 1.61(W59) and the Integrative Probability p increases to 5.37% (W60).
wrote, “My definition of an aftershock is an
event within the region of aftershocks of a main event” on
Moreover, Jones extends aftershock probability even to those earthquake he did not claim as "aftershock", such as No. 2, 4-6, 8-9, 11-17, 25, and even claimed as "Not aftershock" such as No. 18, 21, 22 and 24 to make hits smaller plus scores, and misses bigger minus scores artificially.
Correct extending ‘aftershock’ probability to not aftershock prediction
Column Y corrects the use of aftershock probability for quakes not claimed to be aftershocks. The Normalized Score z increases to 1.96(Y59) and the Integrative Probability p decreases to 2.50% (Y60), smaller than 5%, Jones’ own significant threshold for publication.
Correct No. 1, 7, 10 and 20 so called ‘aftershock’ prediction
Column AA corrects the 4 so called ‘aftershock’ predictions that were incorrectly labeled aftershocks. The Normalized Score z increases to 2.46(AA59) and the Integrative Probability p decreases to 0.69% (AA60), much smaller than the 5% publication threshold.
About Prediction 2001/03/20
Jones wrote, “It seems he made a prediction on 2001/03/20 and then send in a new prediction to replace it on 2001/03/21”. Our paper has two tables of predictions: Table 1 by the clouds and Table 4 by geoeruptions. He forgot to review Table 4, whose No. 7 was the prediction on 2001/03/20.
Jones blames all misses on the precursor, but miss analysis shows their causes from satellite data problems, earthquake data problems and my lack of experience as a pioneer on the prediction, detailed in (28, 29).
exaggerates two misses
to five to lure attention. He replaces his
own largest magnitude rule by our average magnitude
rule to make No. 19 and No. 26 misses.
He claims No. 26 a hit, but marks a miss.
He misclassifies 8 independent hits as dependent and takes them out. Then, he puts No. 32 back for a big minus based on a longitude error of 0.01 degree, ten times more
precise than 0.1 degree minimum error of the USGS. He extends ‘aftershock’
probability to not
aftershock predictions to make hits smaller plus
scores and misses
bigger minus scores. He classifies ‘aftershock’
without a scientific
definition. His unproved ‘aftershock’
model looks like a missile to shoot a remote and isolated
place he likes. He claims No. 1,
7, 10 and 20 aftershocks, while the USGS classifies the Newhall earthquake,
By contrast, I adopt
his largest magnitude and utopia “Peer on” to divide between hit and miss. I decline No. 45 as a hit for his utopia “Peer on”. I tolerate
No. 32 a miss by error
degree for his utopia “Peer on”. I also tolerate No. 23 as a miss in spite of a
puzzle of why
The likelihood p using Jones’ individual probabilities and rules after correcting his artificial errors reaches 0.69%, much better than his 5% threshold in spite of blaming those data and experience problems on the precursor. This clearly demonstrates my work to be statistically significant and worthy of publication.
1. Zhonghao Shou & Darrell
Harrington. A study of earthquake prediction by atmosphere precursors.
(manuscript for SRL)
Jones. peer on
4. Richard Jones & Alan Jones. Testing Skill in Earthquake Prediction. (manuscript) 1996
5. Brelsford, W.M. & Jones, R.H. Estimating Probabilities. Monthly Weather Review 95, 570-576 (1967).
6. Alan Jones 5% significance, independence Review Feb.14, 2003
7. Susan Hough. Apart from prediction. Email.
Jones. Put review out. Email
Jones. Permission Yes. Email
Jones. Up to you. Email
Roger attack Email
Hunter. Why did not post Email
Jones I did.
Jones. you may.
Shou. two reasons.
Jones. “fair review”
show in excel. Email
20. Answers. Normal table (http://www.math.unb.ca/~knight/utility/NormTble.htm)
Jones. 3 hits.
Jones. surprise Email
Shou. two examples Email
27. Alan Jones’ missile aftershock model
28. Zhonghao Shou. Earthquake Vapor, a reliable precursor. Earthquake Prediction, 21-51 (ed. Mukherjee, Saumitra. Brill Academic Publisher, Leiden-Boston, 2006).
29. Darrell Harrington & Zhonghao Shou Bam Earthquake Prediction & Space Technology Seminars of the United Nations Programme on Space Applications 16, 39-63 (2005).