Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Vol.14 No.

3 (293~302) ACTA SEISMOLOGICA SINICA May, 2001

Article ID: 1000-9116(2001 )03-0293-10

Short-term precursor of the geomagnetic


¢¢
horizontal field transfer functions
GONG Shao-jing ( ~ ~) TIAN Zhen-li (~/~(-N) QI Cheng-zhu (/t~,~&J~)
HE Shu-min ( N ~ , ~ ) YAN Xiao-mei ( N ~ N ) CHEN Hua-ran (~,~i{g~ )
LI Lian-di ( ~ N )
Seismological Bureau of Tianjin, Tianjin 300201, China

Abstract
In this paper, using the variometer data of Changli station and Baijiatuan station from 1972 to 1997, the in-
ter-station transfer functions of horizontal field and the single-station transfer functions of vertical field have been
calculated. With analyzing the whole changing process of the transfer functions, we have discovered that only the
horizontal field transfer functions arose obvious short-term precursors before Tangshan earthquake. This fact may
provide a new method for short-term prediction of earthquake.

Key words: horizontal field transfer functions; vertical field transfer functions; short-term precursor of
earthquake
C L C number: P318.2+1 Document code: A

Introduction
The history of studying the changes in deep geoelectric conductivity of an earthquake prepa-
ration using the parameters of geomagnetic short-period variations has been about 30 years. At
first, Japanese calculated the real transfer functions as well as the a and b coefficients of Parkinson
vector by taking the measurement of graph and using empirical formula. There are some examples,
such as the 1923 Kanto M=7.9 earthquake (Yanagihara, 1972), the 1966 Tashkent M=5.5 earth-
quake, the 1972 Sitka, Alaska, M=7.2 earthquake (Rikitake, 1979) and so on. Since the 1980s, we
have also discovered the probable changes in deep geoelectric conductivity of Tangshan M--7.8
earthquake (GONG, 1983, 1985). Then we studied the Heyuan M=6.3 earthquake (GONG, 1987),
the Songpan M=7.2 earthquake swarm (GONG, et al, 1989) and the Heze M=5.9 earthquake
(GONG, et al, 1992) using the a and b. Except for Guangzhou station before Heyuan earthquake,
there are significant anomalies before other earthquakes.
Beamish (1982) provided the result on the Carlisle M=5 earthquake by using geomagnetic
horizontal field inter-station transfer functions. He used the data of optical pumping magnetometer,
and the anomalies of horizontal field transfer functions are very obvious. We try to study the hori-
zontal field transfer functions by using the data of variometers, no researcher have done this kind
of work before. From the results of Tangshan earthquake (GONG, et al, 1997) and Hualian earth-
quake of Taiwan (GONG, et al, 1991), we found that the changes of some horizontal field transfer

* Received date: 2000-05-20; revised date and accepted date: 2001-04-19.


Foundation item: Chinese Joint Seismological Science Foundation.
294 ACTA SEISMOLOGICASINICA Vol.14

functions are more obvious than vertical field transfer functions. We only used the data of
variometers during the period from 1972 through 1984 in the 1997 paper and one to two group
transfer functions are calculated every year, so the short-term and imminent earthquake precursors
are found very difficult. In order to examine whether the short-term precursors exist or not and to
test its reliability, we did the following work.

1 Estimation of transfer functions


1.1 Formulae of transfer functions
Transfer functions express the relationship between the output and the input, and describe the
intrinsic features of the system itself. It reflects the character of the underground electric structure
and represents mainly the transverse inhomogeneity of geoelectric structure. When the inducing
field is uniform in the considered area and the incidence angle is 90 °, there are following formulae
after being simplified as:
Z( to)=n ( to) H(co) +B(co) D( to)+ e:( to) (1)

{ H~((9) = C(co)Hn(co ) + Gico)Dn(co ) + eh(CO)


D,(C0) E(cO)Hn(CO) + F(co)D,(co) + ~'d((.0) (2)

These parameters in equation (1) and (2) are complex number and are functions of frequency. The
equation (l) is formula of the single-station vertical field transfer functions, Z(co), H(co), D(co) are
Fourier transformation of three components of geomagnetic field (Z, H, D). Here, D(nT) =
D'×Ho/3438, Ho is H component' annual mean of absolute value.
The subscript "a" stands for the abnormal field and "n" represents the normal field in equa-
tion (2). The geomagnetic field at reference station indicates the normal field. The reference sta-
tion should keep at a distance from the abnormal seismic zone, it and the station at seismological
zone must be of the common source field (inducing field) for quasi-uniform of source field. The
difference between magnetic field at station of abnormal zone and that at reference station repre-
sents abnormal field.
The complex transfer function consists of Re-part and Im-part, for example, A(co)=Au(co)+iA~(co).
A and B are single-station vertical field transfer functions and C, G, E, F are horizontal field in-
ter-station transfer functions. The transfer function is a dimensionless parameter.
1.2 Method for estimating transfer functions
Estimated values of transfer functions are calculated by the complex least squares method, its
formula is as in the paper about Heze earthquake (GONG, et al, 1991). The formula of estimating
errors on Re-part and Im-part are obtained by using the Lagrange Multipliers method and Unit
Vector method. The errors of estimated value on transfer functions indicate the discrete degree in
data of 12-20 events with which transfer functions are evaluated.
In order to reduce the effect of a few specific events on the estimated values, we used the
weighted least square method before. In this paper, the Robust statistical method has been intro-
duced into the complex least square method. As the time service of variometer is not precise, the
time errors of sampling data will bring the abnormal field some dummy frequency components,
and will form the bad data in estimating transfer functions. We have discovered in practice that a
very few bad data will affect the estimated values greatly. The Robust statistical method is more
advanced than the simple weighted method. The Robust method can distinguish the bad data
No.3 GONGShao-jing, et al: GEOMAGNETICHORIZONTALFIELD TRANSFERFUNCTIONS 295

automatically and decrease its weight, and cause the estimated values to tend to stability (Egbert,
Booker, 1986).
The Robust method was a statistical method developed in 80's of 20th century, its main idea
is to select a certain statistical parameter so that the ideal supposition would not be affected by
some departure and the conception of outlier points is introduced. The Robust statistical method
has been developed into a series of theory and some different statistical parameters are put forward,
such as L-estimating, R-estimating. In this paper, we have used the M-estimating that is applied to
evaluation of parameters.
For any given model y(xi, a), there, the x~ is an independent variable, the a is parameter that
will be evaluated and may be a vector. The y is form of function and the y~ is observed dependent
variable's value corresponding to the x~. Introducing a loss function p, the criterion of error is
constituted:
M=Ep [y, y (x~, a)] (3)
Make this criterion minimized, i.e., the derivative of function p is equal to zero.
In general, the function p is no dependent upon both variables - - observed value y~ and fore-
casted value y(x~, a), and only dependent on their difference y,-y (x~, a) or at least upon weighted
value of this difference z=[yc--y (x,, a)]/~,, ~, is weighting factor. When z follows various distribu-
tion, the function p has different expression. When z follows the normal distribution (except out-
lier points), usually suppose p(z)=z2/2, its derivative ~(z)=z. Make Mvariable minimized by itera-
tion repeatedly. The y, is corrected by weight w(z,)= 74,z,)/z,. When these data follow normal dis-
tribution, w(z,)=l; for bad data, w(z3el. In practical problem, in general, the weight w(z) follows
the formula:
1 z, _<P
w(zi) = (4)
P/z~ zi > P

2 Data processing
2.1 Collection of data
We have collected data of variometer in 1972-1997 at Changli and Baijiatuan stations. A
group of monthly transfer functions has been calculated in 1975 and 1976, and data have been
divided into a group per half month, in February and March of 1976, during the period events are
more. In order to compare with data in 1975-1976, a group of monthly transfer functions has also
been calculated in 1989 that belong to normal year and its events are also more.
We choose 12-20 events for computing a group of transfer functions. There are 6 curves
corresponding to three components of both stations for every event. We use the A/D converter for
sampling. Due to the amount of sampling is large, the former data of 1972-1974 and 1977-1984
were used (GONG, et al, 1997). 17 400 curves from 2 900 events have been sampled. For estim-
ating the sampling errors, some curves have been sampled twice, so in fact number of these curves
sampled have reached about 20 000.
2.2 Processing of data
The smaller period is, the larger effect of sampling errors on the estimated value of in-
ter-station transfer functions is. Therefore, we choose the events involving from ten minute to
about one hour period. Because the initial position of event is not at the time line of magnetogram,
296 ACTA SEISMOLOGICASINICA Vol.14

the length of event is three and a half to four hours. The sampling interval is 0.75 min and the
sample length of FFT is 256. Due to number of events is too much, sampling men are four. The
difference of every body's sampling habit enables the sampling errors to increase.
From the digital time series slope were eliminated, moving average were computed and
higher frequency elements were filtered out. The amplitude spectra and the phase spectra have
been obtained through the discrete Fourier transformation, after that convert them into co-spectra
and quadr-spectra. The estimated values of transfer functions are calculated by using co-spectra
and quadr-spectra of every event for same period and using the complex least square method and
the Robust method. In computing the estimated values of transfer functions, the elements, which
the amplitude of horizontal components are lower than twice average noise level, were removed.
Although 12-20 events were selected for a group of transfer functions, at last the number of
the selected elements is different for various period, this is because some events may not consist of
the needed period element. When the number of remained elements is less than ten, we consider
the estimated value only to be a reference.
In this paper, the P-value in Robust method is evaluated through residual errors of all ele-
ments and the control parameter value of iteration stopping is determined by using the error of the
estimated value.
The numerical experiment for testing the convergence of Robust method is listed in Table 1.
In the table, the left give the initial values and the right are Robust estimating values and the errors.
It can be seen from Table 1 that they all can converge upon a unique estimated value lastly, what-
ever supposed initial value is.

'Fable I The convergenceof Robust method


File Au Av Bu By Au t~lu Av ~v Bu 8Bu By t~Bv
L84 -0.152 -0.105 -0.010 -0.038 -0.163 0.0065 -0.102 0.0071 0.0180 0.0103 -0.057 0.0132
1.000 0.000 -0.015 -0.038 -0.164. 0.0066 -0.102 0.0071 0.0180 0.0103 -0.057 0.0135
Q84 0.166 0.077 -0.160 -0.038 0.163 0.0030 0.0766 0.0052 -0.166 00091 -0.035 0.0058
0.001 0.002 0.004 0.004 0.165 0.0040 0.0762 0,0054 -0.165 0,0099 -0.0381 0.0078
CZ02 0.0287 0.0067 0.0868 0.0024 -0.0549 0.0568 0.0035 0.0037 0,997 0.0351 0.0041 0.0031
-0.467 0r0039 1.131 --0.0008 -00552 0.0567 0.0035 0.0037 0.997 0.0351 0.0041 0.0031
--0.139 00026 1.064 0.0039 -0.0550 0.0568 0.0035 0.0037 0.997 00351 0.0041 0.0031
--0.359 0.0045 1.085 0.0000 -0.0550 0.0568 0.0035 0.0037 0.997 0.0351 0.0041 0.0031

In the computing, there is a circumstance that the iteration cannot stop. After detecting every
event, we discover the sampled series of an event at Changli and Baijiatuan station are staggered
about an hour. Before, in the numerical experiment, when the data consist of two groups of more
different situations, the final results will be different for different given initial values, that the re-
sults either trend towards this group, or else trend towards that and decrease weight of this one.
This condition has deviated from the hypothesis that Robust method only applies to a few outlier
points. When this circumstance appears, we only can re-divide.
We consider the transfer functions value calculated by using original data to be initial values,
and final estimated values and its errors are calculated through Robust method. The Robust estim-
ating values are more stable than the initial values, and its estimating errors are less than the initial
error. This effect of horizontal field transfer functions is more obvious than that of vertical transfer
functions.
No.3 G O N G S h a o - j i n g , et al: G E O M A G N E T I C H O R I Z O N T A L F I E L D T R A N S F E R F U N C T I O N S 297

3 Analysis of error
3.1 Errors of digitization
A curve is twice sampled by different people in order to inspect the size of digitizing error, At
and SS represent sampling errors at x axis and at y axis, respectively. The At is time difference of
twice sampling for same characteristic point and the SS indicate the mean-square deviation of
these difference values. The At is an error at time axis and it causes dummy frequency components,
but the SS is an average error aty axis and causes amplitude inaccuracy.

Table 2 Samplingerrors
An 0 ±1 +_2 ±3 ±4 ±5 ±6 ±9 Z PSS SSr~ SS~n
H 60 75 36 17 5 1 0 0 194 0.92 2.09 0.26
D 42 59 45 22 11 11 2 2 194 033 0.82 0.16
Z 14 48 28 17 3 2 0 0 112 0.30 0.70 0.07
Z 116 182 109 56 19 14 2 2 500

In Table 2, At=0.75An, 0.75 min is sampling interval and the n is ordinal number of time se-
ries obtained by using interpolation after sampling. PSS is mean of SS and unit is millimeter. In
Table 2, the SSm,× of a few events is larger, because the paper of magnetogram is not very smooth.
Distribution of At show that time error in the range of 1.5 min is about 81%. Few very large An
should not appear, such as +6 and _+9. That as well as the dispersion in An of D component may
bear relationship with the fact that the extremes o l D are not sharp. So the At reflects time error of
sampling and also mirror the error of selecting extreme point. In a ward, sampling errors in this
paper are larger than those in 1997 paper (GONG, et al, 1997).
3.2 Errors of adopted scale valves
Baijiatuan station is best in China, the errors of adopted scale values were smaller than 1%.
Adopted scale value of Changli in 1972-1979 were selected after drawing measured scale
values on coordinate paper. The measured values in 1972-1973 were more stable and the measur-
ing points were denser, so the errors of adopted scale values were about 1%. Because of adjusting
the instrument, in December of 1973 to June of 1974, in general, the errors of adopted scale values
reached to 2% and reached to 3% in January of 1974 for Z and H components, but for D, the errors
of adopted values were 1%. From July of 1974 to 1979, the errors of adopted values ranged from
1% to 2%. Since the 80's, the errors of adopted scale values were about 1%.
3.3 Analysis of estimating errors
The source field effect and the sampling errors are main reason that enables original data of
estimating transfer functions to disperse. The errors of adopted scale values only bring about sys-
tematic error, its influence upon deviation of estimated values is less than that of above both.
Under premise of uniform source field, zn=O, equations (1) and (2) can be set up (GONG, et
al, 1991, 1997). In fact, the inducing field cannot entirely satisfy the presupposition of uniformity,
therefore it causes source field effect and makes data disperse. Analysis of time sequence of
short-period events at Qingguang station show that the influence of source effect is not entirely
random and has fixed "inertia", but the order of ARMA model is no high usually (GONG, 1983).
On the one hand, the sampling errors of different people made data random dispersion, on the
other hand also give estimated values different systematic errors and lead to fluctuation of esti-
mated values of transfer functions. For parameter estimation, the more the number of the samples
is, the more accurate the estimated values are. However, enough number of transfer functions must
298 ACTA SEISMOLOGICASINICA Vol.14

be obtained in order to detect short-term precursor, but the number of events obtained from
variometer data is restricted. After making numerical experiments, we determine that the number
of events for a group of transfer functions is 12 to 16. So, inadequate event's number also is reason
for the fluctuation. For checking the effect of the errors on estimated values, two group data se-
lected arbitrarily have been re-sampled. Only vertical component Z for Au, Av, Bu, Bv at Changli
has been re-sampled, because the Z component is smaller than H and D components for sin-
gle-station transfer functions, thus its relative errors of sampling will be larger and the influence
on estimated values will also be larger than H and D. For horizontal transfer functions, H and D
have been all re-sampled. The ratios which differ between estimated values in twice to be larger
than the ~ or 2c~ are calculated by using two groups of data. For single-station transfer functions,
the ratios are 36% or 18% respectively. For horizontal transfer function, the ratios are 50% and
13%, respectively and 26% of the differences are larger than 2c~ or 0.1. We have used the larger
estimating errors in two groups as the criterion to judge whether differences are larger than 6 or 2c~
or not. From the mentioned above, only sampling errors bring about already larger fluctuation of
Robust estimated values. Therefore, in following figure, the length of error bar is taken as four
times estimating errors (_+2g).
Table 3 lists the average of estimating errors P6. The initial errors represent the degree of
scatter about original data of a transfer function, the P6 shows the average error level of Robust
estimated values and also relatively reflects the degree of scatter about original data. It can be seen
that the P6 of horizontal field transfer functions are larger than that of vertical field transfer func-
tions, and the size of P6 is 40%-60% of a value. It again indicates that various systematic error of
sampling for various men is also reason to make estimated values fluctuate. Moreover, the appli-
cation of Robust estimating method also made the g vary small.
Table 3 The period response of transfer functions
1" 64.0 48.0 32.0 27.4 22.7 17.0 12.2
A. mean 0.148 0.163 0.176 0.178 0186 0.165 0.139
(Cha) o- 0.020 0.022 0.027 0.031 0.030 0.039 0.053
P6 0.008 0.010 0.014 0.015 0.016 0.018 0.027
A~ mean 0.058 0.067 0.062 0.057 0.056 0.046 0.029
(Bai) cr 0.017 0.015 0.018 0.020 0.016 0.015 0.017
P,~ 0.008 0.008 0.010 0.011 0.010 0.010 0.009
Cu mean -0.070 -0.094 -0.134 -0.144 -0.180 -0.246 -0.379
cr 0.050 0.048 0.072 0.069 0.084 0.093 0.122
P,~ 0.015 0.022 0.037 0.038 0.046 0054 0.069
C, mean -0.096 -0.123 -0.156 -0.157 -0.192 -0.233 -0.253
cr 0.059 0.074 0.094 0.110 0.135 0.137 0.172
P6 0.020 0.027 0.044 0.048 0.053 0.062 0.077
F,, mean -0.106 -0.134 -0.207 -0.224 -0.284 -0.404 -0~552
o" 0,042 0.048 0,118 0.138 0136 0.174 0.173
p,~ 0.012 0.016 0.019 0.017 0.025 0027 0.032
F, mean -0.155 -0.168 -0.243 -0.284 -0.313 -0.337 -0.373
o- 0.051 0.076 0.125 0.113 0.129 0.158 0.215
Po" 0.028 0.034 0.050 0.052 0.064 0.068 0.089
E,, mean -0.060 -0.044 -0.038 0.042 -0.034 -0.022 -0.003
cr 0.025 0.042 0.056 0.058 0.053 0.059 0.075
Po 0.017 0.024 0.033 0033 0.035 0.036 0.045
No.3 G O N GShao-jing, et al: GEOMAGNETIC HORIZONTAL FIELD TRANSFER FUNCTIONS 299

4 Analysis of the results


4.1 T h e period r e s p o n s e o f transfer functions
We have evaluated the mean and the standard deviation a and the P~ of sixteen transfer func-
tions from sixty groups of data during 1985~1997, see Table 3. The a represents the degree of
fluctuation about the estimated values of transfer functions. Table 3 lists the average of some
transfer function corresponding with Figure 1 and Figure 2. It will be seen that there exist evident
period effects for transfer function and its estimating errors.
Due to no strong shock in Tangshan area from 1985 to 1997, the mean and the a in Table 3
can be taken for an average level in normal condition and criterion to determine anomaly.
4.2 T h e time c h a n g e s o f vertical field transfer functions
The time change curves of A~(Cha) and A~(Bai) are shown in Figure 1. In order to detect
whether there exist short-term precursors before Tangshan earthquake, the intervals of horizontal
axis are not equally spaces. The horizontal lines in figures are all represent zero lines. The dotted
lines express there without datum or the estimated value only to be as reference.

•~ ~ . o O . ~ r ~ - - - ~ _ . ~ _ q k ~ _ . _ ~_ . . . . . ~_o; ;
.. o o. ~r'-+. . . . . ~ -

1973 1975 1976 1977 1979~ ~981 1983 1985 1987 ~ ' t'989 ' 1990 1992 ~994 '1996'
~ ~,. Year

.. o ° - ~ ~ ~ - - ~ - - . , - ~ '

32. o o. ~ p , - ~ _ , _ ~ - _ . ~ _ ~ + . ~ _ _ ~ ........

, I I , , I i I __ I L J..:. . ~ ~ i" T T T T T T i"


1973 1975 1976 1977 1979 1981 1983 1985 1987 1989 1990 F992 199,1 1996
Year

Figure 1 The time changes of vertical field transfer functions Au(Bai) and Au(Cha)

At Baijiatuan, there exist no obvious anomalies, most data fluctuate range from 0.01 to 0.08.
It shows that there is no the transverse inhomogeneous underground electric structure in the vicin-
ity of Baijiatuan and it is applied as a reference station.
The time change curves of Au(Cha) seems to not have evident anomalies. The former largest
values and smallest values (GONG, 1997) are still drawn in this figure, but they are also within
±3~. Obviously, these errors of this result are larger, the fluctuation of estimated values has cov-
ered not very large middle-term precursor. It can be seen from Figure 1 that there exists no obvi-
ous and over ±3a short-term precursor of Au for Changli.
4.3 T h e time changes o f h o r i z o n t a l field transfer functions
It can be seen from Figure 2 that Eu does not have evident anomalous changes, they fluctuate
nearby zero line, the fact coincides with the result in 1997 paper.
Cu and Fu exist significant short-term precursors, they arose from February to the first half of
March of 1976. The largest anomaly values of Cu reached to -0.98, -1.36, -1.40, -1.47 for peri-
300 ACTA SEISMOLOGICA SINICA Vol. 14

ods of 17.0, 22.7, 27.4, 32.0 min, respectively, and greatly overstep the scopes of criterions o f + 3 a
in Table 3, they are 10, 16, 20, 20 times as large as the a, respectively. The largest anomaly values
o f F , reached to -0.92, -1.36, -1.26, -1.52 and are 5, 10, 9, 12 times as large as the a.

48. 0.-9.

54. 0 - - 0 .
Lj i I A~, , ~!il~ , I ~Y ulr "4t,v i i J i I I I , __ I { i I J •
-1973 i975 1976 1977 ~979 1981 1983 1985 1987 ' 198~) " '1990 1992 %994 '199d--
Year

17. o . . 0 . 5 ~ r ,~ ~, ~ ~..t/ .+/- r ~ - c ~


27.7 o. 3 L r ~ --- y,.~rl.l.,r ~ ' ~ - .,,i..~.~,,¢r~,.r~r,~,.f~ y ----

4s. 0 0~L : ~ q ~ ~ ' • - ~ " " - " - " ~ ' ; " ' ~ ' - ~ " ~ ~

~'" i ~ i i
aF,. x~ i I [ t
6&" 0 - - 0 . 5L~ .1973
. . . . , ~ .t975
,~. , M , , ~ 1976 1977 1979 21981, i1983, ~98¢~ 119871 .J- 1989 . . 1990
. . . 1992 1994 1996
Year

2 2 . 7 _. 0. 5 L " + - ~ -: . . . . ~ i I : : ; ; : ~

2 7 . , _ 0 . ° ~ r " - " "

~= 2. o o. O~ L ~ = / ~ e , ~ ,/%4,..,. _ ..,_..,__.,..._~ , . . . . .-.

s4. O _ _ o e.IS ~ ~ , _ ~ , , , , ~.,, ~ . . ~ % , . ~ _ . j.~ .


1973 1975 1976 1977 1979 1981 1983 1985 1987 1989 1990 - - ~992 1994 1996
Year

0 _ ~Ax. ............... ~ . . . . . . . . . . . . . .
~'" ~ -o. arJ-T-F=r * ~ - ~ . . . . . . . . . . . . . .

54.0
1973 1975 1976 1977 1979 t981 1983 1985 1987 1989 1990 1992 1994 1996
Year

]7. 0

=2~.~ 0.1~C
"~r-~4"IT';~J~F:'~'~-:~ ~'- - . . . . . .~.,,.+,.e~,,ar,~,,,_,.~.~
32. o - - 0 ,
0
¢8.0_0.5 ~ ~ ~--~-~:~--~.,',~-,-~: - . ~ ;, :- : : ;

1973 1975 1976 1977 1979 198] 1983 19R5 1987 1989 1990 199Z 1994 1996
Year

Figure 2 The time changes of horizontal field transfer functions Cu, Cv, Fu, Fv, Eu

For Cv and Fv, the short-term premonitions at periods of 64.0, 48.0, 32.0 min are more sig-
nificant, this point is different from Cu and F~. The largest anomaly values of Cv reached to -0.61,
-0.83, -0.90 and are 10, 11, 9 times as large as the a respectively. These values of Fv are -0.65,
No.3 GONGShao-jing, et al: GEOMAGNETICHORIZONTALFIELD TRANSFER FUNCTIONS 301

-0.59, -0.61 and 12, 7, 4 times as large as the a.


Cv and Fv seem also to exist middle-term precursor. Their time change curves ascended as
positive value in the first half of 1975, after July 1975, these curves progressively descended and
in the first half of March 1976 reached to negative maximal. Later they have risen step by step and
reached to normal level or positive valve in July of 1976. In comparison with former results
(GONG, et al, 1997), both all have risen as positive values in 1975, after that dropped as negative
values. Due to the longer duration of data in this paper, the adopted normal values differ from the
former, the duration of anomaly is also different.

5 Discussion and conclusions


1) In this paper, there are obvious short-term precursors of some horizontal field transfer
functions, but there are no precursors for the vertical field transfer functions. According to our
numerical simulation result, the positions of extreme of horizontal field and vertical field transfer
functions are different. Maybe in other cases, the evident fore-runner of vertical field transfer
functions will exist, but horizontal field transfer functions have undoubtedly provided a new ap-
proach to detect anomaly.
2) It will be seen from different anomaly shapes of Cu, Fu and Cv, Fv, that every parameter of
transfer functions should be obtained, in order to present entirely the shape of anomalous varia-
tions and detect the middle-term or short-term precursors. Therefore, the method of complex
transfer functions must be adopted to obtain Re-part and Im-part.
3) In order to find out the short-term precursors, the data of transfer function values should be
dense and a group of transfer functions per 3-10 days should be obtained. But, in order to find out
the impending earthquake precursor, a group of transfer functions must be obtained every day. So
the data of variometer cannot be satisfied and fluxgate magnetometer is needed. After the data is
densified, the amplitude of anomaly is larger. In 1997's paper, the largest amplitude of anomaly
only reached to -0.55, but the result in this paper reaches -1.5. Obviously, former results reflect an
average effect.
4) In this paper, we list a great quantity of data to explain the estimated values affected by
sampling errors. The fluctuation of estimated values brought by large sampling errors will cover
originally not large anomalies. Therefore, it is necessary to improve the method of digitization for
variometer's data. Using data of fluxgate magnetometer can avoid digitized errors.
5) The obvious short-term precursor of horizontal field transfer functions appear from the
latter half of the February to the first half of the March in 1976, the underground water short-term
precursor arose from March to May of 1976 (WANG, et al, 1999), the mobile gravity survey dis-
covered that variation of 180x 10-s m/s 2 exists during the three months before the earthquake (MEI,
1993). The appeared time and order of these short-term precursors maybe reflect an intrinsic
mechanism. The earlier appearing short-term precursor of horizontal field transfer functions
maybe bear relationship with that it reflects the variation of deep conductivity. The anomaly of
underground water reflects the shallow dilatancy-expansion process. The accretion of gravity
value shows that the matter is moving into seismogenic zone.
6) The horizontal field transfer functions have significant short-term precursors, this bears
relation to difference of phase and amplitude between two stations' horizontal components, We
will introduce these results in the other paper.
302 ACTA SEISMOLOGICASINICA Vol.14

Author express heartfelt thanks for Professor QIAN Jia-dong, and thank Baijiatuan and
Changli stations for providing valuable data.

References
Beamish P. 1982A geomagnetic precursor to the 1979 Carlisle earthquake [J]. Geophys J R astr Soc, 68:531-543~
Egbert G D, Booker J R. 1986. Robust estimation of geomagnetic transfer functions [J]. Geophys J R astr Soc, 87:173-194.
GONG Shao-jing. 1983 Analysis on the time series of short period geomagnetic events at Qingguang seismographic station [J]. Earth-
quake, (1): 6-11 (in Chinese).
GONG Shao-jing. 1985 Anomalous changes in transfer functions and the 1976 Tangshan earthquake (Ms=7.8) [J]. JGeomag Geoclectr,
37; 503-508.
GONG Shao-jing 1987. The Parkinson vectors at Guangdong Province' stations and time changes of coefficients at Guangzhou station
before and after Heyuan earthquakes [J]. Earthquakes Research, 10(5): 575-582 (in Chinese).
GONG Shao-jing, CHEN Hun-ran, ZHANG Cui-fen, et al. 1997. The anomalous reactions of the geomagnetic horizontal field transfer
functions before Tangshan earthquake [J]. Acta Seismologica Sinica, 10(1 ): 61 ~70.
GONG Shao-jing, WANG Buo-wei, YU Bin. 1989. The variation of transfer function before and after Songpan earthquake [A]. In: Seis-
mological Bureau of Sichuan Province eds. Collected Work of Songpan Earthquake Prediction Scientific Conference [C]. Beijing:
Seismological Press, 95-98 (in Chinese).
GONG Shao-jing, YANG Gui-jun, TIAN Shah, et al. 1992. Research on the time changes of transfer functions at Heze observatory before
and after Heze earthquake of magnitude 5.9 - - A discussion with Wang Qi [J]. Acta Seismologica Sinica, 5(1): 187-195.
Miyakoshi J. 1975. Secular variation of Parkinson vectors in seismically active region of Middle Asia [J]. J Fac General Education,
Tottori Univ, 8:209-218.
MEI Shi-rong 1993. Selected Works of Earthquake Sciences Research Thesis [M]. Beijing: Seismological Press, 192 (in Chinese).
Rikitake T. 1979. Changes in the direction of magnetic vector of short-period geomagnetic variations before the 1972 Sitka, Alaska
earthquake [J] J Geomag Geoelectr, 31: 441-448.
WANG Ya-ling, WANG An-bin, YAN Xiao-mei, et aL 1999. A study of difference evolution image of water head gradient variation and a
prediction method for strong earthquakes risk regions [J]. North China Earthquake Sciences, 17(4): 7~15 (in Chinese)
Yanagihara K. 1972. Secular variation of the electrical conductivity anomaly in the central part of Japan [J]. Memo Kakioka Mag Obs, 11:
1-11.

You might also like