Sie sind auf Seite 1von 25

Chapter 10

The Gutenberg-Richter Law:


Assumptions, Limitations and
Interpretations
Pablo Palacios, Indira Molina, M
onica Segovia

The Gutenberg-Richter Law (GR) is an empirical relation between the magnitude x of some seismic event, and N(x), the number of events with magnitudes higher than x. Ishimoto and Iida (1939) and Gutenberg and Richter
(1944) proposed the following linear relation:
log10 N(x) = a bx

(10.1)

also named Frequency-Magnitude Relation (FMR), where a and b are constants for a xed data set. We utilize here the equivalent expression:
ln N(x) = ln(n) (x x0 ) ;
where

x > x0 ;

>0

(10.2)

(ln(x) + x0 )
, b=
,
ln 10
ln 10
is the slope of the straight line, and x0 is the minimum homogeneity
magnitude, which has been referred to by some authors as the network sensitivity (e.g. Caneva 2002) and also is known as the completeness magnitude
(Sanchez et al. 2004). This value is used to divide the seismic catalogue into
two clusters of information: complete and incomplete catalogues. The incomplete catalogue comprises magnitudes smaller than x0 and assumes that
a=

the network does not have the capability to uniformly detect all events below this threshold (e.g. Knopo and Kagan 1977). The quantity n = N (x0 )
is the total number of seismic events that belong to the complete catalogue.
Some authors have found a non-linear FMR in tectonic regions, showing
a weak negative curvature or deciencies for higher magnitude events (e.g.
Cosentino and Luzio 1976; Cosentino, Ficarra and Luzio 1977; Knopo
and Kagan 1977; Legrand et al. 2004). Similar deviations from linearity
have been found by Villagomez (2000) and Molina (2001) by using volcanotectonic events.
Mogi (1962a, 1962b, 1962c, 1963) and Scholz (1968) performed experiments forcing fractures into rock samples (micro-events) and computed b,
nding a dependence on material and the magnitude of the events. Warren
and Lathman (1970) also performed experimental tests and found that numerical variations in b are inuenced by thermal gradients. Some authors
have found spatial variations in b, among them, Wiss et al. (2001) and
Sanchez et al. (2004). Legrand et al. (2004) suggested that the non-linear
FMR may be understood as the superposition of various processes, such as
classic elastic rupture and hydraulic fracturing. Therefore, deviations from
linearity in the FMR may be related to temporal and spatial variations in
b.
Assuming that the magnitude of the events has a random distribution
and a probability function, we propose that the nonlinearity of the FMR
can be explained as the temporal displacement of the probability function.
Such temporal displacement may be represented using a sequence of random
variables, with each variable related to a xed period of time. This sequence
is known as a stochastic process (Feller 1978; Papoulis and Pillai 2002).
In this paper, the theoretical properties associated with the GR Law
are analyzed and compared to observed data from several volcanoes and
tectonic zones of Ecuador. The minimum homogeneity magnitude is also
deeply analyzed and called the seismic network sensitivity. A new procedure
is proposed to compute the value of the seismic network sensitivity. In
order to explain the origin of the nonlinear FMR, we propose to analyze the
temporal displacements of the magnitude probability function. For the cases
studied here, these displacements have been conrmed by using statistical
condence intervals.

10.1

Data

In this chapter, we analyze the FMR for Cotopaxi, Guagua Pichincha, Tungurahua and Reventador volcanoes, as well as the Pisayambo region (continental crust) and the Subduction zone in the Ecuadorian coast. Figure 10.1
shows the locations of these volcanoes and tectonic zones. The time pe2

riod studied for each data was variable as follows: Cotopaxi ( January 1996
- March 2005), Guagua Pichincha (October 1997 - October 1999), Tungurahua (January 1994 - March 2005), Reventador (March 2003 - March
2005), Pisayambo (January 1990 - March 2005) and Subduction zone (January 1990 - April 2005).
The Ecuadorian seismic network was deployed in 1988 with the rst
seismic stations in the Inter-Andean Valley and on the anks of a few potentially active volcanoes. Two of the most dangerous volcanoes, Cotopaxi
and Guagua Pichincha have had one seismic station operating since 1976
and 1981 respectively. Later, both the coastal areas and the eastern region
began to be monitored seismically. With time, networks were improved on
the more dangerous volcanoes so that hypocentres could be located. Meanwhile, a database of volcanic-seismic and tectonic events was created and
continues to be updated and maintained by the Instituto Geofsico of the
Escuela Politecnica Nacional (IG-EPN).
In general, data recording has been continuous and consistent in terms
of the coda magnitude calculations for the time periods analyzed. For Cotopaxi, Tungurahua, and Reventador volcanoes, the general relation of Lee
and Stewart (1981) has been used. For Pichincha Volcano a coda magnitude was used. For Pisayambo and the Subduction zone, the calculation
of magnitude was performed with a coda function, a relation derived and
equivalent to Mb (local magnitude).
Tungurahua Volcano, located at the center of the Ecuadors Cordillera
Real, has been in an eruptive state since 1999. The eruption has been
monitored by a network of 7 to 10 seismic stations located within 2 to 22
km of the summit crater. The reference station to calculate magnitude is
Retu, which is 2 km from the summit (Molina 2001). Cotopaxi Volcano had
a seismic network of 5 permanent stations until 2001. Its reference station is
VC1, which is located on the north eastern ank at 5 km from the summit.
Since late 2001, following increased seismic unrest, the seismic network was
operated with 8 stations. In 2002, Reventador volcano had its rst station
deployed at 14 km NNE. of the crater. Following the paroxysmal eruption in
late 2002, three closer stations were installed. The reference station CONE
is located 500 m to the NE of the cone, inside the caldera. At present,
the National Seismic Network, composed of 22 stations, covers much of
the Inter-Andean Valley and spans about two-thirds of the country. The
reference station utilized here is also VC1.
Since 1998, Guagua Pichincha and Tungurahua have exhibited unrest
and eruptions. The background seismicity of both volcanoes rose dramatically and a variety of events have been recorded. In late 2001, Cotopaxi
Volcano renewed its seismic activity with an increased number of volcanoseismic events. In 2002 Reventador Volcano erupted following little volcano3

seismic activity.
The Pisayambo Seismic Region, which has seen about 35% of the total
number of events recorded by the National Seismic Network in Ecuador,
has exhibited important activity, sometimes characterized as swarm-like
and at other times similar to main event - aftershock sequences. Despite
the fact that IG-EPN has increased the number of seismic stations, the
seismic network sensitivity for this zone remains constant. This is due to
the location of the Pisayambo Region, between Cotopaxi and Tungurahua
volcanoes, both having been monitored since the installation of the seismic
network.
On the other hand, the seismic activity on the coast, particularly in the
subduction zone, has apparently changed since the beginning of the national
seismic network. For several years, the seismic activity has increased with
the occurrence of several energetic and numerous seismic swarms compared
to the period between 1990 and 1995.

10.2

Method

10.2.1

Analysis of the GR Law

Properties of the GR Law


Using expression 10.2, it is possible to dene a continuous random variable,
which relates an earthquake or volcano-tectonic event to its magnitude x.
If X is this random variable and x its random result, the magnitude probability function is exponential, truncated at x0 :

exp [ (x x0 )] , x > x0
(10.3)
f (x) =
0,
x < x0
The probability that X is between the values a1 and a2 is:
 a2
f (x) dx
P (a1 < X < a2 ) =

(10.4)

a1

Expression 10.3 satises the conditions f (x) 0 and P (X > x0 ) = 1. It


must be noted that the probability for an event with magnitude smaller
than x0 is null:
(10.5)
P (X < x0 ) = 0
There are two assumptions implicit in the use of the GR law in this application:
1. The mechanical behaviour of fracturing rocks can be described by a
random physical process: Expression 10.2 or 10.3 implies the use of
4

a random model to describe the fracturing of rocks or the sliding of


faults. We are not using a deterministic model. This is reasonable,
based on the complexity of volcano-tectonic systems.
2. It is possible to evaluate the minimum absolute sensitivity of the seismic network: A seismic network may not be able to detect all seismic
events, especially when they are very small or far from the network.
The minimum homogeneity magnitude x0 is usually identied as the
network sensitivity value.
Estimation of the parameter with a fixed x0 and
discarding incomplete catalogue data
When the probability function 10.3 is used, x0 is xed and the data of
the incomplete catalogue is discarded before estimating the parameter.
In order to determine x0 , the FMR curve is drawn and the linear portion
of the curve is demarcated. The x0 magnitude is the minimum value of
this linear portion. Commonly, x0 is chosen by visual inspection. Another
equivalent means of choosing x0 is to draw the histogram related to the
probability function 10.3 with the data grouped into classes. Then, x0 is
the minimum magnitude of the class with maximum frequency. Caneva
(2002) presents a similar method for choosing x0 .
After discarding the data belonging to the incomplete catalogue, there
are two ways to estimate the parameter of the GR law. First, one can
assume that the data are from a continuous random variable with a probability function of type 10.3, but without grouping the data in classes. Aki
(1965) used this assumption and calculated the estimator of the param using the maximum likelihood method (Kalbeisch 1979; Papoulis
eter ()
and Pillai 2002):
n
1
1

=
, X=
Xj
(10.6)
n j=1
X x0
where the sum is based on the complete catalogue {X1 , X2 , . . . , Xn } of a
random sample.
Second, can be calculated by using data grouped in classes. To
compute the probability of each class, relation 10.4 is used with a1 and a2
as the limits of each class. For this case Garca and Palacios (2003), also
using the maximum likelihood method, proposed:
 

if i
1
i=1

= ln 
(10.7)

i=1 if i n
Where is the class width, fi is the absolute frequency of the class i, and
n is the number of observations in the complete catalogue.
5

In order to study the temporal behaviour of (for any interval of time),


x0 is xed using the whole data. Following this procedure, the value, x0 , is
unique. Therefore, samples of sub-periods are taken in a sequential way in
the time domain and its corresponding is calculated using equation 10.6
or equation 10.7. This procedure is justied when the network has had no
relevant changes during the period of study, and assuming that x0 is the
network sensitivity. We dene this procedure as P0 .
Estimation of the parameter with a variable x0 and eliminating
incomplete catalogue data
From relation 10.3 note that x0 is the displacement value of the probability
function in the magnitude domain. If x0 represents the xed network sensitivity, no relevant changes in this value should be expected. The temporal
variations of x0 can be calculated for each sample (related to a period of
time) using the maximum frequency class. The incomplete catalogue data
of each sample are discarded before calculating by using (Equation 10.7).
We dene this procedure as P1 .
Estimation of x0 and parameters, without eliminating incomplete catalogue data.
Now consider x0 and as parameters to be estimated. If the incomplete
catalogue corresponds to events beneath the uniform detection threshold,
a better estimate of x0 must be smaller than that value obtained through
empirical choice, because the relation given in equation 10.5 needs to be
satised. This is why all sample data (complete and incomplete catalogue)
should be included. Let N be the total number of observations in the whole
sample. The likelihood function l (x0 , ) is:



x0
(10.8)
l (x0 , ) = N ln () X

However this function does not have a maximum x0 , because the system
l/x0 = 0; l/ = 0 does not have a solution. Therefore, the maximum
likelihood method can not be applied.
In order to estimate x0 and it is possible to use the method of moments (Kalbeisch 1979), solving the following equation system:

(10.9)
E X k = mk , k = 1, 2
where

N
1  k
X .
mk =
N j=1 j

. Using this method with the exponentials (EquaNote that m1 = X


tion 10.3), the expected values of X and x2 are:
1

2
2x0
+ 2
E(X 2 ) = x20 +

E(X) = x0 +

(10.10)
(10.11)

Then, the solutions of the system (eq. 10.9) for non-grouped data are:

(10.12)
x0 = m1 m2 m21
= 

1
m2 m21

and the estimators for data grouped in classes are:



1
x0 = m1
2 + 4 (m2 m21 )
2


2 + 4 (m m2 ) +

1
2
1
= ln 

2 + 4 (m2 m21 )

(10.13)

(10.14)
(10.15)

We dene this procedure to estimate x0 and (without discarding the


incomplete catalogue data) as P2 .
A criterion for a better parameter estimation of the GR law
The standard deviation of a sample is:

N
(m2 m21 )
S=
N 1

(10.16)

Comparing equations 10.16 and 10.12, x0 is approximately one standard


deviation smaller than the sample mean m1 . For data grouped in classes
and using equation 10.14, we can write the deviation m1 x0 as follows:


1
4(N 1) 2
1
2 + 4 (m2 m21 ) =
2 +
(10.17)
S
2
2
N 
2



N 1
N

1 +
= S
N
2S N 1

 
2

1
N 1
N

1+
S
N
2 2S N 1
7

2S

N
N 1

This approximation is valid if


< 1 . The dierence between
x0 and m1 (for grouped data) is bigger than for non-grouped data. The
distinction between grouped and non-grouped data decreases with a small
. In addition, the half of total number of classes with width , roughly
belongs to an interval of magnitude with length 2S. Therefore, the dierence between x0 and m1 is considered approximately equal to one standard
deviation.
If we suppose that there is some statistical theoretical source of random
events, which has a proportion p of events with magnitudes smaller than
m1 x0 , the sample data from that source could have a sample proportion
p that might be dierent than the expected proportion p . This dierence is
given only as a random eect of the sampling process. On the other hand,
it is known that with a 100(1 )% probability ( is the complementary
probability) the sample proportion p belongs to the following interval (Hines
and Montgomery 1993):

p(1 p)
(10.18)
p p + z
N
where z is the percentile of the Gaussian Standard Distribution. The
maximum sample proportion of this range is dened as:

p(1 p)
pmax (, p, N) = p + z
(10.19)
N
In particular, if the data of each sample are approximately related to a
Gaussian Distribution, the expressions 10.12 and 10.14 suggest that around
15.9% of the data might have magnitudes smaller than x0 , so in this case
p = 15.9%. With = 5%, the probability that the sample proportion p
belongs to the interval 10.18 is 95% and, as a consequence, z = 1.64 and
pmax is:
0.6
(10.20)
pmax (N) = 0.159 +
N
We dene a Big Incomplete Catalogue (BIC) if the proportion of the incomplete catalogue is more than pmax of the sample data, otherwise, it would
be a Small Incomplete Catalogue (SIC).
Therefore, for a BIC sample may be biased with an over-estimation
if the P1 procedure is used. The over-estimation occurs because the decaying exponential level is high when the incomplete catalogue information is
discarded. Also with the P1 procedure, x0 is an over-estimation.

10.2.2

The Central Limit Theorem

The displacement of a probability function can be proven through statistical


methods. A measure of the central tendency of a data set is the sample
mean. However if two or more samples are taken from a stationary random
source (for instance, from tossing a coin or throwing a die) dierent means
may be found only because the sampling process is random and generates
dierent data sample sets. On the other hand, it is possible to have physical
changes of the random source (for instance, the balance of the coin or die
can be changed intentionally), and, as a result, dierent sample means can
also appear. To distinguish whether sample mean dierences are due to
any physical change of the random source (we take each source as the whole
volcanic or tectonic region), condence intervals for the population mean
and the Central Limit Theorem (Feller 1978; Papoulis and Pillai 2002) could
is a random variable
be used. This theorem asserts that the sample mean X
that follows approximately a Gaussian Law, with the population mean
(a parameter of the random source) as its expected value, and N as its
standard deviation, where is the population standard deviation, and N
the number of sample events. To get a good approximation N 30 is
required, and the sample standard deviation S instead of could be used.
With a (1 )100% condence level, the population mean may belong to
the following interval:
z/2 S X
+ z/2 S
X
N
N

(10.21)

with z/2 being the percentile of the Gaussian Standard Distribution. The
interval 10.21 is known as the Confidence Interval for the population mean
. For instance, with 95% condence, belongs to the following interval:
+ 1.96 S
1.96 s X
X
N
N

(10.22)

For two or more given samples, if the condence intervals overlap, then
belongs to all of them and no physical change of the source can be proven.
However, if the condence intervals do not overlap, it is possible to assert
that the source has had some physical change with a high condence level.

10.3

Results

10.3.1

Sampling Process

For each volcano or tectonic zone, the magnitudes of the events were sorted
in the time domain. We took samples of 100 events, each one with two
9

consecutive samples overlapped by 50%. Therefore 50 events belong to


consecutive samples. A descriptive statistical analysis showed that a class
width equal to 0.2 (Cotopaxi and Guagua Pichincha) or 0.25 (Reventador,
Tungurahua, Pisayambo and Subduction zone) was convenient. If we select
a class width with values smaller than 0.2, each sample histogram has an uneven shape. This observed histogram does not t the theoretical histogram
obtained through the function 10.3. The left and right border of magnitude
for each class is opened (not included) and closed (included) respectively.
Each sample was labelled with the date corresponding to the last event of
the sample. This label is named final date.

10.3.2

FMR for the Studied Cases

Figure 10.2 shows that the observed FMR values obtained for each geographical region is non-linear or has some deviation at higher magnitudes.
Here the entire data set for each geographical region is considered as a
unique sample. The minimum homogeneity x0 and b values that appear in
the plots of Figure 10.2 were calculated using the P0 procedure (for grouped
data). The black dots and bars are related to the complete catalogue. The
white bars and open circles are related to the incomplete catalogue. If
the classical expression 10.6 is used, the estimates of b values are slightly
smaller than those obtained with equation 10.7. The b value estimates for:
Cotopaxi, b = 1.20; Pichincha, b = 1.13; Tungurahua, b = 0.76; Reventador b = 0.92; Pisayambo Zone, b = 1.08; and Subduction Zone, b = 1.38.
Therefore, the deviations in the observed FMR at high magnitudes are independent of whether the data are grouped or not. When a test of quality
of tting was performed, the null hypothesis, which proposed that the data
were from a law like equation 10.3, was refused for all cases studied here.
This result was also conrmed for the Subduction Zone, which shows an
apparently good t (Figure 10.2(f)).

10.3.3

Estimates of x0 and b Using the P0 , P1 and P2


Procedures

ln 10)
For each sample the minimum homogeneity x0 and the b value (b = /
were calculated with the P0 , P1 and P2 procedures. Figure 10.3 shows the
time-varying estimates of x0 . The estimate of x0 using the P0 procedure is
single-valued for each geographic region, and is the same as in Figure 10.2.
For SIC samples of all studied cases, the dierences found in x0 were small
when P1 and P2 procedures were compared. In Figure 10.3, (b) and (f) show
the SIC samples. Therefore, for SIC samples, eliminating or not eliminating
the incomplete catalogue data, does not signicantly change x0 . On the
10

other hand, for BIC samples, x0 values found with the P1 procedure are
higher than ones calculated with the P2 procedure.
Figure 10.4 shows estimates of b using the P0 , P1 and P2 procedures.
When the P0 and P1 procedures are used, time periods with high b values
may appear. The peaks of the dashed lines (values of the P0 procedure) and
thin lines (values of the P1 procedure) in Figure 10.4 show this behaviour
for the studied cases. However, when the P2 procedure is used (thick line),
the highest b values disappear. For SIC samples the estimates of b derived
from the P1 and P2 procedures do not show large dierences. A similar
situation was observed with the estimates of x0 because the discarded data
of SIC samples have little inuence on the estimation of parameters.

10.3.4

BIC and SIC Samples

For samples of size N = 100, and, taking into account the assumptions
of equation 10.20, the maximum sample proportion is pmax = 21.9. This
value divides the SIC and BIC samples for our study cases. Table 10.1(a)
shows the proportion of BIC samples in each case, and it is easily noted
that these types of samples are common. Using the P2 procedure all cases
had 100% of SIC samples. The exception was Reventador Volcano, with
only one BIC sample (data from 13 Jun 2003 08h03 to 26 Jun 2003 04h01
GMT). For this sample the normality assumption of Equation 10.19 was not
satised. Considering all information for each case as a single sample, the
proportion of events belonging to the incomplete catalogue was computed.
Table 10.1b shows these results. The minimum homogeneity magnitudes
were calculated using the P0 procedure and the values are shown in the
Figure 10.2. Table 10.1(b) shows dierent pmax values because each case
has a dierent number of events.

10.3.5

Confidence Intervals for the Studied Cases

To prove if there are displacements of the probability function related to


physical changes, condence intervals for the population mean have been
calculated. Following expression 10.22, an interval with 95% condence has
been used, and the results of each case are presented in Figure 10.5. Some
periods show non-overlapped condence intervals. An example of each case
has been selected and marked with a gray rectangle in gure 10.5. The gray
periods are: Cotopaxi (22 September 1999 - 9 November 2001), Guagua
Pichincha (9 January 1998 - 22 February 1999), Tungurahua (13 December
1996 - 11 March 1998), Reventador (1 October 2004 - 18 October 2004),
Pisayambo Zone (8 November 1996 - 29 April 1997), and Subduction Zone
(21 April 2002 - 25 January 2005). Three time ordered samples have been
11

chosen from these gray periods, and its smoothed frequency polygon (from
the histogram) is drawn. The numbers 1, 2, and 3, show the sequence, and
are located at the minimum homogeneity magnitude (calculated using the
P1 procedure). The dashed line (rst sample), the solid thin line (second
sample) and the solid thick line (third sample) show the same sequence.
The periods marked in gray are related to non-overlapped condence
intervals, which are statistical evidence of displacements of the probability
function. With the exception of Pichincha Volcano, these displacements are
from low to high magnitudes. For Pichincha Volcano a gray period related
to a displacement from high to low magnitudes was selected.

10.3.6

Correlation Between x0 and Sample Mean


Magnitude

We found good correlations between the minimum homogeneity magnitude


and sample mean magnitude for each case. The correlations of these two
variables are: Cotopaxi 87.8%, Guagua Pichincha 95%, Tungurahua 95.3%,
Reventador 93.9%, Pisayambo Zone 92.1% and Subduction Zone 95.7%.
The examples of gray periods of Figure 10.5 show the x0 value with numbers
1, 2, and 3. The correlations appear as a consequence of the displacement of
the probability functions. When displacements are taken into account, both
the minimum homogeneity and sample mean magnitude have displacements
in the same direction.

10.4

Discussion

The experiments performed by Mogi (1962a, 1962b, 1962c, 1963) and Scholz
(1968) on laboratory-scale rock samples show that changes of the b value
might be related to physical processes. There is of course a scale problem
if one tries to extend their conclusions to large-scale volcanic or tectonic
systems. However, suggestions can be drawn from their work. In particular
Scholz (1968) describes three stages during the fracturing process caused
for increasing stresses:
1. At low stress in unconned experiments, micro-fracturing was attributed to frictional sliding on pre-existing cracks and the collapse
of pores.
2. In a second stage, at stresses from about 30 to 50 percent of the
fracture strength, the rock was nearly elastic and few events could be
detected.
3. In the third stage, micro-fracturing resumes and steadily increases
until fracture. In this latter region, the rock becomes dilatant, i.e., if
12

we disregard elastic compression, the rock increases in volume. In this


stage, micro-fracturing radiation was shown due to the propagation
of new cracks.
From the experiments of Scholz and Mogi, it is clear that during the rst
stage the magnitudes of the events are generally low (due to the collapse of
pores). In the second stage, the activity decreases with events of magnitude
greater than those of the rst stage (more elastic energy is accumulated).
And, in the last stage, the propagation of new cracks is linked to event
magnitudes greater than those of the second stage (the propagation of new
cracks and the increase of volume need high energy). This physical process
implies that it is necessary to use a non-stationary probability function
that is not xed in time. As a consequence, when the rst stage occurs, it is
expected (with high probability) that the events have low magnitudes, then
in the second stage, the probability for events of middle magnitude (the
probability function has a displacement) increases, and in the last stage,
the probability of events occurring with high magnitude increases due to a
displacement of the probability function once more. We propose to relate
the stages drawn by Scholz to the displacement of the probability function
of real cases, which can be detected using condence intervals.
When a truncated exponential function like Equation 10.3 is used and
xed in the time domain, a linear FMR appears. However, if there are
displacements from Equation 10.3 it is possible to obtain a non-linear FMR,
because the total result would be a combination of straight lines, that is, a
curved line.
The probability function does not only have displacements from low to
high magnitudes. Intentionally, in Figure 10.5(b), a period from Guagua
Pichincha Volcano, with a displacement from high to low magnitudes, has
been selected. The situation may be understood by taking into account the
complexity of the system. If a structure has inner stresses, then, when some
parts reach the third stage of Scholz, other parts can be beginning the rst
stage, decreasing the sample mean magnitude. Therefore, the system shows
oscillations of the condence intervals that estimate the mean magnitude,
shown in the Figure 10.5.
The correlation of the sample mean and the minimum homogeneity
magnitude shows that the minimum homogeneity is linked to the same physical process, and it explains the correlation. However a non-xed minimum
homogeneity is a serious problem for the GR law, because it assumes that
x0 is the network sensitivity. Using the P2 procedure, a better estimation
of x0 was tried, but correlation with the sample mean was not eliminated.
Using the P2 procedure, the correlation is evident, due to Relations 10.12
and 10.14. On the other hand, the high percentages of BIC samples shown
in Table 10.1 are evidence of a poor application of the GR law. In our study,
13

there are no important changes in the seismic networks that may explain
the changes in the minimum homogeneity values.
Scholz (1968), and others, observed a decreasing b value with increasing
magnitude, and an increasing b value with decreasing magnitude. Those b
values were computed using a unique x0 and all information (the P0 procedure) and not for each sample. This behaviour of the b value may be understood using the displacement of the probability function. With a unique
x0 , the slope of the FMR (b value) decreases with high magnitudes because
the probability function has a displacement with respect to high magnitude
values, and increases upon returning to small magnitudes. The gray periods in Figures 10.4 and 10.5 are the same and the b values (dashed lines in
Figure 10.4) decrease while the probability functions have displacements to
high magnitudes and otherwise increase (gray period of Pichincha volcano).
This behaviour of the b value, with a unique x0 , is also a consequence of the
displacement of the probability function.

10.5

Concluding Remarks

The use of the GR law implies the acceptance of the truncated exponential functions. The minimum homogeneity magnitude is a threshold that
divides the catalogue in two groups: complete and incomplete catalogues.
A theoretical consequence is that the probability of the incomplete catalogue is null. If the incomplete catalogue of some sample has few events
(SIC sample), the use of the GR law may be considered as an acceptable
approximation. However if the incomplete catalogue has many events (BIC
sample) the b value would not be a good estimate. Considering that the
minimum homogeneity and the b value are parameters which need to be estimated, an alternative procedure is proposed taking into account complete
sample information.
The existence of temporal displacements of the probability function,
have been proven in all cases (four Ecuadorian volcanoes and two tectonic
zones) studied in this paper, with high levels of statistical condence. We
consider that these displacements aect the shape of the FMR.
The use of the Central Limit Theorem has two important advantages.
First, it allows one to investigate possible variations of the population mean,
independently of the type of probability function. In other words, regardless
of whether the GR law and the truncated exponential functions are right, the
displacement of the probability function can be detected. Second, variation
in the population mean of a data source implies physical changes, with a
given condence level, and therefore can reproduce some characteristic of
its underlying physical process.
The experimental results of Mogi (1962a, 1962b, 1962c, 1963) and
14

Scholz (1968) may be considered as suggestions about what is happening


physically in some volcanic or tectonic regions. There are two basic problems with the direct application of the experimental results to real cases.
First, with the change of scale between the laboratory and eld conditions,
physical variables can change the expected results. Second, laboratory studies deal with unique individual samples, whereas the eld situation involves
a complex system, with a combination of many samples, each one in a different physical condition. However Scholz (1968) observed three stages in
the fracturing process of rocks, which we have explained in terms of displacements of the probability function.
Finally, the minimum homogeneity magnitude shows a strong correlation with the sample mean magnitude. This may be understood as a result
of the displacement of the probability function. If the probability function
has a displacement, the magnitude related to the maximum frequency (the
minimum homogeneity) roughly has the same displacement, and therefore a
strong correlation can appear. All cases studied here show correlations over
87 percent. Our conclusion is that the minimum homogeneity magnitude
is not related to network sensitivity. In addition, we consider that network
sensitivity may be dened by taking into account the sensitivity of each
seismic sensor, the number of them and their geometrical distribution, the
heterogeneities of the ground and the characteristic properties of the seismic source. A denition which includes these points must give a stationary
network sensitivity value and must agree with observed data collected by
the seismic network.

Further Reading
For seismology you can read Quantitative Seismology by Aki and Richards
(2002). In this classical text book of seismology you can nd the concepts of
stress and strain, which are important for understanding the source physical
process and to go deeper into the assumptions stated in our work. Excellent
texts for an introduction to statistical concepts are in two volumes of Probability
and Statistical Inference by J.G. Kalbeish (1979). In particular, the Maximum
Likelihood Method is discussed in detail in the second volume. The Introducci
on
a la Teora de Probabilidades y sus Aplicaciones by Feller (1978) has very good
discussions about the Central Limit Theorem and Stocasthic Processes.

Acknowledgements
We would like to thank Hugo Yepes and the sta of Geophysical Institute, National Polytechnic School of Quito, for supporting the last three year of research,
and for the maintaining of the networks and the acquiring data process. This
work also was supported with funds of Statistic in Volcanology Workshop as part

15

of the Environmental Mathematics and Statistic Programme founded jointly by


NERC/EPSRC, UK. We are also very grateful to Constanza Bonadona for an
early review of this paper and her useful comments. We also thank Je Jhonson, Patricia Mothes and an anonymous reviewer for their valuable comments,
improving this paper.

16

Bibliography
[1] Aki, K. 1965. Maximum Likelihood Estimate of b in the formula logN =
a bM and its condence limits. Bulletin of the Earthquakes Research
Institute, 43, 237-239.
[2] Aki, K. & Richards, P. 2002. Quantitative Seismology. University Science Books, USA.
[3] Caneva, A. 2002. Magnitud Representativa. Primer Simposio Nacional
Colombiano de Sismologa. Memorias.
[4] Cosentino, P. & Luzio, D. 1976. A generalization of the frequencymagnitude relation in the hypothesis of a maximum regional magnitude. Annales Geophysicae, 29, 3-8.
[5] Cosentino, P., Ficarra, F. & Luzio, F. 1977. Truncated exponential
frequency-magnitude relation in earthquake statistics. Bulletin of the
Seismological Society of America, 67, 1615-1623.
[6] Feller, W. 1978. Introducci
on a la Teora de Probabilidades y sus Aplicaciones. Ed. Limusa, 1989, Mexico.
[7] Garca J. & Palacios, P. 2003. Analisis de los supuestos asumidos en
los modelos de distribucion de sismos y su contraste con datos experimentales, Tesis-Maestra en Estadstica. E.P.N. Quito.
[8] Gutenberg & Richter 1944. Frequency of earthquakes in California.
Bulletin of the Seismological Society of America, 34, 185-188.
[9] Hines, W. & Montgomery, D. 1993. Probabilidad y Estadstica para
Ingeniera y Administracion, Compa
na Editorial Continental S.A., Segunda Edicion, Mexico.
[10] Ishimoto, M. & Iida, K. 1939. Observations sur les seims euregistre par
le microseisograph construite dernierment (I). Bulletin of the Earthquake Research Institute-University of Tokyo, 17, 443-478.

17

[11] Kalbeisch, J. 1979. Probability and Statistical Inference. SpringerVerlag, New York.
[12] Knopo, L. & Kagan, Y. 1977. Analysis of the theory of extremes
as applied to earthquake problems. Journal of Geophysical Research,
82(36), 5647-5657.
[13] Lee & Stewart 1981. Principles and applications of micro-earthquake
networks, Advances in Geophysics, Supplement 2, Academic Press, NY,
USA. Geological Survey Open-File Report 75-311, 113 pp.
[14] Legrand D. et al. 2004. Multifractal dimension and b value analisis
of the 1998-1999 Quito swarm related to Guagua Pichincha volcano
activity, Ecuador. Journal of Geophysical Reserch, 109, B01307.
[15] Mogi, K. 1962. Study of Elastic Shocks Caused by the Fracture of
Heterogeneous Material and its Relations to Earthquake Phenomena.
Bulletin of the Earthquake Research Institute, 40, 125-173.
[16] Mogi, K. 1962. The Fracture of a Semi-innite Body Caused by a Inner
Stress Origin and its Relation to the Earthquake Phenomena (First
Paper). Bulletin of the Earthquake Research Institute, 40, 815-829.
[17] Mogi, K. 1962. Magnitude-Frequency Relation for Elastic Shocks Accompanying Fractures of Various Materials and Some Related Problems
in Earthquakes. Bulletin of the Earthquake Research Institute, 40, 831853.
[18] Mogi, K. 1963. The Fracture of a Semi-innite Body Caused by a Inner
Stress Origin and its Relation to the Earthquake Phenomena (Second
Paper). Bulletin of the Earthquake Research Institute, 41, 595-614.
[19] Molina, I. 2001. Actividad pre-eruptiva del volcan Tungurahua, Trabajo de tesis para optar al ttulo de Geologa, Universidad de CaldasManizales, Instituto Geofsico-Quito.
[20] Papoulis, A. & Pillai, U. 2002. Probability, Random Variables, and
Stochastic Processes. Ed. Mc Graw Hill.
[21] Sanchez, J., McNutt, S., Power, J. & Wyss, M. 2004. Spatial variations
in the Frequency-Magnitude Distribution of Earthquakes at Mount
Pinatubo Volcano. Bulletin of the Seismological Society of America,
94(2), 430-438.
[22] Scholz, C. 1968. The Frequency-Magnitude Relation of Microfracturing
in Rock and its Relation to Earthquakes. Bulletin of the Seismological
Society of America, 58, 399-415.
18

[23] Villagomez, D. 2000. Sismicidad del Volc


an Guagua Pichincha. Tesis
de grado, Escuela Politecnica Nacional, Quito-Ecuador.
[24] Warren, N. & Latham, G. 1970. An Experimental Study of Thermally
Induced Microracturing and Its Relation to Volcanic Seismicity. Journal of Geophysical Research, 75(23), 4455-4464.
[25] Wyss, M., Klein, F., Nagamine, K., & Wiemer, S. 2001). Anomalously
high b values in the South Flank of Kilauea volcano, Hawaii: evidence
for the distribution of magma below Kilaueas East rift zone. Journal
of Volcanology and Geothermal Research, 106, 23-37.

19

(a)
Volcano
Cotopaxi
Pichincha
Tungurahua
Reventador
Tectonic Zone
Pisayambo
Subduction

BIC
Samples
76.1%
58.8%
55.1%
57.0%

(b)
Events in
Incomplete
Catalogue
31.8%
15.0%
44.8%
31.6%

Number
of Events
3365
866
3477
8253

pmax
16.9%
17.9%
16.9%
16.6%

Incomplete
Catalogue
Type
BIC
SIC
BIC
BIC

60.5%
75.0%

10.4%
37.6%

8375
1833

16.6%
17.3%

SIC
BIC

Table 10.1: (a) Percentage of BIC samples for each volcanic or tectonic
zone using the P1 procedure. (b) Percentage of events in the incomplete
catalogue, considering each volcanic or tectonic zone as a unique sample.

20

1
0
Latitude S ()

Colombia

Pacific Ocean
6

2
1
5
3

-1
-2

Ecuador

-3
Per
-4
-5
-83

0
-82

-81

-80 -79 -78


Longitude W ()

50

100 km
-77

-76

Figure 10.1: Location of tectonic zones and volcanic centers. Solid triangles
show the volcanoes: (1) Cotopaxi, (2) Guagua Pichincha, (3) Tungurahua
and (4) Reventador. The gray areas show the limits of the Pisayambo
seismic zone (5) and the subduction zone (6).

21

2.0
1.5
1.0
0.5
0.0

200
150

2.0
1.5
1.0
0.5

1.0

50

0.5

0.0

0.0

2000
1500

3.0
2.5
2.0
1.5
1.0
0.5
0.0

Number of events

x0 = 2.85

b = 1.19

Log 10 (N(x))

3.5

x0 = 1.25

1000
500
0

(f)

4.5
4.0

b = 1.03

4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0

Magnitude X - Lower border of class (width 0.25)

0.60
0.85
1.10
1.35
1.60
1.85
2.10
2.35
2.60
2.85
3.10
3.35
3.60
3.85
4.10
4.35
4.60
4.85
5.10
5.35

Number of events

4000
3500
3000
2500
2000
1500
1000
500
0

Pisayambo Zone 1990 - 2005

Reventador 2003 - 2005

2500

Magnitude X - Lower border of class (width 0.25)

(e)

1.5

-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
1.25
1.50
1.75
2.00
2.25
2.50
2.75
3.00

2.5

2.0

x0 = 1.50

Subduction Zone 1990 - 2005

800
700
600
500
400
300
200
100
0

3.5
3.0
2.5

x0 = 3.95

b = 1.48

2.0
1.5
1.0
0.5

2.45
2.70
2.95
3.20
3.45
3.70
3.95
4.20
4.45
4.70
4.95
5.20
5.45
5.70
5.95
6.20
6.45
6.70
6.95

3.0

Number of events

b = 0.77
x0 = 1.35

Log 10 (N(x))

3.5

2.5

100

(d)

Tungurahua 1994 - 2005

-0.40
-0.15
0.10
0.35
0.60
0.85
1.10
1.35
1.60
1.85
2.10
2.35
2.60
2.85
3.10
3.35

Number of events

900
800
700
600
500
400
300
200
100
0

3.0

Magnitude X - Lower border of class (width 0.2)

Magnitude X - Lower border of class (width 0.2)

(c)

b = 1.18

250

Log 10 (N(x))

2.5

3.5

Log 10 (N(x))

3.0

300

Log 10 (N(x))

b = 1.23

Guagua Pichincha 1998 - 1999

-0.1
0.1
0.3
0.5
0.7
0.9
1.1
1.3
1.5
1.7
1.9
2.1
2.3
2.5
2.7
2.9
3.1

x0 = 2.00

Number of events

(b) 350

3.5

Log 10 (N(x))

800
700
600
500
400
300
200
100
0

Cotopaxi 1996 - 2005

1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
3.8

Number of events

(a)900

0.0

Magnitude X - Lower border of class (width 0.25)

Magnitude X - Lower border of class (width 0.25)

Histogram of observed
absolute frequencies
x0: minimum homogeneity magnitude
Complete catalogue: x > x0 (black bars and dots)
Incomplete catalogue: x <= x0 (white bars and dots)

Observed FMR values:


Log 10 (N(x))
Calculated FMR values:
Log 10 (N(x)) = a - bx

Figure 10.2: Histograms of absolute frequencies and Frequency-Magnitude


Relations (FMR) obtained for four volcanoes and two tectonic zones of
Ecuador. Volcanoes: (a) Cotopaxi, (b) Guagua Pichincha, (c) Tungurahua,
(d) Reventador, and the Tectonic Zones: (e) Pisayambo and (f) Subduction.

22

2003 03 22
2003 05 24
2003 06 30
2003 10 07
2003 11 30
2004 02 06
2004 03 21
2004 04 30
2004 06 18
2004 08 13
2004 09 22
2004 10 01
2004 10 24
2004 11 05
2004 11 07
2004 11 15
2004 11 22
2004 12 18
2005 01 11
2005 02 06
2005 03 07

2.0
1.5

x0

(c)

(e)

3.0

2.6
3.6

2.2
3.2 SIC

x0

1.0
0.5

1994 12 24
1995 10 11
1996 03 13
1996 12 13
1997 11 07
1998 09 18
1999 01 09
1999 02 06
1999 05 03
1999 05 16
1999 06 07
1999 06 28
1999 07 31
1999 08 11
1999 08 21
1999 08 28
1999 09 03
1999 09 09
1999 09 16
2000 01 30
2002 07 11
2003 05 07
2004 11 19

x0
1996 08 14
1997 06 09
1998 02 22
1999 09 20
2000 09 04
2001 06 03
2001 11 24
2001 11 26
2001 12 11
2001 12 13
2001 12 14
2001 12 25
2002 01 10
2002 02 26
2002 03 25
2002 06 07
2002 09 01
2002 10 23
2003 01 28
2003 04 19
2003 10 05
2004 09 03
2005 03 31

1998 10 07
1998 12 24
1999 02 22
1999 03 12
1999 04 14
1999 05 18
1999 06 09
1999 08 28
1999 09 08
1999 09 10
1999 09 19
1999 09 21
1999 09 21
1999 09 23
1999 09 23
1999 10 19
1999 11 08

x0
2.0

x0

(a)

1993 01 09
1994 10 29
1996 04 05
1996 12 10
1997 02 01
1998 02 05
1998 08 05
1999 04 17
2000 05 25
2000 10 18
2001 09 14
2002 04 21
2003 08 02
2005 01 05
2005 01 25
2005 01 28
2005 01 31
2005 02 21

1990 12 15
1992 07 15
1993 06 28
1994 09 27
1995 07 12
1996 03 20
1996 12 13
1997 09 21
1998 03 27
1998 11 05
1999 04 15
2000 02 02
2000 10 28
2001 04 04
2001 10 22
2002 03 17
2002 10 03
2003 05 16
2003 11 21
2004 04 28
2004 11 14

x0

P0
P1
P2

Cotopaxi

(b)

2.4

1.6

1.2
2.0
1.8
1.6
1.4
1.2

Tungurahua

2.5

Pisayambo Zone

3.4
4.4

final date

23
4.0

Guagua Pichincha

SIC

(d)

(f)

SIC
SIC
SIC
SIC

final date
final date

Reventador

1.9

1.5
1.1

0.7

final date
final date

Subduction Zone

SIC

SIC SIC
SIC SIC
SIC
SIC

final date

Figure 10.3: Estimates of minimum homogeneity magnitude with the P0 ,


P1 and P2 procedures. The cases (b) and (f) show that for incomplete catalogues with fewer data (SIC samples), the minimum homogeneity dierences
found with P1 and P2 procedures are small.

2.5
2.0
1.5
1.0
0.5
SIC
SIC
SIC

Subduction Zone
SIC
SIC
SIC
SIC

Guagua Pichincha
SIC

Tungurahua

SIC
SIC

1999 11 08

P1

1999 10 19

P0

1999 09 23

SIC

1999 09 23

Cotopaxi

1999 09 21

1999 09 21

1999 09 19

1999 09 10

1999 09 08

1999 08 28

1999 06 09

SIC

1999 05 18

SIC

1999 04 14

1999 03 12

1999 02 22

1998 12 24

1996 08 14
1997 03 05
1997 09 06
1998 02 22
1999 03 04
1999 12 14
2000 09 04
2001 03 29
2001 06 25
2001 11 24
2001 11 25
2001 11 30
2001 12 11
2001 12 12
2001 12 13
2001 12 14
2001 12 22
2001 12 30
2002 01 10
2002 02 15
2002 03 05
2002 03 25
2002 05 09
2002 07 18
2002 09 01
2002 09 18
2002 11 20
2003 01 28
2003 03 20
2003 05 05
2003 10 05
2004 03 27
2004 12 20
2005 03 31

Pisayambo Zone

1998 10 07

3.0
2.5
2.0
1.5
1.0
0.5
1994 12 24
1995 09 02
1995 10 29
1996 03 13
1996 09 20
1997 01 30
1997 11 07
1998 08 06
1998 12 09
1999 01 09
1999 01 21
1999 02 26
1999 05 03
1999 05 12
1999 05 21
1999 06 07
1999 06 23
1999 07 03
1999 07 31
1999 08 08
1999 08 15
1999 08 21
1999 08 25
1999 08 30
1999 09 03
1999 09 07
1999 09 12
1999 09 16
2000 01 04
2000 04 09
2002 07 11
2003 02 10
2003 11 16
2004 11 19
2005 03 13

Reventador

b
2.5
2.0
1.5
1.0
0.5
0.0

b
2003 03 22
2003 04 19
2003 05 30
2003 06 26
2003 09 16
2003 10 14
2003 11 14
2004 01 07
2004 02 06
2004 03 07
2004 03 31
2004 04 26
2004 05 24
2004 06 25
2004 07 30
2004 09 05
2004 09 22
2004 09 27
2004 10 06
2004 10 21
2004 11 04
2004 11 05
2004 11 06
2004 11 11
2004 11 15
2004 11 19
2004 11 27
2004 12 13
2004 12 31
2005 01 14
2005 02 01
2005 02 15
2005 03 07

b
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0

b
1990 12 15
1992 01 20
1992 09 04
1993 05 08
1994 07 07
1994 11 28
1995 04 13
1995 11 12
1996 03 20
1996 10 13
1997 02 28
1997 08 10
1997 12 11
1998 04 07
1998 09 03
1998 12 31
1999 04 15
1999 10 06
2000 03 22
2000 09 27
2001 02 28
2001 04 18
2001 09 12
2001 12 12
2002 03 17
2002 07 16
2002 11 29
2003 04 23
2003 08 28
2003 12 13
2004 03 31
2004 08 17
2004 11 14
2005 03 23

b
3.5
3.0
2.5
2.0
1.5
1.0
0.5

1993 01 09
1994 01 25
1994 10 29
1995 07 21
1996 04 05
1996 08 05
1996 12 10
1996 12 19
1997 02 01
1997 09 21
1998 02 05
1998 07 01
1998 08 05
1998 12 13
1999 04 17
2000 02 01
2000 05 25
2000 09 29
2000 10 18
2001 03 13
2001 09 14
2002 02 26
2002 04 21
2002 10 02
2003 08 02
2004 05 16
2005 01 05
2005 01 22
2005 01 25
2005 01 25
2005 01 28
2005 01 30
2005 01 31
2005 02 12
2005 02 21
2005 04 13

b
5.5
4.5
3.5
2.5
1.5
0.5

final date (x-axis)

P2

(a)

(b)

(c)

(d)

(e)

(f)

Figure 10.4: Estimates of the b value calculated by using the P0 , P1 and


P2 procedures. The cases (b) and (f) show small dierences in estimates of
the b value found with the P1 and P2 procedures, for data corresponding to
24
SIC samples.

2.4
2.2
2.0
1.8
1.6
1.4
1.2

3.6
3.4
3.2
3.0
2.8
2.6

4.5

4.3

3.7

Tungurahua

Reventador

Pisayambo Zone

Subduction Zone

4.1

3.9
0

30
25
20
15
10
5
0

40

1.05
1.25
1.45
1.65
1.85
2.05
2.25
2.45
2.65
2.85
3.05

1.6

0.40
0.65
0.90
1.15
1.40
1.65
1.90
2.15
2.40
2.65
2.90
3.15

1.8

0.45
0.70
0.95
1.20
1.45
1.70
1.95
2.20
2.45
2.70
2.95

2.0

Guagua Pichincha

2.15
2.40
2.65
2.90
3.15
3.40
3.65
3.90
4.15

2.2

number of events

magnitude

number of events

1996 08 14
1997 09 06
1999 03 04
2000 09 04
2001 06 25
2001 11 25
2001 12 11
2001 12 13
2001 12 22
2002 01 10
2002 03 05
2002 05 09
2002 09 01
2002 11 20
2003 03 20
2003 10 05
2004 12 20

1.7

number of events

1.4
1998 10 07
1998 12 24
1999 02 22
1999 03 12
1999 04 14
1999 05 18
1999 06 09
1999 08 28
1999 09 08
1999 09 10
1999 09 19
1999 09 21
1999 09 21
1999 09 23
1999 09 23
1999 10 19
1999 11 08

magnitude

1.9

number of events

1994 12 24
1995 10 29
1996 09 20
1997 11 07
1998 12 09
1999 01 21
1999 05 03
1999 05 21
1999 06 23
1999 07 31
1999 08 15
1999 08 25
1999 09 03
1999 09 12
2000 01 04
2002 07 11
2003 11 16
2005 03 13

magnitude

2.1

number of events

2003 03 22
2003 05 30
2003 09 16
2003 11 14
2004 02 06
2004 03 31
2004 05 24
2004 07 30
2004 09 22
2004 10 06
2004 11 04
2004 11 06
2004 11 15
2004 11 27
2004 12 31
2005 02 01
2005 03 07

magnitude
2.0
1.8
1.6
1.4
1.2
1.0

Cotopaxi

3.10
3.35
3.60
3.85
4.10
4.35
4.60
4.85
5.10
5.35
5.60
5.85

1990 12 15
1992 09 04
1994 07 07
1995 04 13
1996 03 20
1997 02 28
1997 12 11
1998 09 03
1999 04 15
2000 03 22
2001 02 28
2001 09 12
2002 03 17
2002 11 29
2003 08 28
2004 03 31
2004 11 14

magnitude

2.3

number of events

1993 01 09
1994 10 29
1996 04 05
1996 12 10
1997 02 01
1998 02 05
1998 08 05
1999 04 17
2000 05 25
2000 10 18
2001 09 14
2002 04 21
2003 08 02
2005 01 05
2005 01 25
2005 01 28
2005 01 31
2005 02 21

magnitude

2.5

1.25
1.45
1.65
1.85
2.05
2.25
2.45
2.65
2.85
3.05
3.25

25
20
15
10
5
0

1 2

40

30

40

30

1,2

50

40

30

1,2

3 1,2

2 3

30

(a)

magnitude
Lower border of class (width 0.2)

(b)

20

10

magnitude
Lower border of class (width 0.2)

(c)

20

10

magnitude
Lower border of class (width 0.25)

(d)

magnitude
Lower border of class (width 0.25)

1 2

(e)

20

10

magnitude
Lower border of class (width 0.25)

(f)

20

10

magnitude
Lower border of class (width 0.25)

Figure 10.5: Intervals with 95% condence for population mean and three
smoothed frequency polygons (found from the histograms) of each marked
gray period.
25

Das könnte Ihnen auch gefallen