Nava Et Al (2017) Gutenberg - Richter B-Value MLE and Sample Size

© All Rights Reserved

43 views

Nava Et Al (2017) Gutenberg - Richter B-Value MLE and Sample Size

© All Rights Reserved

- SAT Level 2 Practice Test 14
- Ex. 1 Data Management
- Using Excel 2007 for Linear Regression
- Measures of Variation
- Bio Stats Excercise
- Anova Post Hoc
- Earthquake Ppt
- Math 1.1 Arithmetic 1-41 to 1-50
- Earthquake Overview
- CN658_659_HB.pdf
- 335
- Internet Prezentacija MN KORISNO
- 1st a Introduction
- Midterm Notes
- 066_2011_Shieh_SDEE_EEW
- 1.12_MRT_Ahmad Luay.docx
- Choy Etal 2010 Sismo 1812
- PSY 2003 Chapter 2 Notes
- 0911.4570
- [4] Probabilistic and Statistic

You are on page 1of 9

DOI 10.1007/s10950-016-9589-1

ORIGINAL ARTICLE

and sample size

F. A. Nava & V. H. Mrquez-Ramrez & F. R. Ziga &

L. vila-Barrientos & C. B. Quinteros

Received: 21 November 2015 / Accepted: 23 May 2016 / Published online: 2 June 2016

# Springer Science+Business Media Dordrecht 2016

Abstract The Aki-Utsu maximum likelihood meth- Keywords Gutenberg-Richter b-value . Maximum

od is widely used for estimation of the Gutenberg- likelihood estimation . Aki/Utsu method

Richter b-value, but not all authors are conscious of

the methods limitations and implicit requirements.

The Aki/Utsu method requires a representative esti- 1 Introduction

mate of the population mean magnitude; a require-

ment seldom satisfied in b-value studies, particularly Possibly the most widely used statistical relation in

in those that use data from small geographic and/or seismology is the Gutenberg-Richter (G-R) distribu-

time windows, such as b-mapping and b-vs-time tion (Ishimoto and Iida 1939; Gutenberg and Richter

studies. Monte Carlo simulation methods are used 1944; Richter 1958), which relates the number of

to determine how large a sample is necessary to earthquakes above a given magnitude, N(M), to the

achieve representativity, particularly for rounded magnitude, M, as

magnitudes. The size of a representative sample

weakly depends on the actual b-value. It is shown log10 N M ab M M 1 ; M M 1; 1

that, for commonly used precisions, small samples

where M1 is a lower threshold below which magnitudes

give meaningless estimations of b. Our results give

are not considered. Parameter a is the logarithm of the

estimates on the probabilities of getting correct esti-

total number of earthquakes with magnitude M M1 and

mates of b for a given desired precision for samples

is a constant that depends on the seismicity rate and the

of different sizes. We submit that all published stud-

length of the observation time. Parameter b, widely

ies reporting b-value estimations should include in-

referred to as the b-value, hereafter referred to as b, is

formation about the size of the samples used.

a constant that relates the relative numbers of small to

large magnitudes. Typically, b 1, and M1 is the com-

F. A. Nava (*) : L. vila-Barrientos : C. B. Quinteros pleteness magnitude below which log10 N(M) ceases to

Sismologa, Centro de investigacin Cientifica y de Educacin behave linearly due to insufficient coverage (e.g.,

Superior de Ensenada, Ensenada, B.C., Mexico Wiemer and Wyss 2000).

e-mail: fnava@cicese.mx

Usually M 1 is estimated from the observed

V. H. Mrquez-Ramrez : F. R. Ziga log10 N(M) or from the noncumulative log10 n(M)

Centro de Geociencias, Universidad Nacional Autnoma de histograms, which requires an adequate magnitude

Mxico, Juriquilla, Qro., Mexico sample. However, this is not always the case: Some-

V. H. Mrquez-Ramrez times, M1 may be estimated for a given data set and

SisVoc, CUC, Universidad de Guadalajara, Puerto Vallarta, Jal., then used for a different dataset; not uncommon is

Mexico the use of the completeness magnitude appropriate

128 J Seismol (2017) 21:127135

for the complete data set of a given seismic network M1 M/2 and, instead of Eq. (3), Utsus (1965)

or array in studies involving subsets of the data. formula

Since the G-R relation gives estimates for the

occurrence ratio of earthquakes with magnitudes log10 e

b 4

above a given value, b has been widely used for M M 1 M =2

seismic hazard and risk estimations (e.g., Bender

1983). An inverse relationship between b and the is applicable.

stress level in a given region has been proposed Once M1 has been estimated, Eqs. (3) or (4) have been

(e.g., Scholz 1968; Ghosh et al. 2008), and several used as a simple and straightforward way of estimating b,

studies have found possibly premonitory behavior in which is less cumbersome than that of fitting a straight

the values of b before large earthquakes (e.g., Shaw line to the magnitude histogram. We will refer to this

et al. 1992; Wyss and Wiemer 2000; Enescu and Ito approach to b-value estimation as the Aki-Utsu method.

2001; Mrquez-Ramirez 2012). Regional variations However, when using the Aki-Utsu method, it is

of b have been used to identify zones susceptible to often not taken into account that b and M in the

large ruptures (e.g., Wiemer and Wyss 1997; Ziga abovementioned formulae are population parameters;

and Wyss 2001; Montuori et al. 2010). Since Aki the actual formula used for estimation is

(1981) proposed a relation between b and the fractal

dimension of fault planes, b has been emplyed in ^b log10 e

5

seismic fractality studies (e.g., Singh et al. 2009). b

M M 1 M =2

The many applications of b in seismology, partic-

ularly on the field of seismic hazard estimation, where ^b and Mb are estimates, and ^b will agree with b

indicate the necessity of estimating b correctly in b is representative of, i.e., reasonably close to,

order to obtain meaningful and reliable results. In only if M

what follows, we will address the problem of sam- the true M . This, in turn, means that M b should be

pling and the reliability associated with it. calculated from an adequate sample, i.e., one that in-

cludes enough data, but in many cases, it is estimated

from too small samples, either because data are scarce or

because no attention is paid to the matter of

2 The problem

representativity.

An example of studies where this problem may ap-

Equation (1) implies that magnitudes are distributed

pear is b-mapping, where M1 is estimated for a given

exponentially as

data set in some regin, and then the region is

subdivided in volume elements or cells, and some of

f M eM M 1 ; M M 1; 2 these cells may not include enough data to yield a

reliable estimate (e.g., Khan 2005; Bhattacharya et al.

where = b ln(10) (c.f. Epstein and Lomnitz 1966; 2002; Singh and Singh 2015). Another example is the

Lomnitz 1966, 1974). study of changes in b vs time, where M b is determined

Based on the premise of this magnitude distribution, from the data in time windows, and there is a tradeoff

Aki (1965) showed that the maximum likelihood esti- between the length and/or number of data in the time

mate of b is window and the resolution in time. In these and other

cases, not only data may not be enough, but also the

log10 e overall M1 may not be appropriate for all data subsets.

b 3

M M 1 The question arises: How many data are enough for

b

M to be representative and b^ to be reliable? The answer,

where M is the observed mean magnitude. of course, depends on the desired precision of ^b; so we

In practice, magnitudes are rounded to some will rephrase the question: What is the minimum num-

M, usually M = 0.1, and when magnitudes are ber of data required in a sample, to achieve a desired

rounded, the actual minimum magnitude is precision for ^b with a given probability?

J Seismol (2017) 21:127135 129

This question has been addressed by Shi and Bolt considering magnitudes rounded to M. r is a uniformly

(1982), hereafter referred to as SB, who give an analyt- distributed pseudo-random number in the (0,1) interval,

ical estimate of the ^

b distribution for unrounded magni- generated using the Matlab rand function with the

tudes, as a function of the number of samples N, for Twister generator, discarding magnitudes larger than

stationary b, and for N large enough for Mb to approach a an arbitrary upper limit M2q = 9.2 set to avoid unaccept-

ably large (potentially infinite) magnitudes; since very

normal distribution; the ^b distribution is large magnitudes are very scarce, this limit does not

2 !2 3

rN b b

significantly affect results.

g ^

b exp4N 1 =25 6 We also tried the doubly truncated Gutenberg-

2 ^ 2

^

b Richter distribution, as

How well does Eq. (6) agree with ^b determinations M ln eM 1q r eM 1q eM 2q =;

for samples with small N?

In what follows, we use Monte Carlo methods to but, since both manitude generating schemes yielded

simulate actual sampling in order to explore the distri- essentially the same results, because their outputs differ

bution of ^b for samples of different sizes and different only for very large magnitudes close to M2, and since the

true values of b. Our results let us determine the range of Aki-Utsu estimates assume a singly truncated exponen-

applicability of Eq. (6), explore the effects of small N tial distribution, we present here only the results obtain-

and of magnitude rounding, and establish some quanti- ed from Eq. (7).

tative guidelines for sampling. Figure 1 shows results from 250,000 realizations,

each, of ^b determinations, for a true b = 1.0, from sam-

ples having from 10 to 100,000 data. The thick (blue)

3 Monte Carlo estimation of sampling line shows the mean of all estimated ^b values for each

sample size, and the thin (red) lines show the mean 1

We simulate magnitude sampling, in the (M1, M2) mag- standard deviation (continuous) and 2 standard devia-

nitude range from the Gutenberg-Richter distribution as tions (dashed).

M M 1q lnr=; 7 The graph on the left hand side is for unrounded

magnitudes and shows quantitatively exactly what

where M1q and M2q equal M1 and M2, respectively, would be expected qualitatively; the graph on the

when considering unrounded magnitudes, and equal right side is for magnitudes rounded to the custom-

M 1 M/2 and M 2 + M/2, respectively, when ary 0.1. Both graphs show that estimates from small

size N.Unrounded magnitudes

(left). Magnitudes rounded to 0.1

(right). The thick, continuous

(blue) line is the mean of 250,000

realizations for each of various

sample sizes N samples, and the

thin (red) lines indicate mean

value 1 standard deviation

(continuous) and 2 standard

deviations (dashed)

130 J Seismol (2017) 21:127135

samples, besides being probably overestimated, are To illustrate how individual estimates can be expect-

really meaningless (i.e., completely unreliable) be- ed to behave, we obtained ^b histograms for some repre-

cause standard deviations are extremely large. The sentative sample sizes, true b-values, and desired preci-

estimate averages over a large number of realiza- sions. All histograms were obtained from sets of

tions are overestimated, because small magnitudes 500,000 realizations for each data set length, and the

are much more numerous than large magnitudes. corresponding SB analytical distributions are plotted

Thus, any one ^ b estimation from less than, say, over the histograms for comparison.

300 data can take values over a range of several Figure 2 shows, on the left, the histograms (blue,

tenths of a unit with large probability. continuous lines) with classes ^b 0:010 wide, for ^b

While, for large samples, the mean estimate from estimated from unrounded magnitudes distributed ac-

unrounded magnitudes converges to the true value, for cording to Eq. (2) for b = 1.00 and different sample

rounded magnitudes, large samples lead to an underes- sizes; it also shows the corresponding SB analytical

timation of it, unless a correction for rounding is applied distributions (red, dashed lines). There is extremely

(see, for example, Mrquez-Ramrez et al. 2015), be- good agreement between histogram and distribution

cause the magnitude value at the center of each class is for samples with N 100, while for samples below

not an unbiased representative of the magnitudes in the N 50, the agreement is not so good, because N is not

class. This is a minor effect that will be ignored in what large enough to fulfill the SB assumptions, but it is not

follows. too bad although the modal values do not coincide. As

Fig. 2 Synthetic data histograms for 500,000 realizations for histograms,while the dashed (red) lines are the SB distribution.

b = 1.0 and classes ^

b 0:010 wide. Graph on the left is for The vertical solid line indicates the true b-value, and the vertical

unrounded magnitudes, and on the right for magnitudes rounded dashed line indicates the observed mean value

to M = 0.1. The thick continuous (blue) lines are the

J Seismol (2017) 21:127135 131

Table 1 Probabilities of obtaining the correct value (Pr), an underestimated value (Pr), and an overestimated value (Pr+) for true b = 1.0,

precisions ^

b 0:010, 0.025, and 0.050, and different sample sizes N

b = 1.0

^b

10 1.103 0.385 0.462 0.000 0.538 0.462 0.029 0.508 0.434 0.058 0.508

25 1.037 0.214 0.474 0.018 0.508 0.455 0.055 0.489 0.437 0.092 0.470

50 1.015 0.145 0.478 0.026 0.496 0.464 0.066 0.467 0.426 0.144 0.430

100 1.005 0.101 0.485 0.037 0.478 0.457 0.092 0.451 0.403 0.201 0.395

200 1.001 0.071 0.483 0.060 0.457 0.445 0.138 0.418 0.369 0.282 0.349

500 0.998 0.044 0.489 0.087 0.423 0.420 0.221 0.359 0.313 0.424 0.263

1000 0.997 0.031 0.489 0.124 0.387 0.394 0.306 0.300 0.248 0.570 0.182

2000 0.996 0.022 0.486 0.175 0.339 0.354 0.419 0.226 0.170 0.731 0.098

5000 0.996 0.014 0.479 0.266 0.256 0.277 0.605 0.118 0.066 0.913 0.020

10,000 0.996 0.010 0.473 0.353 0.174 0.202 0.752 0.046 0.018 0.981 0.002

b and its standard deviation s

Fig. 3 Synthetic data histograms for 500,000 realizations for lines are the histograms, while the dashed (red) lines are the SB

magnitudes rounded to M = 0.1, b = 1.0, and classes ^ b 0:0 distribution. The vertical solid line indicates the true b-value, and

25 (left) and ^b 0:50 (right) wide. The thick continuous (blue) the vertical dashed line indicates the observed mean value

132 J Seismol (2017) 21:127135

Fig. 4 Synthetic data histograms for 500,000 realizations for for are the histograms, while the dashed (red) lines are the SB

magnitudes rounded to M = 0.1 classes ^ b 0:50 wide, and distribution. The vertical solid line indicates the true b-value, and the

b = 0.8 (left) and b = 1.2 (right). The thick continuous (blue) lines vertical dashed line indicates the observed mean value

mentioned before, the mean ^b values significantly over- realizations, so they should not be interpreted as figures

estimate the true b-values for samples below N 100. arrived to analytically, but as probable values that, how-

However, as shown on the right in Fig. 2, the situa- ever, do not change significantly between different

realizations.

tion is quite different for ^b estimated from the same

From this table, it can be seen that the probabilities of

magnitudes as in the previous case, rounded now, as is

usual in practice, to one decimal place. The histograms getting a correct estimate of b = 1.00 with ^b 0:010

show gaps due to the fact that means from only a few precision are extremely low (<0.1) for samples of up to

rounded magnitudes cannot take all values over the 500 magnitudes and are low (<0.5) for samples as large

specified magnitude range; hence, possible magnitude as 10,000. Thus, results from studies claiming to

distiguish changes in b of 0.01 from less than about

means result in peaks for the corresponding ^b estimates.

20,000 events should be taken with a (large) grain of salt.

This appearance of preferred values is reduced as sam-

Of course, reducing the desired precision increases

ples get larger, but samples larger than 200 are needed

the probability of obtaining a correct, albeit less precise,

before the histogram resembles the SB distribution.

result. Figure 3 presents histograms for the same round-

Table 1 lists, for each N, the histogram mean and

standard deviation values, as well as the probabilities ed magnitude data set for ^b 0:025 (left) and ^b

0:050 (right), and the corresponding probabilities are

Pr ^b b , Pr ^ b < b , and Pr ^b > b , denoted by listed in Table 1. For 0.025 precision, a reasonably good

Pr, Pr, and Pr+, respectively, for each given ^b; these agreement between histogram and the SB distribution is

probabilities are obtained from the Monte Carlo obtained for N 100, and estimating the correct b attains

J Seismol (2017) 21:127135 133

b = 0.8

^b

10 0.885 0.310 0.457 0.024 0.519 0.434 0.046 0.519 0.412 0.092 0.496

25 0.831 0.172 0.459 0.029 0.511 0.445 0.058 0.497 0.416 0.117 0.467

50 0.814 0.117 0.469 0.042 0.489 0.448 0.084 0.468 0.398 0.176 0.426

100 0.806 0.081 0.471 0.052 0.477 0.435 0.125 0.440 0.372 0.247 0.381

200 0.802 0.057 0.470 0.069 0.461 0.414 0.177 0.409 0.331 0.340 0.329

500 0.799 0.036 0.463 0.112 0.425 0.380 0.273 0.348 0.250 0.518 0.148

1000 0.799 0.025 0.451 0.157 0.391 0.335 0.378 0.286 0.174 0.678 0.148

2000 0.798 0.018 0.433 0.220 0.346 0.276 0.515 0.209 0.093 0.838 0.068

5000 0.798 0.011 0.395 0.338 0.267 0.174 0.726 0.101 0.019 0.972 0.009

10,000 0.798 0.008 0.357 0.456 0.187 0.093 0.873 0.035 0.002 0.998 0.000

an even chance from samples of about 4000 events. For As can be seen from the histograms and the tables,

0.05 precision, 0.5 probability is attained for less than for small samples, correct estimation of b is easier and

1000 events, and there is a high probability of a correct the fit of the histograms to the SB distribution is better,

estimation from about 2000 events on. for small than for large b-values.

Since the magnitude population depends on the b- For small samples in all cases, the mean and the

value, results are slightly different for different values of mode in the histograms tell different stories: If a mea-

b; for considerations of space, we will only show here surement is repeated many (thousands of) times and b is

the histograms for ^b 0:050, for the representative estimated from the mean, the result would be

values b = 0.80 and b = 1.20 (Fig. 4), and list the overvalued; on the other hand, any single measurement

resulting probabilities in Tables 2 and 3, respectively. will probably result in an overvalued estimation.

b = 1.2

^b

10 1.321 0.458 0.456 0.000 0.544 0.456 0.035 0.509 0.456 0.035 0.509

25 1.241 0.255 0.474 0.022 0.504 0.474 0.022 0.504 0.452 0.066 0.482

50 1.216 0.174 0.492 0.016 0.492 0.476 0.048 0.476 0.445 0.110 0.445

100 1.204 0.120 0.496 0.033 0.471 0.473 0.078 0.449 0.430 0.165 0.405

200 1.198 0.084 0.496 0.048 0.454 0.467 0.118 0.415 0.405 0.233 0.362

500 1.195 0.053 0.513 0.075 0.412 0.454 0.187 0.359 0.363 0.358 0.279

1000 1.194 0.037 0.522 0.104 0.374 0.442 0.258 0.300 0.313 0.487 0.199

2000 1.193 0.026 0.536 0.143 0.321 0.422 0.350 0.230 0.249 0.636 0.115

5000 1.193 0.017 0.556 0.213 0.231 0.380 0.501 0.120 0.144 0.827 0.028

10,000 1.193 0.012 0.584 0.269 0.147 0.335 0.618 0.047 0.067 0.930 0.003

134 J Seismol (2017) 21:127135

Union, Washington, p 566574

Bender B (1983) Maximum likelihood estimation of b values for

In the results presented above, no correction was made magnitude grouped data. Bull Seism Soc Am 73:831851

for the fact that the mean from rounded magnitudes is Bhattacharya P, Majumdar R, Kayal J (2002) Fractal dimension and

not equal to the Btrue^ mean from unrounded magni- b-value mapping in northeast India. Curr Sci 82:14861491

tudes, because the magnitude representative of a given Enescu B, Ito K (2001) Some premonitory phenomena of the 1995

Hyogo-Ken Nanbu (Kobe) earthquake: seismicity, b-value

class is not an unbiased estimator of the magnitudes and fractal dimension. Tectonophysics 338:297314

contained in the class (e.g., Bender 1983; Tinti and Epstein B, Lomnitz C (1966) A model for the occurrence of large

Mulargia 1987; Mrquez et al. 2015), nor were mean earthquakes. Nature 211:954956

magnitude estimates corrected for large magnitude sam- Ghosh A, Newman A, Amanda M, Thomas A, Farmer G (2008)

Interface locking along the subduction megathrust from b-

pling. However, such corrections are small and would

value mapping near Nicoya Peninsula, Costa Rica. Geophys

not significantly change the results. Res Lett 35:L01301. doi:10.1029/2007GL031617

From our results, it is clear that results from papers Gutenberg B, Richter C (1944) Frequency of earthquakes in

dealing with b estimates from less than 12,000, 2000, California. Bull Seismol Soc Am 34:185188

and 500 magnitude data for precisions of 0.01, 0.025, Ishimoto M, Iida K (1939) Observations sur les seisms enregistrs

par le microseismograph construit dernirement (I). Bull

and 0.50 magnitude units, respectively, are valueless. Earthquake Res Inst Univ of Tokyo 17:443478

Some papers show results from 10 data, but most do Kamer Y, Hiemer S (2015) Data-driven spatial b-value estimation

not even report the amount of data used, so their results with applications to California seismicity: to b or not to b.

are not reliable; we will not reference here examples of doi: 10.1002/2014JB011510.

Khan P (2005) Mapping of b-value beneath the Shillong plateau.

these useless papers but instead recommend a strong

GondwanaRes8:271276.doi:10.1016/S1342-937X(05)71126-6

caveat to the readers of these papers. Kijko A (1988) Maximum likelihood estimation of Gutenberg-

Our analysis has considered complete and correct Richter b parameter for uncertain magnitude values.

rounded magnitudes; dealing in practice with incom- PAGEOPH 127:573579

plete data sets and uncertain magnitude estimations Kijko A, Selevoll M (1989) Estimation of earthquake hazard

parameters from incomplete data files. Part I. Utilization of

(Kijko 1988; Kijko and Selevoll 1989, 1992) would extreme and complete catalogs with different threshold mag-

require even larger samples. nitudes. Bull Seism Soc Am 79:645654

It is also clear that papers reporting b estimates Kijko A, Selevoll M (1992) Estimation of earthquake hazard

should always report the size of the sample(s) used for parameters from incomplete data files. Part II. Incorporation

of magnitude heterogeneity. Bull Seism Soc Am 82:120134

estimation, in order that the reader may know how

Lomnitz C (1966) Statistical prediction of earthquakes. Rev

reliable the measurements are and how likely are results Geophys 4:377393

from small samples to be overvalued. Lomnitz C (1974) Global tectonics and earthquake risk. Elsevier

For spatiotemporal variations in b to be significant, Sc. Pub. Co., CH.

whatever the method used for mapping (e.g., Wiemer Mrquez-Ramrez V. (2012) Anlisis multifractal de la

distribucin espacial de sismicidad y su posible aplicacin

and Wyss 2002; Kamer and Hiemer 2015), each b map premonitora. Exploracin de un posible mecanismo para la

should have a sister N map. fractalidad mediante modelado semiestocstico. PhD Thesis,

Programa de Posgrado en Ciencias de la Tierra, Centro de

Investigacin Cientfica y de Educacin Superior de

Acknowledgments We sincerely thank two anonymouis re-

Ensenada, Baja California, Mxico.

viewers for useful comments and suggestions. Thanks to Dr. J.L.

Mrquez-Ramrez V, Nava F, Ziga F (2015) Correcting the

Brioso for his guidance and patience. Many thanks to Jos Mojarro

Gutenberg-Richter b-value for effects of rounding and noise.

for technical support.

Earthq Sci 28:129134. doi:10.1007/s11589-015-0116-1

Montuori C, Falcone G, Murru M, Thurber C, Reyners M,

Eberhart-Phillips D (2010) Crustal heterogeneity highlighted

by spatial b-value map in the Wellington region of New

References Zealand. Geophys J Int 183:451460. doi:10.1111/j.1365-

246X.2010.04750.x

Richter C (1958) Elementary seismology. W H Freeman and Co, USA

Aki K (1965) Maximum likelihood estimate of b in the formula Scholz C (1968) The frequency-magnitude relation of

log(N) = a - bM and its confidence limits. Bull Earthq Res microfracturing in rock and its relation to earthquakes. Bull

Inst Tokio Univ 43:237239 Seismol Soc Am 58:399415

Aki K (1981) A probabilistic synthesis of precursory phenomena Shaw E, Carlson J, Langer J (1992) Patterns of seismic activity

In: Simpson DW Richards PG (eds) Earthquake prediction: preceding large earthquakes. J Geophys Res 97(B1):479488

J Seismol (2017) 21:127135 135

Shi Y, Bolt B (1982) The stand ard error of th e Wiemer S, Wyss M (1997) Mapping the frequency-magnitude

magnitude-frequency b value. Bull Seismol Soc Am distribution in asperities: an improved technique to calculate

72:16771687 recurrence times? J Geophys Res 102(B7):1511515128

Singh C, Singh S (2015) Imaging b-value variation beneath the Wiemer S, Wyss M (2000) Minimum magnitude of completeness

Pamir-Hindu Kush region. Bull Seismol Soc Am 105:808 in earthquake catalogs: examples from Alaska, the Western

815. doi:10.1785/0120140112 United States, and Japan. Bull Seismol Soc Am 90:859869

Singh C, Singh A, Chadha R (2009) Fractal and b-value mapping Wiemer S, Wyss M (2002) Mapping spatial variability of the

in Eastern Himalaya and Southern Tibet. Bull Seismol Soc frequency-magnitude distribution of earthquakes. Adv

Am 99:35293533. doi:10.1785/0120090041 Geophys 45:259302

Tinti S, Mulargia F (1987) Confidence intervals of b-values Wyss M, Wiemer S (2000) Change in the probability for earth-

for grouped magnitudes. Bull Seismol Soc Am 77: quakes in southern California due to the Landers magnitude

21252134 7.3 earthquake. Science 290:1334

Utsu T (1965) A method for determining the value of b in a formula Ziga R, Wyss M (2001) Most-and least-Likely locations of large

329 log n = a - bM showing the magnitude-frequency relation to great earthquakes along the Pacific coast of Mexico esti-

for 330 earthquakes. Geophys Bull Hokkaido Univ 13: mated from local recurrence times based on b-values. Bull

99103 Seismol Soc Am 91:17171728

- SAT Level 2 Practice Test 14Uploaded byLeonardo Florenzi
- Ex. 1 Data ManagementUploaded byArriane T. Bartolome
- Using Excel 2007 for Linear RegressionUploaded byGhe Patani
- Measures of VariationUploaded byjovanializa
- Bio Stats ExcerciseUploaded byAnshuli Trivedi
- Anova Post HocUploaded byDini Hanifa
- Earthquake PptUploaded byHarshada Hikare
- Math 1.1 Arithmetic 1-41 to 1-50Uploaded bysiddique
- Earthquake OverviewUploaded byShakib Aman Sohan
- CN658_659_HB.pdfUploaded byHelmiPerdanaKusuma
- 335Uploaded byVijay Kumar
- Internet Prezentacija MN KORISNOUploaded byevgenija89
- 1st a IntroductionUploaded bysampritc
- Midterm NotesUploaded byIrena Feng
- 066_2011_Shieh_SDEE_EEWUploaded byTara Gonzales
- 1.12_MRT_Ahmad Luay.docxUploaded byAhmad Luay Adnani
- Choy Etal 2010 Sismo 1812Uploaded bymanu_dark07
- PSY 2003 Chapter 2 NotesUploaded byGabe Marquez
- 0911.4570Uploaded byBrandon Mcguire
- [4] Probabilistic and StatisticUploaded byanon_964762231
- Quiz - Week 1Uploaded byRipul Nabi
- Visual Data ActivityUploaded byAnonymous G2io2aXQX
- 001-021 Interrelations Between Levels of Aspiration, Performance, And Estimates of Past Performance.Uploaded byancuta.
- NormalDistribution ExamplesUploaded byDeniz Aras
- CrossValidationReport-residual Buat ClassedUploaded byPuspita Ayuningthyas
- Lab 6 HW Q6.17Uploaded byAdam Kuzminsky
- s11stat200mtA.pdfUploaded byJamie Samuel
- 3 17 08 WhitehillUploaded byAfifah Solehah Hamdan
- Ch2 4 ProblemsUploaded byEhab hoba
- handouts.docxUploaded byHoney Jane Cantillas Lumactod

- An Introduction to ROC AnalysisUploaded byTrần Anh Tú
- Appendix E CASF AlgorithmUploaded bySebastian Astorquiza Trucco
- Mendoza et al. 2010Uploaded byarmandosano
- Do all birds tweet the same.pdfUploaded bySebastian Astorquiza Trucco
- Information credibility on Twitter.pdfUploaded bySebastian Astorquiza Trucco
- 07-05-21+22_ NTNU-Visit_Erlangen_IS-CTO_V4Uploaded bySebastian Astorquiza Trucco
- 04 Sintef PolskaUploaded bySebastian Astorquiza Trucco
- Carter Kim South p256Uploaded bySebastian Astorquiza Trucco
- Revey 2013TAC Workshop_Blast Design.pdfUploaded bySebastian Astorquiza Trucco
- v103n04p233Uploaded byXiaoyu Bai
- Cohesive DevicesUploaded bySebastian Astorquiza Trucco
- Innovation Solutions for Seawater Use in Mining OperationsUploaded bySebastian Astorquiza Trucco
- Evolutionary Algorithms for the Optimization of Production Planningin Uriberground MinesUploaded bySebastian Astorquiza Trucco
- The Influence of Conditional Bias in Optimum Ultimate Pit PlanningUploaded bySebastian Astorquiza Trucco
- Strategies of Investing Into Mining Industry Under Uncertainties of Multidimensional AlternativesUploaded bySebastian Astorquiza Trucco
- Trona Dura Six SigmaUploaded bySebastian Astorquiza Trucco
- Hanks and Kanamori 1979_A Moment Magnitude ScaleUploaded bySebastian Astorquiza Trucco
- Least squares estimation.pdfUploaded bySebastian Astorquiza Trucco
- Vallejos et al (2012)_Seismicity indicators.pdfUploaded bySebastian Astorquiza Trucco
- Woessner y Wiemer (2005)_Assessing the Quality of Earthquake Catalogues Estimating the Magnitude of Completeness and Its UncertaintyUploaded bySebastian Astorquiza Trucco
- Reasenberg and Jones (1994) - Earthquake Aftershocks, UpdateUploaded bySebastian Astorquiza Trucco
- Nava Et Al (2017) Gutenberg - Richter B-Value MLE and Sample SizeUploaded bySebastian Astorquiza Trucco
- Vallejos and McKinnon (2009)_Scaling Laws and Their Implications for Re-Entry Protocol DevelopmentUploaded bySebastian Astorquiza Trucco
- Tutorial on Maximum Likelihood EstimationUploaded byapi-3861635
- Blasting design for obtaining desired fragmentation (Stjepan Strelec, Mario Gazdek, Josip Mesec).pdfUploaded bySebastian Astorquiza Trucco

- Aristotle and the Morality of SelfUploaded byCtorres910
- Dance Swansong Evaluation - UNIT 49Uploaded byLibby Archer
- production formalinUploaded byVirginia Rosales Olmos
- Cause and Effect ExamplesUploaded byZaki Azman
- Marketing ManagerUploaded byapi-122238313
- IMSL Fortran Library User Guide 5.pdfUploaded byAshoka Vanjare
- Zanin, 2017 FSUploaded byDeni Nava
- HP-AP420Uploaded byChristi Walker
- INTRODUCTION - Architecture of Alcohol Addiction Recovery CentresUploaded bySaurabh Singh
- DAA_Brochure tifr.pdfUploaded byNilasha Ghosh
- Market StructureUploaded byKhel Dollison
- PLC1Uploaded byAnonymous DKv8vp
- The Design of Everyday Things by Donald a. NormanUploaded byBoni Aditya
- simulink-kalman-filter.pdfUploaded byjlissa5262
- VedicReport11-25-20143-31-48AMUploaded bySamrat Chowdhury
- Online AuctionUploaded byCharan Arjun
- Practical Exam QuestionsUploaded byReSolumers
- Writing Tips for CPEUploaded bydaniele_catalano
- 3a_AssertionsOverviewUploaded bySatyanarayana Arvapalli
- Program ManualUploaded byangelo2626
- Finite Difference Method for Nonlocal Singularly Perturbed ProblemUploaded byIjmret Journal
- UPONOR MesseUploaded bybatazivo
- Mobile Operating SystemsUploaded byKochcherige Darsha Vishvanath
- Cintra20.3G DSDUploaded bynoskov
- Guidelines for Minor ProjectUploaded byShubham Khera
- Work Preference InventoryUploaded byDany Adex
- Photocatalytic Degradation of Methylene Blue Using TiO2 Impregnated DiatomiteUploaded bychemeng
- LOGIC (Syllabus) FinalUploaded byEarl John Pajaro
- R F Survey SitesUploaded byRam Yadav
- EE264 Introduction to Digital Design Project-2Uploaded byTimsinaKrishna