I statistically discuss the method for estimating the probability of the next characteristic earthquake in a given time interval from the data on date of earthquakes in the past by using lognormal distribution model, based on the Bayesian approach. We assume that
n+1 earthquakes have occurred on a fault or in a source area separated by
n time intervals,
Ti, of which logarithm,
xi=ln(
Ti), are considered to follow a normal distribution,
N(μ, σ
2). Here, μ and σ
2 are mean and variance parameters of a normal population distribution, respectively. Then the likelihood accounted for aseismic period since the last event is defined as
L(μ, σ
2)=(1-
FN(
xp;μ, σ
2))Π
ni=1
fN(
xi;μ, σ
2), where
xp,
FN and
fN are the logarithm of the time interval of aseismic period,
Tp, since the last event, the cumulative distribution function, and the probability density function for
N(μ, σ
2), respectively. We expect that normalized likelihood represents the density distribution of μ, σ
2 based on Bayesian rule with a prior distribution of π(μ, σ
2)=1. If μ and σ
2 are known, the probability of the next earthquake occurring in the forthcoming period from
Tp through
Tp+Δ
T is denoted by a conditional probability,
Pq=(
FN(
xf;, μ, σ
2))/(1-
FN(
xp;μ, σ
2)), where
xf=ln(
Tp+Δ
T). However it is impossible to estimate real values from a small number of data without large error, and we can only know the distribution of parameters to calculate numerically the distribution of
Pq as shown in this paper. The distribution of expected time interval,
T, from the last event to the next event is computed analytically from a law that √
n-3(ln(
T)-
x)/√(
n+1)
s2 follows t-distribution of
n-3 degrees of freedom. Here
x and
s2 are mean and variance of data,
xi=ln(
Ti). And the expected probability of characteristic earthquake occurrence in a time interval Δ
T given that no earthquake has occurred in
Tp, is the mean of variable
Pq or a conditional probability of t-distribution. For the characteristic earthquakes off Miyagi prefecture, Japan, the estimates of probability in the periods of 10, 20 and 30 years after January 2001 are 0.28, 0.62 and 0.80, respectively, using the data set compiled by the Earthquake Research Committee of Japan. These estimates for 20 and 30 years are smaller than those, about 0.8 and larger than 0.9, given by the committee, respectively. Interval estimates of
Pq for those periods are 0.15-0.41, 0.42-0.82 and 0.75-1.0.
View full abstract