Abstract
The quality of earthquake prediction is usually characterized by a two-dimensional diagram n versus τ, where n is the rate of failures-to-predict and τ is a characteristic of space–time alarm. Unlike the time prediction case, the quantity τ is not defined uniquely. We start from the case in which τ is a vector with components related to the local alarm times and find a simple structure of the space–time diagram in terms of local time diagrams. This key result is used to analyze the usual 2-d error sets {n, τ w } in which τ w is a weighted mean of the τ components and w is the weight vector. We suggest a simple algorithm to find the (n, τ w ) representation of all random guess strategies, the set D, and prove that there exists the unique case of w when D degenerates to the diagonal n + τ w = 1. We find also a confidence zone of D on the (n, τ w ) plane when the local target rates are known roughly. These facts are important for correct interpretation of (n, τ w ) diagrams when we discuss the prediction capability of the data or prediction methods.
Similar content being viewed by others
References
Bolshev, L. N. and Smirnov, V. N. Tables of Mathematical Statistics (Nauka, Moscow 1983).
Harte, D., and Vere-Jones, D. (2005), The entropy score and its uses in earthquake forecasting, Pure Appl. Geophys. 162, 1229–1253.
Hanssen, A. W., and Kuiper, W. J. A. (1965), On the relationship between the frequency of rain and various meteorological parameters, Modedeelingen en Verhandelingen, Royal Notherlands Meteorological Institute, 81.
Jolliffe, I. T. and Stephenson, D. B. (eds.), Forecast Verification: A Practitioner’s Guide in Atmospheric Science (John Wiley & Sons, Hoboken 2003).
Kagan, Y. Y. (2007), On earthquake predictability measurement: information score and error diagram, Pure Appl. Geophys. 164, 1947–1962.
Keilis-Borok, V. I. and Soloviev, A. A. (eds.), Nonlinear Dynamics of the Lithosphere and Earthquake Prediction (Springer-Verlag, Berlin-Heidelberg 2003).
Kossobokov, V. G. (2005), Earthquake prediction: principles, implementation, Perspect. Comput. Seismol. 36-1, 3–175, (GEOS, Moscow).
Lehmann, E. L., Testing Statistical Hypotheses (J. Wiley & Sons, New York 1959).
Marzocchi, W., Sandri, L., and Boschi, E. (2003), On the validation of earthquake-forecasting models: The case of pattern recognition algorithms, Bull. Seismol. Soc. Am. 93, 5, 1994–2004.
Molchan, G. M. (1990), Strategies in strong earthquake prediction, Phys. Earth Planet. Inter. 61(1–2), 84–98.
Molchan, G. M. (1991), Structure of optimal strategies of earthquake prediction, Tectonophysics 193, 267–276.
Molchan, G. M. (1997), Earthquake prediction as a decision making problem, Pure Appl. Geophys. 149, 233–247.
Molchan, G. M., Earthquake prediction strategies: A theoretical analysis. In Nonlinear Dynamics of the Lithosphere and Earthquake Prediction (eds. Keilis-Borok, V.I. and Soloviev, A.A.) (Springer-Verlag, Berlin-Heidelberg 2003), pp. 209–237.
Molchan, G. M., and Kagan, Y. Y. (1992), Earthquake prediction and its optimization, J. Geophys. Res. 97, 4823–4838.
Molchan, G. M. and Keilis-Borok, V. I. (2008), Earthquake prediction: Probabilistic aspect, Geophys. J. Int. 173, 1012–1017.
Shcherbakov, R., Turcotte, D. L., Holliday, J. R., Tiampo, K. F., and Rundle, J. B. (2007), A Method for forecasting the locations of future large earthquakes: An analysis and verification, AGU, Fall meeting 2007, abstract #S31D-03.
Shen, Z.-K., Jackson, D. D., and Kagan, Y. Y. (2007), Implications of geodetic strain rate for future earthquakes, with a five-year forecast of M 5 Earthquakes in Southern California, Seismol. Res. Lett. 78(1), 116–120.
Swets, J. A. (1973), The relative operating characteristic in psychology, Science 182, 4116, 990–1000.
Tiampo, K. F., Rundle, J. B., McGinnis, S., Gross, S., and Klein, W. (2002), Mean field threshold systems and phase dynamics: An application to earthquake fault systems, Europhys. Lett. 60(3), 481–487.
Zechar, J.D. and Jordan, Th., H. (2008), Testing alarm-based earthquake predictions, Geophys. J. Int. 172, 715–724.
Acknowledgements
This work was supported by the Russian Foundation for Basic Research through grant 08-05-00215. I thank D.L. Turcotte for useful discussions, which have stimulated the writing of the present paper.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
We are going to prove the statements 3.1–3.10.
Proof for 3.1, 3.2
Obviously, the projection γ w preserves the property of convexity. Therefore, \({{\mathcal{E}}}_w\) and D w are convex at the same time as are \({{\mathcal{E}}}\) and D. If D w degenerates to the diagonal \(\tilde{D}:n+\tau_w=1,\) then the simplex D is given by any of the two equations: \(n+\sum^k_{i=1} w_i\tau_i = 1\) and \(n+\sum^k_{i=1}\lambda_i \tau_i/\lambda =1.\)Hence w i = λ i /λ.
Proof of 3.3.
The simplex D is the convex hull of \((n,{\varvec{\tau}})\) points of the form \(Q(\varepsilon )=(1-\sum \lambda_i \varepsilon_i/\lambda , \varepsilon_1, \ldots , \varepsilon_k),\) where ɛ i = 0, 1. Accordingly, D w is the convex hull of the Q w (ɛ), see (12).
Proof of 3.4.
This statement follows intuitively from dimensionality considerations: The k-dimensional surface \(n({\varvec{\tau}})\) with k > 1 is projected onto the (n, τ w ) plane, hence its image cannot be single-dimensional in the generic case.
In order to prove (13), we note that a convex function on the simplex \(S_n=\{\sum^k_{i=1}\tau_i w_i=u, 0\le \tau_i \le 1\}\) reaches its maximum at one of the edges, specifically, at a point of the form
The use of (10) gives (13).
Suppose the upper and lower boundaries of the image of \(n({\varvec{\tau}})\) are identical and the {n i (τ)} are piecewise smooth functions. Consider all \({\varvec{\tau}} =(\tau_1, \ldots ,\tau_k)\) for which
where τ w is fixed.
Varying, e.g., τ 1 and τ 2, we have after differentiation:
If τ 1, τ 2 are points of smoothness of n i (τ), i = 1, 2, then repeated differentiation of (A1) will give
However, n i ′′(τ i )≥ 0, i = 1, 2. Hence n i ′′(τ i ) = 0, i.e., n i (τ) are locally linear at all points of smoothness. Since n i (τ) are piecewise smooth, it follows that for any discontinuous point τ 1 of n 1(·) one can find a point τ 2 where n 2(·) will be smooth. Consequently, when n 1 is discontinuous at τ, one should replace n 1′(τ 1) with n 1′(τ 1 + 0) and n 1′(τ 1 − 0) in equation (A1). But then we have from (A1) that n 1′(τ) is continuous at τ 1; hence all n i (τ) are linear. Taking the boundary conditions n i (0) = 1 and n i (1) = 0 into account, we have n i (τ) = 1 − τ. However, in that case one has \({{\mathcal{E}}}=D,\) and, in virtue of 3.2, w i = λ i /λ.
Proof of 3.5.
Let Q w be the point where the convex set {ψ ≤ c} is tangent to the convex curve n(τ w ). The function ψ reaches its minimum at the point Q w on \({{\mathcal{E}}}_w,\) because the sets {ψ ≤c} increase with increasing c. Since \(Q_w \in {{\mathcal{E}}}_w,\) the pre-image \(Q=(n,{\varvec{\tau}}) \in {{\mathcal{E}}}.\) At this point φ (Q) = ψ (Q w ) reaches its minimum on \({{\mathcal{E}}},\) hence Q belongs to the surface n(τ).
Proof of 3.6
follows from 2.3.
Proof of 3.7.
Let Q = (n 0, τ 01, ..., τ 0k ) belong to \(n({\varvec{\tau}}).\) If \(w_i = -{\frac{\partial n}{\partial\tau_i}} (Q)/c,\) then the equation
defines the tangent plane to \(n({\varvec{\tau}}).\) Since \(n({\varvec{\tau}})\) is convex and decreasing, it follows that w i ≥ 0 and \({{\mathcal{E}}}\) lie on the same side of the plane (A2). Consequently, a strategy having the characteristics Q = (n 0, τ 01,..., τ 0k ) optimizes the losses \(\varphi =n + c \sum^k_{i=1} w_i \tau_i.\) Using 3.5, we complete the proof.
Proof of 3.8.
By (10) and (16) one has
In the trivial case of I(t), one has n i (τ) = 1 − τ and α + β = 1. Hence
i.e., w i = λ i /λ, i = 1, ..., k.
Suppose that {w i } = {λ i /λ}. The likelihood ratio of measures (17) and (18) at the point ω = (J(t), j) is
Accepting the hypothesis H 1 as soon as L(ω) > c and H 0 otherwise, one has
Here we have used 2.1 and 2.2.
Proof of 3.9.
Let us consider the following testing problem: H 1 versus H 0 with the errors β = P 1(L < c) and α = P 0(L ≥ c) where L(ω) = dP 1/dP 0 is the likelihood ratio. Obviously
where F L is the distribution of L with respect to the measure P 0. But dβ = c dF(c) and dα = −d F(c). Therefore
Applying this relation to the case (16), (17), (18), one has
Here L i is the likelihood ratio dP 1/dP 0 for G i .
Proof of 3.10.
In virtue of 3.3 the set D w is a convex hull of the points (12). Let \(\tau_w(\varepsilon)=\sum^r_{i=1} w_i. \) We need to find the maximum of \(y=\sum^r_{i=1}p_i\) under the conditions
Because y is a linear function, it reaches its maximum at the boundary of the region (A3). Consequently, we will consider y as a function of the variables (p 2 ..., p k-1) given by the equation
where p 1 = y − p 2 − ... − p r , p k = 1 − y − p r+1 − ... − p k-1. The point of maximum of y is formally defined by the following equations
This gives \(p_i=c_1\widehat{p}_i, i=1,\ldots ,r;\) \(p_j=c_2\widehat{p}_j,\, j=r+1,\ldots ,k.\) Taking (A5, A4) into account, we get two equations for c 1 and c 2. Finally we have
The conditions (A5, A4) hold, if 0 ≤ y ≤ 1, that is, when \(\widehat{y}(1+q)\le 1.\) Otherwise we have to consider the vector
This vector satisfies (A3, A4) and gives the maximum possible value for y, y = 1.
The proof is complete.
Rights and permissions
About this article
Cite this article
Molchan, G. Space–Time Earthquake Prediction: The Error Diagrams. Pure Appl. Geophys. 167, 907–917 (2010). https://doi.org/10.1007/s00024-010-0087-z
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00024-010-0087-z