1、Absolute Earthquake Locations with Differential DataBy William Menke and David SchaffLamont-Dohery Earth Observatory of Columbia University(corrected version of 6-23-4)AbstractNot withstanding the commonly-held wisdom that “you cant determine the absolute location of earthquakes using the double-dif
2、ference method”, you can. We present a way of visualizing double-difference data, and use it to show how differential arrival time data can, in principle, be used to determine the absolute locations of earthquakes. We then analyze the differential form of Geigers Method, which is the basis of many d
3、ouble-difference earthquake location algorithms, and show that it can be used to make estimates of the absolute location of earthquake sources. Finally, we examine absolute location error in one earthquake location scenario using Monte Carlo simulations that include both measurement error and veloci
4、ty model error, and show that the double-difference method produces absolute locations with errors that are comparable in magnitude, or even less, than traditional methods. Absolute earthquake locations that are already routinely produced by most implementations of the double-difference method have
5、a better accuracy than has been credited.IntroductionThere has been much recent interest in performing earthquake locations using only relative (differential) arrival times from pairs of earthquakes observed at a common station. This technique, which is often called the double-difference method, has
6、 two important advantages over traditional earthquake location procedures that use absolute arrival times. First, differential arrival times can be measured very accurately using waveform cross-correlation techniques, and often have variances that are orders of magnitude smaller than the variances o
7、f absolute arrival times determined by picking arrivals “by eye”. Second, differential traveltimes can be predicted more accurately than absolute traveltimes, because errors in the velocity model tend to cancel out when traveltimes of neighboring events are subtracted e.g. Slunga et al., 1995; Waldh
8、auser and Ellsworth, 2000; Wolfe, 2002. These advantages allow remarkably high-quality locations. Spatial clusters of earthquakes that, using older methodology, appeared to be only diffuse clouds are now resolved into complex patterns of intersecting fault planes.There is a prevailing wisdom that th
9、e double-difference method also has a major disadvantage, namely that it provides only relative earthquake locations. The method is said to be able to reconstruct the spatial pattern of a cluster of earthquake sources, but not provide information about the overall position of that cluster. An often-
10、heard, rough justification for this idea is that differential traveltimes depend upon the distances between events, so it is those distances that are being reconstructed by the method. The solution process is envisioned as determining a mesh linking the events, with the overall position of the mesh
11、being indeterminate.This idea is at odds with the results of synthetic tests of the double-difference method Waldhauser and Ellsworth, 2000; Menke, 2004. In these tests, a set of earthquakes are placed at prescribed positions in an earth model, traveltimes from these events to hypothetical stations
12、are calculated using ray theory and perturbed with random noise, and differential arrival times calculated from the traveltimes and a set of prescribed earthquake origin times. The double-difference method is then used to locate the earthquakes, with the results being compared to the known locations
13、. Typically, the absolute event locations are well-recovered.Slunga et al. 1995 recognized that the justification given above might be misleading. Their implementation of the double-difference method retained the absolute location of the earthquake sources among the unknown variables, because “if th
14、e absolute location significantly affects the theoretical arrival time differences” the technique “may improve the absolute location”. In this paper, we revisit the question of the sensitivity of differential data to absolute location.Conceptualizing Differential Arrival Time DataConsider two earthq
15、uake sources, p and q, separated by a small distance, s. Let the origin times of these sources be p and q, respectively, the arrival time of the P wave at a receiver at position, x, be tp(x) and tq(x), respectively, the traveltime of the P wave from these sources to a receiver at position, x, be Tp(
16、x) and Tq(x), respectively. The differential origin time is defined as pq=(pq), the differential traveltime as Tpq(x)=Tp(x)Tq(x), and the differential arrival time as tpq(x)= tp(x)tq(x).We first consider the special case of a pair of sources in a box-shaped earth of constant velocity, v. While this
17、case it unrealistic, it serves both to illustrate the general pattern of differential arrival times expected from a pair of sources, and a simple method of using those data to determine the absolute location of the events. Seismic rays in a constant-velocity material are just straight lines, so that
18、 the differential traveltime is just (ba)/v, where b and a are the respective distances from sources p and q to the receiver (Figure 1a). The law of cosines, when applied to triangles pmx and mxq in Figure 1a, implies that (b2a2)=2rscos(), where r is the distance from the midpoint, m, between the tw
19、o events to the receiver and is the angle between this direction and a line connecting the two sources. But algebraically, (b2a2)=(a+b)(ab), and when r is much larger than s, (a+b)2r, sotpq(x) pq + s cos() / v(1)This formula is the well-known Fraunhofer approximation. Its application to differential
20、 location is discussed in Aki and Richards 1980 (Section 14.1.3) and by Frchet 1985, Got et al. 1994, and Rubin et al. 1999. At points far from the pair of events, such as on the surface of the box, the pattern of differential arrival times is dipolar (Figure 1b), with an axis of symmetry along the
21、line, R, connecting the two events. The maximum differential time occurs when =0 and is approximately tpqmax=pq+s/v. The minimum occurs when = and is approximately, tpqmin =pqs/v. The mean differential time, tpqmean=pq, occurs on a plane, S, given by =/2. This plane is the locus of all points equidi
22、stant from the two sources and the locus of all points of equal traveltime (Figure 1c). There is one point, A, on the surface of the box with the value of tpqmax and there is one point, B, with the value of tpqmin. The mean time occurs along line segments, C, the intersection of S with the boxs seve
23、ral faces.Idealized Earthquake Location AlgorithmsSince the differential arrival time on the surface of the box in Figure 1c is a measurable quantity, the positions of points A and B and line C, and the values of tpqmax, tpqmin , tpqmean can, in principle, be determined by observation. The event sep
24、aration can then be calculated as s=v(tpqmaxtpqmin)/2, and the differential origin time can be calculated as pq=tpqmean. Furthermore, the midpoint between the events can be determined. It is on the line, R, connecting A and B, at the point where a plane perpendicular to R intersects C. The events ar
25、e located on the line, distances of s/2 from the midpoint. It is therefore clear that measurements of the differential arrival time for two events are sufficient to determine their absolute locations. Note that we have assumed here that the velocity, v, is a known quantity (that is, it has been prec
26、isely determined by independent means such as a seismic refraction experiment). We discuss the impact of the practical problem of velocity model error later in the paper.Knowledge of the position of the symmetry axis, alone, is sufficient to allow events to be located, provided that differential arr
27、ival time data for several overlapping pairs of events are available. Suppose that we know the position of points A and B for a pair (p,q) of events (lets call these points, Apq and Bpq). Events p and q must lie on the line connecting Apq and Bpq (Figure 1d). Suppose that we also know the position o
28、f points Apr and Bpr for the (p,r) pair of events. Events p and r must lie on the line connecting these two points. Event p must therefore lie at the intersection of the two lines. Similarly, the other two events, q and r, must lie at the intersection of analogous lines. The absolute position of the
29、 events can be determined solely through knowledge of where on the surface of the box the differential arrival time is minimum and maximum. It is therefore clear that having information about overlapping pairs of events helps with the location processNow let us consider the more realistic case of a
30、vertically-stratified velocity structure (that is, v(x)=v(z), where z is depth). Like the constant velocity case, the pattern of differential arrival times also possesses symmetry, in this case mirror symmetry about the vertical plane that contains the two sources (Figure 2a). Once again, the measur
31、ements of the line of symmetry on the surface of the earth can be used to constrain the location of the events: Since events p and q must lie on the vertical plane beneath the line of symmetry of tpq, and events p and r must lie on the vertical plane beneath the line of symmetry of tpr, event p must
32、 lie on the vertical line defined by the intersection of the two planes. Only the horizontal coordinates, (x,y), and not the depth, z, of the event are determined.To proceed further, we must make some assumptions about the way in which the velocity field varies in the vicinity of the events. If it p
33、ossesses strong variability over length scales smaller than the event separation, then the pattern of differential arrival times might be very complicated, or even ill-defined (if, for example, one event was shadowed from a given receiver, while the other wasnt). We will restrict ourselves to cases
34、where it varies sufficiently slowly that ray paths are simply “curvy” versions of the straight lines of the constant-velocity case. Far from a pair of events, we would then expect that the differential arrival time would be at least approximately a generalized version of Equation 1:tpq(x) pq + T cos
35、()(2)Here T, the traveltime from the location of event p to q, is a generalization of the quantity, s/v, in Equation 1. It is computed along the ray, R, connecting the two events (Figure 3). The angle, , is now understood to mean the takeoff angle of a ray from the midpoint between the sources to th
36、e receiver at position, x, measured with respect to the line connecting the sources. A key property of this formula is that the minimum and maximum differential arrival times occur along R.When this approximation holds, then differential arrival times are just the dipolar function from the constant-
37、velocity case, projected along rays that diverge from the source midpoint onto the surface of the earth (Figure 4). In the vertically-stratified case, several special cases are worth examining: When events p and q are separated by a purely horizontal distance increment, then the points A and B are i
38、n the vertical plane containing the sources, and line C is the perpendicular bisector of line AB (Figure 5a); When the events are separated by a purely vertical distance increment, then the point A is directly above the events and the curve, C, is a circle centered on A (Figure 5b). More generally (
39、and not limited to the vertically-stratified case), the location algorithms from the constant-velocity case have simple generalizations: Once the points A and B are known, the ray, R, can be found by raytracing from A to B. The midpoint between the events is the point on R from which a suite of rays
40、, when raytraced from this point perpendicular to R, intersect C. As in the constant-velocity case, when data from several overlapping pairs of events are available, the events can be located by finding points of intersection between rays (Figure 5c).The patterns in Figures 4, 5a, and 5b can be obse
41、rved in actual data. Figure 6 shows four pairs of earthquakes on the San Andreas Fault at Parkfield recorded on the short-period, vertical component Northern California Seismic Network (NCSN). Shaded contours show the theoretical distribution of differential arrival times, calculated by raytracing t
42、hrough an appropriate vertically-stratified earth model. Individual boxes indicate the observed arrival times for each station, which in general are in very good agreement with the underlying theoretical distribution. In Figures 6a and 6b, the pattern of the differential arrival times corresponds cl
43、osely to the expected horizontal separation pattern in Figures 4a & 5a. The pairs are both separated by 310 m leading to tpqmax of about 0.06 s. These measurements are made by waveform cross correlation which have precision below the sample level (0.01 s) sufficient to reveal the pattern in differen
44、tial arrival times for these separation distances. For longer separation distances and larger events, ordinary phase pick data can produce the expected patterns over a broad scale as well (Figures 6c & 6d). The event pair in Figure 6c is separated by 4.5 km, mostly in the vertical, and therefore mat
45、ches best with the patterns in Figures 4b & 5b. The event pair in Figure 6d has a separation of 5.6 km and has the expected theoretical pattern for an inclined separation vector (Figure 4c).The accuracy of the approximation in Equation 2 is difficult to assess in general. We give here an analysis of
46、 one specific case, that of events separated by a purely horizontal distance increment in a vertically-stratified medium (Figure 7a). This configuration is one where we might expect the accuracy of the approximation to be poorest, since the velocity gradient is perpendicular to the line connecting t
47、he sources, and a ray propagating from one source towards the other will have maximum curvature.Let the traveltime from a source at position, (x,z)=(0,z0) to a receiver at position (x,z)=(0,x) be T(x). In many relevant earth models, where v(z) monotonically increases with depth, this function will r
48、esemble the sketch in Figure 7b. The traveltime increases with range, x, and passes through an inflection point at some range, x0, that corresponds to a ray that leaves the source exactly horizontally. The horizontal slowness, dT/dx, reaches a maximum of 1/v(z0) at this range (Figure 7c).Suppose tha
49、t sources p and q are located at depth, z0, with horizontal positions h and h, respectively. The differential traveltime at position, x, is Tpq(x) = T(x+h) T(xh). We now expand this formula in a Taylor series:Tpq(x) = 2h dT/dx + O(h3)(3)To a high degree of accuracy (that is, to order h2), the differential traveltime is maximum when the slowness, dT/dx, is maximum. But dT/dx has a maximum value at the point, x0. As predicted by Equation 2, the m
