Simultaneous Inversion for 3D crustal Structure and ...€¦ · nch and Manfred Koch, Department of...

1
kassel.de - kochm@uni , kassel.de - tmuench@uni ng Hydrology, nch and Manfred Koch, Department of Geohydraulics and Engineeri ü Contact: Thomas W. M [email protected] Schlittenhadrt, rg ö J Simultaneous Inversion for 3D crustal Structure and regional Hypocenters beneath Germany in the Presence of an anisotropic upper Mantle Thomas Willi Münch, Manfred Koch and Jörg Schlittenhardt Introduction As recognized in previous studies (Song et al., 2001, 2004), travel times of Pn-phases across Germany show anisotropic behaviour. Main task here is to study the influence of the upper mantle anisotropy onto the tomographic reconstruction of the seismic velocities in the crust and upper mantle across Germany. Dataset for the 3D SSH (Simultaneous inversion for Structure and Hypocenters, (Koch, 1993)) tomography consists of regional arrival times recorded across Germany between 1975 to 2003. Due to the large number of records (Table 1), good ray-coverage of study area is ensured (Fig. 1). 751 873 895 PMP-Phases 5880 9001 12804 Pn-Phases 15438 20279 46550 Pg-Phases 1223 1812 10058 Events N-obs >7, GAP <180 N-obs > 7 Total Fig, 1: Regional seismic events and ray- coverage (P + S- phases) across Germany. Anisotropy, preliminary investigations Fig. 4: Determination of optimal anisotropy ellipse. For hypocenters fixed (Fig.4a) the optimal anisotropy angle of about 35°NE is obtained. For full inversion (Fig.4b) optimal angle is at 26°NE coinciding better with results of Enderle et al. (1999). One aim of the study is to show influence of upper mantle P n -anisotropy on the seismic inversion. Therefore, P n - ray tracing is corrected by elliptical (azimuthal) anisotropy, quantified by the velocity contrast (%) and angle of the major axis (Fig.2) Fig. 3: Original Pn residuals (blue) and optimally anisotropically corrected. Fig. 3: Effects of anisotropic Pn- correction on the observed travel-time residuals (using a standard 1D- seismic velocity model for Germany). After anisotropic correction with - +2.5% contrast, the residuals nearly lie on a straight line. Fig. 3:Anisotropicall corrected with +- 1% (top) and with +-5% (bottom) 3D-seismic Models, synthetic Tests Synthetic tests with random lateral velocity anomalies An artificial anisotropic traveltime dataset with several anomalies in the four layers (depths=[0-10];[10-20];[20-30]; >30km) of the model (Fig. 5a) is synthesized and re-inverted. The traveltimes are computed using the original hypocenter locations and station records and putting arbitrary velocity anomalies as shown in Fig. 3a into the model. The reconstructed anisotropic models are shown in Fig. 3b and 3d, respectively, without and with traveltime noise (σ = 0.1s) included. Fig. 3c shows the isotropic inversion. The anisotropic (Fig. 5b, Fig. 5d) reconstruction shows a good agreement with the original model (except in layer 3) due to the lack of earthquakes. The RMS of the data fit is also smaller than that of the isotropic reconstructed. The isotropic (Fig. 5c) reconstruction has no resolution in the 1 st layer, produces only artefacts in the three next layers and has a three times higher RMS. Figure 3d: anisotropically inverted model with noisy data, RMS = 0.1656 s Figure 3a: Original model with random lateral pertubations Figure 3b: anisotropically inverted model RMS = 0.0771 s Figure 3c: isotropically inverted model RMS = 0.2324 s Figure 6c: Inversion with 35x35 blocs, RMS = 0.1593 s Figure 6b: Inversion with 25x25 blocs, sigma = 0.1 s, RMS = 0.2069 s Figure 6a: Inversion with 25x25 blocs, RMS = 0.1435 s The checkerboard tests show where a good lateral resolution can be expected, reflecting the inhomo- geneous ray-coverage of the model area. Even with noisy data (Fig. 6b) a good resolution in the first and fourth layer for the 25x25 bloc models is obtained. For the second and third layer at a depth range between 10 and 30 km, good resolution is obtained only in the south-western part of the model. The other areas suffer from lack of seismic events and rays. Fig. 6c shows how the resolved areas are reduced when a 35x35 bloc discretization is used. 3D-seismic Models, Checkerboard-Tests Fig. 7b: Anisotropic 25x25 model; TSS = 15113 s², RMS = 0.8197 s Fig. 7a: Isotropic 25x25 3D mode TSS = 17458 s², RMS = 0.8810 s Fig.8a: Isotropic 35x35 model; TSS = 16995 s², RMS = 0.8693 s Fig. 8b: Anisotropic 35x35 model; TSS = 15767 s², RMS = 0.8373 s The four 3D-tomographic seismic models for the crust and upper mantle exhibit slightly different features, mainly in the upper mantle layer. Based on the objective criteria of the Total Square Sum of the residuals (TSS) and the synthetic tests, the anisotropic models are considered to be more reliable. 25x25 bloc models (Fig. 7) The first layer nearly has the same structure for both model variants, although major shallow geological features (Rhinegraben area, Vogelsberg and Nördlinger Ries) are better recognized in the anisotropic model. Starting with the second layer, structural differences show up between isotropic and anisotropic models Structure in the fourth layer mainly follows some tectonic features in the tectonic map (Fig. 9), i.e. may represent ancient suture zones of the variscan orogeny. 35x35 bloc models (Fig. 8) Compared with the 25x25 models, datafit is improved. But resolution in some areas of the model is not trustable, as shown by the checkerboard tests. Never- theless major structurural features are similar to the coarser model. Fig. 9 : Tectonic map of Germany with major ancient suture zones V. Simultaneously relocated hypocenters Simultaneously with the isotropic and anisotropic optimal 3D velocity models relocated hypocenters show only minor differences. Isotropically computed epicenters (Fig. 9, left) appear to be more clustered in the EW- direction (effect of anisotropic bias??) than anisotropically computed ones. No visual differences in the depth shifts of the events. Fig. 9: Epicentral and hypocentral (depth) shifts for simultaneously with 3D 25x25 optimal velocity models relocated events (left: isoptropic; right: anisotropic model Table 2: Average shifts of epicenters and for different model variants; x= x-shift , y=y-shift; r = total horizontal shift; Φ=angle of shift.

Transcript of Simultaneous Inversion for 3D crustal Structure and ...€¦ · nch and Manfred Koch, Department of...

ka

ss

el.d

e-

ko

ch

m@

un

i,

ka

sse

l.d

e-

tmu

en

ch

@u

ni

ng

Hyd

rolo

gy,

nc

h a

nd

Ma

nfr

ed

Ko

ch

, D

ep

art

me

nt

of

Ge

oh

yd

rau

lic

s a

nd

E

ng

ine

eri

üC

on

tac

t: T

ho

ma

s W

. M j.s

ch

litt

en

ha

rdt@

bg

r.d

eS

ch

litt

en

had

rt,

rgö

J

Simultaneous Inversion for 3D crustal Structure and regional Hypocenters beneath Germany in the Presence of an anisotropic upper Mantle

Thomas Willi Münch, Manfred Koch and Jörg Schlittenhardt

IntroductionAs recognized in previous studies (Song et al., 2001, 2004), travel times of Pn-phases across Germany show anisotropic behaviour. Main task here is to study the influence of the upper mantle anisotropy onto the tomographic reconstruction of the seismic velocities in the crust and upper mantle across Germany.

Dataset for the 3D SSH (Simultaneous inversion for Structure and Hypocenters, (Koch, 1993)) tomography consists of regional arrival times recorded across Germany between 1975 to 2003. Due to the large number of records (Table 1), good ray-coverage of study area is ensured (Fig. 1).

751873895PMP-Phases

5880900112804Pn-Phases

154382027946550Pg-Phases

1223181210058Events

N-obs >7, GAP <180N-obs > 7Total

Fig, 1: Regional seismic events and ray-coverage (P + S- phases) across Germany.

Anisotropy, preliminary investigations

Fig. 4: Determination of optimal anisotropy ellipse. For hypocenters fixed (Fig.4a) the optimal anisotropy angle of about 35°NE is obtained. For full inversion (Fig.4b) optimal angle is at 26°NE coinciding better with results of Enderle et al. (1999).

One aim of the study is to show influence of upper mantle Pn-anisotropy on the seismic inversion. Therefore, Pn- ray tracing is corrected by elliptical (azimuthal) anisotropy, quantified by the velocity contrast (%) and angle of the major axis (Fig.2)

Fig. 3: Original Pn residuals (blue) and optimally anisotropically corrected.

Fig. 3: Effects of anisotropic Pn-correction on the observed travel-time residuals (using a standard 1D- seismic velocity model for Germany). After anisotropic correction with -+2.5% contrast, the residuals nearly lie on a straight line.

Fig. 3:Anisotropicall corrected with +-1% (top) and with +-5% (bottom)

3D-seismic Models, synthetic Tests

Synthetic tests with random lateral velocity anomalies

An artificial anisotropic traveltime dataset with several anomalies in the four layers (depths=[0-10];[10-20];[20-30]; >30km) of the model (Fig. 5a) is synthesized and re-inverted. The traveltimes are computed using the original hypocenter locations and station records and putting arbitrary velocity anomalies as shown in Fig. 3a into the model.

The reconstructed anisotropic models are shown in Fig. 3b and 3d, respectively, without and with traveltime noise (σσσσ = 0.1s) included. Fig. 3c shows the isotropic inversion.

The anisotropic (Fig. 5b, Fig. 5d) reconstruction shows a good agreement with the original model (except in layer 3) due to the lack of earthquakes. The RMS of the data fit is also smaller than that of the isotropic reconstructed.

The isotropic (Fig. 5c) reconstruction has no resolution in the 1st

layer, produces only artefacts in the three next layers and has a three times higher RMS.

Figure 3d: anisotropically inverted modelwith noisy data, RMS = 0.1656 s

Figure 3a: Original model with random lateral pertubations

Figure 3b: anisotropically inverted modelRMS = 0.0771 s

Figure 3c: isotropically inverted model RMS = 0.2324 s

Figure 6c: Inversion with 35x35 blocs, RMS = 0.1593 s

Figure 6b: Inversion with 25x25 blocs, sigma = 0.1 s, RMS = 0.2069 s

Figure 6a: Inversion with 25x25 blocs, RMS = 0.1435 s

The checkerboard tests show where a good lateral resolution can be expected, reflecting the inhomo-geneous ray-coverage of the model area. Even with noisy data (Fig. 6b) a good resolution in the first and fourth layer for the 25x25 bloc models is obtained. For the second and third layer at a depth range between 10 and 30 km, good resolution is obtained only in the south-western part of the model. The other areas suffer from lack of seismic events and rays.Fig. 6c shows how the resolved areas are reduced when a 35x35 bloc discretization is used.

3D-seismic Models, Checkerboard-Tests

Fig. 7b: Anisotropic 25x25 model; TSS = 15113 s², RMS = 0.8197 s

Fig. 7a: Isotropic 25x25 3D mode

TSS = 17458 s², RMS = 0.8810 s

Fig.8a: Isotropic 35x35 model; TSS = 16995 s², RMS = 0.8693 s Fig. 8b: Anisotropic 35x35 model; TSS = 15767 s², RMS = 0.8373 s

The four 3D-tomographic seismic models for the crust and upper mantle exhibit slightly different features, mainly in the upper mantle layer. Based on the objective criteria of the Total Square Sum of the residuals (TSS) and the synthetic tests, the anisotropic models are considered to be more reliable.

25x25 bloc models (Fig. 7)The first layer nearly has the same structure for both model variants, although major shallow geological features (Rhinegraben area, Vogelsberg and NördlingerRies) are better recognized in the anisotropic model.

Starting with the second layer, structural differences show up between isotropic and anisotropic models

Structure in the fourth layer mainly follows some tectonic features in the tectonic map (Fig. 9), i.e. may represent ancient suture zones of the variscan orogeny.

35x35 bloc models (Fig. 8)Compared with the 25x25 models, datafit is improved. But resolution in some areas of the model is nottrustable, as shown by the checkerboard tests. Never-theless major structurural features are similar to thecoarser model.

Fig. 9: Tectonic map of Germany with major ancient suture zones

V. Simultaneously relocated hypocenters

Simultaneously with the isotropicand anisotropic optimal 3D velocitymodels relocated hypocenters showonly minor differences.

Isotropically computed epicenters(Fig. 9, left) appear to be moreclustered in the EW- direction (effectof anisotropic bias??) thananisotropically computed ones.

No visual differences in the depthshifts of the events.

Fig. 9: Epicentral and hypocentral (depth) shifts for simultaneously with 3D 25x25 optimal velocity models relocated events (left: isoptropic; right: anisotropic model

Table 2: Average shifts of epicenters and for different model variants; x= x-shift , y=y-shift; r = total horizontal shift; Φ=angle of shift.