Imaging the Galápagos mantle plume with an unconventional application of floating seismometers

Abstract

We launched an array of nine freely floating submarine seismometers near the Galápagos islands, which remained operational for about two years. P and PKP waves from regional and teleseismic earthquakes were observed for a range of magnitudes. The signal-to-noise ratio is strongly influenced by the weather conditions and this determines the lowest magnitudes that can be observed. Waves from deep earthquakes are easier to pick, but the S/N ratio can be enhanced through filtering and the data cover earthquakes from all depths. We measured 580 arrival times for different raypaths. We show that even such a limited number of data gives a significant increase in resolution for the oceanic upper mantle. This is the first time an array of floating seismometers is used in seismic tomography to improve the resolution significantly where otherwise no seismic information is available. We show that the Galápagos Archipelago is underlain by a deep (about 1900 km) 200-300 km wide plume of high temperature, with a heat flux very much larger than predicted from its swell bathymetry. The decrease of the plume temperature anomaly towards the surface indicates that the Earth's mantle has a subadiabatic temperature gradient.

Figure 1

A bathymetric map of the Galápagos hotspot region. The locations of MERMAID floats at the times of the P wave arrivals is indicated by circles, where the colour indicates epicentral distance (red: Δ < 10°, orange: 10° < Δ < 30°, yellow 30° < Δ < 100°, green: Δ > 100°). Regional land seismometers used are indicated by red triangles. Plate boundaries are shown in magenta. Lines AA’ and BB’ denote the locations of the cross-sections shown in Fig. S3.

Figure 2

Examples of seismograms recorded by MERMAIDs for a range of epicentral distances. (A) May 13, 2014, 7.2 °N 82.3 °W, depth h = 10 km, Mw 6.5, (B) Sep 21, 2015, 31.7 °S, 71.4 °W, h = 35 km, Mw 6.6, (C) May 29, 2015, 56.6 °N 156.4 °W, h = 73 km, Mw 6.7, (D) PKP waves from May 30, 2015, 27.8 °N, 140.5 °E, h = 664 km, Mw 7.8. Delay is defined with respect to the predicted P wave arrival time for model AK135, inverted triangles indicate the picked onset. Distances Δ (in degrees) are listed with each recording. The records of GSN station PAYG, located in a borehole on the Galápagos Island of Santa Cruz, are listed for comparison.

Figure 3

Resolution test for a checkerboard at depths of 135, 316, 587 and 948 km comparing (left) the improved resolution using ISC delays plus MERMAID data with (right) that for ISC delays only. Damping, smoothing and number of iterations was the same in both cases. Colour scale is in percent. The checkerboard boundaries (green lines) follow the cubed Earth parameterization and are 6 voxels wide. Circles indicate the (±2.5%) anomaly in the input model.

Figure 4

Three-dimensional image of the preferred tomographic solution for P-velocity anomalies. An arrow denotes a negative velocity anomaly that is continuous to a depth of about 1900 km, and that is imaged in the cross-sections of Fig. 5. The magenta lines denote plate boundaries. Upper mantle anomalies A-D are discussed in the text.

Figure 5

Tomographic result showing the P-velocity anomaly of the Galápagos plume in two perpendicular depth sections viewed from the East (left) or South (right). The cross sections are slightly warped with depth to track the maximum anomaly; deviations from 90 °W (left) and 0 °N (right) are plotted in the graphs below the plot. The transition zone between upper- and lower mantle is indicated by black lines at 410 and 660 km depth.

Figure 6

The result of filtering the seismograms. We subtract the estimated microseismic signal (bottom plot) from the observed seismogram (center), to obtain the filtered signal (top) used for picking the onset time.

Figure 7

(a) An example of ISC delay standard error estimation over a cluster of closely spaced earthquakes using the method of Voronin et al.. In this case we selected 38 events (yellow dots) and diagonalized the tomographic submatrix for their N = 6842 wavepaths (solid lines). (b) The distribution of the 6842 eigenvalues (logarithmic scale). The rapid fall-off indicates a large redundancy caused by overlapping raypaths. (c) the distribution of the projected data for the 2552 smallest eigenvalues (eigenvalue λ < 0.1). These eigenvalues are associated with directions in data space that are dominated by noise, allowing an estimate of the standard error in the data (0.53 s for this selection). The red curve shows the best fitting normal distribution, (d) averaging over the estimated errors from 196 such clusters we determined that the standard error in the ISC-EHB data was 0.51 s. The blue curve shows the best fitting lognormal distribution.