Auxiliary Material Three-dimensional seismic structure of the Dragon Flag oceanic core complex at the ultraslow spreading Southwest Indian Ridge (49°39′E) Author names and affiliations Minghui Zhao*, Key Laboratory of Marginal Sea Geology, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China Tel: +86-20-89023192(o); Mob: +86-13560200916, Email: mhzhao@scsio.ac.cn Xuelin Qiu, Key Laboratory of Marginal Sea Geology, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China Tel: +86-20-89023157(o); Mob: +86-13622842912, Email: xlqiu@scsio.ac.cn Jiabiao Li*, The Second Institute of Oceanography, State Oceanic Administration, Hangzhou, 310012, China (corresponding author) Tel: +86-571-81963137; Mob: +86-13805790203; Email: jbli@sio.org.cn Daniel Sauter, Institut de Physique du Globe de Strasbourg, CNRS-UMR 7516, 5 Rue Descartes, F-67084 Strasbourg Cedex, France Email: Daniel.Sauter@unistra.fr Aiguo Ruan, The Second Institute of Oceanography, State Oceanic Administration, Hangzhou, 310012, China Mob: +86-13867434884; E-mail: ruanag@163.com John Chen, Peking University, Beijing 100871, China Mob: +86-13683583697; Email: johnyc@pku.edu.cn Mathilde Cannat, Laboratoire de Geoscience Marine, Institut de Physique du Globe de Paris, 75238 Paris Cedex 05 France, Email: cannat@ipgp.fr Satish Singh, Laboratoire de Geoscience Marine, Institut de Physique du Globe de Paris, 75238 Paris Cedex 05 France Tel: +33 (0) 1 83 95 76 58, Mobile: +33 (0) 6 88285593; Email: singh@ipgp.fr Jiazheng Zhang, Key Laboratory of Marginal Sea Geology, South China Sea Institute of Oceanology, Chinese Academy of Sciences, Guangzhou 510301, China University of Chinese Academy of Sciences, Beijing 100049, China Mob: +86-15013109871; Email: zhangjz005@163.com Zhenli Wu, The Second Institute of Oceanography, State Oceanic Administration, Hangzhou, 310012, China Mob: +86-13819102399; E-mail: zhenli318@vip.sina.com Xiongwei Niu, The Second Institute of Oceanography, State Oceanic Administration, Hangzhou, 310012, China Mob: +86-13588307009; E-mail: nnxxww123@163.com Theme: Oceanic Detachment Faults 1 Random Initial Models We built the 100 1D starting models at random by limiting five parameters defining a two-layer media: three P-wave velocities (at the seafloor, at the interface between both layers, and at the bottom of the lower layer, respectively), and the thicknesses of both layers (Auxiliary Figure S1). The five parameters were permitted to vary randomly within certain predefined ranges to prevent from constructing unreasonable models. The velocities at the seafloor, middle interface, and at the bottom of the model were bounded by 2–4 km/s, 5.3–7.3 km/s and 7.3–8.5 km/s, respectively. The thicknesses of both layers were limited to the range of 1–3 km and 7–9 km for the upper and lower layers, respectively. For each Monte Carlo ensemble a 3D starting volume was built by draping the randomized 1D velocity models from the seafloor. Resolution Tests Although the normalized χ2 misfit provides a general indication of the quality of the velocity model, it is necessary to examine the data fit and model quality in greater detail. Checkerboard tests are commonly used for model assessment and resolution analysis, providing information on the lateral and vertical resolution of a velocity model by examining the ability to recover a pre-defined velocity anomaly pattern [Zelt, 1998]. A perturbed velocity model was built by superimposing a 3D sinusoidal velocity anomaly on the best fitting model with a maximum magnitude of 10% background velocity. The sinusoidal function has half-wavelengths of 5 km in the horizontal directions, a half-wavelength of 2 km in the vertical direction (Auxiliary 2 Figure S5). Synthetic travel times were generated by tracing rays through the perturbed velocity models and then adding Gaussian random noise with a standard deviation of 15 ms. These synthetic travel time data were then inverted for the preferred velocity model as the starting model using the same parameters as the previous inversion. We find that the inversion is able to resolve the checkerboard pattern (Auxiliary Figure S5). A good recovery of the velocity perturbation is observed with velocity anomaly of size 5×5×2 km, which indicating a good model resolution in the central areas of the Dragon Flag segment, where ray coverage is high and velocity uncertainty is small (Figure 5–6). These tests show that our inversion cannot constrain velocity variations at depths greater than ~7 km bsl; thus we limit our interpretation of the velocity model to structure above that level (Figure 5–6). The inversion can resolve features as small as 5 km in size at depths shallower than 2 km (<6 km bsl) along the central part of segment (-20≤X≤30 km and -20≤Y≤15 km, Auxiliary Figure S5). At depths between 2–4 km bsf (6–8 km bsl), however, the inversion can only resolve these smaller anomalies between -10≤X≤10 km and -15 ≤Y≤15 km, Auxiliary Figure S5. Auxiliary Figure S1 100 randomized initial models. (a) 100 Monte Carlo ensembles (blue lines) together with ensemble mean (a red line). (b) Standard deviation corresponding to the 100 initial models. Auxiliary Figure S2 Two sections (V4 and V6) transverse to the ridge axis from the 3 3D final velocity models, velocity gradient, velocity anomaly, and velocity uncertainty model (locations are shown in Figure 2). Other symbols are as in Figures 5 Auxiliary Figure S3 Two sections (H4 and H6) parallel to the ridge axis from the 3D final velocity models, velocity gradient, velocity anomaly, and velocity uncertainty model (locations shown in Figure 2). Other symbols are as in Figure 5. Auxiliary Figure S4 Plan view slices at constant depths below the seafloor from (a) the 3D final velocity models and (b) the uncertainty models. Other symbols are as in Figures 2. Auxiliary Figure S5 Checkerboard tests results. Anomalies with half wavelengths of 5 km (lateral)×5 km (lateral)×2 km (vertical). Recovered anomalies are shown for (a) Y=0 km, (b) X=0 km, (c) Z=5, and (d) Z=7 km below the sea level (bsl). Grid lines and yellow +/- symbols show the sinusoidal synthetic velocity anomaly patterns, which are up to 10% in magnitude relative to the background velocity (mean of the 100 initial ensembles, Auxiliary Figure S1). Green circles indicate OBSs’ positions. 4