Multi-Agency Seismic Moment Tensor Dataset from EMSC SeismicPortal, 2014–2026

This dataset supports the consensus moment tensor (MT) method developed at the European-Mediterranean Seismological Centre (EMSC) and described in Bossu et al. (2026). For many earthquakes, several MT solutions are reported by different seismological agencies using different networks and inversion methods. The consensus approach of Bossu et al. (2026) computes a provider-neutral, reproducible representative solution per earthquake by applying robust median/MAD statistics to solutions compared with two complementary metrics: the Kagan (1991) rotation angle (double-couple geometry) and the Rivera–Kanamori (RK) similarity (Rivera & Kanamori, 2014). The repository contains six files: a README, one raw input file, two global consensus result files, and two files specific to the Tonga region case study. **File 1:** `mt_seismicportal_20140101-20260430.json` (53Mo) — Raw input dataset retrieved from the EMSC SeismicPortal. Covers 10,650 seismic events with 34,425 moment tensor solutions (minimum 2 solutions per event), contributed by 28 agencies worldwide, for the period 2014-01-01 to 2026-04-30. Each event includes hypocenter parameters and, for each contributing agency, centroid location, scalar moment, moment magnitude, nodal planes, full tensor components, principal axes, and source decomposition (Vavryčuk, 2015). **File 2:** `consensus_kagan_run_prefilt-31_strictMAD-publish-17-final.json` (40Mo) — Consensus results using the Kagan rotation angle as pairwise dissimilarity metric. Covers 10,646 events; 8,580 (80.6%) meet the publication threshold (MAD ≤ 17°). Each record contains the pairwise Kagan angle matrix, candidate selection details, consensus centroid with uncertainty ellipse, normalized moment tensor, nodal planes, source decomposition, quality grade (A/B/C), and cross-validation against File 3. **File 3:** `consensus_rk_run_prefilt-0.77_strictMAD-publish-0.94-final.json` (43Mo) — Consensus results using the Rivera–Kanamori similarity as pairwise metric. Covers 10,504 events (pure DC solutions excluded); 7,491 (70.3%) meet the publication threshold (median RK ≥ 0.94). Same structure as File 2, with cross-validation against File 2. **File 4:** `tonga_data.json` (540Ko) — Regional subset covering 55 events in the Tonga region (lat −25.7° to −19.1°, lon −176.0° to −173.0°), spanning 2014–2026, with 212 moment tensor solutions from 6 agencies (CPPT, GCMT, GFZ, IPGP, NEIC, SC4) and magnitudes Mw 5.1–6.3. Unlike Files 1–3, which store raw data and consensus results separately, this file merges all three sources: each event record includes the original multi-agency solutions alongside both Kagan and RK consensus results. **File 5:** `tonga_data_results.json` (430Ko) — Inter-event variability results derived from File 4, supporting Figure 2 of Bossu et al. (2026). Organised into two panels (Kagan angle and RK similarity), each containing four series: the consensus mechanism and the three main contributing agencies (GCMT, NEIC, GFZ). For each series, the file stores all pairwise inter-event values, the cumulative distribution curve, the median, event and pair counts, and the fraction of pairs beyond a literature-based threshold (Kagan ≤ 20°; RK ≥ 0.90). Panels are restricted to publishable events: n = 43 events (903 pairs) for Kagan, n = 40 events (780 pairs) for RK. This file is self-contained and allows Figure 2 to be reproduced without recomputing inter-event metrics from the raw moment tensors.