The 18 September 2004, magnitude mbLg 4.6 (a local magnitude computed from Lg-wave amplitudes) earthquake near the Itoiz reservoir dam (Navarra, northern Spain) occurred eight months after the reservoir's first impoundment. A 2009 analysis of the seven main seismic clusters in the region found that the 2004 cluster had a distinctively higher correlation dimension of its epicenter distribution than the other clusters, suggesting a different — possibly reservoir-induced — seismic source. Later work resolved the triggering question itself: the seismicity is reservoir-induced, driven by pore-pressure diffusion through the karstic Eocene Hecho Group. However, the statistical-discrimination question raised by the 2009 study — and left explicitly open by its 2011 companion — remains open: whether the correlation-dimension anomaly survives analysis in three dimensions with depth-resolved catalogs, and whether any fractal or network statistic — including the neighborhood-based clustering coefficient proposed by Tsonis et al. (2008) — can separate a reservoir-triggered cluster from tectonic clusters once mainshock magnitude and cluster-size effects are controlled. A 2011 weighted-network reanalysis by the same group found no distinct feature for the 2004 cluster and explicitly left the discrimination of artificial from natural seismicity as future research. The task is a reproducible, catalog-based determination of these questions, for which a well-supported negative answer is as valid as a positive one.

Background

Reservoir-triggered (also called reservoir-induced) seismicity (RTS or RIS) is earthquake activity set off by the filling or operation of a large artificial reservoir: the water load and, more importantly, the diffusion of elevated pore fluid pressure into fault zones can reduce effective normal stress on pre-existing critically stressed faults and trigger earthquakes that would otherwise have occurred later or not at all. Classic cases include Koyna (India) and Lake Mead (USA), and the phenomenon is reviewed, together with its usual macroscopic indicators — temporal correlation with water level, elevated b-value (the slope of the Gutenberg–Richter frequency–magnitude relation), and migration of hypocenters — in the RTS literature (Gupta, 2022).

The Itoiz dam sits on the Irati river in Navarra, northern Spain; impoundment began in January 2004. On 18 September 2004, an earthquake of magnitude mbLg 4.6 (magnitude computed from Lg-wave amplitudes) occurred with its epicenter about 4 km from the dam, followed by a widely felt aftershock on 30 September 2004. The sequence attracted a cluster of studies. Geomechanical work resolved the causal question: Durá-Gómez and Talwani (2010) concluded that the seismicity is reservoir-induced, driven by diffusion of excess pore pressure (of order 100–500 kPa) along the carbonate megabreccia systems of the Eocene Hecho Group, with subsequent annual filling cycles accompanied by continued epicentral growth; Luzón et al. (2010) modeled pore-pressure changes with a hybrid Green's-function/finite-difference technique and found the direct Coulomb stress change (the combination of shear and normal stress on a fault that promotes slip) from the water load (at most about 0.5 kPa) too small, pointing again to diffusion as the mechanism; Santoyo et al. (2010) computed the subsurface stress field near the reservoir.

The study that generated the present question took a different, statistical route. Jiménez, Tiampo, Posadas, Luzón and Donner (2009) declustered the regional catalog with a link algorithm based on the ETAS model (epidemic-type aftershock sequence model: a stochastic point process in which every earthquake triggers its own aftershocks following the Omori–Utsu decay law, with productivity increasing with magnitude), obtaining seven main clusters. For each cluster they computed the correlation dimension D2D_2 — the fractal dimension of the earthquake spatial distribution, estimated à la Grassberger–Procaccia from the correlation sum C(r)C(r), the fraction of event pairs closer than a distance rr, which scales as C(r)∝rD2C(r) \propto r^{D_2} in a scaling range — and the properties of the complex network formed by treating earthquakes as nodes joined by links according to an ETAS-based space–time–magnitude distance. Key network measures are the clustering coefficient (the probability that two neighbors of a node are themselves connected, measuring local connectivity), the average path length (mean number of links on the shortest path between node pairs), and node strength (sum of link weights). Networks combining high clustering with short path length are called small-world, after the Watts–Strogatz model (a ring lattice in which a fraction of links is randomly rewired). Because reliable hypocentral depths were unavailable for most events, the 2009 analysis used only horizontal epicenter coordinates.

The 2009 study concluded that the cluster representing the 18 September 2004 earthquake has properties distinctively different from the other clusters — most notably a higher fractal dimension of the epicenter distribution — which could indicate a different seismic source produced by the impoundment. Because this anomaly was computed from epicenter coordinates alone, it might be an artifact of the two-dimensional analysis — a vertically elongated cluster projected onto the map plane can acquire a different apparent dimension — and settling it requires catalogs with reliable depth information (3D catalogs). The generative origin of the observed network structure likewise remained unresolved: the 2009 study reported small-world behaviour of the cluster networks, but the 2011 reanalysis found small-world behaviour against one theoretical random model while finding none against weight-shuffled surrogates ('No single answer was found for this question'), and although ETAS-type simulations approached the seismicity better than the random models tried, they may still not fit the data sufficiently well, with BASS-model-based distances proposed as a future alternative. The 2009 reference list includes an alternative, neighborhood-based definition of the clustering coefficient designed for spatially embedded networks (Tsonis et al., 2008), which has not been applied to these clusters. The open issues are those set out in the 2011 companion by two of the same authors, described next.

The direct methodological follow-up, Jiménez and Luzón (2011), reanalyzed the same clusters as weighted complex networks. They found that the clusters were better modeled by ETAS simulations than by random network models, that the topological properties of the clusters group according to mainshock magnitude and number of events rather than by any other characteristic, and that no distinct feature could be obtained for the 18 September 2004 series; on the small-world question the results were inconclusive (small-world behaviour appeared against one null model but not against weight-shuffled surrogates). The authors explicitly posed as open future research whether the method can differentiate seismicity affected by artificial (anthropogenic) processes from naturally produced seismicity, ideally applied at sites without induced seismicity for comparison, together with the calculation of further graph-theoretic properties. A subsequent source-parameter study of 2004–2009 earthquakes near the dam (Jiménez et al., 2015) found that grouping the events with a Self-Organizing Map (an unsupervised neural-network method that projects data onto a low-dimensional grid of clusters) relates them by magnitude and not by other considerations — again no induced-specific signature. No published three-dimensional correlation-dimension reanalysis of the Itoiz clusters and no validation of the Tsonis-style clustering coefficient as an induced-versus-tectonic discriminator is known; discrimination work has largely moved to waveform, spectral, moment-tensor and machine-learning methods, while case studies (e.g., Tretyak et al., 2024, for the Dnister reservoir; Kumar et al., 2013, for Koyna, where the b-value and correlation dimension are used as monitoring quantities rather than as induced-versus-tectonic discriminators) continue to rely on macroscopic correlates such as water-level correlation.

Problem Statement

Determine, with a reproducible catalog-based analysis, whether fractal and complex-network statistics of earthquake clusters discriminate reservoir-triggered seismicity from tectonic seismicity, taking the 2004 Itoiz sequence as the type case. Concretely: (1) recompute the correlation dimension D2D_2 of the clusters near the Itoiz dam — the 18 September 2004 reservoir-triggered cluster and the regional tectonic clusters — in three dimensions for those clusters for which depth-resolved (relocated) hypocentral locations exist (taking the 2004 sequence from its published double-difference relocation, Ruiz et al., 2006), stating the provenance, spatial and temporal coverage, magnitude range and completeness magnitude of every catalog used, while retaining the two-dimensional values for the remaining clusters, to establish whether the higher two-dimensional correlation dimension reported for the 2004 cluster is a genuine source signature or an artifact of the epicenter-only projection; (2) compute, under the 2009 and 2011 convention framework, the network statistics of these clusters — including the standard clustering coefficient and the neighborhood-based clustering coefficient for spatially embedded networks of Tsonis et al. (2008) — together with suitable tectonic control clusters (regional clusters, or clusters from comparable sites without induced seismicity), and test by appropriate null or generative models (ETAS-type simulations and spatial randomizations) whether any statistic separates the triggered cluster from the tectonic controls once mainshock magnitude, number of events and catalog completeness are controlled; and (3) state which statistics, if any, carry a reservoir-triggered signature, interpret the result in terms of the underlying triggering mechanism (pore-pressure-driven, diffusively clustered seismicity versus tectonic sequences), and identify which generative model class reproduces the observed network structure, or report that none of the tested model classes does. A rigorously supported negative answer — that no fractal or network statistic discriminates the triggered cluster after 3D geometry and magnitude/size confounders are controlled — also fully resolves the question. The common objective uniting these components is the statistical separability of triggered from tectonic cluster populations; the 3D recomputation, the statistic-by-statistic discrimination test, and the generative-model comparison are jointly the evidence needed to answer it. The long-term monitoring question (whether Itoiz seismicity persists over decades with cyclic episodes) is observational and outside this window.

References

  1. Jiménez, A., Tiampo, K. F., Posadas, A. M., Luzón, F., and Donner, R. (2009). Analysis of complex networks associated to seismic clusters near the Itoiz reservoir dam. The European Physical Journal Special Topics 174, 181–195. DOI 10.1140/epjst/e2009-01099-1, https://doi.org/10.1140/epjst/e2009-01099-1
  2. Durá-Gómez, I., and Talwani, P. (2010). Reservoir-induced seismicity associated with the Itoiz Reservoir, Spain: a case study. Geophysical Journal International 181, 343–356. DOI 10.1111/j.1365-246x.2009.04462.x, https://doi.org/10.1111/j.1365-246x.2009.04462.x
  3. Luzón, F., García-Jerez, A., Santoyo, M. A., and Sánchez-Sesma, F. J. (2010). Numerical modelling of pore pressure variations due to time varying loads using a hybrid technique: the case of the Itoiz reservoir (Northern Spain). Geophysical Journal International 180, 327–338. DOI 10.1111/j.1365-246x.2009.04408.x, https://doi.org/10.1111/j.1365-246x.2009.04408.x
  4. Santoyo, M. A., García-Jerez, A., and Luzón, F. (2010). A subsurface stress analysis and its possible relation with seismicity near the Itoiz Reservoir, Navarra, Northern Spain. Tectonophysics 482, 205–215. DOI 10.1016/j.tecto.2009.06.022, https://doi.org/10.1016/j.tecto.2009.06.022
  5. Jiménez, A., and Luzón, F. (2011). Weighted complex networks applied to seismicity: the Itoiz dam (Northern Spain). Nonlinear Processes in Geophysics 18, 477–487. DOI 10.5194/npg-18-477-2011, https://doi.org/10.5194/npg-18-477-2011
  6. Jiménez, A., García-García, J. M., Romacho, M. D., García-Jerez, A., and Luzón, F. (2015). Source Parameters of Earthquakes Recorded Near the Itoiz Dam (Northern Spain). Pure and Applied Geophysics 172, 3163–3177. DOI 10.1007/s00024-014-0883-y, https://doi.org/10.1007/s00024-014-0883-y
  7. Santoyo, M. A., Martínez-Garzón, P., García-Jerez, A., and Luzón, F. (2016). Surface dynamic deformation estimates from local seismicity: the Itoiz reservoir, Spain. Journal of Seismology 20, 1021–1039. DOI 10.1007/s10950-016-9578-4, https://doi.org/10.1007/s10950-016-9578-4
  8. Tsonis, A. A., Swanson, K. L., and Wang, G. (2008). Estimating the clustering coefficient in scale-free networks on lattices with local spatial correlation structure. Physica A: Statistical Mechanics and its Applications 387, 5287–5294. DOI 10.1016/j.physa.2008.05.048, https://doi.org/10.1016/j.physa.2008.05.048
  9. Gupta, H. K. (2022). Artificial Water Reservoir-Triggered Seismicity (RTS): Most Prominent Anthropogenic Seismicity. Surveys in Geophysics 43, 619–659. DOI 10.1007/s10712-021-09675-z, https://doi.org/10.1007/s10712-021-09675-z
  10. Kumar, A., Rai, S. S., Joshi, A., Mittal, H., Sachdeva, R., Kumar, R., and Ghangas, V. (2013). The b-value and fractal dimension of local seismicity around Koyna Dam (India). Earthquake Science 26, 99–105. DOI 10.1007/s11589-013-0008-1, https://doi.org/10.1007/s11589-013-0008-1
  11. Tretyak, K., Brusak, I., and Pronyshyn, R. (2024). Reservoir-triggered seismicity: case study of the Dnister Hydro Power Complex (Ukraine). Geofizicheskiy Zhurnal 46(1). DOI 10.24028/gj.v46i1.298659, https://doi.org/10.24028/gj.v46i1.298659
  12. Ruiz, M., Gaspà, O., Gallart, J., Díaz, J., Pulgar, J. A., García-Sansegundo, J., López-Fernández, C., and González-Cortina, J. M. (2006). Aftershocks series monitoring of the September 18, 2004 M=4.6 earthquake at the western Pyrenees: A case of reservoir-triggered seismicity? Tectonophysics 424, 223–243. DOI 10.1016/j.tecto.2006.03.037, https://doi.org/10.1016/j.tecto.2006.03.037