Seismicity of the Ibero-Maghrebian area is related to the plate motion of the Nubian and Eurasian plates, which is divergent in the Azores region and right-lateral to convergent eastward e.g. [8] and references therein [9] . Depending on the authors and the used methodology (magnetic anomalies and/or GPS data), the Nubia-Eurasia Eulerian pole (Eurasia fixed) is located between 18.2˚S and 21.6˚N and 15.9˚W and 30.8˚W [15] [16] . In contrast to the more or less linear Azores-Gibraltar segment of the Nubia-Eurasia plate boundary (Figure 1), the Rif-Betics-Alboran area is a complex mosaic of micro-blocks undergoing variable motion. The Rif and Betics are symmetrical thrust belts undergoing compressional and/or strike-slip regime, while the western Alboran Sea is an extensional (back-arc) basin [17] . The Al Hoceima area is located at the southern segment of a major NNE- SSW fault zone spanning the eastern Betics e.g. [18] , the Alboran Sea [19] and the Central Rif [20] . This shear

zone, called the Trans-Alboran Shear Zone (TASZ) [21] , may represent the eastern termination of the continental western Alboran basin and the eastward transition to the oceanic part (Figure 1). However, the tectonic framework is much more complex in the Alboran basin [22] . The description of the detailed geological history of this area e.g. [17] is beyond the scope of our paper, but knowledge of its present-day kinematics is necessary for understanding its seismic activity and assessing the seismic hazard.

Analyses of GPS data indicate that the direction of convergence is WNW-ESE (Figure 1) with a rate of 4.3 - 5.6 mm∙yr^‒1 [9] and references in their Figure 5 [23] -[25] . The regional model of [24] considers five blocks bounded by fault zones: 1) Iberia and the Atlantic segment north of the AGFZ; 2) Morocco and the Atlantic segment south of the AGFZ; 3) the Betics block; 4) the central Rif and western Alboran block, and 5) the eastern Alboran block. The faults bounding the Betics block are right-lateral, with the exception of the eastern fault, the TASZ, which is left-lateral. Southwards, the presumed faults bounding the central Rif block are left-lateral (the TASZ and the Rif front fault) with the exception of the western fault zone, which is right-lateral. Therefore, with respect to Nubia, a south-westward motion is observed in the central Rif and its foreland [20] [23] [24] [26] in relation to a SW-verging tectonic escape process observed from the surface [27] which may extend at depth [24] . The boundary between both kinematic zones is located along the TASZ, the continental segment of which is located west of Al Hoceima [24] . The amount of slip along this boundary is evaluated at 5.2 to 5.7 mm∙yr^‒1 [24] . Focal mechanism solutions and analysis of the subsequent state of stress in northern Morocco also indicate a NW-SE convergence [9] [28] -[36] in accordance with the kinematic models.

The city of Al Hoceima is located in the central Rif chain, which runs parallel to the northern coast of Morocco (Figure 2). The most conspicuous structures consist of southward-verging thrust sheets which are, from the upper (northernmost) to the lower (southernmost): 1) the Bokkoya (Palaeozoic and Mesozoic-Cenozoic); 2) the Tiziren unit (Middle Jurassic to Early Cretaceous carbonates and flysch series); and 3) the Ketama metamorphic unit (Cretaceous flysch and limestones). Volcanic rocks are represented by the middle Miocene Ras-Tarf andesites [37] [38] . Unconformable Miocene and younger deposits [39] are encountered in the lower Nekor graben, trending N-S, and in the Boudinar basin, east of Ras Tarf Cape [37] [40] . Offshore, the main structures are the Plio-Quaternary Bokkoya basin, which extends until the Alboran ridge [41] and the lower Nekor basin which is the marine continuation of the onshore one [22] [41] [42] .

The NE-SW left-lateral Nekor fault is the major structure in the area e.g. [43] ; however, it has remained seismically inactive since the onset of seismic recordings in Morocco [36] [44] [45] . Other important faults are (Figure 2): 1) The Imzouren (NNW-SSE) and Trougout (N-S) normal faults, which respectively delimit the western and eastern boundaries of the lower Nekor graben; new studies show that the Trougout fault has an oblique (normal/strike-slip) slip rate of 0.73 - 0.95 mm∙yr⁻¹ since 7 ka [46] , and can be followed offshore for about 7 km with a surface offset [42] ; authors [46] calculate a return period for major earthquakes of about 3 ka, which is a relatively long period; 2) The Jbel Hammam fault system (NNW-SSE), consisting of several normal faults with trace lengths of 20 km; 3) The Rouadi fault, a normal fault trending NNE-SSW over 15 km, the eastern block being downthrown; 4) The Bousekkour-Arhbal strike-slip fault (Figure 2), trending NNE-SSW, which crosscuts the whole Bokkoya unit with a sinistral displacement [47] . The latter fault is thought to have been active during the 1994 seismic crisis [11] , but radar interferometry and field studies suggest that the rupture did not reach the surface [48] . However, a recent oceanographic survey [42] has shown that the fault can be followed offshore for about 4 km where it is expressed by an east-dipping normal fault plane (50 m offset of “basement”) with a clear 5 m-surface fault scarp. Other offshore faults are those delimiting and affecting the Nekor graben [41] , and especially the NNE-SSW striking, 4 km long, Bokkoya normal fault whose throw is about 50 m [42] .

Recent multidisciplinary studies e.g. [49] [50] show that the thickness of the crust in northern Morocco decreases northwards from 30 km in the Rif to 12 km in the central Alboran basin, whereas the depth of the lithosphere-asthenosphere boundary in the area decreases from 230 - 250 km in the west, to about 70 km in the Alboran Sea, the isobaths running roughly N-S. Crustal thinning is accompanied by the development of a large thermal anomaly in the Alboran basin [51] . Around Al Hoceima, the deep structure was mainly studied using seismic tomography [36] [52] -[54] . The results obtained by these authors appear strongly related to the network configuration, and especially to the distance between recording stations; for instance, P-wave velocity maps obtained by [52] show in the layer 5 - 15 km a large circular (~100 km diameter), high velocity zone south-east of Al Hoceima, which extends at depth in the layer 15 - 30 km, contiguous to a high velocity zone. The boundary of both zones runs NNW-SSE across the lower Nekor graben. Authors [53] inferred low velocity zones within the layer 5 km beneath the Nekor graben and within the lower crust in the continental area, whereas the most important high velocity zone is located south of Al Hoceima city within the layer 10 km. Tomographic maps of [54] show a low-velocity zone at 15 km depth near Al Hoceima; however, the resolution is not appropriate for determining better orientations. Instead, the higher resolution tomography obtained by [36] shows better constrained velocity zones at 5 - 15 km depth, with a determinable orientation.

Historically, the Al Hoceima area suffered numerous strong earthquakes as shown by the earliest earthquake catalogues [44] [55] . The best documented events were recorded in 1522, 1624, 1790-1791, 1795 and 1800-1803 (M~6?), 1848, 1910-1914. Since the onset of the instrumental period, the early and following authors e.g. [44] [56] -[58] observed that seismicity is largely scattered on the scale of the Betic-Rif Arc. On the base of the data from NEIC, the epicenters of events with magnitudes > 3 from 1^st January 1960 to 3 October 2014 in the Al Hoceima area are shown in Figure 3. Most of the earthquakes are shallow, and their magnitudes are generally moderate, except for the 1994 and 2004 events. The epicenters are clearly aligned along a NNE-SSW trend parallel to the TASZ; however, on the regional scale, epicenters of some well-constrained micro-earthquakes and aftershocks of large events are clearly aligned along NNE-SSW, NNW-SSE and WNW-ESE trends [11] [36] [41] [59] -[61] . The seismic flux (the seismic moment released per year and per unit area) calculated from a 104 year period is one of the largest in the Western Mediterranean, reaching more than 10¹⁹ dyne∙cm∙km^‒2∙yr^‒1 [9] .

Since the 1970’s and until 2007, at least 100 focal mechanisms were determined for large and moderate earthquakes (M > 3.5) of the Al Hoceima area using first motion polarities of P-waves, waveform analysis and moment tensor inversion [62] . Among these, a large number of solutions were determined by seismological agencies (IGN, IAG, MED, INGV) for the aftershocks of the 2004 earthquake. Most fault-plane solutions, including

the largest aftershocks of the two major earthquakes, correspond to strike-slip faulting with a normal component and to normal faulting (Figure 4). A few solutions correspond to reverse faulting or to strike-slip motions with reverse component. The P axes are distributed in the NW and SE quadrants with variable plunge. The solutions obtained for interseismic periods [28] [45] [63] [64] , also indicate that the dominant regime is strike-slip/normal, with a general lack of pure reverse mechanisms. The P axes are distributed along a NW-SE great circle, with largely variable plunges, whereas T axes are NE-SW to ESE-WNW with a shallow plunge [28] [65] .

From GPS data listed by [23] [24] , the general displacements with respect to fixed Africa calculated from the East and North components, are variable: east of the TASZ, permanent station MELI shows a SSE (N173) small motion (1.61 mm∙yr⁻¹) during the period 1999-2007, while nearby temporary station MDAR shows a NNE (N20) moderate motion (2.24 mm∙yr⁻¹) during the period 1999-2004. West of the trace of the TASZ, the motion at temporary stations BBFH (period 1999-2004) and KTMA (period 1999-2005) is SSW (N203 and N229 respectively), and the linear velocity is significant (3.31 and 3.7 mm∙yr⁻¹ respectively).

Coulomb stress change is best expressed by the equation e.g. [4] :

where, and are, respectively, the coseismic changes in Coulomb stress, shear stress and normal stress, and is the effective coefficient of friction, e.g. [66] and references therein, which corresponds to where B is Skempton’s coefficient, comprised between 0.5 and 0.9.

Modeling was carried out on the region of coordinates 34.8˚N - 35.7˚N × 3.3˚W - 4.8˚W with the help of Coulomb 3.2 software [3] . In a first step, we considered the fault related to the 1994 earthquake as the source fault and that of the 2004 shock as the receiver fault. In a second step, we considered the effects of the 2004 earthquake on the nearby region. CSC was resolved both on optimal strike-slip planes and on chosen particular planes parallel to the considered receiver faults. A discussion on this point can be found in [67] . The following subsections expose in detail the model constraints which are necessary as input data to the construction of the model, such as the regional state of stress and the source parameters including the location of the epicenter, the nucleation depth, length and width of the rupture and aftershocks location.

The model constraints are based on a large database collected after the 1994 and 2004 earthquakes by numerous

international research teams that carried out studies on the source parameters of both earthquakes, using various methods such as classical seismic and micro-seismic surveys [11] [12] [36] [41] [68] [69] , seismic wave analyses [33] [36] [70] , radar interferometry [13] [61] [71] -[73] , GPS measurements and SPOT images interpretation [61] [73] and seismic tomography [36] .

Figure 6(A) illustrates the CSC distribution for optimal strike-slip planes after 0.7 m left-lateral slip on fault 1. It clearly shows an increase (≥5 bar, saturation value) in the magnitude of the stresses at the tips of the fault, and a decrease (≤‒5 bar, saturation value) across the plane along a NE-SW oriented zone. Most of the epicenters of the aftershocks taken from [11] (for clarity, only the 80 first ones recorded from 27 to 30 May are shown) plot in the positive CSC zone oriented at a small anti-clockwise angle from the fault, but a large number plot in a negative CSC zone. The receiver fault undergoes positive CSC mainly at its centre.

In the case of CSC resolved on the N20E (=N200E) vertical planes oriented parallel to the main fault (Figure 6(B)), all aftershock epicenters fall within the large negative CSC area situated on both fault blocks. However, a small positive CSC zone affects the receiver fault at its eastern tip.

Finally, when CSC is resolved for specified N300E (N120E) striking dextral strike-slip planes (Figure 6(C)), parallel to the receiver fault, almost-all aftershocks fall in a negative CSC zone. The receiver fault also undergoes positive CSC all along its eastern segment.

Figure 7(A) shows the Coulomb stress distribution after 1 m right-lateral slip on Fault 2 resolved on optimal strike-slip planes. An increase (≥5 bar, saturation value) in the magnitude of the stresses is observed at the tips of the fault, whereas a decrease (≤‒5 bar, saturation value) can be seen across the plane along a NE-SW direction.

The aftershock epicenters determined by [36] (only the first 300 ones from 29 March to 6 April were plotted for clarity) fall within a Coulomb stress increase area (Figure 7(A)). However, the cluster at 35.15˚N × 4.05˚W falls in a shadow zone, which does not match the Coulomb stress change. No changes are observed when the CSC is resolved on planes which are parallel to the main fault (Figure 7(B)), and the cluster at 35.15˚N × 4.05˚W remains in the same CSC decrease zone.

When CSC is resolved on sinistral planes striking N20E, no important changes are observed with respect to the previous panel (Figure 7C). Finally, CSC resolved on normal fault planes such as found by [36] at the

eastern aftershock cluster (Figure 7(D)) shows that the aftershocks also fall in the increase zone, which is in conformity with aftershock triggering.

As indicated in section 1, CSC modeling has already been successfully used for studying local and regional stress changes after major earthquakes, as well as for predicting the loci of occurrence of future ones in several seismic areas in the world. In the present study, CSC modeling was applied to the Al Hoceima region, which is an area of distributed moderate seismicity with relatively short faults.

From our results, CSC modeling was successful in helping understand several observations:

1) CSC after the 1994 earthquake may have led to a stress increase along the eastern segment of the WNW- ESE fault responsible for the 2004 earthquake; therefore, as for the Rissani twin earthquakes of 1992 [14] , the model provides a consistent explanation for the space relationship between both events.

2) Aftershock distribution and reactivation of optimally oriented planes after both earthquakes are also successfully accounted for by CSC modeling.

However, although the geometrical parameters seem to be relatively well constrained on the base of the data taken from the numerous studies of the Al Hoceima earthquakes, there are intrinsic parameters which remain

less constrained. This is the case of the variable input values suggested by the software such as the friction coefficient, the regional stress magnitude and its variability with depth, which are beyond the scope of the present article, focused on regional aspects.

One of the major problems for constraining the relationship between earthquakes and faults in the Central Rif is that a few ruptures were observed during both events [11] [36] [48] [68] , although rupture has been proved to reach the surface offshore [42] , so the faults remained blind and their pattern was only delineated afterwards using focal mechanism solutions, aftershock distribution and radar interferometry. Indeed, the WNW-ESE fault responsible for the 2004 earthquake had no previous morphological or deep features which could have helped mapping it.

The post-seismic field survey of the 2004 earthquake [48] suggests that there seems to be some vertical partitioning of deformation, with thrust sheets at the surface and strike-slip faults at depth. However, this is not true everywhere, because it is obvious that some faults reach the surface, such as the Trougout-Bou Haddoud, Rouadi, Bousekkour-Aghbal, Bokkoya offshore and Boudinar normal faults which have been and still are potentially seismogenic [42] [46] .

Finally, at depth, tomographic maps obtained by [54] show a low-velocity zone at 15 km depth near Al Hoceima; however, the resolution of the maps is not appropriate for determining accurate orientations. Instead, the higher resolution tomography obtained by [36] clearly shows that the aftershock hypocentres are located within high velocity zones at 5 - 15 km depth, whose orientation is in conformity with the chosen faults, which may not be an artifact of the used methodology.

In order to predict which fault planes are most loaded by positive CSC and may be reactivated, we plotted the main known faults of the Al Hoceima area and run the program with Coulomb stress resolved onto planes having the same strike than the major faults recognized in the area.

Figures 7(D) and Figure 8 show CSC after the 2004 earthquake resolved onto the strikes of the regional fault planes located around the epicenter. The figures show that the faults around Al Hoceima city are not affected by an increase in stress, with the exception of the Nekor fault, the south-western segment of which is affected by an increase of about 7 bar. However, it should be emphasized that most stress was probably released through the very numerous aftershocks.

The time interval for the occurrence of the next event is for the moment difficult to determine because it depends on the tectonic load related to the plate motion of Nubia with respect to Eurasia, but also on the regional motion of blocks across the TASZ and the accommodation of motion by faults. Estimations should be possible when a larger and more precise GPS database will be available in the next years.

Anyway, CSC modeling is a tool which is, at a local scale (10,000 km²), an efficient alternative method to probabilistic and deterministic approaches of seismic hazard in the area [86] [87] .