As indicated above, the geochemistry of Miocene (6-12 Ma) volcanic rocks from Sites 977 and 978 can best be explained in the context of subduction. Based on available geochemical data, rocks from the volcanic belt in Figure 1 have similar compositions to the samples analyzed in this study, as is illustrated by the comparison with Cabo de Gata (e.g., Toscani et al., 1990; Fernandez Soler, 1992; see also Fig. 6). The major- and trace-element chemistry of Miocene tholeiitic, calc-alkaline, potassic and lamproitic rocks in the Alboran Region have been interpreted to reflect subduction in the region (Araña and Vegas, 1974; Bellon and Brousse, 1977; Delarue and Brousse, 1974; Foley et al., 1987; Venturelli et al., 1988; Nelson et al., 1986; Torres-Roldán et al., 1986; Venturelli et al., 1984). Therefore, the 400 km northeast-trending volcanic zone may represent a volcanic arc formed by subduction of ocean crust during the Miocene.
Additional support for northeast-trending subduction comes from the depth of earthquake epicenters and seismic tomography. The epicenters of two earthquakes in 1954 and 1973, located at 630 km and 640 km, respectively, beneath southern Spain, are interpreted to reflect the presence of a subducted slab at the base of the upper mantle (e.g., Isacks and Molnar, 1971; Chung and Kanamori, 1976; Udías et al., 1976; Grimison and Chen, 1986). The calculated east-west-trending principal stress axis (Udías et al., 1976) is consistent with northwest-dipping subduction (Blanco and Spakman, 1993; Zeck, 1996). Delay-time tomography shows that the presence of a positive P-wave velocity anomaly between 200 and 700 km depths resembles a cold, subducted, lithospheric slab (Blanco and Spakman, 1993). The southwest-northeast strike and ~600 km length of this anomaly are similar to that of the Miocene volcanic belt (see Fig. 1). Although oceanic lithosphere is generally not thicker than ~100 km, the 250-km thickness (and 600-km length) of the P-wave anomaly no doubt in part reflects thermal equilibration of the surrounding warmer mantle with the colder slab (Zeck, 1996) and the resolution of the tomographic model. Data from the new seismic network in Morocco show a similar P-wave anomaly beneath northeastern Morocco, the Alboran, and southeastern Spain between 150 and 350 km (Seber et al., 1996a).
Although it is difficult to constrain the onset of subduction, the oldest ages of volcanism in the Alboran region (20-25 Ma) suggest that it may have occurred near the Oligocene/Miocene boundary (Fig. 2). This boundary is associated with a major change in the plate kinematics in the North Atlantic region. Not only did movement within the Pyrenees and Biscay cease, but the European-African plate boundary shifted to the Azores-Alboran fracture zone (Roest and Srivastava, 1991). The Oligocene/Miocene boundary also coincides with the beginning of subsidence to form the Alboran Basin and exhumation of metamorphic nappe complexes in southern Spain (Comas et al., 1992; Zeck et al., 1992; Zeck, 1996). These events suggest that the Alboran Basin may have been formed by back-arc processes (Torres-Roldán et al., 1986) and that the Alboran back-arc basin (and possibly other Mediterranean back-arc basins) distinguishes itself from Pacific-style back-arc basins primarily because subduction occurs beneath continental crust under compression from continent-continent collisions.
The lower velocities above 200 km in the tomographic model of Blanco and Spakman (1993) and the lack of seismicity under the Alboran and adjacent regions of southern Spain and northern Morocco (from 2° to 8°W) between depths of 120 and 180 km (Seber et al., 1996b; Fig. 2) are consistent with the proposal that the subducted lithospheric slab is detached (Blanco and Spakman, 1993; Zeck, 1996). Warmer and lighter asthenospheric material from greater depth is likely to flow into the void formed by the detachment and sinking of a cold lithospheric slab. Decompression melting of in-flowing asthenosphere could result in the formation of low-degree, alkali-basalt melts, which suggests that detachment occurred 5-6 Ma or slightly before, when the composition of volcanic rocks changed from primarily calc-alkaline to alkaline (Fig. 2).
Detachment could also explain several other phenomena that occurred at or near the Miocene/Pliocene boundary. Whereas deformation and uplift in the Alboran region moved westward at the end of the Miocene and beginning of the Pliocene, subsidence in the Alboran Basin moved to the east and southeast (i.e., toward and east of the Alboran Ridge; Docherty and Banda, 1995). Release of the downward pull from the subducting slab on the bent portion of the slab and in-flow of warmer and lighter asthenosphere into the gap above the slab could cause uplift above the sinking slab (Wortel and Spakman, 1992; Zeck et al., 1992; Blanco and Spakman, 1993). Temporary uplift of the Gibraltar arc on the western margin of the Alboran Sea 5-6 m.y. ago may in part be responsible for closing the Strait of Gibraltar, which resulted in the Messinian salinity crises in the Mediterranean. Subsidence of the eastern portion of the Alboran in the early Pliocene, however, is difficult to explain unless part of the lithospheric mantle beneath the eastern Alboran was removed or delaminated (Docherty and Banda, 1995). It is possible that during the detachment process a piece of the lithospheric mantle beneath the eastern Alboran may have been removed by the sinking slab. Finally, detachment of the lithospheric slab, or subsequent mantle upwelling, might also trigger delamination of the lithospheric mantle on the overriding plate, which could explain the gap in seismic activity, low Pn-wave velocities, and high seismic-wave attenuation at depths of 20-60 km presently beneath the Alboran Sea and Gibraltar Arc (Seber et al., 1996b).
Although northwest-dipping subduction of oceanic crust in the Miocene beneath the Alboran, followed by slab detachment and lithospheric delamination in the Pliocene, does not provide a unique solution for the geological, geophysical, tectonic, and structural data (e.g., see "Introduction" in Comas, Zahn, Klaus, et al., 1996), it shows that the available data can at least be interpreted in the context of such a model.