On the interaction of the North Andes plate with the Caribbean and South American plates in northwestern South America from GPS geodesy and seismic data

Omar J. Pérez,1 Steven G. Wesnousky,2 Roberto De La Rosa,1 Julio Márquez,1 Redescal Uzcátegui,1 Christian Quintero,1 Luis Liberal,3 Héctor Mora-Páez4 and Walter Szeliga5 1Department of Earth Sciences, Simón Bolı́var University, Caracas 1080A, Venezuela. E-mail: ojperez@usb.ve 2Center for Neotectonic Studies, University of Nevada, Reno, NV 89557, USA 3Department of Geodesy, Central University of Venezuela, Caracas 1050, Venezuela 4Sptial Geodetic Research Group, Colombian Geological Survey, Bogota 111321, Colombia 5Department of Geological Sciences, Central Washington University, Ellensburg, WA 98926, USA

. Tectonic setting of the southern Caribbean and adjacent plates (compiled from Silver et al. 1975;Pérez & Aggarwal 1981;Kellogg & Vega 1995;Pérez et al. 1997b;Trenkamp et al. 2002;Bird 2003;Gorney et al. 2007;Magnani et al. 2009;Weber et al. 2009;Camacho et al. 2010;Kroehler et al. 2011). Main plate boundaries are shown by thick lines. Oceanic segments with teeth indicate subduction with teeth on overthrust plate. The North Andes plate (Trenkamp et al. 2002;Bird 2003) is addressed as a block by several authors (Pennington 1981;. Plate motions (arrows) are relative to stable South America (DeMets et al. 1990;Pérez et al. 2001;Weber et al. 2001;Trenkamp et al. 2002;Pérez et al. 2011).  (Rial 1978;Pérez et al. 1997a;Choy 2001;Global CMT Catalog). They show the dextral dominant nature of the Boconó, Morrocoy and San Sebastián faults. triple junction, complexity arises because relative motion between the Caribbean and South American plates is separated by the North Andes plate and divided, as we will show in this paper, between the major northwest striking Morrocoy and southwest striking Boconó fault systems, respectively ( Fig. 1; Pérez et al. 1997a;Pérez & Mendoza 1998;Bird 2003), with some motion possibly taken up by other faults such as the Oca and Santa Marta faults (Symithe et al. 2015). Only along the San Sebastián-El Pilar fault system does motion occur directly between the Caribbean and South American plates. Yet even here, the fault system accommodating the motion and the GPS velocity field reflecting that motion is distributed across an 80-120 km wide zone (Pérez et al. 2001;Jouanne et al. 2011;Reinoza et al. 2015), including other faults such as the La Victoria fault.
A number of large earthquakes have occurred on the main branches of the triple junction since 1640 (Fig. 2). These include a sequence of two events in 1812 (M w 7.4 and 7.1) that took place within about 1 hr and broke across the triple junction to rupture portions of both the Boconó and San Sebastián faults (Fielder 1961;Altez 1998;Choy et al. 2010). In 1900, an M w 7.6 shock (Pacheco & Sykes 1992) broke the eastern segment of the San Sebastián fault, as recognized from both intensity observations (Centeno-Graű 1900;  Sievers 1905;Choy et al. 2010) and marine geophysics studies (Colón et al. 2015). A number of moderate (M w ∼6+) yet destructive dextral strike-slip shocks are documented (  (Pérez et al. 1997b). The damaging effects of the latter events are described in detail by Centeno-Graű (1940), Grases (1980) and Ernst (1878). Focal mechanisms of small earthquakes along La Victoria fault also show dextral motion along strike (Pérez et al. 1997b). Microseismicity (M w < 4) has been reported on each of the major branches of the triple junction ( Fig. 3; e.g. Pérez et al. 1997b;Pérez & Mendoza 1998). No significant earthquake activity is reported along La Tortuga fault (Fig. 2) or along the Oca and Santa Marta faults ( Fig. 1) since the colony times. The known seismicity record along the Caribbean subduction in northern Colombia ( Fig. 1) is limited to a small quantity of teleseismically located events and microseismicity recorded since 1993.
To check and quantify the style and sense of motion of the various active faults that define the branches of the triple junction and ultimately better define the tectonics of northwestern South America, we here conduct an integrated analysis of a series of new GPS observations collected in north-central and northwestern Venezuela and Colombia during the period from 1999 to 2015 and seismological data, including well located sets of microearthquakes reported by the Colombian (1993) and Venezuelan (2003-2012 broad-band national seismological arrays operated by the Colombian Geological Survey (Colombian Geological Survey 2016) and the Venezuelan Foundation for Seismological Research (FUNVISIS 2016), respectively.
In the next sections, we describe the GPS data acquisition and analysis methods and the velocity vectors and field subsequently obtained. Next, we present details of the models that were used to emulate the velocity field across the San Sebastián and La Victoria faults in north-central Venezuela. Then, we analyse the seismicity and the velocity field observed in western Venezuela, which encompasses the Bocono and Morrocoy faults and northern Colombia where the Caribbean plate is being subducted to the south-east beneath the North Andes plate (Fig. 1) along the South Caribbean Marginal fault (Toto & Kellogg 1992;Pérez et al. 1997b;Trenkamp et al. 2002). Then, we performed a kinematic block modelling for the southern Caribbean plate boundary to further analyse and quantify the regional GPS velocity field and rate and sense of motion along the major active faults that mark the plate boundary. For the analysis we used the modelling approach and associated software 'Blocks' of . Finally, the implications of the geodetic findings for regional tectonics and for the quantification of seismic hazards in the region are discussed.

G P S M E T H O D S A N D A N A LY S I S
GPS data in Venezuela and adjacent regions were obtained using dual frequency receivers during a series of 5-15 d and 12-24 hr surveying sessions between 1994 and March 2015. During this whole period the monument CANOA, located deep within stable South America and Maracaibo (MARA; Fig. 1), as well as USB0 and USB1 (replacing USB0 in 2006; Fig. 2) were occupied 7, 10, 7 and 7 times, respectively since 1999, sites CASI and ALTA in 1999CHUR and AMUA in 1994, 1999and 2014ROQU and CALA in 1994, 1999ARUB in 1999 and2012;ROSA andMONA in 2000 and2014. We also used data collected since 2012 at sites CN38 and CN40, which belong to the COCOnet permanent GPS array as well as data collected in 1994 and 1999 by Pérez et al. (2001) and Pérez et al. (2011) to re-calculate several geodetic vectors reported in those papers using newly acquired data in 2014 at those sites.
In general all Venezuelan sites were occupied simultaneously with CANOA, MARA ( Fig. 1) and USB0 (or USB1, Fig. 2) or all of them. We used 15 or 30 s sampling intervals and an elevation mask of 15 • . Site's positions for each epoch were obtained using simultaneous observations at 8-11 permanent GPS stations of the International GNSS Service located in the Americas and the Caribbean and/or directly relative to CANOA as a fixed reference point to check for consistency. We used the Bernese GPS v. 5.0 (Dach et al. 2007) software package to process the data, following the procedures described by these and other authors (e.g. Jouanne et al. 2011;Reinoza et al. 2015). Additionally, the Space Geodesy Research Group that runs the GeoRED Project in Colombia (Mora-Páez et al. 2016a,b) provided us with the velocity vectors they obtained for six GPS sites in Colombia territory that they processed using the GIPSY-OASIS II software v. 6.3. The corresponding velocity vectors at all sites  used in this study, calculated in ITRF2008 Epoch 2010 (Altamimi et al. 2011), are given numerically in Table 1. The velocities were then calculated in the South America reference frame using the pole of rotation between ITRF2008 and South America provided by Altamimi et al. (2012). They are graphically shown in Figs 3, which were re-calculated in the ITRF2008 frame (Mora-Páez et al. 2016a).

Slip-partitioning associated with the San Sebastián and La Victoria fault zones
In this section, we combine GPS data with simple elastic strain accumulation models and seismic evidence to constrain slip rate estimates along these two major dextral, easterly striking fault zones. The two faults ( Fig. 3) form the Caribbean-South American main plate boundary east of ∼68 • W. Our goal is to quantify how the ∼19.6 mm yr −1 in the N85 • E direction of Caribbean-South American rate of relative motion, indicated by the corresponding pole of rotation (Table 2) derived from GPS geodesy (Pérez et al. 2001), is distributed between these fault zones. N85 • E is also the strike direction of the San Sebastián fault. Fig. 3 shows the seismicity (moment magnitude M w ≥ 2.6, circles) reported by FUNVISIS (2016) for the period 2003-2012 together with the GPS-derived velocity vectors (arrows) relative to stable South America, in north-central Venezuela. About 98.5 per cent of the seismic events reported are microearthquakes in the M w range of 2.6-4.4. Only minor activity is observed in the inner regions of the Caribbean, though a series of seismological stations are located in Venezuelan islands and along the northern coast of the country (see http://www.funvisis.gob.ve/). FUNVISIS catalog shows a fairly constant rate of listing for events with M w ≥ 2.6 during the time span under consideration.
East of ∼68 • W the San Sebastián and La Victoria faults coexist. These fault zones are nearly parallel (Fig. 3), making an angle of ∼5 • -10 • and are separated by a distance that ranges from ∼30-50 km. Recent marine geophysics studies show that at some places the San Sebastián fault is multibranched and the fault zone reaches a width of 10-12 km (Colón et al. 2015). As found out by Pérez et al. (1997b) from a different set of earlier earthquakes, most of the seismic activity in the region straddles both fault zones, occurs in general at focal depths shallower than ∼14 km and focal mechanisms commonly show dextral motion on easterly oriented vertical fault planes (Figs 2 and 4). These observations show that these two fault zones accommodate most of the relative motion between the Caribbean and South American plates in the region.  Fig. 4 with blue stars. Many benchmarks were vandalized since the first occupations in 1999 and thus the number of data points remains relatively few. Nonetheless, the velocities of USB0 and USB1 are very well constrained since all together they have been occupied 14 times since 1999 and their location between the two faults provides strong constraints on the slip partitioning we next describe.
The field (Fig. 5) across these faults is confined to a shear zone <120 km wide and most of the surface deformation is contained within a 70-80 km wide zone encompassing the San Sebastián and La Victoria faults. The dashed line in Fig. 5 is the elastic modelled surface velocities (Okada 1985;Dixon et al. 1998) for a single, N85 • E oriented vertical fault taken up the full Caribbean-South American relative motion (∼19.6 mm yr −1 ). The solid line is in contrast calculated for a system of two subparallel vertical faults (i.e. The La Victoria and San Sebastián faults) separated by a distance of 45 km, in both cases with dextral shear below a locking depth of 14 Km. The two-fault model provides a good fit for the observed data, and corresponds to a dextral shear of 2.6 ± 0.4 mm yr −1 for La Victoria fault and 17.0 ± 0.5 mm yr −1 for San Sebastián fault, below a locking depth of 14 ± 3 km.
The modelling result does not preclude the possibility of some portion of slip modelled on the San Sebastián fault being accommodated on other faults to the north such as the easterly striking La Tortuga fault (Fig. 2). One could, for example, allow up to ∼3 mm yr −1 of slip along La Tortuga fault by decreasing a similar amount of modelled slip on the San Sebastián fault, without violating the observations. We hesitate to argue for this latter possibility because both the seismic record and marine geophysics studies (Colón et al. 2015) do not support the existence of the La Tortuga fault along the entire north-central coast of Venezuela. Similarly, it is possible that the plate boundary may include a third fault farther north along the easterly striking section of the South Caribbean Deformed Belt in the off-shore area (Fig. 1), with a small and probably negligible amount of slip on that portion of the structure (Symithe et al. 2015), expressed by an actively deforming accretionary prism (Gorney et al. 2007;Clark et al. 2008;Magnani et al. 2009;Kroehler et al. 2011). Finally, based on GPS investigations there is general agreement (Pérez et al. 2001;Clark et al. 2008;Jouanne et al. 2011;Reinoza et al. 2015;Symithe et al. 2015) in that east of ∼65.5 • W in eastern Venezuela (Figs 1 and 2) the easterly striking El Pilar fault zone is the main strand of the Caribbean-South America plate boundary, taking up most of the relative motion between the two plates. In our interpretation, both the San Sebastián and La Victoria faults merge around 65.5 • W with the El Pilar fault zone so that their motion is absorbed into the single El Pilar strand.

The interaction of the North Andes plate with the Caribbean and the South American plates in western Venezuela and northern Colombia
Numerous authors (e.g. Dewey 1972;Pennington 1981;Kellogg & Bonini 1982;Kellogg 1984;Toto & Kellogg 1992;van der Hilst & Mann 1994;Malavé & Suárez 1995;Pérez et al. 1997b;Pérez & Mendoza 1998;Trenkamp et al. 2002;Pérez et al. 2011;Bernal-Olaya et al. 2015;Chiarabba et al. 2015;Syracuse et al. 2016;Mora-Páez et al. 2016b) have studied the seismicity and tectonics of northwestern South America. There is a general agreement that along the northwestern tip of the continent there exists a northeasttrending slab of Caribbean plate that is subducting along the South Caribbean Marginal fault to the southeast (Figs 1 and 6) beneath the  North Andes plate (Toto & Kellogg 1992;Trenkamp et al. 2002;Bird 2003). Here we shall re-visit the seismicity of the region using teleseismic locations (m b ≥ 4.5) reported in the revised catalogues of the ISC for the period 1980-2013, as well as microearthquake locations (2.5 < m b < 4.5) reported in Colombian territory by the Colombian Geological Survey (Colombian Geological Survey 2016) for the period 1993-2015 and by FUNVISIS (2016) for the period 2003-2012 in Venezuelan territory, followed by the analysis of GPS-derived velocity vectors and their tectonic implications.
The epicentral locations are plotted as a function of increasing depth in Fig. 6. The distribution of seismicity defines an inclined northeast striking slab dipping southeasterly in northwestern Venezuela and northern Colombia. The top of the interpreted slab is defined by the series of depth contours (dotted lines) shown in the figure. These contours illustrate the northeast strike and southeast dip of the Benniof zone that defines the subduction of the Caribbean slab along the South Caribbean Marginal fault beneath South America. Three northwest-southeast vertical cross-sections (1-3) are shown and described in Fig. 7 and its caption. Each of them contains the earthquake hypocentres (dots) located within the rectangles 1-3 shown in Fig. 6. The horizontal projection of the dip of the subducted slab as manifest by the seismicity (Fig. 6) is oblique to the calculated slip vector (Pérez et al. 2001;Weber et al. 2001) of the Caribbean plate relative to South America at 12 • N, 77 • W (broken arrow in Figs 6 and 8). It is also observed in Fig. 6 that seismicity generally terminates and does not occur northeast of the Morrocoy fault zone, an observation that led Pérez et al. (1997b) to conclude that this fault zone marks the northeastern end of the North Andes plate. Fig. 8 shows a series of GPS velocity vectors (arrows) projected relative to the South American plate reported by several authors (Trenkamp et al. 2002;Pérez et al. 2011;Mora-Páez et al. 2016a,b; this study) at sites within the northernmost regions of the North Andes plate and adjacent Caribbean and South American areas. The GPS vectors at sites located in the northern inner regions of the North Andes plate (marked with blue stars in Fig. 8) show an average magnitude of 15.0 ± 1.0 mm yr −1 and are easterly oriented (average Az 80 • ± 6 • ), which for reference is the motion of site MARA. The 19.6 mm yr −1 relative motion of the Caribbean plate with respect to the South American plate, shown by a dashed arrow at two places in Fig. 8, is calculated from the corresponding pole of rotation provided by Pérez et al. (2001). To shift the reference frame to the North Andes plate we subtract the MARA vector from all other plotted vectors in Fig. 8. The result is shown in Fig. 9 and along with Fig. 8 serves to illustrate the rate and sense of displacements occurring at the boundaries of the North Andes plate.
It is observed with the vectors in Fig. 9 that the Caribbean plate is subducting at rate of ∼5-7 mm yr −1 beneath the North Andes plate and that motion along each the Morrocoy and Boconó faults boundaries is right-lateral. As well, there is an obliquity of motion along the Boconó and Morrocoy boundaries that predicts each to have a component of transpression and transtension, respectively. Thus, the Morrocoy fault can be interpreted as a continental rift apart boundary, as pointed out by Bird (2003), at the same time being a dextral shear zone that transforms the Caribbean subduction in northwestern South America into pure east-west strike-slip motion along the easterly oriented San Sebastián fault in north-central Venezuela.
Similarly, it is observed in Fig. 9 that South America is moving about 15 mm yr −1 at an azimuth of 260 • relative to the North Andes plate. Assuming the relative motion is mostly accommodated by deformation along trend of the Boconó fault system, the GPS vectors and model in Fig. 9 indicate about 9-11 mm yr −1 of dextral motion and 2-5 mm yr −1 of contraction across the fault, consistent with the geodetic results reported by Pérez et al. (2011;see their figs 3 and 4) and Symithe et al. (2015). The interaction between the North Andes and South American plates along the northeast striking Boconó fault zone in the Venezuelan Andes has been studied using seismic and geomorphological data as well as GPS-derived velocity vectors (e.g. Pérez et al. 1997bPérez et al. , 2011Pérez & Mendoza 1998;Wesnousky et al. 2012;Guzmán et al. 2013;Symithe et al. 2015). Our additional GPS vectors confirm their main results which indicate that due to the obliqueness of the Boconó fault relative to the easterly oriented slip of the North Andean plate relative to South America, this slip is partitioned into 9-11 mm yr −1 of dextral motion in the far field of the Boconó fault and a compressive regime perpendicular to the Andes that results in 2-5 mm yr −1 of horizontal shortening across the ranges and seismically active thrust faulting along the foothills (Wesnousky et al. 2012).
Several faults with slip rates estimated on the order 2 mm yr −1 or less occur within the northern tip of the North Andes plate and may also accommodate a small fraction of the Caribbean-South America relative motion (Audemard 1996;Audemard & Audemard 2002;Symithe et al. 2015). The Oca and Santa Marta faults (Figs 8  and 9) are the longest and most prominent of these slower slipping faults (Audemard & Audemard 2002;Audemard et al. 2005;Symithe et al. 2015). These faults have not shown the occurrence of significant earthquake activity since the colony times but palaeoseismicity studies show they have generated earthquakes in recent (Quaternary) times (Audemard 1996;Idárraga-García & Romero 2010). In the next section, we perform a kinematic block motion analysis aimed to better quantify the rate of motion taking place along each of the main active faults of the southern Caribbean plate boundary, with emphasis in our study area, that is, the northwestern tip of South America, including the Morrocoy, Oca and Santa Marta faults.

K I N E M AT I C B L O C K M O D E L L I N G
We use GPS observations in conjunction with the linear block theory and associated code 'Blocks' presented in  to estimate slip rates on the major active faults and partitioning of deformation occurring along the southern Caribbean plate. The kinematic block modelling approach is used to study regional GPS velocity fields and quantify the rates of motions along locked or partially locked active faults that form a plate or block boundary (e.g. McCaffrey 2002; Meade et al. 2002;Reilinger et al. 2006;Loveless & Meade 2010;Saria et al. 2014;Symithe et al. 2015). It can also be used to confirm or reject a proposed fault boundary.
To perform the modelling we assume for analysis the plate boundary geometry provided by Symithe et al. (2015) for the Lesser Antilles, the Trinidad region and northeastern Venezuela (Fig. 10). The geometry is based on their analysis of the seismicity along the Lesser Antilles, the results and interpretations of Weber et al.  . Northwest-southeast vertical cross-sections of earthquake hypocentres whose epicentres are located within the corresponding rectangles 1 in the north to 3 in the south shown in Fig. 6. These cross-sections clearly define the Benioff zone and Caribbean slab subducting to the southeast beneath the North Andes plate in northwestern Venezuela and northern Colombia. Arrows in cross-section 1 show sense of motion relative to South America. Maracaibo (M) Lake and Perijá Peninsula are also shown in Fig. 6 for reference. Faults: SCMf, South Caribbean Marginal; Of, Oca; SMf, Santa Marta; Bf, Boconó. C-L, coast line. In the north (cross-sections 1 and 2) the shallower section of the slab dips ∼10 • above a depth of 40 km and 45 • -50 • below that depth. In the south (cross-section 3) it flattens and the shallower and deeper sections dip ∼5 • and ∼35 • , respectively. These results were used in our kinematic block motion modelling.  Fig. 10) east of ∼72 • W. This belt has been proposed as an accretionary wedge along which the Venezuelan basin is obliquely subducted beneath South America (Jordan 1975;Silver et al. 1975;Talwani et al. 1977;Gorney et al. 2007;Magnani et al. 2009;Kroehler et al. 2011). We also examine the effect of the Oca, Morrocoy and Caribbean Deformed Belt faults by assembling and modelling a series of block geometries in which they are systematically included and excluded. In these cases it is assumed, consistent with geology, that the Santa Marta fault must have a left-lateral component of motion (Paris et al. 2000;Idárraga-García & Romero 2010), and that motion on the Oca fault must be rather small (<2 mm yr −1 ), such as indicated by palaeoseismicity studies (Audemard 1996;Audemard et al. 2000) and the block motion analysis of Symithe et al. (2015).
The set-up of the geometric configuration of the various fault segments used to define each plate or block boundary to be used in all models run was done in the following way: (1) We used a locking depth of 14 km for all strike-slip faults, in agreement with the seismogenic depth reported along the El Pilar, San Sebastián, La Victoria and Boconó faults (all faults are shown in Fig. 1) in seismological studies (e.g. Pérez & Aggarwal 1981;Pérez et al. 1997a,b). (2) In the case of a series of subduction segments along the southern Lesser Antilles and Panamá (Fig. 10) we followed Symithe et al. (2015) and used a constant dip of 16 • and a constant locking depth of 40 km. (3) For the subduction segments along the Southern Caribbean Marginal fault (SCM in Fig. 10) we used a locking depth of 40 km and a dip of 10 • , except along its southwest segment where we used 5 • , all according to the seismicity profiles  shown in Fig. 7. The same locking depth (40 km) and a dip of 13 • obtained by Schmitz et al. (2008) is used for the east section of the South Caribbean Deformed Belt east of 72 • W (CDB in Fig. 10).
(4) All our analyses are done in the South American (fix) reference frame. Following Manaker et al. (2008) we constrain the angular velocity of the Caribbean plate, in our case relative to the South American plate, using the Pérez et al. (2001) pole of rotation given in Table 2. Fig. 10 shows a series of five tested models (Models 1-5). In Models 1-4 we simultaneously solve for block rotation rates; kinematically consistent, fully coupled, fault slip rates; and effective elastic coupling coefficients, whereas in Model 5 we impose and fix a slip rate on two slow slipping faults, namely the Oca fault and the east-striking section of the Caribbean Deformed Belt east of 72 • W, as we later describe. We evaluate each model fit by examining the residual (observed-modelled) velocities estimated by the block model analysis while remaining consistent with the previously mentioned geological constraints on the sense and rate of fault slip of the Santa Marta and Oca faults. We follow the style of Loveless & Meade (2010) in presentation of the results in the figure: the estimated fault slip rates predicted from each modelling of block or plate bounding segments are annotated along each segment. Bold black lines indicate the fault traces, and the corresponding labels give strike (top) and fault-normal (bottom) slip rates and uncertainties. Positive signs are for left-lateral and convergent fault motions, and negative signs for dextral slip and tensional faults. Blue and magenta arrows are the observed and residual (observed-modelled) velocities, respectively. Capital letters in the figure provide fault Model 4 is the block model that best fits the GPS data and geological constraints we used. Model 5 is a hybrid model that incorporates slower slipping faults (O, Oca and CDB, Caribbean Deformed Belt) fixing their slip rates, as explained in the text. Letters within circles in Models 4 and 5 indicate the blocks that configurate each model, whose calculated poles of rotation are given in Table 2. names. The strengths and weaknesses of each model are discussed in the following paragraphs.
Model 1 is rejected because the calculated dextral slip rate along the Oca fault (10.0-10.6 mm yr −1 ) is significantly larger than the slip rate obtained (<2 mm yr −1 ) from palaeoseismicity studies (Audemard 1996; Audemard et al. 2000) and the block motion analysis of Symithe et al. (2015). Indeed, changes in direction or magnitude of velocity vectors across the Oca fault (Figs 8 and 9) are within the uncertainties of the data. The large motion modelled for the Oca fault also leads to unacceptably large dextral slip (4.1 mm yr −1 ) along the Santa Marta fault which is geologically recognized to be a left-lateral fault (Paris et al. 2000;Idárraga-García & Romero 2010).
Model 2 is rejected because the combined dextral slip rate modelled along the San Sebastian (S)-La Victoria (V) fault system is only 13.2 mm yr −1 and thus it is unable to reproduce the observed velocities (18-20 mm yr −1 ) on the islands of Aruba (Ar), Curaçao (Cu), Los Roques (Lr) and La Tortuga (Lt), where the residual velocities (magenta arrows in Fig. 10) are very large, in the range of 4-8 mm yr −1 . Also, the model shows 3.4 mm yr −1 of dextral motion along the east-trending section of the Caribbean Deformed Belt, which clearly violates the observed surface velocity field and subsequent elastic model we derived (Fig. 5) for the north-central Venezuela and south-central Caribbean regions.
Model 3, which includes all faults under consideration, is rejected because as in the case of Model 1, contrary to geology the Santa Marta fault shows a large amount (4.1 mm yr −1 ) of right-lateral slip and the calculated slip rate along the Oca fault (10.6-10.7 mm yr −1 ) is significantly higher than the <2 mm yr −1 indicated by palaeoseismicity studies (Audemard 1996;Audemard et al. 2000;Paris et al. 2000;Idárraga-García & Romero 2010).
Models 4 and 5 in Fig. 10 are the two block models that result in velocity fields that best fit the GPS data and geological estimates of fault slip rates. In general, velocity residuals (magenta arrows, Fig. 10) are within measurement uncertainties at most sites. The difference between the two models is the presence in Model 5 of the eastern section of the Caribbean Deformed Belt (CDB in Fig. 10) where we have fixed the total slip at 0.2 mm yr −1 (from Symithe et al. 2015) along the east-striking section of the belt, and the inclusion of the Oca fault with a fix dextral slip of 2 mm yr −1 (Audemard 1996;Audemard et al. 2000;Symithe et al. 2015). Recall that in Models 1-4 we simultaneously solve for block rotation rates and fault slip rates, whereas in Model 5 we impose and fix a small slip rate on two slow slipping faults, namely the Oca fault and the east-striking section of the Caribbean Deformed Belt east of 72 • W. Thus, Model 5 is a hybrid model that combines geological and geodetic data and results, taking into account slow slipping faults such as the Oca and the Caribbean Deformed Belt east of 72 • W, whereas Model 4 takes into consideration only the faults that take up most of the relative motion between the Caribbean and the South American plates, that is, the faults that conform the main seismically active plate boundary between the two plates.
Taking into account both Models 4 and 5 in Fig. 10, the following observations are made: (1) Subduction rates along nearly northsouth oriented segments of the Lesser Antilles subduction are 19.4-19.9 mm yr −1 , in agreement with the results reported by Symithe et al. (2015). The segments which depart from the north-south strike show a mechanically coherent strain partitioning that is dependent on segment's strike.
(2) The easterly oriented El Pilar (P) fault in northeastern Venezuela and the plate boundary fault zone going across Trinidad (T) and then off-shore to the east both show a dextral slip of 19.3-19.8 mm yr −1 , in agreement with the geodetic results originally reported by Pérez et al. (2001), Weber et al. (2001Weber et al. ( , 2009, and also in agreement with the results provided by the kinematic block motion analysis of Symithe et al. (2015). (3) The northwest oriented Los Bajos-El Soldado (BS in Fig. 10) submarine fault system, located between Venezuela and Trinidad (T) shows dextral slip (13.7 mm yr −1 ) and is transtensional (∼14 mm yr −1 ), a type of motion early suggested by Pérez & Aggarwal (1981). (4) In both Models 4 and 5 in Fig. 10 the dextral relative plate motion in north-central Venezuela is partitioned between the San Sebastián (S, 17 mm yr −1 ) and La Victoria (V, 2-2.3 mm yr −1 ) faults, in agreement with the results of the elastic modelling we presented in an earlier section of the manuscript. (5) Taking into account both models, the Boconó (B) fault shows up to 10.2 mm yr −1 of dextral slip, and a component of convergence in the order of 2-6 mm yr −1 , in agreement with the results we derived in the previous section. (6) The Santa Marta (SM) fault shows a left-lateral slip rate of 1.2-2.0 mm yr −1 , and a convergence rate of 6.6-8.0 mm yr −1 depending on the model, in agreement with early GPS results reported by Trenkamp et al. (2002). (7) Subduction rate along the South Caribbean Marginal (SCM) fault segments ranges from ∼5 to ∼7 mm yr −1 . (8) Both models indicate that the Morrocoy (M) fault is right-lateral (2.9-4.8 mm yr −1 ) and transtensional (∼4 mm yr −1 ). (9) The existence of the Morrocoy boundary is compulsory to explain the observed velocity field in northwestern Venezuela and northern Colombia.

D I S C U S S I O N A N D C O N C L U S I O N S
The observed velocity field and corresponding modellings for the south-central Caribbean and north-central Venezuela regions, east of the triple junction vertex between the North Andes, Caribbean and South American plates (Figs 1, 2, 5 and 10) show that the ∼19.6 mm yr −1 of dextral relative motion of the Caribbean relative to South America is split with ∼17 mm yr −1 taken up by San Sebastián fault and ∼2.6 mm yr −1 by La Victoria fault. This latter rate of motion is highly consistent with geological estimates (∼2.75 mm yr −1 ) of the Quaternary rate of motion of La Victoria fault zone (Audemard et al. 2000) near the epicentre of the 1641 earthquake (Fig. 2), where the fault is multibranched and is composed of several parallel strands. This partitioning of slip between subparallel faults is similar to the phenomenon occurring in the Hispaniola Island in northern Caribbean where the Caribbean-North America easterly left-lateral slip (18-20 mm yr −1 , DeMets et al. 2010) is divided among several eastward-trending faults that compose the plate boundary (Dixon et al. 1998;Calais et al. 2002;Symithe et al. 2015). And in some regards it is the same phenomenon taking place in the southern San Andreas fault system in California, where several subparallel faults accommodate plate motions (e.g. Lindsey & Fialko 2013).
The slip rate along San Sebastián fault is ∼5.5 times larger than the one along La Victoria fault and this may explain the much lower seismic moment productivity of the La Victoria fault compared to San Sebastián fault in the last ∼4 centuries (Pérez et al. 1997a) and qualifies the San Sebastián fault as the main strand of the plate boundary in the region. If slipping at ∼17 mm yr −1 , the ∼100 km long western segment of this fault (Fig. 2) has accumulated about 3.4 m of potential slip since it last broke in 1812 (Audemard 2002;Choy et al. 2010), sufficient to be released during a large strike-slip earthquake of M w > 7, according to the scaling laws derived by Henry & Das (2001) and Wesnousky (2008).
Northwest of the North Andes-South American-Caribbean plates triple junction (Fig. 1) the Morrocoy fault zone marks the northeast boundary of the North Andes plate with the Caribbean plate (Pérez et al. 1997b;Bird 2003). Focal mechanisms along this fault system show northwest oriented dextral slip with a small component of normal faulting about 100 km northwest of the triple junction (Fig. 2), and normal faulting with a strike-slip component (Audemard et al. 2005) farther northwest around 12 • N (Fig. 4). Right-lateral transtension is required across the Morrocoy fault because easterly directed GPS velocity vectors (Fig. 8) increase in magnitude and in azimuth as one moves from the North Andes plate to the Caribbean plate. As it was described in the previous section, from a kinematic block modelling we estimate the amount of dextral slip in ∼3-4.8 mm yr −1 along fault's strike and about 4 mm yr −1 of normal slip in the direction perpendicular to the fault. These observations are consistent with the observed focal mechanisms of seismic events straddling the Morrocoy fault (Figs 2 and 4) and also with geological and geophysical studies carried out in the region (Bellizzia et al. 1976;Beltrán 1993;Audemard et al. 2000;Gorney et al. 2007;, which show the existence of active northwest striking Quaternary faults with normal and dextral slip both off-shore and inland. Fig. 11(a) is a map that compiles the known traces (coloured sinuous lines) of faults active since Tertiary in the regions adjacent to the Morrocoy (M) fault (Beltrán 1993;Audemard et al. 2000;Gorney et al. 2007;Castillo et al. 2017). Large portions of the Morrocoy boundary have not been seismically explored or the information is in the non-public oil industry domain. Fig. 11(b) shows a seismic  Fig. 11a). It shows normal faults cutting across the basement in an area that matches the proposed location of the Morrocoy fault. Farther southeast the fault is apparently buried beneath a northwest trending thrust belt at the base of the Falcon slope.
In the most northwestern regions of the South American continent, using teleseismic and microseismicity data we have improved the knowledge of the geometry concerning the northeast striking, southeast dipping subduction of the Caribbean plate beneath the North Andes plate (Figs 6-8), and have shown this subduction is oblique to the Caribbean-South America slip vector. Also, the modelled rate of southeast subduction of the Caribbean plate relative to the North Andes plate (Fig. 10) is rather small (5-7 mm yr −1 ), which explains the low background seismic activity in the region. Indeed, the Benioff zone in northwestern Venezuela is better defined only when microearthquake locations are considered. This small relative rate of subduction implies that 2000-3000 yr of strain accumulation are needed for a potential giant (M w > 8.5) shock with a co-seismic slip of up to 15 m or more to break the entire 500-600 km long interplate boundary, a rupture length typical for events of this size (e.g . Thatcher 1990;Pérez & Scholz 1997), as it was the case during the 2010 M w 8.8 Chilean earthquake (e.g. Vigny et al. 2011;Wang et al. 2012). At this time scale the seismic history of the region is completely unknown. A smaller yet strong event would need a smaller strain accumulation time. For instance, it would take around 300-600 yr to accumulate 2-3 m of slip, equivalent to the average slip that would be expected in an M w 7.5-7.8 subduction earthquake with a rupture length on the order of 130 km (e.g. Henry & Das 2001;Wesnousky 2008).
Southwest of the triple junction the interaction between the North Andes and South American plates along the Boconó fault zone and Venezuelan Andes occurs in such a way that slip partitioning takes place with 9-11 mm yr −1 of dextral motion in the far field of the fault and 2-5 mm yr −1 of shortening across the Venezuelan Andes. Thus, the slip partitioning exhibited by this region resembles that occurring in Central California where thrust faulting occurs on planes that strike near parallel to the San Andreas fault, as in the 1983 Coalinga shock and related events (Stein & King 1984;Ekström & England 1989;Stein & Ekström 1992). As along the central San Andreas fault, focal mechanisms reported for the area (e.g. Pérez et al. 1997b) show pure dextral motion for microseisms associated with the Boconó fault itself, whereas many microearthquakes located well away from the fault and along the Andean foothills show pure thrust motion on planes dipping towards the Boconó fault.
Several authors have proposed (e.g. Kellogg & Bonini 1982, 1985Kellogg & Vega 1995;Trenkamp et al. 2002) that dextral motion along the Oriente, East Andean and Boconó faults in Ecuador, Colombia and Venezuela, respectively (Fig. 1), is accommodating the northward component of motion of the North Andes plate that results from a squeezing of the plate between the Nazca plate and the Oriente and East Andean faults. While this idea is acceptable and is consistent with our GPS data, the transtensional right-lateral motion we observe along the Morrocoy fault requires that the resulting eastward component of North Andean motion remains less than that of the Caribbean plate. Finally, the GPS coverage for the southern Caribbean region is being significantly improved thanks to the implementation by UNAVCO of a wide aperture GPS array, the COCOnet GPS network, covering the whole Caribbean and adjacent regions. Thus, it is expected that refinements in the knowledge of the kinematics of the region and subsequent quantification of seismic hazards will arise as new observations are collected.

A C K N O W L E D G E M E N T S
New GPS observations in Venezuela were funded by Simón Bolívar University. We thank the Colombian Geological Survey for providing seismic data and a series of GPS velocity vectors of sites in Colombian territory. David Mencin provided us with several unpublished velocity vectors and helped us in the geodetic data processing. UNAVCO provided data from sites that belong to the COCOnet GPS array. We are grateful to John P. Loveless for his help to use the software 'Blocks'. Our former students Yuleika Madriz, Freddy Rondón, Carlos Moncayo, Mizael Bravo and the Physical and Satellite Geodesy Laboratory of La Universidad del Zulia in Maracaibo, Venezuela, took care of the new data gathering. Comments and suggestions made by Editor Juan C. Afonso, Eric Calais and by an anonymous reviewer led to a significant improvement of our work. Roger Bilham provided us with GPS data he collected in Aruba in 2012. In memoriam of LL who passed away during the preparation of the manuscript.