Marine terraces are landforms that record both paleohorizontal and true surface uplift above mean sea level, and thus serve as long-baseline geodetic markers. Here we present a method for extending terrace records back in time by analyzing 2003 National Aeronautics and Space Administration (NASA) airborne LiDAR data to discover a suite of probable marine terraces along a 70-km-long section of the northern California coast. In particular, we present a semiautomated surface classification model, based on slope and surface roughness properties, to identify low-slope and low-relief topographic platforms with characteristics similar to marine terraces. Between Fort Ross and Point Arena (California, United States), the method identifies 851 individual platforms that range in elevation from 100 to 450 m above sea level. The total platform area is 24.5 km2 over a total area of 394 km2; the majority of platforms are found west of the San Andreas fault. We interpret these platforms as a suite of marine terraces because they are generally elongate parallel to the modern coastline and one another, and their elevations increase systematically from west to east in a step-like fashion. Geologic evidence supporting this conclusion includes locally abundant well-rounded and well-polished clasts with heterogeneous compositions, beveled outcrops of the underlying bedrock (German Rancho Formation), and provenance studies indicating transport of clasts contained in probable marine deposits across the San Andreas fault. The platforms contain 21 different elevation bands, each of which is 2–6 m high, suggesting that they may define a flight of 21 distinct marine terraces at elevations between 113 and 405 m above sea level. Matching the 21 observed platform elevations with those predicted from sea-level curves suggests that the surface uplift rate for the northern coast of California has been ∼0.2 mm/yr over the past 2 Ma. This rate is similar to, but lower than the previously determined uplift rate of ∼0.3 mm/yr for a ∼23 m marine terrace in the area.
Marine terraces are important landforms because they record both paleohorizontal and uplift (Bradley and Griggs, 1976; Lajoie, 1986), defined as the height of a feature above sea level after correction for eustatic sea-level change (e.g., England and Molnar, 1990; Molnar and England, 1990). In the case of a marine terrace, this rock uplift corresponds to surface uplift because erosion of the terrace must be minimal for it to be preserved as a geomorphic marker in the landscape (Abbott et al., 1997). Thus, marine terraces can be used to determine surface uplift rates if the sea-level history is known and the terraces can be dated, or they can be used to quantify the record of eustatic sea-level change in cases where the uplift rate is known independently (Lajoie, 1986). Chappell (1974), Bradley and Griggs (1976), and Lajoie (1986) reported pioneering work that developed understanding of marine terraces and their evolution in regions of active tectonics. Anderson et al. (1999) advanced on these defining studies by using marine terraces to investigate landscape evolution.
Because marine terraces can serve as long-baseline geodetic markers that track finite strain since their formation, they have been used extensively to investigate coastal tectonics around the world (Abbott et al., 1997; Armijo et al., 1996; Berryman, 1993; Bloom et al., 1974; Chappell, 1974; Chappell and Veeh, 1978; Lajoie, 1986; Pedoja et al., 2008); a number of studies investigated marine terraces and their tectonic implications in northern California and Oregon (e.g., Anderson and Menking, 1994; Kelsey, 1990; McLaughlin et al., 1983; Merritts, 1996; Merritts and Bull, 1989; Muhs et al., 1990, 2003, 1992; Prentice, 1989; Prentice et al., 1991; Valensise and Ward, 1991). Marine terraces may also potentially be used to distinguish between tectonic models that differ in their predictions of the spatial and temporal patterns of rock and surface uplift, such as those for the Mendocino Triple Junction at the north end of the San Andreas fault in northern California (Fig. 1C; e.g., Furlong and Govers, 1999; Furlong and Schwartz, 2004; Lock et al., 2006; Merritts, 1996; Merritts and Bull, 1989; Trehu and Mendocino Working Group, 1995).
Although the youngest marine terraces are often clearly preserved in the landscape, older generations are commonly obscured due to vegetation cover, burial by younger deposits, decay of the former terrace margins, and stream incision (e.g., Anderson et al., 1999). Thus, to extend records of coastal deformation or sea-level history back in time, it is necessary to develop methods for discovering and characterizing the geometry of these older and more obscure marine terraces.
The northern coast of California provides a natural laboratory to develop such methods and thus extend surface uplift histories. Several generations of marine terraces at elevations <100 m above sea level attest to recent surface uplift in the region, and prior work has focused on mapping and dating these terraces, which have ages of ca. 80 ka to ca. 300 ka (Crosby et al., 2007; Merritts, 1996; Merritts and Bull, 1989; Muhs et al., 1994; Prentice, 1989; Prentice and Kelson, 2006). In contrast, elevations >100 m in the region have received less attention because dense redwood forests generally cover them. Airborne LiDAR (light detection and ranging) data in the area provides the opportunity to look for older terraces that may be hidden beneath this vegetation.
To determine if an older history of marine terraces is preserved in the area, we analyzed a bare-earth digital elevation model (DEM) derived from a 2003 National Aeronautics and Space Administration (NASA) LiDAR survey (Harding, 2004). We focused the investigation on areas west of the San Andreas with elevations >100 m above sea level along the ∼70-km-long stretch of coastline between Fort Ross State Park in the south and Point Arena in the north, in Sonoma and Mendocino Counties (Fig. 1).
In this paper, we present and evaluate a semiautomated surface classification model (SCM) that we developed to identify potential marine terraces on the basis of both their topographic characteristics (slope and relief) and their geomorphic context. We refer to the features identified by the SCM as platforms, which we use without genetic connotation. Our analysis revealed a regionally extensive flight of platforms at elevations of 100–450 m that we interpret to be marine terraces. Although dating these surfaces was beyond the scope of our study, we used the geometry of the platforms and their elevation spacing to identify the uniform surface uplift rate that best explains their elevation distribution. Identification of these platforms extends the record of surface uplift along the northern California coast from 320 ka back to ca. 2 Ma.
This study presents a new approach for identifying marine terraces. Although we developed the method to investigate the uplift history of the northern California coast, it is generally applicable to marine terraces globally because it relies on the topographic characteristics of these distinctive landforms and how they appear in airborne LiDAR data. It should also be applicable to lake shorelines, which can be used as strain markers in continental interiors.
The study area is divided by the northwest-striking, active, right-slip San Andreas fault (SAF), which separates the Gualala block (i.e., Pacific plate) to the west from the Franciscan Complex and Coast Range (i.e., North American plate) to the east (Fig. 1 and Supplemental Fig. 1 in the Supplemental File1) (Blake et al., 2002; Prentice, 1989; Wentworth, 1967). In this area, the SAF generally is east of the modern coastline, coming on shore ∼3 km south of Fort Ross State Park and leaving ∼10 km north of Point Arena (Fig. 1). The SAF is an important potential seismic source (Ellsworth, 1990; Gilbert, 1907; Lawson, 1908), and is the principal fault within the San Andreas system, which also includes predominantly dextral faults that are to the east within the Coast Range (e.g., Fig. 1C). The SAF system as a whole accommodates the majority of the plate motion between the Pacific and North American plates (Wallace, 1990), although in the study area, the SAF sensu strictu only absorbs 35% (∼18 mm/yr; Prentice, 1989; Prentice and Kelson, 2006) of the total ∼50 mm/yr Pacific–North American relative plate motion as determined by Global Positioning System (GPS) measurements (Freymueller et al., 1999; Sella et al., 2002).
Because the SAF has juxtaposed blocks with distinctive geology in the study area, identification of materials on the west side of the fault (i.e., Gualala block) that were derived from rocks to the east (i.e., Franciscan) indicates sedimentary transport across the fault, and can therefore help in understanding landform genesis. The majority of the Gualala block consists of the German Rancho Formation, which is Paleocene to Eocene in age as determined by megafossils and microfossils (Wentworth, 1967), and is characterized by feldsarenite, lithic feldsarenite, and conglomerate (Supplemental Fig. 1 [see footnote 1]) (Anderson, 1998). To the east, the Franciscan Complex consists of the coastal, central, and eastern structural belts (Blake et al., 1985); the central and coastal belts are exposed in the study area (Supplemental Fig. 1 [see footnote 1]). It is important that no distinctive lithologic components of the Franciscan Complex, such as radiolarian chert, blueschists, or metagraywacke, have been reported from within the German Rancho Formation west of the SAF (Anderson, 1998; Wentworth, 1967).
The Mendocino Triple Junction is northwest of the study area, and is a transform-transform-trench triple junction where the Gorda, North America, and Pacific plates conjoin (Fig. 1C; Zandt and Furlong, 1982). The junction is 220 km from Fort Ross and 160 km from Point Arena (Fig. 1), and is migrating northwest at a rate of 50 mm/yr with respect to stable North America (Sella et al., 2002). Because the overall plate motion between the Pacific and North American plates is not all focused on the SAF, rocks on the east side of the study area are only moving away from the triple junction at a rate of ∼18 mm/yr. According to Atwater and Stock (1998), the triple junction passed points on the North American plate at the current latitude of Fort Ross ca. 8 Ma and at the latitude of Point Arena by 6 Ma.
Topographic Characteristics of a Marine Terrace
Marine terraces are shore platforms that form below sea level and are preserved due to surface uplift or sea-level fall (Bradley and Griggs, 1976; Burbank and Anderson, 2001; Muhs et al., 1992; Fig. 2). The intersection between a marine terrace and an adjacent sea cliff is the shoreline angle, which defines an originally horizontal datum that can record deformation or sea-level fluctuations (Fig. 2; Bradley and Griggs, 1976; Chappell, 1974; Lajoie, 1986; Muhs et al., 1992; Pethick, 1984; Scott and Pinter, 2003). The surface of the marine terrace is often used as a proxy for the shoreline angle because the latter is often eroded or obscured by eolian deposits and/or colluvium shed from the adjacent sea cliff (Anderson et al., 1999; Scott and Pinter, 2003).
Marine terraces have characteristic slope, relief, and deposits that enable their identification. In a flight of terraces, individual terrace heights increase systematically landward. The individual terrace surfaces generally slope oceanward 1°–6°, but can be as high as 15° (Anderson et al., 1999; Merritts et al., 1991; Muhs et al., 1992; Scott and Pinter, 2003). This slope increases with age due to both burial of the inner shoreline angle by colluvium and decay of the outer sea cliff (Fig. 2). The original terrace surface is generally smooth, although sea stacks, landslides, and gullies can locally increase roughness (Fig. 2; Bauer, 1952; Lawson, 1894). A thin veneer of marine sediments can unconformably cap the bedrock platform (Clifton et al., 1971; Kelsey, 1990; McLaughlin et al., 1983). With increasing age, however, alluvial, colluvial, and eolian sediments can cover these deposits, and soils will develop (Muhs et al., 1992; Polenz and Kelsey, 1999), potentially obscuring their origin. Pygmy forests are also associated with marine terrace flats (Aitken and Libby, 1994; Northup et al., 1995; Westman, 1978). In summary, although the topographic signature of a marine terrace will degrade over time, preserved portions of the terrace surface should remain generally smooth. Individual terrace surfaces should slope oceanward 0°–15°, heights of adjacent terraces should systematically increase in elevation landward, and wave-worked sediments of nonlocal provenance may cap the terrace surface.
Previous Work on Marine Terraces along the Coast of Northern California
The surface uplift history of the northern coast of California between 80 and 320 ka is well studied, based on mapping of the most recent/lowest elevation marine terraces (Crosby et al., 2007; Merritts and Bull, 1989; Merritts et al., 1991; Muhs et al., 1994, 2003; Prentice and Kelson, 2006; Prentice et al., 2000). However, there is a gap in the terrace record between ca. 320 ka and 3.3 Ma in the study area. As Table 1 indicates, at least 5 generations of marine terraces from Fort Ross to Point Arena have been identified previously, although their reported elevations differ: 19–28, 38–50, 56–75, 95–110, and 145–160 m (Bauer, 1952; Crosby et al., 2007; Huffman, 1972; Lawson, 1894; Muhs et al., 1994, 2003; Prentice, 1989; Prentice and Kelson, 2006). Bauer (1952) mapped 4 marine terraces ranging in height from 30 to 140 m using topographic maps, aerial photographs, and a survey altimeter. Prentice (1989) mapped 5 generations of marine terraces near Point Arena ranging in elevation from 24 to 160 m. Crosby et al. (2007) used aerial photographs and bare-earth DEMs (from the same 2003 NASA LiDAR data set that we used) to manually map the inner edges of the lowermost three marine terraces on the Gualala block at elevations between 22 and 80 m.
Dates on these terraces are rare, but the solitary coral Balanophyllia elegans was found at 2 locations on a ∼23 m elevation terrace near Point Arena, yielding U-series ages of 76 ± 4 ka and 88 ± 2 ka (i.e., marine isotope substage 5a; Muhs et al., 1990, 1994). By correlating shoreline angle elevations with the glacioeustatic sea-level curve, the ∼44 m and ∼65 m terraces have been estimated to be ca. 103 ka and ca. 120 ka (i.e., marine isotope substages 5c and 5e), respectively (Muhs et al., 2003; Prentice, 1989). Likewise, the ∼100 m and ∼150 m terraces have been estimated to be ca. 210 ka and ca. 320 ka (i.e., marine isotope stages 7 and 9), respectively (Crosby et al., 2007; Prentice, 1989). Based on these elevations and ages, surface uplift rates in this area have been estimated to be ∼0.2 to ∼0.6 mm/yr (Crosby et al., 2007; Merritts and Bull, 1989; Muhs et al., 1992; Prentice, 1989; Prentice and Kelson, 2006).
Prior studies also suggested the possible existence of older, topographically higher marine terraces. Bauer (1952) discussed the plausibility of marine terraces at elevations above ∼150 m, but concentrated on mapping terraces below that elevation due to limitations on the accuracy of topographic data available at the time, noting that morphology provides the best way to identify higher/older marine terraces (Bauer, 1952). In a report primarily focused on assessing seismic hazard and landslides, Huffman (1972) mapped the surficial geology along the coast from Fort Ross to Point Arena at a scale of 1:24,000 and identified old marine terrace deposits at various elevations up to a maximum of ∼275 m, although he did not focus on these features and primarily mapped the high-elevation (150–250 m) marine terrace surfaces from aerial photographs with minimal field verification. Huffman (1972) also mapped lower marine deposits, but did not subdivide individual terrace generations; he found that deposits capping the terraces are orange to white, poorly sorted, sand-clay gravel with rounded pebbles and cobbles. Blake et al. (2002) compiled bedrock and surficial maps to create a 1:100,000 geologic map of western Sonoma, northern Marin, and southern Mendocino Counties; the map groups fluvial and marine terraces and shows several such undifferentiated terraces with very limited extent near Fort Ross at elevations up to 250–300 m.
In a reconnaissance expedition, Lawson (1894) noted that marine terraces are preserved at elevations of 425, 360, 230, 135, 105, and 85 m. He identified and mapped marine terraces based on their morphology as expressed on topographic maps. He did not find evidence of wave action at elevations above 85 m, but identified sand and gravel beaches near Fort Ross School (Lawson, 1894). Higgins (1960) described the Ohlson Ranch Formation as marine sedimentary deposits that overlie the Franciscan Complex and crop out in patchy areas at elevations between ∼150 m to ∼520 m. A zircon age from a tuff bed within the Ohlson Ranch Formation provided an age of 3.3 ± 0.8 Ma (Prentice, 1989). A height of 500 m and age of 3.3 Ma for that unit imply a long-term rock uplift rate of ∼0.15 mm/yr. The mapped patches of Ohlson Ranch Formation are entirely east of the SAF and reflect the invasion of a Pliocene sea; implying that ca. 3 Ma the shoreline was east of the SAF (Higgins, 1960; Lock et al., 2006; Prentice, 1989).
A number of different geologic problems have benefited from use of LiDAR data (e.g., Cunningham et al., 2006; Engelkemeir and Khan, 2008; Frankel and Dolan, 2007; Glenn et al., 2006; Gold and Cowgill, 2008; Gold et al., 2007, 2009; Haugerud et al., 2003; Hofton et al., 2006; Olariu et al., 2008). In our study, LiDAR data provided an opportunity to search for previously unrecognized marine terraces at elevations above ∼100 m along the coast of northern California. Studying the morphology of these areas has previously been hindered by dense vegetation and its effect on the accuracy of prior topographic maps.
In 2003, NASA flew an airborne LiDAR campaign along the SAF in northern California (Harding, 2004; TerraPoint, 2004; Prentice et al., 2003). These data have been used to map geomorphic features related to the San Andreas fault (Prentice et al., 2004) and marine terraces at elevations <100 m (Crosby et al., 2007). The LiDAR data were collected using the Airborne LiDAR Terrain Mapping System 4036 with GPS control from a Trimble Ag132 differential GPS receiver on the aircraft and a Trimble 4700 GPS receiver base station, and acceleration measurements from a Honeywell H-464G embedded GPS/inertial navigation system (TerraPoint, 2004). The data were classified using TerraSolid's TerraScan software (http://www.terrasolid.fi/), with points classified as blunder (erroneous point), ground or water, vegetation, or building or structure (see TerraPoint, 2004). The point-cloud data are provided in state plane projection in zone California II, the spheroid is GRS80, the horizontal datum is NAD83 (North American Datum, 1991 Adjustment), the vertical datum is NAVD88, and horizontal units are in U.S. survey feet (1 ft = 0.3048006096 m) with elevation in international feet (1 ft = 0.3048 m). The entire data set spans ∼418 km2 and contains 1.2 × 109 vegetation and ground points with an average point density of 1.2 points/m2. The classified point cloud was gridded into DEMs with a cell size of 6 ft (Harding, 2004). We obtained the data in their gridded format from http://core2.gsfc.nasa.gov/LiDAR/terrapoint/san_andreas/. We did not use the 2007 EarthScope LiDAR data (Prentice et al., 2009) because they are limited to a 1-km-wide strip along the SAF.
Our goal in creating the SCM was to provide a single metric for identifying topographic surfaces with both low slope and low surface roughness, because these are the primary topographic signatures of marine terraces. As explained here, we generated the SCM by linearly combining normalized slope and roughness values. Both values had to be normalized because their ranges differ and normalization converted each to the same range (i.e., from 0 to 1). In addition, both the slope and roughness values for the data set as a whole contained abnormally large maximum values that resulted from computing slope and roughness between areas with and without data, such as along the edge of the DEM or in the ocean. To resolve this problem, we clipped both the slope and roughness data to remove the anomalous values, and then normalized by the upper value of the clipped data range.
To perform the topographic analyses we used the 2003 NASA bare-earth DEM and ESRI ArcGrid software (www.esri.com). To avoid resampling, we performed all analyses using the original DEM data in state plane units with a 6 ft (1.83 m) cell size. We then converted results to meters after the analyses were completed (see Supplemental Text in the Supplemental File [see footnote 1] for ArcGRID code). Because of this, we report some values in feet.
To generate a slope map (Fig. 3B), we used a moving 3 × 3 window (Supplemental Fig. 2A [see footnote 1]) to calculate slope values for each individual cell in the DEM following Burrough and McDonnell (1998) (equations in Supplemental Text [see footnote 1]). We then clipped the slope map by setting all slopes >15° to null. We chose 15° as the upper limit because that is the maximum value expected for preserved marine terrace remnants. Marine terraces typically slope only a few degrees when first formed, but slopes generally increase over time due to erosion and deposition of colluvium along the terrace edges. This value range encompasses slope angles of 1.6°–3.4° calculated by Crosby (2006) for select traverses across the youngest marine terrace surfaces in the study area using the 2003 LiDAR data set. To normalize the slope values, we divided all values by 15°, which converted the slope to a range of 0–1.
Surface roughness (Fig. 3C) is the standard deviation of the slope (Frankel and Dolan, 2007), larger values indicating greater roughness. Here again we used a 3 × 3 moving window to calculate the standard deviation for each cell in the DEM (equations in Supplemental Text [see footnote 1]). The maximum roughness value for the data we analyzed was 43.04 (Fig. 4A), reflecting spurious values in the ocean and along the edge of the DEM. To remove these anomalous results, we clipped the roughness map by setting to null all values >4.0 (Fig. 3C). We chose 4.0 as the maximum value because it incorporates 90% of the roughness values and removed the long tail of anomalously high values (Fig. 4A). We then normalized by dividing all surface roughness values by 4.0, which converted the roughness to a range of 0–1.
Mapping Platform Perimeters
Two basic approaches were possible for delineating the platform margins: use ESRI ArcGIS (http://www.esri.com/software/arcgis/index.html) to automatically extract polygons around raster values of a particular value, or digitize them manually. Although more time intensive, we chose to manually map the edges of continuous platforms with SCM values <0.3 using the map of SCM values, a hillshade image from the bare-earth DEM, and aerial photographs (National Agricultural Imaging Program, 2005). We mapped at a scale of 1:5000 using ESRI ArcGIS version 9.2.
We manually mapped platform margins because doing so allowed us to avoid areas where an automated analysis would produce false positive or false negative results. For example, referring to aerial photographs during the mapping allowed us to identify areas such as ponds, lakes, or graded areas (e.g., an airport) that would generate false positives in an automated analysis. Manual mapping also allowed us to recognize geologically implausible situations such as “platforms” that are inland of, but at lower elevation than, an adjacent platform to the west. We mapped platforms east of the SAF at elevations lower than those to the west because the two fault blocks could juxtapose different shoreline positions or have different histories of surface uplift.
Manual mapping also allowed recognition of situations where an automated approach would not map a platform, but where topographic context indicated such a feature was present (i.e., a false negative). For example, features such as remnant sea stacks, landslide deposits on the platform, or small drainage networks incised into a platform are all likely to have local SCM values above the particular value used to automatically extract polygons. However, if these features do not define the actual edges of the platform, then they should be included within the platform polygon.
Determining Platform Elevations
Due to the temporally pulsed nature of marine terrace formation, the platform surfaces should concentrate into discrete elevation bands if they are marine terraces. To look for such a pattern, we evaluated the distribution of platform elevations by extracting elevations for every pixel contained within, or crossed by, a platform polygon and then plotting the number of pixels versus elevation. Peaks in the resulting elevation-frequency plots reflect increased concentrations of platform elevations at a given height, which we call elevation bands. We selected peak positions and widths manually, using the peak widths to estimate the widths of the elevation bands. We also extracted platforms automatically by clipping the SCM to retain only values between 0 and 0.3, extracting the number of pixels and their individual elevations in the clipped data, and then plotting the results to compare the elevation distribution of the manually mapped platforms with those generated automatically.
To evaluate the spatial distribution of elevations within the platforms and their geometric distribution in the landscape, we colored the DEM according to the elevation bands identified in the elevation-frequency plots described here. Specifically, we first clipped the original LiDAR DEM using the shapes of each manually mapped platform, discarding data outside the platforms. We then plotted all cells within a single elevation band (e.g., 330 ± 6 m) using the same color (e.g., purple). Cells that are within the platform polygons but have elevations outside the defined elevation bands were left blank.
The elevation analysis relied on several assumptions. First, it presumed that the platforms have not been folded or tilted at wavelengths smaller than the typical platform length along strike, so that the platforms will be within a narrow (<50 m) elevation range. Second, it presumed that peaks in the elevation curve represent the average original elevation of the platform. Although the preferred elevation to determine on a marine terrace is that of the inner edge, these features are generally the first to be covered by diffusion of the sea cliff. Finally, it assumed that the automated point-cloud classification provides a robust bare-earth DEM that does not produce artificially smooth surfaces or extrapolate height estimates over large areas. At present these assumptions are relatively untested.
Surface Classification Model
Because the locations of marine terraces <100 m above sea level are well constrained by prior work (Crosby et al., 2007; Merritts and Bull, 1989; Merritts et al., 1991; Muhs et al., 1994, 2003; Prentice and Kelson, 2006; Prentice et al., 2000), we only mapped platforms above 100 m. The SCM revealed a suite of 851 platforms at elevations ranging from ∼100 m to ∼450 m between Fort Ross and Point Arena (Fig. 5; Supplemental Fig. 3 [see footnote 1]). The total platform area is 24.5 km2 over a total area of 394 km2. The majority of platforms are west of the SAF, only ∼8% (i.e., 1.9 km2 of the 24.5 km2 total) being mapped to the east. The maximum length of a platform is ∼1.5 km measured parallel to the modern shoreline and along the strike of the SAF, with an average length of ∼0.2 km. The average platform width is ∼0.14 km perpendicular to the shoreline.
Performance of SCM on Known Terraces
To test the accuracy of the SCM, we evaluated the extent to which it successfully identified the inner edges of the four youngest marine terraces independently mapped by Crosby et al. (2007). These inner edges are at the bases of risers or cliffs that should have slopes >15°, precluding them from being identified as platforms in the SCM. However, a marine terrace flanks this riser or cliff, and should appear as a platform in the SCM. Thus, we expected the inner edges mapped by Crosby et al. (2007) to be near the boundary between a strip of high to null SCM values (the riser) and adjacent area of low SCM values (the terrace). Maps showing both SCM values and the inner-edge locations mapped by Crosby et al. (2007) indicate that the model accurately highlighted terraces and the intervening risers where they are preserved (Fig. 6).
Plots of elevation frequency for the areas encompassed by the platforms show multiple peaks, indicating that platform elevations are clustered into distinct elevation bands (Fig. 7). The manually mapped platforms contained 21 different elevation bands ranging in elevation from 113 to 405 m (Table 2; Fig. 7A). Not all bands have peaks of the same amplitude or width, indicating that some platforms are not as common or as well defined as others. Comparison of the elevation distributions of the manually mapped (Fig. 7A) and automatically extracted (Fig. 7B) platforms shows that both the overall data distribution and the elevations of individual peaks are broadly similar. For example, both curves show concentrations of peaks in the ranges 100–150 m and 275–300 m, with minima at ∼220 and ∼310 m. Essentially all peaks picked from the manual mapping overlap within error with peaks in the elevation distribution from the automatically extracted platforms, and very few additional peaks are present in the latter that are absent from the former.
Maps showing the elevation bands as colored contour intervals crossing the platforms (i.e., color bands) help to clarify the location and spatial distribution of the elevation bands within the landscape (Fig. 8; Supplemental Fig. 7 [see footnote 1]). It is important to note that there is not a one-to-one correlation between the manually mapped platform polygons and the elevation bands, so that some polygons contain multiple bands. All 21 color bands are present at the north end of the field area (Fig. 8), where maximum elevations are highest. The maximum elevation at a given location along the fault sets the maximum number of bands that could possibly occur at that area, and in general, color bands with elevations at or below this upper limit are typically present. A number of the color bands, particularly those at lower elevations (113–180 m), are present along most of the coastline. Individual platforms typically contain multiple color bands, and inside these individual platforms the color bands generally increase in elevation systematically from southwest to northeast. However, there are some cases in which a single higher elevation band will define an island that is flanked both to the southwest and northeast by bands of lower elevation.
It is important to note that in these maps (Fig. 8; Supplemental Fig. 7 [see footnote 1]), it is given that the color bands within individual polygons will be elongate along the coastline, because the topography of the Gualala block on which they occur is generally that of an elongate ridge trending roughly parallel to both the San Andreas fault and the coastline. As such, this aspect of the color bands simply reflects the regional strike of the topography. Likewise, it is also given that band elevations between platforms will increase systematically from southwest to northeast, because we excluded platforms for which this was not the case during the manual mapping.
To check the surface classification method, we conducted reconnaissance field work at 57 sites on 4 separate platforms, all at an elevation of ∼280 m (Fig. 1). Because we had limited access to private property, most platforms we visited were in public or state parks or along public roads. The 280 m platforms in Salt Point State Park and Bower Park (Fig. 1) are characterized by different vegetation patterns than surrounding areas (e.g., no trees or pygmy forests) and clastic deposits with well-rounded and polished cobbles, including clasts of radiolarian chert (Supplemental Text and Supplemental Figs. 5 and 6 [see footnote 1]). At Bower Park, outcrops next to a baseball field expose medium-grained, poorly indurated, massive sandstone of the German Rancho Formation, overlain by an unconsolidated and poorly sorted clastic deposit containing highly polished and well-rounded clasts (Supplemental Figs. 5A–5D [see footnote 1]). Clast compositions in the overlying unit are varied and include minor radiolarian chert, sandstone, and volcanic rock. A roadcut ∼9 km northwest of Bower Park exposes a beveled outcrop of German Rancho Formation overlain by a medium-grained clastic sedimentary deposit with well-rounded clasts of radiolarian chert, sandstone, and volcanic rock (Supplemental Figs. 5E–5H [see footnote 1]).
The shapes of the platforms, their concentration into discrete elevation bands, and their geology where examined in the field all suggest that the platforms are most simply interpreted as a flight of marine terraces. For example, most individual polygons have long axes that parallel the modern shoreline. They also tend to be aligned along their long axes to form chains of polygons that roughly parallel the modern shore and the SAF (Fig. 5). In addition, many platforms have long edges that parallel those of neighboring platforms at slightly higher or lower elevations. Although the areas of the platforms are generally smaller than those of the youngest marine terraces in the area, this is expected due to greater erosion with increasing terrace age. Analysis of topographic curvature indicates that the platforms most likely originated as tabular landforms that have diffused around their margins, and are not simply ridge crests (Supplemental Text, Supplemental Fig. 4 [see footnote 1]).
In more detail, the presence of more than one color band in many of the larger polygons suggests that either we grouped together more than one platform with closely spaced elevations when manually mapping, or that we defined too narrow an elevation range for the size of the bands. We think it is more likely that the manually mapped polygons contain multiple platforms, because geomorphic decay of the intervening riser is expected to make it difficult to see such boundaries on a map of SCM values.
Results from field reconnaissance are consistent with a marine origin for the platforms observed at an elevation of ∼280 m, although more field work is needed to check the platforms at the other elevations. Pygmy forests such as those observed in the field are known to be associated with marine terrace flats (Aitken and Libby, 1994; Northup et al., 1995; Westman, 1978). Likewise, the degree of polishing of well-rounded clasts is consistent with a marine origin for deposits unconformably overlying beveled German Rancho Formation in Bower Park (Supplemental Fig. 5 [see footnote 1]). The presence of heavily fractured radiolarian chert in unconsolidated sediments capping the German Rancho Formation (Supplemental Fig. 6 [see footnote 1]) also strongly suggests a marine origin for the 280 m platforms because they indicate sedimentary transport from source areas in the Coast Range that now are across a large, fault-parallel valley and on the opposite side of the SAF. The simplest explanation for the presence of these clasts is that they were transported across the fault when the platform was at sea level, prior to surface uplift and formation of the intervening valley. The German Rancho Formation, which is the predominant unit underlying the platforms, contains numerous conglomerate horizons, raising the possibility that the clasts were derived from in situ weathering. However, Wentworth (1967) stated that no Franciscan-sourced chert is found in any geologic unit west of the SAF in the study area, including the German Rancho Formation, making in situ weathering an unlikely source.
This model predicts that marine terraces in the vicinity of Santa Cruz, California, have a life expectancy of ∼500 ka, based on a 500 m spacing of major streams, 0.3 mm/yr rock uplift rate, and 30° bedrock failure angle (Anderson et al., 1999). In contrast, in the Fort Ross–Point Arena area, the life expectancy of marine terraces should be ∼720–2200 ka, because in this area the mean spacing of major drainages is ∼1500 m, and previously reported surface uplift rates vary from ∼0.2 to ∼0.6 mm/yr (Crosby et al., 2007; Merritts and Bull, 1989; Muhs et al., 1992; Prentice, 1989; Prentice and Kelson, 2006). This calculation uses the surface uplift rates as a proxy for the rate of stream incision and an assumed hillslope failure angle of 30°. Under these conditions, the maximum elevation of preserved marine terraces is predicted to be ∼430 m, slightly higher than the maximum platform height we observed (405 m).
Implied Surface Uplift Rate
The analysis revealed a suite of markers that, if confirmed to be marine terraces and successfully dated, would yield a rich record of surface uplift rates and their potential variations in space and time. Unfortunately, no absolute ages are available for the platforms we mapped, thus our analysis does not yield independent surface uplift rates. However, because the SCM analysis revealed a large number of platforms, it was possible to use the platform elevations to explore the surface uplift rate that best explains their vertical distribution (Fig. 9). We emphasize that the following analysis is exploratory in nature, and further work is needed to test the assumptions upon which it relies. In particular, the analysis assumes that the rate of surface uplift in the study area has been uniform in both space and time. If this was the case, then dividing the observed platform elevations by an estimated surface uplift rate will yield a set of predicted platform ages that can be compared with the times of known sea-level highstands.
We began the rate analysis by assuming a range of potential surface uplift rates between 0.05 and 1.2 mm/yr and then combined each rate in this range with the platform heights to predict an age for each platform. We then quantified the degree of correlation between the predicted platform ages and the known history of sea-level highstands. Finally, we determined a best-fit surface uplift rate by finding the rate that maximized the match between the predicted platform ages and the sea-level curve.
We assumed that surface uplift rates were uniform through time and that the platforms most likely formed after 3 Ma. We selected a 3 Ma maximum age because at that time the shoreline in the Coast Range was located east of the study area (Lock et al., 2006), as indicated by marine deposits of the Ohlson Ranch Formation (Higgins, 1960) that contain a tuff dated as 3.3 ± 0.8 Ma (Prentice, 1989). However, the platforms could be older than 3 Ma, because at that time the Gualala block would have been ∼50 km south of its present location. We picked highstand ages from a 3 Ma sea-level curve for North America and assigned each an error of ±5 ka to account for potentially fast (<5 ka) changes in sea level (Bintanja and van de Wal, 2008). For platform heights we used those of the 21 elevation bands reported in Table 2, using the height uncertainties to estimate maximum and minimum ages.
To compare known highstand and model-predicted platform ages, we used a simplified approach to quantify their degree of correlation. In particular, we created equally weighted synthetic curves for both the known highstand and model-predicted platform ages in which values decrease linearly from 1 to 0 from the age to each of the uncertainty bounds (Fig. 9A). We then summed the areas where the two curves overlap to obtain a single number that quantifies the degree to which the curves match for a given surface uplift rate (Fig. 9B). We repeated this process for different rates, effectively stretching or compressing the curve of predicted platform ages relative to that for the highstands. To identify the matches between the sea-level highstand ages and the modeled platform ages, we plotted the values representing the degree of curve overlap versus the surface uplift rate (Fig. 9C). The resulting curve has peaks at surface uplift rates of 0.20, 0.63, and 1.18 mm/yr (Fig. 9C), indicating the best matches within the context of our simplified curve-correlation approach.
To evaluate these matches more robustly, we used a method suggested by Horsfield (1975), in which present-day marine terrace elevations are correlated with a sea-level curve to predict surface uplift rates (Fig. 10; see also Lajoie, 1986; Merritts and Bull, 1989). In this approach, the platform heights are plotted on the y-axis and the sea-level curve on the x-axis to create a surface uplift rate–prediction diagram, on which the slope of lines connecting marine terrace elevations with peaks in the sea-level curve represent the surface uplift rate (Fig. 11; Horsfield, 1975; Lajoie, 1986; Merritts and Bull, 1989). If the surface uplift rate correlates well, then the lines that represent the surface uplift rate should intersect peaks of the sea-level curve.
Of the 3 best-matching rates (0.20, 0.63, and 1.18 mm/yr), a surface uplift rate of 0.2 mm/yr over the past 2 Ma provided the highest correlation between terraces and sea-level highstands, with only a few platforms failing to correspond to peaks in the sea-level curve (Fig. 10A). For this analysis, we included the four lowest marine terraces reported in previous work (Crosby et al., 2007). The 100 m marine terrace is one such elevation that does not match well with the sea-level curve at a rate of 0.2 mm/yr (Fig. 10A). All the other lines that do not correspond with peaks have large errors associated with their elevations or are secondary peaks on the plot of platform elevations (Fig. 7B). A rate of 0.2 mm/yr is at the low end of the previously estimated surface uplift rates of ∼0.2–0.6 mm/yr, based on the elevations and ages of the youngest marine terraces in the area (Crosby et al., 2007; Merritts and Bull, 1989; Muhs et al., 1992; Prentice and Kelson, 2006).
In addition, we found that surface uplift rates of 0.63 mm/yr and 1.18 mm/yr do not correlate well with the sea-level curve (Figs. 10B, 10C). Rates >0.6 mm/yr are essentially too fast, causing multiple platform elevations to correlate to a single sea-level highstand (Figs. 10B, 10C). Although coseismically driven surface uplift could produce multiple terraces for a single highstand, to do so requires surface uplift of ∼5 m per event; this seems unreasonable considering that the maximum surface uplift per earthquake appears to be ∼2.8 m in this region (Merritts, 1996).
The analysis in the previous section assumed that the platforms underwent spatially and temporally uniform rates of surface uplift. If this was not the case, then a single tilted marine terrace could produce multiple elevation bands (i.e., in the case of tilting, the number of observed elevation bands may be larger than the true number of different terrace generations). More specifically, if tilted, the same-aged platform will have different elevations in different parts of the study area and therefore could appear to be two different terrace generations if not contiguous along strike.
To test for local tilt, we examined platforms in the elevation range of 60–225 m (∼200–750 ft) along a 10.7-km- (35,000-ft) long section of coastline near Point Arena (Fig. 8). We selected this area because it preserves the largest range of platform elevations and has good platform preservation, particularly at lower elevations. First we rotated the data to obtain an east-west strike for the SAF, and then binned the distance along strike into 15.24 m (50 ft) increments, with the long axes of the bins perpendicular to the SAF (Fig. 11A). We then extracted elevation values (integers in feet) from within each mapped platform crossed by the bin and computed the frequency with which a given elevation value occurs within the bin. We generated a grid in which the x-axis was the along-fault distance, the y-axis was elevation, and pixel values within the grid were the frequency with which that elevation value occurred within the bin (Fig. 11B). On this plot, concentrations of frequency values will reflect a platform surface, and a line through those points will yield its slope along the strike of the SAF (Fig. 11B). The data suggest that the platform surfaces within the 10.7-km-long subset of the study area locally tilt north by ∼0.079° (Fig. 11C).
The significance of this tilt and the extent to which it violates the assumption required to estimate the uniform uplift rate remain unclear. One possibility is that the tilting reflects local faulting or folding in the vicinity of Point Arena and does not affect the whole area. Another possibility is that there are other similar tilts along strike, but that they vary in direction to define a subtle set of low-amplitude folds. The most problematic scenario is one in which the tilt is regional scale, and extends with the same magnitude (0.079°) and direction (north-side down) across the entire study area, in which case a 150 m surface at Point Arena should be 247 m high near Fort Ross State Park, 70 km to the south. If this terrace is ca. 500 ka, based on a 0.3 mm/yr uplift rate at Point Arena derived from coral ages from the ∼23 m terrace, then the surface uplift rate at Fort Ross would be ∼0.5 mm/yr. The impact of tilt in this hypothetical scenario is large because of the large distance along strike over which it is applied (∼70 km), and not because of the magnitude of tilt, which is actually small (0.079°). More work is needed to investigate the spatial extent of the gentle tilt observed at the north end of the study area, to explore the detailed geometry of the platforms along the coastline, and to evaluate the significance of such patterns in terms of local or regional structure.
Relationship of Surface Uplift to Tectonic Models
Rock and surface uplift along the coast of northern California (Fig. 1) is typically attributed to the Mendocino Triple Junction and an associated slab window beneath North America (e.g., Furlong and Govers, 1999; Furlong and Schwartz, 2004; Lock et al., 2006; McLaughlin et al., 1983; Merritts, 1996; Merritts and Bull, 1989; Trehu and Mendocino Working Group, 1995; Zandt and Furlong, 1982). In such models, the pattern of surface uplift both varies along strike of the San Andreas fault and is locked to the triple junction. Because the triple junction moves relative to North America, the uplift pattern will slide along the margin. More specifically, as the triple junction and slab window move relative to the North American plate, points on the plate will migrate from north to south through spatially variable patterns of surface uplift, causing the uplift rate to vary over time (Figs. 12A, 12B). Models of this process, such as the Mendocino crustal conveyor (Furlong and Govers, 1999), predict temporally and spatially varying surface uplift rates that can differ by ∼3 mm/yr over a distance of 400 km along strike (Fig. 12).
It is important to note, however, that triple junction migration does not readily explain surface uplift of marine terraces west of the San Andreas fault because points on the Gualala block and Pacific plate should not move relative to the Mendocino triple junction, assuming the plate is rigid and the boundary is fixed relative to the plate interior (Fig. 12C). In addition, such points do not overlie the slab window, which should be east of the SAF. As the schematic diagram in Figure 12C shows, a passive marker (e.g., the gray box) on the Pacific plate will remain the same distance from the Mendocino Triple Junction and the slab window over time. Because the Pacific plate is separated from North America by a transform boundary, it should be possible for rock and surface uplift of North America to occur independently of the Pacific.
Transpression provides a more likely explanation for surface uplift in the study area. Models of transpressional deformation predict more temporally and spatially uniform uplift rates (<0.5 mm/yr; Anderson, 1994; Bennett et al., 2003; Freymueller et al., 1999; Spotila et al., 2007) produced by a component of Pacific–North American relative plate motion that is perpendicular to, and convergent across, the northern SAF.
We developed and evaluated a semiautomated SCM for analysis of LiDAR data to identify potential terraces on the basis of their low slope, low relief, and geomorphic context. We refer to the features identified by the SCM as platforms, a term we use nongenetically. Our analysis revealed a regionally extensive flight of different platforms at elevations of 100–450 m along the northern coast of California. These platforms contain 21 distinct elevation bands (Table 2). Geologic evidence consistent with the interpretation of at least some of these platforms as marine terraces includes (1) a lack of vegetation or the presence of pygmy forests, (2) locally abundant well-rounded, well-polished clasts of varying compositions, (3) clasts of radiolarian chert sourced from Franciscan on platforms west of the fault and, and (4) beveled outcrops of German Rancho Formation overlain by thin layers of unconsolidated clastic deposits with sandy matrix and well-rounded clasts.
Comparison of the distribution of platform elevations with sea-level highstands reveals that rates faster than 0.6 mm/yr are unlikely in the study area. The best-fit surface uplift rate we find is 0.2 mm/yr for the past 2 Ma, at the low end of previously estimated rates for the area, which are ∼0.2–0.6 mm/yr, based on the elevations and known or inferred ages of the youngest marine terraces in the area (Crosby et al., 2007; Merritts and Bull, 1989; Muhs et al., 1992; Prentice and Kelson, 2006). In addition, we find that platforms are tilted gently northward by 0.079° on the Gualala block west of the SAF in the 10.7-km-long section of coastline near Point Arena, in the northernmost part of the study area. We interpret this tilting to reflect short-wavelength (5–10 km) deformation of the terraces in this part of the study area due to folding or fault block rotation. Further detailed work is needed to determine the geometry and extent of such tilting. If such tilting is found to be regionally extensive, then the number of platform generations (21) and uplift rate (0.2 mm/yr) determined here will need to be revisited. This study demonstrates that LiDAR data reveal a long-lived geomorphic record of surface uplift along the coast of northern California, suggesting that future dating efforts could provide a rich record of surface uplift in the region.
This paper is dedicated to Kurt Frankel. Supported by National Aeronautics and Space Administration grant EOS/03-0663-0306, National Science Foundation CI-Team grant OCI-0753407, the Northern California Geological Society, the University of California Davis Engineering Scholarship program, and the University of California Davis Department of Geology Durrell Fund. We thank Harvey Kelsey and an anonymous reviewer for their comments and Carol Prentice for her assistance in the field and knowledge of the area. Bowles thanks the staff and rangers at Fort Ross and Salt Point State Parks, Philip Mooney for field assistance, Mike Oskin for ArcGIS assistance, and Erin Burkett for her MATLAB expertise and help in coding the frequency analysis. Early drafts benefited from reviews by Mike Oskin, Louise Kellogg, Ryan Gold, Adam Forte, and Peter Gold.