Quantification of cliff retreat in coastal Quaternary sediments using anatomical changes in exposed tree roots

Methods of cliff retreat reconstruction

For a precise quantification of the cliff retreat based on root exposure, two parameters are needed: the number of annual rings since exposure (NRex) and the thickness of the eroded soil layer (Er). When a root loses its soil cover, a series of anatomical changes (cell size, tangential growth, compression wood) will occur in terms of ring growth, firstly because of the effect of the exposure itself (e.g. variations in edaphic temperature and humidity, reduction in soil cover pressure), but also because of mechanical stress (e.g. abrasion) that a root will undergo when exposed. As proposed by Corona et al. (2011b), a reduction of cell lumen area in earlywood tracheids by about 50% can be used as a distinct sign of exposure in conifer roots (Figure 3). To calculate the annual erosion rate Era, Er is divided by the number of rings formed since the year of exposure (NRex):

(1)

Sampling strategy and root-ring analysis

At the study site, 56 root cross-sections (six from buried and 50 from exposed roots) were sampled from 16 different roots of 16 P. halepensis Mill. trees, and at a minimum distance of 50 cm from the stem basis. This distance has been chosen as (i) stem movement induced by ongoing growth tends to pull roots upwards (Stokes and Berthier, 2000), and as (ii) roots close to the stem basis and growing near the soil surface often experience bending stress resulting from stem displacement (Watson, 2000), and therefore exhibit asymmetric growth structures in cross-sections. The position of exposed roots with respect to the present cliff surface was documented in detail before the root was actually cut and data recorded on the stratigraphic position, distance of the root section from the tree trunk, aspect, and slope (for a complete description of the method, see Corona et al., 2011b). The horizontal distance between the root and the cliff surface (Er) was determined with a depth gage (accuracy ±1 mm); in cases where the distance between the root and the cliff exceeded 25 cm, a metal ruler (±5 mm) was used instead. Root sample locations were recorded using a Trimble GeoExplorer (with < 1 m accuracy) and roots were then positioned in a geographical information system (GIS; ArcGIS 10.1; Kennedy, 2009) as geo-objects, where erosion rates could be linked as attributes to each single root section.

In the field, the root sampling strategy was defined according to geomorphic units (i.e. rocky point, pocket beach) and to the stratigraphic profile of the cliffs (Figures 2A and 2B). Samples were distributed within the two geomorphic units and the four strata of the stratigraphic profile. In order to determine Nrex, sampled roots were then cut into discs about 2 cm thick and air-dried for ~30 days, before they were prepared for macroscopic analysis (i.e. sanded sequentially with 60, 80, 320, and 600 grit-sanding belts). Ring-width data were obtained on four radii per cross-section using a LINTAB measurement device (Rinn and Jäkel, 1996). Thereafter, we prepared root discs for microscopic analysis. Small cubes were cut from the cross-sections (maximum 2 cm × 4 cm) from which micro-sections where cut with a Reichert sliding microtome (thickness of cuts ~15 μm). The microsections were treated with sodium hypochlorite (NaOCl) solution, deionized water, and soluble safranin before they were dehydrated with alcohol and xylol (Schweingruber, 1978; Arbellay et al., 2012). The microsections were mounted on slides, embedded in Canada balsam, and dried at 60°C for 24 hours. Microsections were then observed and photographed with a digital imaging system under optical microscopy. Measurement of cell lumen area in earlywood tracheids was performed with the semi-automated WinCELL2005 software. Following Rubiales et al. (2008), cell lumen area was determined through an averaging of 12 cell measurements per growth ring. In addition, and in order to detect potential anatomical changes that could occur before root exposure (Corona et al., 2011b) and to assess the maximum depth at which anatomical changes may occur, we compared cell changes in exposed sections with those observed in buried sections embedded into the cliff (−2 to −18 cm, Table 3) (Corona et al., 2011b).

Anatomical changes in roots

The innermost rings of the roots sampled were dated to between ad 1976 (R16.4) and 1999 (R4.3), with a mean root age of 22 years and a standard deviation of ±6 years (Table 3). The wood anatomical structure of the exposed root sections R17.1 and R17.2, shown in Figure 3 for the years 1988–2011 and 1989–2011, respectively, illustrates anatomical changes that occurred in each section after exposure. Both samples were taken half way up the cliff (strata 2 and 3). These roots show thin cell walls and large cell lumina in earlywood tracheids until 1996 (R17.1) and 1999 (R17.2), respectively. From 1997 (2000) onwards, root rings form thicker-walled tracheids and more stem-like wood structures, with very distinct latewood and cell lumen area reductions > 50% (from 2560 to 982 μm², or −62% for R17.1; and from 3600 to 900 μm², or −75%, for R17.2, respectively). Similar sudden and sharp relative reductions in cell lumen area as those described earlier are observed for all exposed roots chosen for analysis between 1991 and 2006 (mean: 2000; Table 3). Conversely, the buried parts of the roots sampled in this study (i.e. R5.3 to R5.5 and R16.5 to R16.7, located at depth ranging from −2 to −18 cm within the strata) are characterized by large tracheids with thin and poorly lignified cell walls as typical for buried roots.

Heights of the exposed parts of the 50 cross-sections of roots – as measured in the field (Er) – varied from 2 to 87 cm (mean: 30.4 cm), representing an overall mean erosion rate of 2.1 ± 1.2 cm yr⁻¹ at the study site.

Variability of erosion rates

At the scale of individual roots, erosion rates varied between 0.6 (R7 and R13) and 3.9 cm yr⁻¹ (R5 and R16, Table 3). The largest erosion values were observed in roots (R3, R4, R5) sampled at Pointe de la Tufière located in the northeast (NE) part of the study site (Figure 4) and where the cliff is exposed directly to waves and storm surges. As a consequence, cliff retreat varies between 2.3 and 3.9 cm yr⁻¹ here. Conversely, those roots sampled on cliffs backing La Courtade pocket beach (R1, R6–R10, R13, R15), sometimes protected by a vegetation buffer (R13, R15), recorded remarkably lower erosion rates ranging from 0.6 to 1.9 cm yr⁻¹.

At the root section scale, heterogeneous erosion rates are reconstructed for sections within the same root. For example, high variability between sections of the same root is observed in R11. Here the ratio between the smallest and largest erosion rates is 1:3.6 (R11.2 in stratum 2 with 0.75 cm yr⁻¹; R11.1 in stratum 3 with 2.7 cm yr⁻¹). Similar differences are observed in R1 with a ratio of 1:3 between R1.1 (stratum 2, 0.4 cm yr⁻¹) and R1.3 (stratum 1, 1.2 cm yr⁻¹). According to the stratigraphic profile, highest erosion rates are reconstructed for root sections located in the upper part of the profile, i.e. in the sandy-gravelly stratum 4 (3.8 ± 0.8 cm yr⁻¹). Conversely, lower (1.8–2 cm yr⁻¹), but more variable, cliff retreat rates are measured in strata 1 (sandy stratum), 2, and 3 (torrential strata, Figure 5).

Reconstruction of cliff profiles

Dendrogeomorphic analyses do not only allow detection of root exposure processes with annual resolution, but also enable reconstruction of past cliff surface positions with centimetric precision. The use of exposed roots intersecting several strata therefore also allows the reconstruction of the spatio-temporal evolution of cliff profiles, as exemplified in Figure 6. The combined analysis of root positions and years of exposure thus provides a chronological framework to study cliff evolution and permits determination of erosion processes. By way of example, erosion at the level of root R3 occurred in three stages. The first stage corresponds to the first year of root exposure in 1991 at the level of sample R3.6, where anatomical changes can be observed in the cell structure of the root. This portion of the root was located in a top layer of fine sand (stratum 1). By contrast, no significant cell variations are observed in the other sections in that year. The second stage occurred in 1994 when anatomical changes start to occur in sections R3.2 to R3.5. This stage corresponds to a marked retreat in strata 1, 2, and 3, and is interpreted as the impact of a small landslide out of the notch in the weak sandy stratum. The third stage was initiated in 1996 and corresponds to a slow evolution of the top of the profile (stratum 4) as revealed by anatomical changes in R3.1 and to a delayed exposure of the foot cliff revealed by the exposure of R3.8 in 1998.

Occurrence of anatomical changes in buried roots

In this study, we investigate the wood anatomical reaction of roots of P. halepensis Mill. to denudation and determine the timing of cliff retreat in the sandy-gravelly cliffs of Porquerolles Island. Anatomical reactions in roots have been used repeatedly to assess exposure dates, but never so far in a marine context. Recent work by Corona et al. (2011a, 2011b) in marly badlands of the southern French Alps demonstrated that changes in root cell anatomy and the related reduction of tracheid cell lumen area start to emerge as soon as the soil is reduced to about 3 cm, thus resulting in a bias of reconstructed sheet erosion. In order to ascertain for the existence of this bias in coastal cliffs, we based our analysis on a systematic and high resolution, quantitative assessment of changes in cell lumen area in exposed as well as buried roots sampled at various depths below the current cliff surface (Table 3).

Results demonstrate that the reduction in cell lumen area of earlywood tracheids does not occur in buried sections of the root, thus confirming that the reduction in the earlywood tracheids lumen area corresponds exactly to the year at which the root section was exposed. At our study site, this absence of a bias can be explained by the milder climatic conditions of Porquerolles Island where frost is unusual. In the southern French Alps, by contrast, the triggering of anatomical changes in buried roots is attributed to freeze–thaw cycles that increase the vulnerability of the xylem to cavitation (Pittermann and Sperry, 2003). Also the results demonstrate that the abrupt cellular metamorphosis observed after exposure is induced by instantaneous erosion processes. Similarly, sharp reductions of the cell sizes in roots have been observed in Fraxinus excelsior L. root sections exposed along riverbanks in the Swiss Alps (Hitz et al., 2008a). At the same time, values found in the present study clearly differ from the gradual evolution of cell lumen areas observed in the marly badlands of the southern French Alps (Corona et al., 2011b). Such differences result from the nature of erosion processes involved in root exposure at each of the regions: i.e. continuous and regular erosion processes such as rill-wash in the badlands of Draix (Rovéra and Robert, 2005), and the sudden processes such as landslides or wall collapses at Porquerolles Island, or debris-flow events in the Swiss torrents.

Reliability of erosion rates derived from dendrogeomorphic measurements

At Porquerolles, average erosion rates derived from root-ring analysis of P. halepensis Mill. are reaching rates up to 20 times higher than those calculated using tree roots in continental environments (soil erosion, bank erosion, and gully retreat) (Table 2). Such discrepancies are probably related to the high sensitivity of the soft sandy-marly deposits to sudden erosion processes resulting from the mechanic impact of waves and swells at the foot of the cliffs. In detail, erosion rates are highest in the top (stratum 4) and basal (stratum 1) sandy strata (≥ 2 cm yr⁻¹). Similar results have been obtained at Carry-le-Rouet (Provence, France), where erosion rates three times higher in sandy-marly deposits than in calcarenite or conglomeratic deposits (Premaillon et al., 2016), thus confirming that soft lithologies are more affected by erosion than indurated deposits (as also reported elsewhere by Moore and Griggs, 2002; Collins and Sitar, 2008; Lee, 2008; Young et al., 2009b).

Erosion rates at our site vary between 0.6 and 3.9 cm yr⁻¹, with maxima observed on rocky points (at the root scale) and in stratum 4 (at the root section scale). By contrast the lowest retreat rates are reconstructed in those portions of the cliff protected by La Courtade pocket beach. These results are in line with Earlie et al. (2015a) who observed strong spatial variability (0.01–0.37 m yr⁻¹) in the recession rates of cliffs located on the south-western UK peninsula in relation with varying boundary conditions (i.e. rock mass characteristics, cliff geometries, beach morphology) and forcing parameters (i.e. significant wave height and peak wave period) (Earlie et al., 2015b).

Quite interestingly, and despite their wide geographic extent over the Mediterranean region, only a very limited literature exists on rock coast recession, especially in soft rock lithologies (Earlie et al., 2015b). Giuliano (2015), for instance, estimated erosion rates in Mediterranean sandy-gravelly cliffs of the Massif de la Nerthe (Carry-le-Rouet, France) through the comparison of (i) yearly-resolved LiDAR data with (ii) a multi-decadal, diachronic analysis of orthorectified aerial photographs covering the period 1924–2011. Cliff retreats derived from both methods are 1.1 cm yr⁻¹ over a 17 month period and 4 ± 0.3 cm yr⁻¹ over a 88-year period, respectively (Giuliano, 2015). These erosion rates are within the range of erosion rates reconstructed from exposed roots at Porquerolles Island (0.6–3.9 cm yr⁻¹). Our results are also comparable with cliff retreat rates (2–6 cm yr⁻¹) derived from diachronic analyses in the Calanque des Maures (France, Figure 1) on similar lithology (Giuliano, 2015). Apart from the methods used to derive erosion rates (i.e. photogrammetry, LiDAR and diachronic analysis versus root exposure), discrepancies between Porquerolles and Carry-le-Rouet may result from the different resolution of both studies. While Giuliano (2015) computed quasi-continuous cliff retreat rates for a 1-km long coastline at Carry-Le-Rouet, our study site is restricted to La Tufière headland and two locations points along one beach and does not exceed a total length of 150 m.

Noteworthy, and in a Mediterranean context (Figure 7, Table 1) the cliff retreat rates derived from dendrogeomorphic reconstructions are much smaller than those measured in sandy-silty environments of Greece (50 cm yr⁻¹, Xeidakis et al., 2007), in bio-calcarenites of Italy (20 cm yr⁻¹, Andriani and Walsh, 2007), and in the quartzose environments of Israel (20 cm yr⁻¹, Zviely and Klein, 2004) (Table 1, Figure 7). At the hemispheric scale, by contrast, our results are in the same order of magnitude as erosion rates derived from 257 photogrammetric analyses on sandstone cliffs in the Algarve region (1.9 cm yr⁻¹, Marques, 1997), North Yorkshire (3 cm yr⁻¹, Agar, 1960) or the Larache region (Morocco, 8 cm yr⁻¹, Marques, 2003). Conversely, they are low compared to those available for California on sandstone and sand-claystone cliffs (15–30 cm yr⁻¹, Moore and Griggs, 2002; Young et al., 2009b; Table 1, Figure 7) and for macrotidal cliffs in northern Europe (10–50 cm yr⁻¹) (see e.g. Dewez et al., 2007; Lim et al., 2010; Dewez et al., 2013; Letortu et al., 2014; Michoud et al., 2015).

However, even if the Mediterranean context is not prone to frequent and spectacular cliff collapse events like those of the chalk cliffs of northern Europe (e.g. Duperret et al., 2002; Costa et al., 2004; Regard et al., 2012; Dewez et al., 2013), acceptability of risk in the vicinity of cliffs is subjected to increasing residential, touristic, and economic pressures (Giuliano, 2015).

Contributions and limitations of the approach

The centimetric resolution of dendrogeomorphic reconstructions and the multi-decadal time windows typically covered by roots facilitate the comparison of averaged erosion rates with meteorological records. Dendrogeomorphic analyses also require less time to be accomplished and exhibit a much better cost–benefit ratio than most other techniques used to infer erosion (Table 4). The analysis of sequential aerial photography and satellite imagery, sometimes supplemented by historic maps, has remained the main method to reconstruct decadal-scale records of cliff recession (Table 1). These sources cover roughly the same time windows (up to one century) as tree roots but at a much lower spatial resolution as a result of (i) shrinkage and stretching of the physical document over time, in addition to (ii) general issues of accuracy and precision associated with map production in the case of old maps (Snyder, 1987); as well as a result of (iii) the intervals between image acquisition, in the case of aerial surveys, that do not permit to detect overly slow or very rapid changes along rock coasts (Trenhaile, 2011). The dendrogeomorphic approach may be also considered as very complementary to the shorter time series obtained through repeat monitoring and on long-term erosion rates derived from radioisotopes (Regard et al., 2012). Erosion pins, for example, have proven useful in changes caused by weathering effects, but remain limited spatially as the surface of surveyed plots rarely exceeds a few square meters and furthermore monitoring of the nail networks is most often limited to a few years (Lim et al., 2005). Similarly, terrestrial LiDAR provides resources to study the spatial distribution of sea-cliff activity and erosional processes at sub-annual timescales (e.g. Lim et al., 2005; Young et al., 2009a; Lim et al., 2010; Quinn et al., 2010; Young et al., 2011; Barlow et al., 2012), but typically only over very short periods that does not exceed a few years. Similarly, terrestrial laser scanning (TLS) from the beach or shore platform can monitor changes in the cliff face, including the detachment of rock fragments of only a few centimeters in size up to large falls, slides, and flows (Rosser et al., 2005) though facilitating high resolution three-dimensional (3D) mapping of sea cliff morphologies. Similarly, high precision measurements such as repeat drone surveys coupled with seismic and/or wave data or video surveillance, enable sea–cliff interactions to be captured (see e.g. Letortu et al., 2018). Yet, such a quantitative characterization of sea–cliff activity remains a challenging task mainly due to the frequent sampling intervals required to capture such short duration, spatially heterogeneous and often non-linear natural phenomena (Katz and Mushkin, 2013). On larger timescales, typically in the range 100–10 000 years, the use of radio-nuclides (e.g. Regard et al., 2012) provides important time-averaged constraints for the cumulative geomorphic effect of the suite of erosional processes that drive coastal cliff retreat, but possibly lacks the temporal resolution to identify causes and drivers of erosion. In all of the earlier examples, the replication of measurements and spatial resolution of results are often hampered by the cost of measurements and heavy instrumentation. On a spatial plan, our study refines future sampling protocols in the cliff context. Reconstructions will be more accurate if samples are acquired carefully every 10 cm on the same root. This first study also reveals that, in the case of cliffs, sub-horizontal root sections represent the most valuable specimens to establish mean erosion rates for a given stratum. In addition, vertical sections can be used as complementary data in order to reconstruct the evolution of the cliff profile. Documentation of such detailed patterns of annual cliff profile evolution will enable comparison of rates of geomorphic processes with hydrodynamic and climatic events that have affected the major study area over the past decades.

In summary, the main advantages of the dendrogeomorphic approach to quantify cliff retreat relate to (i) the quantification of cliff retreat in undocumented areas, (ii) with annual resolution. The approach is complementary of the shorter time series obtained with repeat monitoring or more coarsely resolved records obtained from aerial photographs or cosmogenic radio-nuclide as it enables (iii) to quantify cliff retreat rates with a centimetric resolution and to reconstruct the evolution of the cliff geometry, at the plot scale, for intermediate time window (up to one century) thus offering (iv) the possibility to infer micro-geomorphic and climatic controls on the timing of cliff retreat.

Conversely, the key limitation of root-based analyses of erosion is related to the presence of trees and shrubs in the study area and to the age of roots available for analysis. Dendrogeomorphic reconstructions of erosion rates are also limited to partly exposed, alive roots with growing tips still in the ground (Krause and Eckstein, 1993; Stoffel et al., 2013). Indeed, the water absorbing part – or terminal system – of the root needs to remain underground to work properly and to prevent the death of the root. It implies that an exposed part of a living root is older than the erosion process (Vandekerckhove et al., 2001; Hitz et al., 2008a). In addition, the method relies on a sufficient number of exposed trees and roots (two or three roots per tree, five to 10 trees) and data processing requires the destruction of samples. Furthermore, recent advances in vegetation-based reconstructions of erosion indicate that large exposed roots would underestimate the real values of the erosion rates (Haubrock et al., 2009; Bodoque et al., 2011; Stoffel et al., 2013). In a similar way, roots exposed over different periods in time and/or showing different ages may exhibit significant discrepancies in mean erosion rates. Finally, many of the ecosystems in regions with distinct seasons and the presence of trees are clearly dominated by broadleaved species (angiosperms). Despite their abundance, not least in areas affected by erosion, they have only rarely been used so far to reconstruct erosion processes, presumably as a result of their more complex wood anatomy and the existence of frequent growth anomalies (Cherubini et al., 2003). Some of the limitations of reconstructions with Mediterranean broadleaves have been illustrated by Bodoque et al. (2005). Roots of broadleaved trees growing in the more temperate climatic zones of Europe appear to be less affected by cambium stress and the formation of false or double rings and therefore have been used occasionally to infer erosion processes in the past (Malik, 2006; Hitz et al., 2008a, 2008b). Based on the earlier limitations and in an attempt to increase accuracy of reconstructions (while reducing the effect of biases), further calibration and validation of dendrogeomorphic results are needed on sites where erosion rates are monitored continuously and with high accuracy.