Tracking the behavior of rocky coastal cliffs in northwestern Spain

Rocky coastal regions are often inaccessible due to steep slopes and high relief. Remotely sensed data can, therefore, be useful, but they often have low spatial and temporal resolution and, in the case of airborne LiDAR, if not publically available, are costly to obtain. This paper reports on the use of high-resolution images from unmanned aerial vehicles (UAVs) and Structure-from-Motion (SfM) photogrammetric techniques, supplemented by a series of orthophotos and aerial LiDAR, to examine changes in rocky coastal cliffs from 2002 to 2018. The study was conducted over an 800 m-long, orthogneiss-dominated coastal section in northwestern Galicia, Spain. Cliff changes are due, primarily, to rockfalls, resulting from weathering and wave undercutting, which cause talus deposits to accumulate at the cliff foot. These deposits provide temporary protection to the cliff from wave action, until destabilized and removed by wave erosion and shallow landslides. Cliff recession rates are affected by changing conditions within a cycle and are dependent, in part, on when a survey is conducted. The data suggest that rates of cliff recession are increasing in this region and that the plan shape of the coast, which consists of headlands and bays, is continuing to evolve. Most coastal landslides in this region help to transport and dispose of talus fallen from the cliff. In contrast to landslides that remove intact materials from the cliff face, and are, therefore, primary erosional mechanisms, talus landslides are triggered by wave erosion and probably simultaneously, by storm wave spray and splash, rather than by heavy rainfall.


Introduction
Rock coasts are being impacted increasingly by human activities and infrastructure, which are, in turn, at growing risk from natural hazards promoted by climate change, including the effects of rising sea level and possibly increased storminess (Del Río and Gracia 2009;Trenhaile 2014a; Moore and Davis 2015;Vann Jones et al. 2015;Alessio and Keller 2020). Risk assessment, hazard mitigation, and coastal management in these environments require appropriate data on mass movements and rates of cliff recession, and the formulation of inventories and data-bases that record these events (Nandi and Shakoor 2010;Arbanas et al. 2014;Steger et al. 2017;Esposito et al. 2017;Valenzuela et al. 2018).
This paper is concerned with rocky, coastal cliffs in northwestern Spain. Although not specifically directed towards managerial issues or concerns, the type of data that were collected, and the analysis that was conducted, are fundamental precursors for hazard mitigation and risk assessment on this coast. It has been difficult in the past to acquire the requisite data from rocky coasts that are inaccessible due to steep, unstable slopes and high cliffs. The ability of unmanned aerial vehicles (UAVs) to fly at low altitudes and to traverse difficult terrain provides new opportunities to monitor the behavior of these coasts and, when combined with Structurefrom-Motion (SfM) photogrammetric techniques, to perform 1 3 757 Page 2 of 18 very-high-precision analyses at fairly low cost (Muñoz Narciso et al. 2017;Warrick et al., 2019;Medjkane et al. 2018;Westoby et al. 2018;Horacio et al. 2019;Buchanan et al. 2020). The purpose of this study was twofold, to assess the utility of UAVs to monitor changes to coastal cliffs, and to use the data, supplemented by air photographs and LiDAR, to provide insights into rates and patterns of slope modification on rock coasts. Consequently, there were two study timeframes, a primary one based on the analysis of highresolution UAV images from 2016 and 2018, and a secondary or extended period from 2002 to 2018, based on the additional data from air photographs (orthophotos).

Study area
The study was conducted along an 800 m section of the coast in northwestern Galicia, northwestern Spain (Fig. 1). Most of the area consists of a shallow bay between two headlands, backed by cliffs that are generally more than 50 m high and fronted by sandy beaches interspersed with boulders and areas of bare rock (Blanco Chao and Pérez Alberti, 1996). Exposed rock surfaces, which are generally restricted to the upper parts of the cliffs, slope seawards at gradients of > 75°. The presence in many places of talus reduces the gradients in the middle to lower portions of the cliffs to 30-60° and, where the debris is being transported away from the cliffs, to 5-16° near the base. Cliff profiles tend to be roughly linear although their shape may be punctuated by steep scarps and associated depressions. Near-vertical wave-eroded scarps or concave ramps are also quite common in the talus at the cliff base, and convex slope elements in the exposed rock near their summits. Mass movements occur as rockfalls from bare rock surfaces and as shallow slides, generally originating in the middle to lower portion of the profiles, in the wave-eroded debris (Fig. 2). The study area is dominated by densely jointed and intensely fractured and weathered biotitic-muscovitic orthogneisses, with a zone in the extreme southwest of schists and paragneisses of medium and high metamorphic grade, and small porphyritic granite dikes. Most analysis in this paper was directed towards six discrete sectors, labeled A-F, where mass movements appeared to be particularly active during the primary 2016-2018 study period (Fig. 1d).
The northwestern coast of Galicia is mesotidal, with an average range of approximately 2 m and a maximum range of 4 m (Puertos del Estado 2020). It is the most energetic coast on the Iberian Peninsula, with waves that can exceed 10 m in height during winter storms (Puertos del Estado 2020). Northwesterly waves predominate, especially during the most energetic events (Fig. 3a). The cliffs in sectors B, C, and D are oriented towards these dominant storm waves, but sectors A, E, and F are more sheltered, due not only to their orientation but also because of the protective effect of the adjacent headlands. Simulated (SIMAR) data from a station (3030038) approximately 2.19 km from the study area were used to represent wave conditions over the extended time period (Fig. 1b). To check their validity, the simulated data were correlated against the corresponding values from the two closest wave buoys; each located about 77 km from the study area (Puertos del Estado, 2020). The non-parametric Spearman's Rank correlation coefficients (ρs or rho), 0.914 for Villano-Sisargas and 0.904 for Estaca de Bares, were statistically significant (p ≤ 0.05). Significant wave height (Hs) was > 4 m for 5.54% of the 2016-2018 study period, the highest waves being especially prominent during winter, when Hs > 5 m was exceeded for 3.79% of the time (Fig. 3a).
A storm event was defined as a period, of at least 6 or 12 h, when the mean hourly wave height exceeded the 95th percentile of the hourly wave height over the entire study period (in this case H s.95 > 4.11 m). The average interval between storm events was 14 days, with shorter intervals in winter and especially in January 2018 (Table 1). The simulated data suggested that the stormiest periods were 2014-2016 and 2017-2018. The highest waves occurred most frequently from 2008 to 2010, with a general tendency for a higher frequency of high waves (H s > 5.5 m) before rather than after 2014 (Table 1; Fig. 3b).
Meteorological data for the extended study period were obtained from the Meteogalicia station at Ferrol, about 8.5 km from the study area (Meteogalicia, 2020). Annual rainfall fluctuated between approximately 850 and 1750 mm, with a tendency for the wettest years to have been before 2007 and after 2012 (Fig. 3c). To assess the effect of persistently wet conditions on mass movement activity, the precipitation data were analyzed to determine the total amount of rain falling over consecutive, discrete, or overlapping, 5-or 10-day periods (5-or 10-day running totals). Rainfall totals were represented by the number of days within 5-or 10-day intervals that exceeded the specified totals. Rainfall frequency and intensity, adjusted for differences in the number of days in each period, were broadly similar, with notable exceptions being 2017-2018 (period VII), which had far fewer 5-day wet events than other periods, and 2014-2016 (period V), which had a particularly high frequency of 10-day periods with more than 50 mm of rainfall (Table 2). Soil moisture data were also available for 2012-2018 from the Meteogalicia station at O Val, about 4 km from the study area (data were not collected before 2012) (Meteogalicia, 2020). The data from this station exhibited strong seasonal behavior, with minimum values in summer and maximum values in winter. Values were closely related to rainfall with soil moisture being highest in the wet winters of 2013.

Materials and methods
The primary source of data for this study was two UAV flights in May 2016 and September 2018. The first flight utilized a 20.0 MP ILCE-6000 camera mounted on a Microdrones UAV model md4-200, and the second a FC6310 camera, with the same resolution as on the first flight, on a DJI Phantom 4 Pro UAV (Table 3). Flight planning and data collection were based on the study area, ground control points (GCPs), and the camera. Both UAV flights were made at a height of 40 m above the ground. The frontal and lateral overlap between images was 80% and each pixel in the study zone had at least 5 images. A Stonex S8 GNSS receiver was used to locate GCPs, which had a vertical accuracy of 10 mm and a horizontal accuracy of 5 mm.
The UAV data were used to produce two digital surface models (DSMs) with Pix4dMapper (2014) software. The images were processed and filtered to remove erroneous data and damaged files, and processed, using SfM (Structurefrom-Motion) techniques, to transform the information into The slide in sector C is at the mouth of a valley and on a much lower part of the cliff. The distribution of pixels affected by surface lowering or rising is more complex, possibly due in part to redistribution of loose material by stream water running down the cliff face. Note the scarp behind cliff top in sector B three dimensions in Point Cloud format. To compare cliff morphology in the two flights, a new raster file was created to produce a DEM of Difference (DoD), with positive values corresponding to areas where the surface had been elevated and negative values to areas where the surface had been lowered. A fixed value, equal to the root-mean-square error (RMSE) calculated for the two flights (0.2 m), was used for the Limit of Detection (LoD) (Wheaton et al. 2010), rather than a variable value related to systematic spatial errors (James et al. 2017). ArcGIS 10.5 and Geomorphic Change Detection (GCD) software (Wheaton et al. 2010) were used to record positive and negative changes in elevation, > 0.2 m, in the DSMs for the two flights. Recorded changes in elevation were used to calculate surficial and, based on the number and total area of the 5 cm pixels (the resolution of the 2016 images), volumetric differences in the 6 sectors. In a  few small areas, places apparent changes in elevation were caused by changes in the height of the herbaceous vegetation, which varied according to the time of year and meteorological conditions over the preceding months. In such cases, where the effect of vegetation was clear, the relevant pixels were removed from the analysis. DEMs were also obtained from LiDAR surveys on June 30th, 2010, and September 7th, 2015 (NGI, Instituto Geográfico Nacional) (Table 3), although their resolution was lower than for the UAV images. Four sets of orthophotos, taken from 2008 to 2017, lower-resolution orthophotos from 2005, and rectified air photographs from 2002, were also used in the study to provide additional insights into cliff behavior and to estimate longer-term rates of cliff retreat in this area (Table 3); all air photographs and orthophotos were obtained from the NGI.
Positive (seaward advance) and negative (landward advance) changes in the position of the cliff top were measured for the entire study area. This was accomplished using the Digital Shoreline Analysis System (DSAS) tool in Arc-GIS 10.5 and orthophotos digitized to a scale of 1:500, which provided a suitable degree of detail for accurate Table 2 The number of consecutive 5-or 10-day periods with total rainfall amounts exceeding the listed value *Number of times when rainfall exceeded the listed value (> 150 mm or 200 mm to > 50 mm) over a 5-or 10-day period. Discrete events occurred when the requisite rainfall totals were exceeded over a single 5-or 10-day period. Overlapping events occurred when the running totals continued to exceed the requisite totals over a prolonged period (> 5 or 10 days) of heavy rainfall. Periods between surveys (column 1) are defined in Table 1 Period  I  983  0  3  33  101  0  11  70  150  267  II  1123  5  22  54  122  15  27  86  185  313  III  746  0  0  9  48  0  0  29  82  193  IV  1505  1  7  32  140  0  14  95  214  422  V  608  1  10  21  82  7  13  57  130  225  VI  387  0  0  17  35  0  0  13  57  85  VII  449  0  0  2  20  0  0  8  50  105   Table 3 Technical data for UAV, LiDAR, and air photo surveys RMSE is the root-mean-square error a The DSM after point cloud transformation had a spatial resolution of 0.5 m, and a Limit of Detection (LoD) of 0. measurement. Changes were recorded with reference to the distance of the cliff top from a baseline generated automatically in the beach area; these distances were measured along 320 transects spaced at 5 m intervals along the coast. Changes were recorded over various periods since 2002, using as a common endpoint the location of the cliff edge (cliff top) on the high-resolution UAV images of 2018. Three shoreline change parameters were calculated for each transect: the net shoreline movement (NSM), the change in cliff top location from the beginning to the end of a given period of time (while debris from cliff erosion can extend the shoreline seawards, the effect is temporary and is contrary to the permanent landward retreat of the cliff top due to erosion); the End Point Rate or EPR, an annual rate of change derived by dividing the NSM by the number of years in a record; and the Linear Regression Rate (LRR), obtained by fitting a least-squares regression line to all the shore location/time data (Sytnik et al. 2018;Himmelstoss et al. 2018;USGS 2018). The EPR is largely used in this paper for discussion and comparative purposes. One of the most characteristic elements of coastal cliffs, as viewed in vertical imagery, is their width, defined as the horizontal distance between the location of the slope summit and base. The UAV survey data were used to record positive and negative changes in this parameter, between 2016 and 2018. This required additional measurements to be made to locate the toe of the slope, using the same technique and the same transects as for the cliff top.
A variety of mechanisms can cause portions of the surface of a cliff face to be lowered or elevated. The use of the terms erosion and deposition is therefore restricted in this paper to situations where it is reasonable to assume the origin of any changes in surface elevation, such as due to wave erosion or the deposition of material at the foot of a cliff. Where the cause is less clear, or attributable to other mechanisms, the non-generic terms surface lowering and surface rising (or their contracted forms lowering and rising) are used instead of erosion and deposition. The terms sector and period are used throughout the paper to refer to specific areas (A, B, C, etc.) and intervals (I, II, III, etc.) between aerial surveys, respectively (Table 1; Fig. 1). Although erosional data are provided for each survey interval (I-VII), the emphasis in this paper is between periods of differing duration that terminate in the more reliable, higher resolution data from 2018 (Table 4). This ensures that, in contrast to comparisons between pairs of earlier surveys (say between periods I and II), at least one end point in each pair (for 2018) can be considered to be a fairly precise representation of shoreline location at that time. Furthermore, differences in mean erosion rates over longer and shorter periods, terminating in 2018, are indicative of erosion rates during the earlier part of the longer period (for example, differences in rates

Results
Because of the large number of data-producing pixels from the UAV images (2016 and 2018), (ranging from approximately 250,000 for sector E to almost 2 million for sector D), non-parametric Spearman Rank and parametric Pearson correlation coefficients ranged from -0.87 to -0.95 and were significant (p < 0.01) for the elevation of the cliff face and the distance from the cliff top in each sector. The correlations for the relationship between local gradient and distance from the cliff top were also significant at the p < 0.01 level, but much lower and more variable between sectors (− 0.24 to − 0.34 for sectors A, C, and E, and − 0.02 to − 0.16 for sector D), reflecting the degree of irregularity in the surface of the cliff faces. Although the gradient generally decreased seawards, reflecting a tendency towards profile concavity, it increased in sector B (both correlation coefficients were 0.31), where a more convex profile had a gently sloping segment near the cliff top and a much steeper slope seaward (Fig. 4, profiles B01 to B03).
The UAV data showed that between 2016 and 2018, approximately 76% of the cliff face within the 6 sectors experienced measurable lowering (negative values) or rising (positive values), as expressed by vertical changes in each pixel. Most changes in elevation were fairly small, however, and the − 0.36 m mean value was close to the LoD (Figs. 4, 5, 6). These changes in elevation on the cliff face produced an overall decrease in volume in the 6 sectors of 4048.16 m 3 . The highest percentage loss of material occurred in sectors C (53.8%) and D (71.8%) which, in the latter case, also had the greatest absolute loss (1403.3 m 3 ), in part due its large surface area. There are several possible explanations for increases in elevation and volume in sectors A, E, and F, but the most plausible is due to the decrease in density and increase in volume that occurs when intact rock is eroded and deposited in loose, poorly packed accumulations with numerous inter-clast voids. This promotes surface rising over lowering, so that the latter is more likely than the former to be below the LoD, and therefore to be under-reported.
The changes in elevation and volume recorded from the 2016 and 2018 UAV surveys generally reflected the transfer of material from the central to the lower portions of the cliffs (Fig. 2e). Most changes involved increased basal scarping, with an increase in local slope gradient, by wave erosion at the foot of the cliffs (Fig. 4, profile A02, Fig. 5, profile D04), or the burial and elimination of cliff foot scarps by the further deposition of slope material, resulting in a decrease in local slope gradient (Fig. 4, profile B01; Fig. 6, profile F02).
Although the resolution was much lower than in the UAV surveys, a similar analysis was conducted on the DEMs constructed from LiDAR data from 2010 and 2015. The general shape of the UAV and LiDAR cliff profiles was similar, but there were some significant differences in the distribution of areas of surface rising and lowering in each sector. In comparison with the UAV surveys for 2016-2018, the 2010-2015 LiDAR data suggested that there was previously: less lowering and more rising at the cliff foot in sector A; more rising and less lowering in the upper cliff and less rising at the cliff foot in sector B; rising instead of lowering in the upper cliff in sector C; dominance of rising over lowering in sector D; rising instead of lowering in the lower cliff and lowering instead of rising in the upper cliff in sector E; and dominance of low amounts of rising over almost the entire cliff face in sector F.
The mean EPR was consistently low for each sector over the 2002-2018 period, ranging from 0.01 m yr −1 in sector E to 0.11 m yr −1 in sector D. Rates increased progressively through time, although they occasionally declined temporarily from one period to the next in individual sectors. There was no consistent pattern in the sectors which recorded the highest and lowest rates of erosion during each period. Rates determined from the high-resolution data of 2016 and 2018 were much greater than in the earlier parts of the extended study period, ranging from a high of almost 0.5 m yr −1 in sectors D and E to a low of 0.18 m yr −1 in sector A. Even higher rates of recession were recorded from 2017 to 2018, although they were based in part on lower-resolution orthophotos from 2017 (Table 4). Apart from an area immediately east of sector D, there were no strong regional patterns in cliff recession rates at the individual transect scale. Along the rest of this coast rates varied significantly over short distances, and between adjacent transects, producing a distribution that appeared essentially chaotic in nature (Fig. 7). Transect rates of retreat were frequently from 0 to < 0.01 m yr −1 in the earlier part of the 2002-2018 period, but they had increased to > 0.02 m yr −1 at most transect sites by about 2016, or somewhat earlier, and these rates became increasingly dominant up to 2018. Increases in transect rates of erosion were, therefore, consistent with the trends that were evident in the mean rates at the study area and sector scales (Table 4). The mean change in the width of the cliff face for the entire 320 transects within and between the 6 sectors, and to the immediate east of sector A, was − 0.09 m, signifying a small decrease in width over this two-year period (2016)(2017)(2018), with recession at the cliff top being slightly less than erosion of intact rock or removal of loose debris at the cliff foot. Changes in cliff face width varied considerably in amount and direction with location along the study area, however, with significant increases in width in sector A and between sectors E and F, and much greater reductions in width in sectors B and F (Table 5).  (Table 5), being frequently roughly equal or greater than the mean change in width in each location. There was a particularly high value in sector F due to the addition and depletion of significant amounts of landslide debris from the foot and the lower part of the cliff (Figs. 2c, 6, profiles F01-F05), and a low value in the area immediately east of sector A.
The images were also used to map and record the occurrence of mass movements, primarily landslides, between 2002 and 2018. Movements that were large enough to be identified were recorded for each period, for areas within and adjacent to the study sectors. Most slides were restricted to the lower part, rather than the whole, of a cliff face (Fig. 2), although a few did extend from near the top to the bottom of the slope. Slides were most frequent from 2008 to 2010 (18 events), and 2016 to 2018 (21 events), and least common between 2010 and 2014 (5 events). Movements occurred in all six sectors and in the areas between the sectors between 2002 and 2018 but were least frequent in sectors A and C (1 event in each sector) and to a lesser extent in sector F (3 events).
There was little relationship between rates of cliff recession and a variety of meteorological conditions. Furthermore, while there is evidence to suggest that cliff recession may have accelerated through time, being fastest from 2016 to 2018 and especially 2017 to 2018 (Table 4), this increase was not matched by increases in annual rainfall (Fig. 3c), storm frequency or intensity, (Table 2), or soil moisture (Fig. 3d).

Discussion
The profiles of rocky cliffs have been characterized as essentially passive expressions of elevational differences in erosional susceptibility to marine and subaerial processes (Emery and Kuhn 1982). Cliffs also function as transportation systems, as a component of the rock cycle that delivers the products of erosion and weathering to the upper intertidal zone, where they can be removed by waves. There is a continuum of transportational cliff types defined according to the time that debris is stored at the cliff foot, thereby   protecting it from wave erosion. At the one end are sloping, transport-limited (TL) cliffs that are covered in debris and no longer being eroded by, the sea (Fig. 8a). This condition could arise through a fall in relative sea level, the protection afforded by the growth of a fronting barrier or pebble beach, or construction of breakwalls or other human infrastructure.
At the other end are steep, supply-limited (SL) cliffs where the rapid disposal of fallen debris inhibits sediment storage and permits essentially continuous wave erosion of the cliff base. These cliffs may develop where there are high waves, mobile, fine-grained debris, or slow debris production (Fig. 8b).
Between the TL and SL extremes are cliffs which for convenience will be termed periodically transport-limited (pTL) in this paper. These cliffs experience cyclical development including periods when the cliff foot is exposed and periods when it is protected by talus from wave erosion (Fig. 8c). These types of cliff are cascading systems whereby weathering and oversteepening by waves deliver debris from the cliff face to the cliff foot. This material accumulates, at its angle of repose, until undercut and steepened by wave action, in single or multiple storm events, destabilizing the deposit and promoting its removal by landslides and wave action (McLean and Davidson 1968;Castedo et al. 2017;Trenhaile 2020). In some areas, debris removal may also be facilitated by increasing exposure to wave action as the toe of a growing talus migrates seawards. Talus deposits can result from abrupt and catastrophic cliff failures, the more gradual addition of individual rock fragments with high transport (disposal) thresholds (due to debris size and shape), or low transport capacity and competency in weak or protected wave environments. In addition to talus-dominated periods of transport limitation, pTL cliffs can also be periodically SL in some fairly resistant rocks. This occurs when sediment production essentially ceases, or is maintained at a very low level, while a notch is cut to its maximum supportable depth. SL and TL conditions can also operate simultaneously on pTL cliffs consisting of a variety of rock types. This can occur, for example, where there are alternating beds of limestone and shale, with the latter shedding fine, easily disposable material while thick, structurally strong limestones provide the roofs for deep undercuts and, upon collapse, large debris accumulations (Trenhaile 1972).
Assuming fairly homogeneous geological and morphogenic (waves, tides, etc.) conditions, cliff recession rates and cliff face width would slowly decline to almost zero on TL cliffs, and be quite uniform in SL environments, albeit punctuated by chaotic perturbations, and therefore fairly predictable over time ( Fig. 8a and b). Conversely, there are likely to be marked variations in cliff morphology and rates of erosion in pTL environments. The decline in cliff recession rates would vary according to the rate of debris accumulation, for example, ranging from a gradual reduction due to the addition of individual clasts to an abrupt cessation caused by a sudden, catastrophic slope failure. Similarly, changes in cliff width would mirror the rate of talus growth, with maximum width being attained when the talus extended furthest seawards. Cliff recession would then be essentially zero during the time required to remove the debris, and to destabilize the cliff (Trenhaile 2014b(Trenhaile , 2020 (Fig. 8c).
The predominance of talus suggests that the cliffs in the study area are primarily pTL, although there is also a SL component due to the supply and rapid disposal of small rock fragments at their foot. A distinction can be made, however, between pTL cliffs in the bay that are in different stages of development. In the north, around sector B, the cliffs have talus deposits at their foot, whereas those in the south are essentially bare rock faces with a prominent notch under sectors C and D (Figs. 1c,2d). While the entire bay is well exposed to storm waves from the northwest, wave attenuation over the intertidal and subtidal shore platform in the north is likely to be much greater than over the narrower and predominantly sandy beaches in the south (Fig. 1c and d). While strong waves promote cliff undercutting and collapse, the dominance of notches and steep, bare cliffs in the south suggests that, while talus periodically protects the cliff foot, it is removed much faster than it is produced. Conversely, in the north, more attenuated waves may result in slower cliff erosion and debris removal, with the predominance of talus in this area suggesting that debris removal operates more slowly than debris production.
The erosional data provided conflicting evidence of the effect of talus accumulations at the cliff foot (Table 4). While the lack of talus may have contributed to faster erosion in sectors C and D than in sector B from 2016 to 2018, for example, erosion was faster in sector B than in the other two sectors from 2014 to 2016. In addition to the opposing effects of talus accumulation and wave attenuation rates in these sectors, the lack of consistent relationships between talus deposition and rates of cliff erosion may be attributed to the low temporal resolution of the erosional data and the lack of correspondence between survey intervals and the duration of debris accumulation and removal cycles, resulting in cliff recession rates that are dependent, in part, on when the survey was conducted.
The sectors outside the bay are sheltered from northwesterly storm waves by the two headlands (Fig. 1), and by: a boulder and cobble beach and a wide, largely submarine shore platform in sector A; a sand and boulder beach in sector E; and a wide sandy beach in sector F. The lower part of the cliffs in sectors A and F is draped in debris in various stages of removal, which is consistent with the assumption that debris accumulation and removal are dominant where there is weaker wave action. This assumption must be tempered by the occurrence of an essentially bare and exposed cliff face in section E, however, which may be the least exposed to vigorous wave action in the study area. There are several possible explanations which may account for the lack of talus in this sheltered area, including the occurrence of a porphyritic granite dike which releases fine, easily transported material, and a 24 m-high cliff which produces less debris than the much higher cliffs elsewhere in this region.
An important question regarding the development of rock coasts, which also has implications for coastal management and planning for climate change, is whether the plan shape of irregular, crenulated (headland and bay) coasts trends towards a state of dynamic equilibrium. This possibility is based on the assumption that, as an irregular coast develops, erosion of the more resistant rocks on headlands by increasingly refracted waves must eventually accomplish the same amount of erosion as the weaker, attenuated waves on the less resistant, and often beach-protected, rocks in bays (Muir-Wood, 1971;Trenhaile 1987, pp. 269-270;2002;2019).
There is no evidence that the plan shape of the Galician coast was in equilibrium over the short periods covered in the present study (Table 4). Rates of cliff recession in some sectors were commonly several times greater than in other sectors, even over the 2016-2018 period covered by highresolution UAV imaging. There was no consistent pattern in the occurrence of sectors with particularly high or low recession rates, with the possible exception of persistently lower values in zone A, which is more sheltered from storm waves than most other sectors. There was also little evidence of plan shape equilibrium at the transect scale, for the sectors and areas adjacent to, or between, the sectors. The largest area with essentially uniform cliff recession rates was at the rear of the shallow bay east of sector D, which exhibited relatively rapid retreat in the 2002-2018 and 2016-2018 periods (Fig. 7). The most prominent headlands in the study areas, between sectors A and B in the north and D and E in the south, experienced a range of recession rates that were lower than in the C-D bay from 2016 to 2018, although rates were comparable between the tip of the southern headland and the C-D bay from 2002 to 2018.
The lack of evidence for plan shape equilibrium is not surprising given the limited time covered in this study and the cyclical pTL nature of coastal development in this area. Under such circumstances, it is likely that any effect of differences in wave height and rock resistance between headlands and bays would be obfuscated by the influence of other factors, including the protection or enhanced erosion (through abrasion) afforded by beach material Trenhaile 2016), the drainage and water-holding characteristics of the rock, and possibly changes in sea level and storminess.
Landslides and other mass movements can be triggered by high pore water pressures produced during periods of heavy rainfall or snowmelt (Polemio et al. 2000;Duperret et al. 2005;Massey et al. 2013;Gordo et al. 2019;Li et al. 2019). There is limited evidence of this effect in northwestern Galicia, despite reports in the local press of landslides occurring during winter and spring (Abelairas 2017), including one in sector F in March 2017, following a period when 87.4 mm of rain fell over 10 days (Meteogalicia 2020). The general lack of a relationship between rates of cliff retreat and annual to daily rainfall and groundwater data in this area ( Fig. 3c and d) might be attributed to the poor temporal resolution of the data. The function of most of the slides described in this paper is to transport and help to dispose of previously dislodged and fallen debris, however, rather than to contribute directly to the erosion of intact rock. Furthermore, while heavy rainfall may help to trigger landsliding in the talus deposits, this might be accomplished more effectively by the splash and spray of large storm waves as they simultaneously erode and destabilize the debris toe. Therefore, heavy, or prolonged rainfall may not be the primary driving force for the landslides along this coast.
The coast of northwestern Galicia experiences pTL conditions, with cascading systems operating on the cliff face and on the talus deposits below. Cyclical behavior, therefore, results in different parts of the study area being at different stages of development at the time of each survey. Asynchronous behavior must account, in part, for the spatial and temporal variations in cliff recession rates and cliff width reported in this paper. Although it is difficult, due to wide spatial and temporal variations, and essentially stochastic behavior, to correlate global rates of cliff recession with such intuitively important variables as climate, wave regimes, and rock hardness and structure, values around 0.03 m yr −1 may be considered characteristic of resistant rocks, 0.10 m yr −1 of moderately resistant rocks, and 0.23 m yr −1 of low resistance rocks (Prémaillon et al. 2018;Stephenson et al. 2021). The data discussed in this paper, therefore, suggest that recession rates in the weathered rocks of the study area are broadly characteristic of rocks of similar resistance in other areas. Recent rates, since 2016, are particularly rapid, however, and may reflect, not only the fractured nature of the rocks, but also the exposed nature of this coast and its vigorous wave environment.
It has been suggested that cliff erosion rates will increase due to rising sea level and possibly increased storminess (Dickson et al. 2007;Ashton et al. 2011;Trenhaile 2011Trenhaile , 2014aGómez-Pazo and Pérez-Alberti 2017;Limber et al. 2018), but the data from hard rock coasts are, at present, ambiguous and contradictory (Lee et al. 2001;Hurst et al. 2016;Young 2018;Swirad et al. 2020). There is little evidence that apparently increasing rates of recession in the study area were triggered by increases in wave height or precipitation, and it is questionable whether such a marked increase in rates could be attributed to an approximately 0.06 m rise in global sea level from 2002 to 2018 (Table 4; Fig. 3b and c) (NASA 2020). Consequently, lacking specific evidence to the contrary, one can only conjecture at present that any increase in the rate of cliff recession in this area is the result of perturbations related to cyclical variations in geomorphological activity. Determining whether these cycles are experiencing, and being modified by, the effects of rising sea level and other elements of climate change, which could account for a consistent increase in recession rates with time, will require longer erosional records from the study area and from similar rock coasts than is presently available.

Conclusions
Images obtained from UAVs, supplemented by a series of aerial photographs and aerial LiDAR, were used to study rocky cliff evolution from 2002 to 2018 in northwestern Galicia, Spain. The main conclusions of this research are as follows: a) Cliffs migrate landwards in cycles characterized by a series of stages involving rockfalls, talus sediment storage, destabilization through wave undercutting and landsliding, and renewed cliff exposure and erosion. b) Cliff recession rates vary in time and space due to changing conditions within a cycle and the time when a survey is conducted. c) The data suggest that rates of cliff recession have been progressively increasing during the study period. d) There was no evidence of plan shape equilibrium between headlands and bays in this area; and e) Coastal landslides, which are disposal mechanisms for talus deposits fallen from the steep cliff face, are triggered by wave erosion and probably storm wave spray and splash, rather than by heavy rainfall.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.