the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Apr 2024
Using UAV-Derived Point Clouds to measure high resolution cliff dynamics in soft lithologies: Demons Bluff, Victoria, Australia
Abstract. Unoccupied aerial vehicles (UAVs) have revolutionised data collection on the Earth’s surface. Through aerial photogrammetry, very high-resolution digital surface models can be produced enabling contemporary research on landscape stability. There are however significant limitations in the imaging of vertical and overhanging landforms when using aerial platforms for data collection. Rather than reconstructing change from a digital surface model, direct analysis of generated point clouds is likely the key for understanding landform change in these morphologically complex environments. In this study, UAV’s were used to collect aerial imagery generating a high spatio-temporal resolution timeseries comprising of thirty point cloud datasets spanning four years for the steeply sloped Demons Bluff cliff located on the Great Ocean Road, Victoria, Australia. A method was developed to analyse the large quantity of 3D point cloud datasets. It was then possible to capture changes in cliff face morphology, enabling us to enhance our understanding of the erosional processes in coastal cliff environments. Over the study, the retreat rate for the upper half of the cliff face was 0.67 m/year (0.60 m3/m/yr). Cliff erosion was found to be dominated by 9 high magnitude, cliff-top collapses that exceeded 1,000 m3 (up to 9,500 m3) and terminated mid-way down the cliff. Below this point, several slab detachments were observed. Pre-collapse deformation was detected before collapses > 300 m3 with success (75 % of the time) in areas with a complete timeseries over the four-year period. Deformation was observed to occur in two ways, and both were observed to be eventuate in high magnitude (> 1,000 m3) collapse. The first which occurred prior to most high magnitude collapses (> 500 m3) was caused by the expansion of tension cracks behind the cliff-top, and the second which was characterised by smaller collapse volumes (> 100–500 m3) was initiated by rock slabs fracturing and cleaving away from the cliff face and was proceeded by high magnitude failure (< 1000 m3). An additional 14 instances of lean have been identified that are yet to result in collapse and should continue to be monitored to assess the success of this method with the forecasting of future collapse locations. Ultimately, this provides the ability to identify potential locations for future collapses which could aid in the development of an early warning system for cliff collapse to improve the management of volatile cliff environments that pose threats to infrastructure and public safety.
- Preprint
(1694 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-659', Anonymous Referee #1, 25 Apr 2024
The paper describes a case study to monitor cliff dynamics over 4 years covered by 30 UAV point clouds. The tools used are standard like point cloud cleaning followed by common M3C2. I am sure the study is highly relevant for understanding the local situation and how it is evolving over time. This might be also very important for the local community. The researchers invested a lot of time and have a great dataset at hand. However, I could not find any new scientific findings / concepts regarding the process of cliff erosion going beyond the description and quantification of the local situation. It remains a case study, as presented in the current manuscript, which is not showing significant relevance and learning moments for probably many others working in the field of 3D monitoring. The senors and methods used are already standard in geosciences.
Citation: https://doi.org/10.5194/egusphere-2024-659-RC1 -
CC1: 'Reply on RC1', Daniel Ierodiaconou, 02 May 2024
I would like to thank RC1 for your constructive comments. Whilst we agree that the tools are standard in terms of point cloud analytics we think the study provides novelty in the application due to the following reasons
-The high frequency data collection over such a period is unique and required in case studies to understand the broader implications of soft cliff retreat and hazard implications. In addition this has been done using innovative citizen science program (see Ierodiaconou et a;. 2022) that received a Eureka Prize in 2020
-Such studies on soft cliffs are unique with such frequency of information and we do provide a conceptual model for the location that we think would be applicable more broadly to soft cliffs globally but would need to be tested. We can make this point more clear in the revised manuscript
-This study has highlighted how a combination of point cloud analysis and high temporal resolution data collection can inform management decision making regarding hazardous coastal cliffs and we demonstrate how this has been put in to practice to improve safety of beach users.
cheers
Daniel Ierodiaconou
Citation: https://doi.org/10.5194/egusphere-2024-659-CC1 -
CC2: 'Reply on CC1', David M. Kennedy, 16 May 2024
We thank the reviewer for their thoughts and comments. They are correct that we utilize a local laboratory in southern Australia to test the methodological approach undertaken. Soft rocky coasts are a relatively understudied geomorphic system and are very sensitive to changes in boundary conditions. The difficulties in obtaining precise and accurate data of their erosion have hindered their study due to their inaccessibility and high rates of change.
This work aims to test the precision and accuracy of drone-based data collection for measuring such environments. It is true that drone-derived photogrammetry is widely used to produce DSM’s in geomorphology, but overhangs and area of shadowing produce many errors for the derived surfaces from vertical imagery. Our point-cloud approach tests an approach for minimizing these errors to enable valid scientific hypotheses to be tested. Another unique aspect of this study is that the collection of our UAV-imagery is conducted primarily by citizen-scientists, and while the precision and accuracy of this method has been proven for sandy beaches it has yet to be properly tested on vertical rapidly eroding cliffs.
We therefore believe the study is globally important. The study proves the unique point-cloud approach is highly relevant for measuring rates of erosion of vertical soft-rock cliffs on the open ocean coast. The field laboratory in southern Australia provides a perfect environment for the methodological exploration. We agree that new models of the processes driving soft rocky coast evolution are unlikely to be produced from the dataset at this stage, yet the results produced in the methodological approach do provide further steps in our understanding of soft cliff evolution.
Citation: https://doi.org/10.5194/egusphere-2024-659-CC2 -
RC2: 'Reply on CC2', Javier Leon, 28 Aug 2024
The manuscript presents a methodology for the analysis of UAV-derived point clouds to describe the dynamics of coastal cliffs. The presented change-detection workflow is not novel, except for the use of UAV-derived point clouds instead of the more commonly TLS-derived point clouds. The analysis is undertaken using CloudCompare and common algorithms such as SOR and M3C2.
In my opinion, the value of the manuscript resides on the use of a unique UAV dataset (50 months of high-res data along 1.5 km). This allows for a detailed description of cliff erosion/deposition and the development of a collapse mechanism model. However, this is not clearly communicated in the introduction, where the aim of the study comes across more like a methodological contribution to change-detection analysis. I suggest this is reworded.
The Discussion needs to be more critical and include a comparison between the advantages/disadvantages of UAV-derived point clouds (photogrammetry) and lidar (TLS and/or UAV). Further, suggest including a limitations/future research paragraph at end of discussion (e.g. use of citizen science programs, etc).
Some specific comments:
- L18 include LOD
- L41 UAV includes UAV-lidar?
- L54: The analysis of point clouds is moving very fast (e.g. Deep learning and individual tree detection ). Suggest you include relevant literature.
- Fig. 1 Suggest removing inset b and including place names.
- Can you merge Fig. 2 and 3? If not, Fig. 3 can improve by adding more detail (e.g. algorithms used).
- Can you adapt Fig. 4 to a cliff example?
- L143 Reword first sentence.
- L143-L155 Break paragraph as hard to follow.
- Fig. 7: Is this better suited as a Discussion figure (e.g. conceptual model based on results)?
- L196: Is this Discussion? Were waves measured? What are high-energy wave conditions?
Citation: https://doi.org/10.5194/egusphere-2024-659-RC2 -
AC2: 'Reply on RC2', Todd Doran, 23 Sep 2024
The manuscript presents a methodology for the analysis of UAV-derived point clouds to describe the dynamics of coastal cliffs. The presented change-detection workflow is not novel, except for the use of UAV-derived point clouds instead of the more commonly TLS-derived point clouds. The analysis is undertaken using CloudCompare and common algorithms such as SOR and M3C2.
We agree that our approach uses standard methods and applies them in a unique way. As noted by Javier, analysis of UAV-derived point clouds is novel and we find it surprising that point-clouds derived from UAV-based photogrammetry have to date, not been analysed to a great degree, especially in regards to studying soft, vertical sea cliffs. The power of our study is that we show that such point clouds from high resolution timeseries can be accurately and precisely used to understand change in complex geomorphic environments.
In my opinion, the value of the manuscript resides on the use of a unique UAV dataset (50 months of high-res data along 1.5 km). This allows for a detailed description of cliff erosion/deposition and the development of a collapse mechanism model. However, this is not clearly communicated in the introduction, where the aim of the study comes across more like a methodological contribution to change-detection analysis. I suggest this is reworded.
We thank Javier for this observation and agree that the dataset which is available (www.marinemapping.org/vcmp-citizen-science), has the potential to advance our knowledge of cliff dynamics. In fact, it was with this intention that the data was first investigated. During this process, the lack of scientific approaches to analyse and quantify this magnitude of UAV-derived point cloud datasets quickly became apparent as the primary factor in limiting knowledge advancement. As a result, significant work was required to prove the utility of the described approach. An additional important factor is that the extent of the dataset is owed to collection being undertaken by citizen-scientists. In addition to the duration and frequency of datasets, it has been highlighted in previous studies that the accuracy of citizen science UAV mapping datasets is very good (Pucino et al., 2021).
We agree that the focus of the work could have been clearer. As a result, the abstract, introduction and discussion has been reworded in the following ways to better convey this message, and place less of a spotlight on the methodological approach. See below for the amendments with the original text in italics, deleted text in bold, and amended text in standard font.
Abstract
Unmanned Aerial Vehicles (UAVs) and Structure for Motion (SfM) photogrammetry have revolutionised data capture for hazardous coastlines. The increased usability of UAVs, coupled with automated flight planning, has enabled citizen scientists with basic UAV handling skills to capture highly accurate data in coastal cliff landscapes. This advancement has facilitated the generation a four-year long timeseries, with a bi-monthly sampling resolution, of point cloud datasets for a 1.5 km stretch of the steeply sloped Demons Bluff cliff, located on the Great Ocean Road, Victoria, Australia. In this study, UAV-SfM derived point clouds were analysed to describe the erosional dynamics of the soft-coastal cliff at Demons Bluff. Changes in cliff face morphology exceeding 0.07 m were consistently captured, allowing for the development of a cliff collapse mechanism model. Over the study, the retreat rate for the upper half of the cliff face was 0.67 m/year (0.60 m3/m/yr). Cliff erosion was found to be dominated by nine high magnitude cliff-top collapses, each exceeding 1,000 m3 (up to 9,500 m3), which terminated mid-way down the cliff. Below this point, several slab detachments were observed. Pre-collapse deformation was detected prior to collapses greater than 300 m3 with success (75 % of the time) in areas with a complete timeseries over the four-year period. Deformation was observed to occur in two ways, both of which cumulated in high magnitude (> 1,000 m3) collapse. The first, preceding most large collapses (> 500 m3) was caused by the expansion of tension cracks behind the cliff top. The second, associated with smaller collapse volumes (> 100-500 m3), was initiated by rock slabs fracturing and cleaving away from the cliff face. An additional 14 instances of seaward displacement of cliff face have been identified that have not yet resulted in collapse. Ultimately, this study highlights the benefits and potential for using UAV-SfM derived point clouds for the monitoring of hazardous cliff environments. Benefits extend from ease of data capture and generating extensive timeseries to the analytical insights it can provide. UAV SfM point clouds offer a promising low-cost alternative to cliff monitoring compared to commonly used techniques such as Terrestrial Laser Scanning (TLS).
Unoccupied aerial vehicles (UAVs) have revolutionised data collection on the Earth’s surface. Through aerial photogrammetry, very high-resolution digital surface models can be produced enabling contemporary research on landscape stability. There are however significant limitations in the imaging of vertical and overhanging landforms when using aerial platforms for data collection. Rather than reconstructing change from a digital surface model, direct analysis of generated point clouds is likely the key for understanding landform change in these morphologically complex environments. In this study, UAV’s were used to collect aerial imagery generating a high spatio-temporal resolution timeseries comprising of thirty point cloud datasets spanning four years for the steeply sloped Demons Bluff cliff located on the Great Ocean Road, Victoria, Australia. A method was developed to analyse the large quantity of 3D point cloud datasets. It was then possible to capture changes that exceeded 0.07 m in cliff face morphology, enabling us to enhance our understanding of the erosional processes in coastal cliff environments. Over the study, the retreat rate for the upper half of the cliff face was 0.67 m/year (0.60 m3/m/yr). Cliff erosion was found to be dominated by 9 high magnitude, cliff-top collapses that exceeded 1,000 m3 (up to 9,500 m3) and terminated mid-way down the cliff. Below this point, several slab detachments were observed. Pre-collapse deformation was detected before collapses > 300 m3 with success (75 % of the time) in areas with a complete timeseries over the four-year period. Deformation was observed to occur in two ways, and both were observed to be eventuate in high magnitude (> 1,000 m3) collapse. The first which occurred prior to most high magnitude collapses (> 500 m3) was caused by the expansion of tension cracks behind the cliff-top, and the second which was characterised by smaller collapse volumes (> 100-500 m3) was initiated by rock slabs fracturing and cleaving away from the cliff face and was proceeded by high magnitude failure (< 1000 m3). An additional 14 instances of lean have been identified that are yet to result in collapse and should continue to be monitored to assess the success of this method with the forecasting of future collapse locations. Ultimately, this provides the ability to identify potential locations for future collapses which could aid in the development of an early warning system for cliff collapse to improve the management of volatile cliff environments that pose threats to infrastructure and public safety.
Introduction
Cliffs are prominent landforms that comprise ~52 % of coastlines globally (Young and Carilli, 2019). Projections for sea level rise and Changed intensity and frequency of storms is forecast to enhance future rates of cliff recession (Limber et al., 2018, Lim et al., 2010). This has implications for coastal planning and public safety, signifying a need to better understand the processes of cliff retreat and identify the environmental drivers of morphometric change in cliff landscapes.
Quantitative methods to analyse cliff change advanced over the past few decades. Methods of data capture have evolved from ground-based physical field survey using levels and theodolites (Smith and Zarillo, 1990), to comparison of satellite imagery or cartographic charts (Bray and Hooke, 1997), terrestrial photogrammetry (Gulyaev and Buckeridge, 2004), terrestrial and aerial laser scanning (Dewez Thomas et al., 2013, Young et al., 2014), with Unoccupied Aerial Vehicles (UAVs)-based photogrammetry surveys now becoming a more common approach (Esposito et al., 2017). Of these methods, the LiDAR-based aerial and terrestrial laser scanning have dominated cliff erosion studies in the past decade (Letortu et al., 2018, Rohmer and Dewez, 2013, Theodore et al., 2020, Young et al., 2009, Terefenko et al., 2019, Bezore et al., 2019, Katz and Mushkin, 2013), owing to their ability to produce 3D point clouds (hereafter denoted as “point clouds”) that allow visualisation and analysis of landscapes in three dimensions. More recently, UAVs and Structure from Motion (SfM) photogrammetry have been used to generate high resolution point clouds of cliff environments (Mancini et al., 2017, Dewez Thomas et al., 2013). UAV-SfM photogrammetry of cliff environments provide several key benefits, including eliminating the need for site accessibility below hazardous cliff faces and enabling fast survey times with a low occurrence of occlusions (Letortu et al., 2018). Advancements in the UAV usability, particularly in automated flight planning, have increased the ability of pilots with basic UAV training to collect accurate data and has expanded the possibilities for drone mapping to be conducted by citizen scientists. This allows for reductions in operational costs for maintaining timeseries data. Coupled with fast survey times, there is a greater ability to build extensive timeseries with high sampling resolution (Ierodiaconou et al., 2022), allowing for more detailed descriptions of coastal change processes.
More recently, UAVs with the ability to collect Structure from Motion (SfM) photogrammetry have provided a cost-effective way to collect data with high spatial resolution and capability to produce point clouds (Mancini et al., 2017, Dewez Thomas et al., 2013), thereby allowing for the generation of extensive multi-temporal datasets (Ierodiaconou et al., 2022).
Despite the truly three-dimensional nature of point cloud datasets, it is more common for geomorphic analysis to be performed using digital elevation models (DEM) that are a derived product from the point cloud. The downside of DEMs is that they are pseudo-3D and have a limited ability to visualise and measure change across vertical and undercut surfaces (Lague et al., 2013, Li et al., 2021). However, the analysis of point clouds overcomes these limitations and provides the ability to accurately measure change and describe erosion processes. For example, Alessio and Keller (2020) utilised point clouds for five small (< 145 m width) coastal cliffs in Santa Barbara, California. They found retreat of cliff toe occurred first, with the middle and upper portions of cliff eroding back at a later time until they patched the position of the toe, whereby the cycle would reinitiate. As illustrated by Hendrickx et al. (2022), point clouds can also be used to detect the precursory deformation on rock walls prior to major collapse in mountainous landscapes. In Matter Valley, Switzerland, the changes in point clouds allowed identification of rock slabs cleaving, which resulted in debris falls that eventually lead to more major collapse of larger areas. Similarly, Abellan et al. (2010) at Pallars Jussà, Spain, identified areas of deformation prior to small rockfall event and measured a maximum horizontal displacement value that could be used to predict the locations of future collapses. However, few studies have utilised an extensive timeseries of point clouds generated solely by UAV photogrammetry to analyse cliff retreat and to our knowledge, there has been limited application to determine whether such approaches can identify pre-collapse deformation.
Point cloud analysis of cliffs over a timeseries has traditionally been conducted using point clouds derived from TLS (Young et al., 2021, Alessio and Keller, 2020, Terefenko et al., 2019). Consequently, the use of UAV-SfM-derived point clouds to monitor steep, soft-rock coastal cliffs over extensive timescales with high spatial resolution has been relatively unexplored. This study aims to test the ability of a four-year timeseries, consisting of bi-monthly UAV-SfM point cloud datasets (> 25 datasets) collected by citizen scientists, to describe the process of erosion for a soft-rock cliff spanning more than 1.5 km, along a high-energy, open coast.
Despite the highlighted upsides of point cloud analysis, a methodology is yet to be developed for an ongoing temporal analysis of point clouds for a single steep coastal cliff containing high quantities (> 20) of UAV generated datasets. A higher spatio-temporal resolution of analyses is important for informing management decisions surrounding rockfall hazards, or localised measures of cliff top retreat. This study aims to develop a methodology for the analysis of large datasets of near vertical coastal cliffs. It utilised a unique citizen-scientist UAV dataset (Ierodiaconou et al., 2022) collected bimonthly over four years along > 1.5 km of coast to test the ability of point clouds to detect meaningful change in coastal cliffs along a high-energy open-coast.
Discussion
In this study, a timeseries of UAV-SfM derived point clouds collected by citizen scientists was effectively used to monitor cliff erosion over 50 months along a 1.5 km stretch of shoreline. The level of accuracy and precision achieved by the point clouds, along with the ability to visualise 3D change using this low-cost method of data collection enabled the tracking of precursors to cliff collapse, namely the seaward cleaving of large rock slabs from the cliff face. It also allowed the description of collapse mechanisms and the quantification of retreat rates. At Demons Bluff, cliff collapse is caused by the expansion of cliff-top fractures that eventually result in moderate-to-high magnitude collapses. Rock lean has been monitored previously using TLS point clouds of a cliff at Catalonia, Spain (Abellán et al. (2010)). Here, surveying of a small area of sea cliff (150 m wide, 25 m high) detected block leaning that exhibited a displacement of 0.06 m prior to collapse (< 10 m3). In this study, however, deformation could not be detected prior to failure of larger magnitude. It was highlighted that the temporal resolution of the surveys was inadequate, and that deformation and the subsequent collapse likely occurred between the surveys. In our study, the method emphasised the importance of high spatio-temporal resolution, as the seaward displacement of cliff was detected prior to 75% (N = 8) of collapses greater than 300 m3 within the upper cliff, in sections 7 – 11. This finding is similar to Kromer et al. (2018) who utilised TLS point cloud analysis and clusters of high temporal resolution sampling across different seasons, achieving a 90 % success rate in identifying deformation prior to larger collapses. Kromer et al. (2018) also found that the duration between deformation and eventual collapse ranged between hundreds – thousands of days. At Demons Bluff, the timing between cliff deformation and collapse was shown to occur over a similar range. Tension cracks persisted behind the cliff top for the duration of the study (> 1000 days), while another instance of creep led to collapse at the same location where a high magnitude collapse event had occurred 12 months earlier.
Studies that have conducted TLS point cloud analysis have generally focused on smaller study sites, allowing the cliff face to be treated as a single entity. The use of UAV-SfM-derived point clouds enable fast survey rates and allowed a larger area of cliff to be mapped. Treating the cliff face as a single entity for an area spanning 1.5 km would require excessive computational power and cause deterioration in the ability to measure incremental change as measurements are averaged over larger surface areas (Neverman et al., 2016). For example, for volumetric change, segmentation permitted a LoD of ± 22 m3 and ± 24 m3 for the lower and upper halves of the cliff, respectively, across each 100 m section. This allowed for the identification of at least 298 erosional events. If like other studies, the cliff face was treated as a single entity, the LoD would increase to ± 302 m3 and the number of erosion events detected would decrease to 76. By segmenting the cliff into 100 m sections along the shore and dividing it vertically based on the local geology, it became possible to highlight localized retreat rates. Localised retreat rates between sections were observed to double the mean annual retreat rate. This result is similar to Benjamin et al. (2020) who split over 20 km of Jurassic age mudstones, limestones and sandstones cliffs in North Yorkshire, UK into 100 m wide alongshore bins to identify the localised measures of retreat. They also found that erosion rates exhibited spatial and temporal variations with different regions experiencing annual retreats from < 0.001 m to 1.63 m.
In this study, point clouds have been effectively used to monitor cliff erosion over 50 months along a 1.5 km stretch of shoreline. This level of accuracy and precision, along with the ability to visualise 3D change, has proven to be able to track the precursors of cliff collapse, namely the initial seaward cleaving of large slabs of bedrock in soft carbonate mudstones. Previously, point clouds have been used to analyse the cliff face as one entity over smaller sites (Abellán et al., 2010, Earlie et al., 2018, Alessio and Keller, 2020, Gilham et al., 2019). However, this becomes less effective over larger sites due to a deterioration in the ability to measure incremental change caused by measurements being averaged over greater surface areas (Neverman et al., 2016). The ability to highlight localised retreat rates was provisioned by segmenting the cliff into 100 m sections in an alongshore direction and subsequently divided in half vertically based on the local geology. On average, segmentation permitted a LoD of 22 m3 and 24 m3 for the lower and upper halves the cliff respectively, across each 100 m section. This allowed the identification of at least 298 erosional events. If like other studies, the cliff face was treated as one entity, the LoD would increase to +/- 302 m3 and the number of erosion events detected would decrease greatly. Segmentation also provided insight into localised retreat rates where sections observed to retreat in excess of 100 % greater than the mean annual retreat rate. This result is similar to Benjamin et al. (2020) who split over 20 km of Jurassic age mudstones, limestones and sandstones cliffs in North Yorkshire, UK into 100 m wide alongshore bins to allow the identification of localised measures of retreat. They also found that erosion rates contained spatial and temporal variations with different regions experiencing annual retreats from < 0.001 m, up to 1.63 m.
One observed benefit of the point cloud approach is a greater ability to detect the seaward displacement of the cliff face. In the case of Demons Bluff, this is caused by the expansion of cliff-top fractures that eventually result in moderate - high magnitude cliff collapse. Rock lean has been monitored previously using point clouds generated from terrestrial laser scans at Catalonia, Spain (Abellán et al. (2010)). Here, surveying of a small area of sea cliff (150 m wide, 25 m high) could detect block leaning that exhibited a displacement of 0.06 m prior to collapse (< 10 m3). In this study, deformation could not be detected prior to failure of larger magnitude. They highlighted that the temporal resolution of survey was inadequate, and that deformation and the subsequent collapse likely occurred between the surveys. In this study, our method was able to emphasise the importance of high spatio-temporal resolution as the seaward displacement of cliff was detected prior to 75% (N = 8) of collapses > 300 m3 within the upper cliff in sections 7 – 11. This was similar to Kromer et al. (2018) who utilised clusters of high temporal resolution sampling within different seasons and found a high success rate of 90 % in identifying deformation prior to larger collapses. However, in our study, outside of sections 7 – 11, precursory information was not detected for collapses of similar magnitude (> 300 m3). This is likely due to the length and temporal resolution of the timeseries being smaller and not optimal for capturing the development of deformation prior to collapse and may benefit from a longer timeseries. Alternatively, as the baseline survey for some of the areas to the east did not occur until later in the study and there is potential that deformation had already occurred and was not detected, even if the timeseries was complete.
A high coincidence in identifying block lean in areas with complete timeseries illustrates that a temporal resolution of 6 – 8 weeks appears to be sufficient to identify deformation and gain an indication of a region’s relative stability at Demons Bluff. Conversely, the lack of detection outside of sections 7 – 11 highlights that sustaining the timeseries is imperative for being able to identify deformation and potentially hazardous areas in the future. Kromer et al. (2018) also identified that the duration between deformation and eventual collapse ranged between 100s – 1000s of days. The timing between cliff deformation and collapse at Demons Bluff was shown to occur over a similar range. Tension Cracks were shown to exist behind the cliff top for the duration of the study (> 1000 days), while conversely, another instance of creep led to collapse at the same location where a high magnitude collapse event had occurred just 12 months earlier.
The Discussion needs to be more critical and include a comparison between the advantages/disadvantages of UAV-derived point clouds (photogrammetry) and lidar (TLS and/or UAV). Further, suggest including a limitations/future research paragraph at end of discussion (e.g. use of citizen science programs, etc).
Two paragraphs have been added to the end of the discussion that evaluate the trade-offs between using TLS and UAV SfM and the study limitations/future research.
Discussion
UAV-SfM-derived point clouds offer an alternative method for quantifying cliff retreat and capturing the process of collapse mechanisms. Traditionally, TLS has been the primary method of data capture for this type of monitoring. However, the ease and speed of data capture of UAV SfM in areas where accessing the fronting beach to perform TLS scans is either unsafe or cumbersome, offers an advantageous means of survey. In addition to this, there are several other advantages to using UAV SfM over TLS for mapping cliff environments. Consumer-grade airframes fitted with factory RTK modules and high-resolution cameras that can produce accurate, high-resolution point clouds, have become increasingly affordable (< $10,000). With the integration of automated flight planning, citizen scientists can be trained to fly UAVs and collect accurate, high resolution point cloud data of coastal environments (Pucino et al., 2021, Ierodiaconou et al., 2022). The use citizen science volunteers thereby reduce operating costs for building and maintaining a timeseries, while also building capacity and empowering local communities. In comparison, TLS systems are more expensive to purchase ($20,000 to > $100,000) and take considerably more time to survey smaller coverages than what can be achieved by UAV surveys (Guisado-Pintado et al., 2019). Their reliance on site accessibility and appropriate scanner angles is also a major limitation when considering hazardous cliff environments, like Demons Bluff. The higher cost of TLS is owed to its capability of capturing higher accuracy data (sub-centimeter) (Guisado-Pintado et al., 2019, Del Río et al., 2020). This is due to the stationary nature of the TLS while a scan is being performed, in contrast to UAVs, where imagery is being captured while the airframe is in motion. TLS point clouds are more prone to occlusions in complex cliff faces and bias scanning positions caused by accessibility issues (Letortu et al., 2018, Westoby et al., 2018), whereas a UAV provides manual control of the camera angle inflight to ensure a good representation of the cliff-face in imagery. The main drawback of UAV-SfM is its reliance on favourable weather conditions (no rain, low wind) for data capture. While TLS can operate in a wider array of conditions, it is more restricted by tidal conditions in areas where waves impact the cliff base. The miniaturisation of sensors has increased the availability of UAV mounted LiDAR systems that have been applied to coastal geomorphological studies (Shaw et al., 2019, Pinton et al., 2022). However, these sensors remain costly to purchase and typically require larger UAVs with higher maximum takeoff weights. Depending on the country, regulations may require additional licensing with associated costs to operate these larger airframes. This reduces the viability of using citizens scientists to conduct data collection and subsequently, increase the operating costs of maintaining the timeseries.
This analysis focuses on detecting larger movements compared to previous studies (Dewez Thomas et al., 2013; Hendrickx et al., 2022; Kromer et al., 2018). It remains to be explored whether trade-offs exist in this dataset between the extent of coastline mapped and the ability to detect smaller changes. For instance, other studies analysing TLS (terrestrial laser scanning) point clouds for smaller sites have employed clustering algorithms to monitor smaller magnitude rockfalls (Micheletti et al., 2017; Tonini & Abellan, 2014) and predict the spatial distribution of future rockfalls (Abellán et al., 2010). If this ability is absent/reduced within the data, there could be adjustments in data collection protocols that allow for a scalable approach to mapping, where high-priority areas are captured at higher spatial resolution. Such an approach would also test the feasibility of using citizen scientists for data collection, as tasks would be more dynamic as areas of interest would shift in line with management priorities. Altering mapping areas across surveys could introduce potential degradations in data quality or increase the risk of human error, leading to equipment damage.
Conclusion
The use of UAV-SfM-derived point clouds captured by citizen scientists presents a cost and time effective alternative to traditional methods for monitoring cliff erosion and identifying collapse mechanisms. The ability to map larger areas introduces computational limitations associated with the analysis of point clouds. However, segmenting the large study area into 100-metre-wide alongshore bins provided a timely way of performing high resolution change detection and quantifying localised measures of retreat and hotspots. Given the observed evolution of this UAV technology in recent years, the forthcoming development of components associated with UAV-SfM, such as Inertial Measurement Units and RGB sensors, may lead to faster flight times with more accurate and higher quality data. This may see UAV-SfM become the favoured method of data collection for cliff monitoring efforts, particularly for organisations who have limited budgets to perform surveys, and in areas where site accessibility poses safety risks.
This study has highlighted how a combination of point cloud analysis and high temporal resolution data collection can inform management decision making regarding hazardous coastal cliffs. In the future, with a greater understanding and confidence around the timing of failure, it may become possible to define areas at threat of collapse and flag environmental conditions that may induce failure to promote the safety of beach users and the preservation of coastal infrastructure. Future research priorities should be placed on the continued monitoring of the areas where lean was identified but collapse is yet to ensue. This may enhance our understanding into the timing of failure and the success of this method in forecasting the locations for collapse based on cliff deformation and help inform the development of a cliff collapse early warning system. Furthermore, this study focused on relatively large erosion events and highlighted the morphological processes of moderate-high magnitude (> 300 m) cliff failure. Future work should also be directed to identify if precursory cliff deformation can be detected prior to smaller failures and rockfalls (Benjamin et al., 2020, Gilham et al., 2019, Rosser et al., 2013). Subaerial and marine processes are known to affect rates of cliff retreat and morphological change. Future work should begin establishing whether relationships exist between weathering processes such as precipitation and wave action and cliff-face deformation and the timing of subsequent failure. This would further aid the development of an early warning system tool for management that may be applicable to other similar, neighbouring cliffs near Demon’s Bluff that also experience high tourist visitation rendering the provision of safety of highest importance.
Some specific comments:
- L18 include LOD
LOD will be included in revison
- L41 UAV includes UAV-lidar?
It does not include UAV-Lidar. However, UAV-Lidar is now discussed in the evaluation of TLS + UAV SfM point clouds within the discussion.
- L54: The analysis of point clouds is moving very fast (e.g. Deep learning and individual tree detection). Suggest you include relevant literature.
Acknowledgement of rockfall detection algortihms (DBSCAN) has been included. Vegetation-related ML/DL classification is out of the scope for this study. Deep learning for change detection is a recent development for coastal cliffs but remains rudimentary (classification of: Accretion, Erosion, no change,0.5194/isprs-annals-V-3-2022-649-2022). Therefore, we do not see enough relevance to include it in our study.
Introduction
Despite the truly three-dimensional nature of point cloud datasets, it is more common for geomorphic analysis to be performed using digital elevation models (DEM) that are a derived product from the point cloud. The downside of DEMs is that they are pseudo-3D and have a limited ability to visualise and measure change across vertical and undercut surfaces (Lague et al., 2013, Li et al., 2021). However, the analysis of point clouds overcomes these limitations and provides the ability to accurately measure change and describe erosion processes, including the classification of rockfall events (Schubert et al., 2017).
- Fig. 1 Suggest removing inset b and including place names.
Resolved in revision
- Can you merge Fig. 2 and 3? If not, Fig. 3 can improve by adding more detail (e.g. algorithms used).
Remaining separate and will add detail regarding algorithms in revision.
- Can you adapt Fig. 4 to a cliff example?
Figure will be adapted to cliff example in revision.
- L143 Reword first sentence.
Resolved in revision.
- L143-L155 Break paragraph as hard to follow.
Resolved in revision.
- Fig. 7: Is this better suited as a Discussion figure (e.g. conceptual model based on results)?
There are arguments for including in either the discussion or results. This will be carefully considered in the revision.
- L196: Is this Discussion? Were waves measured? What are high-energy wave conditions?
Resolved in revision.
Citation: https://doi.org/10.5194/egusphere-2024-659-AC2
-
AC2: 'Reply on RC2', Todd Doran, 23 Sep 2024
-
RC2: 'Reply on CC2', Javier Leon, 28 Aug 2024
-
CC2: 'Reply on CC1', David M. Kennedy, 16 May 2024
-
AC1: 'Reply on RC1', Todd Doran, 23 Sep 2024
We thank RC1 for their comments and taking time to read this manuscript. We believe the novel aspect of this study is the cost-effective and community-empowering approach to data collection that utilises citizen scientists and consumer grade airframes and sensors to capture accurate, high-resolution UAV SfM data, from which cliff deformation could be identified and erosional processes could be described. This low-cost approach provides opportunities for organizations and coastal management groups with limited budgets to perform surveys and build extensive timeseries to monitor cliffs, especially in areas where accessibility poses safety risks, rather than being reliant on more expensive methods for obtaining point cloud data, like TLS.
While we agree that new models of soft, rocky coast evolution have not been identified in this study, it is evident that UAV SfM-derived point clouds offer a promising alternative for detecting cliff processes that have typically been observed with TLS-derived point clouds. This was highlighted by the development of a collapse mechanism model, consistently supported in the point cloud data, throughout the timeseries. Additionally, other collapse mechanisms described in previous studies, such as failure around the peripheries of earlier collapses and cascading events (where smaller collapses precede larger ones), were also observed. Therefore, the analysis of long-term, citizen science-collected UAV SfM point cloud timeseries has global implications, demonstrating that low-cost data collection can be used to build detailed timeseries capable of identifying erosional processes previously observed in more cost prohibitive, TLS point clouds.
Citation: https://doi.org/10.5194/egusphere-2024-659-AC1
-
CC1: 'Reply on RC1', Daniel Ierodiaconou, 02 May 2024
-
RC3: 'Comment on egusphere-2024-659', Anonymous Referee #3, 28 Oct 2024
Overall, this article describes an application case of long-term monitoring of a coastal cliff retreat in soft lithologies, using data obtained by drone photogrammetry. The dataset described is very rich and interesting, and the paper shares the approach followed as well as useful information about a specific area and setting. As a general comment, despite there is no significant scientific advancement presented, I think this paper is worth publishing, perhaps as a Brief communication or possible specific article categories dedicated to case studies. As it is now, considering its structure and content, the paper is closer to a project report than a scientific paper. As a more specific comment, the method section dedicates space to description of common methods rather than explaining better the important approaches to optimize the application of existing algorithms to the local situation. For example, it would have been useful a figure to explain the added value of projecting a raster onto the cliff face from the x-axis rather than the z-axis in the computation of the 2.5D volume. Also I am not sure figure 2 should be cited as it is, because it is just an excerpt of a figure of a different paper, which I think would have required author's permission. My last specific comment is about acronyms, which use should be always preceded by a clear explanation.
Citation: https://doi.org/10.5194/egusphere-2024-659-RC3 -
AC3: 'Reply on RC3', Todd Doran, 31 Oct 2024
We thank the reviewer for their kind words recognizing the importance and quality of the dataset collected and reported in this work. The length and frequency of this UAV-SfM time series of point clouds are unique, as prior point cloud analyses of vertical sea cliffs have relied on Terrestrial Laser Scanning (TLS), for which data collection is hindered by limited site accessibility, particularly in hazardous cliff environments, and the high costs of sensor purchase. The use of UAV-SfM point clouds overcomes these limitations, and as shown in our results, similar geomorphic processes can be observed over time. This enables managers of such environments to collect and analyze hazardous cliff data, where prohibitive costs and safety risks would have otherwise made this unfeasible. This is a global issue, and our field laboratory at Demons Bluff was specifically chosen to demonstrate the utility of our technique for researchers and managers worldwide. We agree that the initial submission was less focused on the novel aspects of this study, offering a broad narrative on cliff erosion trends and morphology change. However, in light of revisions made following suggestions by other reviewers, the study now focuses more directly on the finding of identifying pre-collapse deformation using UAV-SfM point clouds and its potential implications. Therefore, the conclusions are relevant for any site with rapidly eroding, near-vertical cliffs, including environments such as the chalk cliffs along both sides of the English Channel and the till-dominated cliffs characteristic of lacustrine and estuarine shores in North America. This work thus advances intellectual understanding in rock coast geomorphology.
Citation: https://doi.org/10.5194/egusphere-2024-659-AC3
-
AC3: 'Reply on RC3', Todd Doran, 31 Oct 2024
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 306 | 103 | 32 | 441 | 19 | 19 |
- HTML: 306
- PDF: 103
- XML: 32
- Total: 441
- BibTeX: 19
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1