**3.1 UAV survey design and initial SfM quality controls**

On the 19th and 20th of September 2019, 11 days following SD, a survey consisting of eight overlapping UAV flight grids (**Figure 4A**) was flown at BB. Images were collected with a Mavic 2 Pro quadcopter and covered an alongshore extent of 2.5 km. Across shore, UAV flight paths were programmed to capture images from the nearshore, foreshore, backshore, and back dune zones with 80% frontal- and 70% side-overlap. The UAV was flown at an altitude of 50 m corresponding to a ground sampling distance of 1 cm/pixel. In each flight grid, at least 10 bright white bucket lids (30 cm in diameter) were used as GCPs and spread in a non-uniform placement across the foreshore, beach, and foredune (**Figure 4A**). The geographic location of each GCP was surveyed using a Global Navigation Satellite Series GPS. The high number of GCPs was designed to increase the accuracy of the geographic coordinate conversion by providing a diverse set of elevational references and to systematically build in redundancy that allow for GCP

### **Figure 4.**

*UAV flight and GCP locations for the Brackley Beach (BB) east and BB west surveys (A). Quality control measures were taken during post-processing including removing photos that generated a limited number of ATPs, typically occurring in areas of homogenous vegetation or prominent peaked vegetation (B) and breaking waves in the nearshore (C). Also, only GCPs that were clearly visible and non-deformed were selected (D) while all others were left unmarked (E) in order to improve accuracy during geo-referencing.*

*Monitoring Storm Impacts on Sandy Coastlines with UAVs DOI: http://dx.doi.org/10.5772/intechopen.91459*

topographic surface and near shore bathymetry, up to 5 m water depth. The survey was flown at an altitude of 400 m above the ground surface and maintained a sampling density of 2.72 points per meter (p/m) for the bathymetric and 18.6 p/m for the topographic surfaces. Point clouds were then geo-referenced using a total of 141 ground control points with a root mean square error (RMSE) of 0.05 m for the vertical transformation. The point cloud data was then classified for surficial and supra-surficial elements (e.g., vegetation, water surface, etc.). Following classification of the point cloud, only bathymetric and topographic surface classes were maintained to produce a 1 m 1 m 'bare earth' or digital terrain model (DTM) that

will serve as a pre-SD baseline for comparisons to post-SD UAV surveys.

*Spatial Variability in Environmental Science - Patterns, Processes, and Analyses*

On the 19th and 20th of September 2019, 11 days following SD, a survey consisting of eight overlapping UAV flight grids (**Figure 4A**) was flown at BB. Images were collected with a Mavic 2 Pro quadcopter and covered an alongshore extent of 2.5 km. Across shore, UAV flight paths were programmed to capture images from the nearshore, foreshore, backshore, and back dune zones with 80%

corresponding to a ground sampling distance of 1 cm/pixel. In each flight grid, at least 10 bright white bucket lids (30 cm in diameter) were used as GCPs and spread in a non-uniform placement across the foreshore, beach, and foredune (**Figure 4A**). The geographic location of each GCP was surveyed using a Global Navigation Satellite Series GPS. The high number of GCPs was designed to increase the accuracy of the geographic coordinate conversion by providing a diverse set of elevational references and to systematically build in redundancy that allow for GCP

*UAV flight and GCP locations for the Brackley Beach (BB) east and BB west surveys (A). Quality control measures were taken during post-processing including removing photos that generated a limited number of ATPs, typically occurring in areas of homogenous vegetation or prominent peaked vegetation (B) and breaking waves in the nearshore (C). Also, only GCPs that were clearly visible and non-deformed were selected (D) while*

*all others were left unmarked (E) in order to improve accuracy during geo-referencing.*

frontal- and 70% side-overlap. The UAV was flown at an altitude of 50 m

**3.1 UAV survey design and initial SfM quality controls**

**3. Methodology**

**Figure 4.**

**72**

removal if elevated error values are recorded. Common causes of error can include GPS survey (e.g., limited number of satellites or line of sight), pedestrian interference during UAV flights, or user identification error during post processing.

Images of all connected flight grids were collectively processed (i.e., BB West and BB East) using the commercial SfM software Pix4D. Alternative SFM software including Agisoft Metashape [36, 39, 40, 42–45] and Fledermaus v-7 [38] have been previously used and described. The remainder of this section will focus on the general Pix4D workflow used in this study. Initially, a target number of 10,000 ATPs were generated from each overlapping image. Next, Automatic Aerial Triangulation (AAT), Bundle Block Adjustment (BBA), and camera calibration were optimized for all images. Uncalibrated cameras, a result of errors in the internal (e.g. vibrations) or external (e.g., position and orientation) camera parameters, were deactivated to remove potential topographic deformation during point cloud generation.

Furthermore, flight lines on the periphery of the survey grid had difficulty finding a sufficient number of ATPs due to areas of homogenous vegetation (**Figure 4B**), selecting prominent features such as treetops from multiple angles (**Figure 4B**), or breaking waves in the nearshore (**Figure 4C**). As a result, severe over- or under-estimation of the surface can occur by misclassifying reference elevation values. In an attempt to reduce survey error, images were clipped to the primary area of interest (i.e., the beach-dune zone). This ensured an optimal number of ATPs present in each image and removed potential edge or 'bowl' effects that can distort the point cloud in areas away from the GCP locations [40]. Next, GCPs that were clearly visible in the flight images were zoomed into and marked at their centroid (**Figure 4D**) to ensure proper pixel selection. GCPs that were blurry (**Figure 4E**), warped, displayed limited contrast, or not entirely visible were not selected because un-proper pixel selection can also introduce error during geographic transformation. Marked GCPs were then used for geographic conversion resulting in a vertical RMSE of 0.028 and 0.021 for BB West and East, respectively.

#### **3.2 DSM and DTM generation and environmental uncertainty**

Following initial processing, a point cloud was generated and consisted of 2.65 <sup>10</sup><sup>8</sup> points with a density of 2000 points per m<sup>2</sup> . The point cloud was classified using Pix4D's predefined class groups including ground, road surface, high vegetation, building, and human made objects, in order to improve DTM filtering. Point interpolation was completed using an inverse distance weighting (IDW) approach to generate a universal DSM (i.e., retaining all elevational classes) and filtered DTM (i.e., retaining only the ground elevation class) at a 1 m 1 m resolution for direct comparisons to the pre-SD LiDAR DTM.

Prior to measuring topographic changes between surfaces, an initial assessment of the quality of the LiDAR DTM and UAV DSM and DTM displayed significant irregularities (**Figure 5A**). For instance, the LiDAR DTM was generated using green laser that can penetrate up to 5 m depth and full wave form infrared laser that can penetrate the vegetation canopy. This provides a fully integrated bathymetric and topographic surface transitioning from the nearshore—to backshore zones (**Figure 5A**). SfM is not able to penetrate the water column and has difficulty measuring ground points in vegetated areas. As a result, the SfM DSM captures noise above the surface that is associated with wave breaking and run up in the nearshore and foreshore zones (**Figure 5A** and **B**). In the backshore, the vegetated crest and lee slope is overestimated on average by 0.5–1m(**Figure 5A**) and represents variability in vegetation density and canopy height (**Figure 5B**).

Alternatively, the standard Pix4D DTM filtering method almost completely removed the foredune in areas that recorded significant scarping. **Figure 5A** shows

vegetation. Difficulty arises when demarcating a seaward boundary due to the fluctuation of the water line in the intertidal zone (**Figure 5B**). The higher high water (HHW) line, 0.8 m above mean sea level MSL [49], or average of the highest annual tide levels was chosen as the seaward boundary. The HHW line is typically above the fluctuating water line and provides a repeatable method to measure spatially comparable topographic changes through time, regardless of the yearly, monthly, or daily variability in tide ranges and cycles. The landward extent of the DSM was limited to the foredune scarp because it marks the boundary between non-vegetated and vegetated surfaces, post-SD (**Figure 5B**). By removing significant environmental uncertainties associated with the water and vegetation line, we improve our confidence in monitoring topographic changes occurring between the

within the detectable range of the SfM derived DSM. Here, *z*<sup>2</sup> and *z*<sup>1</sup> represent the elevation of the surface in time series two (i.e., post-SD) and time series one (i.e., pre-SD), respectively, and *x m*ð Þ and *y m*ð Þ are the length and width of the raster pixels ð Þ *p* . To assess the accuracy of the *ΔV* measurements, the propagated error (*PE* (m)); Eq. (2)) reports the magnitude of error ð Þ *e m*ð Þ associated with both time series. Common *e* metrics include the RMSE [36, 38–40, 43, 44] and standard deviation error (*σ*, [50, 51]). Once the *PE* is determined, the volumetric change

each *p* and allows for a range of uncertainty (i.e.,�*ΔVU*) to be reported with the *ΔV*

q

To minimize the uncertainty of the *ΔV* measurements, a threshold is applied to remove values of low magnitude topographic change. The magnitude of change, away from (i.e., �)*PE*, is measured using a statistical t-score approach (*t*; Eq. (4); [52]). A minimum 95% confidence level (*CL*95%; Eq. (5)) threshold of �1.96, valid for large population sizes, is determined based on a two-tailed test that accounts for both negative and positive values that correspond to erosion and deposition, respectively. Absolute *t* values that exceed 1.96 are preserved representing areas of low uncertainty, while *t* values less than 1.96 are classified as zones of high uncer-

*<sup>t</sup>* <sup>¼</sup> *<sup>z</sup>*<sup>2</sup> � *<sup>z</sup>*<sup>1</sup>

*CL*95% <sup>¼</sup> j j*<sup>t</sup> for t*j j<sup>&</sup>gt; <sup>1</sup>*:*<sup>96</sup>

Topographic changes, measured between the pre-SD LiDAR DTM and post-SD UAV DSM, display high magnitude and continuous erosion that mark the foredune

�

*else* ∅

ffiffiffiffiffiffiffiffiffiffiffiffiffi *e*2 <sup>1</sup> <sup>þ</sup> *<sup>e</sup>*<sup>2</sup> 2

*PE* ¼

); Eq. (1)) from pre- and post-SD are reported

); Eq. (3)) provides a universal error value associated with

*ΔV* ¼ ð Þ *z*<sup>2</sup> � *z*<sup>1</sup> *xy* (1)

*ΔVU* ¼ *n p*ð Þ*PE* (3)

*PE* (4)

(2)

(5)

pre-SD LiDAR DTM and post-SD UAV DSM.

*Monitoring Storm Impacts on Sandy Coastlines with UAVs*

*DOI: http://dx.doi.org/10.5772/intechopen.91459*

**3.3 Volumetric change and uncertainty**

Volumetric changes (Δ*V* (m<sup>3</sup>

uncertainty (Δ*VU* (m<sup>3</sup>

**4. Results**

**75**

measurements (e.g., [50, 51]).

tainty and removed from *ΔV* measurements.

#### **Figure 5.**

*Across shore profiles comparing the pre-storm Dorian (SD) LiDAR DTM and post-SD UAV DSM and DTM (A). Inconsistencies occur in the UAV surveys due to over-estimation of elevation values due to environmental uncertainties including the waterline (A) in the intertidal zone (B) and vegetation (A) in the vegetated back dune zone (B). Also, under-estimation of elevation values occurred due to over-filtering of the UAV DTM has almost entirely removed the foredune following SD.*

a profile that was taken from a highly eroded foredune at BB East. Notice, only a small crest remains visible more than 2 m below and 10 m landward of the dune crest measured by the LiDAR DTM (**Figure 5A**). The removal of the foredune from the UAV DTM is indicative of over-filtering, not storm erosion. While good agreement exists between the SfM DTM and DSM over the non-vegetated upper beach surface (**Figure 5A** and **B**), the DTM filtering method could not resolve significant breaks in slope. Therefore, our UAV DTM was not capable of quantifying topographic changes associated with a distinct scarped foredune and will not be used for our post storm measurements.

To limit the environmental uncertainty associated with the UAV DSM, and to create a repeatable survey methodology, topographic change measurements must be confined to areas not affected by potential error introduced by water or

*Monitoring Storm Impacts on Sandy Coastlines with UAVs DOI: http://dx.doi.org/10.5772/intechopen.91459*

vegetation. Difficulty arises when demarcating a seaward boundary due to the fluctuation of the water line in the intertidal zone (**Figure 5B**). The higher high water (HHW) line, 0.8 m above mean sea level MSL [49], or average of the highest annual tide levels was chosen as the seaward boundary. The HHW line is typically above the fluctuating water line and provides a repeatable method to measure spatially comparable topographic changes through time, regardless of the yearly, monthly, or daily variability in tide ranges and cycles. The landward extent of the DSM was limited to the foredune scarp because it marks the boundary between non-vegetated and vegetated surfaces, post-SD (**Figure 5B**). By removing significant environmental uncertainties associated with the water and vegetation line, we improve our confidence in monitoring topographic changes occurring between the pre-SD LiDAR DTM and post-SD UAV DSM.

#### **3.3 Volumetric change and uncertainty**

Volumetric changes (Δ*V* (m<sup>3</sup> ); Eq. (1)) from pre- and post-SD are reported within the detectable range of the SfM derived DSM. Here, *z*<sup>2</sup> and *z*<sup>1</sup> represent the elevation of the surface in time series two (i.e., post-SD) and time series one (i.e., pre-SD), respectively, and *x m*ð Þ and *y m*ð Þ are the length and width of the raster pixels ð Þ *p* . To assess the accuracy of the *ΔV* measurements, the propagated error (*PE* (m)); Eq. (2)) reports the magnitude of error ð Þ *e m*ð Þ associated with both time series. Common *e* metrics include the RMSE [36, 38–40, 43, 44] and standard deviation error (*σ*, [50, 51]). Once the *PE* is determined, the volumetric change uncertainty (Δ*VU* (m<sup>3</sup> ); Eq. (3)) provides a universal error value associated with each *p* and allows for a range of uncertainty (i.e.,�*ΔVU*) to be reported with the *ΔV* measurements (e.g., [50, 51]).

$$
\Delta V = (\mathbf{z}\_2 - \mathbf{z}\_1)\mathbf{x}\mathbf{y} \tag{1}
$$

$$PE = \sqrt{\mathbf{e}\_1^2 + \mathbf{e}\_2^2} \tag{2}$$

$$
\Delta VU = n(p)PE \tag{3}
$$

To minimize the uncertainty of the *ΔV* measurements, a threshold is applied to remove values of low magnitude topographic change. The magnitude of change, away from (i.e., �)*PE*, is measured using a statistical t-score approach (*t*; Eq. (4); [52]). A minimum 95% confidence level (*CL*95%; Eq. (5)) threshold of �1.96, valid for large population sizes, is determined based on a two-tailed test that accounts for both negative and positive values that correspond to erosion and deposition, respectively. Absolute *t* values that exceed 1.96 are preserved representing areas of low uncertainty, while *t* values less than 1.96 are classified as zones of high uncertainty and removed from *ΔV* measurements.

$$t = \frac{z\_2 - z\_1}{PE} \tag{4}$$

$$\text{CL}\_{95\%} = |t| \begin{cases} \text{for} & |t| > \text{1.96} \\ \text{else} & \text{@} \end{cases} \tag{5}$$

### **4. Results**

Topographic changes, measured between the pre-SD LiDAR DTM and post-SD UAV DSM, display high magnitude and continuous erosion that mark the foredune

a profile that was taken from a highly eroded foredune at BB East. Notice, only a small crest remains visible more than 2 m below and 10 m landward of the dune crest measured by the LiDAR DTM (**Figure 5A**). The removal of the foredune from the UAV DTM is indicative of over-filtering, not storm erosion. While good agreement exists between the SfM DTM and DSM over the non-vegetated upper beach surface (**Figure 5A** and **B**), the DTM filtering method could not resolve significant breaks in slope. Therefore, our UAV DTM was not capable of quantifying topographic changes associated with a distinct scarped foredune and will not be used for

*Across shore profiles comparing the pre-storm Dorian (SD) LiDAR DTM and post-SD UAV DSM and DTM (A). Inconsistencies occur in the UAV surveys due to over-estimation of elevation values due to environmental uncertainties including the waterline (A) in the intertidal zone (B) and vegetation (A) in the vegetated back dune zone (B). Also, under-estimation of elevation values occurred due to over-filtering of the UAV DTM has*

*Spatial Variability in Environmental Science - Patterns, Processes, and Analyses*

To limit the environmental uncertainty associated with the UAV DSM, and to create a repeatable survey methodology, topographic change measurements must be confined to areas not affected by potential error introduced by water or

our post storm measurements.

*almost entirely removed the foredune following SD.*

**Figure 5.**

**74**

scarp along BB (**Figure 6A** and **B**). The scarp line consistently eroded into the seaward base of the foredune by >1 m, with higher magnitude erosion of the frontal section of the foredune of over 4 m recorded at eastward extent of the BB East survey (**Figure 6B**). Relatively low magnitude deposition of sediment of 0.2–0.4 m is observed in discontinuous areas of the beach, becoming more prominent at BB west. Despite these areas of significant erosion or depositional changes, 42% of BB (**Figure 6A** and **B**) is classified as a zone of uncertainty (i.e., below the *CL*95% threshold). Low magnitude topographic changes of >�0.11 and <0.11, corresponding to a j j*t* <1.96, often mark an area of transition between sediment eroded from the foredune and deposited onto the beach. This area of transition or slope relaxation is typical of a 'winter' or flattened beach profile following storm events and was observed over the entirety of BB (e.g., **Figures 2** and **3B**).

Post-SD, a total volume change (*ΔV*) of �11,004 m<sup>3</sup> and volume change uncertainty (*ΔVU*) of �4704 m3 was recorded (**Table 1**). High magnitude erosion, accounting for 49% of the detectable surface (**Table 1**), is the primary geomorphic change driver; however, this is associated with significant *ΔVU* accounting for 42% of the cumulative volume change value. To reduce the uncertainty associated with low magnitude elevation changes, the topographic change threshold ð Þ *CL*95% was applied. As a result, *<sup>Δ</sup><sup>V</sup>* remained similar at �11,323 m3 but *<sup>Δ</sup>VU* was reduced to �2659 m<sup>3</sup> (**Table 1**). Notice that the higher magnitude erosional values now represents 57% of the *CL*95% surface and *ΔVU* is reduced, accounting for only 23% of the cumulative volume change. Furthermore, after applying the topographic change threshold the observational area was reduced by 43% and *ΔVU* by 54% (**Table 1**). By accounting for the uncertainty introduced by the survey error, and associated with low magnitude elevation changes, we have systematically reduced *ΔVU* and improve our confidence in reported volume changes post-SD.

**4.1 Alongshore variability**

*reported for all erosional, depositional, and total change surfaces.*

*Monitoring Storm Impacts on Sandy Coastlines with UAVs*

*DOI: http://dx.doi.org/10.5772/intechopen.91459*

Erosion (m<sup>3</sup>

Area (km<sup>2</sup>

Area (km<sup>2</sup>

**Table 1.**

**Figure 7.**

**77**

*Images from pre-SD (C) and post-SD (D).*

Deposition (m<sup>3</sup>

Total change (m3

Total area (km<sup>2</sup>

Alongshore, the eastern extent of the BB survey recorded the highest magnitude topographic change. This is particularly evident in areas that experienced erosion of the full-frontal section of the foredune (**Figure 7A** and **B**). In these locations, the scarp line forms at or behind the former crest line, leading to slope failure,

*A topographic change profile sampled from Brackley Beach (BB) east was measured from the pre-storm Dorian (SD) LiDAR DTM and and post-SD UAV DSM (A) BB east experienced significant scarping that eroded the frontal section of the foredune and lead to large volumetric losses (A and B). Aerial UAV images from pre-SD (C) and post-SD (D) show the removal of the stoss slope and formation of a steep scarp at the former crestline.*

) �15,975 � 2349 �15,201 � 1509

) 4971 � 2354 3878 � 1149

) �11,004 � 4704 �11,323 � 2659

) 0.038 (49%) 0.025 (57%)

) 0.039 (51%) 0.019 (43%)

) 0.077 0.044

*Observed and threshold (CL95%) volume change* ð Þ *ΔV , volume change uncertainty* ð Þ *ΔVU , and area are*

**Observed** *CL***95%**

#### **Figure 6.**

*Elevation changes at Brackley Beach West (A) and East (B), measured from the pre-SD LiDAR DTM and post-SD UAV DSM. A high magnitude and continuous erosional scarp is visible (i.e., in dark red) along the beach-dune boundary with the highest magnitude change occurring at the eastern extent of Brackley Beach East (B). Lower magnitude deposition (i.e., in blue) was observed to develop intermittently on the upper beach surface, becoming more continuous at Brackley Beach West (A).*

*Monitoring Storm Impacts on Sandy Coastlines with UAVs DOI: http://dx.doi.org/10.5772/intechopen.91459*


**Table 1.**

scarp along BB (**Figure 6A** and **B**). The scarp line consistently eroded into the seaward base of the foredune by >1 m, with higher magnitude erosion of the frontal section of the foredune of over 4 m recorded at eastward extent of the BB East survey (**Figure 6B**). Relatively low magnitude deposition of sediment of 0.2–0.4 m is observed in discontinuous areas of the beach, becoming more prominent at BB west. Despite these areas of significant erosion or depositional changes, 42% of BB (**Figure 6A** and **B**) is classified as a zone of uncertainty (i.e., below the *CL*95% threshold). Low magnitude topographic changes of >�0.11 and <0.11,

*Spatial Variability in Environmental Science - Patterns, Processes, and Analyses*

corresponding to a j j*t* <1.96, often mark an area of transition between sediment eroded from the foredune and deposited onto the beach. This area of transition or slope relaxation is typical of a 'winter' or flattened beach profile following storm events and was observed over the entirety of BB (e.g., **Figures 2** and **3B**).

tainty (*ΔVU*) of �4704 m3 was recorded (**Table 1**). High magnitude erosion, accounting for 49% of the detectable surface (**Table 1**), is the primary geomorphic change driver; however, this is associated with significant *ΔVU* accounting for 42% of the cumulative volume change value. To reduce the uncertainty associated with low magnitude elevation changes, the topographic change threshold ð Þ *CL*95% was applied. As a result, *<sup>Δ</sup><sup>V</sup>* remained similar at �11,323 m3 but *<sup>Δ</sup>VU* was reduced to �2659 m<sup>3</sup> (**Table 1**). Notice that the higher magnitude erosional values now represents 57% of the *CL*95% surface and *ΔVU* is reduced, accounting for only 23% of the cumulative volume change. Furthermore, after applying the topographic change threshold the observational area was reduced by 43% and *ΔVU* by 54% (**Table 1**). By accounting for the uncertainty introduced by the survey error, and associated with low magnitude elevation changes, we have systematically reduced *ΔVU* and

*Elevation changes at Brackley Beach West (A) and East (B), measured from the pre-SD LiDAR DTM and post-SD UAV DSM. A high magnitude and continuous erosional scarp is visible (i.e., in dark red) along the beach-dune boundary with the highest magnitude change occurring at the eastern extent of Brackley Beach East (B). Lower magnitude deposition (i.e., in blue) was observed to develop intermittently on the upper beach*

*surface, becoming more continuous at Brackley Beach West (A).*

improve our confidence in reported volume changes post-SD.

**Figure 6.**

**76**

Post-SD, a total volume change (*ΔV*) of �11,004 m<sup>3</sup> and volume change uncer-

*Observed and threshold (CL95%) volume change* ð Þ *ΔV , volume change uncertainty* ð Þ *ΔVU , and area are reported for all erosional, depositional, and total change surfaces.*
