Download as pdf or txt
Download as pdf or txt
You are on page 1of 44

Estimation of grain sizes in a river through UAV-based SfM photogrammetry

Thesis

Presented in Partial Fulfillment of the Requirements for the Degree Master of Science in

the Graduate School of The Ohio State University

By

Tyler Wong

Graduate Program in Environment and Natural Resources

The Ohio State University

2022

Thesis Committee

Dr. Steve W. Lyon, Advisor

Dr. Kaiguang Zhao

Dr. Sami Khanal

1
Copyrighted by

Tyler Wong

2022

2
Abstract

Unmanned aerial vehicles (UAVs) have an increasingly relevant role in the field

of hydrology and water resources management. Their affordability and ease of use in

comparison to traditional field-based methods have made research on their applications

increase rapidly in the past decade. One application of UAVs to the hydrology of river

systems is the estimation of particle sizes within a channel. This project investigated the

ability of UAV imagery and Structure-from-Motion (SfM) photogrammetry to estimate

grain-size distributions within a reach along the Olentangy River. To do this, we selected

a study reach within the Highbanks Metro Park that was approximately 250 m in length

and 50 m in width. We flew a DJI Mavic 2 Pro quadcopter UAV and collected imagery

of subaerially exposed grains throughout gravels bars within this study reach. These

images were processed using a SfM workflow that yielded point clouds and orthomosaics

from which we extracted multiple topography-based and image-based metrics to be used

as proxies for grain sizes. We then calibrated statistical regression models to predict the

D50 and D84 grain size percentiles from these grain size proxies. While previous literature

has suggested that topographic roughness metrics outperform image textural metrics for

statistical grain size estimation, our study showed that the statistical models that were

calibrated based on image textural properties performed better than those that were

calibrated based on point cloud roughness properties. This contradiction may reflect the

ii
unique nature of our study site where the grains were dominated by smaller particles in

comparison to other studies. The smaller grain sizes in our study area would have likely

produced less significant topographic signatures in comparison to larger grains, which

makes topographic roughness difficult to accurately measure and apply to statistical grain

size estimation techniques. The results of this study suggest that topography-based grain

size estimation may not be adequate for all sites, and further work on analyzing the range

of grain size characteristics for which topography-based and image-based techniques

perform better should be done to improve the applicability of these techniques. Doing so

will help river scientists and managers to easily assess the physical, chemical, and

biological dynamics that occur within rivers.

iii
Acknowledgments

This work was supported by fellowships from The Ohio State University

Graduate School and the Ohio Agricultural Research and Development Center. Thank

you to Dr. Steve Lyon for mentoring me throughout the past two years. Planning and

carrying out this project seemed dauting at times, but you always managed to give me

new perspectives about my research and guide me through any obstacles I faced. Thank

you to Dr. Kaiguang Zhao and Dr. Sami Khanal for serving on my committee and

providing feedback on my ideas for this thesis project. I am grateful to Dr. Khanal for

allowing me to borrow her drones for this project and assisting in planning the field

survey. I would also like to thank John Brett, Shannon Pace, and Kushal KC for their

help in the field.

iv
Vita

2020............................B.S. Geology, Temple University, Department of Earth and

Environmental Science

2022............................M.S. Environment and Natural Resources, The Ohio State

University, School of Environment and Natural Resources

Fields of Study

Major Field: Environment and Natural Resources

v
Table of Contents

Abstract ............................................................................................................................... ii
Acknowledgments.............................................................................................................. iv
Vita...................................................................................................................................... v
List of Tables ................................................................................................................... viii
List of Figures .................................................................................................................... ix
Chapter 1. Introduction ....................................................................................................... 1
Chapter 2. Methods ............................................................................................................. 6
2.1. Field experiment ...................................................................................................... 6
2.1.1. Site description.................................................................................................. 6
2.1.2. Collection of calibration imagery ..................................................................... 8
2.1.3. UAV survey ...................................................................................................... 9
2.2. Data processing ........................................................................................................ 9
2.2.1. Ground truth data ............................................................................................ 11
2.2.2. SfM photogrammetry ...................................................................................... 11
2.3. Statistical grain size estimation .............................................................................. 12
2.3.1. Roughness height (rh) ..................................................................................... 13
2.3.2. Standard deviation of elevations (σz) .............................................................. 13
2.3.3. Image entropy ................................................................................................. 14
2.3.4. Calibration and validation of grain size proxies ............................................. 14
Chapter 3. Results ............................................................................................................. 18
3.1. Ground truth data ................................................................................................... 18
3.2. Calibration and validation relationships ................................................................ 19
Chapter 4. Discussion and conclusion .............................................................................. 27
4.1. On the efficacy of image-based versus topography-based grain size proxies ....... 27
4.2. Limitations and recommendations ......................................................................... 29
vi
4.3. Conclusion ............................................................................................................. 31
Bibliography ..................................................................................................................... 32

vii
List of Tables

Table 1. The 95% confidence intervals (CI) for the predicted D50 and D84 based on the
roughness and texture metrics. .......................................................................................... 24

viii
List of Figures

Figure 1. (a) Location of the study reach. (b) Google Earth satellite imagery around study
reach, which is shown in yellow. (c) Drone imagery of the study reach. The direction of
flow is shown by the white arrow. (d) Example of calibration image from the study area. 7
Figure 2. Processing workflow diagram. .......................................................................... 10
Figure 3. Examples of the different roughness metrics calculated for one of the
calibration plots, including the RGB image from the orthomosaic, the entropy image, the
nadir view of the roughness height point cloud, and the nadir view of the standard
deviation of elevation point cloud. Each example corresponds to the most optimal kernel
size. ................................................................................................................................... 15
Figure 4. The normalized histogram distributions of the different roughness metric for
one of the calibration plots and their average values (shown by the red vertical lines). .. 16
Figure 5. The Wolman pebble count distribution (shown by the red dashed line) and the
b-axis grain size distributions for the 13 calibration plots obtained through photosieving.
........................................................................................................................................... 18
Figure 6. D50 calibration and validation relationships using the best kernel size for each
roughness metric. In the calibration plots, the red solid line shows the line of best fit. In
the validation plots, the black solid line shows the line of best fit and the red dashed line
shows the 1:1 line.............................................................................................................. 22
Figure 7. D84 calibration and validation relationships using the best kernel size for each
roughness metric. In the calibration plots, the red solid line shows the line of best fit. In
the validation plots, the black solid line shows the line of best fit and the red dashed line
shows the 1:1 line.............................................................................................................. 23
Figure 8. The relationships between R2 and kernel size for the calibration and validation
models on D50 and D84. For the entropy metric, the pixel radius is multiplied by the
spatial resolution of the orthomosaic. ............................................................................... 24
Figure 9. Maps of predicted D50 values over the entire surveyed area. Each map uses the
most optimal model found for each respective metric. Each pixel is 1 m2 in size. ........... 25
Figure 10. Maps of predicted D84 values over the entire surveyed area. Each map uses the
most optimal model found for each respective metric. Each pixel is 1 m2 in size. ........... 26

ix
Chapter 1. Introduction

The quantification of grain sizes is a common component of river assessments as

the distribution of grain sizes in a fluvial setting is linked to a river’s physical, chemical,

and biological functioning. Grain sizes in rivers can range from clays, which can be as

small as 1 μm, to boulders, which are larger than 256 mm. Grains may be measured along

three orthogonal axes: the longest a-axis (the “length” of a grain), the intermediate b-axis

(the “width” of a grain), and the shortest c-axis (the “thickness” of a grain). An example of

their relevance to river processes is how grain sizes within a river relate to sediment

transport driving erosional and depositional forces, which directly influence aggradational

and degradational forcings (Dade and Friend, 1998; Church, 2002). The geomorphological

characteristics of a river system such as its channel morphology and network geometry are

thereby reflective of the river’s grain size distribution (Frissell, 1986; Leopold, 1992;

Benda et al., 2004). These geomorphological properties can affect the ecological health

within rivers through their influence on substrate texture and habitat quality (Newson and

Newson, 2000; Wohl et al., 2015). Information on grain size distributions is also necessary

for a variety of models relating to riverine morphodynamics and hydraulics (Carrivick and

Smith, 2019). Because of this, grain size assessments are crucial to sustainable river

management practices.

1
Traditional methods of assessing grain size typically involve field-based and

laboratory-based techniques such as pebble counts of sieve analyses (Wolman, 1954; Rice

and Church, 1964). Such methods may often be laborious and costly - and these methods

are typically inadequate for describing spatial variations of grain sizes throughout a river

(Graham et al., 2005; Woodget and Austrums, 2017; Danhoff and Huckins, 2020). As an

alternative to these methods, remote sensing techniques have been proposed as more

efficient methods for quantifying grain sizes in rivers.

In general, remote sensing techniques for grain size assessment may be classified

as either image-based or topography-based. Image-based techniques were initially

developed based on imagery taken from manned aircraft, satellites, or close-range

photography. From these images, image-based techniques generally aim to either measure

visible grains directly, which is known as photosieving (Graham et al., 2005), or use

metrics of image texture as proxies for grain size related through statistical relationships

with observed grain sizes (Carbonneau et al., 2004). With regard to this latter aspect,

topography-based methods are similar in that they calibrate statistical relationships

between grain size and metrics; however, the difference is that the metrics are primarily

computed from topography data such as point clouds or elevation models (Brasington et

al., 2012). A central limitation for these early remote sensing techniques was resolution of

images or topographic data such that assessments were often limited to either the patch-

scale, as was the case with statistical topography-based methods, or the catchment-scale,

as was the case with statistical image-based methods. Furthermore, these techniques

required expensive equipment and needed a significant amount of field and processing

2
work (Woodget and Austrums, 2017). Because of these aspects, early remote sensing

techniques for estimating grain sizes were difficult to apply to the reach scale. In addition,

these remote sensing analyses are usually limited to subaerial rather than submerged grains

due to the difficulty in correcting for refraction through water. As such, multiple scenes or

data collected at various flow conditions (i.e., high and low flows) becomes required to

fully explore the grain size distributions. To address some of these limitations around

resolution and the need for repeatable scans, recent research has focused on the

transferability of standard remote sensing techniques for assessing grain sizes to data

derived through unmanned aerial vehicles (UAVs).

The use of UAVs in the river sciences in general has gained significant interest in

recent years such that UAVs have been used to study a variety of aspects of rivers beyond

grain size distributions. Examples of such applications include streamflow measurements,

bathymetric mapping, geomorphological change detection, and stream ecological

assessments (Carrivick and Smith, 2019; Acharya et al., 2021; Vélez-Nicolás 2021). A

primary reason for this increasing interest is the increasing ease of operation and

affordability of UAVs. In addition, the ability to attach various sensors to UAVs and deploy

them at high spatial and temporal resolutions have made them versatile instruments in the

hydrological and river sciences (Rhee et al., 2018; Carrivick and Smith, 2019; Tomsett and

Leyland, 2019; Acharya1 et al., 2021; Vélez-Nicolás 2021). UAVs are able to quickly

survey large areas at a speed and spatial resolution suitable for reach-scale analyses and

are also easy to deploy multiple times, offering a higher temporal resolution for remote

sensing in riverine environments. The wide range of possibility offered by UAV

3
technology permits the coupling of remote sensing of grain sizes with other hydrological

measurements of interest, which makes them an attractive tool for fluvial investigations.

Investigations on the transferability of early grain size estimation techniques to data

collected from UAVs have shown the ability for reach-scale assessments, effectively

bridging the gap in scales (Woodget and Austrums, 2017).

So, how are UAVs being utilized for grain size assessments? Primarily, processing

UAV imagery through workflows based on Structure-from-Motion (SfM) photogrammetry

yields three-dimensional point clouds and orthomosaics. These products can be used for

both topographically-based and image-based approaches for grain size estimation. For

topographically-based approaches, UAVs are significantly more cost-effective in

comparison to LiDAR systems that have previously been used for such studies (Brasington

et al., 2012), and studies that have tested the transferability of these approaches to UAV

data have observed strong calibration relationships between UAV-derived topographic

roughness and grain size. Vázquez-Tarrío et al. (2017) for example applied a

topographically-based approach to a braided gravel-bed river and reported a maximum R2

value of 0.89 for the calibration relationship between grain size and point cloud roughness.

Studies that have used image-based approaches have also shown to be successful.

Tamminga et al. (2015) for example reported a calibration R2 value of 0.82 for the

relationship between grain size and orthomosaic image texture. However, the difference in

performance between topography-based and image-based approaches is poorly

understood. Woodget and Austrums (2017) found that models calibrated on point cloud

roughness, which had a maximum calibration R2 value of 0.80, performed better than those

4
calibrated on image texture, which had a maximum calibration R2 value of 0.48. In contrast,

Woodget et al. (2018) suggested that image-based methods may outperform topography-

based ones depending on certain site-specific characteristics. That work highlighted that

there are a range of conditions such as the shape, size, and imbrication of grains where

image-based and topography-based approaches perform optimally.

Because of this, there is ambiguity in the current literature on the efficacy of image-

based techniques in comparison to topographically-based techniques when using UAV-

derived point clouds and orthomosaics. To address this ambiguity, the purpose of this paper

is to compare the performance of statistical grain size models based on image properties to

those based on topographic properties. We will apply these approaches to a large stream in

the Midwestern United States where the grains consist of moderately imbricated medium-

sized gravels. We are interested in how the results of this analysis compare with previous

studies and their respective grain characteristics. Specifically, how do image-based metrics

of grain size compare to topography-based metrics when applied to our stream of interest?

Answering this will aid our understanding around the range of conditions in which

topography-based and image-based statistical grain size estimation approaches perform

better. Ultimately, this improved understanding will help enable UAV-assisted grain size

estimation techniques to become more robust and accessible for river scientists and

managers.

5
Chapter 2. Methods

2.1. Field experiment

2.1.1. Site description

We conducted our experiment along a reach of the Olentangy River in Columbus,

Ohio (Figure 1). The length of our study reach is approximately 250 m long, and the

bankfull width along this reach is approximately 50 m. The Olentangy River is a low-

gradient meandering stream and is classified as a Rosgen Type C stream. Our study reach

is located within the Highbanks Metro Park, and the banks along this reach contain forests.

We limited our study to a series of gravel bars along this reach, which is common for

remote sensing assessments of grain size due to the impact of water refraction on image

properties (Woodget and Austrums, 2017). These gravel bars consist of alluvial sands,

gravels, and cobbles. The bedrock at this location is the Ohio Shale, and colluvial shale

fragments are present throughout the gravel bars due to erosion of the banks. Discharge at

this location is influenced by a dam that is approximately 30 km upstream. A stream

gauging station maintained by the United States Geological Survey (USGS) is located near

our study site.

6
a b

Figure 1. (a) Location of the study reach. (b) Google Earth satellite imagery around study
reach, which is shown in yellow. (c) Drone imagery of the study reach. The direction of
flow is shown by the white arrow. (d) Example of calibration image from the study area.
We conducted our experiment in early September 2021 during a period of low flow.

For the 2021 water year, the USGS stream gauge reported an annual mean discharge of

556.9 ft3/s, and the minimum and maximum daily mean discharge were 32.5 and 4750 ft3/s.

On the day of our survey, the mean daily discharge was 79.5 ft3/s. This was within the 20th
7
percentile of daily mean discharges for the 2021 water year. Weather conditions were clear

and sunny. We initially performed a pebble count by traversing the gravel bars in a grid

pattern and measuring each sample with a gravelometer. We sampled a total of 110 grains

throughout the bars. The purpose of this pebble count was to act as a comparison to our

remote sensing analysis.

2.1.2. Collection of calibration imagery

We collected calibration imagery over subaerially exposed grains. To do this, we

constructed a 3 ft x 3 ft frame from PVC pipe. We then selected locations throughout the

gravel bar for calibration plots. These plots should ideally be heterogeneous in terms of

grain size throughout each individual calibration frame, and together they should all

represent a range of grain sizes. To ensure an even distribution of plots throughout the

study area, we placed plots near the downstream edge of a bar, in the center of a bar, and

near the upstream edge of a bar. We distributed a total of 13 calibration plots throughout

the bars. This number of samples follows the findings from Vázquez-Tarrío et al. (2017)

who used a sensitivity analysis to show that a minimum of 9 or 10 calibration plots are

recommended for calibrating statistical grain size models. To collect calibration images,

we placed this frame over a selected location. We then marked two opposite corners of the

frame and geolocated these markers with a Trimble RTK GPS unit. We took images of

these plots with a smartphone camera (Figure 1b). This camera used a 64 MP Sony IMX682

sensor and has a 35 mm equivalent focal length of 25.4 mm. These images had a resolution

of 9248 x 6936 pixels. To take these images, we positioned the camera approximately

8
parallel to the ground and held the camera at a height such that the calibration image fully

captured the frame.

2.1.3. UAV survey

We manually flew a DJI Mavic 2 Pro at an altitude of approximately 4 m and took

images throughout the gravel bars. This drone contains an integrated camera with a 20 MP

(5472 x 3648 pixels) 1” CMOS sensor, and it has a 35 mm equivalent focal length of 28

mm and a field of view of 77°. This drone uses a 3-axis gimbal, which works to stabilize

the camera and reduce blurriness in the collected imagery. While flying the UAV, we

collected RGB images at nadir configuration and collected images with a visually

estimated degree of overlap. Previous literature has shown that motion blur may negatively

affect the quality of remote sensing grain size analysis (Woodget et al., 2018). To minimize

the effect of motion blur, we flew at a low velocity and decelerated the drone when taking

images. We separated the survey into four flights, and we collected a total of 427 images.

2.2. Data processing

A summary of our processing workflow is shown by Figure 2. In the following, the

various steps and iterations are described in detail.

9
Figure 2. Processing workflow diagram.
10
2.2.1. Ground truth data

The images of the plots that we took with a smartphone served as our ground truth

data that would be used to calibrate and validate the statistical grain size models. The

calibration frame served as a constant scale for the image so that the grains that were visible

in the image could be measured. To estimate the grain size distributions within these plots,

we used the photosieving software BASEGRAIN (Detert, 2020). We initially scaled each

image according to the fixed size of the calibration frame. The average resolution of these

photos was approximately 0.2 mm/pixel. The default software parameters tended to over-

segment individual grains. To account for this, we adjusted the parameters to process the

image more smoothly by decreasing medfiltsiz10 and increasing facgraythr1. After the

photosieving process was completed, we post-processed the delineated grains by manually

excluding non-grain objects (e.g., leaves, branches, etc.) and merging over-segmented

grains. Previous work has shown that truncating the grain size distributions for the

calibration data improves the performance of statistical grain size models (Vázques-Tarrío

et al., 2017). We truncated the grain size distribution at a minimum b-axis length of 10 mm,

which is the default parameter for the program.

2.2.2. SfM photogrammetry

We used Pix4Dmapper to process drone images and processed all of the images

over each of the four flight as separate projects. Our GPS measurements served as ground

control points for georeferencing the photogrammetric data. From the default calibration

11
parameters, we changed the Internal Parameters Optimization setting from “All” to “All

Prior” to account for the flat and homogeneous nature of our imagery. For the point clouds,

we selected a high point density. The resolution of the orthomosaics were set to be equal

to the ground sampling distance of the imagery, which averaged approximately 0.8

mm/pixel among the separate flights. The mean RMS georeferencing errors for these

projects ranged from 0.01 m to 0.05 m.

We imported the processed point clouds and orthomosaics with our GPS

measurements into QGIS for further analysis. Computing image and point cloud metrics

over the entire study area would require too much computer storage and processing power,

so we first crop the calibration plots from the orthomosaics and point clouds. To do this,

we vectorized the calibration plots into polygons based on the geolocated corners of the

plots. We then created 1 m buffers around each of these polygons. These buffers prevent

the edges of the plots from affecting the computation of grain size proxies. The

orthomosaics and point clouds were then masked by these buffers to isolate the calibration

plots and the area surrounding them. After the computation of following grain size proxies,

the resulting point clouds and images were clipped to the original shape of the plots.

2.3. Statistical grain size estimation

Using these clipped calibration plots, we computed three different point cloud and

image metrics that have previously been used as proxies for grain sizes. For each point or

pixel in the point cloud and orthomosaic, these metrics, which serve as proxies for grain

size, were calculated from the surrounding neighborhood within a certain search radius or

12
kernel size. The average metric value was then extracted from each calibration plot. Given

the grain size percentiles obtained from the photosieving analysis and the average values

for each grain size proxy, a statistical grain size model can be produced through linear

regression. The grain size metrics that we considered are described below.

2.3.1. Roughness height (rh)

The roughness height is a topography-based metric that corresponds to the

deviation from the best-fitting plane of a neighborhood of points in the point cloud

(Woodget and Austrums, 2017; Vázquez-Tarrío et al., 2017; Woodget et al., 2018; Chardon

et al., 2020). To compute this metric, we used the roughness tool provided by

CloudCompare (CloudCompare 2.12, 2022). This tool permits various kernel sizes to be

used, which define the radius of the neighborhoods around each point. The result of this

computation is a point cloud where each point contains a scalar value that is equal to its

roughness height for a given kernel size.

2.3.2. Standard deviation of elevations (σz)

The standard deviation of elevations is a topography-based metric that has

commonly been used in statistical grain size estimation (Heritage and Milan, 2009;

Vázquez-Tarrío et al., 2017; Pearson et al., 2017; Groom et al., 2018). To calculate this

metric, we used the KDTree class from SciPy, which is a scientific computing library for

the Python programming language (Virtanen et al., 2020). We imported the point cloud

and queried neighboring points within a given radius using the functions from this class.

13
For each point in the point cloud, we queried its neighborhood, computed the standard

deviation of the elevations within this neighborhood, and assigned this value to the

respective point in the point cloud.

2.3.3. Image entropy

The entropy of an image is a measurement of its texture and has been used to

estimate grain size (Carbonneau et al., 2005; Dugdale et al., 2010; Woodget and Austrums,

2017; Woodget et al., 2018). To calculate this third metric, we converted the calibration

plots clipped from the orthomosaic to grayscale images and used the entropy function from

scikit-image, which is an image-processing library for Python. In this case, the entropy of

an 8-bit image is defined as


255

𝐸𝑛𝑡𝑟𝑜𝑝𝑦 = − ∑ pi log 2 (pi )


i=0

where pi is the frequency of the gray level 𝑖 in the normalized image histogram. For each

pixel, a disk with a given radius is used to extract neighboring pixels. The results of this

computation are entropy images where each pixel contains the calculated entropy over its

neighboring pixels.

2.3.4. Calibration and validation of grain size proxies

To determine the optimal radius or kernel size for these metrics, we used an iterative

approach similar to the one presented by Woodget et al. (2018). For the topography-based

metrics, we repeated the computations described above for kernel sizes ranging from 10 to

14
0 mm in a step size of 10 mm. For the image-based metric, we repeated the computation

for kernel sizes ranging from 1 to 10 pixels in a step size of 1 pixel. Given the resolution

of our orthomosaic, this translates to a range of kernel sizes from 8 to 80 mm.

Figure 3. Examples of the different roughness metrics calculated for one of the
calibration plots, including the RGB image from the orthomosaic, the entropy image, the
nadir view of the roughness height point cloud, and the nadir view of the standard
deviation of elevation point cloud. Each example corresponds to the most optimal kernel
size.
Figure 3 shows examples of the three different grain size proxies for one of the

calibration plots. In the entropy image, larger grains are represented by lower entropy

15
values. In the roughness height and standard deviation of elevation point clouds, larger

grains are represented by higher values. Figure 4 shows the corresponding distributions of

values for the three metrics. For each of possible kernel sizes and proxies, we extracted the

average proxy values from all 13 plots, which we used to calibrate the statistical grain size

models.

Figure 4. The normalized histogram distributions of the different roughness metric for
one of the calibration plots and their average values (shown by the red vertical lines).
We compared the performance of the different grain size proxies and kernel sizes

through a leave-one-out cross-validation approach (LOOCV). One of the advantages of

LOOCV is that it is a more robust method for model validation when using a small sample

size such as ours. To do this, we first excluded one of the calibration plots from the dataset.

The remaining plots were then used to calibrate a univariate linear regression model of the

plots’ grain size percentiles as a function of their average proxy value. We used this linear

model to predict the grain size percentile of the plot that had been excluded from the

calibration dataset. To repeat this for another plot, we excluded a different plot from the

calibration dataset and calibrated a new linear model that was used to predict a grain size

percentile for the excluded plot. This process was repeated until grain size percentiles for

each plot had been predicted. To quantify the quality of the validation result, we plot the

observed grain size percentiles against the predicted grain size percentiles and compute the

16
R2 and slope of the line of best fit. We repeated this LOOCV procedure for each of the

three grain size proxies of interest over a range of kernel sizes.

17
Chapter 3. Results

3.1. Ground truth data

Figure 5 shows the grain size distributions obtained through the Wolman pebble

count and the photosieving analysis for the 13 calibration plots. Due to the use of a

gravelometer, we focus only on the b-axis measurements obtained from photosieving for

this analysis. In comparison to the Wolman pebble count, the photosieving distributions

underestimate the frequency of small grains due to the truncation made at 10 mm.

Figure 5. The Wolman pebble count distribution (shown by the red dashed line) and the
b-axis grain size distributions for the 13 calibration plots obtained through photosieving.

18
However, the range of D50 and D84 measurements obtained from the photosieving

analysis includes the measurements made from the Wolman pebble count. We observed

D50 measurements from 13.0 to 15.8 mm and D84 measurements from 18.9 to 29.1 mm

throughout the 13 calibration plots. The D50 and D84 measurements from our pebble count

were 14.1 and 25.1 mm, respectively. Therefore, we trust the validity of the calibration

data for the purposes of predicting the D50 and D84 percentiles.

3.2. Calibration and validation relationships

Figure 6 shows the strongest calibration and validation relationships for the three

metrics based on the b-axis D50 values from the photosieving analysis of the 13 calibration

plots. Figure 7 similarly shows this for the D84 measurements. For the D50, the entropy

metric shows the highest validation R2 of 0.59 and slope of 0.65. The validation

relationships for the other metrics were significantly weaker, with the roughness height

metric having a validation R2 value of 0.28 and slope of 0.35 and the σz metric having a

validation R2 value of 0.20 and slope of 0.28. For the D84, the validation relationships were

slightly stronger for all of the three metrics than they were for the D50 relationships. The

entropy metric had the highest validation R2 of 0.62 and slope of 0.66. The roughness

height metric had a validation R2 of 0.30 and slope of 0.37. The σz metric had a validation

R2 of 0.25 and slope of 0.32. These figures show that these models generally underestimate

grain size, but this effect is least pronounced in the case of the entropy metric. These

patterns are similar for the calibration relationships where, for the D50 relationships, the

entropy metric has the highest calibration R2 value of 0.68, the roughness height metric has

19
a calibration R2 of 0.46, and the σz metric has the lowest calibration R2 of 0.39. For the D84

relationships, the entropy metrics has a calibration R2 value of 0.71, the roughness height

metric has a calibration R2 of 0.48, and the σz metric has the lowest calibration R2 of 0.43.

In the calibration models for the topography-based metrics, there are several outliers that

worsen the model performance that have a high topographic signature but low measured

grain size.

Figure 8 shows how the performance of these metrics relate to the kernel size used

to process them. For these relationships, the optimal kernel size for all of these metrics is

between 30 and 40 mm. This kernel size represents the distance that most accurately

captures the textural and roughness signatures that are present in the orthomosaics and

point clouds. When the kernel size is too small, the grain size proxy becomes skewed by

local deviations in texture or roughness so that the kernel window does not include an

adequate number of grains to calculate a representative value for image texture or point

cloud roughness. Conversely, when the kernel size is too large, the grain size proxy

becomes less sensitive to textural and roughness variations, which causes the grain size

proxy to inadequately detect textural and roughness features and have a weaker correlation

with grain size.

Table 1 shows the 95% confidence intervals for the predicted values of D50 and D84

based on the entropy model compared to the values from the pebble count. The D50 from

the Wolman pebble count is within the confidence interval for the entropy-predicted D50.

However, the D84 from the pebble count is slightly higher than the confidence interval for

the entropy-predicted D84. The confidence intervals for the D50 and D84 both of the

20
topography-based predictions are also shown. For the D50, these models accurately

captured the grain size obtained from our pebble count. However, for the D84, these models

tended to underestimate the grain size from our pebble count more than the entropy-based

model. This is an agreement with the low validation slopes observed for the two

topography-based models, which translates into overestimation of grain size for smaller

particles and underestimation for larger ones.

Figures 9 and 10 show the application of the calibration models to the remainder of

the survey area that had been imaged in the study. These maps show that the models based

on roughness height and σz predict similar grain sizes throughout the gravel bars but

contrast strongly with the entropy-based model. In general, the topography-based models

predict larger grains in areas where the entropy-based model predicts small grains. This is

likely due to the aforementioned tendency for the topography-based models overestimate

the size of smaller grains.

21
Figure 6. D50 calibration and validation relationships using the best kernel size for each
roughness metric. In the calibration plots, the red solid line shows the line of best fit. In
the validation plots, the black solid line shows the line of best fit and the red dashed line
shows the 1:1 line.
22
Figure 7. D84 calibration and validation relationships using the best kernel size for each
roughness metric. In the calibration plots, the red solid line shows the line of best fit. In
the validation plots, the black solid line shows the line of best fit and the red dashed line
shows the 1:1 line.
23
Figure 8. The relationships between R2 and kernel size for the calibration and validation
models on D50 and D84. For the entropy metric, the pixel radius is multiplied by the
spatial resolution of the orthomosaic.

Table 1. The 95% confidence intervals (CI) for the predicted D50 and D84 based on the
roughness and texture metrics.
Pebble count Entropy 95% CI Roughness height 95% CI σz 95% CI
D50 14.12 mm 13.73 – 14.62 mm 13.81 – 14.51 mm 13.83 – 14.51 mm
D84 25.14 mm 21.29 – 24.35 mm 21.55 – 24.01 mm 21.61 – 23.99 mm

24
Figure 9. Maps of predicted D50 values over the entire surveyed area. Each map uses the
most optimal model found for each respective metric. Each pixel is 1 m2 in size.
25
Figure 10. Maps of predicted D84 values over the entire surveyed area. Each map uses the
most optimal model found for each respective metric. Each pixel is 1 m2 in size.
26
Chapter 4. Discussion and conclusion

4.1. On the efficacy of image-based versus topography-based grain size proxies

Our results demonstrate that the image-based entropy metric computed from a

UAV-derived orthomosaic performed better at estimating D50 and D84 measurements than

topography-based metrics computed from UAV-derived point clouds (Figures 6 and 7).

The calibration and validation R2 values for the entropy models indicate a moderately good

predictive relationship between image entropy and sedimentological properties in this river

of interest that failed to be captured by the topography-based grain size proxies

The poor performance of the topography-based metrics may be attributed to the

characteristics of the grains within our study site such as their size, shape, and degree of

imbrication. Metrics for predicting grain size from topography have traditionally been

applied in rivers where the grains are larger, which would make fine-scale topographic

variations more significant (Woodget and Austrums, 2018). The range of grain sizes

studied in this paper are finer than previous UAV-based grain size estimation studies that

have typically focused on topography-based metrics. Smaller grains result in less

significant fine-scale topographic variations between particles, which worsens the ability

to correlate point cloud roughness to grain size. Additionally, a large portion of the grains

within our study area are shales that are flatter in shape, which would result in very little

fine-scale topographic variability near those grains. Furthermore, topography-based

27
metrics have been shown to be sensitive to the imbrication of grains, where a high degree

of imbrication limits the ability to use fine-scale topographic variations as a proxy for grain

size (Pearson et al., 2017; Casado et al., 2020). We observed a slight degree of imbrication

in grains throughout the study area (see Figure 2), which would have contributed to the

poor performance of the topography-based metrics.

The entropy metric, which is computed from only a two-dimensional image of the

grains rather than a three-dimensional point cloud, performed better than the topography-

based metrics as it does not rely on the limited fine-scale topographic variability throughout

our study area. Instead, entropy represents the textural characteristics of the orthomosaic,

and our results show that this textural metric correlates well with grain sizes. This finding

agrees with previous work on grain size estimation though image texture. For example,

Dugdale et al. (2010) suggested that entropy is a suitable metric for estimating smaller

grain sizes as it amplifies the contrasts between grain boundaries through the logarithmic

component in its computation. Because of this, entropy appears to be a more robust proxy

for grain size in areas with smaller grains where the light-dark boundaries between particles

tend to be less clear.

Recent research on techniques for grain size estimation from UAV imagery and

SfM photogrammetry further support our findings. For example, Woodget et al. (2018)

similarly compared how image-based and topography-based metrics correlate to grain size,

and they found that models calibrated on the roughness height metric yielded weaker

validation relationships when compared to those calibrated on image entropy. Their study

area had b-axis D84 measurements of approximately 0.07 m, while ours were approximately

28
0.025 m (Figure 8). Since we observe a similar contrast between the image-based and

topography-based metrics, the results of this paper provide further evidence that image

textural approaches are more accurate than topographic roughness approaches in settings

where grain sizes are small. Future research should attempt to clearly define the ranges in

grain size for which image textural metrics and topographic roughness metrics each

perform optimally.

4.2. Limitations and recommendations

A graphical overview of the processes discussed in this paper is shown in Figure 3.

This UAV-based workflow presented in this study has demonstrated promise at estimating

grain size percentiles in rivers with inexpensive equipment. The statistical grain size

models calibrated based on orthomosaic image texture were shown to perform best in this

stream characterized by slightly imbricated medium-sized gravels. However, further

research into the conditions that influence the performance of these metrics still needs to

be done. This study demonstrates that processing both image-based and topography-based

grain size proxies using UAV imagery is straightforward to translate into grain size

estimates, and investigating a variety of kinds of grain size proxies when conducting UAV-

based grain size estimation may prove to be a robust and comprehensive method of

quantifying grain size distributions in rivers.

One specific source of error that we observed in the topography-based models was

an outlier that produced a high topographic signature but a low grain size percentile

measurement (see Figures 6 and 7). This could be due to a limitation in our method of

29
producing the calibration measurements. Our method of using photosieving assumes that

each grain lies flat on the surface such that the a- and b-axes are fully expressed from the

images taken with a nadir perspective and that the c-axis is perpendicular to the surface.

The microtopography near that particle is then controlled by the c-axis. When the grains

lie at an angle and are not flat, the photosieving measurements made from the nadir angle

become smaller as the c-axis becomes visible and the expression of the longer a- and b-

axes increases. This in turn allows the a- and b-axes to contribute more to the

microtopography, inflating the computed metrics. The majority of the grains from this

specific plot could have been oriented this way, causing a high topographic signature in

the point cloud but a small footprint in the nadir calibration imagery. Because of this, future

efforts for statistical grain size estimation should attempt a variety of methods for

measuring the calibration data set such as physical pebble counts in addition to

photosieving.

For the purposes of grain size estimation in rivers characterized by smaller grains

through the use of UAV-based SfM, orthomosaic quality is, therefore, vital for accurate

sedimentological analyses. Our method of manually piloting the drone allowed us to limit

blurriness in the captured images. However, this method was slow and required multiple

battery changes to fully cover the study area, which may prove problematic for larger and

less accessible study areas. In the future, improving the battery life and the use of a higher

resolution camera may expedite the image collection process. Future research to determine

the optimal balance between image resolution, flight height, and survey time for these

techniques may also aid in the application of these techniques. An additional concern for

30
this methodology is the amount of fieldwork required to collect calibration data, and

possible future research can possibly extract such calibration data from low-flying drone

surveys.

4.3. Conclusion

We tested different methods of estimating grain size from UAV-based SfM

photogrammetry over several gravel bars along a large meandering river in Ohio. Statistical

grain size models were calibrated on three different image textural and topographic

roughness metrics that had been calculated from the resulting orthomosaic and point cloud.

This paper supports recent findings that image textural metrics may provide more accurate

grain size estimates in comparison to topographic roughness metrics when used in settings

characterized by smaller grains; however, further work to clearly define the ranges where

these metrics perform optimally still needs to be done.

31
Bibliography

Acharya, B. S., Bhandari, M., Bandini, F., Pizarro, A., Perks, M., Joshi, D. R., Wang, S.,
Dogwiler, T., Ray, R. L., Kharel, G., & Sharma, S. (2021). Unmanned aerial
vehicles in hydrology and water management: Applications, challenges, and
perspectives. Water Resources Research, 57(11).
https://doi.org/10.1029/2021WR029925
Benda, L., Andras, K., Miller, D., & Bigelow, P. (2004). Confluence effects in rivers:
interactions of basin scale, network geometry, and disturbance regimes: effects of
confluences on river morphology. Water Resources Research, 40(5).
https://doi.org/10.1029/2003WR002583
Brasington, J., Vericat, D., & Rychkov, I. (2012). Modeling river bed morphology,
roughness, and surface sedimentology using high resolution terrestrial laser
scanning. Water Resources Research, 48(11), Article 11.
https://doi.org/10.1029/2012WR012223
Carbonneau, P. E., Lane, S. N., & Bergeron, N. E. (2004). Catchment-scale mapping of
surface grain size in gravel bed rivers using airborne digital imagery: mapping
grain size in gravel bed rivers. Water Resources Research, 40(7), Article 7.
https://doi.org/10.1029/2003WR002759
Carrivick, J. L., & Smith, M. W. (2019). Fluvial and aquatic applications of Structure
from Motion photogrammetry and unmanned aerial vehicle/drone technology.
Wiley Interdisciplinary Reviews: Water, 6(1), e1328.
https://doi.org/10.1002/wat2.1328
Casado, M. R., Woodget, A., Gonzalez, R. B., Maddock, I., & Leinster, P. (2020).
Unmanned Aerial Vehicles for Riverine Environments. In Unmanned Aerial
Remote Sensing (pp. 55-75). CRC Press.
Chardon, V., Schmitt, L., Piégay, H., & Lague, D. (2020). Use of terrestrial photosieving
and airborne topographic LiDAR to assess bed grain size in large rivers: a study
on the Rhine River. Earth Surface Processes and Landforms, 45(10), 2314–2330.
https://doi.org/10.1002/esp.4882
Church, M. (2002). Geomorphic thresholds in riverine landscapes: geomorphic
thresholds. Freshwater Biology, 47(4), 541–557. https://doi.org/10.1046/j.1365-
2427.2002.00919.x
CloudCompare (version 2.12) [GPL software]. (2022). Retrieved from. http://www.
cloudcompare.org/.
Dade, W. B., & Friend, P. F. (1998). Grain‐size, sediment‐transport regime, and channel
slope in alluvial rivers. The Journal of Geology, 106(6), 661–676.
https://doi.org/10.1086/516052
Danhoff, B. M., & Huckins, C. J. (2020). Modelling submerged fluvial substrates with
structure‐from‐motion photogrammetry. River Research and Applications, 36(1),
128–137. https://doi.org/10.1002/rra.3532

32
Detert, M., Kadinski, L., & Weitbrecht, V. (2018). On the way to airborne gravelometry
based on 3D spatial data derived from images. International Journal of Sediment
Research, 33(1), 84–92. https://doi.org/10.1016/j.ijsrc.2018.02.001
Dugdale, S. J., Carbonneau, P. E., & Campbell, D. (2010). Aerial photosieving of
exposed gravel bars for the rapid calibration of airborne grain size maps. Earth
Surface Processes and Landforms: The Journal of the British Geomorphological
Research Group, 35(6), 627-639. https://doi.org/10.1002/esp.1936
Frissell, C. A., Liss, W. J., Warren, C. E., & Hurley, M. D. (1986). A hierarchical
framework for stream habitat classification: Viewing streams in a watershed
context. Environmental Management, 10(2), 199–214.
https://doi.org/10.1007/BF01867358
Graham, D. J., Rice, S. P., & Reid, I. (2005). A transferable method for the automated
grain sizing of river gravels: method for grain sizing river gravels. Water
Resources Research, 41(7), Article 7. https://doi.org/10.1029/2004WR003868
Groom, J., Bertin, S., & Friedrich, H. (2018). Assessing intra-bar variations in grain
roughness using close-range photogrammetry. Journal of Sedimentary Research,
88(5), 555–567. https://doi.org/10.2110/jsr.2018.30
Heritage, G. L., & Milan, D. J. (2009). Terrestrial Laser Scanning of grain roughness in a
gravel-bed river. Geomorphology, 113(1–2), 4–11.
https://doi.org/10.1016/j.geomorph.2009.03.021
Leopold, L. B. (1992). Sediment size that determines channel morphology. Dynamics of
gravel-bed rivers, 297-311.
Newson, M. D., & Newson, C. L. (2000). Geomorphology, ecology and river channel
habitat: mesoscale approaches to basin-scale challenges. Progress in Physical
Geography: Earth and Environment, 24(2), 195–217.
https://doi.org/10.1177/030913330002400203
Pearson, E., Smith, M. W., Klaar, M. J., & Brown, L. E. (2017). Can high resolution 3D
topographic surveys provide reliable grain size estimates in gravel bed rivers?
Geomorphology, 293, 143–155. https://doi.org/10.1016/j.geomorph.2017.05.015
Rhee, D. S., Kim, Y. D., Kang, B., & Kim, D. (2018). Applications of unmanned aerial
vehicles in fluvial remote sensing: an overview of recent achievements. KSCE
Journal of Civil Engineering, 22(2), 588–602. https://doi.org/10.1007/s12205-
017-1862-5
Rice, S., & Church, M. (1996). Sampling surficial fluvial gravels; the precision of size
distribution percentile sediments. Journal of Sedimentary Research, 66(3), 654–
665. https://doi.org/10.2110/jsr.66.654
Tamminga, A., Hugenholtz, C., Eaton, B., & Lapointe, M. (2015). Hyperspatial remote
sensing of channel reach morphology and hydraulic fish habitat using an
unmanned aerial vehicle (UAV): A first assessment in the context of river
research and management. River Research and Applications, 31(3), 379–391.
https://doi.org/10.1002/rra.2743
Tomsett, C., & Leyland, J. (2019). Remote sensing of river corridors: A review of current
trends and future directions. River Research and Applications, 35(7), 779–803.
https://doi.org/10.1002/rra.3479
33
Vázquez-Tarrío, D., Borgniet, L., Liébault, F., & Recking, A. (2017). Using UAS optical
imagery and SfM photogrammetry to characterize the surface grain size of gravel
bars in a braided river (Vénéon River, French Alps). Geomorphology, 285, 94–
105. https://doi.org/10.1016/j.geomorph.2017.01.039
Vélez-Nicolás, M., García-López, S., Barbero, L., Ruiz-Ortiz, V., & Sánchez-Bellón, Á.
(2021). Applications of unmanned aerial systems (UASs) in hydrology: a review.
Remote Sensing, 13(7), 1359. https://doi.org/10.3390/rs13071359
Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M., Reddy, T., Cournapeau, D.,
Burovski, E., Peterson, P., Weckesser, W., Bright, J., van der Walt, S. J., Brett,
M., Wilson, J., Millman, K. J., Mayorov, N., Nelson, A. R. J., Jones, E., Kern, R.,
Larson, E., … Vázquez-Baeza, Y. (2020). SciPy 1.0: Fundamental algorithms for
scientific computing in Python. Nature Methods, 17(3), 261–272.
https://doi.org/10.1038/s41592-019-0686-2
Wohl, E., Bledsoe, B. P., Jacobson, R. B., Poff, N. L., Rathburn, S. L., Walters, D. M., &
Wilcox, A. C. (2015). The natural sediment regime in rivers: Broadening the
foundation for ecosystem management. BioScience, 65(4), 358–371.
https://doi.org/10.1093/biosci/biv002
Wolman, M. G. (1954). A method of sampling coarse river-bed material. Transactions,
American Geophysical Union, 35(6), 951.
https://doi.org/10.1029/TR035i006p00951
Woodget, A. S., & Austrums, R. (2017). Subaerial gravel size measurement using
topographic data derived from a UAV-SfM approach. Earth Surface Processes
and Landforms, 42(9), 1434–1443. https://doi.org/10.1002/esp.4139
Woodget, A. S., Carbonneau, P. E., Visser, F., & Maddock, I. P. (2015). Quantifying
submerged fluvial topography using hyperspatial resolution UAS imagery and
structure from motion photogrammetry. Earth Surface Processes and Landforms,
40(1), 47–64. https://doi.org/10.1002/esp.3613
Woodget, A. S., Fyffe, C., & Carbonneau, P. E. (2018). From manned to unmanned
aircraft: Adapting airborne particle size mapping methodologies to the
characteristics of sUAS and SfM. Earth Surface Processes and Landforms, 43(4),
857–870. https://doi.org/10.1002/esp.4285

34

You might also like