Assessing soil erosion with unmanned aeria vehicles
1. ASSESSING SOIL EROSION
WITH UNMANNED AERIAL
VEHICLES FOR PRECISION
CONSERVATION
Joby Czarnecki, Anna
Linhoss, Lee Hathcock,
John Ramirez Avila, and
Timothy Schauwecker
Mississippi State University
2. THE STORY OF CATALPA CREEK
Catalpa Creek is the main
drainage channel for
Mississippi State University,
which is located at the
headwaters
Catalpa Creek flows through
two of MSU’s research facilities
The Watershed is impaired for
sediments, pathogens, and
nutrients
Faculty and state agency
personnel successfully applied
for 319(h) status for the
3. IN OTHER WORDS….
Our University has world class researchers working in conservation,
but our research facilities are disconnected from their expertise
4. RESEARCH ACTIVITIES WITHIN THE
WATERSHED
Surveying
Vegetation survey
Cross-sectional survey
Monitoring
Water quality
Flow rates
Modeling
Hydraulic modeling
Hydrologic modeling
Landscape evolution
modeling
Designing
New best management
practices
New methodologies for
research
5. STUDENT ENGAGEMENT WITHIN
THE WATERSHED
In 2 years of the project:
37 undergraduate and 7 graduate students involved in research
Students from 9 academic departments in 4 colleges and 2 universities,
engaged with faculty members from 5 academic departments in 3
colleges and 1 research center
60 students utilizing the watershed for class projects
6. STRUCTURE FROM MOTION
Structure from Motion (SfM) creates 3D
structure from a combination of multiple
viewing angles
Structure from Motion is called "image
parallax" in the remote sensing parlance
Unmanned aerial vehicles (UAV) are flown
with high levels of overlap between
successive flight lines to create a series of
stereopairs
Parallax - Displacement or difference in the apparent position of an object viewed
along two different lines of sight and is measured by the angle or semi-angle of
inclination between those two-lines.
7. SFM VERSUS LIDAR
SFM
Passive remote sensing process
Can be performed with RGB
cameras on low-cost UAVs
Does not penetrate vegetation
well, if at all
Performs poorly on uniform
scenes
LIDAR
Active remote sensing process
Requires a LIDAR sensor, as well
as additional hardware for the
UAV
With enough returns, can
penetrate some vegetation
Can be used on uniform scenes
8. APPLICATION OF SFM TO SOIL
EROSION
1. Determine the limitations to
the technology and the
accuracy obtainable
2. Assess usefulness in popular
models
3. Determine the suitability of
this data for decision making
by water resource managers
4. Produce a method and best
practices that end users could
adopt fairly easily, affordably,
and quickly
9. MATERIALS AND METHODS
Materials
1. COTS UAV systems from DJI
2. AgiSoft image processing
software
3. DroneDeploy and Pix4DModel
cloud-based image processing
4. Ground control points and
calibration structures
5. Survey grade GPS unit
6. Traditional survey equipment
Methods
1. Conduct UAV mission with
high overlap between flight
lines
2. Process image data with
desktop and cloud-based
software
3. Collect ground truth data
4. Evaluate agreement in
elevation values
• Between SfM surfaces produced with
varying techniques
• Between SfM surfaces and ground truth
11. LIMITATIONS AND LESSONS LEARNED –
GROUND TRUTH
Increased size of dots coincides with decreased
precision of GPS instrument (as reported by the
instrument under vertical dilution of precision). Small
dots with blue, pink, or red are acceptable.
12. LIMITATIONS AND LESSONS
LEARNED – FLIGHT
Generally ± 1m difference on average between elevations for surfaces processed with short
paths or long paths only. Disagreement between surfaces highest in areas with vegetation
present, with brown indicating higher elevation in short paths and blue representing higher
elevation in the long path.
Long -
Short
13. LIMITATIONS AND LESSONS LEARNED
– PROCESSING
Yellow areas indicate negligible differences in elevation. Brown (lower density values dominate) and blue (higher
density values dominate) areas indicate disagreement of up to 3 meters between resolutions indicated in each
image.
Ultra -
High
High -
Med
Ultra -
Med
14. LIMITATIONS AND LESSONS LEARNED
– SOFTWARE
Proprietary software
Pros:
Complete control over processing
options
Unlimited images
Relatively inexpensive for educational
license
Cons:
Learning curve
Computationally intensive
Hard to share outputs
Verdict:
More control, but more taxing
Necessary for research applications
Cloud-based service
Pros:
Data turnaround in < 24 hours
Sharable weblink
Cons:
Accuracy?
Monthly fee (albeit low)
Limited to 1000 images per model
Desktop install for analysis
Elevated license required for
downloading DEM
Verdict:
Good for simple site characterization
An easy option with reasonable visual
output
*DroneDeploy,
Pix4DModel
15. CONCLUDING REMARKS
Promising findings:
This project demonstrated UAV images are capable of providing spatially-
explicit, fine- to medium-temporal scale data.
When the need for high accuracy is secondary to identification of eroded
areas and periodic review of landscape change in response to practice
placement, UAV-based SfM surfaces represent a low-cost, rapid turn-
around solution.
With a cloud-based, paid processing service, surfaces were generated
within 24 hours. The service was simple to use and provided a sharable
weblink to a 3D model.
Cautions:
Every choice matters and will affect accuracy.
In some cases, the proprietary software that took more than a week to
produce a SfM surface, which could be a consideration for high-temporal
16. FUTURE DIRECTIONS
Explore glacier research for dealing with undercutting
Incorporation of tracers
Vegetative effects by type and percent cover
17. FUNDING SOURCES AND CONTACT
INFORMATION
Structure from Motion research
in Catalpa Creek is supported by
funding from:
National Institute of Food and
Agriculture, USDA, Hatch funds
The Mississippi Agricultural and
Forestry Experiment Station
Mississippi Water Resources Research
Institute, under USGS 104B
Joby Czarnecki
Assistant Research Professor
Geosystems Research Institute
Joby.Czarnecki@msstate.edu
662-325-5972
Editor's Notes
Civil, LA, ABE, Geosci, Forestry, WFA, Chem, Ag Econ, Systems
Ground control points (GCP) are a NECESSITY
UAV GPS accuracy is relatively low
Many say 20 is a standard for number of ground control points
Placement of control points is important
Highly erodible areas
Surface warping
Increased erosion potential
Landscape paint is easy to pick out in vegetative indices
Mowing and weedeating issues
Compounding effect of error due to horizontal precision is problematic
Vegetation is problematic for year-around monitoring for maintaining cross-sectional accuracy and also for change detection
Mismatches between ground truth and SfM surfaces create dilemmas about accuracy of both
These results are invariate of re-scaling. It’s simply an issue of interpolation in the areas we have poor coverage. Processing options matter, especially when vegetation is present. These maps indicate the difference between ultra high, high, and medium cloud density in processing. Raster calculator also assigns the lower resolution to the output, but resolution is not the driver for these differences.