application of multi-source uav data to assess...
TRANSCRIPT
Application of multi-source UAV data to assess revegetation efforts on waste rock
Tim Whiteside, Renée Bartolo & James BoydenEnvironmental Research Institute of the Supervising Scientist
Outline of this talk• Background to the Trial
Landform at Ranger• Project aim• Data capture and analysis• Some preliminary results• Concluding comments &• What’s next
5 Jan 12
3
Location• Northern Australia
• Mining lease established before declaration of Kakadu National Park
• Surrounded by Kakadu National Park
• Tropical monsoonal climate with distinct wet & dry seasons
• ~1550 mm annual rainfall
Ranger Uranium Mine• Scheduled for closure
in 2026
• Final landform will be mostly waste rock
• Established a trial landform to understand waste rock behaviour
• Information about:
– Material transport
– Hydrology
– Vegetation
• Helps inform closure criteria, calibrate models etc.
Construction
Mixing
■ Construction 2008-2009■ Built in layers■ Waste rock & lateritic
material
■ Mixed using an excavator■ Levelled with large
bulldozer■ Estimated cost of $10
million
7
Trial Landform (TLF)
1 2 3 4
Plots 1&2 Waste rock only
Plots 3&4 Waste rock &laterite mix
Vegetation• Tubestock planted on EP1
& EP4 - March 2009
• Direct seeding occurred on EP2 & EP3 -July 2009
• Some infill planting -tubestock on EP1 & EP4 –Jan 2010
• Direct seeding had poor germination, infill planted with tubestock Jan 2011
• Relied in natural recruitment since
Need images of the vegetation Perhaps before and after
6 May 12
10 Dec 09
TLF vegetation now• On site for 8-9 years
• Tree stem diameters > 5 cm
• No understorey yet on plots 1&2
• Due to substrate plots 3&4 has understorey (mostly weeds) and is managed
To assess ongoing viability of an established revegetation effort on waste rock
Resilience – fire, seasonality, substrateSpecies composition
Using UAS data – major benefit is data frequency and full site coverage
Primary aim of this project
Bungee launch
Skycam UAV - Swampfox
Parachute recovery
3DRobotics – X8+
Gryphon Dynamics – X8 1400
Year Date Fixed Multi RedEdge HS LiDAR RGB
2016 17 May
28 July
14 October
16 November
2017 18 January
26 April
18 May
Data captures to date - TLFFixed wing flights at 100 m altMulti rotor MS and HS flights at 70 mLidar at 30 m
• RGB and multispectral data processed using Pix4D Mapper
• MS radiometric calibration to reference panel
• Up to 8 GCPs used – DGPS recorded
• All RGB and MS data co-registered to May 2016 data
Data processing
Multispectral time series
Controlled burn18-19 May 2016
Multispectral time series
Multispectral time series
Multispectral time series
Multispectral time series
Multispectral time series
Object-based analysis of cover pre- and post-burn
• Multiresolution segmentation
• Random Forest classifier
• Accuracy assessment
• RGB data used for segmention
• Based on mainly on multispectral data
• Point based from mosaic
Object-based approach
Features used
Random Forest parameters
AccuraciesTrees vs non-trees92% accuracy
Weeds vs non-weeds85% accuracy200 stratified random points per class
AccuraciesTrees vs non-trees91% accuracyWeeds vs non-weeds95%200 stratified random points per class
Building up a field reference dataset
Conclusion
• UAV data enables high temporal analysis of rehabilitation sites at the plant level.
• Enables the monitoring of seasonal effects (eg phenology, water stress), plant fate and fire effects for the whole of a site.
So far:
• Strengths
‒ Coverage for whole of site ~ 30 min
‒ Cost effective multi-temporal data set
• Challenges
– Remote locality
– Terrain
– Erectophile leaves
– Sparse canopies
– Wind and thin stems
Strengths and challenges
Further work• Continued data collection including hyperspectral
and LiDAR data
• Using the multi-temporal UAV data to describe:
‒ Further analysis of structural, spectral and textural information to discriminate tree and ground cover
‒ Plant object fate analysis as site matures further
‒ Surface behaviour – including particle movement
Thank you.
Any questions?