School of ComputingFaculty of Engineering
DART workshop: Airborne remote sensing
David Stott
Overview
● This presentation covers the airborne side of DART:– LiDAR– Spectroscopy (imaging and field)– Aerial photography
● About cropmarks on arable land:– The nature of contrasts in vegetation marks– How we can use this to improve detection
Aims and Objectives
● Look at contrasts over time:– How they change with weather– How they change with land-use
● How to best detect contrasts with different sensors:– What can the sensors detect?– What is the best context to deploy them in?
Archaeological vegetation marks
● Soil differences influence the development & health of the crop
● Visible as local variation in the crop canopy:– Stress and vigour:
● Variations in foliar chemistry● Extreme condtions
– Canopy structure:● Leaf Area Index (LAI) ● Tillering / early growth stage development
Some challenges
● High dimesnionality of data:– Hyperspectral can have 100s of bands– Full wave-form LiDAR– Lots of redundancy
● No unique spectral signature
Some challenges
● High dimesnionality of data:– Hyperspectral can have 100s of bands– Lots of redundancy – Full wave-form LiDAR
● No unique spectral signature– Brute force / classifcation approaches are
problematic– Changes in soils, land use and crop
Methodology: Ground-based
● Multi-temporal ground measurements (monthly):– Spectro-radiometry
● ASD FieldSpec Pro● 350-2500nm @ 3nm (ish) sampling interval
Fiber-optic probe
Tired old laptop (needs an LPT port…)
Reflectance panel
Instrument(20Kg of backpain)
Methodology: Ground-based
● Multi-temporal ground measurements (monthly):– Spectro-radiometry
● ASD FieldSpec Pro● 350-2500nm @ 3nm(ish) sampling interval
● Crop height & density:– Ceptometer (leaf area index)– Surface coverage (near-vertical photos)– Tillering in early growth-stage
Airborne data
● NERC ARSF:
– Eagle and Hawk hyperspectral (VIS-SWIR)
– Full waveform LiDAR
– Survey camera
● Geomatics Group:– CASI hyperspectral
– Discrete LiDAR
– Orthophotography
Methodology: Analyses
● Python software:– Spectral analysis (imagery & spectroradiometry):
● Continuum removal● Vegetation analyses● Red edge position
– LiDAR:● Multi-temporal vegetation mass● Full-waveform
Jun 14th 2011
Jun 29th 2011
Jul 15th 2011
Spectral differences
● Example from Diddington June-July 2011– Spectradiometry shows
good contrast– Continuum removed
spectra from 670nm absorption feature
– Band normalised by area
So...
● Spectroradiometry shows good contrast:– Variations in foliar pigmentation change rapidly– Variations in crop structure remain fairly similar
● Can we use the LiDAR to detect the biomass variations?– Higher spatial resolution (~0.4m vs 1m)
Full waveform
● Looked at correlation between hyperspectral and full waveform LiDAR– Reflectance @ 1062nm (Hawk) & intensity @
1064nm (LiDAR)
Full waveform
● Correlation between archaeological features and full waveform LiDAR
Dataset t p
Vegetation height 42.9721 2.2E-016
Peak sum 12.968 8.56E-014
Maximum intensity 7.9123 1.327E-015
Peak width 0.4164 0.3385
Full waveform
● Sensor only resolves a single return over low, sparse crop
● Very little variation in the width of the return● Intensity is usable● Best results came from using vegetation height
model derived from discrete returns
Conclusions so far
● Different sensors and techniques required on a field by field basis
● This is hard:– Variability of the archaeology– Variablility of its context– Small things in big data– (not even mentioning the weather...)
● Spatial resolution is not the be all & end-all
Further work
● LiDAR:– Scan angle– Spatial analyses
● Statistical comparison of sensors– Comparison of contrasts on and off the features
● Writing it all up
Acknowledgements
● NERC ARSF● Royal Agricultural College● Thornhill Estates● DART community