Alex Lee - A New Dataset For Forest Height Across Australia: Pilot Project To Calibrate Icesat Laser Data With Airborne Lidar

Pilot project to calibrate ICESat satellite laser data with airborne LiDAR. Developing a new dataset for forest height and cover across Australia Alex Lee1, Peter Scarth2, Adam Gerrand3. 1. The Fenner School, Australian National University and DIGO 2. Queensland government SLATS team, QDNRW. 3. Bureau of Rural Sciences

Pilot Project Aims
To investigate the Australian use of ICESat satellite laser data for the National Forest Inventory, to improve forest sampling of height and cover, national reporting and monitoring.


Collaboration between the Queensland government Statewide Land and Tree Survey (SLATS) team, ANU, BRS, and CSIRO.


SLATS investigating ICESat data to help improve their Foliage Projective Cover (FPC) and vegetation change detection processes
CSIRO looking at ICESat data to improve continental DEM derived from satellite radar, with potential for vegetation height layer.

What is ICESat?


ICESat (Ice, Cloud, and land Elevation Satellite) is a spaceborne LIDAR platform, primarily used to measure ice change at the poles, using the Geoscience Laser Altimeter System (GLAS)
The laser transmits very short pulses of infrared light and visible green light
Photons reflected back to spacecraft from Earth and from the atmosphere, including the inside of clouds, collected in a 1 metre diameter telescope.
Laser pulses at 40 times per second will illuminate spots (footprints) approximately 70 meters in diameter, spaced at 170-meter intervals
Data is free to download from NASA ~ ICESat website: http://ICESat.gsfc.nasa.gov/)

What does the data look like?
GLAS received waveform (below, red) typical of returns from tree cover on flat ground
transmit pulse waveform is 7 ns (1 m) wide at half the maximum amplitude (black), and alternate threshold (dotted line),
alternate signal start & end (horizontal blue lines) and centroid (horizontal dashed blue line),
‘‘standard’’ Gaussian fit and centroid (black dashed line),
‘‘alternate’’ fits (cyan), & alternate model fit from sum of alternate Gaussians (thick blue line)
Received waveform & transmit pulse amplitudes are scaled separately
Tree cover shown is illustrative; it does not correspond to the location of the waveform.

Where is the data collected?


Transects approx 25km apart (vary 500m - 75km)




Data collected approx every 6 months since Jan 2003




Variation in lasers mean that shift in data collection (appears transects are “moving” west to east) -> 50-500m offset for subsequent collection periods

ICESat Calibration – NE Victoria


27 overlap locations




94 footprints within airborne LiDAR




Range of environments: • Floodplain forests • Foothills grazed woodlands • Montane tall forests




Project primary calibration site

ICESat Calibration – Injune, central Qld • 18 footprints within airborne LiDAR • Airborne LiDAR located within 150 500 x 150m sampling units. Study area is 220,000ha in total • Foothills grazed woodlands and open forests • dominated by Eucalypt (poplar box, ironbarks) and • Callitris species, with • Angophora, & Acacia also present (mainly brigalow). • Secondary calibration site.

ICESat Calibration – SE Queensland
Brisbane River Lidar – 20 footprints overlap airborne LiDAR at 5 locations




SEQ Private Native Forest project lidar – approx 11 overlap locations. Airborne LiDAR not yet checked for number of footprints

ICESat Calibration – Other potential locations Near Mt Isa

AAMHatch intersected ICESat footprints with their existing LiDAR holdings (for 2006)

Near Townsville

Near Rockhampton

Bundaberg Near Toowoomba

Additionally, SLATS team have ongoing LiDAR calibration sites throughout Queensland

Near Coober Pedy Near Grafton

Near Walgett

Near Taree Near Dubbo Near Mildura Near Hay

Wimmera Mallee Pipeline project

Near Kingston SE Near Lismore

Near Bunyip

ICESat attribute extraction
For accurate vegetation comparison, the ICESat footprint shape and size needs to be accurately portrayed within the airborne LiDAR. Parameters for this have been extracted.
Different lasers on satellite used over time – changes footprint shape

Selecting LiDAR in the ICESat footprint
The ICESat data point is queried for the centre location coordinates, diameter, azimuth and shape (‘eccentricity’ = circle → ellipse)
These parameters are input into an elliptical formula, where each LiDAR return is checked to see if it occurs within the footprint area

Extracting physical attributes from LiDAR
A range of physical attributes were extracted from LiDAR data: • Slope – absolute ( from max – min elevation in footprint), and mean slope from 20 x 20m cells. • Mean elevation across footprint • Elevation range (max veg – min ground) • Max tree height (m) • Predominant tree height (m) across 10 x 10m cells • Foliage cover (% returns) @ 0.5 and 2 m height • Crown cover (%) (from crown delineation results)

Extracting Height Information
3 potential vegetation height parameters:
centroid_height - distance from centre of ground pulse to centre of highest veg pulse
fit_height - third parameter in the weibull distribution used to fit the cumulative vegetation profile - ( p[1] *exp (-p[2] * ( height x / p[3] )p[4] ))
veg_height - Height where the cumulative FPC greater than 2m crosses 95%

Ground Elevation Comparisons – NE Victoria
NE Victoria - mean elevation difference 1.67 m (stdev = 3.12 m, range –3.04 → 12.66 m)
Many locations had ICESat with generally higher ground elevation, possibly due to: • denser canopies, • greater understorey presence, and • effect of slope in steeper terrain. • Definite ecozone / elevation trend


Those that recorded lower ICESat elevations were generally in riparian zones, where the 70-100m footprint could be influenced by terrain lower down river banks

Ground Elevation Comparisons –Qld


Injune - mean elevation difference was 0.16 m (sd = 1.31 m, range -2.0 → 2.73 m).




Close match to airborne lidar elevation possibly due to: • open canopies found in semi-arid environment, • generally flat terrain Indicates that ground elevation results may be good for most of Australia







Brisbane River mean elevation difference of 1.71 m (sd = 4.52 m, range -18.86 → 4.59 m).




ICESat value was mostly lower than LiDAR value,




Possibly larger footprint area recording range of elevation across stream bank slope, rather than at the footprint centre point higher up bank




The large 19 m difference observed at one site could be the result of the ICESat footprint overlapping a riverbank cliff

Ground Elevation Comparisons by Ecozone (NE Victoria)

Validation of ICESat attributes with LiDAR
Best forest height comparisons were for max elevation range and predominant height


Crown and Foliage Cover comparisons rather poor

Quality Assessment for ICESat result by Ecozone Good = within 10% or 5m of LiDAR value, Poor = >20% or 10m of LiDAR
Foliage cover Good – 33% Marg. – 26% Poor – 41%


Mean canopy height Good – 54% Marg. – 18% Poor – 28%

Vegetation Case Study

Vegetation Case Study Results

Vegetation Case Study Results
Higher tree cover + lower slope = improved ICESat veg structure extraction

Midslope

Riparian Strip Ridgetop

ICESat Version 26 and 28
Version 26 – Aug 2006
Version 28 – Dec 2006
Version 26 • ~ 1.9 million footprints • FPC, 3 potential veg heights • Range of locations • Older processing algorithms


Version 28 • • • •

~ 2.6 million footprints Different FPC, 3 veg hts Less transects Newer algorithms, more consistently applied (older data reprocessed)

• 2008 update ~ 4 million footprints across Aust

Version 26 & 28 continental results
Issues with Version 26 vs NFI data:

Version 26 continental summary

• Only 50% non-forest, (NFI ~ 80%) • Too much open forest (by ~ 30%) • Too much low forest (by ~25%)


Version 28 improvements: • Non forest within ~5% • All height classes within 5% wrt NFI • Open and closed forest within 6% • Woodland still underestimated

Version 28 continental summary

Enhanced national forest reporting
NFI could report forest structure distributions as well as class summaries
Provides improved assessment and monitoring of more subtle changes in height and cover
Improved calibration of other remotely sensed data when using continuous, rather than categorical, data

Conclusions
ICESat data has been successfully extracted
Compared to airborne Lidar for 94 points at 3 main sites
Strong agreement in ground elevation with airborne LiDAR
Weaker but still reasonable relationship for forest height and cover in some cases: • Better for flatter terrain, with more open & shorter forests • Less reliable for taller dense forests on steeper terrain

Conclusions cont…
Potential for forest vertical structure – but further calibration required for consistent extraction across environments
Currently over ~4 million footprints across Australia, great sampling tool with preliminary results showing promise compared to existing NFI data
National collaboration encouraged to share information • build a set of shared forest height and inventory data that can be used to calibrate, validate and develop modelling techniques