Integrated Sampling Scheme Project
Overview
Objectives ~ Methods ~Results ~ Further Information
Principle Investigators
Philip Tickle (formerly BRS & CRCGA, now GeoScience
Australia)
Alex Lee (formerly BRS, now PhD candidate, ANU)
Christian Witte (QDNRME)
Jenet Austin (formerly BRS, now CSIRO)
Kerstin Jones (QDNRME)
Richard Lucas (UWA / UNSW)
Introduction
The Integrated Sampling Scheme project was initiated in 1999 as part of the CRC Greenhouse Accounting (CRCGA) Program A – Accounting for Carbon. Within this program the Biomass carbon section (Project A2) had the main objective of improving the measurement and prediction of spatial and temporal variations in woody biomass carbon storage
The objective of the sub-project (A2.4) – Remote Sensing of Forest and Woodland Biomass – was to
“Develop
and evaluate the potential of remote sensing technologies for providing spatial
estimates of above-ground carbon at project and regional scales”
This CRC sub-project was initiated because transparent and verifiable biomass estimation methods are needed at site, project, regional and national scales, for national carbon accounting and trading. Obtaining reliable field-based measurements is very labour intensive and costly, so new techniques need to be developed to estimate above ground biomass using remote sensing technology. This project was established to investigate how new and existing remote sensing technologies can be integrated with field assessments of biomass to produce the information needed to estimate carbon in forests (which is contained in biomass) at a wide range of scales.
The primary project partners, collaborating under the CRCGA, were:
v Bureau of Rural Sciences
v Queensland Dept Natural Resources & Mines
v Queensland University of Technology
v Collaboration with ARC SPIRT (radar project) affiliated agencies included University of New South Wales, and the University of Wales, Aberystwyth (UK).
With the submission of a comprehensive
summary paper to the Remote Sensing of Environment journal in early 2004, this
project is now substantially completed.
Objectives
A primary aim of the work is to reliably estimate biomass at very fine scales for sample-based assessment and monitoring, as well as calibration of other larger scale wall-to-wall mapping sensors.
The objectives of this project are:
1.
Reduce
the dependency on field-based gathering of information about forests and
woodlands. Field data is still a vital
part of a sampling system, but by reducing the need to gather lots of field
data we can save money and time.
2.
Develop
robust links between above ground (eg trees and shrubs) measurements of biomass
from remotely sensed data, and below ground (eg roots) measurements, as well as
growth models, so a more complete understanding of all biomass carbon is
gained.
3.
Develop
an understanding of the limitations of remotely sensed data.
4.
Provide
a framework for bringing together into one system measurement, validation and
auditing / monitoring at all scales
5.
Use
ultra-high resolution airborne camera and Lidar systems to intensively sample
forests and to test, check and improve the data gathered from other more
coarser resolution remote sensing systems (eg. Radar, Landsat TM)
6.
Develop
routine, repeatable and economical procedures for estimating biomass at site,
project, regional and national scales using a range of remote sensing
platforms, such as LIDAR, Large Scale Photography (LSP), Radar, and Landsat TM,
among others.
Remote Sensing and Field Data
Introduction
We are using a range of data sources to gather information on forests and woodlands at a range of scales, from the individual tree level (using field data) through to broad regional assessments across hundreds of kilometres using remotely sensed data. By using remote sensing we are able to gather much of the same information that we can in the field, but over a much wider area, more consistently over time, and for a much reduced cost. The use of an integrated sampling strategy allows us to extrapolate findings over a larger area with confidence. The range of data sources are shown below.
Organisation of remote sensing data acquisition
With the knowledge that AIRSAR radar data were to be acquired over the Injune study area, a sampling framework for the collection of finer spatial resolution remote sensing data was designed and implemented as full coverage was not achievable given time, distance and cost constraints. The sampling scheme also facilitated comparison of estimates generated from using wall-to-wall mapping undertaken as part of other studies (QDNR, 2000) and also to provide operational experience in the implementation of sampling frameworks that may be adopted in future regional and national inventory programs (e.g., the National Forest Inventory).
A systematic sampling scheme was selected, as knowledge of the floristics, structure and biomass of the forests and woodlands was too limited to allow application of efficient stratified sampling methods. Furthermore, the state of the forests and woodlands had changed rapidly over recent years, largely because of extensive clearance of vegetation within the area, thereby preventing the use of historical spatial layers for stratification.
Based on these
considerations, the systematic sampling scheme for the 37 x 60 km study area
allowed the acquisition of Large Scale Photography (LSP) pairs across a grid containing 150 (10
columns and 15 rows) points located 3.7
x 4 km apart in the east-west and north-south directions respectively (Figure
1). Within each 800 x 800 m (64 ha)
area represented by the LSP pairs (herein referred to as Primary Photo
Plots or PPPs), a Primary Sampling Unit
(PSU) 150 x 500m (7.5 ha) was located within the 60 % stereo photo
overlap. Each of the 150 PSUs
was then subdivided into 30 systematically numbered Secondary Sampling Units
(SSU) which were 50 x 50 m (0.25 ha) in area.
Using this scheme, data could be analysed and summarised for each of the
150 PPPs and PSUs (4500 SSUs) that represented 5.3 % (3.9% for only the stereo
area), and 0.5 % of the 222,000ha study area respectively.
Figure 1: Layout of the PPP, PSU and SSU grid.
Click here for a larger version of Figure 1.
Stereo aerial photography
For each of the 150 PPPs, and using pre-defined
coordinates, large scale (1:4000) stereo colour aerial photographs (in negative
format) were acquired on the 11th July, 2000 by QASCO Surveys Pty.
Ltd on behalf of the Queensland Department of Natural Resources Landcare
Centre. Photographs were taken using an RC20 large format photographic camera
from late morning to mid afternoon. The
effective swath width was 920 m and, for each photo principle point, GPS
coordinates were recorded to within a nominal precision of ± 20 m absolute
location. As 150 PSUs were sampled, 300
frames of photographs were obtained. The primary objective
of acquiring the aerial photography was to facilitate the subsequent planning
of field data collection. However, interpretation of these data also allowed
the floristic composition and structure (e.g., canopy cover) of the forests and
disturbance across the Injune study area to be sampled and described.
In 2003 an
assessment of historical air photography was undertaken by BRS in conjunction
with QDNRME. For the study area it was
found that there was a range of photography (at a range of scales) dating from
the 1960’s through to 1995, for many of the field sites, within the Queensland
government archives. Those that were
available were scanned at high resolution and added to the database. A full listing of this data can be found on
the (Data
Table Page), and on the (Timeline page). Further research found that the first
photography over the area was in 1948, and we are currently in the process of
acquiring these images, to enable a longer time series analysis of vegetation
change.
LiDAR data
Airborne scanning LIDAR data were captured over a
one-week period commencing August 24th 2000 using an Optech 1020
scanning LIDAR mounted in a Bell Jet Ranger helicopter. The Optech 1020 measured
5,000 first and last returns and the intensity of each return per second. The
LIDAR operated within the NIR spectrum with a beam divergence of 0.3
milliradians, a footprint of approximately 9 cm and an average sampling
interval of < 1 m. Data were acquired flying in an east-west direction (and
centred on each PSU row), at a nominal altitude of 250m and a swathe width of
approximately 200 m. A GPS base station was established for all flights. With
full differential GPS corrections, in addition to pitch, yaw and roll
compensation from an Inertial Navigation System (INS), coordinates were
guaranteed to an absolute accuracy of < 1m x and y directions and < 0.15
m in the z direction. These data were
acquired to obtain information on the height, crown area and vertical structure
of individual trees and forest stands as well as ground surface digital
elevation models (DEMs) and as a mechanism for scaling-up biomass to the
landscape (lidar biomass
page link)
Airborne hyperspectral data
Hyperspectral CASI (28th August, 2000) data (1 m spatial resolution; 14 wavebands; see Figure 2) and HyMap (3rd September, 2000) data (2.5 m spatial resolution, 120 wavebands, for a subsection of the sampling units) were acquired initially to support the upscaling of biomass and structural attributes to the landscape as differences in the reflectance of tree species and communities allows their discrimination. For the main tree species, foliar chemistry (photosynthetic pigments) and moisture content (through destructive harvesting) were recorded and leaf sections (cell structure) taken. Field based spectral information of leaves associated with the main species were also collected to facilitate a better understanding of the information content of hyperspectral data.
Figure 2.
Hyperspectral HYMAP data with CASI data inset.
Field data
Field data were collected between July 14th and
August 3rd from 12 of the 150 sample units , and included:
v
Within
31 50 x 50 m secondary sampling units (SSUs; up to 4 per subunit), the
locations of all trees > 5 cm diameter (at 130 cm) and their diameter (at 30
and 130 cm), height and crown dimensions.
Each tree was identified to species. A further 3 10x10m plots were
established in recently cleared areas, that had subsequent regrowth.
v
Within
10 x 10 m subplots, an assessment of understorey (< 5 cm) vegetation and
structure.
v
Digital
photographs of at least every 10th tree measured (for some SSUs, all
trees were photographed).
v
Foliage
projected cover observations acquired at 1 m intervals along three 50 m
transect lines and canopy photographs taken at 5 m intervals along the same
transect lines.
v
Soil
dielectric constant and moisture content, measured for each SSU through a
combination of Time Domain Reflectometry (TDR) and gravimetric methods.
v
For
the majority of trees harvested, spectral reflectance measured using an ASD
Fieldspec radiometer, of leaves extracted from different heights and aspects
within the canopy, thus providing comparative reflectance curves for species
discrimination purposes.
Destructive harvesting
In the weeks following data collection, destructive harvesting of Callitris glaucophylla (22 individuals) was undertaken in conjunction with the QDPI Tropical Beef Centre, primarily to generate new allometric equations for this species such that the biomass of components (i.e., leaves, branches and trunks) could be estimated from tree size (e.g., diameter) measurements. Harvesting of Eucalyptus melanaphloia (Silver-leaved ironbark; 5 individuals) and E. populnea (Poplar Box; 7 individuals) was undertaken to establish the validity of existing equations generated elsewhere in Queensland (Burrows et al., 1998).
Additional sampling for total above ground biomass estimation was undertaken for several other key species, including Angophora leiocarpa. All trees harvested were divided into the major components (leaves, branches and trunks) and also subcomponents (i.e., branches < 1 cm, 1-4 cm etc.). These equations were applied subsequently to each tree recorded within the SSU to provide an estimate of component and subcomponent biomass per hectare (Mg ha-1). For the main species, leaf thickness was also measured using callipers and leaves were photographed against a reference scale.
Topographic SAR (TOPSAR)
On 2st September, 2000, a 12 x 60 km strip of TOPSAR data was acquired for the western section of the study area. Full coverage was not provided because of high winds at altitude. The data were acquired to evaluate their potential for retrieving stands heights.
Polarimetric SAR (POLSAR)
On 3rd September, 2000, the AIRSAR POLSAR acquired four strips (~ 12 x 80 km) of
fully polarimetric multifrequency SAR (POLSAR) C- (~5.6 cm wavelength), L- (~ 24
cm wavelength) and P- (~ 68 cm wavelength) band data across the entire PPP
grid. The incidence angle at which the PSUs were observed varied from 29 to 59o. POLSAR data were acquired to evaluate their
potential for retrieving forest biomass and structural diversity and mapping
across the landscape.
Figure 2. P-band (~ 68 cm
wavelength), L-band (~ 25 cm) and C-band (~6 cm) data of the western section of
the Injune study area
MODIS/ASTER Simulator (MASTER)
On 3rd September, 2000, MASTER data (50 bands, visible to thermal) were acquired at the same time as the POLSAR and for the entire study area. These data were acquired to support the mapping of species and communities across the study area.
Landsat TM
Landsat imagery provides a regional scale assessment for land cover mapping. With a resolution of 25m x 25m for each pixel, broad vegetation and cover types can be identified. Landsat imagery can be calibrated using the finer scale remotely sensed data so that very large areas can be assessed accurately. Our study area is 37km x 60km, but the full Landsat scene covers 185km x 170km. Using many images collected in different years, we can track changes in the land cover (such as vegetation clearing for agriculture, or vegetation regrowth from abandoned farmland). We have been using Landsat imagery from 1972 through to 2000 to get an idea of the broad scale changes that have happened at the field site in Queensland. See here for further information (Land Cover Change page).
Landsat Hyperion
Hyperion data (20 m spatial resolution, > 200 wavebands) were acquired in January and March, 2001, diagonally across the study area. These data were acquired to better understanding the information content of hyperspectral data at the landscape rather than tree level and allowed a better understanding of the advantages (or otherwise) of these data over Landsat sensor data.
A full description of the data used in this project can be found here (data table page)
Results
LSP
Following delivery of the photographs, the main vegetation types observed within the overlap area of each of the 150 stereo pairs were delineated manually and described in terms of their species composition, height, cover and disturbance. The community composition of the woodlands was then established by summarising the occurrence of dominant and co-dominant species and their associations within each of the 4500 SSUs.
Analysis of the LSP established that open forests and woodlands occurred in 90 % of the PSUs, with the remainder containing non-forest (generally actively-used pasture). Although diverse, higher biomass (> 80-100 Mg ha-1) woodlands were generally dominated or co-dominated by the coniferous species White Cypress Pine (Callitris glaucophylla; coded CP-), Silver-leaved ironbark (E. melanaphloia; SLI) and Smooth-barked apple (A. leiocarpa; SBA), whilst those of lower biomass were dominated by Poplar Box (E. populnea; PBX) and other Eucalyptus species (EUS). Most areas of regrowth were dominated by Brigalow (Acacia harpophylla; BGL).
Further information
Alex Lee
PhD Scholar
School of Resources, Environment & Society
Australian National University
Canberra
ACT 0200,
Australia
Email: alex.lee@anu.edu.au
Tel: +61 (2) 6125 0348
Fax:+61 (2) 6125 0746
A number of
conference papers have been presented as part of this research, and a
international journal article is currently under review. Links to these can be found below.
Journal Papers & Conferences
A list of journal paper and conference presentations can
be found in the Publications page.
