Heard Island - Remote Sensing project UTAS & Australian Antarctic Division (AAD)
*Note: the content of this page is based on an article that was published in the Australian Antarctic Magazine
Changing environment
Research on Heard Island has shown that local climatic conditions are continuing to change. These changes, in turn, are having an effect on the island’s environment. Between 2000 and 2004, for example, Brown Glacier, on the island’s east coast, lost 8 million cubic metres of ice a year, compared to the 50 year average of 3 million cubic metres a year (Australian Antarctic Magazine 7: 9, 2004). In some coastal areas that were previously ice-covered, there are now large areas of bare ground and lagoons.
The Australian Antarctic Division manages the World Heritage listed Territory of Heard Island and McDonald Islands (HIMI) and the HIMI Marine Reserve. Part of this role includes monitoring and reporting on changes to the environment and conservation values, and pursuing necessary management actions. However, the cost and logistical challenges of getting to Heard Island preclude regular on-site monitoring. These practical constraints are acknowledged in the Heard Island and McDonald Islands Marine Reserve Management Plan, which promotes the development of practical, cost-effective and low-impact remote monitoring techniques.
Satellite imagery
Accordingly, the AAD has over the past year implemented a project to develop techniques to use satellite imagery to detect change in features on Heard Island. The project is led by AAD’s Policy Branch, which administers the HIMI Territory and Marine Reserve, and involves a team of scientific and technical experts from the Australian Antarctic Data Centre (AADC) and the University of Tasmania.
The process involves obtaining high resolution satellite images, such as from the WorldView-1 and QuickBird earth imaging satellites. These satellites detect the near-infrared light wavelengths reflected by vegetation, allowing different land cover types, such as bare rock and plant communities, to be identified. The satellites have a resolution of 50 cm and 60 cm, respectively, and can collect all the required information in one pass of the island.
DigitalGlobe WorldView-1 image of Heard Island acquired 23 March 2008
A WorldView-1 image of Heard Island (23 March 2008) was purchased by the AAD and UTAS in June 2008. The WorldView-1 imagery contains two separate image strips that cover the whole island. These strips were acquired at slightly different times from different angles during the satellite overpass. The discrepancy in acquisition angle has resulted in a geometric offset between the two image strips. A geometric rectification step is required to correct this offset and create a single image for the whole of the island that is geometrically accurate. Orthorectification was carried out with a 10 m Digital Elevation Model (DEM) acquired by RADARSAT in 2002 and the WorldView-1 RPC information. The two image strips were merged to form one image mosaic. A resampled version of the image at 5 m resolution can be download in KMZ format for viewing in Google Earth from here (~15MB)
WorldView-1 image strip of the eastern side of Heard Island (P001)
WorldView-1 image strip of the western sideof Heard Island (P002)
The latest 50 cm resolution WorldView-1 image covering the whole of Heard Island (acquired 23 March 2008, copyright DigitalGlobe).
3D view of the WorldView-1 image. Click on the image to download a KMZ file to view a resampled version (5 m resolution) of the image in Google Earth (~15MB).
The satellite images must be corrected for spatial distortions arising from topographical variations (such as mountains and gullies) in the earth’s surface and the tilt of the satellite as it passes over these features. This ‘orthorectification’ is done with the help of a digital elevation model (DEM) showing the terrain in 3D, on-ground photos, and global positioning system (GPS) data previously collected (Australian Antarctic Magazine 7: 10-11, 2004).
This video shows a Google Earth Tour of the new WorldView-1 (23 March 2008) and IKONOS (Jan. 2004) imagery of Heard Island.
Incredible detail can be obtained from WorldView-1 imagery. This image shows the snout of the Stephenson glacier with its deep crevasses.
Fifty centimeter resolution imagery allows us to see a group of Elephant Seals on the beach.
Feature extraction and Change detection
Once corrected, important features such as the coastline, glacial extent, vegetation and lagoons are manually digitised in a geographic information system (GIS). The eventual aim is to automate this time-consuming process using pattern-recognition software. Instead of analysing each individual pixel, as is common in traditional feature extraction methods, homogeneous (similar looking) objects would be derived from the image by image segmentation (see figure). Object characteristics such as spectral (light) properties, shape, size, texture and context can then be used to classify the object into a meaningful class and produce up-to-date maps. This so-called ‘object-based image analysis’ approach aims to simulate the way humans visually analyse imagery.
Features such as the coastline, lagoons, and glacial extent is digitised from the satellite imagery.
This small portion of a 60 cm resolution QuickBird image of Brown Lagoon, acquired on 30 January 2006,
was used for testing automated feature extraction. The image on the left is a false colour composite highlighting
vegetated areas in red. The image on the right is the result of object-based image analysis, which aims to simulate
human vision. The first step is to identify image objects that correspond to real-world features (snow patches,
lagoon, vegetation patches, etc.). The objects are then classified into land cover classes, resulting in a thematic
map with information that can be integrated with an existing GIS database. This automated analysis allows for
more efficient feature extraction than manual digitisation techniques.
The project also involves the development of automated techniques to identify changes in the coastline, glacial extent, and vegetation cover. Change detection is based on a comparison of two or more images. Preliminary results show that there has been an increase in vegetation between 1991 and 2006. Comparing QuickBird images from 2003 and 2006 revealed that detailed changes in vegetation communities can be automatically identified and mapped.
This map shows the glacial extend (as derived from satellite imagery) for several epochs dating back to 1947. Signifcant glacier retreat can be observed.
Photo by E.H.M (Tim) Ealey taken in 1949. The photo shows the establishment of Spit Camp. The glacier in the foreground is Stephenson Glacier. Behind Stephenson Glacier is Brown Glacier and behind Brown is Compton Glacier.
Photo by Eric Woehler taken in Dec 2003. The photo shows the same area as Ealey's photo, but glaciers have retreated. The photo wasn't taken in exactly the same position, as the coastline has moved inland about 20 metres.
This image was produced by overlaying the change detection result from the 1991 SPOT and 2006 QuickBird image comparison over a digital elevation model of Heard Island (with the SPOT image in the background). The white arrow indicates an area of vegetation increase on the Stephenson moraine (significant changes appear in red). The photographs of Stephenson moraine (below), taken in 1987 (left) and 2004 (right), show the changes in this area as observed in the field – note the increase in cushion plant cover in the foreground of the 2004 photo. Photo credit:Jenny Scott.
Within this remote sensing project we created orthorectified images from the Worldview-1 and QuickBird satellites and updated topographic maps of the island by extracting the coastline, glacial extent and lagoons, and we mapped detectable changes in these features.
A major objective of the project was to further develop the capability of the AADC to assist with applying a similar approach to the remote monitoring of the Australian Antarctic Division’s other areas of management and research interest.
EWAN MCIVOR1, ARKO LUCIEER2, URSULA HARRIS1 and ANGELA BENDER1
1 Australian Antarctic Data Centre, AAD
2 University of Tasmania