By: Cassidy Newland (AIMS)
It’s as strange as it sounds… researchers are looking for whales in the Kimberley from a vantage point of 770 kilometres above sea level using the Worldview 2 Satellite.
The Satellite captures a spatial resolution of 0.46 metre pixels in the Panchromatic band and a spatial resolution of 1.6 metre pixels in the 8 band Multispectral image (Figure 1).
The Panchromatic band captures light across the visible spectrum in a single band and is often displayed as a grey scale image. It has a higher spatial resolution of up to 0.46m, but a lower spectral resolution than the multispectral. The multispectral has 8 bands extending from the far blue through the visible spectrum to the near infra-red providing higher spectral resolution but the spatial resolution is only up to 1.6m.
A number of surveys by different organisations had previously been carried out in the Kimberley by plane, boat and some from land and for the first stage I looked for an image in the World view 2 archive to match one of these.
Although there is a large archive of imagery since the satellite commenced operation in 2009, matching an exact date was difficult and we were lucky to find several images matching surveys conducted in August 2010. I acquired one image which I expected would have the most whales and began the task of looking for them.
At 12-16 metres long you might expect a Humpback whale to be relatively easy to distinguish, but in the multispectral image with 1.6m pixels, a whale seen fully surfaced will only be up to 10 pixels long. Figure 2 shows a simulation based on an actual aerial photo of what a whale might look like in a satellite image.
In actuality only a single fully surfaced whale was distinguishable, but with a lot of interpretation and a little imagination a range of whale related features were identified including partially surfaced whales, submerged whales, the foot print of recently surfaced whales, bubble rings from below and even what appears to be a bubble net.
These features are examined in each band to see where they can be best distinguished and determine what band or combination of bands will be used in the analysis to identify further features. There is also testing of the similarity of features within a type and testing of separability of each type. The result is a refined set of features which can then be used to train and verify success of the remote sensing techniques used.
The techniques used included thresholding where cut-off values are defined manually, supervised classifications using the features as training sites and unsupervised classifications where clusters are identified statistically. Of these, thresholding and unsupervised classification provided the best results.
Challenges were noise from shallow water, swell and turbidity, but it was possible to identify surfaced whales, whale footprints, some submerged whales and boats.
ABC Kimberley’s Erin Parke talks to WAMSI/AIMS researcher Michele Thums about satellite tagging humpback whales:
The $30 million Kimberley Marine Research Program is funded through major investment supported by $12 million from the Western Australian government’s Kimberley Science and Conservation Strategy co-invested by the WAMSI partners and supported by the Traditional Owners of the Kimberley.