Heard Island is a pretty cloudy place most of the time. However, there are occasional times when the weather clears, particularly on the southeastern (leeward) side of the island. On rare occasions, the northwest and southwest sides of the island come out from the clouds as a satellite passes over.
For the past two years, I have been watching Heard Island using true-color imagery from four satellites: Terra, Aqua, Landsat 8, and EO-1. I have posted previously about satellite imagery from these instruments. Although every image of the Island I have seen has clouds in it covering a portion of the island, I was curious whether or not I had accumulated clear imagery of the entirety of Heard Island.
In part, this question was spurred by a follower on Twitter asking about eruptive activity at Heard. I had to admit I didn’t really know whether the activity was low-level and continuous (like Kilauea) or more intermittent. Given that our knowledge of its eruptive activity is primarily from satellite observations, do the satellite “thermal anomalies” correspond to short eruptive events, or simply a cloud-free view of the volcano?
For high-resolution imagery of Heard Island, the datasets of interest are from EO-1 ALI, and Landsat 8 OLI. The two MODIS instruments (one on Aqua, one on Terra) are moderate-resolution, and while 250-m resolution is sufficient for some purposes, this one needs more. Looking through the archives, I was able to find EO-1 ALI data primarily for Mawson Peak and points southeast, and Landsat 8 OLI covered much of the island, particularly the northwest.
Not only is having cloud-free, high-resolution data important for me, but I want the data to be recent. There has been a retreat of up to 5.5 km for some of the glaciers since 1947, and the Google Maps imagery of that area (Stephenson Lagoon) is horribly outdated. Fortunately, I found most of the island covered in large swaths with images from 2014 onward, and mostly 2016. There was even good imagery from when I was on Heard Island! Our ship, the Braveheart, is visible as a few white pixels in Atlas Roads (just north of Atlas Cove), slightly closer to the Azorella Peninsula than to the Laurens Peninsula. The tents and campsite are too small and darkly colored to be visible on this image.
A small portion of the island between Atlas Cove and Mawson Peak was the most difficult to find. With the topography of the island, the steady stream of wind, and the humid air, the 2.5 km by 2.5 km region was cloudy pretty much all the time. Eventually, using the EO-1 ALI instrument and going back to early 2010, I found a reasonably clear image of it.
Once I had the images (after combining true-color and panchromatic brightness data in QGIS), I needed to stitch them together. Thanks to the wonderful QGIS training manual, I was able to create vector (polygon) layers which corresponded to the clear region of each image (plus some surrounding ocean). At this point the troublesome mostly-cloudy spot became evident, and the search was on for imagery to fill the void.
Finally, I tried to put them together. This turned out to be more trouble than it was worth for my purposes, having only five images. Several of the images had differing resolutions (10 m/pixel for EO-1 ALI, 15 m/pixel for Landsat 8 OLI). Additionally, since I was handling these in their raw format, color balances/exposures were not consistent across images. I decided it best, then, to leave them separate, and sent them around to the Heard Island Expedition team.
Soon I had an email from the expedition leader: he was very interested in the imagery, but it wasn’t opening in Google Earth. Some searching later, I found that Google Earth works best with a certain map projection (EPSG:4326), and when exporting the GeoTIFF, I needed to select “rendered image” rather than “raw data”. I re-exported the images, zipped them up, and tested it out on another computer: success! This Google Earth friendly imagery is now available here (17 MB zip).
One continuation of this project would be to keep looking through the documentation on GeoTiffs to find out how to make the rendered images use a transparent, not white, border where there is no data. That would likely let me create a virtual raster catalog to load all of them in one go, rather than having to load them separately.