Error, Certainty, and Sharing Results
I love American Geophysical Union’s weekly newspaper, “EOS”. It’s a great weekly summary of what’s going on in the earth sciences. I make it a point to stay current on it, even at least skimming an item if it’s not directly in my area of study. A front page article this spring really caught my eye, “An Obituary For Sandy Island”. It ended up getting popular press coverage in places like The Washington Post. Coastal change is one of my main areas of study, so I dug in to it. I was thinking it was a story of storm surge wiping out a barrier island, or something like that. Instead of another piece of data on geophysical dynamics, though, I got a nice insight in to the importance of how we acquire and process data.
The article explained that a pumice raft likely fooled a 1870s whaling ship in to charting an island in the Coral Sea. The error was eventually found and the island eradicated from most paper maps. Several times. But not the paper maps that were converted in to a popular database. And it turns out parents everywhere are right: once it’s posted to the internet, it’s there almost forever. Sandy Island was accidentally included and enjoyed a new digital life for quite a time, until this past October when researchers collecting data in the area noticed an island the size of Washington, DC was on some maps but not others. Being scientists, they knew “I don’t know” should always be followed up with “Let’s find out!” Finally, somebody actually went to check on the island. It’s been removed from active databases (I checked Google Earth, but it’s fixed there). So hopefully it’s been killed off once and for all, unless islands are included in the Zombie Apocalypse, I suppose.
There are a lot of good points in the saga of Sandy Island for scientists and anyone who uses data. That’s pretty much everyone these days. Who collected the data? How’d they get it? What’s its provenance? How has it been checked and validated? Sandy Island is a success story, though. It started with an observation, which wasn’t totally wrong: that pumice raft would have been cool to see. It was reasonable to believe there was yet another island in that region: there are about 25,000 in Oceania anyway. It was usually marked as unsubstantiated or “Existence Doubtful”. When solid contrary evidence was found, the island was promptly 86’d. That’s how science works: Make the best use of data to form the best current hypothesis and update as new data come along.
That’s easy enough in cartography, to a point. The place in question either exists or it doesn’t and you can go and check on it if you need to. But how do we do that in day-to-day work with data? A lot of people smarter than me have spent careers developing best practices for just that. Next week, at VISualize 2013, some of them will show what they’ve been doing to acquire and analyze the best data possible for getting a handle on global change and environmental monitoring. I’ll be there, presenting with colleagues from Esri and running a seminar on Landsat 8. I’d love to see you there! So, stop by if you can. It’s not hard to get to, at the World Wildlife Conference Center, in Washington, DC. We had a spot booked on Sandy Island for a super cheap rate, but the travel logistics turned out to be a nightmare.