IEEE Spectrum June, 2014 - 12

the satellites' visible light and infrared data.
Each pixel of satellite imagery is characterized by both its color and its infrared signature, and the algorithms then compare
the data for that pixel across time to detect
changes. A switch from green to brown, for
example, is a bad sign. Hansen pioneered
this technique in his earlier research on the
Congo Basin, where the ground was very often obscured by clouds. Rather than throw
out the cloudy images, Hansen developed
ways to create composite pictures using
many days' worth of images. "We learned
how to work pixel by pixel," he says.
When Landsat data became available at no
charge in 2008, Hansen worked with Google
Earth Engine to apply his model globally, looking at 143 billion pixels with 30-meter resolution. By tracking the pixels over months and
years, the model corrects for seasonal changes to forests and can distinguish between
crops and woodlands. The collaborators published their results last November, revealing
a net loss of 1.5 million square kilometers of
forest between 2000 and 2012. Those calculations, the researchers noted, took 1 million
CPU-core hours on 10 000 computers.
Thau says that Google Earth Engine aims
to take the pain out of this big-data research.
In typical cloud computing, he says, researchers have to manage the distribution
of their computing tasks across the network.

With Earth Engine, however, researchers
simply use a programming interface to enter their queries, which get "parallelized"
automatically, Thau says.
By creating the Global Forest Watch website, the World Resources Institute aims to
give the public access to all that big data too.
Dan Hammer, chief data scientist at the organization's Data Lab, says he expects that
government agencies, businesses, researchers, and advocacy groups will use the site to
get a better picture of forest management.
At the environmental action group Rainforest Action Network, agribusiness campaigner Gemma Tillack says the new tool may
be particularly useful in Indonesia, where
rain forests are falling to make way for palm
oil plantations. Her group is asking 20 big
food corporations to guarantee that the palm
oil used in their products is being grown on
legal, sustainable plantations. Some companies are establishing responsible palm oil
procurement policies, she says, and Global
Forest Watch could help them implement
those policies. "They need to find out where
the palm oil they're buying is coming from,
and then they'll need to monitor the actions
of their supply-chain partners," she says.
Time will tell if the site will make any difference in the seemingly inexorable advance
of bulldozers, but Hammer is optimistic.
"I'm consistently surprised by how much
open access to data can fix things
in the world," he says. Many policy decisions are delayed by arguments over conflicting information,
Hammer notes, but the objective
data provided by Global Forest
Watch has the potential to eliminate such confusion. "That would
let people get to the hard questions
about what should be going on in the
forests rather than what is going on,"
he says. -eliza str ick l a nd
A version of this article appeared
online in April.
troublE in indonEsiA: Global Forest
Watch shows deforestation in Indonesia,
where land is often cleared for palm oil
plantations.

JUn 2014

nORTh aMERICan

SPECTRUM.IEEE.ORG

GloBal ForesT WaTch

Google is developing this platform to host
petabytes of earth science data and to give
researchers a straightforward way to use it.
"They log on, access all the data, and run their
own algorithms," explains David Thau, the
senior developer advocate for Google Earth
Engine. Thau and his colleagues work with
scientists to develop useful analysis functions,
and then they "get out of the way," he says, and
let researchers conduct their investigations.
Google Earth Engine is currently available to
thousands of research partners, and the company plans a general release down the line.
Global Forest Watch is the result of a convergence of projects. The World Resources
Institute's Data Lab had been working on a
forest-clearing-alert system for the tropics
based on data from the MODIS (Moderate
Resolution Imaging Spectroradiometer), instruments that ride aboard NASA's Terra and
Aqua satellites. Meanwhile, Matthew Hansen,
a professor of geographical sciences at the
University of Maryland, had been collaborating with Google Earth Engine on a global
map of deforestation; his project used images
from the Landsat satellites operated by NASA
and the USGS. Both data sets are now used to
create Global Forest Watch; MODIS provides
better temporal resolution, while Landsat
provides exemplary spatial resolution.
The researchers' algorithms create the
site's dramatic map of forest loss using

http://SPECTRUM.IEEE.ORG

Table of Contents for the Digital Edition of IEEE Spectrum June, 2014