objective, which they accomplished last year,
involved replicating the results from that proof of
concept with public data available through Google
Earth Engine, a cloud-computing platform that
stores large satellite-image data sets. “We chose to
start with public data because it helps us establish
a baseline for what we can accomplish with
minimum cost, and there are a variety of different
satellite collections available on the platform,”
Mitchell explains.
Their focus now, Mitchell says, is on developing
a general monitoring model and then using it “to
make as many predictions as we can.” They hope,
for example, to be able to estimate the total annual
carbon emissions of individual plants, and they’d
like to eventually interpolate between predictions
to create “granular profiles” of each facility.
AIMING HIGH
As of now, according to Isabella Soldner-Rembold,
a data scientist on the Carbon Tracker team, the
program is using data from just a handful of satellites.
Different satellites have different orbits, she explains,
and the imagery each one is able to provide depends
on its location at different times of day.
The coalition doesn’t own the satellites in its
network, but instead relies on the data feeds from
commercial and government satellites like the
European Space Agency’s Copernicus Sentinel-2.
Such satellites are preferable to anything they
might launch independently because many have
been collecting data for years. “When training
machine learning algorithms, more training data
generally increases the accuracy of the model,”
Soldner-Rembold says. The initiative is still in the
research and development phase, but the team
hopes to release data as early as fall 2020, and to
eventually—perhaps within two years—have data
from enough satellites to enable round-the-clock
monitoring “of every power plant across the entire
globe.”
Her current focus, Mitchell says, involves
using machine learning techniques with Earth
Engine to process not only their satellite data
but also “ground truth” data from power plants
themselves. The coalition’s AI algorithms, she
explains, can analyze a range of indicators of
power plant emissions, whether it’s a thermal
infrared image showing signs of heat or certain
colors in an image suggesting the presence of
smoke. This information, in turn, can help them
determine whether a facility is a “baseload
plant” that operates most of the time, or if it’s
more likely a “peaker plant” that only generates
electricity during very high demand. (Peaker
plants are typically high emitters of CO 2 and other
pollutants.) “This data is helping us see things we
could never see before, and to really understand
the actual impact” of any given power plant at any
time, she says.
Looking ahead, Mitchell and Soldner-Rembold
believe the data their organizations make public
will be of great interest to a wide variety of
stakeholders. Investment firms, for instance,
may use it to better understand the financial
implications of climate change and to steer their
clients away from high-cost polluters. Activists may
turn to the same information to lobby politicians
for policy changes, and governments keen on
meeting the goals of the Paris Agreement may use
it to drive their countries toward a cleaner future.
In the end, Mitchell predicts, the people it will
benefit most “are those who are willing and really
want to initiate change, but right now are unable
to because they don’t have the resources.” If the
project helps to turn the tide, “I think we would call
it a success.” ■
See Dell Technologies’ vision for
advancing sustainability by 2030.
DellTechnologies.com/ProgressMadeReal
39