Researchers from IBM and the University of Texas at Austin
have applied advanced analytics to river systems, weather and sensor data, to
predict the Guadalupe River’s behavior more than a hundred times the normal
speed. Floods are the most common
natural disaster in the United States, but traditionally flood prediction
methods are focused only on the main stems of the largest rivers – overlooking
extensive tributary networks where flooding actually starts, and where flash
floods threaten lives and property.
IBM's new flood prediction technology can simulate tens of
thousands of river branches at a time and could scale further to predict the
behavior of millions of branches simultaneously. By coupling analytics software
with advanced weather simulation models, such as IBM's Deep Thunder,
municipalities and disaster response teams could make emergency plans and
pinpoint potential flood areas on a river.
“Effective flood preparedness can be looked at as a large
scale computing problem, with a huge number of relevant data and
independencies,” said Frank Liu, Research Staff Member at IBM Research –
Austin, “Using advanced models to simulate the scores of tributaries of large
rivers along with other relevant real-time information such as weather, we are
better able to give people valuable advance notice of a flood.”
As a testing ground, the team has applied the model to
predict the entire 230 mile-long Guadalupe River and over 9,000 miles of
tributaries in Texas. In a single hour the system can currently generate up to
100 hours of river behavior.
"Combining IBM's complex system modeling with our
research into river physics, we've developed new ways to look at an old
problem," said Ben Hodges, Associate Professor at UT Austin Center for
Research in Water Resources. "Unlike previous methods, the IBM approach
scales-up for massive networks and has the potential to simulate millions of
river miles at once. With the use of river sensors integrated into web-based
information systems, we can take this model even further."
Speed on this scale is a significant advantage for smaller
scale river problems, such as urban and suburban flash flooding caused by
severe thunderstorms. Within the emergency response network in Austin, Texas,
professors from University of Texas at Austin are linking the river model
directly to NEXRAD radar precipitation to better predict flood risk on a
creek-by-creek basis.
In addition to flood prediction, a similar system could be
used for irrigation management, helping to create equitable irrigation plans
and ensure compliance with habitat conservation efforts. The models could allow
managers to evaluate multiple “what if” scenarios to create better plans for
handling both droughts and water surplus.
The project is currently
being run on IBM’s Power 7 systems, which accelerate the simulation and
prediction significantly, allowing for additional disaster prevention and
emergency response preparation.
No comments:
Post a Comment