Imagine that a major hurricane will make landfall within the next 24 hours over a heavily populated coastal city like Houston, Texas. As the storm approaches the mainland, a continuing deluge of rainfall already blocks several primary evacuation routes. You are responsible for giving emergency managers an accurate forecast of the flash-flood potential in broad areas threatened by constantly worsening conditions, as rain bands sweep across the coastline and streams begin to rise beyond their banks.

That is your nightmare and your challenge, if you are Gordon Wells of the Center for Space Research (CSR) at The University of Texas at Austin (UT Austin). You have at your disposal hundreds of gigabytes of detailed elevation data for the region, collected by airborne LIDAR instruments. You have recent high-resolution orthoimagery from satellite and aerial surveys. You can run a sophisticated hydraulic and hydrologic model to simulate floods, using real-time NEXRAD Doppler radar data for estimating rainfall accumulation. Finally, you can access Houston's geographic information system (GIS), containing information about the stormwater drainage system, automated stream gages, evacuation routes, and critical infrastructure.

Now comes the crucial question, "one we ask ourselves every day," says Wells. "Even though the necessary data are at hand, can we make predictions of the impending disaster with the accuracy, detail, and timeliness required to offer critical guidance to all of the emergency management people who will deal with the consequences of the storm?"

There are two possible approaches to an answer to Wells's question, both involving computation. One is to simulate a large number of possible flash-flood scenarios well in advance of any actual storm. When the storm arrives, one attempts to match the real-time observed conditions to the most relevant scenario.

Unfortunately, according to Wells, even a very large collection of flood simulations could not capture the spatial and temporal complexities of rainfall distribution and watershed response during an actual flood. "We would be operating with scenarios that were far too general to give us the information we need," Wells says.

The second approach is to create real-time model simulations of metropolitan flooding as the events take place. Where would CSR find the required on-demand computational firepower?

CSR has long worked with the Texas Advanced Computing Center (TACC) at UT Austin on data-intensive processing requirements for NASA satellite missions. Notable among these are the monthly terrestrial gravity models produced from the Gravity Recovery and Climate Experiment (GRACE) launched in 2002.

CSR also operates a direct broadcast satellite receiving station as part of its Mid-American Geospatial Information Center (MAGIC) program. Wells and his group take advantage of TACC resources in processing and producing data products from a variety of satellite-borne remote sensing instruments. "We are engaged in a daily struggle to convert large-scale streams of data into timely information for state and federal emergency management agencies, regional and local governments, academic institutions, TV and radio broadcasters, and the public," he says.

The MAGIC satellite receiving station collects direct transmission of remotely sensed data from orbiting satellites and combines these data with geographic and demographic information to produce geospatial analyses deliverable within minutes in the case of such life-threatening emergencies as wildfires, tornadoes, flash floods, and hurricanes.

In the past three years, under the direction of Jay Boisseau, TACC has become one of the premier academic computing centers in the nation. In addition to supplying archival storage for data from CSR and other UT institutes (TACC's new system can store more than 2 petabytes--2 thousand trillion bytes), TACC operates several very large supercomputing systems. One, called Longhorn, is a large-scale IBM Power 4 System. The largest system, called Lonestar, is a Cray-Dell Linux cluster with 1024 Intel Xeon processors. Lonestar has a theoretical peak speed of 6.12 trillion floating-point operations per second (6.12 teraflops). Most recently installed is Maverick, a Sun Fire E25K server for high-end computation, visual rendering, and analysis. Maverick has 128 UltraSPARC IV multithreaded processors and a unique set of 16 dual-video-out commodity graphics processors. It is currently the most capable machine available for visualizing large-scale data at interactive frame rates. "Accurate real-time flood modeling demands this kind of responsive terascale visualization," Wells says.

In September 2003, TACC received a multimillion-dollar award from the National Science Foundation to become a participant in the TeraGrid, the nation's largest academic grid computing project. The TeraGrid links the resources of nine universities and national laboratories over the world's fastest dedicated network (running at 40 gigabits per second). With more than 40 teraflops of combined computing power and scientific visualization resources, the TeraGrid is an ideal platform on which to test a real-time, on-demand, and eventually nationwide flood prediction capability.

One operating mode of the TeraGrid is the support of science gateways, portals that incorporate TeraGrid resources into the day-to-day work of specific scientific communities. The Flood Modeling Science Gateway is one of ten such gateways, one of two directed at geoscience applications (the other is concerned with atmospheric modeling). Other gateways supply community resources for biosciences, chemistry, physics, and astronomy.

To address the need to provide real-time flood hazard forecasts for emergency management, TeraGrid participants from TACC, led by TACC scientist William Barth, will join with Wells and CSR participants and with the Center for Research in Water Resources (CRWR), also at UT Austin, and other TeraGrid participants at Oak Ridge National Laboratory (ORNL) and Purdue University. Together, they will develop the capability to model flood events in real time.

Their first project will be the parallelization of the Map2Map model developed by Professor David Maidment and his team from CRWR. This rainfall-map-to-inundation-map program is based on the ESRI ArcHydro data model (ESRI is a company formed from the Environmental Systems Research Institute). Map2Map can incorporate real-time rainfall estimates from the National Oceanic and Atmospheric Administration's NEXRAD Doppler radar network into a hydraulic/hydrologic model of a watershed area. The result is a prediction of inundation surfaces for affected areas. "Parallel processing will let us regenerate flood surfaces in near real time when we trigger the model with sequences of NEXRAD inputs," Wells explains.

The ORNL group, directed by Budhendra Bhaduri, is preparing dynamic population data for the test-case area around Houston, Texas. Their model traces the movements of people over a 24-hour cycle, weekdays and weekends. Where will people be concentrated when a dangerous hurricane makes landfall? ORNL can examine the impact of floods that occur at different times of the day, using a transportation model containing alternate evacuation routes tied to the dynamic population database.

The visualization at the beginning of this article shows the results of calculations performed to test the elements of the gateway system.

Large-scale flooding along the Brays Bayou in central Houston was triggered by heavy rainfall during Tropical Storm Allison on June 9, 2001. The flooding, which severely affected the Texas Medical Center (Baylor and UT medical facilities), caused a total of $2 billion worth of damage.

High-resolution elevation data (gathered for the Federal Emergency Management Agency using airborne light imaging detection and ranging [LIDAR] systems) were combined with NEXRAD precipitation data in the prototype real-time flood hazard model. To predict the impact of urban floods, the hazard analysis system combines a great deal of urban infrastructure data, including the LIDAR-derived elevations, building structure footprints, transportation routing information, stormwater drainage network data, and maps of utility conduits.

In a parallel computing environment, the rainfall-map-to-flood-map (Map2Map) application can generate a rapid series of predicted inundation surfaces according to changing rainfall inputs. Such predictions will allow emergency managers to make better informed decisions about the impacts of urban flooding on evacuation routes, critical infrastructure, and vulnerable populations.

"It is extremely rewarding for TACC to work with these talented researchers on problems that really make a difference in people's lives," Boisseau says. "We're passionate about using advanced computing technologies to solve important problems, and the Flood Modeling Science Gateway is work that showcases the value that supercomputers, visualization systems, databases, and grids have for society."

The geospatial information community may be among the first to benefit from portals to the TeraGrid, but it will be one of many. Just as the DOD-sponsored ARPANET evolved over the course of three decades to become the modern commodity Internet, the TeraGrid is a harbinger of the coming of a broad-reaching national cyberinfrastructure. It brings data processing, storage, and visualization capabilities to large numbers of users in the same way that today's Internet offers the standard digital content of text, graphics, music, and video.

When the Purdue Terrestrial Observatory (PTO) becomes operational later this year, the TeraGrid network will allow the CSR and PTO satellite receiving stations to exchange data collections, compare datasets, perform data quality analyses, and generate composite data products within minutes of data reception. As more such stations link to the TeraGrid and other high-speed networks, Wells says, "I foresee tracking satellites through a series of near real-time data exchanges that collect and integrate data from portions of the same orbital pass, from Canada across Latin America." And as grid supercomputing extends beyond the realm of university research, the outcome will likely change the way everyone views and uses geospatial information.

---

Portions of this article appeared earlier in Earth Observation Magazine, and we thank the magazine for permission to reprint these here.

Research Feature - April 4, 2005