Focusing radar's lensRemote sensing tools reveal ecological secrets
This article is the fourth in a series on the nine NASA HPCC Earth and Space Sciences (ESS) Project Science Team II Grand Challenge Investigations. by Jarrett Cohen |
Issue 4, January 1998 Welcome to the on-line version of NASA's Insights Magazine.Insights was published by the High Performance Computing and Communications (HPCC) Program Office. Program Manager:
|
|
As a young mountain climber, Jeff Dozier witnessed several avalanches safely from above. Their naked power drew him to the science of snow. "I learned that there are many interesting questions about water and vapor transport in snow, and about snow's electromagnetic 'signature.' I combined my hobby and my job," said the dean of the School of Environmental Science and Management at the University of California, Santa Barbara (UCSB).
Of greatest concern to Dozier and fellow Californians is how much water melted snow will produce in the spring: mountain runoff supplies the bulk of water for agriculture, homes and workplaces. The one method to reliably predict this "snow water equivalence" is synthetic aperture radar, or SAR. Aircraft, the space shuttle and satellites have been the mounts for SAR antennae, which bounce radar signals off the earth and use the return time to gauge the distance. Radar's self-illumination and long wavelengths make SAR an any-time, all-weather instrument. For Dozier, "SAR is the only remote sensing technology that sees down into the snow and is the only way to measure the amount of snow" across large areas.
However singular its traits, SAR is useless without computers. "The signals look like white noise because the radar doesn't have a lens," said Grand Challenge Investigator Dave Curkendall of NASA's Jet Propulsion Laboratory (JPL). "Like the aperture controlling how much light enters a camera lens, SAR gets a wonderful aperture from the motion of the spacecraft. The computer then supplies the lens that focuses the image." Curkendall estimates that it requires 800 billion floating-point operations (flops) to "focus" one image taken by NASA's Shuttle Imaging Radar-C (SIR-C), for example. Processing regional data like that for snow water equivalence becomes a Grand Challenge problem and thus a task for today's fastest supercomputers. To tame this "computational bear," Curkendall's team is exploiting SAR's unique characteristics for three crucial ecological problems.
Melted snow forecasts High-performance computing enables widespread measurement of snow water equivalence, "the major unsolved problem in snow hydrology," Dozier said. The goal is to process all the snow scenes from SIR-C's 10-day flights of April and October 1994, months when melting occurs. Validating SIR-C predictions with 1994 runoff data will "show what we could do with a permanent orbiting SAR in space," he said.
Mountainous areas such as the Sierra Nevada, Alps, Himalayas and Andes are the primary interest for the validation. Since there will be hundreds of scenes involving trillions of bytes, UCSB scientists are adapting their workstation-conceived algorithm for parallel computers. The snow algorithm transforms raw SIR-C signals into images that are analyzed to determine which areas are snow-covered, the quantity of snow (usually expressed by snow water equivalence) and if snow is wet or dry (wet snow adds to runoff). Combining these factors with weather forecasts pinpoints snow water equivalence down to as accurate as 20 centimeters. Field measurements of the same variable can only cover a few sites in any given time period. Today, researchers ski across mountains to extract cores and weigh snow by hand and then compare the measurements with river runoff data. "It gives a pretty good estimate, but like all empirical methods it occasionally goes wrong, with errors of 30 percent in some years," Dozier said. Taking advantage of SAR's ability to "look at widely independent regions," the snow algorithm "calculates all possibilities," said Jiancheng Shi, assistant research scientist at UCSB. "Previously, local conditions limited the measurements."
With a U.S. SAR satellite and the snow algorithm's maturation, better runoff forecasts could become reality. In California alone, "a one percent improvement would be worth millions of dollars," he said. "We are working with the California Department of Water Resources to translate research results into operational practice." Cracking earthquakes open As with the water supply's ebbs and flows, a worrisome ecological threat for California is earthquakes. SAR's penetration of earth's surface and high spatial resolution are proving key to studying tectonic plate movements in the state's southern portion.
Capturing movement demands repeat-pass SAR interferometry, which correlates data from multiple satellite orbits. Scientists at San Diego's Scripps Institution of Oceanography are focusing on the European Earth Resource Satellites (ERS-1 and ERS-2), using interferometric SAR technology largely developed at JPL. "You take two images that span a tectonic event and fuse them into an interferogram," said David Sandwell, Scripps professor of marine geophysics. "The satellite doesn't always come back to the same position, adding a parallax effect. What you see is the topography; you have to remove the topographic signature." Peeling away that signature hopefully will open up a sweeping view of Southern California deformation every 35 days, he said. A complementary network of 50 Global Positioning System (GPS) satellite stations has disclosed total movement along the 400 kilometer plate boundary to be 45 to 50 millimeters per year, "about the rate at which your fingernails grow," Curkendall said.
|
"SAR gives us the possibility of determining the deformation over the continuum of Southern California, not just a selection of points, but a spatially dense picture," said Yehuda Bock, research geodesist at Scripps. "The reason to combine the two is that these GPS positions are known to a millimeter or so. We use them as calibration points, so the map of deformation is more accurate." What GPS and earlier approaches cannot do is "get a handle on inhomogeneity in the crustal deformation, including localized effects that may be due to non-tectonic sources, such as water withdrawal," Scripps professor of geophysics Jean-Bernard Minster explained. "SAR's regional snapshot can give us a detailed understanding of how the Earth prepares for earthquakes." With no detection of a particular tremor signal, this knowledge will not likely lead to earthquake prediction. However, "a fault may be sliding and relieving its stress," Sandwell said, pointing to safe areas for buildings and roads. "Predicting the absence of a big earthquake is almost as important as attempting to predict the earthquake location and time."
Rejoicing at floods Much more relevant than prediction in the Amazon rainforest are regular and timely surveys showing inundation (or flooding) and deforestation zones. SAR again is uniquely qualified for the duty, as Brazilian farmers practice "slash and burn" agriculture. "They fell the larger trees and burn them" for crop fertilizer, said Tony Freeman, a radar instrument manager at JPL. "The land is completely obscured by smoke, but radar sees right through it." Japan's Earth Resources Satellite-1 (JERS-1) can cover the United States-sized region in two months, encompassing the flood (April-June) and dry seasons (October-November) that JPL and others are mapping. Freeman and collaborators have completed a JERS-1 map of 1995's dry season. The satellite took 1,500 70-square kilometer frames of data. When processed, "there is some overlap between image scenes, much like when you take landscape portraits," Freeman said. "We use correlation matching to determine what that overlap is, then run a large matrix inversion program to co-register images." JPL-developed software pieced together the frames into a seamless mosaic.
Herb Siegel, JPL The dry season map clearly traces "the shape and size of the river course, which had been slightly unknown," Freeman said. "We can see what areas are inundated during this season across the whole Amazon; this is the primary product. The secondary product is a map of the forest, deforestation and land cover." Classification studies will calculate these categories every 100 meters. "It takes several years to establish deforestation from optical satellite photos," Freeman reflected. Using SAR, "we have the ability to monitor this phenomenon and floods every six months," leading to adjustments in local strategies. "When the Amazon floods big time, the people are extremely happy," he said. Crops are better because of washed-in silt, and fish are bigger and more plentiful. "Fisherman could take more fish from the river if they know it will be superabundant in one year." The mosaics also are of interest to scientists studying global change because the areas where there is standing water tend to produce the greenhouse gas methane. Real-time impacts Building mosaics of such large regions is one function of the Digital Light Table software for "examining images too large for traditional viewing programs," said JPL's Herb Siegel, chief computational scientist for the Grand Challenge team. "The program is a window on the disk," only using workstation memory to pan and zoom through on-screen images. The Digital Light Table will be folded into a Scalable SAR Software Suite (S4) aimed at making SAR more usable. Interferometry processing, drawing on the Scripps algorithms, is one emphasis. Speeding up SAR processing in general will be another prime contribution.
Three years after its two flights, just over 20 percent of the SIR-C data has been processed into image products ordered by researchers. "The instrument takes about 15 seconds to record an image. SIR-C's dedicated supercomputer takes one hour to process that 15 seconds," said Craig Miller, JPL software developer. With NASA HPCC supporting development and access to the latest scalable parallel computers, S4 is nearing 100 seconds per image. A 100 billion flops milestone due in 1999 will "reduce it to 50 seconds, near real-time," Miller said.
"You'll never exploit the potential of SAR until you do it in real-time," Freeman stressed. Speed is especially critical if NASA launches a U.S. SAR satellite. With real-time processing, "if you have a big snowfall or disaster, you could analyze the data in days," Shi said. Similarly, Minster envisions a weekly picture of Southern California tectonic plate evolution. "Right now, we only get data as targets of opportunity," he bemoaned.
Anticipating these kinds of explorations, JPL is tooling S4 for distributed computing. "We're moving the computation all over the country" via high-performance networks to wherever parallel systems are available, Siegel said.
"A scientist should be able to say 'I would like to form an interferogram between the ERS-2 passes over Southern California on July 15 and August 31' and initiate a run that finds that data, brings the software to large-scale machines, does the processing and returns the result in a visual way," Curkendall said. "That whole sequence will transform Earth science data processing from a product to a capability."
|
|
| The three-panel view (above) shows successively higher resolution views of the Amazon River basin as mapped by the Japanese Earth Resources Satellite-1. Flooding is clearly seen below tree canopies (bright yellow) and in open water areas (blue). |