Pigs fight for their lives as floodwaters from Hurricane Floyd invade a North Carolina farm. Such disasters caused by El Niño and La Niña help motivate the NASA Seasonal to Interannual Prediction Project to improve forecasts of the phenomena and their effects on the climate.

Drought dried up the Middle and South Atlantic states. It was the worst in two generations.

Then Hurricane Floyd arrived, with floods entombing one million farm animals in water and mud.

Last summer's disasters were influenced by the hand of La Niña, a periodic cooling of the eastern Pacific Ocean along the equator.

La Niña and its warm-water brother, El Niño, contribute to major shifts in the atmosphere's jet streams, impacting weather worldwide. Knowing their timing and strength before they appear again would help people better prepare for extreme conditions.

Figure NOAA/PMEL/TAO Project Office

The ocean and atmosphere interact to produce El Niño, a periodic warming (shown in red) of the eastern Pacific Ocean along the equator. A weakening of tropical trade winds in the west spawns an eastward-moving wave. The sea surface warms as the wave lowers the thermocline, the dividing line between warm and cool waters, in the east. Increased atmospheric convection further weakens the trade winds and supports the warm surface waters. larger pix

"If Florida is wet or dry, that is going to affect its orange crop. In the summer or fall that follows, will there be more or fewer hurricanes?" asks Michele Rienecker, a NASA oceanographer seeking to improve such short-term predictions by combining supercomputer-based climate models and satellite observations.

Across the United States, "Californians will worry if they will have an anomalously wet season," a typical response to El Niño, she says. "State authorities want to plan for beach erosion and landslides."

Rienecker leads the NASA Seasonal to Interannual Prediction Project, or NSIPP, at Goddard Space Flight Center. "We want to help operational agencies such as the National Weather Service in how they assimilate satellite observations into their forecast models," she says.

Embarking on the quest to develop better forecasting tools, NSIPP is currently focusing on El Niño-La Niña and their effects on the North American climate.

Getting the past right

"Not all events are the same, which makes it difficult to predict El Niño," Rienecker says. "With 1997-98, forecasts did well once the event was underway, but models didn't do so well in predicting the onset of that event."

To improve upon that prediction record, NSIPP is initially applying its techniques to past events. In these retrospective forecasts, they look at predictability and how one event differs from another.

In making forecasts, NSIPP must use numerical models of both the ocean and the atmosphere, because in the real world they interact to produce El Niño and La Niña. "The tropical trade winds blowing over the equatorial ocean make waters in that region slosh back and forth," explains Max Suarez, atmospheric scientist and NSIPP investigator. "That sloshing affects the temperatures that produce El Niño, which in turn creates the winds that tend to support the temperatures."

NSIPP runs up to 50 simulations per forecast, each with slightly different starting conditions. Averaging the results better accounts for the chaotic atmosphere, which is not predictable beyond about two weeks. Greater fidelity also comes from dividing the simulated globe into as many boxes as possible so the physics equations can be solved for smaller regions.

Visualizations of sea surface temperature mapped onto sea surface height show El Niño (January 1998) and La Niña (June 1998) conditions. Red traces warm water, and blue traces cool water. As demonstrated in side-by-side comparisons, NSIPP's oceanic model produces a fairly realistic match with satellite measurements. larger pix

The CRAY T3E supercomputer at Goddard ably meets these demands with its 1,024 parallel processors. NSIPP scientists say it would have taken years to develop parallel code had they not participated in NASA HPCC's Earth and Space Sciences Project. Instead, they were ready to do forecasts within weeks of installing the CRAY T3E.

Harnessing this computational engine, NSIPP works toward building a forecasting system that can produce the next El Niño or La Niña. The only way to predict the future is to couple the climate models so the simulated ocean and atmosphere can exchange information as they do on Earth. A coupled model dynamically creates values such as winds and sea surface temperatures. "If the model's climate is realistic, we expect it will be useful for predicting future climate anomalies up to a year in advance," says Sonya Miller, NSIPP support scientist.

"Not all events are the same, which makes it difficult to predict El Niño." Michele Rienecker Goddard Space Flight Center
NSIPP oceanographer Michele Rienecker and atmospheric scientist Max Suarez use the 1,024-processor CRAY T3E supercomputer at Goddard Space Flight Center. larger pix

NSIPP tests the coupled model's El Niño-La Niña statistics against past observations, as well as against results from the oceanic model driven by satellite observations of the atmosphere. The standalone oceanic model's output is fairly accurate, but some problems arise from the coupling. For example, the coupled model correctly predicts the current La Niña's occurrence, but it gets the timing of the 1997-98 El Niño wrong when predicting a year ahead. Overall, the modeled climate drifts and will eventually affect the quality of the forecast. Drift manifests in discrepancies such as modeled sea surface temperatures that are one or two degrees colder than actual observations.

"Slight differences from observed features can introduce errors in the oceanic model; these errors will be amplified by the atmosphere," says Augustin Vintzileos, NSIPP senior research scientist. NSIPP is researching which elements of their climate models are responsible for drift. Possible culprits include oversimplifying representation of some physical processes and having too few computational boxes.

NSIPP drives its atmospheric model with sea surface temperature measurements to forecast El Niño-La Niña's effects. These tables show North American precipitation during winter 1983's El Niño and winter 1989's La Niña. NSIPP scientists compare the mean (upper left) of nine simulations with observations (upper right). Coloring shows how much precipitation varies from normal. larger pix

Advances in realism are expected as NSIPP takes the pioneering step of blending satellite observations of the ocean with the oceanic model. Known as data assimilation, this technique also accounts for model and data errors to provide a better starting point for forecasts. "If you want models to forecast tomorrow's weather conditions, you first need to let them know about today's," says Christian Keppenne, NSIPP senior research scientist.

Weather forecasters regularly assimilate atmospheric data, but NSIPP concentrates on the ocean because its slow-changing nature bears the memory of El Niño and La Niña. Sea surface height measurements are especially important for studying these phenomena. By assimilating height information into the ocean model, scientists can locate the chief indicator of El Niño-La Niña: anomalies in the thermocline -- the dividing line between warm surface waters and the cooler waters of the deep ocean.

El Niño lowers the thermocline in the eastern equatorial Pacific, sustaining warm waters at the surface. La Niña raises the thermocline so strong winds can more easily lift up cold water. "Assimilating surface data into the model improves the representation of sub-surface oceanic processes such as changes in the depth of the thermocline," Keppenne says.

Seasonal climate

By refining its models, NSIPP also anticipates improved forecasts of El Niño-La Niña influences on climate, which is how weather is organized for a season or longer. "The dominant response to the El Niño signal is right over North America, so the United States has the greatest interest in the forecast problem," Suarez says.

This visualization depicts the ocean-atmosphere dynamics of the 1997-98 El Niño, the strongest on record. A red tongue of warm, raised water extends westward from the South American coast. Black arrows indicate sea surface winds. larger pix

The central question the nation's modelers are trying to answer is "How big is the signal relative to the noise, or variability, of weather?" says Siegfried Schubert, meteorologist and NSIPP investigator. Schubert stresses that even a perfect climate model could not produce a perfect forecast because of atmospheric chaos. "We're up against a property of nature," he says. Nevertheless, NSIPP is making progress by averaging results from multiple runs of the atmospheric model driven by sea surface temperature observations. "We have a good start on understanding how much the details of sea surface temperatures matter for response in different regions," Schubert says.

If sea surface temperatures dominate interactions between the Pacific Ocean and atmospheric circulation over North America, an event's climatic effects are more predictable. NSIPP simulations show that to be the case during El Niño and the winter season, as both extend strong jet streams over the continent. Model runs of these conditions in winter 1983's major El Niño confirm relatively high predictability by reproducing more precipitation than normal on the West Coast and less precipitation than normal on the East Coast.

Prediction becomes more difficult when sea surface temperatures do not dominate the atmospheric response over North America, but NSIPP is rising to the challenge with its research on forecasting summer precipitation. To understand something like last summer's roller coaster weather pattern, "local interactions between the atmosphere and land must be simulated correctly," Suarez says. NSIPP entrusts this responsibility to its atmospheric model's land surface component, which covers the space "from the trees down," as Suarez puts it. Relevant to El Niño-La Niña studies, the component includes water runoff, evaporation and -- most critical of all -- soil moisture.

Higher soil moisture appears to increase precipitation by allowing greater evaporation. Using the land surface component shows that "predictability could be enhanced if you could predict the state of the soil moisture in addition to the El Niño signal," says Randal Koster, hydrologist and NSIPP investigator. He says the key discovery thus far is that "land has the greatest impact on predictability in the transition zones between very dry and very humid regions." These transitions are common in central North America, which bodes well for future successful forecasts.

Sonya Miller, NSIPP support scientist, and Ted Krumbach, animator in Goddard's Scientific Visualization Studio, discuss new visualizations of sea surface height anomalies. Incorporating satellite-observed height data into NSIPP's oceanic model will enable greater realism in representing sub-surface processes involved in El Niño and La Niña. larger pix

Focusing on specifics

In NSIPP's efforts to predict El Niño-La Niña's widespread effects, "gross climatological features do well at 2-degree resolution," or a computational box 170 miles wide over the continental United States, Schubert says. "That resolution is probably not good enough if we want to know if it will be raining over Wisconsin or Maryland. With between 1 and 1/2 degree, we would be happy."

Climate models are hungry -- double their resolution, and they want eight times more computation. NASA will have such power within a few years as it acquires a supercomputer able to sustain more than one trillion floating-point operations per second. For an interim solution, NSIPP uses an innovation called a stretched grid that concentrates the boxes in a particular part of the globe. A collaborative project with the National Weather Service will apply this technique to North America, especially to discern how East Coast droughts are linked to La Niña.

As the climate community incorporates knowledge from such studies and from NSIPP's strides in coupled models and data assimilation, they will be that much closer to generating accurate forecasts far enough ahead so governments and citizens can plan accordingly.

Credits for Insights Magazine go to the following people along with the writers and photographers who contribute to each issue and the researchers and specialists whose material is highlighted:

Program manager: Dr. Eugene L. Tu
Deputy program manager: Dr. William R. Van Dalsem
Editor, writer, photographer and designer: Judy Conlon
Associate editor: Sue Reynolds
Web designer and programmer: Sue Cox

Insights was published by the HPCC Program office and produced by Raytheon contractor staff at NASA Ames Research Center.