ESRL Quarterly Newsletter - Summer 2010

Pseudo Storms, Key Decisions

OSSEs mimic reality to help NOAA prioritize resources

It’s notoriously difficult to know when a hurricane is about to spin up into a monster, or when a storm will settle down a notch, sparing a coastal city of major damage. To predict those shifts in intensity and other changes, weather researchers and forecasters need better observations of hurricanes and storm environments. But what’s the most effective way to gather those observations?

Launch a new billion-dollar satellite? Buy a few multi-million-dollar unmanned aircraft systems, UAS, to gather observations? Or send instruments out over the ocean on relatively inexpensive balloons?

To help NOAA make such decisions, Yuanfu Xie, Nikki Privé and their colleagues are creating Observing System Simulations Experiments or OSSEs—worlds of make-believe observations and simulated weather. OSSEs are designed to help researchers plan future observing systems, and how to best use the data from those systems.

“This is what you do when you want to know how much impact a new observing system could have on numerical weather prediction—you set up an OSSE,” Privé said. “We are supporting NOAA’s UAS program (Unmanned Aircraft Systems) and others, to determine the best potential use.”

Privé, Xie, Scott Mackaro, and other ESRL Global Systems Division scientists are working with colleagues at the National Centers for Environmental Prediction, NOAA’s Atlantic Oceanographic and Meteorological Laboratory , AOML, NASA, and the European Center for Medium-Range Weather Forecasts to finish building two OSSEs this year: A global one, focused on improving hurricane track forecasts, and a finer-resolution regional OSSE, targeting hurricane intensity.

In the case of the global system, the interagency research team has already created a “nature run,” which represents global weather and atmospheric conditions for the entire year of 2006, at a resolution of about 40 km. The run includes 10 hurricanes and typhoons that formed between June and August, three of them near the United States.

It was tricky enough to create a complete and accurate representation of Earth’s weather for an entire year, but next came the task of creating synthetic observations from existing meteorology networks and proposed new ones. Creating observations is an exacting science, said GSD’s Steve Weygandt, who has worked on OSSEs for about a decade now. The simulated observations cannot be perfect—they need to be riddled with the same kinds of errors and missing data points that plague reality.

“Everything is in specifying the errors,” Weygandt said. “If synthetic observations from proposed new systems are too perfect, it’s very easy to get an overly optimistic result about the effect of a proposed observing system.”

Synthetic observations for the global OSSE will be complete this summer, Xie and Privé said, and then can be used to run weather prediction experiments toward the end of the summer. How skillfully does the model predict weather, given conventional observations? Does skill improve if measurements from a new observing system—say a system of UAS with dropsondes—are assimilated?

“We can do this because we know the truth,” Xie said. “We know what (weather) actually happens.”

“You can even use the models to tell you about where to collect observations,” Weygandt said. For example, which improves forecasts more significantly: Observations from near the center of a hurricane, or around the edges of it?

“Some really Interesting questions emerge about reality,” Privé said. In the global model nature run, one of the three hurricanes that swing near the United States passes directly over Cuba. Would a system of UAS airplanes—similar to military drones—be allowed to conduct atmospheric surveillance over that country?

Xie and his colleagues will ask researchers at AOML for guidance about which UAS to “fly” in the simulation, along what routes, and carrying what payloads. “Then we can ask, could we do a better job if we took a different route?” Xie said.

The regional OSSE is also nearly ready for experiments, Privé said, and although it is at significantly finer resolution than the global one, it has not yet been calibrated. The resolution of about 4 km, however, will allow researchers to investigate hurricane intensity changes (hurricane intensity forecasts have not improved in 20 years; hurricane track forecasts have improved 50 percent in that time). That OSSE will focus on three days in the development of a single storm: 2008’s Hurricane Ike.

Privé said she and her colleagues will be cautious when interpreting the results of the two OSSEs. “Both of these are really case studies,” she said. Results may speak to potential forecast improvement in the specific forecast models used: GFS/GSI in the case of the global OSSE; and HWRF/GSI in the case of the regional one.

Even given those limits, OSSEs represent the best way to evaluate expensive systems, without actually deploying them, Privé and Weygandt said.

“What else could we do? Fly a new satellite for two months and study the global impact of the observations on forecast skill and other statistics? What if it doesn’t help?” Privé asked.

Xie and Weygandt estimated that OSSEs—which can cost $1 million, given the many agencies, computing power, and people involved—cost a small fraction of a new observation system, whether a satellite or UAS-based observation network.