|Most daily weather forecasts still don't reliably predict how much rain will fall on a given day. NCAR scientists are part of a national program to improve estimates of rainfall from radars and computer forecast models. With UCAR input, the National Weather Service (NWS) is exploring how to improve predictions and translate them into probabilities useful to emergency managers. And NCAR technology is helping to get a better fix on how much water vapor resides in the atmosphere--a first step toward knowing how much of it might fall on us.|
Forecasters at the U.S. Weather Bureau (forerunner of today's National Weather Service) in Roswell, New Mexico, conduct one of the first experiments to classify rain chances in terms of probabilities. The predictions showed about one-fifth the skill of a perfect forecast.
|The signals from NCAR's S-Pol radar can be separated into horizontally and vertically polarized components--valuable for distinguishing rain from hail.|
Sometimes a degree can make all the difference. When the atmosphere is poised for action, one degree Celsius (1.8 degrees Fahrenheit) can produce a downpour out of a clear, quiet afternoon.
NCAR scientist Andrew Crook has shed new light on this truth long recognized by forecasters in the Great Plains. Hot air often flows eastward from the desert Southwest in spring and summer, forming a cap several thousand feet above the prairie. The cap allows warm, moist air to simmer below. Sometimes the surface air heats up enough to "break" the cap, and thunderstorms explode into being. Other times the afternoon heat isn't enough, and paltry rain clouds choke as they hit the cap's dry air.
Using computer simulations, Crook examined the vertical profiles of temperature and moisture used by forecasters on a series of days across the storm-prone plains. He found that the typical uncertainties in measuring temperature and moisture variations--about 1 degree C or about one part water vapor per thousand parts of air--were often enough to mask the ability of the lower atmosphere to break its cap and form storms. The implication: Even the best meteorologist may be unable to tell farmers whether they'll get a sprinkle or a deluge on a given night.
Few problems have been so persistently vexing to forecasters. It's been more than thirty years since the National Weather Service added precipitation probabilities to its regular daily forecasts. While these indicate how much of a region should be affected by storms, they don't reveal the amount of rain that might fall. One reason is that showers and thunderstorms occur on a smaller scale than that mapped by routine observations and computer forecast models.
Computers are now poised to take the next step--assigning rainfall amounts to a range of probabilities. UOP's Cooperative Program for Meteorology, Education and Training (COMET) is teaming with the NWS to test this notion. COMET trains thousands of NWS forecasters each year through residence courses in Boulder and, through CD-ROM modules and Web sites, at their home offices. COMET also provides seed money for cooperative research, and probabilistic quantitative precipitation forecasting (PQPF) is a major theme in COMET's recent training and research.
What does PQPF look like? As part of a COMET-NWS pilot study conducted with the University of Virginia, an automated system rated the odds of rain in the Upper Allegheny River Basin for a 24-hour period (October 19 1996) as follows:
An array of numbers like this might be too much for sleepy viewers of a late-night weathercast, but the probabilities are potential gold to emergency managers, farmers, and other users. For example, a high probability of heavy rain (say, an inch or more) could change how hydrologists, road crews, or park rangers operate on a given day.
To strengthen the science behind such forecasts of the future, a number of NCAR scientists are part of a national initiative, the U.S. Weather Research Program. One of the USWRP's main goals is to improve quantitative forecasts of rainfall and snowfall. In a USWRP report, a team led by Michael Fritsch (Pennsylvania State University) writes, "What people need to know most about the weather is, Will it rain or snow, and if so, how much?" Moreover, says Richard Carbone, an NCAR senior scientist who is serving as the science lead for the USWRP, "Floods are the most costly weather-related disaster for the United States and most other nations."
Like any great artist, nature needs good material to work with. A thunderstorm may convert as little as 10% of its moisture into rainfall, so plenty of water vapor is needed for heavy rains to occur. But adequate, cost-efficient measures of the vertical distribution of water vapor are currently lacking.
"Knowing just the surface humidity isn't enough. Flash floods tend to happen most often when the whole atmosphere is moist," notes NCAR scientist David Parsons. "However," he adds, "most current techniques for sensing water vapor remotely are very expensive and often labor-intensive." Though limited in space and time, radiosondes (twice-daily weather balloons) are still the main way to gauge water-vapor levels at a variety of heights on a routine basis. Laser-based radars (lidars) can sense water vapor quickly at high precision, but they are pricey, and--ironically--they are hobbled when thick clouds and rain are present. Weather satellites provide useful measurements over oceans, but they have difficulties in the atmosphere just above land areas.
A UOP/NCAR team is working on better three-dimensional representations of water vapor through a multiyear USWRP grant and support from the U.S. Department of Energy's Atmospheric Radiation Measurement program. The team has done an end run around some of the above problems with inexpensive new tools that make use of the Global Positioning System (GPS) array of navigation satellites. A GPS receiver on the ground experiences a tiny delay in receiving signals from space as the signals are refracted and thus slowed down by water vapor. This delay can be used to infer an important value: how much water vapor is present in a vertical column above the receiver. The numbers don't provide the vertical distribution of a radiosonde's output. However, the data are dependable (unhindered by clouds and rain) and frequent (every half hour), and the GPS receivers can be deployed widely at low cost.
UOP's Teresa Van Hove derived water vapor fields from GPS data taken at 14 sites across Oklahoma and Kansas in a 1997 test. Researchers then combined the water vapor fields with other high-precision data in a complex technique called four-dimensional variational data assimilation. NCAR's Yong-run Guo and colleagues found that this technique improved the initial picture of water vapor, temperature, and wind used in short-term forecast models. According to Parsons, the next step is more simulations to learn "exactly why we're getting the improvements," followed by a field experiment over a larger area to examine changes in forecast skill.
"GPS networks, infrastructure, and technology are evolving rapidly," says Christian Rocken, head of GPS meteorological research at UOP. "I would not be surprised if data from hundreds, if not thousands, of GPS sites are routinely used for numerical weather forecasts in just a few years." Under the guidance of Michael Hardesty (NCAR/National Oceanic and Atmospheric Administration) NCAR is working with NOAA and the National Aeronautics and Space Administration (NASA) to develop and test a low-cost lidar that, like the GPS technique, might be used routinely in the decades to come.
When it comes to very short term forecasts (nowcasts) of rainfall, radar is the tool of choice. The recently completed NWS network of Doppler radars (WSR-88D) has a precipitation-estimation feature that maps cumulative amounts during a storm. A team led by NCAR's James Wilson, Rita Roberts, Andrew Crook, and Cindy Mueller is pushing the radar's skills into the near future through the NCAR Thunderstorm Auto-nowcaster. This software-and-hardware package extrapolates radar data and adds automated scientific reasoning to predict storm initiation, growth, and dissipation for precise locations in a 30- to 60-minute forecast window. In the summers of 1997 and 1998, the Auto-nowcaster went through tryouts at the NWS forecast office in Sterling, Viginia, and at the U.S. Army's White Sands Missile Range.
|Precipitation research brings together Christian Rocken, David Parsons, Michael Hardesty, and Yong-run Guo.|
To help improve rainfall estimates, the NWS may add dual polarization to its Doppler radars. Groundwork toward this end is being laid in a USWRP project led by NCAR's Edward Brandes, Jothiram Vivekanandan, Roy Rasmussen, and Wilson. NCAR has its own dual-polarization radar, dubbed S-Pol, that can be easily dismantled and shipped to research projects that place S-Pol side by side with the WSR-88D for easy comparison. Previous work has shown that rainfall estimates made from Doppler radar reflectivity have large biases that vary from storm to storm. In certain storms they overestimate actual rainfall; in others they underestimate it. The biases seem to be rooted in differences among rainstorms, especially in the size distribution of their raindrops. Polarized radar signals show promise in providing a quasi-independent estimate of rainfall that is less prone to bias.
"The road from pilot research to an operational system is still a long one, but it has been charted," wrote Fritsch and colleagues in their 1998 USWRP report. It may be years before the public hears anything like "a 50% chance of an inch of rain" in their daily forecast. But with research proceeding at full steam, it won't be long before such numbers are available to decision makers--the people who control how much we pay for food, whether an afternoon's baseball game is called off, or when we need to leave our homes for higher ground.
On the WebU.S. Weather Research Program
UOP/Cooperative Program for Operational Meteorology, Education and Training/Outreach Program
UOP/GPS Science and Technology Program
NCAR/Atmospheric Technology Division/S-Pol radar