The Back Page

How's the Weather?

High-resolution weather forecasts from meteorological models are essential to air-quality predictions, which have become a routine component of daily urban forecasts nationwide. In a region such as Los Angeles, a high-resolution model is necessary to represent the small-scale phenomena that influence dispersion of pollutants (or chemical or biological agents). In addition to charting regional scale winds, the model must accurately capture the circulation of sea and land breezes and the convectively driven upslope circulation in the surrounding mountains. The accuracy of fine-scale, limited-area forecast models depends, among other factors, on the quality of boundary and initial conditions supplied to the model; the number, distribution, and quality of atmospheric observations; and the methods used to analyze them.

A sample forecast from www.aerospaceweather.com.

Aerospace has been conducting research geared toward improving fine-scale, real-time weather predictions over the Los Angeles basin through the optimal assimilation of space-based and local weather observations. One result of this research is a system that automatically issues daily 36-hour forecasts and posts them on a publicly accessible Web site, www.aerospaceweather.com. Aerospace is making these forecasts available to demonstrate the benefits of satellite weather data for air-quality monitoring and emergency response planning. The forecasts contain hourly predictions extending through late afternoon of the following day.

Scatter plots like this, plotting the observed temperature versus the predicted temperature, are used to validate and improve the forecast model.

A widely used weather prediction model known as MM5 has been configured to run with data assimilation and analysis software to automatically generate pseudo-operational daily weather forecasts at 5-kilometer resolution. The system analyzes operational data from the Air Force Weather Agency that includes surface weather reports and weather-balloon observations, aircraft reports, cloud-drift readings from geostationary satellites, and moisture and surface wind speed over oceans from the DMSP Special Sensor Microwave Imager. Additional buoy data are pulled from the National Buoy Data Center. Data are also automatically pulled from surface stations operated by the South Coast Air Quality Management District of California and the Bureau of Land Management as well as NOAA's Forecast Systems Laboratory boundary layer profiler and ground-based GPS receiver networks. To make an acceptable forecast over a domain with significant ocean area, the MM5 model also requires fairly accurate measurements of sea surface temperature, which are obtained from the Navy via the Air Force Weather Agency.

Sample snapshot of the predicted rain and surface winds from the animated loops available for each forecast. A winter rainstorm is approaching Los Angeles.

Particle dispersion simulation of a toxic plume release driven by MM5 winds over the Los Angeles area.

A multiprocessor Cray computer uses a three-dimensional variational analysis (3DVAR) program to assimilate observational data for two nested domains, one at 15-kilometer resolution, and one at 5 kilometers. The 3DVAR system was selected because of its ability to assimilate a variety of observations, including those that are not direct measures of the model state variables. The system characterizes observations by type, each with its own error statistics. Linearized observation operators relate the values of the model state variables at the analysis time to observed quantities. The goal is to specify the variables so as to minimize the difference between the analysis and observations and a prior estimate of the model state (known as the background).

Data assimilation occurs in three phases, spaced 6–7 hours apart. The information from the previous phase is used to enhance the background for the next phase. The time delay allows data to be culled from numerous sources, including satellites, which often have latencies of a few hours. The final data assimilation phase is used to generate a forecast out to 36 hours.

Verification software was developed to compare the MM5 output to observations. The program can compare temperature, relative humidity, dew point, mixing ratio, total precipitable water, wind speed, and wind direction. It performs separate comparisons for sounding and surface-station data. Verification scatter plots for a given model run are posted to the Web site two days after that run. Individual time series of model forecasts and corresponding surface observations for a number of air-quality monitoring sites are also posted.

Work is still under way to establish a continual data-assimilation cycle and to assimilate additional data sources. Researchers also hope to complete background error statistics and to optimize the model configuration based on feedback from verification efforts. The MM5 system may also be coupled with a more sophisticated land-surface model and draw upon high-resolution sea surface temperature readings. Additional plans call for deploying a transportable lidar system to locations in the Los Angeles basin to verify MM5 model forecasts and using MM5 wind forecasts in a multiple-particle dispersion model developed at Aerospace to predict the path of a toxic release.

To Winter 2005 Table of Contents