Abstract

A new urban parameterization for a fast-running dispersion prediction modeling system suitable for emergency response situations is introduced. The parameterization represents the urban convective boundary layer in the dispersion prediction system developed by the National Atmospheric Release Adv... Show moreA new urban parameterization for a fast-running dispersion prediction modeling system suitable for emergency response situations is introduced. The parameterization represents the urban convective boundary layer in the dispersion prediction system developed by the National Atmospheric Release Advisory Center (NARAC) at Lawrence Livermore National Laboratory. The performance of the modeling system is tested with data collected during the field campaign Joint Urban 2003 (JU03), held in July 2003 in Oklahoma City, Oklahoma. Tests were performed using data from three intense operating periods held during daytime slightly unstable to unstable conditions. The system was run in operational mode using the meteorological data that would be available operationally at NARAC to test its effectiveness in emergency response conditions. The new parameterization considerably improves the performance of the original modeling system, by producing a better degree of pattern of correspondence between predictions and observations (as measured by Taylor diagrams), considerably reducing bias, and better capturing directional effects resulting in plume predictions whose shape and size better resemble the observations (via the measure of effectiveness). Furthermore, the new parameterization shows similar skills to urban modeling systems of similar or greater complexity. The parameterization performs the best at the three JU03 sensor arcs (1, 2, and 4 km downwind the release points), with fractional bias values ranging from 0.13 to 0.4, correlation values from 0.45 to 0.71, and centered root-mean-square error being reduced more than 50% in most cases. The urban parameterization has been tested with grid increments of 125, 250, 500 and 1000 m, performing best at 250 and 500 m. Finally, it has been found that representing the point source by a Gaussian distribution with an initial spread of particles leads to a better representation of the initial spread induced by near-source buildings, resulting in lower bias and improved correlation in downtown Oklahoma City. Show less