IMPROVE is using the Fifth Generation Penn State / NCAR Mesoscale Model (MM5, Grell et al. 1994) as the modeling test bed. The MM5 is a nonhydrostatic sigma-coordinate model with multi-level nesting capability and physical parameterizations for clouds and precipitation, surface properties, boundary layer processes, and radiative transfer. Real-time forecast runs with the MM5 are currently being conducted on a daily basis on the University of Washington’s 14-processor Sun server. The present configuration is a twice-per-day 48-h forecast on a 36-km domain that covers most of the northeastern Pacific Ocean and part of western North America, a 12-km nest that covers Washington, Oregon, and offshore waters, and a 4-km nest that covers western Washington. This configuration was used during the field project, with an additional 4-km domain added to focus on the IMPROVE frontal and orographic study areas.
The MM5 includes two key components that are essential for IMPROVE. The first is a choice of several sophisticated bulk microphysical parameterization (BMP) schemes: the Reisner scheme (Reisner et al. 1998), the NASA-Goddard scheme (Tao and Simpson 1993), and the CSU-RAMS scheme (Cotton 1982)]. In the past 15 years, the MM5 modeling system has undergone a progressive improvement in terms of its BMP. In the mid-1980s, the MM4 simulated only warm rain processes (Hsie et al. 1984). Non-mixed-phase cloud ice and snow were added by Dudhia (1989). Reisner et al. (1998) implemented a mixed-phase BMP similar to those found in state-of-the-art cloud-resolving models, with five hydrometeors categories (cloud water, cloud ice, rain, snow, and graupel), as well as predicted number concentrations of cloud ice, snow, and graupel. The Reisner BMP is being used as the primary test bed for IMPROVE. The other sophisticated BMPs that have recently been added as options in the MM5 (the NASA-Goddard and CSU-RAMS schemes) will also be evaluated in an attempt to determine which components of these two schemes consistently produce the best results.
The second key component is an adjoint version that allows for 4DVAR data assimilation. Zou et al. (1995) developed an adjoint version of the MM5 which NCAR has made available to the research community. The first step in making accurate simulations of the observed cases is to produce the best possible initial condition using the 4DVAR system. It will be run at a coarse resolution (36 km) with simplified microphysics for periods of 12 h—these constraints are necessary to account for the large computational demand of the 4DVAR system. The initial conditions produced by this process will then be used to initialize higher resolution model simulations with sophisticated microphysics. The University of Washington has conducted experiments using the adjoint/4DVAR version of the MM5 to refine the initial conditions for the UW’s real-time MM5 simulations using only standard surface and upper-air data, with positive results.
Data from a variety of sources are being assimilated, including all available surface and rawinsonde observations, commercial aircraft (ACARS) data, and cloud-tracked winds from satellite imagery. Special project data are being used, including soundings, research aircraft observations of wind and temperature, and Doppler radial velocity measurements. Although a 4DVAR system has already been developed for the MM5, additional programming work is being carried out to incorporate the various data types into the 4DVAR system. The data assimilation method does not guarantee a perfect simulation, so rigorous verification against all available data, especially dual-Doppler wind fields and in situ wind and temperature data, is being carried carried out.
The important step in the verification process for each IMPROVE case study is to confirm that the model is capturing observed airflow, thermal, and moisture structures. After confidence is gained in these aspects of the simulations, the microphysical aspects of the simulations will be evaluated. Simulated reflectivity fields will be calculated and checked against radar-observed reflectivities, and model precipitation fields will be compared with rain gauge data. Although rain gauge data is not available offshore for the IMPROVE frontal study, the Convair-580 measured microphysical variables very close to the ocean surface. Predicted mixing ratios and particle number concentrations for each hydrometeor type will be verified by aircraft measurements, and the assumed particle size distributions will be tested. Over a larger volume of the cloud, dominant precipitation types will be derived from S-Pol polarimetric radar data.
After the MM5 model simulations have been compared to observations, and key discrepancies in the simulated microphysical fields have been identified, sensitivity tests will be performed on various assumptions and components of the BMP to improve the simulated microphysics. Many aspects of BMPs have been the subject of recent testing in mesoscale modeling studies, including the representation of intercept parameter for the snow size distribution (Reisner et al. 1998); collection efficiencies, fall speeds, and particle density for graupel and snow (Brown and Swann 1997); cloud ice sedimentation rate (Manning and Davis 1997); ice initiation (Manning and Davis 1997); ice multiplication rate (Brown and Swann 1997); and snow fallspeeds (Colle and Mass 1999). However, these studies were limited in that they were either confined to a small number of cases and microphysical situations, or they did not have in situ cloud microphysical measurements for verification of the model. IMPROVE will rigorously test these and other components of the BMP in a large number and variety of cases. Other issues IMPROVE is addressing are the added benefits of additional predicted variables, such as number concentrations of hydrometeors or new hydrometeor species (such as drizzle), and the importance of continental versus maritime cloud droplet number concentrations.
Tests of sensitivity to all of these processes are being performed in IMPROVE, with emphasis on making adjustments to the BMP that will be widely applicable to a variety of microphysical scenarios, rather than tuning the BMP for specific situations. Another important aspect of the BMP testing and improvements to model algorithms is an emphasis on cost-effectiveness in terms of CPU time. In the operational forecasting environment, where computers are pushed to the limit to produce timely forecasts, any improvements in the physical parameterizations must be weighed against the CPU "price tag". Ultimately, the improvements to BMPs that derive from IMPROVE will face their most stringent testing in the operational environment, where QPF improvement can be measured over many seasons or years.