Regional mesoscale models are becoming the central tools for the operational forecasting of local weather systems and  quantitative precipitation forecasts (QPF) for periods of 0-48 h. During the past three decades, model resolution has continuously increased with advances in computer technology, and model parameterizations of physical processes has become more sophisticated. Even so, improvements in QPF have been comparatively slow. One reason for this, at least in regions of complex terrain, is that operational models are only now approaching the resolution necessary to resolve key terrain features that have a direct and significant impact on precipitation. Several recent studies (Bruintjes et al. 1994; Colle and Mass 1996; Gaudet and Cotton 1998; and Westrick and Mass 2000a) have shown that, when run at sufficiently high resolution (down to ~10 km), mesoscale models can realistically simulate observed precipitation structures over complex terrain. Yet significant systematic deficiencies in model precipitation can be found even when complex terrain is adequately resolved. For example, in studies of MM5 and Eta model forecasting of wintertime precipitation in the Pacific Northwest (Colle et al. 1999; Colle and Mass 1999; Westrick and Mass 2000b), model QPF was consistently too high on the windward side and too low on the lee side of orographic barriers. Therefore, increased resolution alone is insufficient to address the challenge of improving QPF.

With the exceptions of resolution and initial condition accuracy, the key model component that effects QPF is the parameterization of clouds and precipitation. State-of-the-art mesoscale models running at <10 km resolution now resolve most of the precipitation processes at the grid scale. Therefore, these models have increasingly relied on sophisticated bulk microphysical parameterizations (BMPs) (e.g. Cotton 1982; Lin et al. 1983; Rutledge and Hobbs 1983, 1984), which until recent years were used primarily in cloud resolving models. These schemes allow for the explicit prediction of cloud and precipitation mixing ratios based on a complicated array of empirically and theoretically derived sources, sinks, and exchange terms between the different hydrometeor types. In spite of their complexity, evidence of flaws in these schemes is ample (e.g. Manning and Davis 1997; Colle and Mass 1999). However, these flaws usually come to light only when the BMP is verified in an indirect way, such as by comparing forecast and observed precipitation or satellite cloud cover for many case studies. Such comparisons reveal that problems exist in the BMP, but they do not clearly point to the origin of those problems. The only way to clearly determine the source of problems in a BMP (and to correct them) is to comprehensively verify the microphysical assumptions and predicted mixing ratios in model simulations, using a combination of in situ (airborne) and remotely sensed (radar) microphysical observations. In addition, it is critically important that these observations be taken concurrently with observations of wind, temperature and humidity, so that errors in the simulated microphysics can be isolated from errors in these other predicted fields. To date, few dedicated efforts have been made to verify and improve BMPs in this manner. This fact has been recognized by both the Eighth and Ninth Prospectus Development Teams of the U.S. Weather Research Program (Fritsch et al. 1998 and Smith et al. 1999, respectively), which have placed high priority on observational verification of the parameterization of cloud and precipitation microphysics in numerical weather prediction models.

Specifically, the Improvement of Microphysical PaRameterization through Observational Verification Experiment (IMPROVE) is  aimed at comprehensively checking and improving the BMP currently implemented in a mesoscale model that has been extensively used for both research and operational forecasting by many groups—the Penn State/NCAR Mesoscale Model (MM5, Grell et al. 1995). Although the specific goal of IMPROVE is to improve QPF accuracy in the MM5, the results will carry over to QPF improvement in other national forecast models, such as the Weather Research and Forecast  (WRF) model now under development.

A unique opportunity exists at this time to carry out the studies outlined above. This is due to the convergence of several recent scientific and technological advances pertaining to numerical modeling, data assimilation, in situ cloud microphysical observations, and radar observations. In the case of in situ observations, there has been a recent breakthrough in our ability to obtain unambiguous quantitative information concerning the concentrations and size distributions of both liquid and solid cloud and precipitation particles, namely, the development of the Cloud Particle Imager (CPI; Lawson and Jensen 1998).  The CPI uses a pulsed laser to produce detailed 2-D images of cloud particles as small as 10 µm, allowing for the determination of the phase of such small particles. By comparison,  (PMS-type) digital optical imaging probes lack the resolution to determine the phase of particles smaller than about 200 µm (Korolev et al. 1998).

Secondly, recent advances in radar technology have enhanced the acquisition of wind and microphysical information. The first advance is the implementation of airborne dual-Doppler measurement via the fore- and aft-scanning technique (FAST; Frush et al. 1986; Hildebrand 1989; Jorgensen and Smull 1993). This observational method is capable of providing dual-Doppler data within an approximately 40-km wide volume centered along an aircraft flight path. During the COAST field project, the University of Washington (UW) was the first to successfully use this technique to derive 3D wind fields over complex terrain (Colle and Mass 1996). Another advance is the use of bistatic receiving antennas in conjunction with a ground-based radar for multi-Doppler velocity measurements (Wurman et al. 1993). This technique has already been successfully demonstrated in the field (Wurman 1994). Bistatic antennas are easier and more cost-effective to deploy and operate than are multiple radars, and they measure Doppler velocity components that are precisely simultaneous with those measured at the primary radar. Another significant advance in radar technology is the development of techniques to infer microphysical information from dual-polarized radar measurements (Doviak and Zrnic 1993). These techniques provide information on the dominant precipitation type in the sampled volume, which can be used to verify model outputs in a qualitative manner and to clarify the representativeness of the more quantitative (but spatially more limited) airborne in situ observations.

Finally, advances in computer power and mesoscale modeling make it feasible to run the MM5 mesoscale model in a real-time forecast mode at 4-km resolution during the course of a field study, and to make a large number of after-the-fact sensitivity test runs on field case studies down to 1.3 km resolution. The development of the variational method of 4D data assimilation (4DVAR, Lewis and Derber 1985; Courtier and Talagrand 1987) makes it possible to incorporate not only direct measurements of wind, temperature, pressure, and humidity, but also indirect measurements of atmospheric conditions, such as radar reflectivity and Doppler radar radial velocities (Sun and Crook 1997, 1998), satellite radiometer measurements (Andersson et al. 1994), and radio occultation data (Zou et al. 1995). Furthermore, the 4DVAR method can incorporate these data at non-synoptic times. The 4DVAR method adjusts the initial condition to maximize agreement between the model solution and observations during the forecast period, without requiring artificial nudging terms.  This technique will be used in IMPROVE research model simulations to maximize the accuracy of the simulated meteorological parameters (air flow, temperature, pressure, humidity), thereby isolating errors in the microphysics that are due to flaws in the BMP. The Atmospheric Sciences Department at the University of Washington operates a powerful 28-processor computer that runs twice-a-day 4-km mesoscale model forecasts for the local area, as well as frequent research simulations.

The primary goal of IMPROVE is to utilize comprehensive and quantitative measurements of cloud microphysical parameters in a variety of mesoscale frontal features to improve the representation of cloud and precipitation processes in mesoscale models. Therefore, IMPROVE field studies focused on precipitation systems that are relatively simple (in terms of dynamical mechanisms) and predictable, but which exhibit a variety of cloud microphysical processes that permit testing of a wide spectrum of forcing terms in bulk microphysical parameterization (BMP) schemes. To this end, two IMPROVE field studies, both carried out during 2001, focused respectively on frontal and orographic clouds and precipitation in the Pacific Northwest. In the winter, the Pacific Northwest is an ideal location to study precipitation systems both offshore and over orography, with numerous cyclonic storm systems making landfall from November through February.

IMPROVE-1, the Washington Offshore Frontal Field Study, was carried out off the coast of Washington State during the period 4 January--14 February 2001. The advantage of studying frontal systems over an oceanic domain with weak sea-surface temperature gradients is that they are driven by large-scale dynamical processes, which are typically well simulated in mesoscale models. Furthermore, because the lower boundary is spatially uniform, the structures of precipitation features can often be verified by observations in spite of modest position errors.

IMPROVE-2, the Oregon Cascades Orographic Field Study, was carried out in the Oregon Cascade Mountains during the period 26 November--22 December 2001.  Orographic precipitation systems are good candidates for IMPROVE studies because much of the forcing is tied to the terrain, and the terrain is precisely known. Thus, in situations where essentially steady flow impinges on a topographic barrier and the upstream conditions are known, the dynamical response to that flow is highly deterministic, provided the forecast model can properly resolve the key terrain-forced dynamics (Colle and Mass 1996). In addition, terrain-forced flow produces large gradients in cloud microphysical variables and processes, providing a good test bed for the model microphysics.

The Department of Atmospheric Sciences, University of Washington, is well suited to carry out both the field work and data analysis required to meet the goals of IMPROVE. In the 1970s the Department's Cloud and Aerosol Research Group (CARG), under the direction of Professor Peter Hobbs (an IMPROVE PI), carried out pioneering studies of orographic and frontal precipitation  systems in the CASCADE and CYCLES  Projects, respectively (e.g., Hobbs, 1973, 1975, 1978; Hobbs et al., 1971, 1975; Houze et al., 1976; Matejka et al., 1980). Rutledge and Hobbs (1983, 1984) advanced the understanding of warm frontal and narrow cold-frontal rainbands by studying CYCLES case studies with a BMP implemented in an idealized 2D kinematic model. Over a period of 32 years (1970-2002), the CARG instrumented and utilized a series of research aircraft for state-of-the -art measurements of clouds and precipitation.

The  COAST project (Bond et al., 1977), also carried out by IMPROVE PIs (Professors Clifford Mass, Robert Houze and Brian Colle), used airborne dual-Doppler data to define the mesoscale airflow within frontal systems encountering coastal topography, and used these data to verify mesoscale model simulations. However, COAST did not focus on obtaining concurrent in situ cloud microphysical measurements.

IMPROVE PI Professor Robert Houze and Brad Smull participated in the MAP project , which concentrated on airflow over the Swiss Alps and the processes that lead to flooding during the passage of baroclinic storm systems over the Alps. Dual-Doppler radar measurements were made by the NOAA P-3 and the NCAR Electra aircraft. The NCAR S-Pol radar data, together with vertically pointing S-band radar measurements, provided information on cloud microphysical processes. However, one of the shortcomings of MAP was a lack of in situ sampling of ice particles and raindrops just above and below the melting level.

Together, the conduct of the CASCADE, CYCLES,  COAST and MAP Projects, analyses of those data sets, and experience with the MM5 model, provide the PIs of IMPROVE with invaluable experience in the key observational  and analysis strategies needed to achieve the goals of IMPROVE.

References