WRF Microphysical Adjustments with CloudSat
On the time and space scales of regional weather, accurate forecasts of cloud cover are required to predict the diurnal temperature cycle and likelihood of precipitation. Clouds and precipitation disrupt transportation networks, and in severe cases, may contribute to flooding, property damage or agricultural losses. Many of these problems may be alleviated through risk mitigation strategies, enhanced by accurate weather forecasts issued in the form of watches and advisories. Numerical models assist with the issuance of these operational forecast products. Gains in forecasting will come from improved simulation of clouds and their microphysical processes, achieved through steady increases in computer resources and forecast models that operate at cloud resolving resolution, rather than current convective parameterization schemes. Accurate short-term weather forecasts have a demonstrable benefit to society, but will also translate to the improved simulation of present and future climate, as global models transition to the use of cloud-resolving models in the form of superparameterizations. Improving cloud processes in operational, daily weather forecasts will translate to greater forecast skill on relatively short time periods, a primary goal of the SPoRT Center.
The NASA CloudSat 94 GHz Cloud Profiling Radar was launched, as a member of the A-Train of Earth Observing Satellites, in order to obtain vertical profiles of cloud layers and properties and builds on the robust heritage of ground-based 94 GHz profiling systems (Stephens et al. 2002). Data from CloudSat may be used to compare the properties of real clouds to their counterparts, as simulated within a numerical model. Although cloud resolving models offer a wide range of microphysics packages, CloudSat is currently being used to evaluate the performance of the Goddard six-class, single-moment microphysics scheme (Tao et al. 2008) as implemented within the Weather Research and Forecasting (WRF) Model. Due to the operating frequency of the CloudSat radar, the focus of current work is on cold-season, midlatitude cyclones producing light to moderate snowfall. Forecasts of these systems are not as dependent upon mesoscale processes and are well observed by synoptic-scale observation networks within the continental United States. These cyclones produce cloud cover and precipitation over multiple states, often leading to difficult forecasts for these high impact events. Within the WRF model, forecasts of precipitation and type depend upon the correct evolution and distribution of water mass among various hydrometeor classes. Meanwhile, forecasts of surface and profile temperatures depend upon diabatic processes in the form of latent heat exchange and the interaction of solar and terrestrial radiation with the modeled cloud shield.
Figure 1. Cross section of CloudSat 94 GHz radar reflectivity profiles obtained in Eastern Nebraska and Western
Iowa at 0830 UTC on 13 February 2007. Surface observations reported light to moderate snowfall with WSR-88D
radars also suggesting a northward decrease in reflectivity.
Toward the aforementioned goals, CloudSat observations have been searched to locate orbital segments containing observations of clouds and precipitation associated with cold-season midlatitude cyclones. These orbital segments are assumed to represent a distinct feature, such as clouds generated by warm frontal ascent (see Figure 1 above), so that a comparable feature may be examined within a WRF model forecast. Once the modeled feature is identified, representative model profiles are extracted and converted to an equivalent CloudSat radar reflectivity through application of the QuickBeam radiative transfer model (Haynes et al. 2007) . Properties of the observed and modeled clouds are compared through a contoured frequency with altitude diagram (see Figure 2 below, Yuter and Houze 1995), which depicts a probability distribution function of radar reflectivity at each altitude level. Deficiencies within the model forecast are noted, based on reflectivity characteristics. Preliminary work has focused on the snow crystal size distribution prescribed within the Goddard scheme. Currently, the Goddard scheme uses an inverse-exponential size distribution as in Gunn and Marshall (1958), where the intercept parameter is fixed. Other parameterization schemes have included an intercept that is temperature dependent, based on observational campaigns. Operating under the assumption that the modeled snow profile is reasonable, varying snow crystal size distributions are applied to determine which assumptions produce a better fit to CloudSat observations. This comparison effort is complicated by the remote sensing characteristics of the 94 GHz radar. At 94 GHz, oscillations in radar backscatter occur as the target diameter increases, so that an increase in target size does not consistently generate an increase in target diameter. In order to supplement CloudSat observations, the NWS Weather Surveillance Radar-1988 Doppler (WSR-88D) network is leveraged as a supplemental observation. The WSR-88D network is most sensitive to precipitation and operates at a frequency where reflectivity is more sensitive to increases in target diameter.
Figure 2. Comparison of CloudSat reflectivity CFADs. [Top] CloudSat observations. [Middle] Radar reflectivity
CFAD at 94 GHz, simulated from WRF profiles believed to be representative of CloudSat observations using snow
crystal distribution characteristics assumed within the Goddard scheme. [Bottom] As in the Goddard case but
simulating 94 GHz reflectivity using the distrbution characteristics of Brandes et al. (2007).
Observations by Brandes et al. (2007) of snow crystals in upslope Colorado snowstorms have suggested that the distribution slope parameter could be parameterized as a function of temperature. This size distribution has been implemented within the QuickBeam model and used in calculation of WSR-88D reflectivity. CloudSat and the WSR-88D network observed light to moderate snowfall to the northwest of a midlatitude cyclone on February 13, 2007. This system was simulated well by the WRF model, with only minor displacement of the simulated snowfall and cloud features, versus observations. When applying the snow distribution within the Goddard scheme, CloudSat reflectivity is underestimated in the lowest 3 km, while WSR-88D reflectivity is greatly overestimated. Application of the Brandes et al. (2007) distribution increases CloudSat reflectivity toward observed values, while narrowing the WSR-88D reflectivity distribution to more appropriate values and a mean profile that provides a better fit to observations (see Figure 2 above). Similar findings have occurred for two other cold-season cyclones, suggesting that there may be value in applying the Brandes et al. (2007) parameterization or a similar methodology. Simulated reflectivity will also be sensitive to the snow content within the vertical profiles, as well as any change in vertical distribution. It should be noted that there is no guarantee that the implementation of a different size distribution will produce comparable snow profiles.
Future work in this area will be targeted toward identifying additional case studies for model simulation and evaluation. Assuming that additional cases indicate similar, potential improvements from a Brandes et al. (2007) style of parameterization, this new distribution will be implemented within the WRF/Goddard scheme framework. There are also opportunities to investigate snow terminal velocities as an additional parameterization. Analyses based upon a new size distribution will examine changes to microphysical evolution of forecast clouds and their similarities to CloudSat and WSR-88D radar reflectivity.
