Testing of new parameterization schemes in collaboration with outside users on MM5 has continued. New radiative, boundary-layer, cumulus, microphysics, and land-surface parameterizations have undergone testing by Jimy Dudhia, Anyu Wang (visitor, Zhongshan University), and Yubao Liu (visitor, Chinese Academy of Meteorology), in collaboration with Pennsylvania State University, NOAA, ETL, and NASA, Goddard, and will be made available to the user community in 1996. Work by David Gill, William Skamarock, and Wesley Jones (Silicon Graphics Inc., SGI) has continued to allow MM5 to run efficiently on workstation platforms, and (in collaboration with the Argonne National Laboratory) on massively parallel machines. David Hart and Gill have developed a capability for displaying model output with a three-dimensional graphics package (Vis5D).
A workshop and a series of meetings have been called to consider the next-generation model, MM6. Experts from NCAR's Scientific Computing Division and the Argonne National Laboratory have been consulted on programming aspects for a new model that would need to be portable and efficient on a variety of computer architectures and to be state-of-the-art for the year 2000. A basic outline for a new code has been formulated to meet these needs.
Mark Stoelinga (visitor, University of Washington) and Jordan Powers (ASP postdoctoral visitor) developed a coupled environmental modeling system comprised of the MM5, the Princeton-Geophysical Fluid Dynamics Laboratory (GFDL) Ocean Model, and the Great Lakes Environmental Research Laboratory (GLERL)-Donelan Wave Model. The coupled model is being used to investigate the effects of incorporating high-resolution atmospheric data on Lake Erie circulation and wave forecasts and of incorporating detailed sea-state and heat-flux data on atmospheric forecasts. The system relies on PVM (Parallel Virtual Machine) code for linking the model components, and can be run on separate machines for computational load distribution. The system can derive and output a multitude of diagnostic fields during a simulation, thus enabling real-time analysis of the forecasts.
Simon Low-Nam, Wei Wang, and Gill, in collaboration with Alexis Lau (visitor, Hong Kong University Institute of Science and Technology, HKUST, joint appointment with RAP), working on the Hong Kong airport project, have developed a stand-alone MM5 system to provide short-term predictions of the environmental conditions at the new airport site. The system is relocatable and can be operated on a high-end workstation platform. The project involves close collaboration with other groups within NCAR (including RAP). The MM5 model will provide the prediction of vertical profiles of temperature and wind fields up to three hours to the Terrain Induced Windshear and Turbulence algorithm, which is currently under development. The MM5 group has completed the first phase of the training of the Royal Observatory personnel in the use and maintenance of the system. Training of the HKUST personnel is also being undertaken in order to facilitate the technology transfer to Hong Kong. The MM5 work has been reported in the "Feasibility and Concept Development Study, Phase II" (1995) document recently submitted to the Royal Observatory.
Verification of 10-day mesoscale MM5 simulations, centered on the DOE Atmospheric Radiation Measurement (ARM) Cloud And Radiation Test-Bed (CART) site in Oklahoma/Kansas, by Dudhia, revealed a persistent underestimation of the daytime surface temperature that was traced to overestimation of cirrus cloud cover in the model. Improved representation of the diurnal heating was achieved by including ice sedimentation effects that produce a more realistic lifetime for cirrus clouds generated almost daily over the Rocky Mountains in the summer and advected to the Southern Great Plains.
MM5 simulations of an Oklahoma cold front at multiple grid resolutions from 180 km to 2.22 km have been carried out by Dudhia to investigate the differences between parameterized and resolved convective systems. The Grell and CCM2 convective schemes have been tested. A tendency for coarser-resolution simulations to underpredict convective mass flux has been found for this case, and this is attributable to the lack of parameterized convection in a stably capped region just ahead of the cold front where convection develops in the high-resolution simulations. It would require a modification of the trigger function in parameterized convection to correctly represent this case at coarse resolution.
Kevin Manning has conducted a statistical study of the performance of MM5 during the Winter Icing and Storms Project 1994 (WISP94) real-time forecasting exercise using satellite and radiosonde data. The satellite data clearly showed that during WISP94, MM5 produced far too much high cloud (also noted by Dudhia, see above). The radiosonde statistics revealed a cold bias throughout the troposphere and moist biases at low (~800 mb) and high (~350 mb) levels. Sensitivity studies of several cases have demonstrated that the extensive high cloud is a result of too many small ice crystals being generated at low temperatures according to Fletcher's (1962) equation, and that simple modifications to the microphysics scheme can greatly improve the forecast of high clouds. These same modifications reduce the moist bias at 350 mb. A more complex radiation parameterization than was feasible for use in real time reduces the cold bias throughout the troposphere and improves the temperature forecast, but only if the forecast of high clouds is reasonable. A cold bias and a moist bias at low levels remain. The cold bias at night and the moist bias (night and day) may be due to excessive mixing by the boundary-layer parameterization, which would tend to produce too deep a boundary layer and transport too much moisture away from the surface to the top of the boundary layer. The cold bias during the day may be due to insufficient surface heating by solar radiation, but the cause of this is not currently known.
Xiaolei Zou led a major effort to develop a four-dimensional variational data assimilation system based on the adjoint of the MM5 model, with significant contributions by Yong-Run Guo and Wei Huang (visitor, Peking University). Over the last year, adjoints of several physical parameterization schemes have been developed, including a Kuo-type cumulus parameterization scheme, the Grell cumulus convection scheme, a grid-resolvable-scale nonconvective precipitation parameterization, a cloud microphysical parameterization, a surface-energy budget, atmospheric radiation, and a high-resolution Planetary Boundary Layer (PBL). Effort was also put into the multitasking and parallelization of the MM5 adjoint model. The parallelization was shown to save up to 50% in computing resources.
Zou continued a collaboration with John Derber and Joseph Sela (National Center on Evironmental Prediction, NCEP, formerly the National Meteorological Center, NMC), on the development of a four-dimensional variational data assimilation using the full-physics NCEP operational global spectral model and its adjoint. The efforts during the past year have focused on the implementation of the adjoint of several model physical processes, including atmospheric radiation and moist physics. We intend to use this data assimilation for Global Positioning System (GPS) data impact assessment in the near future. The results of the four-dimensional variational real-data assimilation using adjoint models including partial and full physical processes will also be compared with the results from the perturbation method, where a full nonlinear trajectory and an adiabatic adjoint model are used in data-assimilation experiments.
An important problem for the development of an adjoint for moist physics is the existence of on-off switches for model physical parameterization. Zou first performed a theoretical analysis based on idealized examples. She showed that if on-off switches are kept the same as in the nonlinear model for the basic state around which the linearization is performed, a sufficient condition for the tangent linear model (TLM) and adjoint model to provide accurate gradient estimates is that the perturbation solution of the nonlinear model or the tendency equation be continuous at switching points. Otherwise, the accuracy of the tangent linear model and adjoint models depends on the relative magnitude of the jump caused by on-off processes with respect to other terms in the tendency equations. Zou subsequently performed numerical experiments using the MM5 adjoint with moist physics, and showed that such errors caused by on-off switches do not seem to impact the validity of the TLM and the adjoint model. She found that the presence of on-off switches increased the nonlinearity of the system more than that of the loss of accuracy in the TLM and adjoint model. The former makes the TLM solution deviate from the true perturbation solution after a few hours of model integration.
Zou, Ying-Hwa Kuo, and Guo performed a series of observing-system simulation experiments (OSSEs) to assess the impact of GPS measurements on Numerical Weather Prediction (NWP), using the adjoint of an adiabatic version of the MM5 model. They showed that the assimilation of atmospheric refractivity using the four-dimensional variational data assimilation (4DVAR) method is very effective in recovering the vertical profiles of water vapor. The accuracy of the derived water-vapor field is significantly better than that obtained through the traditional retrieval technique. They also found that the impact of GPS atmospheric refractivity data depends critically on the data density.
In an effort to more realistically incorporate the measurement characteristics of the GPS refractivity data, Zou and Kuo carried out another series of OSSEs using a hemispheric version of the MM5 model with a full-physics adjoint. They showed that it is not necessary that the observations of refractivity be on a regular grid. Assimilation of randomly distributed refractivity data with a realistic averaging length provides valuable meteorological information. The analysis enhanced through GPS data assimilation led to an improved prediction of an intense marine cyclone.
Kuo, Zou, and Guo carried out a series of numerical experiments using a 4DVAR system based on the NCAR/Pennsylvania State University (PSU) mesoscale model, MM5, and its adjoint. The assimilation of the precipitable water data obtained from the Severe Environmental Storms and Mesoscale Experiment (SESAME) 1979 special 3-h soundings was shown to recover the vertical structure of water vapor and to improve the quality of moisture analysis. The improved moisture analysis led to significant improvement in short-range precipitation forecasts. They found that including moist physics in the 4DVAR system reduced the systematic biases of the model, allowed a better fit between the model and observed data, and resulted in improved assimilated fields, and consequently, a better short-range prediction.
In order to assess the impact of rainfall data on short-range precipitation forecasts, Zou and Kuo performed a series of 4DVAR experiments using the 3-h rainfall data collected in the Wichita Falls tornado case of SESAME 1979. They showed that the assimilation of observed rainfall data significantly improved the position and intensity of model rainfall forecasts and notably reduced the occurrence of predicted rainfall that was not observed. The positive impact of rainfall assimilation extended 12 h beyond the assimilation period. They also showed that the assimilation of precipitable water is highly valuable. The assimilation of precipitable water constrains the large-scale model error growth in the moisture field, while allowing the model to develop mesoscale structure through rainfall assimilation.
Over the last year, Juanzhen Sun and N. Andrew Crook (joint appointment with RAP) continued to develop a variational four-dimensional data analysis system using a cloud model and its adjoint to assimilate observations from single or multiple Doppler radars. By fitting the model solution to radial velocity and reflectivity observations, an atmospheric state containing wind, thermodynamic, and microphysical information is obtained. These fields are used for either diagnostic studies or model initialization.
The technique was further tested against observations of moist convection in the troposphere. Using observations from CaPE (Convection and Precipitation/Electrification Experiment), a case involving several isolated storms generated along the boundary of a sea breeze was studied to evaluate the technique in retrieving microphysical data. Reasonable structures of the convective storms were obtained using information from two radars. Single radar retrieval degraded the quality of the data, but not significantly.
The adjoint variational data assimilation technique for retrieval of the thermodynamic fields was compared with the traditional technique of Gal-Chen and Hane using simulated data of a collapsing cold pool. In most of the cases examined, it was found that the adjoint method was able to retrieve the buoyancy field more accurately than the traditional technique. In the future, the data assimilation technique will be further developed and tested using supercell and multicell storm data.
In a cooperative effort with Roy Rasmussen (joint appointment with RAP), Edward Szoke completed a preliminary evaluation of the WISP94 nowcasting effort in support of research on conditions requiring ground de-icing of aircraft. The experimental nowcasting supported the United Airlines ground de-icing program at Denver which also involved an advanced weather display. A total of 12 nowcasting events took place over the two-month WISP94 period, with seven separate mechanisms identified that produced mesoscale structures in the snowfall. While high-resolution numerical simulations with MM5 sometimes identified the structure that occurred, Doppler radar was the most effective nowcasting tool. However, the radar was most effective (in terms of a forecast for an hour or more) when the mechanism producing the snowfall could be identified fully; otherwise nowcasting often reverted to simple extrapolation.
Szoke continued the cooperative efforts with NOAA, Forecast Systems Laboratory (FSL), and the Denver NWS Forecast Office as part of the Experimental Forecasting Facility (EFF). He worked shifts for both public and aviation forecasting, releasing operational forecasters to pursue joint research projects designed to improve the interactions between research and operations. Szoke also engaged in several of these cooperative research projects.