The role of convection parameterization in the simulation of the winter temperature and precipitation fields over Iran using Regional Climate Model (RegCM3)

Convection affects the climate through its role in the redistribution of energy and moisture in the atmosphere, and subsequently producing clouds and precipitation. Even with the recent improvements in computational power, the numerical models of the atmosphere still have to run in spatial resolutions that are too coarse to capture the local scale processes such as convection. For this reason, and because of the importance of convection for the surface climate, parameterization schemes have been developed to empirically upscale convection to the scale of the model grid areas.
This study aims at evaluating the impact of different types of parameterizing convection on the simulations by Version 3 of the Regional Climate Model (RegCM3: Dickinson et al., 1989; Giorgi, 1989) of precipitation and air temperature. The convection schemes currently coupled with RegCM3 are Anthes (1997), Betts (1986) and Grell (1993), which may be closed using either the Arakawa-Schubert (1974) or the Fritch and Chappell (1980) closure schemes. The simulations are conducted for the four-month period of December 1998 to March 1999 (inclusive) with 45 × 45 km grid spacing over a domain having 60 and 70 grid points along, latitude and longitude respectively and centered in Iran at 34 ˚E and 48 ˚N. The initial and boundary conditions are derived from the NCEP/NCAR reanalysis. RegCM3 was run four times, keeping all the components of the model and the initial and boundary conditions the same, by each time coupling one the convection schemes (Anthes, Betts, Arakawa, Fritch and Chappell) with the model. To minimize the impact of possibly incorrect initial conditions, we assumed one month as the model's spin-up period, and analyzed the results for the three months of January to March 1999.
The simulated monthly mean precipitation and air temperature as well as the spatial distribution of the model outputs using different schemes are intercompared and compared with observations from the Climate Research Unit (CRU) at the University of East Anglia, United kindom. The results show that the grid-scale mean monthly and mean seasonal (winter) near-surface air temperature simulated by RegCM3 coupled with different convection schemes agree very well with the corresponding observed values. The slope of the regression line of the simulated mean winter temperature against observations is very close to one and varies between 0.973 (Fritch-Chappell) and 0.997 (Kuo-Anthes), with the coefficient of determination (R2) in the range of 0.941 to 0.944, respectively. The degree of agreement between the simulated monthly mean temperature with observations for the three months is somewhat lower that of the mean seasonal with the slope of regression line varying between 0.908 and 0.939 and coefficient of determination between 0.935 and 0.938. It is concluded that RegCM3 is highly effectual in simulating air temperature, irrespective of the type of the convection scheme used. The differences between the simulated mean temperature using different schemes are very small.
On the other hand, the effectiveness of RegCM3 in simulating monthly and seasonal precipitation is very low. Differences between simulated monthly and seasonal precipitation using the four convection schemes are negligible. Although the geographical distribution pattern of precipitation is well simulated by the model, the simulated monthly and seasonal precipitation regressed against observation shows that RegCM3 generally underestimates precipitation during the winter months. The slope of the regression lines significantly differs from unity, varying between about 0.570 and about 0.715. The highest coefficient of determination found for the four schemes, during the three months and the season is smaller than 0.280.
Given the insignificant differences among the model simulations using any of the four convection schemes, the simplest form of convection parameterization with the lowest computational costs, i.e. the Kuo-Anthes scheme, proved to be most appropriate available scheme for medium-range weather predictions in Iran.