Resolution dependence

In this subsection, the dependence of the simulated results to horizontal grid spacing is presented.

Fig. 4. The same as Fig. 3, except for the cases with Ax of (a) 250 m, (b) 500 m, (c) 1km, (d) 2 km, and (e) 4 km.

Figure 4 shows the temporal and height variation of dust concentration, upward flux of dust, and cloud boundary for the cases with various grid spacings. The diurnal features found in the control simulation can also be seen for the cases with the coarser grids except the 4-km grid case in which a later but more sudden development of upward dust flux is seen. In the coarsest grid case, the unexpectedly sudden and intense cumulus convection during 1400-1500 LT resulted in the highest value of column integrated dust content among all the cases examined here. The column dust contents at the end of the simulation period for the control and the 500-m grid runs are 0.63 gm-2 and 0.80gm~2, respectively, while the content for the 4-km run is 1.1 gm~2. It should be noted that the column content obtained by the control run is consistent with the estimation by satellite remote-sensing data.20

The resolution-dependence is further demonstrated by examining the difference of boundary-layer development. The PBL activity is diagnosed in terms of TKE that is computed in the SGS turbulence closure scheme.14'19 Figure 5 compares the diurnal variation of the boundary-layer development for the cases with Ax = 250 m, 1km, and 4 km. In the 250-m grid case, the PBL motion starts to intensify at 0900 LT, and the boundary layer

Fig. 5. The same as Fig. 4, except for turbulent kinetic energy (contoured at 0.5 m2 s 2).

smoothly deepens, which leads to cumulus development at 1100 LT as seen in Fig. 4(a). A similar development can still be identified for the 1-km grid case. Note that the TKE values in Figs. 5(a) and 5(b) are significantly larger than those in the 100m run (see Fig. 2(b)); this is due to the enhanced eddy viscosity, which takes into account the effects of nonlocal mixing in the sensitivity runs. On the other hand, although an increase of boundary-layer activity is seen for the 4-km grid case, the increase is much slower; thus, the cumulus development does not occur before noon and is significantly retarded. Considering that both boundary-layer and cumulus convection play a critical role in enhancing dust emission and transport,6 the slow boundary-layer development in the coarsest grid case seems to be a reason for the significant difference in the dust transport from the finer-grid cases.

These results indicate that the grid spacing of 4 km, which is well in the range of explicit representation of deep convection in mesoscale systems,21 is not sufficient to resolve shallow and deep convection under the present fair weather condition. In particular, the scales of updrafts for shallow convection are typically smaller than 4 km. The 4 km grid spacing is actually arguable, because there is no robust, satisfactory solution for convection and turbulence parameterization in the simulations.22 The 4-km mesh, on the other hand, is obviously not coarse for regional-scale (let alone global-scale) simulations of convective dust transport, and hence most studies on atmospheric transport aim at performing simulations with grid spacings of 1-10 km; therefore, a proper parameterization for activating convection that induces dust emission and transport is necessary in this range of grid spacing.

One possibility for the better representation of the processes is to include a cumulus parameterization in the 4-km simulation. We further performed a 4-km simulation that included either the Kuo or the Kain-Fritsch cumulus parameterization built in the ARPS model,15 and found that the boundary-layer development was not improved. This result is reasonable, since those parameterizations are not suitable for parameterizing both shallow and deep convections. Another possibility is to improve the TKE prediction by enhancing the production terms in the TKE closure equation; however, this requires a detailed analysis of a turbulence simulation data set. In the following, we examine a simple representation for facilitating the development of both boundary-layer and cumulus convection and hence enhancing dust production.

0 0

Post a comment