4. Sensitivity tests

In this section, the sensitivity of the results to various aspects of the model setup is explored. Using the default configuration (see section 2b) as the control, the July-August 1989 pair of simulations was repeated with three alternative model configurations. The first alternative used the Chen-Cotton (1983) radiative transfer scheme, and the second used the Kuo convective parameterization. The third alternative used the default physics options, but at a 40-km grid spacing. Using both the default physics and grid spacing, three additional experiments were conducted to examine the impact of initial soil moisture and the distribution of SST. The first experiment used alternative initial soil moisture, the second an alternative SST distribution, and the third included both of these alternatives.

The factors explored in this analysis were chosen because they were thought to be of first-order importance to the nature of the model fields examined in this study (e.g., the surface energy budget and convective precipitation). However, these factors are by practical necessity a limited subset of those that could impact the results. Numerous other aspects of the model setup could potentially influence the results. For example, these simulations did not employ explicit cloud microphysics, because the resulting precipitation totals with use of a convective parameterization alone appeared adequate to capture the magnitude of the observed rainfall. For the sake of brevity, only the rainfall difference fields produced from the two land-cover scenarios within a given set of sensitivity simulations are shown (Fig. 20). In the case of experiments concerning soil moisture and SST, only the difference field wherein both factors are incorporated is provided. However, a histogram is provided that shows the grid-average rainfall (for both landcover scenarios) for all four combinations of soil moisture and SST experiments (Fig. 21).

Fig. 20. Difference field of accumulated precipitation (mm; 1993 minus pre-1900) from the simulations for Jul-Aug 1989 incorporating (a) the Chen-Cotton radiation scheme, (b) the Kuo convection scheme, (c) default physics at 40-km grid spacing, and (d) default physics, with weekly observed SST and VIC model initial soil moisture. [click on each of the images to view a larger version]

Fig. 21. Histogram of grid-average accumulated rainfall (mm) of all possible combinations of SST and soil moisture sensitivity experiments with the Jul-Aug 1989 simulations for both land-cover scenarios. [larger image]

Much of the convective rainfall over the Florida peninsula during July and August is associated with circulations that are forced by spatially varying surface properties—particularly the sea breezes. Thus, it is reasonable to expect that the parameterized radiation, which provides the primary forcing for the surface sensible and latent heat fluxes, could have an impact on the nature of those circulations, and hence the distribution of convective rainfall. The Mahrer-Pielke (1977) scheme (the default) accounts for the presence of water vapor, but it does not explicitly account for cloud liquid water or ice in determining the radiative transfer. However, during the typical diurnal scenario, circulations such as the sea breezes are often well developed before the onset of significant convective cloudiness and rainfall. Thus, it is suggested that the Mahrer-Pielke scheme, which is desirable because of its computational efficiency, is adequate for the simulations presented in this study. Nevertheless, it is prudent to examine the impact of using an alternative, such as the Chen-Cotton scheme, that does incorporate the effects of cloud water and ice. As Fig. 20a illustrates, the spatial distribution of the precipitation difference field is quite similar to the control field (Fig. 5, bottom). The axis of increase through the Kissimmee River valley is not as distinct as in the control case, but the general pattern of increase over the interior peninsula and decreases on adjacent sides is realized with the use of the Chen-Cotton scheme. The grid-average decrease is 18% of the pre-1900 total. The magnitude of this decrease is somewhat larger than the control, which was 11%. This result indicates that accounting for clouds in the radiative transfer has magnified the impact of land-cover change. A detailed investigation of the possible reasons for the difference with the control is not the primary focus of this study. The point emphasized here is that, even with an alternative radiation treatment, the difference realized because of changing the land cover is qualitatively similar and quantitatively not far removed.

Figure 20b shows that the model configuration with the Kuo scheme produced significantly less precipitation when the 1993 land cover is employed. The grid-average decrease is 13% of the pre-1900 total, which is very close to the percentage decrease in the control. The percentage change is very similar, despite the fact that, as with the 1973 results shown in Pielke et al. (1999), the overall Kuo scheme totals for both land-cover cases were significantly less than those produced by the Kain-Fritsch configuration (maps of these totals are not shown here). For the pre-1900 land-cover case, the Kuo configuration resulted in a grid-average total of 198 mm, whereas for the 1993 case this total was 173 mm. The Kain-Fritsch scheme yielded 342 mm for the pre-1900 case and 305 for the 1993 case.

The horizontal grid spacing would be expected to have significant impacts on the results, regardless of the choices for various physical parameterizations. Moreover, the effects of the physical parameterizations on the grid scale tendencies are themselves strongly influenced by the horizontal grid spacing. In other words, the horizontal grid spacing and the options selected for the physical parameterizations cannot be considered as mutually exclusive categories in this sensitivity analysis because these factors are inherently interactive. This is true especially for the convective parameterization, because convective schemes are designed precisely for a particular range of grid spacing. Figure 20c shows that the results at 40 km, produced with the same physics options used in the control (Kain-Fritsch convection and Mahrer-Pielke radiation), yielded a similar pattern of change in convective rainfall when the pre-1900 land cover is replaced with the 1993 dataset. The percentage decrease, 9.5%, is close to the 11% noted for the control (10 km) case. The overall precipitation totals in both land-cover scenarios (not shown here) were considerably larger than for the 10-km cases, with local maximums exceeding 700 mm. The 10-km maximums were less than 450 mm (see Fig. 5, top two panels). This means that the 10-km totals were in much closer agreement with the observed magnitudes (Fig. 3). Regardless, the important point emphasized here again is that the difference in precipitation totals between land-cover scenarios at either resolution is qualitatively and quantitatively (as a percentage change of the pre-1900 total) similar. This result suggests that the impact of changing the land cover is consistent, regardless of the model grid spacing. Here, this sensitivity was evaluated by examining the results from a model configuration with larger grid spacing than that assigned to the control. However, it is acknowledged that the results could also differ if the model grid increment were smaller than the 10-km control value. At grid increment sizes much below this value, where the model setup could be configured to resolve convection explicitly, the representation of circulations such as the sea breezes could change markedly (and highly nonlinearly) with small changes in grid spacing (Weaver et al. 2002). Here, however, the analysis has been purposely designed to provide results that correspond to a range of grid increments over which the Kain-Fritsch convective parameterization is typically applied.

Other aspects of the model setup besides the physical parameterizations and grid spacing could have similar and even greater impacts on the results of changing the land cover. Two variables in particular that are believed to have a significant impact on warm season convective rainfall in regional/seasonal climate modeling simulations are the specification of SST and initial soil moisture (Walker and Rowntree 1977; Mintz 1984; Atlas et al. 1993; Paegle et al. 1996; Fennessey and Shukla 1999). Using the July-August 1989 simulations as the control, a factorial set of experiments was undertaken (i.e., one addressing initial soil moisture, a second addressing the specification of SST on the regional mode grid, and a third addressing both factors in combination). Note that any effects of SST anomalies at remote locations, such as the equatorial Pacific, are realized through the lateral boundary conditions as specified directly by the NCEP-NCAR reanalysis data. As such, the question of interest here concerns SSTs in the adjacent coastal waters of the RAMS domain and what impact their specification may have on the resulting distribution of convective precipitation on the regional model grid.

In the control simulations, the SSTs were defined based on monthly climatology, as described in section 2b. For the sensitivity studies, a weekly observed dataset from NCEP was used (Reynolds and Smith 1994). For the period July-August 1989, the 2-month average of the weekly observed fields is quite different from the corresponding monthly climatological data (Fig. 22). Note that the Gulf Stream is locally several degrees warmer. In addition, the climatological dataset, which combines many years of multiple observations, is spatially smoother than the weekly data, which are based on a coarse-grid objective analysis of limited observational data.

[larger image]	[larger image]
Fig. 22. Jul-Aug average SST (K) for the weekly observed NCEP data from (left) 1989 and (right) the monthly climatological dataset.

For the sensitivity tests involving initial soil moisture, the alternative initialization was derived from the dataset produced by the University of Washington with the Variable Infiltration Capacity (VIC) model [see Maurer et al. (2002) for a description of both the VIC model and the production of this dataset]. These data are based on the execution of the VIC hydrologic model, as forced with observed rainfall and other observed and derived fields required to provide forcing to the VIC soil hydrologic budget. To initialize RAMS soil moisture, VIC soil moisture was first converted to a percent saturation value corresponding to VIC soil properties. These percentage values were in turn converted to volumetric water content values for RAMS that correspond to the spatially distributed hydraulic properties provided by the FAO soil properties dataset (see section 2b). The 2-m vertical average value derived from VIC was assigned to all vertical levels of a RAMS/LEAF-2 soil column. Figure 23 shows the initial soil moisture for both the pre-1900 and 1993 land-cover scenarios. Note that the initial field is different for the two different land-cover cases because of the imposed saturation condition for swamp and marsh classes. The spatial distribution of these classes is different between the two land-cover datasets. Also note the large area of the Everglades that has volumetric water contents greater than 0.8. In the FAO soil type database, these areas are designated as organic types, which have a very large porosity (i.e., soil moisture saturation value). As discussed by Baker et al. (2001), these properties of the land surface of the Florida peninsula, and their interplay with soil moisture, could have substantial impacts on the nature of local circulations and the interplay of those circulations with the sea breezes.

[larger image]	[larger image]
Fig. 23. Initial soil moisture (m3 m-3.) as derived from the VIC model valid for 1 Jul 1989 for the (left) pre-1900 land cover and (right) 1993 land use.

Figure 21 provides the grid-average rainfall (both land-cover scenarios) for the factorial set of SST and soil moisture experiments. There are substantial differences in the absolute totals among the experiments. However, the difference between land-cover scenarios within a given sensitivity experiment is consistent, lending further credence to the robustness of the impact of land-cover change on convective precipitation in these simulations. The decrease relative to the pre-1900 totals for all four experiments is in the range of 10% to 12%, which is identical to the range for all three periods simulated with the control setup. The spatial nature of the difference field for the experiment incorporating both factors (Fig. 20d) is consistent with the results presented above, with substantial increases in precipitation directly over the Kissimmee River basin and decreases along the adjacent sea-breeze fronts.

Go back to Mesoscale Circulations | Go ahead to Comparison: Model & Observed Trends