448 12. Model Evaluation velocity using a nonlinear bottom boundary condition, with the drag [(12-50)] computed for a linear lower boundary condition. Among their results they found that the drag was enhanced compared with linear theory for mountains with a gentle upslope and steep downslope terrain. 12.6 Comparison with Observations The validation of a model using observations can be cataloged into two general classes: (1) subjective evaluation, or (2) point and pattern quantitative validation. Using subjective verification, one or more of the predicted fields are qualitatively compared against observations of a related phenomena. Pielke (1974a), for example, compared the simulated vertical motion at an elevation of 1.22 km over south Florida with the observed locations of rain showers as seen via a 10-cm radar located at Miami. The justification for the comparison is that the primary control for rain shower development over south Florida during a synoptically undisturbed summer day is the location and intensity of the low-level convergence. Since 1.22 km is approximately at the top of the planetary boundary layer, the predicted vertical velocity at that level yields an appropriate estimate for low-level convergence. Figure 12-13 illustrates one such comparison for June 29, 1971, 93h after sunrise. As evident in the figure the wishbone pattern of the rain showers is closely correlated with the distribution of vertical motion at 1.22 km. Point to point correspondence between model prediction and observation of the same meteorological parameter provides a quantitative test of model skill. Keyser and Anthes (1977) utilize a useful technique where if (i) @;and é, are individual predictions and observations at the same grid point, spectively; (ii) % and o,,, are the average values of @; and ¢;,,, at a | respectively; and (iii) # N is the number of observations, then a fs E= {x (- a : ba { antl i N 7 (ooo doari#5} ’ a e ba - oe — Poon)?/# wt : 4450 12 Model Evaluation TABLE 12-1 Error Analysis of Model Predicted Winds and Temperatures Using (12-53) for an East-West Cross Section from Naples to Just North of Fort Lauderdale, Florida® Variable E Ey ° So, Euy/E Eva/oum aim s~*) 3 31 12 22 10 14 a(m s~') 22 12 08 12 05 10 TCO) SA 2.8 39 46 05 06 “ See Fig. 12-13 for the location (from Pielke and Mahrer, 1978). and a 4a a= {5 (;- ores} can be used to determine the skill of the model results. The parameter E is the root mean square error (rmse), Eyp the rmse after a constant bias is removed and ¢ and ¢,,. the standard deviations of the predictions and the observations tespectively.'© Keyser and Anthes found that the rms error can be reduced significantly when a constant bias is removed. Such a bias, they suggested could be a result of an incorrect specification of the initial and/or bottom and lateral boundary conditions. Skill is demonstrated when (I) o =~ O55 (2) E 1 since the ratio would be expected to be unity dom chance. A necessary condition for perfect skill is F, = 1 since the = convergence zone would be covered with echoes in that case.'7 methodology is illustrated schematically in Fig. 12-14 for an idealized ibution of radar echoes and convergence. Results for an actual model ition of a sea breeze over south Florida (for July 1, 1973) from Pielke and (1978) are given in Table 12-2, where the ratio of F:/F, was greater unity in 26 out of 30 categories. The ratio was larger than 2.0 for 20 of the wries. By contrast, F, was much less than unity indicating that most of the ‘gence Zones were not covered by rain showers —a result indicating that eze convergence alone does not completely explain the spatial ility of radar echoes over south Florida during the summer. application of this analysis procedure to other meteorological les, such as cloud cover, rainfall, etc., is straightforward. Simpson et al.