Special session of California 2000 field study - Air Pollution Modeling and Its Application - page 522

2. Model Description, Initialization and Validation

These simulations employed the most recent versions of WRF (v.3.0.1, v.3.1)
and MM5 (v.3.7.4) with three nests telescoping to 4 km resolution. Initializations
included three-hourly analyses from the Eta model (ETA) and the North American
Regional Reanalysis (NARR), and six-hourly Global Forecast System Final Analyses
(FNL). For WRF, simulations were made using the Noah and Thermal Diffusion
(TD) land surface models, RRTM/Dudhia and CAM radiation parameterizations,
and YSU and MYJ planetary boundary layer (PBL) schemes, with and without
analysis nudging. Model tests spanned three platforms (Xeon, Athlon, PowerPC),
two operating systems (Linux and Mac OS X) and three compilers (Intel, Portland
Group and IBM). Comparable MM5 runs were also made.
We seek accurate representations of PBL temperature, wind and humidity across
complex terrain. However, validation data largely consists of hourly observations
from Remote Automated Weather Stations (RAWS) and mainly airport-based
Automated Surface Observing System (ASOS) sites. For August 1-2, 2000, 140
representative stations in the 4 km nest with fewer than 25% of hours missing and
post-interpolation elevation discrepancies <200 m were identified ( Fig. 1 , Table 1).

Table 1. Representative station information

Type Number of stations Avg. elevation (m) Elevation range (m)

ASOS 39 243 1-1958

RAWS 101 957 5-2679

RAWS subset 32 242 5-459

3. Temperature Diurnal Cycle Assessment

The present analysis focuses on the temperature diurnal cycle for the 24 h period
commencing 00Z August 1, from over 50 simulations initialized at variety of times
back to 00Z July 31. As winds were light and variable in the absence of large-scale
forcing, the boundary layer circulation was sensitively dependent on differential
heating reflecting mesoscale landuse and topographic variations, stressing a model's
weakest components.
Surprisingly large variation among forecasts of average 2 m temperature for the
combined RAWS and ASOS dataset was found ( Fig. 2a) . Many forecasts evinced
a nighttime warm bias and all ostensibly underpredicted temperature the following
afternoon; the TD scheme consistently overpredicted 2 m dewpoints ( Fig. 2b) .
However, inspection revealed a substantial difference in forecasts at ASOS and