Intercomparison of nocturnal turbulence formulations in WRF model at a network of stations in southern Brazil

WRF model is used to simulate the nocturnal boundary layer for 31 nights in July 2012. The domain covers Rio Grande do Sul state, in southern Brazil, including 26 stations, arranged in a network with a typical distance of 100 km between adjacent stations. Observations of 2-m temperature, specific humidity and 10-m winds are compared to the WRF outputs using 4 different turbulence schemes: Yonsei University (YSU), Mellor-Yamada-Janjic (MYJ), standard Bougeault–Lacarrére (Boulac) and a modified, more turbulent, version of it (Boulac-m). The overall performance of the turbulence formulations varies both by station and by night. YSU, the most turbulent of all schemes compared, provides the smallest root mean squared error for the entire period, but is shown to consistently overpredict temperature, both in calm and windy nights. MYJ has a comparable overall performance to YSU, but it tends to overpredict temperature in windy nights, while underpredicting it in calm nights. Boulac parameterization is the least turbulent of all, and because of that, it underpredicts temperature in all nights except the calmest ones. The increased turbulence in Boulac-m modification makes their overall results very similar to those obtained with MYJ. Spatially, the stations with poorest performance in all schemes are located at altitudes lower than their surroundings. At higher stations, schemes YSU and MYJ show a consistent improvement, which is not observed in either version of Boulac. The main reasons for the difference in model performance among nights and sites are discussed and suggestions for improvement are presented.