Sensitivity of the WRF model to PBL parametrisations and nesting techniques: evaluation of wind storms over complex terrain

Abstract

Abstract. Simulating surface wind over complex terrain is a challenge in regional climate modelling. Therefore, this study aims at identifying a set-up of the Weather Research and Forecasting Model (WRF) model that minimises systematic errors of surface winds in hindcast simulations. Major factors of the model configuration are tested to find a suitable set-up: the horizontal resolution, the planetary boundary layer (PBL) parameterisation scheme and the way the WRF is nested to the driving data set. Hence, a number of sensitivity simulations at a spatial resolution of 2 km are carried out and compared to observations. Given the importance of wind storms, the analysis is based on case studies of 24 historical wind storms that caused great economic damage in Switzerland. Each of these events is downscaled using eight different model set-ups, but sharing the same driving data set. The results show that the lack of representation of the unresolved topography leads to a general overestimation of wind speed in WRF. However, this bias can be substantially reduced by using a PBL scheme that explicitly considers the effects of non-resolved topography, which also improves the spatial structure of wind speed over Switzerland. The wind direction, although generally well reproduced, is not very sensitive to the PBL scheme. Further sensitivity tests include four types of nesting methods: nesting only at the boundaries of the outermost domain, analysis nudging, spectral nudging, and the so-called re-forecast method, where the simulation is frequently restarted. These simulations show that restricting the freedom of the model to develop large-scale disturbances slightly increases the temporal agreement with the observations, at the same time that it further reduces the overestimation of wind speed, especially for maximum wind peaks.The model performance is also evaluated in the outermost domains, where the resolution is coarser. The results demonstrate the important role of horizontal resolution, where the step from 6 to 2 km significantly improves model performance. In summary, the combination of a grid size of 2 km, the non-local PBL scheme modified to explicitly account for non-resolved orography, as well as analysis or spectral nudging, is a superior combination when dynamical downscaling is aimed at reproducing real wind fields.

Impact of Grid Nudging Parameters on Dynamical Downscaling during Summer over Chinese Mainland

“…This hypothesis is also verified by the ME analysis of the simulated geopotential height at different levels (data not shown). Furthermore, a very consistent result was the overestimation of wind in the CTL, and was in good agreement with previous studies [20,33,44], which believed that it was partly due to the poor representation of the unresolved topography. Gómez-Navarro et al [20] pointed out that the bias could be reduced effectively by a change of PBL scheme, the use of nudging, and improvement of horizontal resolution.…”

“…Recently, research on the nudging application and comparison of two nudging schemes (grid nudging and spectral nudging) has been very common [17][18][19][20][21][22][23]; however, a consistent conclusion is still lacking. Logically, spectral nudging is superior to grid nudging [24], but many findings [17,22,23] have shown that, in some cases, grid nudging is better than spectral nudging.…”

Perspectives of regional paleoclimate modeling

“…). A standard resolution for present‐day and future RCMs is 10–25 km, and is increased to 1–3 km for specific applications . For paleoclimate applications, the spatial resolution has usually been limited to about 50 km due to the demand of longer simulation periods, requiring a high amount of computational resources.…”

An intercomparison of mesoscale models at simple sites for wind energy applications

“…The choice of groupings was based mainly on two criteria: (1) it was possible to form groups with at least six members in each group and (2) each of the options was highlighted in the literature as being important for model performance (Hahmann et al, 2015b;Gómez-Navarro et al, 2015;Carvalho et al, 2012;Draxl et al, 2014). Several other setup options were considered: MM, LSM, land cover, spin-up time, and data set used for initial and boundary conditions, but either it was not possible to group them in a meaningful way, or they were deemed of too little importance based on previous studies.…”

“…They showed that the Pleim-Xiu SL scheme (Pleim, 2006) combined with the ACM2 PBL scheme (Pleim, 2007b) gave the smallest errors for wind speed and wind energy production estimates across the sites, while the quasi-normal scale elimination (QNSE) SL and PBL schemes (Sukoriansky et al, 2005) gave smaller errors for offshore sites. In a similar study, Gómez-Navarro et al (2015) analyzed the sensitivities of the WRF model to the choice of PBL scheme and grid spacing in complex terrain in Switzerland. They found that using a modified version of the YSU PBL scheme, which accounts for effects of unresolved topography (Jiménez and Dudhia, 2012), in combination with the smallest grid spacing (2 km) and analysis nudging gave the best agreements with measurements during a number of wind storms.…”