Results of numerical simulations of squall lines in Transbaikalia region of Russia are shown in the article. Simulations have conducted by the advanced version of Weather Research and Forecasting (WRF) model for the events recorded over the period from 2010 to 2017. WRF model is configured to provide calculations on two nested domains with grid spacings of 3 and 9 km. Simulations with two types of planetary boundary layer (PBL) parameterizations are compared (Yonsei University and Mellor-Yamada-Janjic schemes). Approach to determine squalls in model output based on values of simulated vertical and horizontal wind speed is proposed. Description and product samples of operational method for forecasting squall lines and strong winds in Transbaikalia region of Russia are presented.
A version of WRF-ARW model with two nested domains (Fig. 1) is used to simulate squall lines on the territory of
Transbaikalia region, Russia. The spatial resolution of outer domain (332×250 grid points) is 9 km and for inner
domain (618×339 grid points) is 3 km. Detailed representation of topography and land use (resolution of data is 30″)
in the inner domain allows to get a more realistic mesoscale circulation over the complex-terrain region during a
squall event. There are 31 eta-levels in the vertical dimension of model domains with the most detailed resolution in
PBL; there are 11 levels for the height of 2 km AGL. Initial and boundary data are given by 3-h interval product of
the Global Forecasting System (GFS) [
Both versions of the WRF-ARW model are calculated on Altix S4700 (peak performance is 0.7 TFlops). One run for lead time of 24 hours takes near 2 hours of machining time. Output production with time step of 1 minute is available for analysis. Procedure of calculations includes three main steps as follow: WRF preprocessing of initial and boundary data provided from GFS output; calculation of the WRF-ARW model; analysis of output WRF data to diagnose squall events.
A significant contribution to intensifications of near ground wind is made by turbulent motion [
We have tested YSU and MYJ parameterizations to define the most suitable scheme of PBL for squall forecasting based on WRF-ARW model. The comparison of output model data for these schemes was conducted. The skew-T aerological diagrams for simulated squalls events were obtained. This diagram shows a vertical profile of temperature, dew point and winds in the atmosphere on particular layers above a point of interest (a point where squall was recorded).
Figure 2 shows simulated skew-T diagrams for Mogocha at 0531 UTC on 14 June 2011 where squall (10-m wind speed of 23 m/s) took place at this time. We calculate some energetic indices to define possibility of convection initiation.
Main indicator of atmospheric instability is convective available potential energy (CAPE). CAPE is the positive buoyancy of an air parcel. We use mixed layer CAPE; this index is calculated for air parcel from 100-mb layer AGL. The vertical resolution of squall vortex is near 1.5-2 km AGL and mixed layer CAPE shows how many energy is available here for developing of convection. Convective inhibition (CIN) is a negative buoyancy of an air parcel. It describes the power of inhibitory layer (air parcel has to overcome this layer to initiate convection). (1)
LFC where z – start height, LFC is level of free convection, EQL – equilibrium level (level of neutral bouncy), Tv,p – virtual temperature of parcel, Tv,e – virtual temperature of environment, g is 9,81 m/s.
LFC Tv, p − Tv,e CIN = g
dz , z
Tv,e
EQL Tv, p − Tv,e CAPE = g
dz
CAPE and CIN for squall event in Mogocha (at 0531 UTC on 14 June 2011) for both PBL schemes are noticeably different. Value of CAPE is 784 J/kg and CIN is -176 J/kg for simulation with MYJ parameterization. For YSU scheme values of these indices are near zero (parcel trajectory does not cross temperature line on Fig. 2b). Figure 2 also shows that 0-1 km wind shift is 16 m/s and 2 m/s for MYJ and YSU schemes respectively. Therefore, MYJ generate stronger atmospheric convection than YSU scheme.
Further analysis of simulated skew-T aerological diagrams during the squalls shows that YSU scheme of PBL generate weaker convection (based on mixed layer CAPE and CIN) and associated wind in comparison with MYJ parameterization. Simulations with MYJ scheme spawn more squall events, sometime 10-m wind speed in these events is above 27-30 m/s, but YSU scheme generates wind speed no more than 25-27 m/s.
Skew-T diagrams provide a lot of information but vertical wind speeds do not take into consideration. To take into consideration vertical wind speed we develop special method for identification of squalls in model output. 3.1
On pilot stage we tested several methods to identify squall events from model output [
N V1 (i, j ) = max( M 10 m , M 1 ) + (max( wrd ) − min( wrd ) ), (2) k =1 where M1 – wind speed at the closed to the ground model level, M10m – wind speed at 10 m, wrd – values of vertical wind speed inside a circle of radius 30 km centered at point (i, j), k – number of a model level (numbering from bottom to top), N – number of vertical levels under consideration.
Parameter V1 takes into consideration difference between maximum and minimum values of vertical wind speed at model levels in PBL within the circle centered at a point of interest. Thus, last term of (2) captures characteristics of a horizontal roll vortex. Values of V1 are calculated with time step of 1 minute automatically during a run of WRFARW model.
We suggest that a model squall in grid point (i, j) takes place if the value of V1 in this point exceeds threshold of 18 m/s no longer than 30 minutes. A model squall is suggested to be correct if it locates within the circle of radius 30 km centered at the point where observed squall is recorded, maximum value of V1 in this area exceeds 18 m/s and time difference between simulated and observed squall events takes no more than 3 hours.
Maps of parameter V1 are plotted to find out the position of squall lines and define accuracy of simulations. Figure 3 shows position of squall lines based on values of V1. The squall took place (10-m wind speed is 24 m/s) on point 30823 at 0815 UTC on 28 July 2010. Maximum values of V1 are 20-21 m/s and 22-25 m/s for YSU and MYJ parameterizations respectively.
Analysis of V1 maps shows that both PBL parameterizations integrated to WRF-ARW model sometimes spawn false squalls (especially MYJ scheme). The amount of false alarms varies depending on a year but it exceeds 50% and 70% for YSU and MYJ schemes respectively.
However it should be understood that meteorological observational points record no more than 10-15% of squall events. Majority of false alarms is possible to eliminate through the analysis of wind regime of every geographical distinct of Transbaikalia region (i.e., prolonged strong winds are blowing along the west coast of lake Baikal). Therefore, false alarms are not major challenge. 3.2
Standard measures of skill in dichotomous categorical forecasts are used to compare the quality of model
simulations. We take into consideration values of critical success index (CSI) [
CSI =
x x + y + z
x x + y ,
POD = ,
FAR = ,
bias = z x + z x + z x + y , (2) where x is a number of correct forecasts, y is a number of missed events and z is a number of false alarms. CSI and POD measure skill in squall forecasting, bias compares number of simulated and observed events and FAR shows false alarms. These measures are suitable for forecasting of rare events because of a large number of correct rejections do not take into consideration and total sampling size does not affect result.
Table 1 shows the measures of skill in squall forecasting for both variants of WRF -ARW model. Probability of detection is 39% and 67% for YSU and MYJ schemes respectively. CSI value of YSU is better than for MYJ according to large values of false alarm ratio and bias. WRF-ARW model with MYJ scheme generates no more than three times that really observed squalls.
New cluster CRAY XC-40 was installed in Regional Specialized Meteorological Center of Khabarovsk, Russia (RSMC Khabarovsk) in 2018. This cluster includes 60 nodes of two 18-cores Intel Broadwell processors with Dragonfly topology and has total peak performance of 76 TFlops. Therefore, we get new directions to improve current method for squall and strong wind forecasting. We plan to use the latest stable version of WRF-ARW model of one domain with grid step of 1 km not only for the territory of Transbaikalia region but also extend the forecasting area to Amur region where strong squalls take place as well. Furthermore, vertical resolution of WRF-ARW model will expanded to 45-60 levels to improve representation of temperature, humidity and wind profiles.
Some new parameterizations of PBL are introduced in WRF model like the Mellor‐Yamada‐Nakanishi‐Niino level
2.5 (MYNN) [