A sensitivy study of microphysics schemes of the wrf-arw model for low-level wind shear forecast in José Martí International Airport during convective storms.

2.1 WRF-ARW and RAP model

The WRF v.3.9 model (Skamarock et al., 2008Skamarock, W.C.; Klemp, J.B.; Dudhia, J.; Gill, D.O.; Barker, D.M.; Wang, W. & Power, J.G. 2008. “Description of the Advanced Research WRF Version 3”. NCAR Tech Note NCAR/TN 475 STR, UCAR Communications, Boulder, Colo.) was used with ARW dynamic core. The simulations comprised a 12/4 km two-way nested domains (Figure 1) and 34 verticals levels. Briefly, the setup includes for both domains (Coll-Hidalgo et al., 2021Coll-Hidalgo, P.; Pérez-Alarcón, A. & González-Jardines, P.M. 2021. “Evaluation of Microphysics Schemes in WRF-ARW Model for Numerical Wind Forecast in José Martí Airport”. In: Proceedings of The 3rd International Electronic Conference on Atmospheric Sciences. <https://sciforum.net/paper/view/conference/8121> [Consulted: July 15, 2021]): Rapid Radiative Transfer Model longwave radiation parametrization (Mlawer et al., 1997Mlawer, E.J.; Taubman, S.J.; Brown, P.D.; Iacono, M.J. & Clough, S.A. 1997. “Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave”. Journal of Geophysical Research: Atmospheres, 102(D14): 16663-16682, ISSN: 0148-0227, DOI:10.1029/97JD00237.), Dudhia shortwave radiation scheme (Dudhia, 1989Dudhia, J. 1989. “Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model”. Journal of Atmospheric Sciences, 46(20): 3077-3107, ISSN: 0022-4928, DOI:10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.), Unified Noah land-surface model (Tewari et al., 2004Tewari, M.; Chen, F.; Wang, W.; Dudhia, J.; LeMone, M.A.; Mitchell, K.; Ek, M.; Gayno, G.; Weigel, J. & Cuenca, R.H. 2004. “Implementation and verification of the unified NOAH land surface model in the WRF model”. In: Proceedings of the 20th Conference on Weather Analysis and Forecasting/16th Conference on Numerical Weather Prediction, Seattle, WA, USA; 17.5, pp. 11-15.), Grell-Freitas Ensemble cumulus parametrization (Grell & Freitas, 2013Grell, G.A. & Freitas, S.R. 2013. “A scale and aerosol aware stochastic convective parameterization for weather and air quality modelling”. Atmospheric Chemestry and Physics Discussion, 13(23): 23845-23893, ISSN: 1680-7316, DOI:10.5194/acp-14-5233-2014.), Mellor-Yamada-Janjic planetary boundary layer (Janjić, 2001Janjić, Z.I. 2001. “Nonsingular implementation of the Mellor-Yamada level 2.5 scheme in the NCEP Meso Model”. NCEP Office Note 437.). The model was initialized and forced at the boundaries, with every three hourly updates, by 0.500 Global Forecast System (GFS) forecast outputs, which are freely available at https://nomads.ncep.noaa.gov/cgi-bin/filter_gfs_0p50.pl.

Storms simulations were performed with four selected microphysics schemes: Lin (Lin et al., 1983Lin, Y.-L.; Farley, R.D. & Orville, H.D. 1983. “Bulk parameterization of the snow field in a cloud model”. Journal of Applied Meteorology and Climatology, 22(6): 1065-1092, ISSN: 1558-8424, DOI:10.1175/1520-0450(1983)022<1065:BPOTSF>2.0.CO;2.), Morrison 2-moment (Morrison, Thompson & Tatarskii, 2009Morrison, H.; Thompson, G. & Tatarskii, V. 2009. “Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes”. Monthly Weather Review, 137(3): 991-1007, ISSN: 1520-0493, DOI: 10.1175/2008MWR2556.1.), WSM5 (Hong et al., 2004Hong, S.-Y.; Dudhia, J. & Chen, S.-H. 2004. “A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation”. Monthly Weather Review, 132: 103-120 ISSN: 1520-0493, DOI: 10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2.), and WSM6 (Hong & Lim, 2006Hong, S.-Y & Lim, J.-O.J. 2006. “The WRF single-moment 6-class microphysics scheme (WSM6)”. Journal of the Korean Meteorological Society, 42: 129-151, ISSN: 1598-3560, DOI:10.1155/2010/707253.). Table 1 shows the main species of prognostic variables in these schemes. The selection was based on their use and performance in numerical weather forecast systems operating in Cuba (Sierra et al., 2015Sierra, L.M.; Ferrer, H.A.L.; Valdés, R.; Mayor, Y.; Cruz, R.; Montejo, I.; Rodríguez, C.F.; Rodríguez, N.; Roque, A. 2015. “Automatic mesoscale prediction system of four daily cycles”. Technical Scientific Report. DOI: 10.13140/RG.2.1.2888.1127 ; Pérez-Bello et al., 2019Pérez-Bello, A.; Mitrani, A.I.; Díaz, R.O.; Wettre, C. & Robert, L. H. 2019. “A numerical prediction system combining ocean, waves and atmosphere models in the Inter-American Seas and Cuba”. Revista Cubana de Meteorología, 25(1): 109-120, ISSN: 2664-0880 <http://rcm.insmet.cu/index.php/rcm/article/view/459> [Consulted: January 15, 2021]). The forecasts were for 54 hours, started from the storm observation date at 0000 UTC.

To verify the accuracy of the LLWS forecast was used the Rapid Refresh model (RAP) analysis (Benjamin et al., 2016Benjamin, S.G.; Weygandt, S.; Brown, J.M.; Hu, M.; Alexander, C.R.; Smrinova, S.; Olson, J..; James, E.P.; Dowell, D.C.; Grell, G.A.; Lin, H.; Peckham, S.E.; Smith, T.L.; Moninger, W.R.; Kenyon, J.S.; Manikin, G. 2016. “A North American Hourly Assimilation and Model Forecast Cycle: The Rapid Refresh”. Monthly Weather Review, 144: 1669-1694, ISSN: 1520-0493, DOI:10.1175/MWRD-D-15-02421.) (https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/rapid-refresh-rap). In the RAP analysis, the horizontal wind components are updated variables in Gridpoint Statical Interpolation analysis system (GSI)/RAP assimilation. According to Benjamin et al. (2016)Benjamin, S.G.; Weygandt, S.; Brown, J.M.; Hu, M.; Alexander, C.R.; Smrinova, S.; Olson, J..; James, E.P.; Dowell, D.C.; Grell, G.A.; Lin, H.; Peckham, S.E.; Smith, T.L.; Moninger, W.R.; Kenyon, J.S.; Manikin, G. 2016. “A North American Hourly Assimilation and Model Forecast Cycle: The Rapid Refresh”. Monthly Weather Review, 144: 1669-1694, ISSN: 1520-0493, DOI:10.1175/MWRD-D-15-02421., the analysis method used for those variables is GSI hybrid ensemble-variational analysis. For wind components, the use of GSI permits assimilation of NOAA 405 MHz profiler wind data (decommissioned 2014), boundary layer (915 MHz) profiler wind data, radar data, aircraft data, surface/METAR-land, surface/mesonet-land, bouy/ship and GOES atmospheric motion vectors. Although the study area was near the boundaries of the RAP domain, the RAP analysis data was used as preliminary source for evaluation.

2.2 Observational data

The study area was "José Martí" International Airport (Figure 2a). It was selected because it constitutes the airport with the highest number of air operations in Cuba. Furthermore, the airport is surrounded by terrain with complex orography. "José Martí" International Airport is located in the inland forecast region of the Artemisa, Havana, and Mayabeque provinces (Figure 2b).

Sosa (2018)Sosa, A. 2018. “International Airport “José Martí” low-level wind shear characteristics during 2012-2017 period”. Bachelor Thesis, Higher Institute of Technologies and Applied Sciences, Havana, Cuba. describes the behaviour of the LLWS over "José Martí" International Airport. LLWS is associated with meteorological systems: cold front (18.36 %), anticyclone (53.06 %), and tropical wave (10.20 %) (Sosa, 2018). The author identified the following synoptic patterns as the most frequent in which low-level wind shear occurs:

Sosa (2018)Sosa, A. 2018. “International Airport “José Martí” low-level wind shear characteristics during 2012-2017 period”. Bachelor Thesis, Higher Institute of Technologies and Applied Sciences, Havana, Cuba. refers to that significant wind field variations over the airport are often reported under storms. As case studies, four storms associated with synoptic patterns that cause dangerous conditions at this aerodrome were selected, as shown in Table 2.

Meteorological data from the weather station in the "José Martí" International Airport was used to evaluate the surface forecast. The storms were frequently placed in the aerodrome vicinity (Figure 3). In addition, to better capture details of simulations were utilized observations from weather stations placed in the inland forecast region of the Artemisa, Havana, and Mayabeque provinces (Figure 2b).

3.1 Post-processing WRF-ARW output files

In this paper, it was used the WRF-ARW output variables T2 (temperature at 2 m, K), PSFC (surface pressure, Pa), Q2 (vapor mixing ratio at 2 m, kg/kg), U (x-wind component, m/s), V (y-wind component, m/s), U10 (x-wind component at 10 m, m/s), V10 (y-wind component at 10 m, m/s), RAINC (accumulated total cumulus precipitation, mm), Air Force Weather Agency (AFWA)_LLWS (AFWA_diagnistics (Creighton et al., 2014Creighton, G.; Kuchera, E.; Adams-Selin, R.; McCormick, J.; Rentschler, S. & Wickard, B. 2014. “AFWA diagnostics in WRF”. <http://www2.mmm.ucar.edu/wrf/users/docs/AFWA-Diagnostics-in-WRF.pdf>. [Consulted: February 12, 2020]): 0-2000 ft wind shear, m/s). Relative humidity was determined using Clausius-Clapeyron (Iribarne & Godson, 1981Iribarne, J.V. & Godson, W.L. 1981. “Atmospheric Thermodynamics”. In Geophysics and Astrophysics Monographs, 2^nd ed.; B.M. McCormac, Ed.; Kluwer Academic Publishers: Boston, MA, USA; pp. 259.) as follow:

Where:

T= temperature (K)

p= pressure (Pa)

q= specific humidity or the mass mixing ratio of water vapor to total air (dimensionless)

T₀= reference temperature (typically 273.16 K)

3.2 Calculation of LLWS

LLWS was determined by calculation for components (OACI, 2015OACI. 2005. “Manual on Low-Level Wind Shear”. Doc 9817 AN/449, pp. 1-8 <https://standards.globalspec.com/std/471313/9817> [Consulted: March 10, 2021]), as a difference between wind vectors from one point in space to another. LLWS is defined as a vector difference from surface to 500 m (Boilley & Mahouf, 2008Boilley, A.; Mahouf, J.-F. & Lac, C. 2008. “High resolution numerical modeling of low level wind-shear over the Nice-Côte d’Azur Airport”. In: 13^th American Meteorology Society Conference on Mountain Meteorology, Session9A.2, United States of America, pp. 11-15, Available: <https://ams.confex.com/ams/13MontMet17AP/webprogram/Paper141049>, [Consulted: August 22, 2021] .; Díaz-Zurita et al., 2021Díaz-Zurita, A.; Góngora-González, C.M.; Lobaina-LaO, A.; Pérez-Alarcón, A. & Coll-Hidalgo, P. 2021. “Numerical mesoscale and short-term wind forecast for the “José Martí” International Airport of Havana”. Revista Brasileira de Meteorologia, 36(2),171-182, ISSN: 1982-4351, DOI:10.1590/0102-77863610007.) or 600 m (OACI, 2015OACI. 2005. “Manual on Low-Level Wind Shear”. Doc 9817 AN/449, pp. 1-8 <https://standards.globalspec.com/std/471313/9817> [Consulted: March 10, 2021]; Urlea & Pietrisi, 2015Urlea, A.-D. & Pietrisi, M. 2015. “Study of the Low Level Wind Shear using AMDAR reports”. EGU General Assembly Conference Abstracts, pp. 2185, <https://ui.adsabs.harvard.edu/abs/2015EGUGA..17.2185U> [Consulted: August 21, 2021]; Iribarne & Godson, 1981Iribarne, J.V. & Godson, W.L. 1981. “Atmospheric Thermodynamics”. In Geophysics and Astrophysics Monographs, 2^nd ed.; B.M. McCormac, Ed.; Kluwer Academic Publishers: Boston, MA, USA; pp. 259.). In this paper, the forecast includes LLWS speed and direction, in sub-layers from 0-600 m (Figure 4a). These layers make it possible to detect and forecast wind shear caused by smaller-scale storms. Furthermore, the sub-layers LLWS forecast is necessary to support meteorological service. The forecast accuracy, in this case, strongly dependent on the scale on which the wind shear operates relative to the size of the aircraft.

The density of the nodes in the neighborhood of the airport in the domain of 4 km (d02) of the resolution is low. For this reason, the rectangular grid developed by Díaz-Zurita et al. (2021)Díaz-Zurita, A.; Góngora-González, C.M.; Lobaina-LaO, A.; Pérez-Alarcón, A. & Coll-Hidalgo, P. 2021. “Numerical mesoscale and short-term wind forecast for the “José Martí” International Airport of Havana”. Revista Brasileira de Meteorologia, 36(2),171-182, ISSN: 1982-4351, DOI:10.1590/0102-77863610007. was used for wind field simulation. This grid takes into account the orientation (60⁰ from the north) and the length of the runway (4 km). Five points are matched: one in the center of the runway (MID), one at each headland, and the other two points at 1 km from the center. The grid with a longitudinal resolution of 0.87 km and a latitudinal resolution of 0.5 km is shown in Figure 4b.

Based on Díaz-Zurita et al. (2021)Díaz-Zurita, A.; Góngora-González, C.M.; Lobaina-LaO, A.; Pérez-Alarcón, A. & Coll-Hidalgo, P. 2021. “Numerical mesoscale and short-term wind forecast for the “José Martí” International Airport of Havana”. Revista Brasileira de Meteorologia, 36(2),171-182, ISSN: 1982-4351, DOI:10.1590/0102-77863610007. results, for wind interpolation, we use the natural neighbor method. In addition, following Díaz-Zurita et al. (2021)Díaz-Zurita, A.; Góngora-González, C.M.; Lobaina-LaO, A.; Pérez-Alarcón, A. & Coll-Hidalgo, P. 2021. “Numerical mesoscale and short-term wind forecast for the “José Martí” International Airport of Havana”. Revista Brasileira de Meteorologia, 36(2),171-182, ISSN: 1982-4351, DOI:10.1590/0102-77863610007. recommendations, a correction to the interpolated WRF-ARW wind field is applied with a consistent mass model.

This model is based on the equation of continuity for an incompressible air mass moving in a two-dimensional domain, Ω, with a velocity field $\overrightarrow{u}\left(u,v,w\right)$ :

If the constant air density is considered for the entire domain, the equation becomes:

Which joins the impenetrability condition on the ground ${\Gamma }_b$ , thus constituting the boundary condition:

From conditions (3 $\overrightarrow{\nabla }\bullet \overrightarrow{u}=0\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }in\mathrm{ }\mathrm{ }\mathrm{ }\Omega$ ) and (4 $\overrightarrow{\eta }\bullet \overrightarrow{u}=0 in {\Gamma }_b$ ), the consistent mass models pose a least-squares problem with the velocities to adjust $\overrightarrow{u}\left(u,v,w\right)$ from the observed ${\overrightarrow{u}}_0\left(u_0,v_0,w_0\right)$ in the Ω domain, according to the functional:

where { $u\left(x,y,z\right)$ , $v\left(x,y,z\right)$ , and $w\left(x,y,z\right)$ , } are the wind components calculated by the model through fit; { $u_0\left(x,y,z\right)$ , $v_0\left(x,y,z\right)$ , and $w_0\left(x,y,z\right)$ , } are the components of the initial field, interpolated from the observations, and ${\alpha }_{1, }{\alpha }_{2, }{\alpha }_{3 }$ are the Gaussian precision modules (Montero et al., 2006Montero, G.; Montenegro, R.; Escobar, J.; Rodríguez, E. & González, J. 2006. “Modelización y simulación numérica de campos de vientos orientados a procesos de contaminación atmosférica”. Instituto Universitario de Sistemas Inteligentes y Aplicaciones numéricas en Ingeniería, Universidad de Las Palmas, Gran Canaria <http://www-lacan.upc.es/projects/CLI, Modelización Numéricas de Problemas Medioambientales de Convección-Difusión-Reacción> [Consulted: August 21, 2021]). Considering ${\alpha }_1$ and ${\alpha }_2$ identical, for horizontal directions the functional to minimize (5 $E\left(u,v,w\right)=\iiint \left[{{\alpha }_1}^2{\left(u-u_0\right)}^2{{+\alpha }_2}^2{\left(v-v_0\right)}^2+{{\alpha }_3}^2{\left(w-w_0\right)}^2\right]dxdydz$ ) is:

The search field $\overrightarrow{v}\left(u,v,w\right)$ will be the solution to the problem:

Find $\overrightarrow{v}$ ∈ K such that,

This problem is equivalent to finding the saddle point at ( $\overrightarrow{u,\Phi }$ ) of the Lagrangian:

The technique of Lagrange multipliers allows obtaining the saddle point of the expression (9 $\overrightarrow{v}=\overrightarrow{v_0}+T\overrightarrow{\nabla }\overrightarrow{\Phi }$ ), $L\left(\overrightarrow{u},\lambda \right)\le L\left(\overrightarrow{u},\Phi \right)\le L$ ) such that the solution field is obtained from the Euler-Lagrange equations:

Where Φ is the Lagrange multiplier and $T=[T_h,T_h,T_v]$ is the transmission diagonal tensor:

If ${\alpha }_1$ , ${\alpha }_2$ are considered constant throughout the domain, the variational formulation leads to an elliptic equation defined in Φ. Indeed, substituting equation (9 $\overrightarrow{v}=\overrightarrow{v_0}+T\overrightarrow{\nabla }\overrightarrow{\Phi }$ ) in (3 $\overrightarrow{\nabla }\bullet \overrightarrow{u}=0\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }in\mathrm{ }\mathrm{ }\mathrm{ }\Omega$ ) results:

Which is completed by the null Dirichlet condition at permeable boundaries (vertical domain boundaries)

And Neumann's condition in the raincoats (terrain and upper border)

Considering $T_h$ and $T_v$ constant, equation (12 $-\overrightarrow{\nabla }\bullet \left(T\overrightarrow{\nabla }\right)=\overrightarrow{\nabla }\bullet \overrightarrow{u_0}$ ) becomes:

Eliminating the vertical component (two dimensions) was obtained:

This methodology guarantees the conservation of wind direction due to the impenetrability conditions.

3.2 Evaluation

In this work, for the forecast verification was calculated some statistical metrics: mean systematic error or BIAS, standard deviation (SD), root mean squared error (RMSE) and Pearson correlation coefficient. Precipitation is a dichotomous variable; therefore, it is convenient to construct a contingency table. According to Jolliffe and Stephenson (2003)Jolliffe, I. & Stephenson, D. 2003. “Forest Verification: a practitioner’s guide in atmospheric science”. In: John Wiley Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England, pp. 37-97., some performance measures were computed. These measurements are determined from each category defined in Table 3.

The hit rate (H) is the proportion of occurrences of precipitation event that where correctly forecast.

False Alarm Rate (F), is the proportion of non-occurrences that were incorrectly forecast.

Proportion Correct (PC) of forecast is given for equation (19 $PC=\frac{a+d}{n} ;n=a+b+c+d$ ).

False Alarm Ratio (FAR) is the proportion of forecasts of occurrences that were not followed by an actual occurrence. FAR is a sample estimate of the conditional probability of a false alarm given that occurrence was forecast.

Critical Success Index (CSI) can be regarded as a sample estimate of the conditional probability of a hit given that the event of interest was either forecast, or observed, or both. Perfect skill (H=1, F=0) gives a CSI with the maximum value of 1.

Gilbert’s skill score (GSS) is a modification of CSI to allow for the number of hits that would have been obtained purely by chance. In equation (22 $GSS=\frac{a-a_r}{a-a_r+b+c}$ ), $a_r$ is the number of hits expected by forecasts independent of observations (pure chance) given by:

To obtain the best microphysics scheme to reproduce the LLWS derived from storms the analysis was developed pre-, in-, and after- the occurrence of the event Sari, Baskoro, & Hakim (2018)Sari, F.P.; Baskoro, A.P. & Hakim, O.S. 2018. “Effect of different microphysics scheme on WRF model: A simulation of hail event study case in Surabaya, Indonesia”. In: Proceedings of the American Institute of Physics Conference, 1987(1), 020002, DOI:10.1063/1.5047287..

4.1 Impact of different microphysical parameterizations on surface variables

(Coll-Hidalgo et al., 2021Coll-Hidalgo, P.; Pérez-Alarcón, A. & González-Jardines, P.M. 2021. “Evaluation of Microphysics Schemes in WRF-ARW Model for Numerical Wind Forecast in José Martí Airport”. In: Proceedings of The 3rd International Electronic Conference on Atmospheric Sciences. <https://sciforum.net/paper/view/conference/8121> [Consulted: July 15, 2021]) found simulated wind surface field evaluation as a difficult criterion to consider the best microphysical scheme for wind forecast. However, the authors pointed out that WSM6 and Lin occasionally exhibit better scores. The wind at 10 meters is the lower height of LLWS calculation layers, therefore, include other sensitivity meteorological surface field analysis for determinate the best WRF-ARW configuration is a priority. In this paper, the evaluation process of surface variables is focus on temperature, surface pressure, relative humidity, and precipitation.

The schemes did not show significant differences among themselves. In previous research for a convective storm over the Nepal Himalayas (Shrestha, Connolly & Gallagher, 2017Shrestha, R.K.; Connolly, P.J. & Gallagher, M.W. 2017. “Sensitivity of WRF Cloud Microphysics to Simulations of a Convective Storm Over the Nepal Himalayas”. The Open Atmospheric Science Journal, 11(1): 29-43, ISSN: 1874-2823, DOI: 10.2174/1874282301711010029.) and a hail event over Surabaya, Indonesia (Sari, Baskoro, & Hakim, 2018Sari, F.P.; Baskoro, A.P. & Hakim, O.S. 2018. “Effect of different microphysics scheme on WRF model: A simulation of hail event study case in Surabaya, Indonesia”. In: Proceedings of the American Institute of Physics Conference, 1987(1), 020002, DOI:10.1063/1.5047287.), the authors reported that the microphysics schemes are not sensitive to surface properties.

From all of the variables, relative humidity has the worst results. It performs poorly in hourly bias (Figure 5). The relative humidity results are not good about 1800 UTC in rainy season (Figure 5a), it is probably because a poor representation of the size and position of storms. For the dry season, about 2100 UTC, during the daily minimum relative humidity, maximum biases are showed (Figure 5b). The rainy season biases are bigger than dry season ones, this behavior strongly indicates the combination of storms relative humidity field deviations and the overestimation of the minimum daily value.

Temperature surface field biases matrix depicted the best skill, the maximum overestimation for both rainy and dry season are at 0000 UTC (Figure 6). According to the Figure 6a, under storms (approximately 2100 UTC) the temperature forecast overestimates observations. Once more, every scheme produced slightly different hourly bias.

The same slightly different microphysics performance is describing by surface pressure (Figure 7). In this case, skills are strongly dependent on the change of surface pressure due to meteorological phenomena. The dry season storm case analysed was derived from the synoptic pattern cold front. One of the main characteristics of this type of phenomenon is the falling surface pressure. A cold front crossed over the study area between 24 and 27 forecast hours. Biases for this period show large bias in the data (Figure 7b).

The Table 4 exhibits the indices that describe the correspondence between forecasts and observations of precipitations events. These values are comparable with those obtained in the evaluations developed for WRF-ARW configurations in Cuba (Sierra et al., 2015Sierra, L.M.; Ferrer, H.A.L.; Valdés, R.; Mayor, Y.; Cruz, R.; Montejo, I.; Rodríguez, C.F.; Rodríguez, N.; Roque, A. 2015. “Automatic mesoscale prediction system of four daily cycles”. Technical Scientific Report. DOI: 10.13140/RG.2.1.2888.1127 ; Pérez-Bello et al., 2019Pérez-Bello, A.; Mitrani, A.I.; Díaz, R.O.; Wettre, C. & Robert, L. H. 2019. “A numerical prediction system combining ocean, waves and atmosphere models in the Inter-American Seas and Cuba”. Revista Cubana de Meteorología, 25(1): 109-120, ISSN: 2664-0880 <http://rcm.insmet.cu/index.php/rcm/article/view/459> [Consulted: January 15, 2021]). In this investigation, the authors verify the precipitation forecast in domains, periods, and from data that lead from the numerical experiments in this paper. However, SisPI precipitation events verification (Sierra et al., 2015Sierra, L.M.; Ferrer, H.A.L.; Valdés, R.; Mayor, Y.; Cruz, R.; Montejo, I.; Rodríguez, C.F.; Rodríguez, N.; Roque, A. 2015. “Automatic mesoscale prediction system of four daily cycles”. Technical Scientific Report. DOI: 10.13140/RG.2.1.2888.1127 ) have a FAR of 0.724 and a CSI of 0.113 for a 27 km domain and microphysics scheme WSM5. The same microphysics in our numerical experiments configurations have FAR results above 0.529 and lower than 0.894. CSI range from 0.113 to 0.347. The extremes CSI values correspond to the June 29, 2012 and January 27, 2019 case studies date. In a perfect skill, CSI should have a maximum value of one, and FAR be equal to zero.

The best performance for June 29, 2012 was obtained with Morrison 2-moment and Lin schemes. In the figure 8 are represented the simulated precipitation field with Morrison at 2100 UTC and the location of the surface weather stations.

Even so, the high detection of precipitation events is probably associated with the fact that Morrison is characterized by producing a wide area of stratiform precipitation associated with storms (Morrison, Thompson & Tatarskii, 2009Morrison, H.; Thompson, G. & Tatarskii, V. 2009. “Impact of cloud microphysics on the development of trailing stratiform precipitation in a simulated squall line: Comparison of one- and two-moment schemes”. Monthly Weather Review, 137(3): 991-1007, ISSN: 1520-0493, DOI: 10.1175/2008MWR2556.1.).

During the dry season, the Lin scheme has the better results. January 27^th exhibits the highest detection of events. The flux imposed by the extratropical cyclone and the LTPF influence propitiated the wide distribution of precipitation areas, which persisted temporarily.

In general, the model did well for temperature, the rest of the variables have superior biases in presence of storms. Large errors in relative humidity were also found by Somos-Valenzuela and Manquehual-Cheuque (2020)Somos-Valenzuela, M. & Manquehual-Cheuque, F. 2020. “Evaluating Multiple WRF Configurations and Forcing over the Northern Patagonian Icecap (NPI) and Baker River Basin”. Atmósfera, 11(815), 19, ISSN: 0187-6236, DOI:10.3390/atmos11080815. over the Northern Patagonian Icecap (NPI) and Baker River Basin. Tested schemes produced slightly different BIAS. Some remarkable differences were depicted for precipitation events verification. It is because the precipitation field should be sensitive to storm features simulation as position, size, and convection intensity.

4.1 LLWS derived from storms sensitivity

June 29^th, 2012 is the storm that caused the highest wind speeds records on the runway, for this case study, we focus on the influence of convection intensity in the wind field at height.

In absence of storms, wind barbs verticals profile not show significant differences. On the other hand, microphysics schemes derived dissimilarities winds barbs verticals profiles for weak and deep convection. We dismiss the effect of the storm position, which enforces both size and intensity simulated features. The larger variations are depicted in lower levels, the same altitudes with maximum reflectivity values. For WSM5 and Lin schemes, under deep convection conditions (Figure 9a) more strong winds were simulated above 1 km. In reflectivity representation, also, WSM5 and Lin performed the maximum reflectivity cores in lower levels (Figure 9b and Figure 9d respectively). Coll-Hidalgo et al. (2021)Coll-Hidalgo, P.; Pérez-Alarcón, A. & González-Jardines, P.M. 2021. “Evaluation of Microphysics Schemes in WRF-ARW Model for Numerical Wind Forecast in José Martí Airport”. In: Proceedings of The 3rd International Electronic Conference on Atmospheric Sciences. <https://sciforum.net/paper/view/conference/8121> [Consulted: July 15, 2021] pointed out perceptible variations in the profile-mixing ratio of hydrometeor particles for a storm over “José Martí” International Airport the June 29, 2012. Finally, our results indicate that each hydrometeor particle concentration predicted by tested microphysics has an influence on vertical wind profiles, particularly, for inferior heights. On behalf of our purpose, it is a substantial reason that aimed to evaluate LLWS derived from storms sensitivity of microphysics schemes.

In figure 10, we can see the Pearson correlation coefficient between the wind shear speed in the 0-2000 ft layer obtained from the AFWA-diagnostics module and the methodology applied in this investigation. It is observed how the correlation between variables is lower for the case of July 3, 2016, while for the rest of the cases the linear relationship oscillating between 0.75 and 1. The configuration with the WSM5 scheme shows the highest correlation in the cases of the rainy season, while Lin exhibits the best correlation in the case January 27, 2019.

Figure 11 shows Taylor diagrams form evaluation of LLWS speed derived from rainy season storms studies cases over “José Martí” International Airport. In WRF-ARW LLWS speed forecast verification with RAP model, all the layers of calculation have a SDs lower than 4.5 m/s, with an average maximum RMSE of 4 m/s. Pearson correlation coefficient shows negative values, schemes often depict a moderate to strong inverse correlation. Therefore, this performance may be related to use spatial resolutions. In our numerical experiments, spatial resolution is 4 km, while RAP model resolution analysis data is 13 km, which make it impossible to capture local features.

Accuracy differences between before-, in- and after- storms LLWS speed forecast are very high when compared to statics among calculation layers. In addition, for the time lapse before- occurrence of storms in the study area all microphysics have similar results, as it can been see in the Taylors diagrams (Figure 11a, d, g, h). At the same period, the highest layer shows weak positive correlation of WSM6 scheme, but whit the biggest SD and RMSE. During storms influence we can see larges variations in microphysics behavior. Microphysics performance depends on which static is considered to evaluate results. In storms lapse, Morrison 2-moment has a poorly correlation, with the lowers SD and RMSE. WSM5 and Lin schemes before- storms have SD and RMSE above 1.50 m/s, which increasing after-storms upper to 2.40 m/s. Is necessary take into account that the designed experiments differ from the RAP configuration in notable aspects for obtaining the variable. The microphysics scheme varies between the configurations, also others such as the surface layer and the planetary boundary layer. The wind-shear obtained from the WRF-ARW is sensitivity to variations in these parametrizations (Storm & Basu, 2010Storm, B. & Basu, S. 2010. “The WRF Model Forecast-Derived low-level wind shear climatology over the United States Great Plains”. Energies, 3, 258-276, ISSN: 1996-1073, DOI: 10.3390/en3020258.; Carvalho et al., 2012Carvalho, D.; Rocha, A.; Gómez-Gesteira, M. & Santos, C. 2012. “A sensitivity study of the WRF model in wind simulation for an area high wind energy”. Environmental Modelling & Software, 33: 23-34, ISSN: 1364-8152, DOI: 10.1016/j.envsoft.2012.01.2019.). Likewise, aspects that the WRF-ARW and RAP experiments do not share, such as data assimilation and the vertical levels used must be considered.

Figure 12 shows wind roses of LLWS at different heights calculated over “José Martí” International Airport. We compute LLWS from RAP analysis, WSM5 and Lin microphysics, because this schemes showed the better performance in our experiments. Wind roses corresponds to dry season study case, in this period, storms were results of a minor convection intensity, and LLWS was more related to cold front wind discontinuities. In our experiments, to obtained wind at upper levels, we interpolate to a grid with longitudinal resolution of 0.87 km and a latitudinal resolution of 0.5 km, from a 4 km resolution data and we applied a consistent mass model correction. For LLWS from RAP data we interpolate directly to point over airport form 13 km resolution.

Near the aerodrome, there is multiple obstacles with heights comparable to the lower layer altitude. LLWS speed is less intense than RAP forecast at this layer. Also, LLWS direction frequency distribution differs more than other heights. Mass model correction was applied for guarantees the conservation of wind direction due to the impenetrability conditions, this is the probably reason of remarkable differences. For the rest of layers, RAP analysis, WSM5 and Lin scheme depict similar LLWS direction distributions frequencies. Wind roses for rainy season study cases did not show the same homogeneity in distributions, it is consequence of LLWS due to local phenomena. Besides tropical waves imposed a synoptic flux, RAP domain not include the south of western Cuba, and LLWS is more relative to small-scale process, therefore, wind roses depict substantial variations (Figure 13).