Evaluating the performance of a WRF microphysics ensemble through comparisons with aircraft observations

Yun Fu ,Henghi Lei ,Jie-fn Yng ,Zhibo Go

a Key Laboratory of Cloud-Precipitation Physics and Severe Storms, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China

b University of Chinese Academy of Sciences, Beijing, China

c State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing, China

Keywords:Aircraft observation Microphysics ensemble Particle number concentration Liquid/ice water content

ABSTRACT Aircraft observation data obtained in a mesoscale convective system are compared to Weather Research and Forecasting (WRF) model simulations using four microphysics schemes (Morrison, WSM6, P3, SBM) with different complexities.The main purpose of this paper is to assess the performance of the microphysics ensemble in terms of cloud microphysical properties.Results show that although the vertical distributions of liquid water content(LWC) and ice water content (IWC) simulated by the four members are quite different in the convective cloud region, they are relatively uniform in the stratiform cloud region.Overall, the results of the Morrison scheme are very similar to the ensemble average, and both of them are closer to the observations compared to the other schemes.Besides, the authors also note that all members still overpredict the LWC by a factor of 2–8 in some regions, resulting in large deviation between the observation and ensemble average.

1.Introduction

Clouds affect the amount of solar radiation balance reaching the Earth’s surface and the atmospheric water cycle process through changes in their horizontal extent and vertical structure ( Lohmann and Feichter 2005 ).Thus, accurately forecasting cloud properties is important for numerical weather prediction (NWP) models.However, the uncertainties in the model’s initial conditions (ICs) and boundary conditions (BCs),and imperfections in the model’s microphysics schemes, will result in the largest uncertainties of cloud simulation in NWP models ( Xu et al.,2019 ).Numerous studies (e.g., Stensrud et al., 1999 ; Bright and Mullen 2002 ) have shown that an ensemble forecast, in terms of its ensemble mean, performs comparably to, or better than, a similarly single deterministic model in aspect of short-range forecasting of the precipitation.

Ensemble forecast methods include perturbation of the ICs, BCs, and physics schemes.Stensrud et al.(2000) demonstrated that both ICs/BCs and physics perturbations play important roles in ensemble modeling,but that physics perturbations are more skillful than the ICs/BCs when the large-scale forcing is relatively weak.

Li and Tang (2012) used the Weather Research and Forecasting(WRF) model with different perturbed physics schemes to analyze the forecasting performance of meteorological fields in southeastern China.They reported the ensemble mean performed better than any single model member for most meteorological fields.Ji et al.(2014) evaluated the ability of WRF using multi-physics ensembles to simulate storm systems that develop offthe eastern coast of Australia.They demonstrated that the ensemble average gives higher skill than the median performer within the ensemble.

However, the studies mentioned above mainly focused on the forecasting of precipitation and meteorological fields in terms of temperature, wind, humidity, and pressure; there are relatively few studies that have compared with aircraft observations to study the simulation ability of the ensemble forecast in aspects of fine structures of cloud microphysical properties, such as the vertical distributions of liquid water content/ice water content (LWC/IWC) and number concentrations.

In this paper, we utilize the WRF model, version 4.0, including four microphysics schemes, to simulate a mesoscale convective system (MCS)precipitation event that occurred on 22 May 2017 in Hebei Province,China.Bulk parameters in terms of LWC, IWC, and particle number concentrations obtained by airborne instruments in different regions are compared to ensemble model results.The main purpose of the study is to evaluate the performance of the microphysics ensemble to simulate the cloud microphysical properties.

Table 1 Summary of instruments and parameters used in this study.

2.Data and methods

2.1. Aircraft observation

On 22 May 2017, five spiral flight observations were performed in different regions of an MCS to obtain the cloud microphysical characteristics.In this study, we focus on the second spiral (from 0810 UTC to 0823 UTC), the third spiral (from 0832 UTC to 0850 UTC), and the fourth spiral (from 0925 UTC to 0940 UTC).The airborne particle measurement system used in the study comprised a cloud droplet probe(CDP), a cloud imaging probe (CIP), a high-volume precipitation spectrometer (HVPS), a cloud particle imager probe (CPI), a Nevzorov LWC probe, a Nevzorov total water content (TWC) probe, and an Aircraft-Integrated Meteorological Measurement System.Detailed information regarding each probe is provided in Table 1.In order to include all small and large particles with diameters from the submillimeter to centimeter scale, the cutoffsize between CIP and HVPS was set to 900μm and only particles greater than 100μm were taken into account because of the inaccuracy of the depth of field in diameters less than the 100-μm size range ( Baumgardner and Korolev 1997 ).The IWC was calculated by TWC minus LWC.The Nevzorov TWC probe may have had some errors, ranging from 10% to 20% (e.g., Korolev et al., 2013 ).The average and median values of every 500 m are taken in this paper to reduce random errors due to the small sample volume of airborne probes.

2.2. Model configuration

In this study, the initial and lateral boundary conditions were taken from the 1°×1°final global analysis of the National Centers for Environmental Prediction, while the WRF model with a three-way interactive grid-nesting procedure was used.The simulation time was from 0600 UTC to 1200 UTC 22 May 2017.The coarse grid domain of 106 × 106 grid points had a horizontal grid spacing of 9 km, the intermediate grid domain of 151 × 151 grid points had a horizontal grid spacing of 3 km, and the inner grid domain of 190 × 250 grid points had a horizontal grid spacing of 1 km.All model simulations had 40 vertical levels.

Other model configurations were as follows: the long-/shortwave radiation schemes used were RRTM/Dudhia, and land the surface scheme was the Noah land surface model.The planetary boundary layer physics scheme was the YSU (Yonsei University) scheme, while cumulus convection parameterization was only used in the coarse domain.To investigate the potential effects of different microphysics schemes on cloud properties, four schemes with different complexities were employed, including a single-moment scheme (WSM6), a double-moment scheme (Morrison), a predicted particle properties scheme (P3), and a spectral bin microphysics scheme (SBM).Detailed descriptions of these schemes can be found in Hong and Lim (2006),Morrison et al.(2009) ,Morrison and Milbrandt (2015),and Khain et al.(2010).The microphysical properties predicted by the four microphysics schemes and their detailed information are summarized in Table 2.We chose these four microphysics schemes because they not only include the bulk scheme with different complexities, but also the detailed bin scheme.

3.Results

3.1. Precipitation and radar reflectivities

The 1-h cumulative precipitation (0800–0900 UTC, 0900–1000 UTC)of the ensemble average and observation is shown in Fig.1 (a–d).The locations of the second, third, and fourth aircraft spiral observations are marked by the black rectangles.The results illustrate that the simulated precipitation distribution is very similar to the observation.Stratiform precipitation is distributed in Shanxi and Hebei provinces from southwest to northeast, and the concentration of convective precipitation is mainly in the southwest of Hebei Province.Despite some deviation in intensity and position, the main structure of this MCS precipitation is roughly reproduced by the ensemble average.The aircraft is in the convective cloud (CC) in the second and third spirals and in the stratiform cloud (SC) in the fourth spiral.

Fig.1 (e–g) demonstrate the time series of the maximum reflectivities of the radars and from the microphysics ensemble forecasting within the observation areas (dashed black lines) during the second, third, and fourth flight spiral observations.It can be seen that the true cloud system may develop faster than the simulated results, so it causes some errors between the simulated and observed radar reflectivity.In general, the maximum radar reflectivities simulated by each member differ little, except for the time lag of the SBM scheme in the third spiral.The observed radar reflectivity is very similar to the ensemble average results, especially in the fourth spiral.The second and third spirals of observation can be regarded as the CC, since the radar reflectivity is over 35 dBZ in most parts, which could be taken as the threshold of a convective echo.The fourth spiral of observation, with a uniform reflectivity of 30 dBZ,can be regarded as the SC.In this paper, we compare the observational and predicted cloud microphysical characteristics in the second, third,and fourth spirals.

3.2. LWC

Fig.2 demonstrates the vertical distributions of the LWC from the observation and microphysics ensemble.It shows that the results of the average values (every 500 m) are slightly larger than or similar to the results of the median values (every 500 m).The observation values below are taken as the average values.It can be seen that the simulation results of each microphysics scheme are quite different in the CC region,but are relatively uniform in the SC region.The simulation results of SC are closer to observations and the value of LWC is smaller compared to CC.

Fig.1.(a, b) 1-h cumulative precipitation of the observation and ensemble average at 0800–0900 UTC 22 May 2017.(c, d) 1-h cumulative precipitation of the observation and ensemble average at 0900–1000 UTC.The black rectangles represent the locations of the second, third and fourth spiral aircraft observations.The hourly 0.1°precipitation data are from the CMORPH dataset.(e–g) Time series of maximum radar reflectivity within three spiral observation regions and that from the microphysics ensemble.The dashed black lines represent the start time and end time of the observation.The maximum composite radar reflectivity is used as the observation data.

Table 2 Summary of microphysics schemes used in this study.Q c,Q r,Q i,Q s,and Q g represent the mass mixing ratios of cloud, rain, ice, snow, and graupel,respectively.N n,N c,N r,N i,N s,and N g represent the number concentrations of cloud condensation nuclei, cloud, rain, ice, snow, and graupel, respectively.

Fig.2.(a) Vertical distributions of LWC in the second spiral.The light gray line represents the LWC from the different microphysics schemes.The red line is the ensemble average.The black line represents the result of the observation(average values every 500 m) and the standard deviations at both ends.The green line represents the result of the observation (median values every 500 m)and the standard deviations at both ends.The mixing ratios of cloud and rain are used as the results of the simulation.(b) As in (a), but in the third spiral.(c)As in (a), but in the fourth spiral.

Fig.3.(a) Vertical distributions of IWC in the second spiral.The light gray line represents the IWC from the different microphysics schemes.The red line is the ensemble average.The black line represents the result of the observation(average values every 500 m) and the standard deviations at both ends.The green line represents the result of the observation (median values every 500 m)and the standard deviations at both ends.The mixing ratios of ice, snow, and graupel are used as the results of the simulation.(b) As in (a), but in the third spiral.(c) As in (a), but in the fourth spiral.

The height of the 0 °C isotherm level observed during three spirals is different (3700–4000 m).In the CC region (second and third spiral),the simulated LWC in the third spiral is greater than that in the second spiral, especially the height below the 0 °C isotherm ( Fig.2 (a and b)).The LWC simulated by the SBM scheme in the third spiral is the smallest compared to the other microphysics schemes due to the time lag of the SBM scheme, as shown in Fig.1 (f).In general, the ensemble average simulation in the CC region is closest to the observation results, and the simulation results of the Morrison scheme are better than the other microphysics schemes.A similar result can also be found in the SC region(fourth spiral).The simulation results of all microphysics members are slightly larger than observation in the fourth spiral, and the supercooled liquid water changes with height simulated by the ensemble average basically coincide with the observation ( Fig.2 (c)).Besides, we also note that, for the third and fourth spirals, all members overpredict the LWC by a factor of 2–8, especially the height below the 0 °C isotherm (warm cloud), leading to the results of the ensemble average being greater than those of the observation.

3.3. IWC

The vertical distributions of the IWC in three spirals are demonstrated in Fig.3.In the second spiral, the simulated IWC by the Morrison and SBM schemes is slightly larger than the observation, while the simulated IWC by the WSM6 and P3 schemes is smaller.The results of the ensemble average are closer to the observation compared to each member.For instance, the observed IWC is 0.45 g m?3at 5050 m,while the value of IWC simulated by the ensemble average is 0.38 g m?3( Fig.3 (a)).In the third spiral, except for the SBM scheme owing to the time lag, the simulated results are larger than observed.Compared with the different microphysics schemes, the results of the ensemble average are more robust compared to each member ( Fig.3 (b)).However, in the SC, the results of the different microphysics schemes are quite similar and the simulated IWCs are almost two times larger than observed.For instance, the observed IWC is 0.28 g m?3at 4550 m,while the value of IWC simulated by the ensemble average is 0.52 g m?3( Fig.3 (c)).

It should be noted that in the cold part of the CC region (above 5000 m), the observed IWC curve slightly increases with height, while the simulated IWC curves decreases with height.Similar to the LWC,the ensemble average simulation in the CC and SC regions is relatively better, being closest to the observation results, and the results of the Morrison scheme are relatively better than for the other microphysics schemes.

Fig.4.(a) Vertical distributions of the ice-phase particle and rain number concentration in the second spiral.The light gray line represents the ice-phase particle and rain number concentration from the different microphysics schemes.The red line is the ensemble average.The black line represents the result of the observation (average values every 500 m) and the standard deviations at both ends.The green line represents the result of the observation (median values every 500 m) and the standard deviations at both ends.The number concentrations of rain, ice, snow, and graupel are used as the results of the simulation.(b) As in (a), but in the third spiral.(c) As in (a), but in the fourth spiral.

Fig.5.(a) Vertical distributions of the cloud number concentration in the second spiral.The light gray line represents the cloud number concentration from the different microphysics schemes.The red line is the ensemble average.The black line represents the result of the observation (average values every 500 m)and the standard deviations at both ends.The green line represents the result of the observation (median values every 500 m) and the standard deviations at both ends.(b) As in (a), but in the third spiral.(c) As in (a), but in the fourth spiral.

3.4. Number concentration

Because the WSM6 scheme is a single-moment scheme, which does not predict the particle number concentration, and the Morrison scheme does not predict the cloud drop number concentration, three members are used for the prediction of ice-phase and rain number concentration( Fig.4 ), and two members are used for the prediction of cloud number concentration ( Fig.5 ).Fig.4 shows the vertical distributions of icephase particle and rain number concentration in three spirals.It shows that the results of the simulation reproduce the observations both in terms of trends and orders of magnitude in the three spirals.Meanwhile,the simulated results of SC are better than those of CC.No matter which spiral, there is little difference in the raindrop number concentration below the 0 °C isotherm simulated by the different microphysics schemes,but there are big differences in ice-phase number concentration above the 0 °C isotherm.The possible reasons are as follows: The nucleation scheme for ice crystals adopted by different microphysics schemes may lead to these results.The nucleation schemes used in Morrison and P3 are parameterized following Cooper (1986) and the number concentration of ice nuclei (IN) is a function of temperature, while the nucleation scheme used in SBM is described following Meyers et al.(1992) and the number concentration of IN is a function of supersaturation with respect to ice.Thus, the predicted number concentrations of IN in the Cooper scheme are greater than in the Meyers scheme, especially for very cold temperatures.Moreover, the ice-phase process is more complicated, mainly including riming, accretion, and melting processes than the warm-rain process.

In general, compared with the observation, the particle number concentrations predicted by the SBM scheme are smaller, while the results of the P3 scheme tend to overpredict particle number concentrations,and the Morrison scheme is closer to the observation.Moreover, regardless of which spiral, the results of the ensemble average and Morrison scheme are better compared with the other schemes.

Fig.5 shows the vertical distributions of cloud number concentration obtained during three spirals.It shows that the results of the simulation are closer to the observations both in terms of trends and orders of magnitude in the three spirals, and the simulated results of SC are better than those of CC.The cloud number concentrations in CC are larger than in SC, both for aircraft observation and the model simulation.Compared with the P3 scheme, the SBM scheme is better and the ensemble average is an effective means to reproduce the observation.It should be noted that if we take the value of 106m?3 as the cloud region, the range of CC simulated by the ensemble average is larger than that of the observation, which extends to 6500 m in the second spiral and 7000 m in the third spiral, while the cloud region of observation extends to 5500 m whether in the second or the third spiral ( Fig.5 (a and b)).However,the range of SC simulated by the ensemble average is similar to that of observation, which extends to 4800–5100 m ( Fig.5 (c)).

4.Conclusions

In this paper, the observed microphysical properties, in terms of LWC, IWC, and particle number concentration, in different regions of an MSC, are compared to the microphysics ensemble to assess its capability of predicting cloud microphysical properties.The results can be summarized as follows:

(1) The vertical distributions of LWC and IWC simulated by the WSM6,Morrison, P3, and SBM schemes are quite different in the CC region,but they are relatively uniform in the SC region.Overall, the results of the Morrison scheme and ensemble average are closer to the observations compared to the other schemes.

(2) For the third and fourth spiral, all members overpredict the LWC by a factor of 2–8, especially at heights below the 0 °C isotherm,leading to the results of the ensemble average being greater than those of the observation.Moreover, the observed IWC curve slightly increases with height in the cold part of the CC region, while the simulated IWC curves decrease with height.

(3) The vertical distributions of ice-phase and rain number concentration can be reproduced by the ensemble average of three members(the Morrison, P3, and SBM schemes).Moreover, the range of CC simulated by the ensemble average of two members (the P3 and SBM schemes) is larger than that of the observation, while the range of SC simulated by the ensemble average of two members is similar to the observation.

Although we realize that the potential importance of model microphysics differences to the creation of ensembles needs to be explored vigorously, the results presented in this paper are only based upon one aircraft observation case; more data –especially those obtained in deeply developed weather systems –will be utilized to compare with the microphysics ensemble in a future study.

Funding

This work was supported by the National Key R&D Program of China[grant number 2018YFC1507900 ] and the Demonstration Project of Artificial Precipitation Enhancement and Hail Suppression Operation Technology at the Eastern Side of the Taihang Mountains [grant number hbrywcsy-2017-2 ], and was sponsored by the National Natural Science Foundation of China [grant numbers 41530427 and 41875172 ].