Microphysical Processes Producing High Ice Water Contents (HIWCs) in Tropical Convective Clouds during the HAIC-HIWC Field Campaign: Evaluation of Simulations Using Bulk Microphysical Schemes

Regions with high ice water content (HIWC), composed of mainly small ice crystals, frequently occur over convective clouds in the tropics. Such regions can have median mass diameters (MMDs) < 300 μm and equivalent radar reflectivities < 20 dBZ. To explore formation mechanisms for these HIWCs, high resolution simulations of tropical convective clouds observed on 26 May 2015 during the High Altitude Ice Crystals High Ice Water Content (HAIC-HIWC) international field campaign based out of Cayenne, French Guiana, are conducted using the Weather Research and Forecasting (WRF) model 5 with four different bulk microphysics schemes: the WRF single-moment 6-class microphysics scheme (WSM6), the Morrison scheme and the Predicted Particle Properties (P3) scheme with oneand two-ice options. The simulations are evaluated against data from airborne radar and multiple cloud microphysics probes installed on the French Falcon 20 and Canadian National Research Council (NRC) Convair 580 sampling clouds at different heights. WRF simulations with different microphysics schemes generally reproduce the vertical profiles of temperature, dew-point temperature and winds during this event compared 10 with radiosonde data, and the coverage and evolution of this tropical convective system compared to satellite retrievals. All of the simulations overestimate the intensity and spatial extent of radar reflectivity by over 30% above the melting layer compared to the airborne X-band radar reflectivity data. They also miss the peak of the observed ice number distribution function for 0.1 <Dmax < 1 mm. Even though the P3 scheme has a very different approach representing ice, it does not produce greatly different total condensed water content or better comparison to other observations in this tropical convective system. Mixed15 phase microphysical processes at−10 ◦C are associated with the overprediction of liquid water content in the simulations with the Morrison and P3 schemes. The ice water content at −10 ◦C increases mainly due to the collection of liquid water by ice 1 https://doi.org/10.5194/acp-2020-1045 Preprint. Discussion started: 31 October 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
High concentrations of small ice particles ingested into jet engines can cause power-loss and damage events (Lawson et al., 1998;Mason et al., 2006). They can also cause air data probe failures (Duviver, 2010). Regions with high ice water content (HIWC), composed of mainly small ice crystals and median mass diameters (MMDs) as low as 300 µm, frequently occur over oceanic convective systems (Mason and Grzych, 2011;Ackerman et al., 2015;Leroy et al., 2016b). Such HIWC regions, with 25 relatively low equivalent radar reflectivities (Z e ) (often less than 20 dBZ; Mason et al., 2006;Fridlind et al., 2015;Protat et al., 2016;Wolde et al., 2016;Leroy et al., 2017), are hard to detect with pilot radars onboard commercial aircraft and are thus potentially hazardous.
In order to explore the processes responsible for the occurrence of HIWC regions and the associated ice crystal properties within tropical convection, the High Altitude Ice Crystals -High Ice Water Content (HAIC-HIWC) international field cam-30 paigns (Dezitter et al., 2013;Strapp et al., 2016a) and the HIWC RADAR campaign  were conducted. Data collected during these campaigns have been being analyzed to understand HIWC conditions (Leroy et al., 2015(Leroy et al., , 2016aProtat et al., 2016;Wolde et al., 2016;Korolev et al., 2020), to develop warning products that can identify HIWC regions 35 (Yost et al., 2018;Bedka et al., 2019;Harrah et al., 2019;Haggerty et al., 2020), and to characterize the high altitude HIWC environment to assess a new ice crystal aircraft certification envelope.
In general, total condensed water content (TWC) values in these campaigns reached as high as 4.1 g m −3 averaged over 0.93 km (0.5 nautical mile) distance scales, and even up to about 2 g m −3 over 185 km distance scales. Average MMDs in HIWC zones greater than or equal to 1 g m −3 increased with temperature, from ∼326 µm at −50 • C to ∼708 µm at −10 • C 40 . Leroy et al. (2015Leroy et al. ( , 2016aLeroy et al. ( , 2017 showed that MMDs decrease with increasing TWC and decreasing temperature, indicating small ice crystals are responsible for HIWC regions at high altitudes for both the Darwin and Cayenne datasets. Wolde et al. (2016) found the relationship between the ice water content (IWC) and radar equivalent reflectivity factor followed a power-law fit with coefficients dependent on temperature. However, they found the pilot X-band weather radar on the Canadian National Research Council (NRC)'s Convair-580 aircraft did not have adequate sensitivity to detect 45 HIWC regions when calibrated using the NRC X-band research radar. Nguyen et al. (2019) proposed a retrieval method for IWC using the specific differential phase (K dp ) and differential reflectivity ratio (Z dr ) data from X-band dual-polarization airborne radar. This method was demonstrated to be superior to the power-law fits between IWC and reflectivity as accounting  (Milbrandt and Yau, 2005) also produced IWC and ice particle number concentration differing from observations during the HAIC-HIWC Cayenne project, which they hypothesized was due to the poor representation of SIP processes in the microphysics scheme. 70 Other tropical convective clouds have also been observed and simulated. For example, McFarquhar and Heymsfield (1996) found the numbers of smaller particles (D < 100 µm) close to the convection were one order of magnitude higher than the numbers found further away according to data obtained during the Central Equatorial Pacific Experiment (CEPEX). Meanwhile, McFarquhar and Heymsfield (1997) indicated the shapes of the ice PSDs in the tropics substantially differ from those in the midlatitudes, especially at temperatures < −40 • C. Lohmann et al. (1995) showed that the simulated average IWC by a 75 coarser resolution (∼125 km × 125 km) general circulation model (GCM) agreed well with the observed IWC during CEPEX, especially with respect to the relationship between IWC and temperature, whereas the model underestimated the variability of simulated IWC within each temperature bin. Chen et al. (1997) indicated the main sources of ice particles are frozen cloud droplets and interstitial aerosol particles, and the number concentration of ice particles is influenced strongly by the amount of condensation nuclei in convective inflows according to the simulations and a sensitivity experiment for cases during CEPEX 80 by using a one-dimensional microphysical model. Ackerman et al. (2015) conducted three 3D cloud-resolving model (CRM) simulations of MCSs observed on 23 January 2006 during the Tropical Warm Pool International Cloud Experiment (TWP-ICE) with bulk and bin microphysics schemes, but were not able to produce HIWC regions (IWC > 2 g m −3 and Z e < 30 dBZ). Lang et al. (2011) greatly reduced model bias of excessively large reflectivity values (e.g., 40 dBZ) in the middle and upper troposphere in the simulation of a continental convective case observed during the Tropical Rainfall Measuring Mission (TRMM) 85 Large-Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) through modifying a single-moment bulk microphysics scheme in the Goddard Cumulus Ensemble model; however, there was much less improvement for an oceanic MCS observed during the TRMM Kwajalein Experiment (KWAJEX).
As indicated by the review above, the numerical studies on HIWC phenomenon to date have not been able to capture HIWC phenomenon well. This has been attributed to biases in particle properties, parameterized PSDs, and microphysical 90 processes. The lack of knowledge about processes generating HIWC regions suggests that further numerical simulations are needed to explore the microphysical pathways producing HIWCs. Qu et al. (2018) indicated MY2 greatly overestimated the graupel content and hypothesized that HIWC will be better estimated by the next generation of microphysics schemes (e.g., the Predicted Particle Properties (P3) microphysics scheme, Morrison and Milbrandt, 2015). Therefore, this study will be the first test of the P3 scheme in simulating the HIWC phenomenon and in comparing P3 to other bulk microphysics schemes to 95 determine whether P3 is better able to predict HIWCs in a high resolution numerical weather prediction (NWP) context. In this study, a tropical oceanic MCS on 26 May 2015, which was well sampled during the Cayenne field campaign, is simulated at a high resolution with 1-km horizontal grid spacing. The numerical simulation experiments and their evaluation are described in this paper. In an upcoming companion paper, attention will be focused on sensitivity experiments varying some parameters within the microphysics scheme to enhance understanding of processes leading to the formation of small crystals in HIWC 100 regions.
The next section describes the tropical oceanic MCS sampled on 26 May 2015. Section 3 introduces the collected data and how they were processed. Section 4 shows the simulated fields and their evaluation against observations. A summary and conclusions are presented in section 5.

105
The tropical MCS observed on May 26 2015 initiated and developed over the tropical Atlantic Ocean north of Cayenne, French Guiana. The MCS was not associated with obvious synoptic-scale flow features, such as clearly identifiable highs or lows. From the soundings at Cayenne, a deep moist absolutely unstable layer (MAUL) existed over the area where the MCS occurred, and the MAUL was maintained as the MCS developed, consistent with the conceptual model of a MCS proposed by Bryan and Fritsch (2000). There were mainly easterly (westerly) winds below (above) 350 hPa. The first convection initiated 110 and developed over the ocean near the coast in the early morning. The convection moved eastward over the course of the day due to the upper westerly winds. Subsequently, new convective cells continually initiated and developed in a similar location and gradually moved eastward to merge with old convective cells that were present over the ocean to form a large and long-lived MCS.
The convective system was sampled by two research aircraft, the French SAFIRE Falcon 20 and Canadian NRC Convair 115 580, during the HAIC-HIWC field campaign. Figure 1 shows the observed brightness temperature from GOES-13 geostationary 4 https://doi.org/10.5194/acp-2020-1045 Preprint. Discussion started: 31 October 2020 c Author(s) 2020. CC BY 4.0 License. satellite channel 4 (10.8 µm) at 1045 UTC 26 May 2015, tracks of the two flights (Fig. 1a), as well as the height of the aircraft above mean sea level and air temperature at the flight levels (Fig. 1b). Both aircraft sampled close to the convective core of the storm as shown by the tracks of the aircraft through the lowest cloud-top brightness temperature (Fig. 1a). The SAFIRE Falcon 20 sampled at three height levels, i.e., ∼7, ∼10 and ∼11.5 km, corresponding to temperatures of about −10, −30 and −45 • C, 120 respectively. The NRC Convair 580 sampled mainly at ∼7 km and a temperature of around −10 • C (Fig. 1b).

Data
The SAFIRE Falcon 20 was equipped with cloud microphysics instrumentation, including a Cloud Droplet Probe (CDP2), Two Dimensional Stereo Imaging Probe (2D-S), Precipitation Imaging Probe (PIP) and Isokinetic Evaporator Probe (IKP-2, Strapp 125 et al., 2016b). The NRC Convair 580 was equipped with an X-band (9.41 GHz) cloud airborne radar (NAX, Wolde et al., 2016) including three antennae (nadir, zenith and side-looking) and similar cloud microphysics instrumentation. The two optical array probes, 2D-S and PIP, recorded 2D images of ice crystals nominally in the size range of 10-1280 and 100-6400 µm, respectively. The diode resolutions of 2D-S and PIP are 10 and 100 µm, respectively. The size distribution data with uncertainty of 10%-100% (Baumgardner et al., 2017) are processed following the general approach described in McFarquhar et al. (2017), 130 with only center-in particles accepted, and corrections for out-of-focus particles (Korolev, 2007), shattered particles (Field and Heymsfield, 2003;Field et al., 2006;Korolev and Field, 2015) and particles partially within the photodiode array applied (Heymsfield and Parrish, 1978). Due to a poorly defined depth of field for small particles (Baumgardner et al., 2012) and the potential of shattered artifacts only n(D) for D max > 50 µm are considered here. Composition PSDs ranging from 0.05 to 12.845 mm merged from the 2D-S and the PIP were derived at a 5-s time resolution with a crossover of 400 µm between 135 probes Leroy et al., 2017). The IKP-2 bulk TWC probe was designed specifically for these campaigns to measure the high-speed, high-TWC environment, up to at least 10 g m −3 at 200 m s −1 , with a target accuracy of 20% Leroy et al., 2017).
Radar reflectivity data from the X-band airborne radar installed on the NRC Convair 580 (Wolde et al., 2016), and TWC measured by the IKP-2, and PSDs measured by the 2D-S and PIP installed on the SAFIRE Falcon 20 and on the NRC Convair

Model setup
The WRF model Version 4.1.3 (Skamarock et al., 2019) is used to simulate the tropical oceanic MCS event on 26 May 2015.

145
Two one-way nested domains with 3-and 1-km horizontal grid spacing and 51 vertical levels are adopted (Fig. 2). The ERA5 reanalysis data available every 1 hr with 0.25 • × 0.25 • horizontal grid spacing (https://rda.ucar.edu/datasets/ds633.0) are used for initial and boundary conditions. The model is run from 0000 to 1800 UTC 26 May 2017 for 18 hr with a spin-up time of the first 6 hr. Physical parameterization schemes include the revised Rapid Radiative Transfer Model (RRTMG) longwave and shortwave radiation scheme (Iacono et al., 2008), the Yonsei University (YSU) planetary boundary layer (PBL) scheme (Hong 150 et al., 2006), the MM5 similarity surface layer scheme (Beljaars, 1995), and the unified Noah land-surface scheme (Tewari et al., 2004). The cumulus parameterization scheme is not activated in this study.
Four bulk microphysics schemes, namely the WRF single-moment 6-class (WSM6) microphysics scheme (Hong and Lim, 2006), the Morrison double-moment scheme (Morrison et al., 2009) and the Predicted Particle Properties (P3) microphysics scheme with one-and two-ice options (Morrison and Milbrandt, 2015;Milbrandt and Morrison, 2016) are used for separate 155 simulations. The simulations using the WSM6 and Morrison microphysics schemes are referred to as the WSM6 and MORR runs hereafter, respectively. These microphysics schemes are used because the PSDs of ice species are parameterized differently in these schemes. Both the WSM6 and Morrison schemes predict the mixing ratios of five cloud hydrometeor species, including cloud water, rainwater, cloud ice, snow and graupel 1 , while the number mixing ratios for all species except cloud water are also predicted in the Morrison scheme. The P3 scheme predicts bulk ice properties (e.g., mean particle density) rather than predicting 160 separate species of ice with fixed properties (e.g., cloud ice, snow and graupel). P3 uses one or more "free" ice categories to represent all ice-phase hydrometeors, which can eliminate the unphysical "conversion" processes between different traditional ice categories (Morrison and Milbrandt, 2015;Milbrandt and Morrison, 2016). In this study, the options of one-and two-ice categories in the P3 scheme are used, referred to as P3-1ICE and P3-2ICE hereafter, respectively. Technically P3-1ICE and P3-2ICE are two configurations of the same scheme, but they have notably different treatments of ice which is the basis on which 165 all the microphysics schemes were chosen, so these are referred to as different schemes. Output data in the model domain d02 with 1-km horizontal grid spacing are analyzed in this study. It should be noted that cloud ice, snow, and graupel in WSM6 and MORR and both two categories of ice in P3-2ICE are treated as ice particles to compare with the observed ice particles, because the observed ice particles are not separated into different categories.

Estimation of X band radar reflectivity 170
The computations of simulated radar reflectivity are performed using the Rayleigh approximation which is applicable at the X-band given the size of typical ice particles (Ryzhkov et al., 2020). The relations for reflectivity from rain (Z r ), graupel (Z g ), snow (Z s ), and ice (Z i ) are derived in detail in Appendix A. The total equivalent radar reflectivity factor (Z e ) in units of dBZ can thus be attained using (1)    5).

215
To obtain a statistical comparison between simulated and observed radar reflectivity, Contoured Frequency by Altitude Diagrams (CFADs) (Yuter and Houze Jr, 1995)  It should be noted that radii from 20 to 200 km in the 20-km intervals were tested, and the results were similar. A radius of 100 km was adopted because the standard deviation between the observed and simulated reflectively was the least when using this radius threshold. The sampling method used here is similar to the one used by Borderies et al. (2018).
In general, the simulations overestimate the X-band radar reflectivity above the melting layer (∼4.7 km). Figure 7 shows 230 CFADs and cumulative CFADs of simulated and observed X-band radar reflectivity above 5 km. The CFADs are shown only above 5 km, because the formulae for calculating the simulated radar reflectivity in section 3.3 do not consider the effect of melting, and this study mainly focuses on HIWC regions. From Fig. 7e, 95% of the observed reflectivities are < 30 dBZ above 6 km. The most frequently observed radar reflectivity is around 25 dBZ at heights of 5-7 km and ∼15 and ∼20 dBZ at a height of ∼8 km (Fig. 7e). The simulated radar reflectivity shows broader distributions with larger values than the observations 235 (95% of the observed reflectivities < 30 dBZ above 6 km), and maxima in radar reflectivity can reach 50 dBZ, especially for WSM6, P3-1ICE and P3-2ICE (Figs. 7a, c and d). There are 95% of the simulated reflectivities < 44 dBZ in WSM6, < 41 dBZ in MORR, < 45 dBZ in P3-1ICE, and < 47 dBZ in P3-2ICE above 6 km. The simulated radar reflectivity in MORR has a narrower distribution with 70% of reflectivities between 34 and 42 dBZ at 5 km (Fig. 7b), which better resembles the observation with 70% of reflectivities between 24 and 36 dBZ at 5 km (Fig. 7e). The other simulations have broader 240 distributions with 70% of the reflectivities between 30 and 44 dBZ in WSM6 (Fig. 7a), between 17 and 46 dBZ in P3-1ICE ( Fig. 7c), and between 25 and 48 dBZ in P3-2ICE (Fig. 7d) at 5 km. The radar reflectivity in all simulations extends above zenith-viewing Doppler airborne radar shows a peak-to-peak correlation between them. This suggests there may be stronger updrafts in the simulations associated with the higher extended simulated reflectivity. Overall, all the simulations overestimate 245 the intensity and spatial extent of radar reflectivity. By examining each component of reflectivity, the overestimation of radar reflectivity above the melting layer in WSM6 and MORR mainly results from the overprediction of graupel (not shown), which is similar to the tropical MCS simulations of Lang et al. (2011) and Qu et al. (2018). The P3 scheme, which was expected to yield better estimates of HIWCs, does not reduce the biases in simulated radar reflectivities. It should be noted that these biases may be also related to the aircraft sampling statistics. The NRC Convair 580 operations had to avoid the cloud regions with 250 high reflectivity due to safety regulations, and thus it did not approach high reflectivity regions (red zones on the pilot's radar) and within 30 nautical miles (∼55.56 km). Therefore, sampling statistics is partly biased due to exclusion of cloud regions with high reflectivity.

Cloud microphysical properties
Samples are selected to examine the observed and simulated cloud microphysical properties using the same method as used   and ∼264% (∼2.5 × 10 5 m −3 mm −1 ) respectively at D max of 0.1 mm (Figs. 9c and d). None of the simulations capture the 275 peak of the observed PSD with the median of ∼2.6 × 10 5 m −3 mm −1 near D max of 0.3 mm (Fig. 9). There are no obvious peaks in PSDs for 0.1 mm < D max < 1 mm in WSM6, MORR, and P3-2ICE (Figs. 9a, b, and d). There is a PSD peak with a median n(D) of ∼9.6 × 10 5 m −3 mm −1 near D max of 0.17 mm in P3-1ICE (Fig. 9c). Medium particles (0.3 mm < D max < 1 mm) are dominant in the observations and WSM6, while small particles make the main contributions in MORR, P3-1ICE and P3-2ICE (Table 1).

280
At the −10 • C level (Fig. 10), all simulations underestimate the median PSD for D max < 1 mm, especially P3-2ICE with an underestimate by ∼94% (∼4.0 × 10 4 m −3 mm −1 ) at D max of 0.1 mm. Large particles contribute 58.6% of N 0.1−12.845mm in P3-2ICE, which is very different from the observations (Table 1). Compared to the observations, MORR has almost the same PSD spread and median for D max of 0.1 mm (Fig. 10b). All of the simulations miss the peak of the observed PSD with a median of ∼1.5 × 10 5 m −3 mm −1 near D max of 0.3 mm (Fig. 10).

285
Overall, the simulated PSDs at the three temperature levels shown in Figs. 8-10 have biases in various degrees with respect to their shapes, magnitudes and spreads compared to observations. The observations are concentrated around the smaller crystal sizes than are the simulations for most of temperature levels expect P3-1ICE at −45 and −30 • C. It should be noted that the PSD comparison strongly reflects different assumptions built into the schemes regarding PSD shapes, namely inverse exponential PSDs are assumed in WSM6 and MORR, whereas a gamma PSD with a shape parameter (µ) that varies with the 290 slope (λ) following the observations of (Heymsfield, 2003) is assumed for the P3 scheme.  . Given the main purpose here is to compare the particle sizes between observation and simulation for simplicity, the definition of D e given by, is used, where K is the number of ice species. Since D is maximum dimension, only one number distribution function that 300 includes all data is used for the observations, and therefore K = 1 for observations. Generally all the simulations, especially MORR, reproduce the TWC reasonably well at the three temperature levels. In particular at the −10 • C level the 25th and 75th percentiles of TWC in all the simulations cover the same order of magnitude as the observations (Fig. 11a). The differences in N 0.1−3mm among the simulations are quite large (Fig. 11b). At the three temperature levels, WSM6 and especially MORR underestimate the number concentration (Fig. 11b). This is associated with 305 the underpredicted small particles and overpredicted large particles in WSM6 and MORR compared to the observations (Figs. 8-10). Thus, WSM6 and MORR produce larger D e compared to the observations at the three temperature levels (Fig. 11c).
At −30 • C, the median N 0.1−3mm values in WSM6 (∼0.4 × 10 5 m −3 ) and MORR (∼0.2 × 10 5 m −3 ) are underestimated by ∼50% and ∼75%, respectively, consistent with the underestimate of particle number near the peak of the observed PSD at D max of ∼0.3 mm (Figs. 9a and b). The N 0.1−3mm at −45 • C in P3-1ICE is about one order of magnitude larger than observed (Fig. 11b) mainly due to many more small particles for 0.1 mm < D max < 0.4 mm (Fig. 8c). Similarly, compared to the observations, P3-1ICE overestimates N 0.1−3mm at −30 • C, with the median overestimated by ∼129% (∼10 5 m −3 ) (Fig.   11b). This is explained by an overestimate of particle number at D max of ∼0.1 mm (Fig. 9c). Accordingly, P3-1ICE produces smaller D e than the observations at −45 and −30 • C with an underestimate of median D e by ∼38% and ∼46% respectively (Fig. 11c). Due to the larger PSD spreads in P3-2ICE at −45 and −30 • C (Figs. 8d and 9d), N 0.1−3mm and D e in P3-2ICE 315 accordingly have a larger spread than the observations (Figs. 11b and c). This occurs even though the spread of TWC between P3-2ICE and observations is similar (Fig. 11a). At −10 • C, values of N 0.1−3mm from all of the simulations, in particular P3-2ICE, are about one order of magnitude smaller than observed (Fig. 11b), implying larger mean particle size than observed ( Fig. 11c). This is mainly attributed to the underestimate of small particle number for D max < 1 mm, especially near the peak of the observed PSD at D max of ∼0.3 mm, and overestimate of large particles in all of the simulations (Fig. 10). The simulated 320 vertical velocity is in general stronger than in the observations, especially at −45 and −10 • C (Fig. 11c), corresponding to the higher extent of simulated radar reflectivity (Fig. 7).

Cloud microphysical processes
As discussed above, all the four microphysics schemes underpredict the number concentration by about one order of magnitude at −10 • C compared to the observations (Fig. 11b), although they predict similar TWC to the observed TWC (Fig. 11a). In this here. The lack of small ice crystals in HIWC regions at this temperature would extend to a lack of small ice crystals at higher 335 altitudes. The subsamples whose observed and corresponding simulated TWCs at −10 • C are larger than 1 g m −3 are selected from the total samples (1778) at −10 • C (Fig. 10) to conduct a composite analysis. There are 509, 488 and 427 paired samples selected for MORR, P3-1ICE and P3-2ICE, respectively (Table 2).
From Table 2 . Therefore, both MORR and P3, in particular P3-2ICE, substantially underpredict the ice particle number for 0.1 mm < D max < 3 mm and overpredict the vertical motion in the HIWC regions, which results in stronger and higher-extended simulated radar reflectivity (Fig. 7).

350
From the vertical profiles of water content for each hydrometeor class in MORR (Fig. 12a), graupel is dominant above the 0 • C layer up to ∼8-km height, with snow being second most important. Snow is the dominant hydrometeor category above 8 km in terms of mass content. The aforementioned underestimate of ice particle number concentration at −10 • C is associated with the total number concentration (< 4 × 10 4 m −3 ) of cloud ice, snow and graupel (Fig. 12b). According to Eq. (A17), under the condition of similar simulated and observed TWCs, an underestimate of ice particle number concentration, especially graupel, From vertical profiles of microphysical conversion rates in MORR (Fig. 13), the main source terms of cloud ice content are ice nucleation at −45 • C and vapor deposition at −30 and −10 • C, and the main sink terms are collection by snow and 375 autoconversion to snow. The net conversion rate of cloud ice (Qi_TEND, sum of all microphysical conversion rates including sedimentation) at −10 • C is negative (Fig. 13a). The net number concentration tendency of cloud ice (Ni_TEND) at −10 • C in MORR is ∼−20 m −3 s −1 , mainly due to the accretion of cloud ice, autoconversion to snow and sublimation (Fig. 13d).
The collection of cloud water by snow to form graupel is the main production term of graupel particle number, while it is 385 offset by the sedimentation and sublimation terms. Finally, the net number concentration tendencies of both snow and graupel (Ns_TEND and Ng_TEND) are near 0 m −3 s −1 at −10 • C (Figs. 13e and f). Therefore, the collection of cloud water by graupel is the key source term of total IWC at −10 • C in MORR, which increases the mean mass/size of graupel and does not directly influence its number. This is associated with the strong simulated reflectivity above the melting layer.
From vertical profiles of microphysical conversion rates in P3-1ICE (Fig. 14a), the main production terms of ice content are 390 vapor deposition at −45 and −30 • C, collection of cloud water by ice, vapor deposition and collection of rain water by ice at −10 • C. As for the first ice category in P3-2ICE, in addition to the same main production terms as in P3-1ICE, there is another source term merging from the second ice category due to similar mean mass-weighted diameters between the two ice categories (Fig. 14c). The net tendency of the two ice categories in P3-2ICE (i.e., Qi_TEND + Qi2_TEND) is ∼1.05 × 10 −3 g m −3 s −1 at −10 • C, which is much larger than that in P3-1ICE (∼0.02 × 10 −3 g m −3 ). It is mainly due to the stronger collection of 395 cloud water and rain water by ice in P3-2ICE, which may be associated with the greater cloud water and rain water content at −10 • C in P3-2ICE than P3-1ICE (Figs. 12c and e). The aforementioned collection of cloud water and rain water by ice does not increase the ice particle number. Although the deposition nucleation can increase the ice particle number in P3-1ICE and P3-2ICE, it is small (less than 0.5 m −3 s −1 ) at −10 • C. Merging ice categories does not increase the total ice particle number in P3-2ICE. The sedimentation of ice number is ∼3.3 and ∼3.4 m −3 s −1 at −10 • C in P3-1ICE and P3-2ICE respectively, 400 which dominates the net ice number concentration tendencies at −10 • C in both P3-1ICE (∼2.8 m −3 s −1 ) and P3-2ICE (∼3.1 m −3 s −1 ). The much lower number concentration in P3-2ICE than P3-1ICE (Table 1) is likely due to aggregation associated with collection between the two ice categories in P3-2ICE (Fig. 14).
To summarize briefly, due to the overprediction of LWC in MORR, P3-1ICE and P3-2ICE above the melting layer, there exist obvious mixed-phase processes at −10 • C. The IWC at −10 • C increases mainly due to the collection of liquid water by ice 405 particles, which does not increase ice particle number but increases the size of ice particles. The lower ice particle numbers in the simulations could also be associated with excessive aggregation and/or missing SIPs,such as collision-induced breakup and "freezing-drop-shattering" proposed by Korolev et al. (2020). The large ice particles and lower ice particle numbers contribute to strong simulated radar reflectivity. Introduction of parameterizations for the missing SIPs may be able to overcome some of these model limitations, as will be examined in the future.
A tropical oceanic convective system observed on 26 May 2015 during the HAIC-HIWC international field campaign based out of Cayenne, French Guiana was simulated using the WRF model. Observation data from radiosondes, GOES-13 geostationary satellite, airborne radar, and cloud microphysics instrumentation were used to assess the simulated convective system in terms of the thermodynamic and dynamic environment, storm coverage, evolution and structure, and microphysical properties. The 415 major results are summarized as follows: (1) By comparing simulated and observed soundings, all of simulations using different microphysics schemes replicate temperature with average bias within 1.6%, dew-point temparture with average bias within 6%, wind speed with average bias within 14% and wind direction with average bias within 36 • , with the MORR scheme giving closest agreement with observations.

420
(2) WRF basically reproduces the coverage and evolution of this tropical MCS based on a comparison between simulated and observed brightness temperature with the average bias in storm coverage (brightness temperature < 232 K) by ∼−34.3% in WSM6, ∼30.0% in MORR, ∼12.9% in P3-1ICE, and ∼2.3% in P3-2ICE. Thus, WSM6 underestimates the storm coverage, and P3-2ICE produces the closest storm coverage to the observation.
(3) In general, all of the simulations overestimate the intensity and spatial extent of radar reflectivity above the melting layer 425 compared to the observed X-band radar reflectivity. There are 95% of the simulated reflectivities < 44 dBZ in WSM6, < 41 dBZ in MORR, < 45 dBZ in P3-1ICE, and < 47 dBZ in P3-2ICE above 6 km, while 95% of the observed reflectivities are < 30 dBZ above 6 km. The radar reflectivity > 0 dBZ in all simulations extends above 14 km, whereas the observed radar reflectivity > 0 dBZ is mainly below 14 km. (6) Mixed-phase processes play an important role at −10 • C due to the overprediction of LWC in MORR, P3-1ICE and 440 P3-2ICE above the melting layer. Stronger simulated radar reflectivity in MORR above the melting layer results from large graupel particles associated with greater graupel water content and fewer graupel particles compared with in-situ observations.
Rapid growth of graupel mass mainly through collecting cloud water but with limited increase in graupel number mainly by conversion of snow to graupel through collection of cloud water above the melting layer leads to large mean graupel sizes in water while the net ice number concentration tendencies are near 0, which generates large mean ice particle sizes. The large ice particles generate strong radar reflectivity, partially explaining the bias of simulated radar reflectivity with P3-1ICE and P3-2ICE.
It should be noted that simulations of deep convection at different model resolutions can be much different. To examine the sensitivity of model resolution, CFADs and cumulative CFADs of radar reflectivity are also calculated using the simulation 450 data from the 3-km domain (Fig. 15). Although there are some differences in specific values of reflectivity, the intensity and distribution of reflectivity from the 3-km simulations (Fig. 15) are basically consistent with those from the 1-km simulations ( Fig. 7). Although this does not prove that the conclusions with respect to the differences in behavior among the microphysics schemes will be the same at all resolutions, it does at least indicate that the results in this study have some generality. It provides something useful about the microphysics schemes to numerical forecast guidance for HIWCs in current high-resolution NWP 455 models, which are now routinely run at O(3 km) and now more and more often at O(1 km). However, a caveat is that 1 km is not cloud-resolving O(100 m), thus horizontal entrainment is still not being resolved at this resolution, which affects the amount of liquid available for riming growth in updrafts. Therefore, there still exist uncertainties in the 1-km simulations.
In conclusion, the Morrison and P3 microphysics schemes generally outperform the WSM6 scheme in simulating this tropical oceanic MCS as evident from examining the simulated soundings, brightness temperature, radar reflectivity, ice particle 460 size distributions, total water content and number concentration. However, the Morrison scheme underestimates the number concentration at different temperature levels compared to the observations. This indicates that large ice particles, especially graupel, are overpredicted in this scheme, which is similar to Qu et al. (2018)'s simulation of a different tropical MCS using a different model and microphysics scheme. Even though the P3 scheme has a much different approach for representing ice, it does not produce greatly different TWC or better comparison to the observations using either one-or two-ice categories.

465
This suggests that other aspects need to be considered, such as microphysical process rate formulations or parameters. To enhance understanding of processes leading to the formation of small crystals in HIWC regions, sensitivity experiments varying parameters within the P3 microphysics scheme (e.g., mass-dimensional relations, size distribution parameters, microphysical conversion rates or representation of different processes like secondary ice production) will be examined in a future paper.
Code and data availability. Appendix A: Formulae for X band radar reflectivity factor Most cloud models with bulk microphysics parameterizations predict either mass content (one-moment schemes) or mass content and number concentration (two-moment schemes) for a number of hydrometeor categories. In one-moment schemes, number concentration can be obtained diagnostically. They also commonly assume inverse exponential or gamma size distri-butions of hydrometeors with respect to particle maximum dimension. This allows the use of simple analytical formulae for converting mass content and number concentration to radar reflectivity factor if the scatterers are small compared to the radar wavelength so that the Rayleigh approximation can be used. A gamma size distribution is represented by where N 0 , µ and λ are the intercept, shape and slope factors respectively. The total number concentration N t is thus given by

A1 Calculation of radar reflectivity factor for Rain
For rain, the radar reflectivity factor Z in the Rayleigh approximation is given by the well known formula where D eq is the equivolume diameter of raindrop, and D eq = D for liquid water species in bulk schemes. The liquid water 485 content LW C is defined as where ρ w is the density of water. Taking into account Eq. (A2), integration of Eq. (A3) and Eq. (A4) yields and which results in the following formula for estimating Z from LW C and N t with an inverse exponential size distribution assumption (µ = 0) in WSM6, MORR and P3: A2 Calculation of radar reflectivity factor for ice particles

495
The radar reflectivity factor for ice is defined as (Ryzhkov and Zrnić, 2019, Eq. (5.14)), where ρ i is the density of solid ice sphere, assumed here to be 0.917 g cm −3 , and where ε w and ε i are the dielectric constants of water and solid ice, and |K w | 2 = 0.930 and |K i | 2 = 0.176 in this study. Taking 500 into account that the mass of ice particle m s can be expressed as Eq. (A8) can be written in a different form often used in cloud models (e.g., Hogan et al., 2006): where D eq is the equivalent volume diameter of an ice particle (Ryzhkov and Zrnić, 2019).

505
The m s -D max relations are commonly used (e.g., Locatelli and Hobbs, 1974;Mitchell, 1996;Finlon et al., 2019;Ding et al., 2020) where D max is the maximal dimension of ice particles. These relations are often represented as power-law dependencies Then Eq. (A11) can be rewritten as where a 0 = aη b . In Eq. (A13), the difference between maximal dimension D max and equivolume diameter D eq is taken into account with a scaling factor η = D max /D eq assuming a constant aspect ratio of ice particles across the size spectrum.

A2.1 Constant m-D relation across the size spectrum
In WSM6 and MORR, a and b in the m s -D max relations are assumed to be constant across the size spectrum for snow and 515 graupel particles, implying densities of snow and graupel are constant. This leads to the following expressions for Z and ice water content IW C: and 520 so that It is important that Eq. (A16) is not sensitive to the variability of the prefactor a in the m s (D max ) power-law relation (A12) and is minimally affected by the variability of the exponent b in Eq. (A12). In WSM6 and MORR, b = 3 and µ = 0, thus Eq.
(A16) can be simplified as 525 Z = 720 A2.2 Variable m-D relation across the size spectrum P3 represents each ice category much differently than WSM6 and MORR. In P3, a and b in the m s -D max relations [Eq.

530
Author contributions. YH, WW, GMM designed the study. YH did the calculations, with support from WW, GMM, XW and HM. AR developed the method for calculating X-band radar reflectivity. MW, CN, AS, AVK, and IH processed the original observational datasets.
YH wrote the original draft with contributions from all coauthors, and all coauthors contributed to review and editing.
Competing interests. The authors declare that they have no conflict of interest.
Acknowledgements. This work was supported by the National Science Foundation (Award Numbers: 1213311 and 1842094