A persistent, weakly forced, horizontally extensive mixed-phase boundary layer cloud observed on 4–5 May 1998 during the Surface Heat Budget of the Arctic Ocean (SHEBA)/First International Satellite Cloud Climatology Project (ISCCP) Regional Experiment–Arctic Clouds Experiment (FIRE–ACE) is modeled using three different bulk microphysics parameterizations of varying complexity implemented into the polar version of the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5). The two simpler schemes predict mostly ice clouds and very little liquid water, while the complex scheme is able to reproduce the observed persistence and horizontal extent of the mixed-phase stratus deck. This mixed-phase cloud results in radiative warming of the surface, the development of a cloud-topped, surface-based mixed layer, and an enhanced precipitation rate. In contrast, the optically thin ice clouds predicted by the simpler schemes lead to radiative cooling of the surface, a strong diurnal cycle in the boundary layer structure, and very weak precipitation. The larger surface precipitation rate using the complex scheme is partly balanced by an increase in the turbulent flux of water vapor from the surface to the atmosphere. This enhanced vapor flux is attributed to changes in the surface and boundary layer characteristics induced by the cloud itself, although cloud–surface interactions appear to be exaggerated in the model compared with reality. The prediction of extensive mixed-phase stratus by the complex scheme is also associated with increased surface pressure and subsidence relative to the other simulations. Sensitivity tests show that the detailed treatment of ice nucleation and prediction of snow particle number concentration in the complex scheme suppresses ice particle concentration relative to the simpler schemes, reducing the vapor deposition rate (for given values of bulk ice mass and ice supersaturation) and leading to much greater amounts of liquid water and mixed-phase cloudiness. These results suggest that the treatments of ice nucleation and the snow intercept parameter in the simpler schemes, which are based upon midlatitude observations, are inadequate for simulating the weakly forced mixed-phase clouds endemic to the Arctic.
Clouds remain one of the largest uncertainties in climate and weather modeling. The role of clouds is even less well understood in the Arctic due to sparse observations (Curry et al. 1996). Simulations of Arctic cloud cover and precipitation differ widely among models and often fare poorly compared with the limited observations available (e.g., Curry et al. 1996, 2000; Tao et al. 1996; Walsh et al. 2002; Zhang et al. 2002). These modeling difficulties have been attributed to cloud characteristics in the Arctic that are unique in many ways compared with cloud characteristics observed in other regions (e.g., Curry et al. 1996, 2000). A correct simulation of cloud properties is needed to address Arctic cloud–radiative–surface interactions that may impact global climate (e.g., Lynch et al. 1995; Curry et al. 1996). The important role of the Arctic in global climate and uncertainty in model simulations motivated the 1997–98 Surface Heat Budget of the Arctic Ocean (SHEBA; Uttal et al. 2002) and the First International Satellite Cloud Climatology Project (ISCCP) Regional Experiment–Arctic Clouds Experiment (FIRE–ACE) (Curry et al. 2000). These projects used research aircraft and ground-based remote sensors to characterize cloud properties over the central Arctic Ocean.
One of the important findings during SHEBA/FIRE–ACE was the observed frequency and persistence of supercooled liquid water and mixed-phase stratus (MPS) throughout the year (Curry et al. 2000; Shupe et al. 2001; Intrieri et al. 2002; Shupe and Intrieri 2004). Liquid water was present 73% of the time based on lidar depolarization measurements (Intrieri et al. 2002). There is no simple temperature–phase relationship describing supercooled Arctic clouds. Liquid water was observed by aircraft at temperatures as low as −23°C, while ice was found at temperatures as warm as −4°C (Curry et al. 2000). Intrieri et al. (2002) infer the presence of liquid water at temperatures as low as −34°C using lidar depolarization ratios measured during SHEBA. Correctly predicting cloud phase is crucial in climate and weather simulations since liquid water dominates the radiative effects of clouds (Shupe and Intrieri 2004; Zuidema et al. 2005) and strongly impacts the vertical structure of the boundary layer (BL) through cloud-top radiative cooling and the generation of negative buoyancy (e.g., Pinto 1998; Harrington et al. 1999, Jiang et al. 2000; Wang et al. 2001).
Bulk microphysics schemes were initially developed for mesoscale modeling (e.g., Lin et al. 1983; Rutledge and Hobbs 1984; Dudhia 1989) and have subsequently been implemented in climate (e.g., Ghan and Easter 1992; Fowler et al. 1996; Lohmann et al. 1999) and cloud-resolving (e.g., McCumber et al. 1991; Kruger et al. 1995; Wu et al. 1998; Xu and Randall 1996; Jiang et al. 2000) models. These schemes typically include separate prognostic equations for the mixing ratios of cloud liquid, ice, rain, and one or more solid precipitation species (e.g., snow, graupel, hail). Detailed parameterizations of precipitation and evaporation/condensation processes are generally incorporated. The hydrometeor size spectra are represented by distribution functions (e.g., lognormal, gamma). A more recent improvement has been the prediction of particle number concentrations in addition to the mixing ratios (i.e., two-moment schemes) (e.g., Levkov et al. 1992; Ferrier 1994; Harrington et al. 1995; Meyers et al. 1997; Reisner et al. 1998; Girard and Curry 2001; Morrison et al. 2005a). Predicting two moments of the size distribution increases the degrees of freedom associated with the hydrometeor spectra, improving calculations of the microphysical processes and radiative transfer (Meyers et al. 1997).
Models incorporating bulk microphysics parameterizations generally have difficulty correctly predicting the observed persistence of MPS and supercooled liquid water in the Arctic. Curry et al. (2000) showed that different single-column model (SCM) simulations of the May 1998 period of SHEBA/FIRE–ACE underestimated the liquid water path (LWP) and occurrence of supercooled water. These biases were mostly attributed to the incorrect representation of MPS as entirely crystalline. Morrison et al. (2003) conducted a detailed comparison of simulated column cloud properties with SHEBA observations for the period of 1 April–16 May 1998. They found that the model reasonably predicted the mean cloud boundaries and fraction, but had difficulty correctly partitioning the cloud phase, resulting in a substantial underprediction of LWP. Poor simulation of Arctic MPS has been attributed to uncertainties in the specified ice crystal number concentration (Girard and Curry 2001; Morrison et al. 2003). Several modeling studies have suggested that the lifetime of Arctic MPS is linked to the number of ice-forming nuclei (IN) available (Pinto 1998; Harrington et al. 1999; Jiang et al. 2000), with decreased IN concentrations resulting in more persistent MPS. Girard and Curry (2001) improved the prediction of cloud phase and LWP in column simulations of SHEBA/FIRE–ACE (relative to a bulk scheme that only predicted the mixing ratios, i.e., a single-moment scheme) by incorporating prognostic equations for the ice crystal and droplet number concentrations. Morrison et al. (2005b) used an SCM to reveal improvements in the prediction of cloud phase, LWP, and surface radiative fluxes during the April–May period of SHEBA using a two-moment scheme with detailed treatments of ice nucleation and droplet activation. One difficulty noted in all of the column studies, however, was uncertainty associated with the specified large-scale dynamics and advective forcing, which were based on forecast model output due to lack of observations (Morrison and Pinto 2004).
There has been little systematic testing of bulk microphysics schemes in mesoscale simulations of Arctic MPS. The ability of mesoscale models to correctly predict MPS has important implications for weather forecasting due to the persistence and large horizontal extent of these cloud systems (e.g., Curry et al. 2000; Intrieri et al. 2002; Zuidema et al. 2005). In particular, realistic treatment of supercooled water can improve forecasts of aircraft icing fields (e.g., Thompson et al. 1997; Reisner et al. 1998). Mesoscale models have also been used to study interactions and feedbacks in the Arctic climate system (e.g., Lynch et al. 1995; Wei et al. 2002; Rinke et al. 2004). A realistic treatment of Arctic MPS is crucial for studying cloud–climate–surface feedbacks due to the strong radiative impact of these clouds on the surface (Shupe and Intrieri 2004; Zuidema et al. 2005). Such feedback studies are especially important in light of GCM simulations indicating strong Arctic warming in response to increased greenhouse gases (e.g., Houghton et al. 1990, 1995) and recent observations of rapid warming of the Arctic surface (Chen et al. 2002; Serreze et al. 2000; Stone 1997).
Morrison and Pinto (2005) evaluated mesoscale simulations of a persistent MPS deck observed on 4–5 May 1998, during SHEBA/FIRE–ACE. They found that the polar version of the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR–Penn State) Mesoscale Model (MM5; Bromwich et al. 2001) reasonably predicted the macro- and microphysical characteristics of the MPS using a new two-moment bulk microphysics scheme. In this paper, we extend the study of Morrison and Pinto (2005) by simulating a similar case study using this scheme and two other bulk microphysics schemes currently available in MM5. Results using these two schemes and the Morrison and Pinto (2005) scheme are compared with in situ observations and remotely based retrievals obtained during SHEBA/FIRE–ACE. Differences in the simulations are examined in terms of the cloud microphysics, radiative transfer, surface characteristics, and synoptic-scale atmospheric dynamics, and various interactions and feedbacks between them. Sensitivity tests are described that investigate how differences in the parameterized microphysics impact the simulated MPS. The paper is organized as follows. The instrumentation and case description is given in section 2. Section 3 provides a brief description of the MM5. An overview of the three microphysics schemes is given in section 4. Section 5 presents results and describes interactions between the cloud microphysics, synoptic-scale dynamics, and surface fluxes in the model. Sensitivity tests are described in section 6. Summary and conclusions are given in section 7.
2. Observational data
SHEBA is detailed in Uttal et al. (2002). A heavily instrumented icebreaker ship was frozen into the multiyear sea ice on 1 October 1997 at 75.27°N, 142.68°W, and allowed to drift for 1 yr across the Beaufort and Chuckchi Seas. FIRE–ACE was conducted in coordination with SHEBA and used aircraft to measure in situ cloud and aerosol characteristics near the SHEBA site in spring and summer (Curry et al. 2000). Temperature and relative humidity profiles at SHEBA were measured by rawinsondes launched two to four times per day. Surface and near-surface meteorological measurements obtained near the Atmospheric Surface Flux Group tower are described in Persson et al. (2002b). Surface pressure was measured with a Vaisala digital barometer with an uncertainty of ±0.3 mb. Surface turbulent latent and sensible heat fluxes were obtained by taking the median of eddy correlation measurements taken at five different heights along the tower. Upwelling and downwelling shortwave and longwave radiative fluxes were measured using Eppley radiometers and pyrgeometeors, with uncertainty of ±4 W m−2. Surface temperature, Ts, was obtained from the radiative surface temperature measured by an Eppley radiometer, assuming a surface emissivity of 0.99. Uncertainty in Ts is ±0.6 K.
Cloud properties were retrieved from a collection of ground-based instruments deployed at SHEBA. A vertically pointing, 35-GHz cloud radar [the millimeter-wave cloud radar (MMCR)] made continuous measurements of reflectivity and mean Doppler velocity up to a height of 13 km. Collocated with the MMCR were a dual-channel microwave radiometer (MWR) and a depolarization micropulse lidar used to remotely determine liquid water path (LWP), cloud phase, and cloud-base height (Westwater et al. 2001; Intrieri et al. 2002). Retrieval techniques for estimating the ice water contents are described in Shupe et al. (2001); LWP is retrieved from MWR brightness temperatures (Han and Westwater 1995; Zuidema et al. 2005). Radar retrievals of liquid water content (LWC) were not performed for mixed-phase clouds since ice tends to dominate the radar signal. Estimates of LWC profiles have been made by assuming adiabatic ascent of a parcel from the lidar-determined cloud base and constraining the column-integrated values to the MWR-derived LWP (Zuidema et al. 2005). In situ cloud properties obtained from an array of instruments flown on the NCAR C-130Q as part of FIRE–ACE were used to validate the remotely sensed data (Zuidema et al. 2005).
Five-kilometer-resolution data from the Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder Project were analyzed to provide retrievals of cloud properties over the Arctic during SHEBA/FIRE–ACE (Maslanik et al. 2001). Pathfinder data products are twice-daily composites of AVHRR images centered on 0400 and 1400 local solar time (∼1500 and 0100 UTC). Acquisition times for a given pixel may have varied several hours from this time. Cloud properties were retrieved using algorithms of the Cloud and Surface Parameter System (CASPR) described by Key (2002). The satellite products agree well with observations and ground-based retrievals obtained during SHEBA/FIRE–ACE (Maslanik et al. 2001).
a. Case description
A detailed description of the case is given by Curry et al. (2000), Zuidema et al. (2005), and Morrison and Pinto (2005). A low-level MPS formed over the western Arctic basin in late April and persisted nearly continuously at SHEBA until mid-May (Curry et al. 2000). The evolution of this cloud on 4–6 May is indicated by time–height plots of the retrieved IWC and LWC (Fig. 1). Thin upper-level ice clouds appeared at 0600 UTC on 4 May over the SHEBA site, merging with and seeding the lower cloud at 1200 UTC before dissipating at 0600 UTC on 5 May. The boundary layer was well mixed from the surface to the top of the low-level MPS, with a minimum temperature of ∼250 K and 2–3-K temperature inversion above the boundary layer. The liquid layer was horizontally uniform near SHEBA (over scales of 10–20 km) as indicated by in situ aircraft data along horizontal flight segments, although the ice hydrometeor field exhibited significant spatial and temporal variability (Lawson et al. 2001; Zuidema et al. 2005). The low-level cloud deck surrounding the SHEBA site during this period was horizontally extensive and its spatial coverage was persistent (Fig. 2). National Centers for Environmental Prediction [NCEP; Medium-Range Forecast model (MRF)] analyses provide the synoptic context for this case (see Fig. 2). The synoptic conditions were dominated by an anticyclone (1017 mb) centered over the Bering Strait and a cyclone (1004 mb) centered over the Beaufort Sea east of the SHEBA site at 0000 UTC 4 May. By 0000 UTC 6 May, a broad anticyclone was present to the east of SHEBA, while a stronger cyclone (∼990 mb) was located in southwestern Alaska. The SHEBA site was dominated by a fairly weak (<5 m s−1) flow from the south and west at low levels during this period. Large-scale dynamic forcing at low levels over SHEBA is indicated by the NCEP reanalysis described by Zuidema et al. (2005). Daily mean 850-mb vertical pressure velocities show that weak subsidence (<1 mb h−1) occurred on 4 May in the grid cell incorporating SHEBA, with weaker ascent on 5 May (see Zuidema et al. 2005).
3. MM5 configuration and forcing
The MM5 is a nonhydrostatic model that includes a number of parameterizations for treating the following: 1) shortwave and longwave radiative transfer, 2) boundary layer (BL) and turbulence processes, 3) surface processes and exchange with the overlying atmosphere, 4) cumulus convection, and 5) cloud microphysics. The nonhydrostatic momentum equations are solved using the time-splitting method for sound wave stability described in Grell et al. (1994). The shortwave and longwave radiative transfers follow Briegleb (1992a, b). Turbulent fluxes in the atmosphere and between the surface and atmosphere are parameterized following the 1.5-order prognostic turbulent kinetic energy (TKE) scheme described by Janjić (1994). Heat transfer through the surface is predicted using a multilayer soil or sea ice and snow model depending upon the surface type (Bromwich et al. 2001). The surface latent and sensible heat fluxes over ocean points are calculated by averaging the separate contributions from open water and sea ice in each grid cell depending upon the specified sea ice concentration. Here, a sea ice concentration of 0.95 is assumed for all ocean points in the model domain broadly consistent with Special Sensor Microwave Imager retrievals (Cavalieri et al. 2006). Turbulent transport is calculated for cloud droplets and ice, but is neglected for precipitation (rain, snow, graupel). The parameterization of shallow and deep cumulus convection is neglected since convection is limited over Arctic sea ice with the exception of convective plumes emanating from leads (e.g., Pinto and Curry 1995). The cloud fraction within a grid cell is unity if the water content predicted by the microphysics scheme is greater than 10−5 g m−3 at any level, and zero otherwise.
MM5 offers the flexibility of grid nesting. We utilize two domains centered on the SHEBA site. The outer and inner domains have grid spacings of 60 and 20 km, respectively. Results from the inner domain are presented here. Simulations are performed with 34 vertical levels and 15 levels in the lowest 1 km. The initial and lateral boundary conditions are specified using the 2.5° NCEP Global Data Assimilation System (GDAS) dataset (Kalnay et al. 1996). The period simulated is from 0000 UTC on 4 May to 0000 UTC on 6 May. Simulations are initialized with no cloud water (i.e., “cold start”). Time-averaged model output over the domain is calculated over the period 1800 UTC 4 May to 0000 UTC 6 May. Eighteen hours of spinup time allows for the formation of a quasi-steady low-level cloud layer as shown in section 5. The quasi-steady nature and large horizontal extent of the low-level cloud deck provide justification for our use of time- and area-averaged values in assessing the performance of the model.
4. Description of the microphysics schemes
The three microphysics schemes used in this study vary widely in detail and complexity. Table 1 summarizes various features of the schemes. Model parameters that are shared between the schemes (e.g., precipitation fall speeds) are specified so that differences in the simulations due to tuning are minimized. These schemes are described below in order of increasing complexity.
The first scheme (hereafter R1) is based on the microphysics package described by Reisner et al. (1998). The R1 scheme was initially developed for predicting supercooled liquid water in the context of midlatitude winter storms but has since been widely used over a range of conditions. This scheme predicts the mixing ratio of four hydrometeor species (cloud droplets, cloud ice, rain, and snow). The partitioning of water between liquid and ice in mixed-phase clouds is calculated in terms of freezing, riming, and condensation/deposition onto the various hydrometeor species. The saturation adjustment method used to calculate droplet condensation assumes zero residual water supersaturation; all excess vapor is instantaneously converted to cloud water. Collection between hydrometeor species is calculated using the continuous collection equation (see, e.g., Pruppacher and Klett 1997). The number concentration of cloud ice, Ni, is specified as a function of temperature based on a composite of midlatitude measurements of ice nuclei concentration following Fletcher (1962). The parameter Ni at temperatures <−30°C is set to the value at −30°C to prevent unrealistically large Ni at colder temperatures. Cloud ice is initiated directly from water vapor or from droplet freezing following Bigg (1953).
The second scheme (hereafter R2) is also based on Reisner et al. (1998), but it differs from R1 by including Ni as a prognostic variable, along with the mixing ratio of graupel, qg. Thus, R2 includes a number of additional parameterized processes for calculating tendencies of Ni and qg. Note that prediction of qg appears to have little impact in this particular case study. A number of other microphysical processes in R2 differ from R1 as well (see Table 1). Since Ni is prognosed, ice nucleation must be explicitly treated. Ice crystals nucleate through deposition/condensation freezing if Ni is less than the number concentration of IN given by Fletcher (1962) (again limited at temperatures <−30° as described above). Crystals can also nucleate by droplet freezing following Bigg (1953). Ice crystal multiplication can increase the crystal concentration through rime splintering (Hallet and Mossop 1974). However, this case study mostly lies outside the assumed temperature range for rime splintering to occur.
The third scheme (hereafter M05) is described by Morrison et al. (2005a) and Morrison and Pinto (2005). The M05 scheme predicts the number concentrations and mixing ratios of four hydrometeor species (cloud droplets, cloud ice, rain, and snow). Thus, a number of additional parameterized processes are included relative to R2, since M05 includes three other prognostic variables. Several other parameterized processes also differ from R2 (see Table 1). A distinguishing feature of the M05 scheme, relative to other bulk microphysics schemes including R1 and R2, is that droplet activation and ice nucleation are calculated from a specified distribution of aerosols. The treatment of ice nucleation includes separate contributions from homogeneous and heterogeneous freezing on deliquescent aerosol, and contact freezing of droplets (see Morrison and Pinto 2005 for details). The potential number of droplets activated is a function of the cloud condensation nuclei (CCN) activity spectrum and an effective vertical velocity (Rogers and Yau 1989), which includes a subgrid vertical velocity component (Morrison and Pinto 2005). The parameterized CCN activity spectrum is a function of the aerosol size, number concentration, and composition following Khvorostyanov and Curry (1999). The subgrid vertical velocity is a function of the predicted TKE following Morrison and Pinto (2005).
Aerosol properties used by the M05 scheme are based on condensation nuclei (CN) and CCN aircraft measurements taken during clear-sky periods in May 1998 as part of FIRE–ACE (Yum and Hudson 2001). We chose aerosol properties measured during clear-sky conditions to avoid low concentrations associated with nucleation and particle scavenging in the cloud layer and below. The aerosol number concentration is given by observed CN concentrations, which ranged from ∼350 to 700 cm−3, with the larger values occurring at higher altitudes. These aerosol concentrations are similar to those obtained by Radke et al. (1984) during an Arctic haze event. Since detailed measurements of aerosol size and composition were lacking, we must infer values for the aerosol parameters needed by the model (i.e., size and composition) from the observed CCN activity spectra and reports in the literature. A soluble volume fraction of 75% (with the soluble portion consisting of ammonium sulfate) is specified based upon the occurrence of high-sulfate aerosols in the Arctic during springtime haze events (Borys 1989). The aerosol size parameters (minimum size and size distribution slope) were found by matching CCN spectra calculated by the model [using the observed number concentration and assumed composition; see Morrison and Pinto (2005)] with observed CCN spectra obtained during the three sampling missions that encountered clear sky conditions.
As described in the next two sections, contact freezing of droplets is the dominant ice initiation mechanism in the simulated mixed-phase cloud using the M05 scheme. In M05, the number of contact IN, NIN,c (units of L−1), is specified following Meyers et al. (1992) as a function of temperature, T (units of K), where
Note that this parameterization is a fit to midlatitude measurements (Meyers et al. 1992) and that NIN,c may have differed substantially during this case (measurements of NIN,c in the Arctic are lacking). This function sets the upper limit of the Ni, which can be realized through contact freezing. The actual nucleation rate is determined by the collection of contact IN by rain and cloud droplets. Here, it is assumed that IN are collected through convective Brownian diffusion (Young 1974). Phoretic forces may also strongly impact local collection rates. However, since phoretic forces depend upon vapor density and temperature differences between a drop and its local environment, their impact will vary between small-scale updrafts and downdrafts that are not resolved by the model. An investigation of the impact of phoretic forces on the aggregate-scale (grid mean) collection rates is left for future work. Sources and sinks of contact IN are neglected; that is, it is assumed that the supply of contact IN is steady.
The microphysics schemes are coupled to the radiative transfer. Cloud optical properties are calculated following Slingo (1989) and Ebert and Curry (1992) for droplets and ice, respectively. The droplet and crystal effective radii are assumed to be 10 and 40 μm, respectively, for the R1 and R2 simulations. For M05, the droplet and crystal effective radii are diagnosed from the predicted mixing ratios and number concentrations. The droplet number concentration predicted by M05 is close to in situ observations obtained on 4 May (Morrison and Pinto 2005). The specified droplet concentration in R1 and R2 is 250 cm−3 based on the observations.
Time–height plots of the simulated LWC and IWC for the grid cell incorporating the SHEBA site reveal significant differences between the simulations (Fig. 4). The M05 scheme predicts a quasi-steady, continuously precipitating MPS similar to observations (see Fig. 1), while R1 and R2 produce low-level ice-phase clouds with very weak precipitation. Ice water contents predicted by M05 are similar to MMCR retrievals and are generally at least an order of magnitude larger than values predicted by R1 and R2. All three schemes underpredict the observed upper-level cloudiness, which potentially impacted the BL stratus through seeding and radiative interactions (Zuidema et al. 2005). As suggested by Morrison and Pinto (2005), this bias may result from deficiencies in the NCEP GDAS upper-level RH used to initialize the simulations. Note that the impact of the upper cloud on the low-level cloud was probably fairly limited since ice water contents in the upper cloud were generally less than a few mg m−3 (Morrison and Pinto 2005).
In the M05 simulation, supercooled liquid water associated with MPS is present across the domain, with the largest time-average LWPs (>30 g m−2) located near the SHEBA site (Fig. 5). In contrast, very little liquid water is produced by R1 or R2 (Table 2). CASPR satellite retrievals of cloud phase during this case are uncertain since mixed-phase clouds were not specifically identified (clouds were retrieved as either all liquid or all ice). The retrieval of cloud phase may be particularly problematic because of the presence of ice clouds overlying the low-level MPS (Key 2002), resulting in a potential underestimation of the liquid fraction. Nonetheless, the satellite-derived liquid cloud fraction of 0.48 (average of the overpasses that occurred during the period) suggests that significant liquid water was present across the domain. Plots of the time-averaged IWP across the domain (Fig. 6) exhibit some common features produced by the three schemes. The largest IWPs occur along the southern edge of the domain associated with extensive mid- and upper-level cloudiness. Two other regions of fairly significant high cloudiness and IWP > 5 g m−2 are found in the northeast and northwest corners of the domain in all three simulations. However, in M05, a region with IWP > 1 g m−2 extends across most of the domain, which is mostly absent in R1 and R2. This region is associated with the extensive low-level MPS. Here, ice is initiated mostly by the contact freezing of droplets, and subsequently grows rapidly by vapor deposition in the water-saturated environment (e.g., a Bergeron–Findeisen mechanism). Cloud-top radiative cooling rates associated with the MPS that exceed 40 K day−1 help to maintain water saturation even in the presence of large-scale downward vertical motion and vapor deposition onto ice particles (Morrison and Pinto 2005). In the absence of MPS and significant cloud-top cooling in R1 and R2 (along with a reduced surface turbulent flux of water vapor as described later in this section, and dehydration due to the initial formation of ice clouds as described in section 6), low-level relative humidity (or ice supersaturation) is lower than it is in M05. Hence, ice depositional growth and IWPs are limited in R1 and R2 even though the ice growth rate (for a given bulk ice mass and ice supersaturation) is much greater relative to M05 (see section 6). The time- and domain-average IWP predicted by M05 is about a factor of 3.5 larger than that predicted by R1 or R2 (see Table 2). The larger IWPs in M05 are also associated with a much larger domain-average surface precipitation rate compared to R1 and R2 (see Table 2). Comparison of the modeled precipitation rate with SHEBA observations is hampered by uncertainty in the observations due to factors of the high-latitude environment (e.g., blowing snow) and the very small precipitation amounts. The M05 scheme predicts a mean precipitation rate for the SHEBA grid cell (calculated between 1800 UTC 4 May and 0000 UTC 6 May) of about 0.5 mm day−1, while measurements indicate a trace of precipitation during the period.
The microphysical characteristics of the predicted low-level cloud exert a large impact on the modeled surface radiative fluxes. One of the most critical factors is the cloud phase as described in previous studies (Shupe and Intrieri 2004; Zuidema et al. 2005). The crystalline clouds predicted by R1 and R2 have much smaller liquid and ice water paths (Fig. 7) and larger effective radii than the MPS predicted by M05; therefore, they have a much more limited impact on the radiative transfer. Time series of modeled and observed surface downwelling longwave (LW) and shortwave (SW) radiative fluxes at SHEBA are shown in Fig. 8. Both R1 and R2 exhibit substantially smaller downwelling LW fluxes at the surface and larger downwelling SW fluxes during the “day” compared to observations (here, “night” and “day” refer to those portions of the diurnal cycle when the solar zenith angle was comparatively large or small, respectively, since the sun was continuously above the horizon at this location and time of year). In contrast, downwelling LW and SW fluxes in M05 are fairly close to observations after formation of MPS in the simulation. The small overprediction of downwelling LW flux in M05 is likely the result of a small positive bias in the cloud temperature (∼2–3 K). Differences in the time-average and domain-average downwelling surface LW and SW fluxes between the three simulations closely track differences in the LWP (see Table 2), since liquid water dominates the radiative impact of the cloud layer. The time- and domain-average cloud optical depth, τ, predicted by M05 is slightly larger than the satellite-derived value; the mean values of τ predicted by R1 and R2 are close to zero.
Differences in the surface radiative fluxes between the simulations in turn have a strong impact on the surface energy budget and surface temperature, Ts (Fig. 9). The values of Ts for the SHEBA grid cell in R1 and R2 are colder than in M05 due to the much smaller downwelling LW flux and hence rapid cooling of the surface at night. The much larger downwelling SW flux in R1 and R2 results in a greater range of Ts over the diurnal cycle. The trend of Ts predicted by M05 is quite close to the observed once MPS forms in the simulation, although values of Ts are consistently too high by about 2 K. This bias is attributed to the somewhat larger downwelling LW radiative flux (see Fig. 8) and upward conductive heat flux (transfer of heat from below toward the surface), which is balanced by the increased upwelling LW flux and increased sensible and latent turbulent heat fluxes associated with a warmer surface. An incorrect conductive heat flux may be due to inadequate snow and sea ice model structure (Persson et al. 2002a), although a detailed evaluation of the surface parameterization is beyond the scope of this study. Note that small biases in the initial Ts may also magnify errors in the modeled surface energy budget.
Large differences in the microphysical characteristics of the cloud layer between the simulations also strongly impact the vertical structure of the modeled BL (Fig. 10). Several studies have shown that strong cloud-top radiative cooling atop Arctic MPS promotes convection via negative buoyancy (e.g., Pinto 1998; Jiang et al. 2000; Harrington et al. 1999; Wang et al. 2001). Modeling studies have suggested that glaciation of MPS can weaken cloud-top cooling and thus production of turbulence, leading to the collapse of the cloud layer (Harrington et al. 1999; Jiang et al. 2000). Strong radiative cooling of the surface in R1 and R2 produces a surface-based stable layer at night for the SHEBA grid cell (e.g., 1200 UTC 5 May). With daytime solar heating of the surface, a shallow (∼100 m) surface-based mixed layer is evident at 0000 UTC 5 May and 0000 UTC 6 May. In contrast, once MPS forms in the M05 simulation, cloud-top radiative cooling results in a well-mixed BL that extends from the cloud top to the surface (note, however, that the BL scheme predicts constant potential temperature rather than equivalent potential temperature). The well-mixed BL in M05 is consistently about 200–300 m shallower than observed. The reasons for this bias are unclear, but could be related to potential deficiencies in the boundary layer scheme, grid-scale vertical velocity, and/or initial profiles. The BL temperatures are 1–2 K higher than observed. However, cloud-top radiative cooling associated with the MPS produces a net cooling of the BL in M05 compared with R1 and R2.
The surface temperature and near-surface static stability exert a large influence on the turbulent flux of water vapor at the surface, Fs, which serves as a source of moisture for the modeled cloud layer as shown later. In the model, Fs is parameterized as a function of the stability-dependent exchange coefficient and gradient of the water vapor mixing ratio, qυ, between the surface and lowest model level. Since qυ at the surface is assumed to be saturated with respect to ice (over sea ice or snow-covered land) or water (over open ocean), it is strongly dependent upon the surface skin temperature. In R1 and R2, and before MPS forms in M05, rapid radiative cooling of the surface between about 0300 and 1200 UTC reduces the qυ at the surface and results in negative values of Fs (meaning a net transfer of water vapor from the atmosphere to the surface) for the ice-covered portion of the SHEBA grid cell (Fig. 11). With rapid warming of the surface in M05 associated with the formation of MPS (see Fig. 9), the qυ gradient changes sign and Fs becomes positive. With daytime solar heating of the surface in R1 and R2, Fs becomes positive at around 2130 UTC 4 May. The observed values of Fs also appear to be related to Ts and hence changes in cloud cover, although the observed Fs differs substantially from the modeled values. During the brief period around 1200 UTC 4 May that the MPS dissipates and observed Ts rapidly decreases (see Fig. 9), Fs drops to near zero. Decreased near-surface static stability in the M05 simulation also increases Fs over open water relative to R1 and R2 (by an average of ∼40%), even though the skin temperature is held constant. The flux over open water provides an important contribution to the grid-average Fs in all simulations. Note that the presence of an optically thick low-level cloud layer in M05 also substantially increases the surface sensible turbulent heat flux, as was found by the modeling study of Wang et al. (2001).
Differences in Fs and surface precipitation rate between the simulations suggest that the microphysics scheme has a significant impact on the bulk water budget. The larger precipitation rate in M05 suggests either an increase in the precipitation efficiency, an increase in the local moisture supply, or an increase in the large-scale 3D moisture advection compared to R1 and R2. The time- and domain-average column-integrated water budget is expressed as (Morrison and Pinto 2004)
where Fs is the area-average surface turbulent flux of water vapor (including both sea ice and open-water contributions), ADV is the vertically integrated 3D advection of water (both water vapor and cloud water), PRE is the surface precipitation rate (PRE is negative since precipitation results in a loss of water from the domain), and R is the residual term equal to the time rate of change of the precipitable water (PW). Modeled values for the budget terms are shown in Table 3. The advection of water vapor is similar between the simulations; since this term is positive, it indicates an overall convergence of moisture across the domain. The surface flux of water vapor in M05 is about 3.5 times larger than it is in R1 and R2. The residual term is positive in all three simulations, indicating an overall moistening of the domain. The residual term is smaller in M05 due to greater precipitation efficiency and somewhat reduced advection (which is mostly the result of significant cloud water advection out of the domain in M05). Increased precipitation efficiency appears to be associated with the much larger cloud-top radiative cooling rates in M05 relative to R1 and R2 (which is equivalent, in terms of supersaturation generation, to increased vertical velocity). Increased precipitation efficiency using M05 relative to R1 and R2 is further suggested by the sensitivity run described below.
To test the impact of Fs on the cloud field, a sensitivity test is run using M05 with Fs = 0 kg m−2 s−1 (for both sea ice and open-water contributions). This run produces less liquid water than the baseline in terms of mean LWP and Fl (Table 4). For the SHEBA grid cell, the MPS dissipates at 0600 5 May, before reforming 4 h later, in contrast to the continuous MPS produced in the baseline M05 simulation (see Fig. 4). These results suggest a positive feedback between MPS lifetime and Fs in MM5: the presence of optically thick MPS increases Fs, which in turn supplies water vapor to the cloud layer and helps to maintain the cloud. Note that the suppression of ice nucleation and especially the snow particle number concentration in M05 appears to play a critical role in the formation and maintenance of the simulated MPS (see section 6). A positive feedback between the cloud-top radiative cooling rate and droplet condensation rate also helps to maintain the MPS (Morrison and Pinto 2005). Thus, M05 is still able to maintain a fairly extensive and persistent MPS even with no surface vapor flux. Reducing Fs in M05 also produces a surface precipitation rate that is about a factor of 2 smaller than the baseline M05 simulation, but still larger than R1 and R2 because of the increased precipitation efficiency using the M05 scheme, which results in a smaller residual term (see Table 3). This suggests that the larger precipitation flux in the baseline M05 run compared to R1 and R2 is due to both an increase in Fs and greater precipitation efficiency.
These results suggest that the MPS lifetime, and especially the precipitation rate, may be related to Fs when the cloudy BL is coupled to the surface. Surface-based mixed layers occurred about 25% of the time during the November–February period of SHEBA, typically under cloudy conditions, and were more frequent during spring and summer (Persson et al. 2002b). Because of large biases in the modeled surface fluxes (see Fig. 10), surface–MPS interactions are probably exaggerated in MM5. Nonetheless, the possibility of this interaction pathway (although much weaker than in the model) is suggested by observations showing a decrease in Fs to values near zero during the brief period around 1200 UTC 4 May when the MPS dissipated and the surface rapidly cooled (see Fig. 9). It should be noted, however, that the observed Fs does not include contributions from open water. A more extensive analysis of the observations is needed to quantify the bulk water budget and cloud–surface interactions in reality over the Arctic Ocean. Close coupling between the clouds and the surface for this case also suggests that an improved prediction of the cloud field and precipitation rate may require an improved treatment of the surface and especially the surface turbulent fluxes.
The widespread occurrence of MPS in the M05 simulation is associated with greater surface pressure relative to R1 and R2 (Fig. 12). The average surface pressure at the end of the period (0000 UTC 6 May) is about 0.6 mb larger in M05 compared to R1 and R2. The simulated sea level pressure and 900-mb winds across the domain at 0000 UTC 6 May using the M05 scheme (Fig. 13) provide the context for these results, and are in reasonable agreement with the NCEP analysis at this time (see Fig. 2b). The surface pressure is closer to observations in M05 compared to R1 and R2 for the SHEBA grid cell for most of the simulation (Fig. 14). The presence of liquid water associated with MPS in M05 promotes much greater LW radiative cooling within the lower troposphere (e.g., cloud-top cooling), which appears to increase the strength of the overlying anticyclone. These results are in agreement with the modeling studies of Curry (1983, 1987) and Pinto and Curry (1997), who found that condensate led to increased anticyclogenesis through increased radiative cooling.
Low-level (<1 km) subsidence across the domain is increased by an average of 0.07 cm s−1 compared to R1 and R2. However, the impact of MPS on the mesoscale dynamics is quite complex, with distinct regions of increased convergence and upward motion in M05 (Fig. 15). The banded nature of these regions may reflect the development of weak circulations in M05 associated with conditional symmetric instability (SI). The possibility of SI circulations exhibiting banding over scales of ∼100 km has been suggested by previous studies (e.g., Bennetts and Hoskins 1979; Thorpe and Rotunno 1989). Areas of weakly negative moist potential vorticity develop in the M05 simulation (presumably due to diabatic mixed-phase cloud–radiation processes), which are mostly absent in the R1 and R2 simulations. A more complete diagnosis of the dynamics is beyond the scope of this paper and is left for future work.
Because optically thick low-level clouds impact the large-scale subsidence and vertical gradients of water vapor, Curry (1987) suggested a positive feedback involving increased anticyclogenesis and hence low-level subsidence that helps to supply the cloud layer with water vapor, increasing the amount of condensate and in turn increasing the anticyclogenesis and subsidence. This feedback is hypothesized to occur because of the common presence of low-level inversions in the water vapor mixing ratio (e.g., Curry et al. 2000; Tjernstrom et al. 2004). Here, the time- and domain-average low-level (<1 km) vertical advection of water vapor is negative in all three simulations (R1, R2, and M05), although this term is less negative in M05 (−7.6 × 10−8 versus about −3 × 10−7 g kg−1 s−1 in R1 and R2). These results indicate that the MPS predicted by M05 impacts the vertical advection of water vapor in MM5. However, this impact is small in terms of the low-level water vapor budget relative to the impact of the MPS–surface interaction described previously. We note that interactions and feedbacks between the clouds and large-scale dynamics in regional-scale models such as MM5 may be limited by the specified lateral forcing (Lynch et al. 1995). However, these interactions are expected to play a greater role in global models, particularly over longer time scales relevant to climate modeling.
6. Sensitivity tests
In this section, several sensitivity tests are described that investigate how different parameterized processes in the microphysics schemes impact the results. We focus on sensitivity testing of the R1 and R2 schemes; sensitivity tests of the M05 scheme for this case study are described in the companion paper (Morrison and Pinto 2005).
Previous modeling studies have indicated that the lifetime of Arctic MPS is sensitive to the ice microphysical processes, especially the Bergeron–Findeisen mechanism (i.e., the growth of ice by vapor deposition at the expense of liquid water due to the lower saturation vapor pressure with respect to ice compared to liquid) (e.g., Pinto 1998; Harrington et al. 1999; Jiang et al. 2000; Morrison et al. 2003). In particular, the lifetime of MPS was shown to be highly sensitive to the ice particle or IN number concentrations. As described in section 4, R1 diagnoses the number concentration of cloud (small) ice, Ni, from the specified IN number concentration (NIN) as a function of temperature following Fletcher (1962), while R2 and M05 predict Ni. In addition, R2 assumes that ice crystals nucleate if Ni decreases below the NIN given by Fletcher. Therefore, the predicted value of Ni in R2 mostly depends upon the specified NIN. In contrast, M05 predicts the ice nucleation rate on aerosol occurring through deposition/condensation–freezing based upon extensions of classical nucleation theory (Khvorostyanov and Curry 2004). However, little ice is nucleated here through deposition/condensation–freezing at temperatures above about 253 K for the assumed aerosol characteristics. Hence, contact freezing dominates ice nucleation in the modeled MPS in M05. Note that numerous laboratory experiments have indicated that particles are more active (i.e., nucleate at warmer temperatures) in contact mode compared with deposition or condensation–freezing modes (summarized by Pruppacher and Klett 1997). While Rogers et al. (2001) describe occasionally significant number concentrations of deposition/condensation–freezing nuclei during FIRE–ACE at temperatures above 253 K measured by a continuous-flow diffusion chamber, overall these values were low (∼50% of measurements were zero). Note that these counts did not directly measure contact nuclei. Previous observations of NIN using filter techniques suggested mean values of NIN < 0.1 L−1 in the Arctic over pack-ice regions (Bigg 1996). Several observations are consistent with the dominant role of droplet freezing in MPS as summarized by Morrison et al. (2005c).
The number concentration of snow particles, Ns, plays a critical role in MPS as described below. The value of Ns is predicted in M05. In R1 and R2, it is diagnosed from the intercept parameter, N0, and the snow mixing ratio, qs:
where ρ is the air density and ρs is the bulk snow density. In R1, N0 is diagnosed from the snowfall rate following Sekhon and Srivistava (1970), and in R2 from the temperature following Houze et al. (1979). These parameterizations were derived from observations in midlatitude clouds.
Various sensitivity simulations are run with the R1 and R2 schemes to quantify the impact of Ni and Ns on the prediction of MPS. M05 predicts values of Ni and Ns about 10 and 5–7 times smaller, respectively, than is predicted or diagnosed by R1 and R2 for the low-level cloud in the SHEBA grid cell. In M05, the mean total (cloud ice plus snow) ice particle number concentration in the MPS in the SHEBA grid cell is about 0.5 L−1. The actual crystal concentration in Arctic MPS is uncertain. Previous observations of Arctic stratus (e.g., Jayaweera and Ohtake 1973; Pinto 1998; Rogers et al. 2001) reported similarly low crystal concentrations (<1 L−1), although these measurements often did not include the smallest ice particles of the distribution. Crystal concentrations of a few per liter were observed in the low-level MPS on 4 May during FIRE–ACE using the 260X optical probe. However, the 260X probe was found to undercount same-sized particles relative to other instruments (Lawson 2003). The Cloud Particle Imager (CPI), which was also operating during the 4 May flight, indicated much larger crystal concentrations (by more than an order of magnitude), although at this point in time it is difficult to assess the accuracy of the CPI-derived concentrations (Zuidema et al. 2005).
Both NIN and N0 are modified in R1 and R2 to produce values of Ni and Ns similar to M05. Modification of NIN impacts the Ni diagnosed or predicted by R1 and R2, while modification of N0 impacts the diagnosed Ns. A weakness of the single-moment approach employed by R1 and R2 is that Ns (or N0) is not directly influenced by NIN (or Ni). In the double-moment approach used by M05, modification of NIN directly impacts Ns (or N0). Thus, the different treatment of ice nucleation in M05 compared to R1 or R2 yields not only much smaller values of Ni, but also much smaller values of N0 than given by the empirical midlatitude formulations.
Three sets of sensitivity simulations are run using both R1 and R2: 1) decrease of N0 by a factor of 10 (proportional to a decrease in Ns by a factor of ∼6 for a given qs, since Ns ∝ N 3/40), 2) decrease of NIN by a factor of 10 (proportional to a decrease in Ni by a factor of 10), and 3) decrease of both N0 and NIN simultaneously by a factor of 10 (see Table 4). Modifying N0 has a strong impact on LWP and Fl in R2, while the response in R1 is much weaker. The differing results are attributed to differences in the partitioning of ice mass, and hence vapor depositional growth (or the Bergeron–Findeisen mechanism), between the cloud ice and snow categories in R1 and R2. In the baseline R2 simulation, the ratio of the deposition rate onto snow and the deposition rate onto cloud ice is 15, while in R1, it is 3.5. Thus, the Bergeron–Findeisen mechanism is dominated by snow deposition in R2 and hence LWP and Fl are much more sensitive to N0 in this scheme compared with R1 (the Bergeron–Findeisen mechanism plays a much greater role in depleting liquid water than riming in these simulations). Note that even though ice depositional growth at a given ice supersaturation is suppressed in the sensitivity tests (and in M05) because of the reduced values of N0, net depositional growth rates are actually increased, leading to larger IWPs and greater precipitation rates. This seemingly contradictory result is due to the substantial increase in ice supersaturation associated with greater cloud-top radiative cooling and a larger surface turbulent flux of water vapor induced by the widespread MPS (see section 5). Modification of NIN has little impact in either scheme. Interestingly, while modifying either NIN or N0 individually in R1 has limited impact, when both parameters are reduced simultaneously, the impact on LWP and Fl is substantial. This occurs because reducing NIN increases the amount of cloud ice autoconverted into snow, while modifying N0 simultaneously reduces the snow deposition rate (for given values of ice supersaturation and snow mixing ratio) and hence the strength of the Bergeron–Findeisen mechanism. When only N0 is modified in R1, the Bergeron–Findeisen mechanism is still fairly strong due to the removal of water vapor by deposition onto cloud ice. When only NIN is modified, more cloud ice mass is transferred into the snow class, increasing the snow deposition rate and limiting the impact of NIN. The strong impact of N0 (or both N0 and NIN) on LWP and Fl appears to be due mostly to direct changes in the strength of the Bergeron–Findeisen process (i.e., modification of the number concentration for a given bulk ice mass changes the bulk particle surface area and hence the vapor deposition rate) rather than indirect changes in the strength of the Bergeron–Findeisen process induced by changes in the particle fall speeds and hence in-cloud residence times. Increasing the ice particle fall speeds (both cloud ice and snow) by a factor of 2 (which is much greater than the change in mean fall speed induced by modification of N0 and NIN in the sensitivity tests described previously), while leaving R1 and R2 otherwise unmodified, has little impact on the results.
While N0, or both NIN and N0, play an important role in the prediction of MPS as described above, another important consideration is the treatment of the various ice nucleation modes, which determines the conditions for which nucleation is allowed to occur. As described previously, M05 predicts ice nucleation through deposition/condensation–freezing only at temperatures less than about 253 K. Thus, ice nucleation in MPS at temperatures above 253 K occurs mostly through contact freezing, requiring the presence of supercooled droplets. In contrast, R1 and R2 allow ice nucleation at water subsaturation at temperatures >253 K (as long as conditions are ice supersaturated in R1 and at least 5% ice supersaturated in R2). By allowing significant ice production at water subsaturation and relatively warm temperatures, R1 and R2 predict an initial burst of low-level ice clouds (see Fig. 4b) that rapidly precipitate to the surface and dehydrate the lower atmosphere. This limits the subsequent production of MPS. In a sensitivity test allowing ice nucleation at temperatures >253 K only when liquid water is present (but otherwise leaving NIN and N0 unchanged), LWP and Fl are increased substantially in R2 (see Table 4). Thus, preventing all-ice clouds from forming in water subsaturated conditions limits dehydration and allows for the initial formation of MPS. Limited ice formation in water subsaturated conditions at temperatures above 253 K is consistent with cloud phase retrievals at SHEBA (encompassing the ∼1 yr duration of the project) indicating that 12% or fewer of clouds with minimum cloud temperatures >253 K were all ice and hence water subsaturated (Morrison et al. 2005c).
7. Summary and conclusions
In this study, we compared simulations of a springtime Arctic mixed-phase stratus observed during SHEBA/FIRE–ACE using three different bulk microphysics parameterizations (R1, R2, and M05) implemented into the polar MM5. M05 predicted significant amounts of liquid water associated with MPS across the model domain and at the SHEBA site, similar to satellite and ground-based retrievals. In contrast, R1 and R2 produced very little liquid water. Note that this improvement comes at a computational cost: run time was increased by about 25% in M05 relative to using R1 or R2. We are currently working to improve the efficiency of the scheme. Note also that the M05 results are sensitive to the specified aerosol characteristics (Morrison and Pinto 2005), which were constrained by observations but nonetheless are uncertain.
The prediction of MPS had a strong impact on the modeled surface radiative fluxes, boundary layer stability, and precipitation rate at the surface. The prediction of optically thin, low-level crystalline clouds rather than MPS in R1 and R2 resulted in larger (smaller) downwelling SW (LW) fluxes at the surface compared to M05. The net impact was to cool the surface in R1 and R2 relative to M05. Cloud-top radiative cooling associated with MPS in M05 led to additional condensation in the cloud layer, the development of a quasi-steady 600–800-m surface-based mixed layer, and a net cooling of the BL compared to R1 and R2. In contrast, the inability of R1 and R2 to form MPS resulted in a strong diurnal cycle in the BL structure with a shallow mixed layer during the day and a strongly stable inversion layer at night. The prediction of MPS greatly enhanced the ice supersaturation and hence IWP and the precipitation rate in M05.
Model results were described in terms of interactions and feedbacks between the clouds, surface, and synoptic-scale dynamics. An analysis of the column-integrated moisture budget for the three simulations showed that the much larger precipitation flux in M05 was the result of an increase in the turbulent flux of water vapor from the surface and greater precipitation efficiency than in R1 or R2. An increase in the surface turbulent flux resulted from warmer surface temperatures and differences in near-surface static stability in M05 associated with the MPS–radiative forcing. The additional moisture supplied by the surface fluxes also appeared to increase the longevity of MPS, representing a positive feedback between the cloud lifetime and the surface in MM5. However, even with the surface vapor flux set to zero, the modeled MPS was still fairly widespread and persistent as a result of the smaller ice particle number concentrations, particularly for snow, in M05 relative to R1 and R2. Because the modeled surface fluxes were significantly biased when compared to observations, interactions between the clouds and surface in MM5 were probably exaggerated compared to reality. A more detailed analysis of the observed cloud cover and surface turbulent flux of water vapor over the Arctic Ocean is needed to quantify these interactions in the real atmosphere. We also note that a realistic prediction of the cloud field, and especially the precipitation rate, may require an improved treatment of the surface, at least for cases when the surface fluxes are coupled to the cloud layer.
The different cloud microphysics schemes had an impact on the synoptic-scale atmospheric dynamics. The prediction of MPS in M05 led to larger surface pressures compared to R1 and R2, which generally improved the prediction of the surface pressure at SHEBA. These results are consistent with previous modeling studies (Curry 1983, 1987; Pinto and Curry 1997) that showed anticyclones tended to be stronger when radiative cooling associated with condensate was included. Subsidence was generally stronger in M05, although the impact of MPS on the vertical velocity field was complex. The difference field of the vertical velocity between M05 and R1 or R2 exhibited a banded structure over scales of ∼100 km, possibly related to differences in the moist potential vorticity and development of weak circulations in M05 associated with symmetric instability. A more detailed analysis of the numerous potential MPS–dynamic interactions was beyond the scope of this study, and is left for future work.
Sensitivity tests showed that much of the difference between the simulations in terms of LWP and fraction of the domain covered by MPS could be explained by their different treatments of the ice microphysics, and particularly the treatments of cloud ice and snow number concentrations and the ice nucleation mode. M05 produced both cloud ice and snow number concentrations that were significantly less than the number concentrations diagnosed or predicted by R1 and R2. The ice nuclei number concentration and snow intercept parameter were modified in R1 and R2 to produce cloud ice and snow number concentrations similar to those predicted by M05. Modification of N0 substantially increased the LWP and amount of MPS predicted by R2, while modification of NIN had little impact. Modification of either N0 or NIN individually had little impact in R1. Interestingly, R1 was much more sensitive to the simultaneous modification of both N0 and NIN. The sensitivity of the LWP to N0, or both N0 and NIN, was due to changes in the rate of water vapor deposition and hence the Bergeron–Findeisen mechanism (for given values of ice supersaturation and bulk ice mass) as the particle number concentrations were modified. The different responses exhibited by R1 and R2 were the result of different partitioning of the ice mass between cloud ice and snow; R2 included significantly more ice mass in the snow category and therefore was much more sensitive to modification of N0. The strong response of R1 and R2 to changes in N0, or both NIN and N0, is consistent with previous studies indicating the strong sensitivity of MPS lifetime to changes in the ice particle number concentration (Pinto 1998; Harrington et al. 1999; Jiang et al. 2000). While recent attention in the literature has focused on the impact of NIN in detailed cloud models of Arctic MPS (e.g., Harrington et al., 1999), due to inherent weaknesses in the single-moment approach employed by R1 and R2, modification of NIN had no direct impact on N0 and therefore it had a limited impact in the simulations. More detailed double-moment microphysics models (including M05) predict the snow number concentration (or N0) and include cloud (small) ice number concentration as a source term for the snow number concentration through autoconversion. The snow number concentration (or N0) in these models therefore directly responds to the modification of NIN. Thus, the different treatment of ice nucleation in M05 compared to R1 or R2 yielded not only much smaller values of Ni, but also much smaller values of N0 than given by the empirical formulations. This suppressed the Bergeron–Findeisen mechanism and therefore helped to maintain liquid water in M05. The current treatments of N0 in R1 and R2, which are based upon midlatitude observations, appear to be inadequate for simulating the long-lived MPS endemic to the Arctic. These results suggest the need to either prognose the snow number concentration or develop new techniques for diagnosing N0 in order to more realistically predict MPS using the simpler schemes.
Different treatments of the various ice nucleation modes also impacted the simulations. In M05, deposition/condensation–freezing was limited at temperatures above 253 K. Thus, the dominant ice nucleation mode in the simulated MPS was contact freezing of droplets (the cloud layer was outside of the Hallett–Mossop rime-splintering zone). Both R1 and R2 allowed significant ice formation in water subsaturated conditions at temperatures >253 K. Thus, low-level ice clouds were initially produced in these simulations, which precipitated and dehydrated the lower atmosphere and subsequently limited the formation of MPS. A sensitivity test allowing ice nucleation in R2 where temperatures were >253 K only when liquid water was present produced a greater MPS fraction and LWP across the domain. These results indicate that the treatment of microscale ice nucleation processes can have a significant impact on the predicted large-scale cloud field.
This research was supported by Grants NSF OPP-0084225 and DOE DEFG03-94ER61771. Microwave radiometer data were obtained from the DOE ARM program. Precipitation and rawinsonde measurements were obtained from the SHEBA Project Office at the University of Washington Applied Physics Laboratory. Surface data were obtained from the SHEBA Atmospheric Surface Flux Group. LWC and in situ IWC data were provided by P. Zuidema (CIRES–NOAA/ETL), and ice retrievals were provided by M. Shupe (CIRES–NOAA/ETL). CASPR AVHRR satellite retrievals were provided by J. Key (NOAA/NESDIS). AVHRR satellite images and NCEP synoptic charts were provided by D. Wylie and the Space Science and Engineering Center at the University of Wisconsin—Madison. The CN and CCN data were made available by J. Hudson (DRI). MWR retrievals of LWP were provided by Y. Han (NOAA/NESDIS). We would also like to acknowledge Dr. Ola Persson for suggesting the possibility of symmetric instability as a dynamic response to the different microphysics schemes. The helpful comments of two anonymous reviewers greatly improved the manuscript.
Corresponding author address: Dr. Hugh Morrison, National Center for Atmospheric Research/MM5 Division, P.O. Box 3000, Boulder, CO 80309. Email: firstname.lastname@example.org