Characterization of ice cloud properties obtained by shipborne radar/lidar over the tropical western Pacific Ocean for evaluation of an atmospheric general circulation model



[1] This study analyzed 95-GHz radar/lidar data collected from the R/V Mirai over the tropical western Pacific to characterize the vertical distribution of ice cloud effective radius reff, ice water content IWC, and in-cloud vertical velocity of the region in conjunction with weather regimes classified by International Satellite Cloud Climatology Project (ISCCP) cluster analysis. Ice clouds observed from the Mirai were roughly consistent with the ISCCP weather regimes; more convectively active regimes had larger amounts of high cloud consisting of deeper cloud with larger ice water path (IWP) and precipitating ice fraction. Ice cloud microphysics of the Center for Climate System Research, National Institute for Environmental Studies, Frontier Research Center for Global Change atmospheric general circulation model (AGCM) was then evaluated using the radar–lidar simulator and ISCCP weather regimes for comparison of the statistics at different scales. The model tended to produce a high cloud fraction that was two times larger in the cirrus regimes but 50% lower in the deepest convective regime. The simulated IWP could only weakly reproduce the observed variety and generally underestimated the observed values despite the weather regimes. Cutoff in the simulated grid mean IWC around 0.1 g−3 was too small, especially above 11 km. The AGCM successfully predicted the observed frequency distribution for reff above 11 km, but produced large overestimation in the peak value below 11 km due to the excessively large fraction of reff ∼100 μm. Establishing a cutoff for cloud ice at reff > 120 μm was found to be quite reasonable, although it would miss some of the larger particles that were observed.

1. Introduction

[2] Ice clouds persist for a long period spanning a wide range of area and have a large coverage over the globe. Since ice clouds interact effectively on the radiation budget and water supply/sink during their existence, appropriate representation of these clouds and the related cloud processes in general circulation models are expected to lead to better understanding of the ice cloud-radiation feedback process and to the better prediction of future climate [Liou, 1986]. In many climate models, much uncertainty still exists in the physical treatment of ice clouds and therefore their radiative effects, especially through the representation of the vertical distribution of ice cloud microphysics in the model [Jakob, 2002]. This is partly because active sensors suitable for ice cloud detection such as cloud radar have not become popular until recently, and also because the variety in the vertical distribution of cloud ice microphysics obtained by these active sensors were less combined with their driving mechanisms for formation and maintenance (e.g., large-scale circulation and deep convection), which have been investigated from passive satellites in the past [Luo and Rossow, 2004]. In this study, ice cloud vertical information obtained by active sensors (i.e., radar/lidar) from the R/V Mirai of the Japan Agency for Marine–Earth Science and Technology (JAMSTEC) in the tropical western Pacific is combined with the International Satellite Cloud Climatology (ISCCP) D1 data to investigate the reproducibility of ice clouds in the Center for Climate System Research–National Institute for Environmental Studies–Frontier Research Center for Global Change (CCSR–NIES–FRCGC) atmospheric general circulation model (AGCM) with the Spectral Radiation–Transport Model for Aerosol Species [Takemura et al., 2005]. Evaluation of the vertical distribution of clouds and also aerosols in the CCSR–NIES–FRCGC-AGCM has been performed in several studies in the past with the same data set [Okamoto et al., 2007, 2008; Nishizawa et al. 2008]. An efficient approach demonstrated by such studies is a comparison of cloud statistics collected by the single-point measurements of active sensors with the clouds produced at the coarser resolutions of a GCM, i.e., subgrid treatment with a vertical overlap assumption for the simulated cloud fields [Jakob and Klein, 1999] and the radar/lidar simulator. The present work extends their study for ice clouds in two new aspects. First is the direct comparison of the reproducibility of reff and IWCgm in the model for ice clouds at least observed by radar, where previous studies dealt with signal-to-signal comparisons of the observed and simulated radar reflectivity Ze and lidar backscattering coefficient βbk only for the lidar/radar overlap region. The analysis method used here provided reff, IWC, and the in-cloud vertical motion, Vair, by combining radar Ze and Doppler velocity VD and lidar βbk. The combination of Ze and VD is powerful for the retrieval of ice cloud microphysics [Mace et al., 2002; Sato and Okamoto, 2006a]; the disadvantage, however, is that the effect of Vair on VD produces uncertainty in cloud microphysics, especially for small particles. Because the method used here can simultaneously retrieve in-cloud Vair and cloud microphysics, the estimations are much improved [Sato et al., 2009]. The CCSR–NIES–FRCGC AGCM also predicts reff from variables such as subgrid-scale vertical velocity and ice nuclei; therefore, ongoing analysis of this retrieval method should allow us to evaluate the parameterization for particle sizing, as a next step. The second aspect is the analysis of the radar/lidar derived ice cloud vertical properties based on the observed cloud regimes, and further evaluation of those simulated profiles in the AGCM. Many studies have tested methods that classify cloud types and relate them to the dynamic and thermodynamic features of their background environments [Jakob et al., 2005]. Among the methodologies, a cluster analysis of the ISCCP D1 data set over a 20 year period in the tropics identified six cloud type/mixture categories based on the joint histogram of cloud top pressure (CTP) and cloud optical thickness (COT) [Rossow et al., 2005]. Because the ISCCP regime classification is based on such joint histogram of variables connected to cloud radiative effect and hydrological balance, it is a good indicator of these impacts. It is expected that the AGCM should have the potential to reproduce some of the variability in the mean cloud field revealed in the ISCCP 2.5 degree domain (i.e., weather regime), which is approximately two to three times larger than the AGCM resolution. However, because the ISCCP weather regimes have been created by subsampling clouds in an approximately 5 km grid within the 2.5° domain, the cloud statistics from the Mirai micro-scale observation could be connected to the weather regime and be used to further evaluate the AGCM. Such consistency between the ISCCP weather regime and cloud statistics derived from point measurements has been demonstrated in several studies. Zhang et al. [2007] found rough agreement in the representation of ISCCP weather regimes as the major regimes in the tropics by performing similar cluster analysis on CloudSat data. They also addressed some issues with the ISCCP weather regimes arising from cloud misclassification. Jakob et al. [2005], using 2 years of ground-based cloud radar data as well as other meteorological and radiative data collected by the Atmospheric Radiation Measurement Program (ARM) at Manus Island, Papua New Guinea, effectively demonstrated that the statistics derived from single-point measurements at a limited site matched the expected results of large-scale observations by the ISCCP, despite the differences in the observations' horizontal and temporal scales. Therefore, a further evaluation of simulated cloud properties based on observed characteristics for each weather regime should contribute to the generalization of model results and an improvement in model performance.

[3] In section 2, we summarize the cruise data and the retrieval algorithm, and investigate the relationship between the observed cloud vertical structure and ISCCP weather regime to roughly show their consistency over the TWP in this 3 months data set. In section 3, we briefly introduce an approach that compares observations to the AGCM simulation, and evaluate the ability of the model to reproduce the ice cloud microphysical features within each weather regime. Finally, section 4 summarizes the study.

2. Observation Data

2.1. Shipborne Radar and Lidar Data

[4] The R/V Mirai has been used to collect data for atmospheric and oceanic studies since 1999. Various cruise tracks have been taken, with particular focus on the western and eastern tropical Indian Ocean to investigate the sea-air interactions in that region. The target radar data in this study were obtained during cruises (coded MR01K05) in the TWP from 25 September to 9 December 2001 (Figure 1). The Special Polarimetric Ice Detection and Explication Radar (SPIDER) system from the National institute of Information Communication and Technology (NICT) and dual-polarization lidar from the National Institute of Environmental Studies (NIES) [Sugimoto et al., 2002] were operated during these cruises. For the analysis, the data were averaged to have vertical and temporal resolutions of 82.5 m and 1 min, respectively. A cloud mask was applied to the lidar/radar data. Ice clouds were considered to be those detected at a temperature below the freezing level. Temperature, pressure, and relative humidity radiosonde measurements were obtained every 3 h during the last half of the observation period, and National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data were used for the first half of the period. A rain gauge was used to measure precipitation. After the cloud mask was applied, a correction for radar signal attenuation due to water vapor and precipitation was performed to allow further analysis of ice microphysics [Okamoto et al., 2008; Sato et al., 2009]. In addition, ship attitude data (i.e., role, pitch, and yaw), the speed and drifting velocity of the ship, and horizontal wind vector data were used to avoid horizontal wind contamination of VD arising from radar antenna inclination from the vertical (which was negligible in most cases). Ship attitude data were unavailable on 10, 16, and 17 October; 6 November; and 6 December. These dates were eliminated from the analysis. In addition, no observations were taken from 12 to 15 October. All of the variables in the time-height cross section for these days are left blank in the figures. In Figure 2, the time-height plot for VD above 6 km obtained by radar during the observation period is shown to illustrate the cloud vertical extent for the period. The record number on the horizontal axis denotes the date, counted from 9 September. Figure 2 shows that periods with fewer high clouds were embedded in continuous deep convective events. This indicates the alternative occurrence of convectively suppressed and active regimes; for further details on this cruise, see Okamoto et al. [2008]. The negative/positive signs of VD indicate downward motion/upward motion, respectively. Vertical change was often found in VD, with stronger negative values (VD < −0.5 m s−1) occurring below 10 km and weaker negative or positive values (−0.5 < VD m s−1) observed above 10 km.

Figure 1.

R/V Mirai cruise tracks for MR01K05.

Figure 2.

Time–height plot of the observed VD obtained during the cruise period.

2.2. Ice Cloud Microphysics Analysis

[5] The ice microphysics and Vair were derived using the method of Sato et al. [2009]. This method combines multiple parameters of 95 GHz radar (Ze, VD, and the linear depolarization ratio LDR) and lidar βbk. The radar quantities in the look-up table are based on discrete dipole approximation (DDA) for the coexistence of column and bullet rosette types with varying mixing ratios [Sato and Okamoto, 2006b]. The method simultaneously obtains reff, IWC, Vair, and/or the habit mixing ratio (depending on the use of LDR) throughout the cloud layers that could be observed by radar. Basically, it is possible to derive four unknowns, i.e., reff, IWC, Vair, and the habit mixing ratio, from four independent observables of the radar and lidar. For cloud portions observed only by radar (which occur mostly in areas where the lidar signals become attenuated), the method effectively expands the retrieval as follows. First, ice microphysics is retrieved for the lidar-radar overlap region of the cloud by the radar/lidar algorithm, which uses the different dependencies on reff of the two sensors, i.e., for the radar Ze∼reff3 and Ze ∼IWC for small ice particle size, while for the lidar, βbk ∼1/reff and βbk ∼IWC [Okamoto et al., 2003]. From these microphysical properties, the radar reflectivity–weighted particle fall speed, Vtz, is estimated, which further enables us to estimate the Vair of the lidar and radar region through Vair = VDVtz. Then Vair for the radar-only region is retrieved from the difference in the radar observables VD and Ze, which reflect the Vair differences. Once Vair is estimated, cloud microphysics for the radar-only region are obtained using Ze and/or LDR and Vtz derived from the correction of Vair in VD. Compared to common lidar and radar algorithms or radar-only (Ze and VD) algorithms, this method overcomes the drawbacks of the two sensors, i.e., the attenuation of the lidar signal by a thick water cloud underneath and the Vair effect on radar VD. Validation of the retrieval method has been performed against ice microphysical properties collected during a series of in situ measurements in both stratiform and convective clouds [Heymsfield et al., 2008] and direct Vair measurement by colocated wind profiler (Equatorial Atmospheric Radar) observations [Sato et al., 2009]. Those studies showed that with the separation of Vair from VD, the retrieval accuracy for reff and IWC is improved by about two and eight times, respectively, compared to the case without Vair correction. In particular, improvements in IWC retrieval are achieved, most significantly at IWC > 0.005 g m−3. This is important because a larger IWC generally has a larger radiative impact. Table 1 summarizes the retrieval accuracy of the method evaluated by direct measurements. Although observational resources are limited, the Vair product is considered to be sufficient for practical use. The greatest advantage of this retrieval algorithm is that it produces small bias errors in the IWC retrieval, and, hence, also in reff [Heymsfield et al., 2008]. Therefore, as indicated in Table 1, a good estimation of IWP is one of the biggest advantages of the method.

Table 1. Summary of the Retrieval Accuracy for Each Variable Against Direct Measurements
VariableRetrieval AccuracyaValidation MethodReference
  • a

    Error denotes the difference from the true value, whereas ratio indicates the retrieved-to-measured value.

  • b

    Cloud average value over ∼12 h.

VairError 0.009 ± 0.119 m s−1bWind profilerSato et al. [2009]
IWCRatio 0.93 ± 0.93In situHeymsfield et al. [2008]
IWPRatio 0.95 ± 0.15In situHeymsfield et al. [2008]

[6] Figures 3a3d show the time-height cross section of Vair, Vtz and the ice cloud microphysics obtained for 1 min and 82.5 m resolution by the method used for the cruise data. We could analyze the Vair and ice cloud microphysics for more than 70% of the radar volumes. Above 6 km, about 57% of the retrieved Vair was updraft, and the mean updraft velocity during the whole cruise increased from about 5 cm s−1 at 10 km to about 20 cm s−1 at around 14–15 km; the mean updraft velocity then decreased with height. These intensities are comparable to the average air motion profile for cloudy scenes in the TWP obtained by the high-frequency wind profiler measurements by Balsley et al. [1988] (Figures 2 and 3). Therefore, it is expected that the effect of Vair on ice cloud microphysics retrieval was successfully removed. During the period, reff varied from 10 μm at high cloud tops to more than 100 μm at low altitudes, whereas IWC ranged between about 10−4 and ∼1 g m−3.

Figure 3.

Time–height plot of the retrieved (a) Vair, (b) Vtz, (c) IWCin, and (c) reff obtained by lidar and radar observation during the cruise period.

3. Results

3.1. Relationship Between Observed Cloud and ISCCP Weather Regimes

[7] Using ISCCP cluster analysis, a joint histogram of CTP (7 bins) and COT (6 bins) obtained within each 2.5° domain every 3 h was estimated and further classified into six categories of “weather states” (WSs) following Rossow et al. [2005]. In the joint histogram, CTPs in the ISCCP “high-cloud,” “midcloud,” and “low-cloud” categories were further divided into two to three equal pressure bins. By the occurrence frequency of deep convective clouds, WS1∼WS3 and WS4∼WS6 were grouped as convectively active and convectively inactive regimes, respectively. WS1 was a more organized deep convective regime, WS2 was a cirrus and cirrostratus regime, WS3 was an isolated deep convective regime that also included many cumulus congestus, WS4 included all of the remnant cirrus clouds that either emerged from convective outflow or were created in situ, WS5 was a trade and shallow cumulus regime, and WS6 was a marine stratus regime. The 3 h daytime-only ISCCP WS product was matched to the ship location to assign the vertical properties of clouds observed every minute by active sensors to a regime. The ISCCP CTP and COT histograms for the WSs matched to the Mirai location are shown in Figure 4. The frequencies at which each WS occurred throughout the whole observation period were 24%, 26%, 20%, 19%, 10%, and 1%, for WS1–WS6, respectively. In the study period, no ice cloud was observed by radar during WS6, and thus statistics for this regime are not presented here.

Figure 4.

ISCCP cloud top pressure-cloud optical thickness histogram of the WSs matched to the Mirai location.

3.1.1. Vertical Distribution of Cloud Amount and Total Ice Amount

[8] In this section we discuss the relationship between the radar-observed total ice cloud fraction CFice and total ice amount and the ISCCP WS category. For the ship observation, CFice was obtained by counting the cloudy pixels of 1 min data for each 40 min period (Figures 5a and 5b). Observed CFice was assumed to be 1 (100%) for a minute during which a 1 min pixel was determined, by the cloud mask scheme, to be cloudy. This time resolution is the same as that of the AGCM. Thick lines/thin solid lines (attached to the thick lines) in the figure correspond to the case when in-cloud dbZe < 0 dBZe was not/was considered in the CFice profile. This threshold is often considered to be a proxy for ice precipitation [Luo et al., 2008; Okamoto et al., 2008] and provides a more realistic ice cloud fraction, not affected by ice-precipitation contamination, with which to later evaluate the model.

Figure 5.

Vertical profile of average cloud fraction obtained every 40 min for the convectively (a) active and (b) inactive regimes. The colored lines in Figure 5a and Figure 5b indicate cases in which a threshold of dBZe,in < 0 dBZe was considered. Note that the profiles below about 4 km include drizzle. (c) Relative humidity profile for each WS obtained by radiosonde measurement.

[9] As shown by Figures 5a and 5b, the frequent occurrence of deep convective clouds inferred from the ISCCP product corresponded well to the increase in high cloud amount found by the single-point radar measurements. Both the altitude and CFice peak value were higher for more convectively active regimes; e.g., CFice − 0.42 for WS1 was two to four times larger than in other regimes. CFice peak altitudes were around 12 km, 11–12 km, and 7 and 11 km for WS1, WS2–WS4, and WS5, respectively. Ice precipitation is inferred from the difference between the two profiles with and without the in-cloud dbZe < 0 dBZe threshold. Distinctive differences also were found in the frequency and height of ice precipitation occurrence. Ice precipitation occurred at higher altitudes in more convectively active regimes; that is., ice precipitation existed from altitudes as high as 12 km, 10 km, and 7 km for WS1/WS2, WS3/WS4, and WS5, respectively. A huge difference in hydrometeor occurrence below the freezing level was found between WS1 and the other WSs. WS1 had the strongest mean vertical updraft of the regimes occurring above about 10 km [Sato, 2009]. Vertical profiles of the average relative humidity for each WS, which were deduced from radiosonde measurement available for the latter observation period (Figure 5c), qualitatively show that WS3–WS5 corresponded to relatively dry atmospheric conditions above 6 km. Note that the cloud fraction profile for the whole observation period was roughly similar to that deduced for the last half of the observation period.

[10] Figure 6 shows the average IWP in log-scale for each WS, which is estimated from integrating the in-cloud IWC above the freezing level. More active regimes had more total ice mass (and also a larger total liquid water path, as indicated by hydrometeor statistics derived from radar Ze and VD below the meting level), and therefore generally had optically thicker clouds and a good relationship with the WS definition observed at a larger scale. The exception was WS5, which had the second largest IWP among the WSs, equaling approximately 3/4 of the IWP for WS1. This may have been affected by the small sample size for WS5 in the radar observation and by the existence of small fraction of deep clouds in ISCCP CTP-COT histogram for WS5 (Figure 4).

Figure 6.

Mean and standard deviation of IWP in logscale estimated for each WS.

[11] In Figure 7, the analog to the ISCCP cluster product, radar observables, and microphysics were classified according to the uppermost CTP observed every minute. The resulting statistics for three categories: 440 hPa < CTP < 310 hPa, 310 hPa < CTP < 180 hPa, and 180 hPa < CTP are shown in the figure. For the same WS, the IWP was larger for clouds with lower CTP (i.e., higher altitude), and for the same CTP category, the IWP was larger for more active (inactive) regimes at low (high) CTP categories (Figure 7a). The mean IWC showed a trend similar to that of IWP; however, the differences among the WSs were not so large at 180 hPa < CTP (Figure 7b). Together, the IWC and IWP results indicate that high clouds are geometrically thicker in more active regimes, and these thicker clouds produce the large mean IWP shown in Figure 6, especially for the two most active regimes (WS1 and WS2). In contrast to these regimes, WS5 had large IWC in the 440 hPa < CTP < 310 hPa category, which contributed to the IWP shown in Figure 6 for this regime. Results for reff and radar observables are also shown (Figures 7c and 7d). For a given WS, Reff became larger with larger CTP categories and showed a trend opposite to that of the IWC. From the VD profile, reff mostly reflected the trend observed in VD. The tendency in dBZe was similar to that of IWP although the differences in dBZe among WS1∼WS4 were less obvious compared to those found for IWP.

Figure 7.

Mean and standard deviation of (a) IWP, (b) IWC, (c) reff, (d) dBZe, and (e) VD in each WS for three CTP categories: 1, 440 hPa > CTP > 310 hPa; 2, 310 hPa > CTP > 180 hPa; and 3, 180 hPa > CTP.

3.1.2. Mean Vertical Profile of Microphysics

[12] Figures 8a8d compare the mean microphysics among WSs as a function of altitude. As mentioned above, highly convectively active regimes have larger IWC at the same altitude compared to less active regimes with ice precipitation from deep clouds especially below 11 km. The rather vertically homogeneous IWC (about 0.3 g m−3) of WS1 as well as the gradual growth of its reff from 20 to 80 μm with decreasing altitude indicate that the number concentration of ice is higher in upper layers and aggregation processes occurred during sedimentation. On average, Reff in WS1 is also the largest among the WSs at the same altitude except at the very top altitudes. Contrary to the gradual increase in reff from high altitudes in WS1, reff for the other regimes started to increase rapidly below 8–10 km (from lower cloud tops as shown in Figure 7c) with an increase in humidity and was associated with a smaller fraction of precipitating ice than in WS1. IWC and reff were the least continuous in the vertical for the least convectively active regime, WS5 and vertical structures of the microphysics were quite different from those of other WSs.

Figure 8.

Mean vertical profile of the retrieved Figures 8a and 8b, IWC and Figures 8c and 8d, reff. Figures 8a and 8c and Figures 8b and 8d show WS1–WS3 and WS4–WS5, respectively. CTP, 1; CTP, 2; and CTP, 3 in Figure 8a correspond to the same CTP categories as in Figure 7.

3.2. Comparison With Ice Cloud Generated in the AGCM

[13] In this section, we evaluate the simulated CFice and IWCgm, which characterize the IWP versus CTP differences among the WSs and reff in the observation cases.

3.2.1. Comparison Method

[14] The observed cloud properties were compared to those simulated by the CCSR–NIES–FRGCG AGCM [Takemura et al., 2005] every 40 min along the cruise tracks in the model's prediction mode; that is., temperature, pressure, and relative humidity estimated by the AGCM were nudged with 6 h NCEP/NCAR reanalysis data. The AGCM treated 20 vertical levels from the surface to about 33 km in height, with a time step length of 40 min, except for the radiation flux calculation, which was converted every 3 h. The model also predicted the concentrations of four types of aerosols: sulfates, carbonaceous aerosols, sea salt, and dust. The direct outputs from the AGCM for ice cloud microphysics were IWCGM and CFice, which were parameterized differently for ice clouds formed by large-scale ascent (stratiform clouds) and those triggered by convection (convective clouds). The IWC generated in cloud (IWCin) can be obtained by dividing the estimated IWCgm by CFice. Note that the IWC from the model includes convective ice. In the AGCM, reff of the ice clouds also was predicted by taking into account homogeneous and heterogeneous nucleation according to the parameterized subgrid-scale vertical velocity and the ice nuclei concentration estimated from simulated carbonaceous and dust aerosol concentrations.

[15] For adequate comparison of the quantities, only AGCM outputs that would be observable by radar were considered, following the same steps as Okamoto et al. [2008]. First, radar Ze was simulated from the IWCin and reff in the AGCM using the radar/lidar simulator [Okamoto et al., 2003]. For this calculation, the subgrid-scale clouds in the AGCM were considered as follows. Each original grid box was subdivided into 100 subcolumns and the cloud fraction was set to 0 or 1 in each subcolumn to distribute the CFice and the microphysics of a model grid box to the corresponding subgrid boxes. Maximum random overlap was assumed; that is., random overlap was assumed for clouds that did not occur in adjacent vertical levels but maximum overlap was assumed for clouds occurring at adjacent levels. The same cloud microphysics (IWCin and reff) was used in each subbox. The simulated Ze values for these subboxes were attenuated according to the amounts of water vapor, oxygen, and lower clouds, and the precipitation profile in the model. The same cloud mask scheme as used for the Mirai data (C1 mask of Okamoto et al. [2008]) was then applied. The threshold for precipitation was the same as that used for the observation data; that is., records with rain rates larger than 0.5 mm h−1 at the model surface were determined to be “precipitating” records. When the simulated Ze for these subboxes did not pass the threshold for radar sensitivity for 1 min, then the cloud fraction and cloud microphysics for the subbox were set to 0. The CFice and grid mean cloud microphysics (also the grid mean IWP, IWPgm) were then recalculated for the comparison by averaging the variables over these subgrid boxes.

[16] In the observation, the IWC and reff were averaged over 40 min at 82.5 m vertical resolution (hereafter, IWCin is used for the 40 min averaged observed IWC). In some cloud portions the microphysics could not be retrieved at the original time resolution; in these cases, the 40 min averages of the microphysics were estimated by averaging the retrieved cloud microphysics as the representative value for the corresponding model grid. The IWCgm was calculated from IWCin and CFice for 40 min (estimated from Ze) at 82.5 m vertical resolution from IWCgm = IWCin × CFice. IWPgm was estimated from the sum of IWCgm in the vertical. Because the IWC in the AGCM did not include the contribution of ice precipitation, a threshold of dbZe,in < 0 dBZe was imposed for both the observation and AGCM data to create the ice cloud statistics.

3.2.2. General Performance of the Ice Cloud Representation in the AGCM

[17] Figures 9a9d show the observed and simulated cloud occurrence and their microphysical properties. The model successfully simulated the overall cloudiness. However, the model tended to produce low ice cloud tops (i.e., occurring at midlevels) at times that did not correspond to the observation, e.g., around record number 25 (60), too-shallow (deep) ice cloud tops were simulated where deep (shallow) ice clouds were observed, and around record number 30, no ice clouds were simulated when shallow ice cloud tops were observed. The IWCgm of the AGCM had more homogeneous vertical structure compared to the observation due to its coarser vertical resolution; that is., the IWCgm profile above 6 km changed by 10−3 g m−3, 10−2 gm−3, and 10−3 gm−3 at the cloud top, midlevel, and ice cloud bottom, respectively. Modeled reff increased with decreasing altitude, similar to the observation. However, the increase in reff to sizes more than 80 μm around 9 km in the model was much steeper than in the observation. Figure 10 shows vertical averages of these properties for the entire observation and simulation periods. Without the dBZe,in < 0 dBZe threshold, the AGCM overpredicted the mean CFice (Figure 10a) by more than 50%, as also reported by Okamoto et al. [2008]. When the threshold was considered, the observed CFice decreased but the CFice profile in the AGCM was insensitive to the threshold because the fraction where dBZe,in > 0 dBZe was small, and therefore the difference between observed and simulated CFice became wider. This result is understandable given that the modeled ice cloud did not contain ice precipitation. In the case of dBZe,gm (Figure 10b), although the simulated peak of the frequency distribution for dBZe,gm was underestimated in the model above 11km (figure not shown), the averaged profile of dBZe,gm became comparable to the observation when the dBZe,in < 0 dBZe threshold was used, essentially due to the removal of precipitating ice (a maximum 3 dB difference between the observation and simulation). Figures 10c and 10d show the performance of the simulated ice microphysics. The simulated IWCgm underestimated the observed value throughout the vertical profile (Figure 10c). The predicted reff was underestimated/overestimated in the model above/below 11 km (Figure 10d). These results indicate that, below 11 km, good estimation of dBZe,gm in the model resulted from the compensation between overestimation in the predicted reff, which generated large dBZe,gm, and underestimation of the predicted IWCgm, which produced lower dBZe,gm. However, this situation also applied above 11 km for the average profile of dBZe,gm. Above 11 km, the model underestimated mean reff but predicted large reff (>40 μm) in one-third of the cases; this contributed to the production of a large dBZe,gm which compensated for the low dBZe,gm produced by the small IWCgm. Sensitivity of cases when dBZe, in <−10 dBZe was taken into account in the comparison for Figures 10a10d was also examined (figure not shown) and turned out to be similar to the dBZe,in < 0 dBZe case except that the simulated CFice became more comparable to the observation below 9 km.

Figure 9.

Time–height plot of IWCgm/reff with a 0 dB threshold for the Figure 9a/9b observation and Figure 9c/9d simulation.

Figure 10.

Vertical profiles of the observed and simulated (a) cloud fraction, (b) dBZe,gm, (c) IWCgm, and (d) reff. Differences in the lines indicate cases with and without the dBZe,gm < 0 dBZe threshold.

3.2.3. The WSs and Reproducibility of Ice Cloud Characteristics in the AGCM

[18] Next we investigate differences in the predictability of ice cloud occurrence by weather regime and the microphysical properties of cloud occurrence in the AGCM. Cloud Fraction

[19] Figures 11a and 11b show the vertical profiles of the simulated CFice for each WS corresponding to Figures 5a and 5b (thin solid line) for the observation. For simplicity, evaluation of the simulated CFice is summarized below for two height categories bounded at 11 km above the freezing level.

Figure 11.

Comparison between (a) observed and (b) simulated CF (dbZe,in < 0dBZe). Note that liquid precipitation is included in the profile below the melting-level altitude in the observation.

[20] Above 11 km, simulated CFice decreased in the order WS2→WS4→WS1→WS3→WS5 and more ice cloud occurred in the cirrostratus and cirrus regimes, WS2 and WS4, respectively. Evaluation of the observed CFice revealed that CFice was not always overpredicted in the AGCM when the WS was introduced. In WS1 only, simulated CFice was underpredicted, by about 50%, due to the frequent occurrence of clear sky. The location of the simulated CFice peak was also underestimated by 2 km. For other WSs, the simulated location of the CFice peak was comparable to observed peaks, but CFice values in the AGCM were overpredicted, i.e., overestimated by approximately two times for the active regimes WS2 and WS3 and by more than four times for the inactive regimes WS4 and WS5.

[21] Below 11 km, the model successfully reproduced larger CFice, as the regime became more convectively active, which was similar to the observation. Around 8 km, the predicted CFice value was underestimated by approximately 50% compared to the observation for WS1 but overestimated by more than two times for other WSs.

[22] These results were less sensitive to the time resolution used for the comparison. Considering the wind speed of 30 m/s, the horizontal scale of the clouds observed by radar in 40 min is 72 km, which is smaller than the GCM horizontal scale of 100 km. Therefore sensitivity analysis was performed, in which CFice values were calculated every 6 h for both observed and modeled data and compared to 40 min data. The results show that whether the AGCM overestimated or underestimated the CFice relative to observed values did not change significantly between the 40 min and 6 h simulations, because the frequency with which a certain CFice bin occurred changed in the same manner for the two time resolutions, in both the observation and in the model. Grid Mean IWP

[23] The observed IWPgm was smaller than the 1 min IWP (Figure 6) mainly due to CFice and the dBZe,in < 0 dBZ threshold. There was a clear shift in the observed IWPgm from the largest value at WS1 (414 g m−2) to the smallest value at WS5 (74.7 g m−2) as the regime shifted from the active to the inactive phase (Figure 12a). The model also reproduced similar trend in IWPgm with that observed though the contrast was much weaker than observation, and the simulated IWPgm was underestimated for all WS (i.e., about 70% for WS1, 45% for WS2, WS3, WS4, and 30% for WS5). Observed IWPgm values were much larger when the cloud top was higher and the frequency at which these highest categories occurred contributed largely to the total IWPgm profile in Figure 12a. An exception was WS5, which had the opposite tendency (Figure 12b). In case of AGCM, the observed relationship between IWPgm and CTP was reproduced in each WS expect for WS5, but IWPgm was generally underestimated compared to observation. In addition, the CTP category that dominated within a WS, were the ones with the largest IWPgm for observation, while it were not in the AGCM except for WS3 (figure not shown), and this also contributed to the difference between the observed and simulated IWPgm in Figure 12a.

Figure 12.

Comparison of observed and simulated (a) total IWPgm (dbZe < 0dB) and (b) IWPgm (dbZe < 0dB) for three CTP categories: 1, 440 hPa > CTP > 310 hPa; 2, 310 hPa > CTP > 180 hPa; and 3, 180 hPa > CTP. The observation is shown in black or in colors in Figure 12a and 12b, respectively, whereas the corresponding simulated profile is plotted by gray bars on the right-hand side of each observation profile. Grid Mean IWC

[24] In the vertical profile, above 11 km, the dBZe,in < 0 dBZe threshold had a minor impact on the comparison and the simulation underestimated IWCgm by orders of magnitude, regardless of the WS (Figures 13a and 13b). Because of the dBZe,in < 0 dBZe threshold, the observed IWCgm below 11 km became smaller than the original profile for the convectively active regimes and the differences among WS1 through WS4 became small. The simulated vertical profile of IWCgm was comparable to observed values below 11 km for the active regimes, but was underestimated for the inactive regimes. For WS5, the underestimation was more than an order of magnitude.

Figure 13.

Comparison of observed and simulated IWCgm for (a) convectively active regimes and (b) convectively inactive regimes.

[25] In the frequency distribution, the observed and simulated frequency distributions of IWCgm for two height categories bounded at 11 km are examined in Figures 14 and 15.

Figure 14.

Frequency distributions of observed and simulated IWCgm for the altitude range of 11–16 km. Solid thick, dashed, and dotted lines correspond to observations with and without the dBze,in < 0 dBZe threshold and the AGCM simulation, respectively.

Figure 15.

Same as in Figure 14 but for the altitude range of 6–11 km.

[26] Above 11 km, the observed peak value of the frequency distribution for IWCgm was located at about 10−1 g m−3 in the observation regardless of WS, whereas the width of the distribution broadened as the regimes became more convectively inactive (Figures 14a14e). The AGCM also successfully reproduced the broader distribution width for more convectively inactive regimes, but the simulated peak value was an order of magnitude smaller than the observed peak. This was due to the too-small cutoff value of IWCgm around 10−1 g m−3 in the model explained as follow. In the model, sedimentation of ice particles P [kg m−3s−1] is parameterized in the form P = {[V0(ρIWCgm)/Δz]δ}IWCgm, where Δz is the cloud layer thickness [m], ρ is the density of air [kg m−3], and others are constants, i.e., Vo = 3.5 and δ = 0.17. Rough estimate of the loss of cloud ice mass according to this equation in 40 min indicates that the removal of cloud ice exceeds the generated ice mass when IWCgm is larger than a few milligrams per cubic meter. Therefore, although there is no explicit cutoff at IWCgm around 0.1 g m−3 in the model, due to the ice sedimentation scheme, the ice cloud cannot contain IWCgm larger than about ∼10−1 g m−3.

[27] The observed peak value of the frequency distribution of IWCgm was around 10−1.5 g m−3 (Figures 15a15e). Because of the ice cloud threshold, the observed and simulated frequency distribution for IWCgm was almost comparable for the active regimes, WS1–WS3. The model underestimated peak values for the inactive regimes by an order of magnitude. In this height range, the cutoff at IWCgm > 10−1 g m−3 in the model seems to be too small. The AGCM often produced too much of a too diluted cloud. Effective Radius

[28] Next, the reff for each WS was estimated with the 0 dBZe threshold, and the observed particle size (Figures 16a and 16b) can be seen to decrease with the vertical profile of reff in Figures 8c and 8d. Above 11 km, the simulated reff was underestimated by about 50% at high altitudes compared to the observation in the active regimes, but was comparable to observed values in the inactive regimes. Below 11 km, the model accurately simulated the gradual increase in the observed reff with decreasing height, although the simulated reff was generally overestimated in the AGCM by more than 20 μm (about 30%) for all WSs.

Figure 16.

Mean vertical profile of reff simulated by the AGCM for (a) WS1–WS3 and (b) WS4 and WS5.

[29] The observed and simulated width and peak value of the frequency distribution for reff were rather similar (around 20 μm) except for WS5 (Figures 17a17d). In the case of WS5, the simulated peak value was similar but the observed distribution peak (50 μm) was larger, which caused the simulated value to be overestimated by 50%. However, the AGCM predicted a larger fraction of reff around 100 μm. This fraction increased as the regime became less convectively active.

Figure 17.

Same as Figure 14, but for reff for the altitude range of 11–16 km.

[30] At 6–11 km, the dBZe,in < 0 dBZe threshold contributed to a reduction in particles with reff > 60 μm in the observation (Figures 18a18d). The observed reff peaks were around 40 μm and 20 μm for WS1–WS4 and WS5, respectively. In the AGCM, reff was concentrated around 100 μm regardless of WS category, resulting in an overestimation. At this altitude range, black carbon and soil dust aerosols act efficiently as ice nuclei for the heterogeneous freezing (i.e., contact and immersion/condensation) processes in the AGCM [Takemura et al., 2009], and especially the contact freezing process becomes dominant at warmer temperatures [Lohmann and Diehl, 2006]. Number concentration of cloud ice generated by contact freezing Qcnt [m−3s−1] is parameterized in the form Qcnt= mioDap4πrlNa,cntNl2/(ρql), where Dap is the Brownian aerosol diffusivity, rl is the volume mean cloud droplet radius, Nl is the liquid droplet number concentration, ρ and ρl are the density of air and water, respectively, ql is the cloud liquid water mass mixing ratio, and Na,cnt is the number concentration of activated ice nuclei for contact freezing among the total number concentration of black carbon and dust [Lohmann and Diehl, 2006]. The coefficients mio and ac, bc, Tio are summarized by Lohmann and Diehl [2006] and Table C1 of Takemura et al. [2009], respectively. In the AGCM, reff is further obtained in proportion to the simulated IWCgm divided by the cloud ice number concentration estimated from the equation above. Recalling that the simulated IWCgm was smaller than observation, this suggests that the AGCM may have underestimated the number concentration of activated ice nuclei (Na,cnt) or/and cloud liquid droplets (Nl), and produced large reff too often as a result. In the AGCM case, the size cutoff for cloud ice occurred around reff = 120 μm. This cutoff seems to be quite reasonable for this study, although a few larger particles, with reff > 120 μm, were observed, especially in convectively active regimes, and this would not have appeared in the AGCM output. When the 0 dBZe threshold was not applied to the observed data (dashed line in Figures 17 and 18), reff values up to about 200 μm were observed both above and below 11 km; these large particles were excluded in the comparison with the AGCM when the 0 dBZe threshold was used.

Figure 18.

Same as Figure 15 but for the altitude range of 6–11 km.

4. Summary

[31] In this study, a synergetic algorithm was applied to the 3 months Doppler radar and lidar data obtained by R/V Mirai over the TWP in late 2001 to retrieve the vertical structure of ice cloud microphysics and vertical air motion in that region. The relationship between the retrieved cloud properties from the ship-based measurement and the weather regime classification of the ISCCP cluster analysis (weather state: WS) was investigated, focusing on CFice, IWP, and CTP. These observed relationships between the WS and ice cloud vertical structure were then used to evaluate the AGCM-simulated clouds along the cruise track with the cloud ice threshold (dBZe,gm < 0 dBZe). The main findings are as follows:

[32] 1. During the cruise, about 70% of the observation period was characterized as a convectively active cloud regime by the ISCCP cluster analysis. Ice clouds observed by the Mirai were roughly consistent with the WS classification in the following ways: First, there was a good relationship between the observed cloudiness and the WS, and more high clouds, indicated by larger CFice and higher CFice peak altitudes, occurred in more convectively active regimes. Second, a larger number of geometrically thicker clouds with high cloud tops, larger IWP, and a higher occurrence of ice precipitation were observed in more convectively active regimes. Third, we examined the relationship of cloud top height to total IWP for each WS and found that clouds with high (low) cloud top contributed more to the total IWP for the active (least active) regimes. Last, the environment of the observed clouds was more humid in more convectively active regimes.

[33] 2. The comparison between observation and AGCM revealed the following: First, the model underestimated the CFice, by about 50%, in the most active regime (WS1) above 11 km, whereas in the other regimes, it overestimated CFice by about two to four times. Although the observation showed the largest CFice in WS1, the model tended to predict large CFice in the cirrostratus and cirrus regimes (WS2 and WS4). Second, IWPgm predicted by the model could only weakly reproduce the observed relation between WS and IWPgm, and were generally underestimated despite the WS and CTP category. Investigation of the simulated IWCgm, which characterizes the IWPgm, showed that the predicted IWCgm was especially small above 11 km, which underestimated the observed value by an order of magnitude. This primarily originated from the cutoff value of IWCgm around 0.1 g m−3 resulting from the too fast ice sedimentation rate in the model when cloud ice exceeds a few milligrams per cubic meter; that is., 0.1 g m−3 was too small. Third, the AGCM-predicted peak value of the reff frequency distribution was generally consistent with the observed peak above 11 km (i.e., about 20 μm); however, the model produced large overestimations below 11 km due to the excessively large fraction of reff around 100 μm. This indicates that the model may have underestimated the activated ice nuclei number concentration or/and the cloud liquid droplet number concentration in the heterogeneous freezing parameterization. The cutoff for cloud ice at reff > 120 μm in the model was found to be quite reasonable, although a small fraction of observed large particles, particularly in convectively active regimes, were missed in the simulation.

[34] 3. In order to improve the modeled cloud ice microphysics, a more reliable ice sedimentation scheme needs to be investigated. One candidate can be the incorporation of simulated reff as an indicator for ice sedimentation rate. In such case, efforts to simulate realistic reff especially at warmer temperatures are required. Since the distribution and fraction of ice nuclei and amount of ice precipitation depends on the atmospheric conditions (e.g., relative humidity and vertical velocity), sensitivity studies classifying these factors by WSs may be effective to adequately reproduce the variety of WSs in the model as observed.


[35] K. Sato was supported by a Grant-in-Aid for Japan Society for the Promotion of Science (JSPS) Fellows (21•6685). This study was also supported by the Ministry of Education, Culture, Sports, Science, and Technology of the Japanese Government through a Grant-in-Aid for Scientific Research (B) (19340132) and by the Global Environment Research Fund of the Ministry of Environment, Japan, B-4. The ISCCP cluster analysis data were obtained from the International Satellite Cloud Climatology Project website maintained by the ISCCP research group at NASA's Goddard Institute for Space Studies, New York. NCEP Reanalysis data were provided by NOAA's Office of Oceanic and Atmospheric Research, Earth System's Research Laboratory, Physical Sciences Division, Boulder, Colorado, from their Web site at Finally, we would like to thank the constructive comments from the three anonymous reviewers to improve the manuscript.