a. The MMCR measurements and retrieval algorithm

The millimeter-wave cloud radar (MMCR) at the SGP site has been described by Moran et al. (1998) and Clothiaux et al. (1995). The radar operates at a frequency of 34.86 GHz (Ka band) with a stated calibration uncertainty of less than approximately 1–2 dB in reflectivity and 10 cm s⁻¹ in mean Doppler velocity and spectrum width. The antenna diameter is 3 m with 57.48-dB gain at the SGP site where the minimum detectable signal is approximately −50 dBZe at 5-km range. The MMCRs at the TWP sites are similar except that their antenna diameters are 2 m with 52.73-dB gain, which makes the MMCR at the SGP site more sensitive by approximately 5 dB. The MMCR used in this study are vertically pointing to within 0.01° of zenith (K. B. Widener 2008, personal communication) so that the Doppler moments contain information regarding the fall speed of the ice crystals and the vertical air motion.

Cirrus cloud identification at SGP is based on the definition of cirrus in Mace et al. (2001). Cirrus identification at the TWP sites is adapted from these criteria. We require the radar echo top to occur at a temperature colder than −35°C, the layer base must be higher than 7 km (which is approximately 400 hPa and a temperature of −10°C), the maximum dBZe in the layer must occur at a temperature colder than −22°C, and the cloud-layer depth must be larger than 180 m or two range bins. As described in Mace et al. (2001), this definition, although arbitrary, helps to exclude deep cloud layers that are not normally classified as cirrus.

The algorithm to retrieve the cirrus microphysical properties and vertical motion as well as the validation and uncertainties of the algorithm are described in detail in Part I. In this algorithm we use the three moments of the Doppler spectrum recorded by the MMCR: the radar reflectivity Ze, the Doppler velocity V_d, and the Doppler spectrum width σ_d. It can be shown (Gossard 1994) that a model describing these measurements can be derived by assuming a convolution between a population of cirrus particles and a turbulent spectrum with some mean vertical motion. Therefore, with an assumption of Rayleigh scattering (Evans and Stephens 1995) for cirrus particles at the 8-mm wavelength of the MMCRs, we further assume that the cirrus particle size distribution can be described as an exponential function (Heymsfield 2003) and that the atmospheric motions can be represented in terms of a mean vertical speed W_m and a double-sided exponential turbulent spectrum with an exponential slope of W_σ. With these assumptions, a forward model of the Doppler moments is inverted for the cirrus particle size distribution and W_m. The primary uncertainties of the algorithm are due to the basic assumptions of the algorithm and the need to represent the particle mass and area in terms of some representative particle habit. Using aircraft data, we determine in Part I that the uncertainties in ice water content (IWC) and mass-weighted diameter D_mass are 50% and 30%, respectively. In the analysis that follows, we apply the algorithm to many years of data and hundreds of thousands of cirrus volumes. We do not account for water vapor attenuation of the radar beam; however, this would bias the radar reflectivity by less than the uncertainty in radar measurements (Clothiaux et al. 1995).

We assume that the retrieval uncertainty is random so that errors of the mean values decrease as the inverse square root of the number of samples. Bias error in the retrievals is difficult to account for, however. Bias in the statistics would occur because of systematic differences between what we assume about the properties of the cirrus and what actually occurs on average. The potential sources of biases would be differences in the functional shape of the particle size distribution (i.e., bimodality) or the particle habit. Our sensitivity studies in Part I showed that the crystal habit assumption is, by far, the most significant source of potential bias. Although our comparison with in situ data in Part I suggests that the bias is minimal, those comparisons were conducted using ice crystal habits that were observed during the flights with which we compared, and the number of actual coincidences was too small to establish a high degree of confidence.

The ice crystal habits influence the retrieval results through the mass– and area–dimensional relationships as described in Part I. Throughout this work, we use mass– and area–dimensional relationships derived from measurements collected in cirrus above SGP during March of 2000 following the general approach of McFarquhar and Heymsfield (1997). If the actual ice crystal habits are systematically different during the periods under consideration here, then we would expect the statistics derived from the long-term data to be biased by some unknown amount. In Part I, we conducted a sensitivity study in which the mass– and area–dimensional relationships were varied by 20%, and we found that the IWC, D_mass, and W_m could vary as much as 75%, 17%, and 15 cm s⁻¹, respectively. We return to this topic below.

From the retrieved particle size distributions, cirrus properties such as the IWC, particle effective radius Re, mass-weighted maximum diameter D_mass, and mass-weighted fall velocity V_fm in the radar sample volumes are calculated as described in Part I. In addition, the optical depth is calculated using the area–dimensional relationships found in Mitchell et al. (1996). We note that because the weighted fall speed of a particle population depends on the ratio of distribution mass to cross-sectional area and the optical extinction depends primarily on the distribution cross-sectional area a retrieval of the distribution mass and fall speed implies a particle distribution cross-sectional area and thereby extinction and, when integrated over the layer, the optical depth.

The ice cloud processes that result in the derived microphysical properties are fundamentally related to updraft velocities within the cloud environment. A full understanding of cirrus cloud properties requires information of this dynamical environment. In this study, the dynamical properties are indicated by W_m in the radar resolution volumes. Given a 0.2° beamwidth, the horizontal scale of measurements at cirrus altitudes is approximately 35 m with depths of 90 m. We assume that these scales approximate cloud-scale motions.

b. MMCR data availability

The MMCRs were designed to operate continuously and autonomously. However, there are significant time periods during which data were not collected because of instrument problems. The percentage of proper MMCR data collection is calculated for each season at the three sites and is shown in Fig. 1. The MMCR at SGP provided useful data more than 90% of the time. However, we do not use the MMCR observations from SGP after 2005 because of a failure of the operational mode that is most sensitive to thin cirrus—this mode outage continued through 2008. The MMCR at TWPC1 was inoperable for a 2-yr period from 2001, which seriously hampers our ability to evaluate seasonal trends at that site.

Among these MMCR observations, there are in total 11 379, 10 370, and 9873 h of cirrus observations at SGP, Manus, and Nauru, respectively. The cirrus cloud occurrence frequency is also plotted as a function of season in Fig. 1 for the three sites. The cloud occurrence frequency at SGP demonstrates a more obvious seasonal variation than in the tropics. This figure also indicates that the cirrus occurrence is greatest at TWPC1 and least at TWPC2, as expected from their geographic placement relative to the western Pacific warm pool.

a. A general look at cloud properties at SGP and TWP

Histograms of cirrus cloud height, temperature, V_fm, D_mass, IWC, and the cloud-scale vertical velocity W_m observed at the three sites are shown in Fig. 2, along with radar reflectivity and Doppler velocity. The mean, median, mode, and standard deviation of these properties as well as other properties such as optical depth and cloud geometrical thickness are listed in Table 1 for all seasons. Given the different sensitivities of the MMCRs, the results at the SGP site are also calculated in terms of clouds with radar reflectivity larger than −40 dBZ, which is roughly the minimum signal at the TWP sites. The increased minimum reflectivity threshold at SGP excludes less than 10% of the total cirrus, which tend to be composed of smaller ice masses at relatively colder temperatures.

In general, the frequency distributions of IWC, particle size, and associated mass-weighted fall velocity in Fig. 2 are skewed. For example, the median and mode of the optical depth and ice water path (IWP) distributions are orders of magnitude smaller than the mean (Table 1). In the following seasonal and diurnal variability analyses, results are mainly shown in terms of mean rather than median or mode to avoid repetition, but the skewness of the distributions should be kept in mind. Figure 2 and Table 1 also show that, in general, the tropical cirrus clouds have slightly larger ice particles with greater mass, which correspond to larger V_fm.

The differences between the two TWP sites are not as large as the differences between the tropics and the midlatitudes. The temperature and height distributions illustrate that the cirrus clouds at Nauru are more peaked and have a relatively smaller variance than those above Manus, and the cirrus at both sites seem to have similar radar-observed top heights (note that the layer top observed by the radar is often negatively biased because of radar sensitivity, as found from analysis of lidar data in Comstock et al. 2002). The cirrus clouds at Manus have an enhanced probability distribution in the 8–10-km height range (see also Mace et al. 2006b). The enhanced probability of occurrence in the 8–10-km region at Manus may be attributed to a closer association with deep convection at that location and may also contribute to the enhanced probability of larger particle size, and IWP for cirrus observed at Manus.

For the cloud-scale vertical motions, the distribution of the vertical velocities at SGP is almost symmetrical between upward and downward motion with a +2 cm s⁻¹ residual of the mean, while the cloud-scale vertical velocities in cirrus at the TWP sites suggest stronger mean upward motion with a residual of about +10 cm s⁻¹. There are also other reports of cirrus cloud–related vertical velocities observed at other scales using aircraft data. Gultepe and Starr (1995) investigated the cloud-scale dynamical structure and turbulence in cirrus using aircraft observations from the First International Satellite Cloud Climatology (ISCCP) Regional Experiment (FIRE). They found that a spectrum of phenomena exist with small-scale convective cells in the cloud system with associated updrafts of 40–70 cm s⁻¹, gravity waves (2–9-km wavelength), quasi-2D waves (∼10–20-km scales) with up to 20 cm s⁻¹ amplitude in vertical velocity, and mesoscale waves (∼170-km wavelength) with associated vertical velocity amplitudes less than 3 cm s⁻¹. Intermediate-scale wave structures were also evident within cloud systems where the associated amplitude in vertical velocity was markedly greater, perhaps exceeding 50 cm s⁻¹. In Heymsfield (1977), using aircraft microphysical measurements and Doppler radar measurements, it was found that the vertical velocities typically range from 10 cm s⁻¹ in warm-frontal overrunning systems to in excess of 50 cm s⁻¹ in clouds associated with closed low or isolated convective cells. From wind profiler measurements over tropical islands, the mean vertical velocity in stratiform cloud regions above 9 km was observed to be in excess of 10 cm s⁻¹ (Balsley et al. 1988; Gage et al. 1991, Jagannadha Rao et al. 2003). Mace et al. (2006a) found that midlatitude cirrus cloud structure and properties are found to be correlated with the large-scale vertical velocity on synoptic scales. On the cloud scale, cloud properties are related to the vertical distribution of ascending and descending air and its temporal change. As indicated in Heymsfield (1975), a cirrus uncinus cloud layer can be defined from top to bottom as generating, growing, and evaporating regions with updraft velocities of 1 m s⁻¹ common in the generating regions.

The enhanced infrared absorption in high-altitude anvil and thin cirrus clouds results in positive heating rates as illustrated by Ackerman et al. (1988). Assuming that all heating is converted to vertical motions as in Ackerman et al. (1988), ascent rates for a radiatively active layer are given simply by W = H/(dθ/dz), where H is the heating rate. Assuming characteristic values of H (30 K day⁻¹) for thick cirrus, and dθ/dz = 3 K km⁻¹ (Garrett et al. 2005), we estimate W ∼ 10 cm s⁻¹. However, the residual vertical motions found by averaging the retrieved vertical motions are about a factor of 1.5–2 greater than what is likely due solely to upper-tropospheric heating rates in cirrus. We cannot discount the possibility that the W_m residuals are biased because of a systematic particle habit difference between the tropical cirrus and the midlatitude cirrus from which we derived the mass– and area–dimensional relationships. However, observations collected during the Tropical Warm Pool International Cloud Experiment (May et al. 2008) showed that cirrus that are more than several hours removed from deep convection are composed primarily of bullet rosettes—the predominant ice crystal habit observed during the March 2000 intensive observing period from which the mass– and area–dimensional relationships were derived. Therefore, while a bias is possible, it is likely that a significantly positive residual in W_m in the tropical cirrus actually exists. According to the radiation modeling study in a recent paper (Corti et al. 2006), the lofting of air by cirrus heating provides a link between deep convection, the first stage of troposphere–stratosphere transport, and the third stage, the Brewer–Dobson circulation. Our results provide preliminary observational evidence of cloud heating–induced lofting in the tropics.

b. Interannual and seasonal variation of cirrus clouds in the midlatitudes and tropics

Cirrus properties vary as the dominant weather patterns shift because of seasonal or interannual changes in large-scale meteorological conditions. The seasonal variation of cirrus cloud properties has a direct bearing on climate sensitivity and must ultimately be simulated properly by numerical models if climate prediction is to be considered realistic (Sanderson et al. 2008). The interannual variations are illustrated well for our particular geographic locations (SGP and Nauru) in Fig. 3 with the monthly mean and standard deviation (vertical line) of cirrus macrophysical, microphysical, and dynamical properties starting from January of 1999. Figure 4 shows the composite annual variation of cloud properties at the SGP and Nauru sites. The mean and standard deviation of cirrus properties in the four seasons are also summarized in Table 1.

For midlatitude cirrus, there are significant seasonal variations (Fig. 3a) with more frequent, thicker clouds occurring during the summer than the winter. The seasonal variations of IWP, optical depth, and particle size basically follow a similar pattern with maxima and minima during the summer and winter, respectively. Similar seasonal variations of cirrus properties in the midlatitudes are reported in several papers. In Sassen and Campbell (2001), the cirrus macrophysical properties in terms of their seasonal and monthly means and variations are described from a 10-yr dataset collected from a University of Utah ground-based facility. Their study shows that midlatitude cirrus cloud-top heights typically correspond to the midlatitude tropopause during nonconvective periods. Using 6 yr of continuous vertically pointing cloud radar data collected at SGP, Mace et al. (2006a) found that the midlatitude cirrus in the warm seasons have relatively larger IWC and particle effective size than those in the cold seasons. However, the seasonal differences in IWC were not found by Wang and Sassen (2002), who used lidar–radar observations. This discrepancy is likely related to different instrumentation because the lidar signals can be attenuated during thick cirrus events while the radar penetrates optically dense cirrus layers and conversely the lidar is able to sense tenuous cirrus that the radar does not.

In Fig. 3b, the monthly-averaged cirrus cloud properties at Nauru seem to exhibit interannual variability with the mean cloud properties increasing on average from 1999 to 2003 and then decreasing thereafter. ENSO is the most important coupled ocean–atmospheric phenomenon that causes climate variability on interannual time scales in the equatorial western Pacific. Kent et al. (1995) and Fu et al. (1996) indicate that correlations exist between cirrus cloud cover and the Southern Oscillation index over the tropics. During a warm event with convection displaced eastward into the central equatorial Pacific, cloud cover decreases over the western Pacific, caused by generally subsident air over this region.

A multivariate ENSO index (MEI; Wolter and Timlin 1993) has been developed mainly for research purposes and is calculated using six main observed variables (sea level pressure, zonal and meridional component of surface wind, sea surface temperature, surface air temperature, and total cloudiness fraction of the sky). Negative values of the MEI represent the cold ENSO phase La Niña, and positive MEI values represent the warm ENSO phase El Niño. The MEI value from 1999 to 2005 is shown in Fig. 5b as cited from the National Oceanic and Atmospheric Administration (NOAA)/Earth System Research Laboratory Web site. The MEI tended to increase gradually from the moderate La Niña that peaked in late 1999. This gradual increase culminated in a weak El Niño event in early 2003. The MEI decreased but remained in the positive range thereafter.

To investigate further the large-scale dynamics near Nauru, we compute the monthly-averaged Hovmöller brightness temperature anomalies averaged between 10°S and 10°N latitude from 130°E to 180° longitude using Geostationary Meteorological Satellite and Geostationary Operational Environmental Satellite data. The result is shown in Fig. 5a. We find that the cold anomaly generally moves eastward from 1999 and approaches Nauru in 2002 as the MEI increases, and it moves westward thereafter as the MEI decreases. The cloud monthly variations in Fig. 3b show that the cirrus thickness gradually increases, consistent with an increase in optical depth, effective particle size, and IWP. These values peak in mid-2002 through early 2003, following the peak of El Niño conditions. However, as the warm phase waned in later 2003, we find that the cloud properties’ trends reverse, with decreasing optical depths, water paths, and smaller particle sizes. The correlation between monthly cloud property tendencies, such as Re and cloud-layer thickness, and the ENSO index is more than 0.7.

c. Diurnal variation of cirrus clouds in the tropics

The cloud properties and vertical motions are averaged by local hour to identify the diurnal cycles of various cloud properties. The hourly means and standard deviations are calculated from all samples in that hour from the 6-yr dataset to form a composite day.

The results at Manus and Nauru are shown in Fig. 6 as a function of local time. The macrophysical properties (cloud height, temperature, and cloud physical thickness) increase almost monotonically from midnight [0000 local time (LT)] to 1200 LT at both sites, followed by a decline thereafter. Only the cirrus observed at Manus show statistically significant variations in cloud microphysical properties, which increase in the morning from 0600 LT and peak in the early afternoon, then decrease thereafter. However the variations at Nauru are statistically insignificant because of the large standard deviation and subtle variations in the hourly averages.

The diurnal variation of in-cloud air motion at Manus shows a pattern that is similar to that at Nauru but with larger amplitude. The mean in-cloud vertical motion decreases from a maximum of +20 (Manus) and 15 (Nauru) cm s⁻¹ around noon to an average of 5 (Manus) and 10 (Nauru) cm s⁻¹ at local midnight.

d. Diurnal variation of midlatitude cirrus clouds in the summer and winter

A composite diurnal cycle of cirrus properties and in-cloud air motions of midlatitude cirrus compiled from all months is also analyzed as a function of local time, and there are no noticeable diurnal variations in the annual average. However, with such a significant transition in meteorological regime between winter and summer at SGP we do expect a strong sensitivity to season (Fig. 7). Although the diurnal cycle in W_m is small at SGP during both seasons, there is a seasonally dependent residual in W_m of approximately ±5 cm s⁻¹. In the absence of significant heating during the winter when synoptic-scale motions are the primary dynamical forcing, we would expect a near-zero residual in W_m. The ±5 cm s⁻¹ residual found here might suggest a systematic bias attributable to the assumed microphysical habits.

Relative to the winter, the summer cirrus at SGP show obvious and significant diurnal variations in most cloud properties. The clouds are observed to be thicker and higher in the later afternoon (1600–2200 LT), whereas the in-cloud air motion has slightly stronger updrafts before that late-day peak in the thickness. The comparison between Figs. 7a and 7b further distinguishes the difference between the summer and winter seasons. During the winter when synoptic-scale weather systems are the predominant source of dynamical forcing, diurnal variations are totally absent. Summer cirrus tend to be higher in altitude and relatively thicker geometrically and optically. Most of all, the in-cloud air motion tends to be upward (5 cm s⁻¹) during the summer. For both the tropics and midlatitudes, the diurnal cycles of cloud height and thickness are in phase with each other whereas the vertical motion and particle size lead the thickness and condensed mass cycles by several hours. It seems plausible that ascent resulting from radiative heating is contributing to the formation and/or maintenance of the cirrus that generates larger particles followed by sedimentation and sublimation in the cirrus as the layers grow, although this speculation must be confirmed using cloud-process models and further analysis of the data.