Ground-based active and passive remote sensing instrumentation are combined to derive radiative and macrophysical properties of tropical cirrus clouds. Eight months of cirrus observations at the Department of Energy Atmospheric Radiation Measurement site located on Nauru Island provide independent retrieval of cloud height and visible optical depth using lidar and radar techniques. Comparisons reveal the millimeter cloud radar does not detect 13% of cirrus clouds with a cloud base higher than 15 km that are detected by the lidar. Lidar and radar cloud heights demonstrate good agreement when the cloud lies below 15 km. Radar and lidar retrievals of visible optical depth also compare well for all but the optically thinnest clouds. Cloud occurrence at Nauru as measured by lidar reveal clear-sky conditions, low clouds, and high clouds occur on average 40%, 16%, and 44% of the time, respectively. Analysis of observed cirrus macrophysical and radiative properties suggests that two different types of cirrus exist in the tropical western Pacific: high, thin, laminar cirrus with cloud base higher than 15 km, and lower, physically thicker, more structured cirrus clouds. Differences in cirrus types are probably linked to their formation mechanisms. Radiosonde profiles of temperature and equivalent potential temperature near the tropical tropopause show a clear transition between neutrally stable and stable air at ∼15 km, which may also explain the presence of two distinct cirrus types. Radiative heating rate and cloud forcing calculations for specific cirrus cases reveal the impact of tropical cirrus clouds on the Earth's radiation budget.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Tropical cirrus clouds, located high in the upper troposphere, may have a significant impact on the Earth-atmosphere system. These high, cold ice clouds can significantly influence radiative cloud forcing [McFarquhar et al., 2000], may help to sustain the warm pool [Prabhakara et al., 1993], and influence stratospheric-tropospheric water vapor exchange [Rosenfield et al., 1998; Sherwood 1999]. Increases in water vapor entering the stratosphere has been linked to increases in stratospheric ozone destruction [Kirk-Davidoff et al., 1999]. Characterizing the location, microphysical, and radiative properties of tropical cirrus will significantly improve our understanding of how these high, often tenuous ice clouds form, persist, and influence the radiative energy budget and dynamics in the upper troposphere.
 Numerous studies have demonstrated the effectiveness of different measurement platforms for studying tropical cirrus clouds. Space-based observations have shown that cirrus cloud systems can extend horizontally up to 1000 km [Winker and Trepte, 1998] and persist for days in tropical regions where convective activity is frequent. The seasonal distribution of thin cirrus clouds in equatorial regions was compiled by Prabhakara et al.  using Infrared Interferometer Spectrometer (IRIS) measurements. They determined that thin cirrus frequency was higher over convectively active regions, including the Intertropical Convergence Zone (ITCZ), the Southern Pacific Convergence Zone (SPCZ), and the Bay of Bengal, and attributed differences in extinction measured by the 10.8 μm and 12.6 μm channels to the presence of small ice crystals in cirrus clouds with a particle size between 1 and 4 μm. A 6-year cloud climatology derived from satellite solar occultation measurements report subvisual cirrus clouds persist over equatorial regions on average 45% of the time at ∼15 km, with a maximum frequency of occurrence of ∼70% at 15.5 km located over Micronesia [Wang et al., 1996]. While satellite retrievals of thin cirrus have improved, there are still some uncertainties when deriving cirrus microphysical properties using passive satellite observations. For example, the occultation technique used during the Stratospheric Aerosol and Gas Experiment (SAGE) II [Wang et al., 1994], has a minimum and maximum optical depth threshold of 2 × 10−6 and 2 × 10−2, respectively. This type of retrieval detects only the thinnest cirrus clouds and has difficulty evaluating the extent that thin cirrus overlap dense clouds.
 Limited measurements of cirrus microphysical properties in the tropics make it difficult to assess accurately the radiative impact of these cold, optically thin clouds. Aircraft in situ observations from three cirrus cloud systems [Heymsfield and Jahnsen, 1974; Booker and Stickel, 1982] summarized by McFarquhar et al.  indicate that subvisual cirrus located near the tropical tropopause with a visible optical depth τ < 0.007 consist of small particles (as small as 3 μm in length), have a low ice water content (IWC ∼ 10−4 g m−3), and can have a radiative heating rate of up to 2 K d−1. Intensive field campaigns, such as the Central Equatorial Pacific Experiment (CEPEX) provide in situ measurements of ice crystal size, concentration, and habit of anvil cirrus. Results demonstrate that cirrus associated with deep convection can have large numbers of small particles, which contribute more to cloud mass and cross-sectional area as height increases and IWC decreases [McFarquhar and Heymsfield, 1996]. Ice particle habits can range from aggregates for clouds with larger values of IWC, to quasi-circular and hexagonal plate-like shapes for small ice crystals. These differences might be related to the distance of the anvil from the convective core. Due to the influence of cloud microphysical properties, particularly IWC, on radiative heating rates, it is important to create a statistically significant database of cirrus cloud properties. Aircraft in situ measurements provide useful insight into the small-scale structure of cirrus microphysical properties; however, there is still a considerable amount of information needed to develop a statistically significant database of cloud properties to describe the unique characteristics of tropical ice clouds.
 Space and airborne lidar measurements have been useful for measuring the location, occurrence, and origin of tropical clouds. Cirrus associated with deep convection measured during the Tropical Ocean and Global Atmosphere/Coupled Ocean-Atmosphere Regional Experiment (TOGA/COARE) using airborne polarization lidar have a high frequency of occurrence of cloud top heights between 16.5 and 17.0 km, which is closely associated with the average tropopause height at ∼16.85 km [Sassen et al., 2000]. Observations from the Lidar In-space Technology Experiment (LITE) [Winker and Trepte, 1998] indicate that cirrus in the tropical western Pacific (TWP) can exist both detached from and closely associated with deep convective clouds. LITE observed thin (<1 km) cirrus layers near the tropopause in the region around the equator between 35°N and 20°S during the 53 hours of data collected over a 10-day period. LITE observations also reveal the presence of two distinct cirrus types: laminar cirrus occurring near the tropopause, and denser cirrus that tend to be deeper and contain more structure. Although LITE provided global lidar measurements of upper tropospheric clouds, the brief time period of the mission does not provide a long-term climatology of cloud properties or frequency of occurrence. This is also the case with ground-based field campaigns, which often last for only 1 month.
 Studies that use ground-based remote sensing instruments to measure tropical cirrus are fairly limited [Platt et al., 1987, 1998; Mather et al., 1998b]. Recently, the DOE ARM program has installed two remote sites in the TWP region on Manus and Nauru Islands [Mather et al., 1998a], which contain a number of passive and active remote sensing instruments. Due to a less frequent occurrence of low clouds and greater data availability, observations at Nauru Island (0.521 °S, 166.916 °E) were chosen to conduct this study of cirrus radiative properties over an extended time period using lidar, radar, and radiometric instruments. This 8-month subset of ground-based remote sensing observations at Nauru provide a unique data set describing the vertical location and radiative properties of tropical cirrus. This data set also provides a basis from which we can infer cloud type, such as the fraction of time when single and multiple-layer cirrus clouds, or low clouds are present. Since satellite retrievals of cirrus cloud properties often work best when performed over oceans with no underlying low clouds, measurements at Nauru (which is a small island with a land area of only ∼21 km2) will also provide a useful data set for comparison.
 The primary goals of this study are (1) to characterize the vertical location and optical depth of tropical cirrus (section 3) and compare results derived from Micropulse Lidar (MPL) data with those obtained using the radar-radiometer algorithm of Mace et al. . Once we identify the strengths and limitations of ground-based remote sensing retrievals, (2) we characterize tropical cirrus in terms of frequency of cloud occurrence and cloud type using height and cloud layers as a guide (section 4). We also relate measurements of cirrus clouds to their formation and persistence mechanisms. Finally, (3) we examine radiosonde thermodynamic profiles (section 5) and calculate radiative heating rates of “typical” cirrus using retrieved optical depths (section 6) to determine the radiative impact of tropical cirrus on the upper troposphere.
2. Instrumentation and Retrieval Techniques
2.1. Lidar-Radiometer (LIRAD) Retrieval
 We derive visible optical depth and cloud height using MPL backscattered energy [Spinhirne, 1993]. The MPL emits maximum pulse of 10–15 μJ at a wavelength of 532 nm with a pulse repetition frequency (PRF) of 2500 Hz. The vertical resolution is 30 m and the receiver field-of-view (FOV) is 100 μrad. Profiles of backscatter cross-section derived from MPL measurements, which are corrected for overlap, deadtime, and afterpulse errors, enable the detection of cloud base and top height for clouds with τ ∼ 2.0. Due to the scattering of light, lidar signals tend to become fully attenuated in optically thick clouds, which prohibits the penetration of the signal through the entire depth of the cloud, resulting in an underestimation of cloud top height as well as visible optical depth. A detailed description of MPL instrument characteristics and data processing techniques is reported by Campbell et al. .
 A combination of the visible wavelength and low output energy can cause some difficulties in the analysis of micropulse lidar backscattered energy profiles. Particularly during daytime hours, the signal-to-noise ratio is low, which makes it difficult to discern weak cloud signals from background noise caused by ambient light. Low daytime signal-to-noise ratios may slightly bias our results to nighttime hours and increase the uncertainty in optical depth retrievals near solar noon. This possible bias is discussed further in section 4.
 We use the cloud mask algorithm of Clothiaux et al.  to identify cloud boundaries from lidar backscatter profiles, and the algorithm of Comstock and Sassen , which is based on the LIRAD method described by Platt and Dilley , to calculate visible optical depth at 532 nm. Visible optical depth is calculated by integrating values of the lidar cloud backscatter coefficient βc (km−1 sr−1) between cloud base zb and top zt. The backscatter-to-extinction ratio k is estimated independently for each profile using the average backscatter coefficient above the cloud as described by Comstock and Sassen . After removing Rayleigh scattering contributions from the total backscatter coefficient, we vary k until the above signal becomes negligible. The primary uncertainty in the optical depth retrieval is in estimating the parameter k, which results in a maximum uncertainty of ∼24% for the retrieved value of τ [Comstock and Sassen, 2001].
 The effect of forward multiple scattering on the returned energy also contributes to uncertainty in optical depth. Since the MPL receiving telescope has a narrow FOV, the contribution of forward multiple scattering to the total cloud optical depth is assumed to be relatively small. According to simulations of the multiple scattering correction factor η [Comstock and Sassen, 2001], as FOV decreases, η increases to between 0.9 and 1.0 for particles with an effective radius re of 20–200 μm. In this study, we set η = 0.9 to compensate for these uncertainties. Although a fraction of the cirrus observed in this study probably consist of small ice crystals that do not contribute significantly to multiple scattering (see discussion in section 3.1), we assume η = 0.9 to correct for the presence of larger particles that probably exist in most of the clouds observed at Nauru.
 The lidar-radiometer retrieval method also combines lidar backscatter profiles with downwelling radiance measurements from the AERI to obtain cloud infrared emittance ε. The AERI measures spectral downwelling radiance with a FOV of 1.3° between 500 and 3300 (cm−1) and has a spectral resolution of 0.5 cm−1 [Smith et al., 1993]. Due to large water vapor amounts in the tropical atmosphere, the radiance emitted by optically thin clouds is often indistinguishable above the infrared radiance emitted by the intervening atmosphere, making it difficult to accurately estimate cloud emittance. Therefore, we will present only the optical depth retrievals, which primarily rely on the lidar backscatter profiles.
2.2. Radar-AERI Retrieval
 To provide a comparison to lidar derived τ and cloud height, we also derive values using the millimeter radar and infrared interferometer algorithm of Mace et al. . This algorithm derives re and ice water path (IWP) from radar reflectivity measurements and infrared emission spectra measured by the AERI between 800 and 1000 cm−1 averaged over 5 cm−1 intervals. The operational millimeter cloud radar (MMCR) located on Nauru has a transmitting frequency of 35 GHz, vertical range (R) resolution of 90 m (up to an altitude of 20.355 km), and a sensitivity of −48 dBZe at 5 km. Due to the 1/R2 dependency of returned power, the sensitivity of the MMCR is −42 and −38 dBz at 10 and 15 km, respectively. In addition to range limitations, the backscattering cross-section of volumes composed primarily of small ice crystals, at microwave frequencies, are often well below the sensitivity of the MMCR. Atlas et al.  discuss the difficulties in using radar frequencies for ice cloud studies, and determine that radar will have difficulty detecting radiatively significant clouds that have reflectivities less than −30 dBZ.
 In the radar-AERI algorithm, we use an iterative approach to identify a first order gamma distribution of equivalent volume ice spheres that simultaneously give the observed layer-mean radar reflectivity and infrared optical properties inferred from AERI radiances. An initial guess of cloud infrared emittance is inserted in an equation relating ε, cloud thickness Δz, radar reflectivity, and modal diameter to IWP and particle size. We apply the radiative parameterizations of Fu and Liou  and the MODTRAN radiative transfer algorithm [Berk et al., 1989] to estimate downwelling radiance, which is compared to the measured AERI radiance for specific spectral bands. This process is repeated by adjusting ε until the measured and modeled radiances agree. Visible optical depth (0.55 μm) is calculated by inserting the retrieved re and IWP into Fu and Liou  parameterizations. Details concerning this retrieval algorithm are explained by Mace et al. , and uncertainties in τ are on the order of 20%.
3. Comparison of Radiative and Macrophysical Properties of Tropical Cirrus
3.1. Cloud Base and Top Height
 Due to differences in scattering properties for visible and microwave wavelengths and sensitivity issues with the MMCR, we expect some differences between cloud base and top heights measured by lidar and radar. Figure 1 illustrates differences in cloud base and top heights detected using radar and lidar by comparing time versus height profiles of normalized lidar backscatter and radar reflectivity. The ability of the MPL to detect ice crystals in the thin cirrus layer between 0100 and 0700 UTC is clearly demonstrated in Figure 1a. However, it is apparent that the low signal-to-noise ratio during the daytime (solar noon occurs at ∼0036 UTC) inhibits our ability to discern the tenuous cirrus layer from the ambient background light. It is also clear in the plot of radar reflectivity (Figure 1b) that the thin cirrus detected by the MPL goes completely undetected by the MMCR. A similar analysis comparing CO2 lidar and Ka band radar measurements by Intrieri et al.  also found that the radar has difficulty detecting ice crystals in thin cirrus clouds. As mentioned in section 2.2, a combination of decreased sensitivity with range and the presence of small particles can prevent the radar from detecting some cirrus clouds. After our initial analysis, the decrease in sensitivity of the MMCR with range appeared to be the predominant explanation since radar reflectivity measurements also did not detect deep convective clouds extending beyond ∼15 km at Nauru. However, after inspecting radar reflectivities from an identical MMCR located on Manus Island (2.058°S, 147.425°E), it is clear that the radar can detect the presence of clouds up to 16–17 km. Therefore, the cirrus clouds above 15 km that are not detected by the MMCR at Nauru probably contain small ice crystals. It is also clear that there is a significant difference in the maximum height of deep convection when comparing MMCR data from the two ARM tropical western Pacific sites during this period.
 A statistical comparison of all high cloud (zb > 7 km) detections at Nauru between April and November 1999 (Figure 2) shows clearly that the lidar detects significantly higher cloud tops. The MMCR misses 45% of cloud top heights above 15 km that are observed by the lidar (Figure 2a). In addition, the MMCR misses all cirrus with zb > 15 km, which is 13% of the cirrus data set detected by the MPL (Figure 2b). Care was taken to remove altocumulus clouds from the high cloud data set, which are identified as thin layers in the middle troposphere (usually ∼200 m thick) in the time versus height displays of lidar backscattered energy. Altocumulus above 7 km contributes less than 0.5% to the entire data set. Cloud base and top height detected by lidar and radar measurements for coincident time periods (Figures 2c and 2d) display good agreement. The offset between zb frequency distributions is only 500 m, which is within the measurement uncertainties of the two instruments. In contrast, cloud top heights that are underestimated by the MPL due to complete attenuation of the lidar beam are unapparent in Figure 2, but are certain to occur in the data set.
3.2. Visible Optical Depth
 The frequency distributions of τ between 0 and 2.0 are quite similar for both radar and lidar retrievals (Figure 3). However, the lidar frequency distribution of τ between 0.0 and 0.5 at a bin resolution of 0.01 for all cirrus observations (Figure 3a) is significantly larger than the radar distribution for τ < 0.08. When lidar observations above 15 km are removed (Figure 3b), the frequency distribution of lidar derived τ < 0.04 decreases and the agreement between lidar and radar retrievals improves. The small discrepancy for τ < 0.04 has two probable sources. First, the uncertainty in lidar optical depths increases for small τ because the calculation of cloud backscatter becomes more sensitive to the value of backscatter-to-extinction ratio. Second, the radar retrieval relies on the measured infrared radiance to obtain cloud emittance, and uses ε to obtain τ. Since emission due to thin clouds is barely detectable above the backgroundwater vapor emission, the retrieval at low optical depths is more sensitive to uncertainties in water vapor path retrievals and spectroscopic errors in MODTRAN. In addition, the inability of lidar signals to penetrate through optically thick cirrus does not significantly affect the frequency of occurrence of large τ because radar results reveal only a small percentage of cirrus have τ > 0.5.
 In order to assess the effect of low daytime signal-to-noise ratios on lidar retrievals of optical depth, we divide the results into four groups according to time of day (Figure 4). The overall shapes of the τ distributions are similar for each case, except for the time period centered around solar noon (Figure 4a), which exhibits a slightly lower percentage in the smallest τ bin (0.0 to 0.1). The frequency of occurrence of low (zb < 7 km), high (zb > 7 km), and no clouds for each time period (Figure 5) indicates that there are only small changes in low cloud frequency at different hours of the day; however, there is a decrease in high cloudiness during the hours around solar noon. Therefore, the reduction in small τ observed in Figure 4a is probably due to the inability of the cloud detection algorithm to discern optically thin clouds from the background noise during periods when the solar zenith angle is small, underestimating the amount of subvisual clouds during this time period. In addition, each day no MPL data is recorded for a period of an hour or less around solar noon. During this time, a cover is put over the lidar receiver to keep direct sunlight from damaging the system. We estimate that the frequency of occurrence of cirrus is low by ∼11% during the solar noon time period due in part to the low signal-to-noise ratio in the daytime, but also due to the lack of observations during the time period when the lidar receiver is covered.
4. Cloud Occurrence and Type
 The widespread nature of cirrus clouds in the tropics has been identified in several satellite observational studies [Liou, 1986; Wang et al., 1996; Prabhakara et al., 1993; Winker and Trepte, 1998]. Here, we use Nauru MPL measurements to estimate frequency of cloudiness in this region of the tropics. Given cloud height, temperature, and optical depth, we can also examine the radiative impact of these clouds.
4.1. Cloud Occurrence
 We processed 5664 hours of MPL data out of a total possible 5856 hours of observations during the 8-month period between April and November 1999. 44% of the observations are high clouds with zb > 7 km and occur with no underlying low, optically thick clouds. Again, altocumulus clouds are removed from the high cloud data set to help isolate cirrus cases from middle and low-level clouds. There are no clouds overhead during 40% of the observations, while low clouds (zb < 7 km plus altocumulus removed from the high cloud data set) occur 16% of the time. By isolating cases where high clouds exist before and after the occurrence of low cloud blockage, we estimate that 69% of low clouds obstruct higher clouds, which equates to 11% of the total time period.
 Monthly frequency distributions of cloud occurrence (Figure 6) clearly exhibit the prevalence of cirrus near Nauru. There is a noticable decrease in high cloud occurrence and an increase in clear-sky occurrence during September in comparison with other months. Several satellite studies [Mapes and Houze, 1993; Chen et al., 1996] have linked the movement of deep convection with the 30–60 day interseasonal oscillation known as the Madden-Julian Oscillation (MJO) [Madden and Julian, 1971]. Ground-based radiometer observations at Manus Island in the tropical Western Pacific also reveal the characteristic periodicity associated with the MJO [Mather et al., 1998b]. The decrease in high clouds observed in this study correlates with a period of reduced convection in the western Pacific, which may be linked to a suppressed phase of the MJO. This implies that cirrus frequency of occurrence is linked with large-scale variability in the atmosphere. However, until a longer times series of lidar observations become available, any apparent correlation between the MJO signature and ground-based observations at Nauru is suspect.
4.2. Cirrus Cloud Types
 Several mechanisms have been identified as possible contributors to the formation and persistence of tropical cirrus. The two most general formation mechanisms are 1) cirrus due to anvil outflows from cumulonimbus clouds, and 2) cirrus formation as a result of synoptic scale uplift [Jensen et al., 1996]. Ackerman et al.  examined radiative heating rates of tropical anvils and suggested that radiative destabilization, caused by a decrease in the heating rate with height, could induce upward motion, which would then lead to ice formation that helped maintain the cloud layer. However, the modeling study of Boehm et al.  indicated that cloud internal dynamics have little impact on cloud maintenance. More recently, specific mechanisms have been linked to cirrus formation. For example, Boehm and Verlinde  link tropopause “trap” cirrus identified using MPL observations from Nauru with large-scale variability near the tropopause and downward phase propagation of Kelvin waves from the lower stratosphere. They also link cirrus occurrence to cold temperature perturbations in the upper troposphere. In a recent study, Hartmann et al.  hypothesize that subvisual cirrus located above optically thick clouds persist due to radiative cooling in the upper layer that is caused by the underlying deep convective cloud. Their simulations predict that the distance between the two layers must be less than 3 km for cooling to occur in the upper layer.
 In order to address the mechanisms described above with actual observations of cirrus clouds, we divide the MPL measurements from Nauru into categories based on height and number of cloud layers. Categories include single-layer cirrus with zb > 15 km, 10 < zb < 15 km, and zb < 10 km, and multiple-layered cirrus clouds with varying distances between layers. Approximately 14% of cirrus observations consist of a single-layer and have zb > 15 km, only 4% have zb < 10 km, and ∼34% have 10 < zb < 15 km. Clouds with zb > 15 km correspond to the high laminar cirrus studied by Winker and Trepte , who observed this cirrus type ∼7% of the time with a cloud fraction of 14% in the equatorial region. Approximately 52% of cirrus clouds with no underlying low clouds (zb < 7 km) occur as single layers, while 48% occur in multiple layers. Table 1 summarizes the frequency of occurrence for single and multiple-layered cirrus, as well as average and median τ for each cloud type. Percentages do not take into account any high cirrus that may exist over low clouds.
Table 1. Frequency of Cirrus Cloud Occurrence Divided Into Categories Based on Cloud Base Height and Distance Between Layers ha
Also listed are average and median τ and cloud thickness Δz for each cloud type displayed in Figure 7. Average τ for multiple-layer cirrus are total values for all layers.
Single layer, zb > 15 km
Single layer, 10 < zb < 15 km
Single layer, zb < 10 km
Multiple layers, h < 0.5 km
Multiple layers, 0.5 < h < 3.0 km
Multiple layers, h > 3.0 km
 Frequency distributions of visible optical depth (Figure 7) for the various categories clearly show that cirrus clouds with zb > 15 km have different characteristics than lower cirrus. The highest observed cirrus are virtually always thinner than 1 km (Figure 8) and have a small τ generally less than 0.1 (Table 1). Single-layer cirrus with zb > 15 km have a much higher percentage of small optical depths (92% with τ < 0.1 in Figure 7a) and the lowest average τ than any other cloud type. Middle layer cirrus (Figure 7b) also display a modal value in the smallest τ bin but have a larger average τ and cloud thickness (Table 1) than the highest cirrus. Single-layer cirrus with zb < 10 km have the largest average τ and the frequency distribution shifts toward τ > 0.5. The average τ for this category may be an underestimate due to the inclusion of cases where the lidar signal is fully attenuated, which can cause an underestimation of τ if there is a substantial amount of the cloud mass above the range where complete attenuation occurs. We think this is the cause of the sharp decrease at τ ∼ 1.0 displayed in Figure 7c. Multiple-layer cirrus distributions of total column cloud optical depth (Figures 7d, 7e, and 7f) do not vary substantially between groups. However, it is noteworthy that the frequency distributions for multiple-layer cirrus are similar to single-layer cirrus with 10 < zb < 15 km (Figure 7b). Although multiple-layer cirrus may include subvisual layers above 15 km, it is only when the entire cloud lies above 15 km that the total column τ is constrained to small values.
 This tropical cirrus cloud data set suggests that two distinct high cirrus types are present over Nauru. The first type is high, physically and optically thin, and appears to be laminar in appearance. The second type is also optically thin; however, it tends to have a larger depth and more structure within the layer. Two distinct cirrus types were also observed by Pfister et al.  during the Tropical Ozone Transport Experiment/Vortex Ozone Transport Experiment (TOTE/VOTE) located in the equatorial region near Hawaii. During several aircraft flights, high, thin cirrus clouds were observed to have either a laminar or more structured appearance using a UV Differential Absorption Lidar. Each of these cloud types were located above 15 km and the laminar cirrus were sometimes found to be tilted upward and poleward.
Pfister et al.  also trace the origin of each cirrus case using trajectory analysis, which showed thicker, structured cirrus were often associated with significant convection that occurred up to 10 days before the cirrus observation. Thin laminar cirrus were attributed to in situ formation; however, presence of water vapor in the upper troposphere is still considered to be convective in origin.
5. Atmospheric Thermodynamic Profiles
 Temporal variations in thermodynamic variables near the tropopause can give some indication of large-scale variability in the upper troposphere and its relationship to cirrus formation and persistence. As mentioned in the previous section, cirrus formation has been linked to temperature perturbations near the tropopause [Boehm and Verlinde, 2000]. Nauru radiosonde temperature profiles (Figure 9a) near the tropopause region reveal noticable temporal variations throughout 1999. In midlatitude regions, the height of the minimum temperature typically defines the position of the tropopause. The tropical tropopause is not as easily defined due to a nearly stable region just below the stratosphere that can extend for several kilometers. The region where the temperature profile is nearly constant and the equivalent potential temperature θe begins to increase, delineates the transition between the troposphere and stratosphere. At Nauru, the tropopause transition layer tends to reside between ∼15 and 18 km (Figure 10). The magnitude of the temperature minimum between 15 and 18 km varies between 183 and 198 K with day to day variations of up to 4 K. Temperature perturbations with periods on the order of a week are linked to cirrus occurrence by Boehm and Verlinde . The annual cycle of temperature near the tropopause is also evident in Figure 9a.
 Equivalent potential temperature (Figure 9b) is also a good indicator of large-scale variability near the tropopause, as well as convective tendency in the lower troposphere. There is a clear transition in both temperature and θe around 15 km that separates the region of near neutral stability below 15 km, which is largely influenced by convection, and the nearly stable region above 15 km (Figures 9b and 10). The cirrus above 15 km are therefore laminar in appearance because the near stable conditions inhibit vertical motions. This height also corresponds to the transition height between laminar (zb > 15 km) and lower, thicker, structured cirrus (where zb < 15 km).
 Anomalies in temperature and θe (Figures 9c and 9d), calculated using a 30-day running average, suggest that large-scale dynamics can influence the thermodynamic properties of the transition region between the upper troposphere and lower stratosphere. Boehm and Verlinde  attribute the clear pattern in temperature anomalies to the downward phase propagation of Kelvin waves. These perturbations appear to extend to roughly 15 km, which implies that the laminar cirrus observed above 15 km exist in a region that is influenced by large-scale dynamics, whereas cirrus below 15 km exist in a regime dominated by moist convection.
 Since the Boehm and Verlinde  study examined radiosonde data only during the Nauru99 Intensive Observing Period, which occurred from 16 June to 17 July 1999, we extend the analysis to April–November 1999. The method we use to correlate cloud occurrence with temperature anomalies is also slightly different. First, we classify each lidar backscatter profile according to cloud height (Table 2), and only consider cases when the lidar detects a cloud base >10 km. Since there are several ways to define cloud occurrence above 15 km, we have used this classification to demonstrate the different correlation for each case. Second, we calculate temperature anomalies at each level (200 m vertical resolution) by subtracting the radiosonde temperature from the 30-day running mean (as in Figure 9c). Each lidar profile is then assigned the temperature anomaly that corresponds to the height of the highest cloud top. Third, we divide each day into 6-hour intervals and calculate the fraction of the time that each cloud class (Table 2) occurs during each interval. The average temperature anomaly is also calculated during each 6-hour interval. Since temperature profiles are available twice per day, we believe that using 6-hour intervals removes any bias that may occur by correlating each lidar profile with the same temperature anomaly. Therefore, we count cirrus occurrence according to each event rather than for each 2-min interval.
Table 2. Description of Cloud Classes Used to Correlate Temperature Anomalies With Cloud Occurrence in Section 5
At least one layer has zb > 15 km
At least some of the cloud is above 15 km (i.e., zb =12.2 km; zt = 15.3 km)
All layers have zb > 15 km with no cloud below 15 km
All layers are below 15 km (including cloud top of highest layer)
 Frequency of occurrence of temperature anomalies for cases when the 6-hour cloud fraction is at least 40% is compared with temperature anomalies for clear-sky cases (Figure 11). The clear-sky temperature anomaly is taken at the height of the 360 K θe surface, which is on average ∼15.4 km. High cirrus with zb > 15 km (classes 1 and 3) do not coincide with negative temperature anomalies as often as found by Boehm and Verlinde . For cases when at least some of the cloud is above 15 km (class 2) there is a high correlation with cold anomalies. It is also interesting that lower anvils with the entire cloud below 15 km (class 4) also have a strong correlation with cold anomalies; however, the anomalies are usually between 0 and −1 K. These statistics suggest that there is still some uncertainty in connecting cirrus occurrence with temperature variations near the tropical tropopause layer, which will be better addressed as a longer time series of data become available.
6. Tropical Cirrus Radiative Heating Rates
 In order to better understand the relationship of tropical cirrus clouds with upper tropospheric and lower stratospheric exchange, we describe several cirrus cases that resemble the different cloud types identified in section 4.2. Heating rate and cloud forcing calculations are performed for a high, thin, single-layer case, a multiple-layer case, and a case where a thin subvisual layer exists above a deep convective cloud to illustrate the radiative impact of tropical cirrus.
6.1. Radiative Heating Rate Calculations
 The correlated K distribution and δ-4 stream radiative transfer code of Fu and Liou [1992, 1993] calculates heating rates from atmospheric thermodynamic profiles, cloud particle size, and IWC. We estimate mean layer extinction coefficient σe = τ/Δz from lidar derived τ and cloud thickness averaged over a specific time period (usually 1 hour) with uniform cloud base and top heights and relatively constant τ. An average lidar backscatter profile provides a weighting function to distribute σe vertically throughout the cloud layer. Assuming a generalized effective size Dge = 10.0 μm, similar to that used by McFarquhar et al.  and Hartmann et al. , we calculate ice water content using the parameterization from Fu . Ackerman et al.  showed that radiative heating rates are much more sensitive to changes in IWC than to changes in ice crystal size. Our simulations predict only an 8% and 0.4% change in maximum net heating rate and outgoing longwave radiation (OLR), respectively, for an increase of Dge from 5 to 15 μm in a subvisual cirrus. Therefore, by assuming Dge, we introduce only small uncertainties into the heating rate calculations. Asymmetry parameter and single scattering albedo are 0.758 and 0.999998, respectively, for the 0.2–0.7 μm band and solar zenith angle θo = 45°.
6.2. High, Single-Layer Cirrus
 The first example is a high, single-layer cirrus case observed on 5 July 1999 (Figure 12). Between 1330 and 1430 UTC, the average τ is 0.022, which falls in the subvisual cirrus category defined as τ < 0.03 [Sassen and Cho, 1992]. A summary of cloud properties for all cirrus cases is listed in Table 3. Although the optical depth is small, radiative heating rates (Figure 12) reveal a maximum net heating rate of 3.1 K d−1 in the cloud layer, mainly due to infrared heating (∼3 K d−1). According to profiles of lidar backscattered energy, this particular cirrus cloud system persists over Nauru for nearly 16 hours. An enlarged view of 11 hours of lidar returned energy (Figure 12a) reveals the detailed structure of the cirrus cloud. Japanese Geostationary Meteorological Satellite (GMS-5) imagery (Figure 12e) indicates that this subvisual cirrus case is not directly associated with deep convection, and therefore is probably influenced by large-scale processes, such as Kelvin waves. Also apparent in the lidar image are a series of undulations with a period slightly less than 1 hour, which appear to be associated with small-scale wave activity in the upper troposphere and lower stratosphere.
Table 3. Average Cloud Properties Used in Heating Rate Simulationsa
IWC, g m−3
Maximum Net Heating Rate, K d−1
IWC represents the mean layer value. Maximum net heating rates for cirrus layers are also listed.
 This example indicates that substantial heating occurs when thin cirrus are present in the upper tropical troposphere. This finding is not necessarily a new one. McFarquhar et al.  also report heating rates for subvisual tropical cirrus on the order of 0.5–2.0 K d−1, which is slightly less than the cirrus displayed in Figure 12. However, τ is slightly higher in our case. There are several studies that attempt to explain the impact of this heating on the upper troposphere. For example, the modeling study of Rosenfield et al.  suggests that a small increase in radiative heating caused by thin tropical cirrus may result in increased temperatures (1–2 K), vertical velocity (0.02–0.04 m s−1) and water vapor amounts (1 ppmv) in the lower stratosphere. Since cirrus clouds exist in tropical regions over 50% of the time and, according to the Nauru data set, clouds with zb > 10 km typically have τ < 0.1 (Figures 3 and 7), it is reasonable to suggest that the increases in vertical velocity and stratospheric water vapor modeled by Rosenfield et al.  might be a prevailing effect of thin tropical cirrus on upper tropospheric dynamics. Since it is difficult to infer vertical lifting using single point lidar or radar measurements, we cannot provide direct evidence of this mechanism. Nauru MPL measurements often show a laminar cirrus layer persisting for many hours with no obvious change in cloud height. If a thin cirrus cloud is rising due to large-scale influence, it is possible that the cloud will cross sloped isentropic surfaces, but in space the cloud altitude remains constant as it passes over a fixed point, such as Nauru. In addition, if upper tropospheric air were not rising, we might expect to observe warming and evaporation of ice crystals in the layer where the cirrus originated. Thermodynamic profiles do not indicate a temperature increase over the lifetime of the cirrus, and MPL backscatter profiles show that cloud layers persist for long periods of time. Therefore, since layer heating does not appear to be contributing to significant ice evaporation, then heating can either result in lifting of saturated air into the stratosphere or away from the layer through horizontal divergence.
6.3. Multiple-Layer Cirrus
Figure 13 displays a multiple-layer cirrus cloud observed on 17 June 1999, which has a distance between layers of 0.5 to 3.0 km. According to GMS-5 imagery (Figure 13e), cirrus appear to be originating from the convective activity located to the southeast of Nauru. The optical depth of the upper layer is 0.076, while the lower layer is 0.065. Heating rate calculations (Figure 13) indicate stronger infrared heating in the upper layer; however solar heating results in significant net heating of ∼2 K d−1 in the lower layer. The height of the upper layer decreases ∼200 m between 0930 and 2359 UTC (Figure 13a). As mentioned previously, it is difficult to deduce from single point data whether the cloud layer is descending, or whether it is a sloped, rising layer that is advecting over the lidar. The undulations apparent in the laminar cirrus case (Figure 12a) are also visible in the upper layer from ∼1800 to 1930 UTC (Figure 13a) and, in this case, have a peak-to-peak wavelength of ∼25 min, which again provides evidence of stratospheric influence on the highest cirrus layers.
6.4. Subvisual Cirrus Over Deep Convective Cloud
 The final case (Figure 14) demonstrates the proposed radiative cooling mechanism of Hartmann et al. . In this case, we use a solar zenith angle of 54°. Radar reflectivity measurements (Figure 14b) indicate a large convective system passes over Nauru between ∼1630 and 1900 UTC on 19 July 1999 with the anvil extending overhead through 2200. Prior to ∼1400 UTC, there is a thin cirrus layer above 15 km visible in the lidar backscatter display (Figure 14a), which is also discernible following the convective event. Subsequent MPL backscatter images (not shown) indicate that significant cirrus layers continued to pass over Nauru for the next 3 days. In this case, we have pieced together lidar and radar measurements to simulate the radiative heating in the column when the thin cloud is present over the convective anvil.
 According to Hartmann et al. , radiative cooling should occur in the upper layer (rather than heating as in Figure 12) because it absorbs less upwelling longwave radiation when a convective anvil is present below. However, in this case net heating rate calculations (Figure 14e) show heating toward cloud top with only small amounts of cooling near cloud base.
 Results of heating rate simulations for each multiple-layer cloud case demonstrate that heating can occur in the cirrus layer even when a thick lower cloud is present. This is apparent in both Figures 13 and 14. In the multiple thin layers case (Figure 13) significant heating occurs in the upper layer despite the presence of lower clouds. If the underlying cloud is sufficiently thick, the magnitude of heating in the upper layer may decrease, but does not necessarily cause radiative cooling. Hartmann et al.  also suggest that after the cirrus layer cools and the anvil dissipates, radiative heating from below will deplete the moisture from the layer, causing the ice to evaporate. However, in the 19 July case, cirrus continue to persist for several days following the convective event. Therefore, at least in this case, some other competing mechanism may be limiting the influence of the radiative cooling mechanism. After inspection of lidar backscatter and radar reflectivity images during April through November 1999, only 14 cases were identified as possible scenarios where thin cirrus persist over deep clouds at Nauru. Although this result is somewhat biased on the low side due to the inability of lidar signals to penetrate through thicker clouds and limitations in radar detection of high, thin cirrus, the Hartmann et al.  mechanism does not explain the persistence of the majority of the Nauru cirrus observations.
6.5. Cloud Forcing
 In addition to heating rates, top-of-atmosphere (TOA) infrared cloud forcing calculations provide a way to assess the radiative impact of tropical cirrus. Infrared cloud forcing, calculated for six cirrus cases, is related to visible optical depth in Figure 15. We estimate that a cloud becomes radiatively significant in terms of OLR when the cloud forcing is approximately 10 W m−2. According to Figure 15, cirrus clouds with τ > 0.06 have a significant cloud forcing. The overall mean optical depth for the entire cirrus sample is 0.15 with a median value of 0.065. According to the cirrus statistics, ∼46% of the entire cirrus data set and ∼82% of the highest cirrus with zb > 15 km have τ < 0.06. Therefore, slightly more than half the tropical cirrus observed at Nauru will significantly affect the OLR. Cloud forcing results are similar to McFarquhar et al. , whose simulations show that infrared heating is stronger than solar cooling for subvisual clouds.
 In this paper, we provide comparisons of remotely sensed cirrus cloud properties using two separate methods. The agreement in cloud base and top heights is remarkable given the fundamental differences between radar and lidar systems. Comparisons of frequency distributions of visible optical depth also compare well given the limitations of each retrieval. However, there are some discrepancies that should be noted. First, it is clear that the sensitivity of the MMCR prohibits the detection of ice crystals above 15 km. This results in the underestimation of cloud top height over 40% of the time, and ∼13% of clouds detected by the MPL are completely missed by the MMCR. Rough estimates of ice crystal effective radius can be made by using the relationship τ = No(πr2Qe)Δz, where No is the number concentration, Δz the cloud thickness, Qe is the extinction efficiency, and r the particle radius. Using τ = 0.019 and Δz = 0.36 km, which are the median values for clouds with zb > 15 km reported in Table 1, we estimate that r is between 9 and 17 μm if No is 0.102 and 0.029 cm−3, respectively. These values of No are the maximum and minimum concentrations reported by Heymsfield and Jahnsen . However, if we use the higher value of No = 0.4 cm−3 simulated by Jensen et al. , r ∼ 4.6 μm. Regardless of the assumed No, it is apparent that cirrus clouds that lie near the tropical tropopause probably consist of small particles, which contributes to the decrease in radar sensitivity to high, thin cirrus. The presence of small ice crystals in tropopause cirrus is consistent with the in situ measurements of Heymsfield and Jahnsen  and Booker and Stickel . Radar and lidar τ comparisons also show agreement except at very low optical depths (τ < 0.04). These differences are attributed to uncertainties in both retrieval techniques for small τ. Although there has been some in situ verification of ground-based retrievals for midlatitude cirrus, where τ is on average higher than tropical cirrus [Sassen and Comstock, 2001], there are, unfortunately, no in situ measurements of subvisual tropical cirrus available that coincide with the MMCR and MPL observations in the tropical western Pacific.
 Frequency of cloudiness estimates derived from MPL measurements show that cirrus occurrence dominates the Nauru observations when compared to low cloud frequency, even when adding together percentages of low clouds with and without cirrus aloft. McFarquhar et al.  draw a distinction between “thin” cirrus that is associated with convection and “subvisual” cirrus that is not. We also find a similar result; however we have made the distinction that subvisual clouds above 15 km are not directly associated with convection, which is a large fraction of cirrus clouds observed at Nauru during 1999. This was also indicated by Winker and Trepte , although they observed tropopause cirrus in only 7% of lidar profiles. It is important to keep in mind that as the height of the tropical tropopause increases during the austral summer (December–Februrary) the height of tropopause cirrus will also increase.
 The water vapor responsible for the formation of isolated cirrus sheets may be due to moisture advected from a remote location where convection occurred up to 10 days prior [Pfister et al., 2001]. During 1999, there is a strong La Niña pattern in the Pacific Ocean, which means that the majority of convective activity is situated to the west of Nauru. Therefore, it is important to keep in mind that the tropical cirrus climatology presented in this paper represents only 1 phase of the El Niño-Southern Oscillation (ENSO) cycle. There is clearly a need to examine a longer time period of observations in order to capture the composite cirrus properties for all phases of the ENSO cycle.
 We have also discussed the relation of the different cirrus cloud types to their formation mechanisms and the radiative impact of tropical cirrus on upper tropospheric dynamics. Our findings indicate that large-scale dynamics play an important role in the formation and persistence of tropical cirrus because more often than not, observed cirrus have radiative properties that are more typical of this generating mechanism than those created by deep convection. In the case where cirrus clouds persist for several hours and appear to be detached from local convective activity implies that the convection does not directly maintain the layer. However, convection occurring several days prior may at least provide the moisture needed to initiate cloud formation. Although the radiative heating produced by subvisual cirrus is small and may not directly impact the radiation budget at the surface, the effect on upper tropospheric dynamics and cirrus persistence may be important. If heating rates are relatively small (Figure 12), either a local temperature change or upward motion will result. A temperature increase would cause cirrus to dissipate in less than a day, while vertical lifting would allow the layer to last for a day or longer [Jensen et al., 1996]. The radiative heating generated by these thin, subvisual layers may also invoke circulations that actually “pump” water vapor into the layer, which may help to sustain the cirrus cloud as well as transport water vapor into the stratosphere [Sherwood, 1999]. Regardless of the formation mechanism, it is apparent that more than half of tropical cirrus clouds in this study will have a significant impact on OLR.
 Although our results provide a good initial data set of tropical cirrus cloud properties as viewed from the ground, there is still potential for improvement. There is a clear need to verify the microphysical properties of high, subvisual cirrus using aircraft measurements. The concise overview of existing in situ measurements presented by McFarquhar et al.  provides a good initial data set that we can use to model tropical cirrus radiative properties. However, most measurements have been in convective anvils with only a few cases measuring tropopause cirrus. Additional in situ data would provide a method for verifying both the radar and lidar retrieval algorithms, as well as our assumption that laminar cirrus consist of small particles.
 Satellite based multiangle viewing instruments such as the Multiangle Imaging Spetroradiometer (MISR) and the multispectral Moderate-Resolution Imaging Spectroradiometer (MODIS) will provide improved detection of thin cirrus layers. Future comparisons between these satellite instruments and the data set presented here will not only help with the validation of satellite retrievals, but will also help assess the climatic implications of tropical cirrus by providing global coverage. The enormous horizontal extent and temporal persistence of high, laminar cirrus will undoubtedly create significant amounts of heating in the upper troposphere. The potential impact of this heating is still uncertain; however, a dynamical response, such as horizontal divergence or vertical lifting, is probable. Further study is also needed to help link the formation and persistence of tropical cirrus to specific mechanisms, which will ultimately determine the radiative impact of cirrus clouds on climate.
 The Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy (DOE). This research was supported by the DOE Office of Biological and Environmental Research under contract DE-AC06-76RL01830 as part of the Atmospheric Radiation Measurement Program.