Overview of MPLNET Version 3 Cloud Detection

Jasper R. Lewis Joint Center for Earth Systems Technology, University of Maryland Baltimore County, Baltimore, Maryland

Search for other papers by Jasper R. Lewis in
Current site
Google Scholar
PubMed
Close
,
James R. Campbell Naval Research Laboratory, Monterey, California

Search for other papers by James R. Campbell in
Current site
Google Scholar
PubMed
Close
,
Ellsworth J. Welton NASA Goddard Space Flight Center, Greenbelt, Maryland

Search for other papers by Ellsworth J. Welton in
Current site
Google Scholar
PubMed
Close
,
Sebastian A. Stewart Science Systems and Applications, Inc., Lanham, Maryland

Search for other papers by Sebastian A. Stewart in
Current site
Google Scholar
PubMed
Close
, and
Phillip C. Haftings Science Systems and Applications, Inc., Lanham, Maryland

Search for other papers by Phillip C. Haftings in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The National Aeronautics and Space Administration Micro Pulse Lidar Network, version 3, cloud detection algorithm is described and differences relative to the previous version are highlighted. Clouds are identified from normalized level 1 signal profiles using two complementary methods. The first method considers vertical signal derivatives for detecting low-level clouds. The second method, which detects high-level clouds like cirrus, is based on signal uncertainties necessitated by the relatively low signal-to-noise ratio exhibited in the upper troposphere by eye-safe network instruments, especially during daytime. Furthermore, a multitemporal averaging scheme is used to improve cloud detection under conditions of a weak signal-to-noise ratio. Diurnal and seasonal cycles of cloud occurrence frequency based on one year of measurements at the Goddard Space Flight Center (Greenbelt, Maryland) site are compared for the new and previous versions. The largest differences, and perceived improvement, in detection occurs for high clouds (above 5 km, above MSL), which increase in occurrence by over 5%. There is also an increase in the detection of multilayered cloud profiles from 9% to 19%. Macrophysical properties and estimates of cloud optical depth are presented for a transparent cirrus dataset. However, the limit to which the cirrus cloud optical depth could be reliably estimated occurs between 0.5 and 0.8. A comparison using collocated CALIPSO measurements at the Goddard Space Flight Center and Singapore Micro Pulse Lidar Network (MPLNET) sites indicates improvements in cloud occurrence frequencies and layer heights.

Publisher’s Note: This article was revised on 11 October 2016 to correct an editing error in the title when initially published online.

Corresponding author address: Jasper R. Lewis, NASA GSFC, Code 612, Greenbelt, MD 20771. E-mail: jasper.r.lewis@nasa.gov

Abstract

The National Aeronautics and Space Administration Micro Pulse Lidar Network, version 3, cloud detection algorithm is described and differences relative to the previous version are highlighted. Clouds are identified from normalized level 1 signal profiles using two complementary methods. The first method considers vertical signal derivatives for detecting low-level clouds. The second method, which detects high-level clouds like cirrus, is based on signal uncertainties necessitated by the relatively low signal-to-noise ratio exhibited in the upper troposphere by eye-safe network instruments, especially during daytime. Furthermore, a multitemporal averaging scheme is used to improve cloud detection under conditions of a weak signal-to-noise ratio. Diurnal and seasonal cycles of cloud occurrence frequency based on one year of measurements at the Goddard Space Flight Center (Greenbelt, Maryland) site are compared for the new and previous versions. The largest differences, and perceived improvement, in detection occurs for high clouds (above 5 km, above MSL), which increase in occurrence by over 5%. There is also an increase in the detection of multilayered cloud profiles from 9% to 19%. Macrophysical properties and estimates of cloud optical depth are presented for a transparent cirrus dataset. However, the limit to which the cirrus cloud optical depth could be reliably estimated occurs between 0.5 and 0.8. A comparison using collocated CALIPSO measurements at the Goddard Space Flight Center and Singapore Micro Pulse Lidar Network (MPLNET) sites indicates improvements in cloud occurrence frequencies and layer heights.

Publisher’s Note: This article was revised on 11 October 2016 to correct an editing error in the title when initially published online.

Corresponding author address: Jasper R. Lewis, NASA GSFC, Code 612, Greenbelt, MD 20771. E-mail: jasper.r.lewis@nasa.gov

1. Introduction

Clouds play a critical role in the earth’s climate system because they are inextricably linked to the hydrological cycle and radiation budget (Liou 1986; Ramanathan et al. 1989; Wielicki et al. 1995; Norris 2000; Stephens 2005; Ringer et al. 2006; Kato et al. 2011). Information about cloud height, thickness, occurrence, and amount are critical inputs for a host of numerical applications involving climate research. Therefore, it is important to have highly accurate and quantitative data records of cloud properties that span several years and geographic regions. Verification of even the most basic modeling processes demands long-term and continuous observations of global cloud occurrence, if there is to be any confidence in their fidelity.

Various methods of determining cloud climatologies exist, each with their own advantages and limitations. Visual observations from the surface (Warren et al. 1985; Hahn et al. 1996; Hahn and Warren 1999) provide cloud fraction and morphological cloud types. However, these can be biased by the quality of technician training, underestimation of high clouds, sparse global coverage, and a lack of nighttime observations. Passive radiometric sensors aboard satellites, which are the core input of the International Satellite Cloud Climatology Project (ISCCP) as well as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Multiangle Imaging Spectroradiometer (MISR), offer a true global representation and have the best (unobstructed) potential view of high clouds (Rossow and Schiffer 1991, 1999; Moroney et al. 2002; Menzel et al. 2008; Marchand et al. 2010). However, these can undersample low-level maritime clouds and underrepresent optically thin cirrus clouds (e.g., Ackerman et al. 2008; Holz et al. 2008).

Active sensors, like lidar and radar (e.g., Platt et al. 1994; Moran et al. 1998; Wang and Sassen 2001), are the primary tools for observing and profiling cloud vertical structure to high accuracy. When flown aboard satellites, like Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO; Winker et al. 2010) and CloudSat (Stephens et al. 2002), active sensors also provide global coverage. Even still, the relatively narrow profiling curtain of current active sensors limits observation densities. In the case of CALIPSO and CloudSat, these missions provide at most two profiles per 24-h period over most regions, which limit studies of the diurnal impact of clouds on the Earth system. Fundamentally, an array of remote sensing methods is needed in order to investigate the complexity of clouds (Schiffer and Rossow 1983; Lazarus et al. 2000; WMO/WWRP 2012).

2. Micro Pulse Lidar Network

The National Aeronautics and Space Administration (NASA) Micro Pulse Lidar Network (MPLNET; Welton et al. 2001; http://mplnet.gsfc.nasa.gov) is a federated network of micro pulse lidar (MPL) systems deployed worldwide in support of basic science and the NASA Earth Observing System program (Wielicki et al. 1995). A benefit of MPLNET is the use of a standardized instrument employing a common suite of data processing algorithms with thorough uncertainty characterization that allows for straightforward comparisons between sites. With sites in polar, midlatitude, and tropical regions and continuous day/night, high-temporal-resolution datasets going back as far as 1999, MPLNET datasets represent a valuable archive for improving our understanding of global cloud macrophysical properties on diurnal, seasonal, and decadal scales.

There have been two versions of MPLNET data processing algorithms to date. The first, referred to as version 1, was released in 2000. Beginning in 2006, the project transitioned to version 2 (V2) data products, which are currently available. Version 3 (V3) data processing algorithms are currently in development.

The V2 level 1 MPLNET data products contain the system diagnostics (solar background counts, instrument temperature and energy, etc.), normalized relative backscatter (NRB), and NRB uncertainty. The NRB is reported at 1-min temporal resolution up to 30 km, typically at 75-m vertical resolution (some sites operate at 30 m). The NRB is defined as
e1
where C is the instrument calibration value, β is the backscatter coefficient from both molecules and particles, T2 is the corresponding total atmospheric two-way transmittance, and z is the altitude. The uncertainty in NRB is computed as
e2
where P(z) is the raw MPL count rate, B is the solar background count rate, A(z, E) is the detector afterpulse, E is the pulse energy, and O is the overlap function. A detailed description of NRB and the calculation of its uncertainty is provided by Campbell et al. (2002) and Welton and Campbell (2002). Level 1 data are available in real time with no quality assurance. Cloud-base and cloud-top heights are identified in the V2 level 1.5 (real time, no quality assurance) and level 2 (not real time, quality assured) MPLNET data products. The V2 level 1.5 data products use temperature and pressure profiles from COESA (1976) to determine molecular calculations, while the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis (Kalnay et al. 1996) is used for V2 level 2 data.

MPLNET V2 data products have been used to distinguish cloud presence in a number of scientific investigations to date. For example, Campbell and Sassen (2008) use data from the South Pole to document polar stratospheric cloud occurrence over multiple seasons. Shupe et al. (2011) consider MPLNET measurements at Ny-Ålesund, Norway, as context for evaluating Artic cloud properties. Others have investigated cirrus contamination of Aerosol Robotic Network (AERONET) aerosol optical depth in Southeast Asia (Chew et al. 2011; Huang et al. 2011) and globally (Huang et al. 2012). Lolli et al. (2013) use collocated 355-/527-nm MPLNET observations to estimate the drizzle droplet size from stratocumulus and stratus clouds.

V2 cloud retrievals underreport the presence of high-altitude clouds, as demonstrated in Fig. 2 in Huang et al. (2011). Therefore, a new V3 cloud detection algorithm has been developed to improve the quality of MPLNET cloud products. Similar to V2, the new algorithm uses a combination of signal-processing techniques to resolve cloud boundaries. However, a multiresolution temporal averaging scheme has been added to increase sensitivity under conditions of low signal-to-noise ratio. Meteorological profiles provided by the Goddard Earth Observing System, version 5 (GEOS-5), atmospheric general circulation model (AGCM; Rienecker et al. 2008; Molod et al. 2012) are now used for all molecular calculations. Specifically, the Forward Processing for Instrument Teams (FP-IT) GEOS-5, version 5.9.1, data are utilized (http://gmao.gsfc.nasa.gov/products). Modeled profiles for this study were subsampled from the GEOS-5 grid containing the Goddard Space Flight Center (GSFC) site location, and interpolated to the MPLNET range and time resolutions (75 m, 1 min, respectively).

The goals here are to describe the new algorithm and demonstrate its performance. We outline changes relative to V2 cloud detection and describe how the new algorithm is applied to a variety of cloudy scenes. We employ one year of data collected at the GSFC MPLNET site (38.99°N, 76.84°W; 0.05 km above MSL) to compare the V2 and V3 results and highlight the impact of our upgraded techniques through differences in macrophysical cloud properties observed from this location. Comparisons to nearby observations from CALIPSO are used to compare the geographical representativeness of the V2 and V3 cloud detections.

3. Version 2 cloud detection algorithm

Examples of daytime and nighttime NRB profiles at GSFC are shown in Fig. 1. Both profiles show high-level clouds with base heights near 10 km and top heights near 13 km MSL. The daytime NRB profile exhibits a relatively lower signal-to-noise ratio compared with the nighttime case due to higher solar background, which makes detection of elevated layers an increasingly difficult task.

Fig. 1.
Fig. 1.

Examples of the NRB at GSFC on 7 Jul 2012 at day (1800 UTC, black) and night (0300 UTC, red). Both profiles are at the base 1-min temporal resolution.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

Layers are identified in the V2 cloud detection algorithm by a combination of two retrieval methods applied to the level 1 data products. The first method requires that the first derivative of the lidar signal exceed a minimum threshold in order to detect a layer. The assumption of strong signal gradients makes this well suited for detecting liquid-phase clouds, which are frequently at lower levels in the NRB profile and have a higher signal-to-noise ratio. This is referred to as the gradient-based cloud detection method (GCDM). The second method is designed for use in cases of low signal-to-noise ratio (SNR) and relies on uncertainties in the lidar signal. This method uses two tunable thresholds and one objective threshold to identify cloud boundaries, and is referred to as the uncertainty-based cloud detection method (UCDM).

Given the relatively low SNR exhibited by the MPL in the upper troposphere at base 1-min resolution (primarily during daytime), no single procedure is used to detect all cloud types at all times. Thus, the merger of these two methods offers the possibility to retrieve the entire cloud vertical structure to the limit of signal attenuation. We describe the basis for each method, as follows.

a. Gradient-based cloud detection

Autonomous methods of cloud detection using gradients in the lidar signal, such as the differential zero-crossing method described by Pal et al. (1992), are well established. The first step in the GCDM is to normalize the NRB using the attenuated molecular backscatter coefficient,
e3
which produces an attenuated scattering ratio R′(z) multiplied by the instrument calibration value as
e4
Here the subscripts m and p denote contributions from molecules and particles, respectively. In clear air, where βp = 0 and Tp is constant, the derivative dR′/dz = 0.
The first derivative of CR′(z) is used to identify clouds in the GCDM and is calculated using a three-point (quadratic) Lagrangian interpolation, scanning the profile from the bottom up (Hildebrand 1987; Bevington and Robinson 2002). Because of increasing uncertainty in the profile with height, the GCDM retrieval is performed only up to a “noise altitude,” defined as the altitude at which the uncertainty, δNRB, exceeds half of the NRB (or conversely, analogous to an SNR of 2). Cloud presence corresponds with an increase in total backscatter, which results in a large positive gradient in CR′(z) with height as seen in Fig. 2. The threshold used to identify the cloud base is defined as
e5
where is the mean value of CR′(z) up to the noise altitude and K is an empirical parameter (unitless), set to 10 for this study. The value of K is chosen empirically so that it is high enough to reject insignificant peaks in the first derivative of CR′(z) (i.e., aerosol stratification in the surface-attached layer or signal noise) while remaining sensitive enough to identify weakly scattering clouds.
Fig. 2.
Fig. 2.

Representation of the GCDM (GSFC; 0233 UTC 7 Nov 2012). (left) The product of the calibration value and the attenuated scattering ratio and (right) its derivative. The vertical dashed lines represent the amin and amax thresholds.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

The cloud base is identified at the altitude bin immediately preceding that where the first derivative of CR′(z) exceeds amax. A negative gradient in CR′(z) occurs near the top of a cloud, corresponding with the decrease in total backscatter and the impact of signal attenuation through the cloud. The threshold used to identify the cloud top is defined as
e6
The algorithm begins by looking for altitude bins above the cloud base where the first derivative of CR′(z) falls below amin. Then the altitude bin where the first derivative initially returns above amin is identified as the cloud top. However, if this condition is not met due to significant attenuation within the cloud, then the noise altitude is reported and the designation as an apparent cloud top is used in accordance with standards agreed upon by the Experimental Cloud Lidar Pilot Study (ECLIPS; Platt et al. 1994). We define apparent cloud-top heights as the altitude where the lidar signal is determined to be completely attenuated following the criteria in section 5c. Otherwise, the cloud top is considered a true cloud top.

Two cloud layers are apparent in Fig. 2. Both cloud bases are identified at the altitudes immediately below the amax exceedances (dashed line, positive derivative). The first (lowest) cloud top can be found using the amin threshold (dashed line, negative derivative). But the derivative never falls below amin for the second cloud layer. Therefore, the noise altitude is used to identify the apparent cloud top. Only true (not apparent) cloud tops are reported in V2 MPLNET cloud products.

b. Uncertainty-based cloud detection

An alternative to algorithms that utilize gradients in the lidar return to identify clouds are approaches that compare cloudy lidar returns to clear-sky returns (e.g., Clothiaux et al. 1998). Similarly, the UCDM uses a theoretical molecular return and the signal uncertainty to detect elevated clouds, and is fully described by Campbell et al. (2008, hereafter C08). The first step in the UCDM is to approximate the value of the instrument calibration value using CR′(z) from Eq. (4). Next, a clear-sky search is performed to locate a normalization region where we can approximate that βp approaches zero over a certain number of range bins, N. The nature of the UCDM allows only for cloud detection at altitudes above the normalization region. The calibration value is approximated in a profile-by-profile process by averaging Eq. (4) over the N bins. C08 stress that this final normalization value, , must be distinguished from C due to unknown transmission losses below the normalization region.

Once has been calculated, a so-called pseudoattenuated backscatter (PAB) and its uncertainty are solved as
e7
and
e8
Equation (8) is then modified by substituting the attenuated molecular backscatter for the PAB to develop an objective threshold,
e9
that is used to differentiate what are first presumed to be cloud returns from molecular return. Range bins above rN that satisfy the minimum detectable scattering ratio for particulates, defined as
e10
are then evaluated to determine whether they represent particulate layer base heights. Range bins that do not meet the condition from Eq. (10) are used to establish “clear air” slots and are disqualified from the particulate base height search. An example of the UCDM is shown in Fig. 3, where the red line indicates the threshold used to distinguish particulate from molecular returns, and r1 and rN indicate the bottom and top of the normalization region, respectively.
Fig. 3.
Fig. 3.

Representation of the UCDM (GSFC; 0300 UTC 7 Jul 2012). The bottom (r1) and top (rN) of the normalization region are indicated by the horizontal dashed lines. The red curve represents the product of and the minimum detectable scattering ratio used to determine the cloud boundaries.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

Running averages of PAB and δPAB are used in conjunction with two additional tunable thresholds, ϕ and κ (both analogous to an SNR), to determine the actual layer base and top heights. The threshold ϕ sets the minimum average value of PAB/δPAB for bins that exceed Eq. (10) in order to identify the layer base. At this point in the algorithm, the UCDM assumes that any such layers detected are hydrometeor clouds, thereby leaving the potential for false detection of elevated aerosol layers. In the absence of supplementary information, however, such as color ratio (Liu et al. 2005) or depolarization (Cho et al. 2008; Omar et al. 2009) and combined with the goal of resolving as much thin cirrus as possible in the low-SNR portions of the NRB profile, this is unavoidable. Mitigation strategies are described further below.

The threshold κ sets the minimum average value of PAB/δPAB for bins that do not exceed Eq. (10) in order to identify clear-air layers and consequently particulate layer tops. The sensitivity of V2 cloud detection to the tunable thresholds is evaluated in C08, and the values chosen for ϕ and κ will depend on the site location and instrument performance parameters. Based on threshold testing performed in C08, reducing ϕ results in a greater number of cloud detections, but it also increases the number of false positives. Increasing ϕ has the opposite effect. Meanwhile, the effect of increasing κ results in increasing the vertical depth of cloud layers because the threshold is not as easily surpassed. To a lesser extent, varying κ can also influence the detection of a cloud-base height, because the detection of clear air can terminate the search for a cloud base.

c. V2 cloud retrievals

An example of V2 cloud retrievals is shown in the top panel of Fig. 4. Results from the GCDM and UCDM are integrated based on the noise altitude (described in section 3a). Clouds occurring below this height are reported from the GCDM. All clouds above the noise altitude are identified using the UCDM. At night, the noise altitude reaches above typical cirrus cloud heights at GSFC. Therefore, the GCDM is almost exclusively responsible for cloud detection. As a result, weakly scattering cirrus can go undetected, since GCDM thresholds are tuned primarily with boundary layer phenomena in mind (i.e., suppression of aerosol identification). This can be observed frequently between 0300 and 0600 UTC in Fig. 4, where cirrus presence is underreported and cloud-base heights are overestimated.

Fig. 4.
Fig. 4.

(top) Version 2 cloud retrievals at GSFC site on 23 Mar 2012 (1-min resolution). Cloud bases and cloud tops are indicated by orange and red markers, respectively. The noise altitude (i.e., δNRB/NRB > 0.5) is represented by the solid white line, and the bottom of the normalization region is represented by the dashed white line (4-min smoothing applied to both for clarity). (bottom) As in (top), but using the V3 algorithm at 1-min resolution. Green markers indicate apparent cloud tops. The bottom of the normalization region is represented by the solid white line (4-min smoothing applied for clarity).

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

In the daytime, the noise altitude shown in Fig. 4 falls between 8 and 9 km and the UCDM is responsible for all cloud detection above it. In several instances, cloud bases (orange markers) are shown while the corresponding cloud tops (red markers) appear to be missing. In these cases the lidar signal is assumed to be significantly attenuated, and therefore no cloud top is reported.

4. Version 3 cloud detection algorithm

The V3 algorithm is based on V2 with a few meaningful changes to the UCDM. Consequently, the changes in V3 represent an update to C08 and how the GCDM and UCDM are merged. A schematic of the V3 cloud detection algorithm at the base 1-min NRB temporal resolution is shown in Fig. 5. Low-altitude obstructions (e.g., fog or low stratus decks) reduce SNR by attenuation of the lidar signal and limit the accuracy of cloud retrievals. Therefore, each profile is screened for these attenuation-limited conditions by a process described in section 5c. If no such obstructions are found, then the first step in the UCDM is to calculate the normalization value.

Fig. 5.
Fig. 5.

Schematic of the V3 cloud detection algorithm at the base 1-min resolution.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

a. Normalization region

As mentioned in C08, it is most practical to find a normalization region to calculate nearest to the instrument in order to increase the depth of the profile analyzed for clouds. In V2, the clear-sky search is initiated at 1 km above ground level (AGL). However, in V3, an altitude of 5 km MSL is chosen to reduce the likelihood of normalizing within relatively homogenous aerosol layers in or near the boundary layer, which results in high values of , thus reducing the ability to detect faint cloud layers. Furthermore, based on recent global cloud and aerosol studies (Holz et al. 2008; Sassen et al. 2008; Campbell et al. 2015), 5 km represents a reasonable height to begin the search for high clouds in tropical and midlatitude locations. For comparison, C08 initiated the search at 3 km AGL for the South Pole site. In the event that it is not possible to perform the normalization step at 5 km, the UCDM attempts to normalize lower in the atmospheric profile iteratively, to as low as 1 km AGL, as done in the V2 algorithm.

The normalization region serves as the boundary between the GCDM and UCDM retrievals in the V3 cloud algorithm, allowing the better-suited method (GCDM for low clouds and UCDM for high clouds) to operate during both day and night. The V3 cloud retrievals in the bottom panel of Fig. 4 can be compared with the V2 retrieval in the same figure to see the relative apparent improvement. We also note that there are conditions when the GCDM may be used to retrieve high-level clouds (e.g., when the normalization step fails) or the UCDM may be used to retrieve low-level clouds (e.g., when the normalization region occurs lower than 5 km). Therefore, the retrieval method for each cloud layer is provided as an output parameter.

b. Objective threshold

The objective UCDM threshold α, defined in Eq. (9), is now adjusted for attenuation within cloud layers in V3. To attenuate the α threshold, we assume an initial unity transmission at the base of the first detected cloud layer. Then the two-way transmittance is calculated at each altitude bin k within the cloud layer as
e11
where βC is the cloud backscatter coefficient, S* is the effective extinction-to-backscatter ratio (or effective lidar ratio), Δz is the vertical resolution of the instrument, and R′(z) is the attenuated scattering ratio given by
e12
The effective extinction-to-backscatter ratio is given by S* = ηSC, where η is the multiple scattering factor (Platt 1973, 1979) and SC is the extinction-to-backscatter ratio of the cloud. Overestimating S* will lead to excessive attenuation of the objective threshold and eventual mischaracterization of clear sky as cloud. Therefore, in this step, a very conservative choice for S* equal to the molecular extinction-to-backscatter ratio (8π/3) is used.
Once the transmittance has been determined, Eq. (9) is modified to recalculate the α threshold as
e13
Attenuating the α threshold does not affect the retrieval of base heights for the lowest cloud layer, but we are able to better resolve cloud tops for optically thick clouds. Furthermore, for multilayer clouds, this change also leads to lower estimates of the base height of the higher cloud and, in some cases, clouds that were completely undetected with V2 can now be retrieved.

c. Multitemporal resolution

Cloud boundaries are reported only at 1-min temporal resolution in the V2 algorithm. However, instances of high solar background reduce UCDM performance. So, as described by C08, multitemporal resolution settings are used in V3. In addition to the base 1-min temporal resolution, the UCDM is performed for intermediate (5 min) and long (20 min) temporal averages using a sliding window centered on a corresponding 1-min profile.

A flowchart describing the process is shown in Fig. 6. Retrievals at the base temporal resolution are used to screen profiles at longer averages, thus reducing the effects of attenuation-limited profile contamination. Within a window of N profiles, k profiles are removed from the average if an attenuating structure is detected below 5 km at the base temporal resolution. If k > N/2, then the entire average profile is rejected.

Fig. 6.
Fig. 6.

Schematic of the V3 cloud detection algorithm for higher temporal averages.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

A combined cloud scene is created using cloud boundaries retrieved from the three temporal resolutions. First, the combined scene uses the cloud boundaries from the base temporal resolution. Next, the intermediate and then long temporal resolutions are used to fill in the missing gaps, as needed. The merger of the three temporal resolutions is performed only for clouds with a top temperature less than −37°C, based on the work of Sassen and Campbell (2001) and Campbell et al. (2015), as they are associated with cirrus clouds and assumed to have larger spatial extent than low clouds. Cloud boundaries and other properties (e.g., temperature and optical depth) are always reported at the highest temporal resolution possible to reduce the effects of cloud smearing caused by averaging. This is broadly consistent with the methodology used for NASA Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) level 2 products and their gridding of multiple spatial resolutions from the selective iterated boundary locator (SIBYL) to their vertical feature mask (Vaughan et al. 2005). A noted difference is that SIBYL also uses an intensity-clearing process to remove features detected at finer resolutions from the coarser spatial averages. No such intensity clearing is performed with the V3 algorithm.

d. False positives

As mentioned previously, the UCDM assumes that signals rising above the α threshold are always caused by clouds. Additional constraints are thus required to reduce the number of instances when noise excursions, elevated aerosol layers, or poor normalizations produce false cloud retrievals. The first constraint establishes a minimum layer thickness of 150 m (i.e., two bins at 75-m resolution) in order to qualify a potential layer as a cloud. Therefore, once an altitude bin exceeds the minimum detectable scattering ratio in Eq. (10), the successive bins within a depth equal to the minimum layer thickness are also required to exceed this threshold in order to establish a cloud base. Similarly, a minimum clear-air distance of 150 m is used to establish a cloud top and to avoid falsely segmenting a single cloud into multiple layers. The second constraint requires that the standard deviation of R′(z) within the detected layer exceeds an empirically determined threshold, σmin, which varies as a function of cloud-top temperature. Cloud layers generally cause large variances in R′(z), either through attenuation effects in otherwise colloidally stable liquid water clouds or since ice crystals fall within cirrus cloud layers, creating complex structures. By contrast, aerosols in the free troposphere settle in stratified stable layers absent of convection and are expected to be homogeneous within each layer. Thus, clouds layers can be distinguished by their relatively large standard deviations of R′(z).

To determine σmin, a dataset was developed consisting of 144 days at GSFC in 2012 and 27 061 retrieved UCDM layers at 1-min resolution (18 308 thin cirrus cloud layers, 3233 noncirrus layers, 5520 aerosol layers) when the particulate type could be reasonably identified from visual analysis (Fig. 7). Thin cirrus clouds are distinguished using a cloud-top temperature threshold of −37°C (Sassen and Campbell 2001; Campbell et al. 2015) and a maximum cloud optical depth (COD) of 0.3 (Sassen and Cho 1992). The COD calculation uses a process described by Chew et al. (2011) and is discussed fully in section 5a. Noncirrus clouds are those with cloud-top temperature warmer than −37°C.

Fig. 7.
Fig. 7.

(left) Representative groupings of cirrus clouds, noncirrus clouds, and aerosols from GSFC (3–5 Aug 2012). (right) Scatterplot of full dataset used to determine σmin threshold. The Xs represent the median values of each group, the thin boundaries enclose 50% of the data nearest to the median, and the thick boundaries enclose 90% of the data. The dashed line indicates the empirically determined σmin. The solid black line indicates another threshold tested, σtest(3).

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

Several choices for σmin were evaluated using error matrices, also known as confusion matrices (Congalton and Mead 1986), and the corresponding values of accuracy and the Matthews correlation coefficient (MCC; Matthews 1975), which are defined as
e14
and
e15
where TP is the number of instances when clouds were correctly identified, TN is the number of instances when aerosols were correctly identified, FP is the number of instances when aerosols were misidentified as clouds, and FN is the number of instances when clouds were misidentified as aerosols. The MCC provides a measure of how well the predicted outcome correlates to observations and ranges from −1 (total disagreement) to +1 (complete agreement). Some of the thresholds evaluated include a conservative threshold σtest(1) = 2 at all temperatures, a relaxed threshold σtest(2) = 0.2 at all temperatures, and a simple temperature-dependent threshold σtest(3) shown in Fig. 7. The choice for σmin that resulted in the highest values of accuracy (0.92) and MCC (0.74) is given by
e16
The error matrices, accuracies, and MCCs for the thresholds described above are provided and in Table 1. Similar results were observed at the 5- and 20-min resolutions but are not shown for brevity. As with any single-channel lidar system, there remains a degree of ambiguity in distinguishing clouds and aerosols with this technique. Strong smoke and dust layers can also produce highly variable scattering ratios and be misclassified as cloud layers. This is especially true at higher altitudes associated with long-range transport, as evidenced by the aerosols occurring at temperatures below −40°C in Fig. 7, which are interpreted as clouds.
Table 1.

Error matrices for CAD used to determine σmin.

Table 1.

The final constraint used to distinguish cloud from aerosol layers is that the estimated COD exceed a threshold, τmin. Through empirical testing, we estimate COD and set τmin = 0.005 based on analysis of these subsets relative to the perception of how noise impacts these subsamples combined with a similar analysis by Thorsen et al. (2011).

We briefly note here that lidars with polarization capabilities have recently been incorporated into the MPLNET project. However, the depolarization products are still in development, so the algorithm presented here does not rely on such data. It remains as a future goal to demonstrate how polarization measurements can be used to improve aerosol–cloud discrimination, once a sufficient amount of data is collected from the new polarized sites.

5. Version 3 algorithm output

A partial listing of the V3 cloud detection algorithm output parameters is provided in Table 2. The output parameters from all temporal averages are gridded to 1-min temporal resolution, as previously described in the combined cloud scene. The number of cloud layers detected, the day–night flag, and the attenuation altitude are given as a single value each minute, characterizing the atmospheric column. All other cloud products and data flags correspond with individual cloud layers and are provided each minute with dimensions equal to the number of cloud layers detected.

Table 2.

MPLNET V3 cloud detection algorithm output.

Table 2.

a. Cloud phase and cirrus cloud optical depth

In the absence of visual cloud observations, as is the case for autonomous lidar measurements made by MPLNET, Sassen and Campbell (2001) recommend using a minimum cloud-top temperature of −37°C to identify cirrus. In the V3 cloud algorithm, we use this thermal threshold to distinguish ice clouds (i.e., cirrus) from all other cloud phases. At present, no attempt is made to distinguish mixed-phase clouds from liquid water clouds.

Campbell et al. (2015) evaluate the −37°C cloud-top temperature threshold globally versus the level 2 CALIOP algorithms that identify ice-phase cloud layers and found that over 99% of clouds satisfying this thermal threshold were classified as ice. Furthermore, 81% of all ice clouds had cloud-top temperatures less than −37°C. They conclude, consistent with the findings of Sassen and Campbell (2001), that this thermal threshold is stable for specifically distinguishing cirrus cloud presence in lidar studies that lack depolarization, though there is some ambiguity in cases of “warm” cirrus that have cloud-top temperatures greater than −37°C.

An estimated COD is calculated for clouds distinguished as cirrus using the procedure described by Chew et al. (2011). Two-way cloud transmission is calculated using Eq. (11). However, now the value of S* = ηSC is selected based on the cloud-top temperature. The effects of multiple scattering depend on the lidar field of view (FOV), the distance between the lidar and the scattering target, and the particle size and density (Bissonnette 2005). MPLs have a very narrow FOV (100 μrad) that minimizes the effects of multiple scattering. Previous studies of cirrus clouds using the MPL have used values of η between 0.8 and 0.9 (Lo et al. 2006; Comstock et al. 2002). Reported values of SC are on the order of 16–18 sr for liquid water clouds (Pinnick et al. 1983, Yorks et al. 2011) and 10–40 sr for cirrus (Sassen and Comstock 2001; Chen et al. 2002; Yorks et al. 2011; Garnier et al. 2015). A value of S* = 16 sr is chosen for layers with cloud-top temperatures warmer than −37°C, and two values of S* = 20 sr and 30 sr are used at colder temperatures where cirrus clouds are expected. The cirrus COD is calculated using two values for the effective extinction-to-backscatter ratio to account for uncertainties in the values of η and SC. We note that due to uncertainty in the lidar ratio for cirrus clouds, these estimates may represent the lower limit of COD.

Next, Eq. (12) is used to solve for the cloud backscatter coefficient and the estimated COD is calculated with the cloud top–dependent S* as
e17
The fidelity of the COD estimate is limited by the choice of S* and the accuracy of cloud boundaries retrieved. The relative error in the lidar-derived optical depth is smallest for low optical depths and proportional to ΔS*/S* as τ approaches zero (Winker et al. 2009). Lidar signals are unable to penetrate through optically thick clouds, which cause uncertainty in the value at the apparent cloud top. In these cases, the estimated COD will be biased low. Similarly, attenuation from the bottommost cloud layer leads to uncertainty in corresponding retrievals of higher clouds for lidar profiles containing multiple cloud layers. As a result, the estimated COD will be most reliable for single-layer, optically thin clouds.

b. Retrieval index

While cloud boundaries are reported only at a single temporal resolution, a retrieval index is included to indicate whether the cloud was also detected at one or more of the other temporal averages. Cloud layers at different temporal resolutions are considered the same if (i) they share a common base or top height within a vertical depth of 250 m or (ii) one cloud layer is completely enveloped within the other.

An example of a combined cloud scene, with corresponding retrieval indices, is shown in Fig. 8. The value of the retrieval index is equal to the sum of the temporal resolutions used to identify the cloud layer. For example, if a cloud is detected at all three temporal resolutions, then the value of the retrieval index is 1 + 5 + 20 = 26. The advantage of the multitemporal averaging scheme can be seen during the day between 1400 and 1500 UTC in Fig. 8. The elevated cloud layer (~15 km) is mostly undetected at the 1-min resolution, but it can be resolved using the longer averages. The cloud layer at ~2 km produces attenuation-limited conditions that prevent use of higher temporal averages for much of the cirrus cloud layer above it.

Fig. 8.
Fig. 8.

(top) NRB from lidar measurements at the GSFC MPLNET site on 7 Jul 2012. (bottom) Retrieval indices showing the temporal resolutions used to detect the combined cloud scene. Square symbols are used to indicate the cloud base, and the Xs are used to indicate for cloud top.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

c. Attenuation altitude

Because the lidar signal can become completely attenuated within optically thick clouds, it is important to determine when a true cloud top is being reported as opposed to an apparent cloud top. Nadir-pointing lidar instruments have an advantage of using the ground return to determine whether the lidar signal has been extinguished. However, with zenith-pointing lidar, that determination is more tenuous. Winker and Vaughan (1994) defined a transmittance index to determine when the lidar signal was fully attenuated based on the percentage of samples above the cloud top that exceeded the background. Other techniques used for zenith-pointing lidar have included the use of a minimum threshold lidar signal along with its slope (Wang and Sassen 2001) and comparisons with molecular profiles (Lo et al. 2006).

In V3, cloud tops (both true and apparent) are reported for all cloud layers along with the altitude at which the lidar signal is determined to be fully attenuated. This attenuation altitude is found by starting at the range bin of the highest reported cloud altitude and incrementally moving upward in the profile until, within a depth of 2 km, (i) the percent difference between the mean pseudoattenuated backscatter and modeled attenuated molecular backscatter falls below some threshold T1, and (ii) either the backscatter signal falls below a minimum value or the percentage of range bins where the backscatter signal is less than zero exceeds a threshold T2.

This application pertains specifically to profiles that contain clouds or other obstructions, since the attenuation thresholds can also be satisfied by other conditions that lead to low SNR (e.g., high solar background). Attenuation-limited conditions from low-altitude obstructions are found with the same search criteria, though the search is limited to the first 2 km above the surface.

6. Results

To demonstrate the effects of the changes implemented in the V3 algorithm, we compare V2 and V3 cloud retrievals for one year at the GSFC MPLNET site. Table 3 shows data sampling statistics for 2012, including the total number of profiles and percentage of time when 1-min NRB measurements were available monthly. Observable profiles are given as the number and percentage of available profiles that are not attenuation-limited below 2 km MSL. Profile attenuation was determined using the V2 method because it is the most restrictive and it ensures an even comparison between the two cloud detection algorithms. The diurnal distribution of data recorded and successful V3 normalizations to calculate are also shown. There is very little differentiation between the distributions of V2 and V3 normalizations, so only the number of successful V3 normalizations are shown for simplicity. Though MPLNET measurements are continuous, there is a decrease in the number of successful normalizations near solar noon in relatively low-SNR conditions.

Table 3.

Summary of data collected at GSFC in 2012 (percentages shown in parentheses).

Table 3.

Because V3 uses a merged cloud scene and V2 is processed only at 1-min resolution, V3 retrievals are evaluated using the base 1-min resolution (V3b) and the merged cloud scene (V3m). Comparisons are limited to cloud-base statistics because cloud tops are not recorded for all V2 retrievals. Comparisons between high-cloud detections from CALIPSO and MPLNET are made in order to demonstrate the geographical representativeness of the V3 cloud retrievals versus V2. Finally, we describe the macrophysical and optical characteristics of cirrus clouds observed over GSFC during this study, again adhering to the −37°C cloud-top temperature threshold described by Sassen and Campbell (2001) and Campbell et al. (2015), using V3 retrievals.

a. Vertical dependence

Figure 9 shows the cloud-base distributions retrieved from the V2 and V3 algorithms, at GSFC during 2012. A bimodal distribution similar to that observed by Winker and Vaughan (1994), with peaks at ~1–2 km and ~9–10 km is apparent. The total number of cloud observations are 269 505, 303 396, and 326 969 for the V2, V3b, and V3m retrievals, respectively. Compared with V2, the number of cloud observations increases by 12.6% and 21.3% for V3b and V3m, respectively. The largest increase in the number of clouds observed occurs at altitudes above 5 km.

Fig. 9.
Fig. 9.

Cloud-base height distributions for GSFC during 2012 for V2 (solid line), V3b (dashed line), and V3m (dashed–dotted line). Vertical axis bin size equals 1 km.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

Because the difference in the number of clouds retrieved shows a clear vertical dependence, we examine them specifically for three subsamples, by defining low clouds as those with base heights less than 2 km, high clouds as those with base heights greater than 5 km, and middle clouds as those with base heights between 2 and 5 km (WMO 1975). The number of lidar profiles for each classification, along with occurrence frequency, is shown in Table 4. Cloud occurrence frequency is defined as the number of lidar profiles containing a particular cloud classification divided by the total number of observable profiles. Regardless of the retrieval method (V2, V3b, and V3m), occurrence frequency is nearly identical for low clouds, which reflects the relative consistency in GCDM application between V2 and V3 at 1-min resolution. High clouds show the largest increase in occurrence frequency. For example, comparing the V2 and V3m algorithms, the occurrence frequency of high clouds increases by 5.3%, attributable to (i) the increased identification of elevated, multilayer cloud decks using an attenuated UCDM threshold; (ii) increased use of the UCDM to identify high clouds at day and night; and (iii) multitemporal application of UCDM to increase SNR. With respect to (i), V2 retrievals resulted in 91% of cloudy lidar profiles containing single-layer clouds. The percentage of single-layer clouds decreases to 83% and 81% for V3b and V3m, respectively.

Table 4.

Number of lidar profiles and occurrence frequency at GSFC in 2012. Note that because a single profile may contain multiple cloud layers, the sum of low, middle, and high clouds does not equal total clouds and may exceed 1.

Table 4.

b. Seasonal dependence

Figure 10 shows the annual cycle for low, middle, high, and total cloud classifications during 2012. The low-cloud occurrence frequency is nearly identical for all three retrieval methods. Middle clouds retrieved using V3b and V3m exhibit a slight separation from V2. The largest differences are again seen with high-cloud retrievals. While the annual cycles for high clouds show similar patterns for all three retrievals, there is an increase in occurrence frequency of 2.4% and 5.3% for V3b and V3m, respectively. The increase in high-cloud occurrence frequency when compared to V2 ranges from 1% to 4% using V3b and from 3% to 8% for V3m. The largest differences for high-cloud occurrence frequency between V2 and V3 occurs during summer months, which is coincident with the period when the sun is at its highest elevation and thus solar background is highest.

Fig. 10.
Fig. 10.

Annual cycle of the occurrence frequency for low, middle, high, and total clouds at GSFC during 2012. The solid line represents V2, the dashed line V3b, and the dashed–dotted line V3m. Three lines are displayed in all four panels. However, all three lines overlie one another for low clouds, and V3b and V3m are indistinguishable for middle clouds.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

c. Diurnal dependence

Differences in the diurnal cycle show similar characteristics as the annual cycles for low and middle clouds. As seen in Fig. 11, V2, V3b, and V3m are nearly identical for low clouds. While V3b and V3m show slight differences from V2, they are indistinguishable from each other. High-cloud diurnal cycles follow the same trends for all three retrievals. However, the cloud occurrence frequency is higher for V3b and highest for V3m. No clear diurnal trend is apparent at GSFC because some changes (e.g., using the UCDM at all times) affect both day and night retrievals. At tropical sites, where the solar background is higher and longer temporal averaging is necessary, there may be a more obvious diurnal trend.

Fig. 11.
Fig. 11.

Diurnal cycle of the occurrence frequency for low, middle, high, and total clouds at GSFC during 2012. The solid line represents V2, the dashed line V3b, and the dashed–dotted line V3m. Three lines are displayed in all four panels. However, all three lines overlie one another for low clouds, and V3b and V3m are indistinguishable for middle clouds.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

d. Geographical representativeness of high-cloud detection

The CALIPSO spacecraft, which was launched in April 2006, flies in a sun-synchronous polar orbit with a 16-day repeat cycle and crosses the equator at about 1330 local solar time (Winker et al. 2009). The primary instrument aboard CALIPSO is CALIOP, which allows for unobstructed, nearly global observations of high clouds, like cirrus. Direct comparisons between CALIPSO and ground-based lidars are challenging because they do not make coincident measurements. However, because of the large spatial extent of cirrus clouds that can range from tens to thousands of kilometers (Lynch and Sassen 2002; Massie et al. 2010), there should be some similarities in the high-cloud occurrence frequencies between CALIPSO and MPLNET.

CALIPSO level 2, version 3, 5-km cloud-layer products are used to determine the occurrence frequency for high clouds (base height > 5 km) with a cloud–aerosol discrimination (CAD) score between 70 and 100. Cloud bases measured from CALIPSO are generally less certain than the cloud tops due to attenuation within layers. However, base heights are used for consistency with ground-based lidar measurements. Comparisons are limited to CALIPSO profiles within 200 km of the MPLNET site and times within ±2 h of the nearest CALIPSO approach. Unlike CALIPSO, which measures atmospheric constituents beneath its orbital path, MPLNET observes only layers advected over the site. To produce equivalent measurement lengths, the mean wind speed above 5 km (determined from GEOS-5) is used to determine the time interval for the comparisons. For example, Fig. 12 shows a CALIPSO overpass on 7 July 2012 that was 180 km west of GSFC at its closest approach. The length of the CALIPSO track within the comparison limits (165 km) is divided by the mean wind speed in order to determine the corresponding MPLNET measurement time (205 min, compared to about 25 s for CALIPSO to travel the same distance). Both CALIPSO and MPLNET indicate the presence of high-cloud layers between 10 and 15 km (Fig. 12). V3 reports more high clouds, partly due to the use multitemporal resolutions, but also because of a change in the method used to determine profile attenuation.

Fig. 12.
Fig. 12.

Comparison of high-cloud observations by CALIPSO and the NASA GSFC MPLNET site 180 km away on 7 Jul 2012. (a) CALIPSO 532-nm total attenuated backscatter; (b) CALIPSO 5-km cloud mask; (c) MPLNET 527-nm NRB; and (d) MPLNET cloud mask for V2, V3b, and V3m. The dashed vertical lines indicate the location and time of the nearest overpass.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

As stated previously, low clouds and fog can obstruct the view of the highest clouds when using ground-based lidars. Therefore, an additional constraint is used to limit the CALIPSO–MPLNET comparison to profiles that are not completely attenuated, referred to as “transparent profiles.” CALIPSO profile attenuation is determined using the opacity flag, and MPLNET profile attenuation again uses the V2 method.

The results of the comparison are given in Table 5. There were 119 instances when a CALIPSO track came within 200 km of GSFC in 2012. The median distance between the two instruments at the closest approach was about 96 km. The occurrence frequencies of profiles with no clouds and high clouds are given relative to the number of transparent profiles. V3m shows better agreement with CALIPSO for occurrence frequencies of no clouds (0.691 and 0.624, respectively) and high clouds (0.235 and 0.332, respectively). Clouds are not only detected more frequently with the V3 algorithm but they are also reported at higher altitudes.

Table 5.

Comparison between MPLNET and CALIPSO cloud detection. Parentheses indicate diurnal observations (day/night).

Table 5.

Comparisons between CALIPSO and MPLNET were also performed at the Singapore MPLNET site (1.30°N, 103.78°E; 0.03 km MSL) in Southeast Asia, where clouds are ubiquitous (Reid et al. 2013) but the high solar background makes cloud detection more challenging. There were 77 instances when a CALIPSO track came within 200 km of the Singapore site in 2012, with a median distance of about 84 km. The percentage of transparent profiles is much lower for both instruments, which can be reasonably attributed to thicker clouds and deep convection found in the tropics. Similar to the comparison at GSFC, V3m is in better agreement with the CALIPSO occurrence frequencies. However, the daytime comparisons are negatively impacted by poor SNR. It should be noted that no changes to the V3 algorithm thresholds were necessary to perform detections at the Singapore MPLNET site, which is consistent with the robustness demonstrated by the V2 algorithm.

e. Macrophysical and optical cirrus properties

Based on the greater detection of high clouds demonstrated above, we characterize cirrus clouds over the GSFC site as detected by the V3 algorithm. We limit the analysis to cases when (i) only cirrus clouds (no underlying liquid water or mixed-phase clouds) were detected in the profile based on the −37°C temperature threshold; (ii) the estimated COD was less than 3, based on the upper-limit for cirrus clouds suggested by Sassen and Cho (1992); and (iii) the attenuation altitude was at least 2 km above the cloud top. The final constraint limits the analysis to “transparent cirrus” cases for which the algorithm is more likely to identify the true cloud top.

The resulting dataset includes 57 351 cirrus clouds. The majority of cloud detections (75%) occur at the base 1-min temporal resolution. The largest occurrence rate of the coarse temporal averages occurs at or near noon and during the summer months, when the solar background is highest.

Table 6 summarizes the seasonal and annual mean characteristics of the transparent cirrus dataset. The monthly variation in the macrophysical properties is shown in Fig. 13. Cirrus clouds over GSFC tend to be higher and thinner (geometrically and optically) in the spring and summer and lower and thicker in the fall and winter seasons. Cirrus clouds also occur more frequently in the spring and summer months. The transparent cirrus dataset is almost entirely (84%–96%) composed of subvisual (COD < 0.03) and thin (0.03 < COD < 0.3) cirrus clouds, where we have made use of the Sassen and Cho (1992) definitions based of the visible cloud color. By this definition, opaque cirrus have optical depths in the range of 0.3 < COD < 3. Uncertainties in the value of the extinction-to-backscatter ratio and cloud-top height could lead to an exaggeration of amount of subvisual and thin cirrus. However, Noël and Haeffelin (2007) similarly found that 90% of completely penetrated cirrus had COD < 0.3.

Table 6.

Transparent cirrus cloud properties.

Table 6.
Fig. 13.
Fig. 13.

(left) Monthly averaged cloud top (thick solid line), cloud base (thin solid line), and cloud depth (dashed red line). (right) Monthly mean cloud optical depth using an effective extinction-to-backscatter ratio of 20 sr (dashed line) and 30 sr (solid line) for the transparent cirrus dataset.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

Frequency distributions of the optical and macrophysical properties are presented in Fig. 14. Here, we show only results using S* = 30 sr. The cloud optical depth peaks in the subvisual range and has a positive skew. The transparent cirrus dataset suggests that the limit at which we are able to resolve molecular signal above cloud, and thus reliably determine the cloud top, occurs near a COD of 0.5–0.8 depending on the choice of S*.

Fig. 14.
Fig. 14.

(a) Cloud-base altitude, (b) cloud-top altitude, and (c) cloud depth for the transparent cirrus dataset. The colors indicate the cirrus type based on the estimated COD (blue for subvisual, yellow for thin, and red for opaque cirrus) using S* = 30 sr. Horizontal axis bin size is 0.5 km for cloud-base and cloud-top altitudes, and 0.25 km for cloud depth. (d) Frequency distribution (black) and cumulative frequency distribution (red). The vertical dashed lines indicate the thresholds for subvisual (0.03) and thin (0.3) cirrus clouds. Horizontal axis bin size equals 0.001.

Citation: Journal of Atmospheric and Oceanic Technology 33, 10; 10.1175/JTECH-D-15-0190.1

A comparison of daytime and nighttime cloud retrievals is provided in Table 7. There are only slight differences in the occurrence frequency between day and night cases. However, the geometric and optical depths are considerably lower in the daytime. The thinning of daytime cirrus may be attributable to difficulty in correctly identifying cloud boundaries due to solar background effects (Thorsen et al. 2013). In the same manner, the daytime retrievals are more likely to be considered as totally attenuated due to the higher solar background reducing the possibility to resolve molecular signal at cirrus heights.

Table 7.

Daytime and nighttime transparent cirrus properties.

Table 7.

7. Summary and discussion

A new cloud detection algorithm, version 3 (V3), has been developed within the NASA Micro Pulse Lidar Network (MPLNET) that uses a combination of retrieval methods and a multitemporal averaging scheme. Most V3 changes represent updates to the version 2 (V2) uncertainty-based threshold algorithm introduced by Campbell et al. (2008). The threshold used to identify cloud presence now accounts for attenuation losses within cloud layers, which allows for better estimation of cloud tops and boundaries of overlying cloud layers in profiles where multiple cloud layers are detected. A more synergistic merging of the gradient-based cloud detection method (GCDM) and uncertainty-based cloud detection method (UCDM) improves nighttime clouds detection of tenuous high clouds. The incorporation of coarser temporal resolutions at intermediate (5 min) and long (20 min) averages improves detection in situations with low SNR (e.g., high solar background). One year of data at the NASA Goddard Space Flight Center (GSFC) in Greenbelt, Maryland, is used to show the effect of these updates on cloud retrievals.

The largest impact of the changes to the cloud detection algorithm is evident with high clouds (those with a cloud base > 5 km), while the diurnal and annual cycles of low and middle clouds exhibit only slight changes from V2 to V3. The high-cloud occurrence frequency increases by over 5% at GSFC when using the V3 merged cloud scene compared with the V2 retrieval. The increase in high-cloud detection could have meaningful implications for studies of radiative forcing, as cirrus are known to both warm and cool the atmosphere depending on their varying physical properties. Furthermore, the ability to detect multilayered cloud scenes is improved with the V3 algorithm. The results show that 91% of clouds in 2012 at the NASA GSFC project site were recorded as single-layer clouds according to the V2 retrieval compared with 81% for V3.

In a brief comparison, V3 retrievals show closer agreement to CALIPSO in both high-cloud occurrence frequency and mean altitude compared to V2. While direct comparisons are difficult to perform between ground-based and satellite-based lidars due to their sampling differences, a careful analysis can be used to address geographical representativeness from the two observational viewpoints. Such a study is the topic of a future work focused on cirrus clouds because their large horizontal spatial extent should lead to relatively high correlations.

The highest and thinnest (both geometrically and optically) cirrus clouds are found during the spring and summer months, which was coincident with the highest cirrus occurrence frequency. There is no significant difference in occurrence frequency between daytime and nighttime retrievals. However, cirrus clouds are thinner (both geometrically and optically) in daytime than nighttime, which may be attributed to increased uncertainty due to the solar background effects. Notably, the limit to which we are able to resolve molecular signal above cirrus clouds occurs between cloud optical depths of 0.5 and 0.8, allowing for uncertainty in the extinction-to-backscatter ratio.

The value of the MPLNET cloud datasets is in its continuous (both day and night) and long-term measurements at polar, midlatitude, and tropical sites using a standard instrument and data processing algorithm. Incorporating the V3 cloud retrievals from MPLNET as part of a multi-instrument investigation will enhance our current knowledge of clouds, in particular cirrus. As it stands, the cloud products provide a unique validation dataset for the modeling community and satellite measurements. With some MPLNET sites now well into their second decade of continuous cloud and aerosol observations, the project has become an integral component of ground-based evaluation of atmospheric processes and verification of NASA satellite missions. This paper thus represents our continuing effort to optimize the fidelity of project datasets for the benefit of the community and in sustaining general scientific inquiry.

Acknowledgments

The authors acknowledge Larry Belcher for processing the V2 lidar data, and the MPLNET PIs and staff for their efforts in establishing and maintaining the GSFC and Singapore sites. The GEOS-5 meteorological data were provided by the NASA Global Modeling and Assimilation Office (GMAO) at GSFC. The NASA Micro Pulse Lidar Network is funded by the NASA Earth Observing System and the NASA Radiation Sciences Program. CALIPSO data were obtained from the NASA Langley Research Center Atmospheric Science Data Center. Author JRC acknowledges the support of NASA Interagency Agreement NNG13HH10I on behalf of MPLNET.

REFERENCES

  • Ackerman, S. A., Holz R. E. , Frey R. , Eloranta E. W. , Maddux B. C. , and McGill M. , 2008: Cloud detection with MODIS. Part II: Validation. J. Atmos. Oceanic Technol., 25, 10731086, doi:10.1175/2007JTECHA1053.1.

    • Search Google Scholar
    • Export Citation
  • Bevington, P. R., and Robinson D. K. , 2002: Data Analysis and Reduction for the Physical Sciences. 3rd ed. McGraw-Hill, 336 pp.

  • Bissonnette, L. R., 2005: Lidar and multiple scattering. Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, C. Weitkamp, Ed., Springer, 43–103.

  • Campbell, J. R., and Sassen K. , 2008: Polar stratospheric clouds at the South Pole from 5 years of continuous lidar data: Macrophysical, optical and thermodynamic properties. J. Geophys. Res., 113, D20204, doi:10.1029/2007JD009680.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Hlavka D. L. , Welton E. J. , Flynn C. J. , Turner D. D. , Spinhirne J. D. , Scott V. S. , and Hwang I. H. , 2002: Full-time, eye-safe cloud and aerosol lidar observation at Atmosphere Radiation Measurement program sites: Instrument and data processing. J. Atmos. Oceanic Technol., 19, 431442, doi:10.1175/1520-0426(2002)019<0431:FTESCA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Sassen K. , and Welton E. J. , 2008: Elevated cloud and aerosol layer retrievals from micropulse lidar signal profiles. J. Atmos. Oceanic Technol., 25, 685700, doi:10.1175/2007JTECHA1034.1.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Vaughan M. A. , Oo M. , Holz R. E. , Lewis J. R. , and Welton E. J. , 2015: Distinguishing cirrus cloud presence in autonomous lidar measurements. Atmos. Meas. Tech., 8, 435449, doi:10.5194/amt-8-435-2015.

    • Search Google Scholar
    • Export Citation
  • Chen, W.-N., Chiang C.-W. , and Nee J.-B. , 2002: Lidar ratio and depolarization ratio for cirrus clouds. Appl. Opt., 41, 64706476, doi:10.1364/AO.41.006470.

    • Search Google Scholar
    • Export Citation
  • Chew, B. N., Campbell J. R. , Reid J. S. , Giles D. M. , Welton E. J. , Salinas S. V. , and Liew S. C. , 2011: Tropical cirrus cloud contamination in sun photometer data. Atmos. Environ., 45, 67246731, doi:10.1016/j.atmosenv.2011.08.017.

    • Search Google Scholar
    • Export Citation
  • Cho, H.-M., Yang P. , Kattawar G. W. , Nasiri N. L. , Hu Y. , Minnis P. , Trepte C. , and Winker D. , 2008: Depolarization ratio and attenuated backscatter for nine cloud types: Analyses based on collocated CALIPSO lidar and MODIS measurements. Opt. Express, 16, 39313948, doi:10.1364/OE.16.003931.

    • Search Google Scholar
    • Export Citation
  • Clothiaux, E. E., Mace G. G. , Ackerman T. P. , Kane T. J. , Spinhirne J. D. , and Scott V. S. , 1998: An automated algorithm for detection of hydrometeor returns in micropulse lidar data. J. Atmos. Oceanic Technol., 15, 10351042, doi:10.1175/1520-0426(1998)015<1035:AAAFDO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • COESA, 1976: U.S. Standard Atmosphere, 1976. NOAA, 227 pp.

  • Comstock, J. M., Ackerman T. P. , and Mace G. G. , 2002: Ground-based lidar and radar remote sensing of tropical cirrus clouds at Nauru Island: Cloud statistics and radiative impacts. J. Geophys. Res., 107, 4714, doi:10.1029/2002JD002203.

    • Search Google Scholar
    • Export Citation
  • Congalton, R. G., and Mead R. A. , 1986: A review of three discrete multivariate analysis techniques used in assessing the accuracy of remotely sensed data from error matrices. IEEE Trans. Geosci. Remote Sens., GE-24, 169174, doi:10.1109/TGRS.1986.289546.

    • Search Google Scholar
    • Export Citation
  • Garnier, A., Pelon J. , Vaughan M. A. , Winker D. M. , Trepte C. R. , and Dubuisson P. , 2015: Lidar multiple scattering factors inferred from CALIPSO lidar and IIR retrievals of semi-transparent cirrus cloud optical depths over oceans. Atmos. Meas. Tech., 8, 27592774, doi:10.5194/amt-8-2759-2015.

    • Search Google Scholar
    • Export Citation
  • Hahn, C. J., and Warren S. G. , 1999: Extended edited synoptic cloud reports from ships and land stations over the globe, 1952–1996. ORNL/CDIAC-123, NDP-026C, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, 76 pp., doi:10.3334/CDIAC/cli.ndp026c.

  • Hahn, C. J., Warren S. G. , and London J. , 1996: Edited synoptic cloud reports from ships and land stations over the globe, 1982-1991. ORNL/CDIAC-77, NDP-026B, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, 45 pp., doi:10.3334/CDIAC/cli.ndp026b.

  • Hildebrand, F. B., 1987: Introduction to Numerical Analysis. 2nd ed. Dover Books, 704 pp.

  • Holz, R. E., Ackerman S. A. , Nagle F. W. , Frey R. , Dutcher S. , Kuehn R. E. , Vaughan M. A. , and Baum B. , 2008: Global Moderate Resolution Imaging Spectroradiometer (MODIS) cloud detection and height evaluation using CALIOP. J. Geophys. Res., 113, D00A19, doi:10.1029/2008JD009837.

    • Search Google Scholar
    • Export Citation
  • Huang, J., Hsu N. , Tsay S.-C. , Jeong M.-J. , Holben B. N. , Berkoff T. A. , and Welton E. J. , 2011: Susceptibility of aerosol optical thickness retrievals to thin cirrus contamination during the BASE-ASIA campaign. J. Geophys. Res., 116, D08214, doi:10.1029/2010JD014910.

    • Search Google Scholar
    • Export Citation
  • Huang, J., and Coauthors, 2012: Evaluations of cirrus contamination and screening in ground aerosol observations using collocated lidar systems. J. Geophys. Res., 117, D15204, doi:10.1029/2012JD017757.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Kato, S., and Coauthors, 2011: Improvements of top-of-atmosphere and surface irradiance computations with CALIPSO-, CloudSat-, and MODIS-derived cloud and aerosol properties. J. Geophys. Res., 116, D19209, doi:10.1029/2011JD016050.

    • Search Google Scholar
    • Export Citation
  • Lazarus, S. M., Krueger S. K. , and Mace G. G. , 2000: A cloud climatology of the Southern Great Plains ARM CART. J. Climate, 13, 17621775, doi:10.1175/1520-0442(2000)013<1762:ACCOTS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liou, K. N., 1986: Influence of cirrus clouds on weather and climate processes: A global perspective. Mon. Wea. Rev., 114, 11671199, doi:10.1175/1520-0493(1986)114<1167:IOCCOW>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liu, Z., Omar A. , Hu Y. , Vaughan M. A. , and Winker D. M. , 2005: Part 3: Scene classification algorithms. CALIOP Algorithm Theoretical Basis Doc., NASA Langley Research Center Tech Rep. PC-SCI-202, Release 1.0, 56 pp.

  • Lo, C., Comstock J. M. , and Flynn C. , 2006: An Atmospheric Radiation Measurement value-added product to retrieve optically thin visible optical depth using micropulse lidar. PNNL Tech. Rep. DOE/SC-ARM/TR-077, 11 pp.

  • Lolli, S., Welton E. J. , and Campbell J. R. , 2013: Evaluating light rain drop size estimates from multiwavelength Micropulse Lidar Network profiling. J. Atmos. Oceanic Technol., 30, 27982807, doi:10.1175/JTECH-D-13-00062.1.

    • Search Google Scholar
    • Export Citation
  • Lynch, D. K., and Sassen K. , 2002: Subvisual cirrus. Cirrus, D. K. Lynch et al., Eds., Oxford University Press, 256–264.

  • Marchand, R., Ackerman T. , Smyth M. , and Rossow W. B. , 2010: A review of cloud top height and optical depth histograms from MISR, ISCCP, and MODIS. J. Geophys. Res., 115, D16206, doi:10.1029/2009JD013422.

    • Search Google Scholar
    • Export Citation
  • Massie, S. T., Gille J. , Craig C. , Khosravi R. , Barnett J. , Read W. , and Winker D. , 2010: HIRDLS and CALIPSO observations of tropical cirrus. J. Geophys. Res., 115, D00H11, doi:10.1029/2009JD012100.

    • Search Google Scholar
    • Export Citation
  • Matthews, B. W., 1975: Comparison of the predicted and observed secondary structure of T4 phage lysozyme. Biochim. Biophys. Acta Protein Struct., 405, 442451, doi:10.1016/0005-2795(75)90109-9.

    • Search Google Scholar
    • Export Citation
  • Menzel, W. P., and Coauthors, 2008: MODIS global cloud-top pressure and amount estimation: Algorithm description and results. J. Appl. Meteor. Climatol., 47, 11751198, doi:10.1175/2007JAMC1705.1.

    • Search Google Scholar
    • Export Citation
  • Molod, A., Takacs L. , Suarez M. , Bacmeister J. , Song I.-S. , and Eichmann A. , 2012: The GEOS-5 atmospheric general circulation model: Mean climate and development from MERRA to Fortuna. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, NASA/TM-2012-104606, Vol. 28, NASA Goddard Space Flight Center, 117 pp.

  • Moran, K. P., Martner B. E. , Post M. J. , Kropfli R. A. , Welsh D. C. , and Widener K. B. , 1998: An unattended cloud-profiling radar for use in climate research. Bull. Amer. Meteor. Soc., 79, 443455, doi:10.1175/1520-0477(1998)079<0443:AUCPRF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Moroney, C., Davies R. , and Muller J.-P. , 2002: Operational retrieval of cloud-top heights using MISR data. IEEE Trans. Geosci. Remote Sens., 40, 15321540, doi:10.1109/TGRS.2002.801150.

    • Search Google Scholar
    • Export Citation
  • Noël, V., and Haeffelin M. , 2007: Midlatitude cirrus clouds and multiple tropopauses from a 2002–2006 climatology over the SIRTA observatory. J. Geophys. Res., 112, D13206, doi:10.1029/2006JD007753.

    • Search Google Scholar
    • Export Citation
  • Norris, J. R., 2000: What can cloud observations tell us about climate variability? Space Sci. Rev., 94, 375280, doi:10.1023/A:1026704314326.

    • Search Google Scholar
    • Export Citation
  • Omar, A. H., and Coauthors, 2009: The CALIPSO automated aerosol classification and lidar ratio selection algorithm. J. Atmos. Oceanic Technol., 26, 19942014, doi:10.1175/2009JTECHA1231.1.

    • Search Google Scholar
    • Export Citation
  • Pal, S. R., Steinbrecht W. , and Carswell A. I. , 1992: Automated method for lidar determination of cloud base height and vertical extent. Appl. Opt., 31, 14881494, doi:10.1364/AO.31.001488.

    • Search Google Scholar
    • Export Citation
  • Pinnick, R. G., Jennings S. G. , Chylek P. , Ham C. , and Grandy W. T. Jr., 1983: Backscatter and extinction in water clouds. J. Geophys. Res., 88, 67876796, doi:10.1029/JC088iC11p06787.

    • Search Google Scholar
    • Export Citation
  • Platt, C. M., 1973: Lidar and radiometric observations of cirrus clouds. J. Atmos. Sci., 30, 11911204, doi:10.1175/1520-0469(1973)030<1191:LAROOC>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Platt, C. M., 1979: Remote sounding of high clouds: I. Calculation of visible and infrared optical properties from lidar and radiometer measurement. J. Appl. Meteor., 18, 11301140, doi:10.1175/1520-0450(1979)018<1130:RSOHCI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Platt, C. M., and Coauthors, 1994: The Experimental Cloud Lidar Pilot Study (ECLIPS) for cloud–radiation research. Bull. Amer. Meteor. Soc., 75, 16351654, doi:10.1175/1520-0477(1994)075<1635:TECLPS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ramanathan, V., Cess R. D. , Harrison E. F. , Minnis P. , Barkstrom B. R. , Ahmad E. , and Hartmann D. , 1989: Cloud-radiative forcing and climate: Results from the Earth Radiation Budget Experiment. Science, 243, 5763, doi:10.1126/science.243.4887.57.

    • Search Google Scholar
    • Export Citation
  • Reid, J. S., and Coauthors, 2013: Observing and understanding the Southeast Asian aerosol system by remote sensing: An initial review and analysis for the Seven Southeast Asian Studies (7SEAS) program. Atmos. Res., 122, 403468, doi:10.1016/j.atmosres.2012.06.005.

    • Search Google Scholar
    • Export Citation
  • Rienecker, M. M., and Coauthors, 2008: The GEOS-5 data assimilation system—Documentation of versions 5.0.1, 5.1.0, and 5.2.0. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, NASA/TM-2008-104606, Vol. 27, NASA Goddard Space Flight Center, 101 pp.

  • Ringer, M., and Coauthors, 2006: Global mean cloud feedbacks in idealized climate change experiments. Geophys. Res. Lett., 33, L07718, doi:10.1029/2005GL025370.

    • Search Google Scholar
    • Export Citation
  • Rossow, W. B., and Schiffer R. A. , 1991: ISCCP cloud data products. Bull. Amer. Meteor. Soc., 72, 220, doi:10.1175/1520-0477(1991)072<0002:ICDP>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Rossow, W. B., and Schiffer R. A. , 1999: Advances in understanding clouds from ISCCP. Bull. Amer. Meteor. Soc., 80, 22612287, doi:10.1175/1520-0477(1999)080<2261:AIUCFI>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., and Cho B. S. , 1992: Subvisual-thin cirrus lidar dataset for satellite verification and climatological research. J. Appl. Meteor., 31, 12751285, doi:10.1175/1520-0450(1992)031<1275:STCLDF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., and Campbell J. R. , 2001: A midlatitude cirrus cloud climatology from the Facility for Atmospheric Remote Sensing. Part I: Macrophysical and synoptic properties. J. Atmos. Sci., 58, 481496, doi:10.1175/1520-0469(2001)058<0481:AMCCCF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., and Comstock J. M. , 2001: A midlatitude cirrus climatology from the Facility for Atmospheric Remote Sensing. Part III: Radiative properties. J. Atmos. Sci., 58, 21132127, doi:10.1175/1520-0469(2001)058<2113:AMCCCF>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., Wang Z. , and Liu D. , 2008: Global distribution of cirrus clouds from CloudSat/Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) measurements. J. Geophys. Res., 113, D00A12, doi:10.1029/2008JD009972.

    • Search Google Scholar
    • Export Citation
  • Schiffer, R. A., and Rossow W. B. , 1983: The International Satellite Cloud Climatology Project (ISCCP): The first project of the World Climate Research Programme. Bull. Amer. Meteor. Soc., 64, 779784.

    • Search Google Scholar
    • Export Citation
  • Shupe, M. D., Walden V. P. , Eloranta E. , Uttal T. , Campbell J. R. , Starkweather S. M. , and Shiobara M. , 2011: Clouds at Arctic atmospheric observatories. Part I: Occurrence and macrophysical properties. J. Appl. Meteor. Climatol., 50, 626644, doi:10.1175/2010JAMC2467.1.

    • Search Google Scholar
    • Export Citation
  • Stephens, G. L., 2005: Cloud feedbacks in the climate system: A critical review. J. Climate, 18, 237273, doi:10.1175/JCLI-3243.1.

  • Stephens, G. L., and Coauthors, 2002: The CloudSat mission and the A-Train: A new dimension of space-based observations of clouds and precipitation. Bull. Amer. Meteor. Soc., 83, 17711790, doi:10.1175/BAMS-83-12-1771.

    • Search Google Scholar
    • Export Citation
  • Thorsen, T. J., Fu Q. , and Comstock J. , 2011: Comparison of the CALIPSO satellite and ground-based observations of cirrus clouds at the ARM TWP sites. J. Geophys. Res., 116, D21203, doi:10.1029/2011JD015970.

    • Search Google Scholar
    • Export Citation
  • Thorsen, T. J., Fu Q. , Comstock J. M. , Sivaraman C. , Vaughan M. A. , Winker D. M. , and Turner D. , 2013: Macrophysical properties of tropical cirrus clouds from the CALIPSO satellite and from ground-based micropulse lidar and Raman lidars. J. Geophys. Res. Atmos., 118, 92099220, doi:10.1002/jgrd.50691.

    • Search Google Scholar
    • Export Citation
  • Vaughan, M., Winker D. , and Powell K. , 2005: Part 2: Feature detection and layer properties algorithms. CALIOP Algorithm Theoretical Basis Doc., NASA Langley Research Center Tech Rep. PC-SCI-202, Release 1.01, 87 pp.

  • Wang, Z., and Sassen K. , 2001: Cloud type and macrophysical property retrieval using multiple remote sensors. J. Appl. Meteor., 40, 16651682, doi:10.1175/1520-0450(2001)040<1665:CTAMPR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Warren, S. G., Hahn C. J. , and London J. , 1985: Simultaneous occurrence of different cloud types. J. Climate Appl. Meteor., 24, 658667, doi:10.1175/1520-0450(1985)024<0658:SOODCT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Welton, E. J., and Campbell J. R. , 2002: Micropulse lidar signals: Uncertainty analysis. J. Atmos. Oceanic Technol., 19, 20892094, doi:10.1175/1520-0426(2002)019<2089:MLSUA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Welton, E. J., Campbell J. R. , Spinhirne J. D. , and Scott V. S. , 2001: Global monitoring of clouds and aerosols using a network of micropulse lidar systems. Lidar Remote Sensing for Industry and Environmental Monitoring, U. N. Singh, T. Itabe, and N. Sugimoto, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 4153), 151–158, doi:10.1117/12.417040.

  • Wielicki, B. A., Cess R. D. , King M. D. , Randall D. A. , and Harrison E. F. , 1995: Mission to planet Earth: Role of clouds and radiation in climate. Bull. Amer. Meteor. Soc., 76, 21252153, doi:10.1175/1520-0477(1995)076<2125:MTPERO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Winker, D. M., and Vaughan M. A. , 1994: Vertical distribution of clouds over Hampton, Virginia observed by liar under the ECLIPS and FIRE ETO programs. Atmos. Res., 34, 117133, doi:10.1016/0169-8095(94)90084-1.

    • Search Google Scholar
    • Export Citation
  • Winker, D. M., Vaughan M. A. , Omar A. , Hu Y. , and Powell K. A. , 2009: Overview of the CALIPSO mission and CALIOP data processing algorithms. J. Atmos. Oceanic Technol., 26, 23102323, doi:10.1175/2009JTECHA1281.1.

    • Search Google Scholar
    • Export Citation
  • Winker, D. M., and Coauthors, 2010: The CALIPSO mission: A global 3D view of aerosols and clouds. Bull. Amer. Meteor. Soc., 91, 12111229, doi:10.1175/2010BAMS3009.1.

    • Search Google Scholar
    • Export Citation
  • WMO, 1975: International cloud atlas, volume I: Manual on the observation of clouds and other meteors. WMO Tech. Rep. 407, 155 pp.

  • WMO/WWRP, 2012: Recommended methods for evaluating cloud and related parameters. WMO Publ. WWRP 2012-1, 34 pp.

  • Yorks, J. E., Hlavka D. L. , Hart W. D. , and McGill M. J. , 2011: Statistics of cloud optical properties from airborne lidar measurements. J. Atmos. Oceanic Technol., 28, 869883, doi:10.1175/2011JTECHA1507.1.

    • Search Google Scholar
    • Export Citation
Save
  • Ackerman, S. A., Holz R. E. , Frey R. , Eloranta E. W. , Maddux B. C. , and McGill M. , 2008: Cloud detection with MODIS. Part II: Validation. J. Atmos. Oceanic Technol., 25, 10731086, doi:10.1175/2007JTECHA1053.1.

    • Search Google Scholar
    • Export Citation
  • Bevington, P. R., and Robinson D. K. , 2002: Data Analysis and Reduction for the Physical Sciences. 3rd ed. McGraw-Hill, 336 pp.

  • Bissonnette, L. R., 2005: Lidar and multiple scattering. Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, C. Weitkamp, Ed., Springer, 43–103.

  • Campbell, J. R., and Sassen K. , 2008: Polar stratospheric clouds at the South Pole from 5 years of continuous lidar data: Macrophysical, optical and thermodynamic properties. J. Geophys. Res., 113, D20204, doi:10.1029/2007JD009680.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Hlavka D. L. , Welton E. J. , Flynn C. J. , Turner D. D. , Spinhirne J. D. , Scott V. S. , and Hwang I. H. , 2002: Full-time, eye-safe cloud and aerosol lidar observation at Atmosphere Radiation Measurement program sites: Instrument and data processing. J. Atmos. Oceanic Technol., 19, 431442, doi:10.1175/1520-0426(2002)019<0431:FTESCA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Sassen K. , and Welton E. J. , 2008: Elevated cloud and aerosol layer retrievals from micropulse lidar signal profiles. J. Atmos. Oceanic Technol., 25, 685700, doi:10.1175/2007JTECHA1034.1.

    • Search Google Scholar
    • Export Citation
  • Campbell, J. R., Vaughan M. A. , Oo M. , Holz R. E. , Lewis J. R. , and Welton E. J. , 2015: Distinguishing cirrus cloud presence in autonomous lidar measurements. Atmos. Meas. Tech., 8, 435449, doi:10.5194/amt-8-435-2015.

    • Search Google Scholar
    • Export Citation
  • Chen, W.-N., Chiang C.-W. , and Nee J.-B. , 2002: Lidar ratio and depolarization ratio for cirrus clouds. Appl. Opt., 41, 64706476, doi:10.1364/AO.41.006470.

    • Search Google Scholar
    • Export Citation
  • Chew, B. N., Campbell J. R. , Reid J. S. , Giles D. M. , Welton E. J. , Salinas S. V. , and Liew S. C. , 2011: Tropical cirrus cloud contamination in sun photometer data. Atmos. Environ., 45, 67246731, doi:10.1016/j.atmosenv.2011.08.017.

    • Search Google Scholar
    • Export Citation
  • Cho, H.-M., Yang P. , Kattawar G. W. , Nasiri N. L. , Hu Y. , Minnis P. , Trepte C. , and Winker D. , 2008: Depolarization ratio and attenuated backscatter for nine cloud types: Analyses based on collocated CALIPSO lidar and MODIS measurements. Opt. Express, 16, 39313948, doi:10.1364/OE.16.003931.

    • Search Google Scholar
    • Export Citation
  • Clothiaux, E. E., Mace G. G. , Ackerman T. P. , Kane T. J. , Spinhirne J. D. , and Scott V. S. , 1998: An automated algorithm for detection of hydrometeor returns in micropulse lidar data. J. Atmos. Oceanic Technol., 15, 10351042, doi:10.1175/1520-0426(1998)015<1035:AAAFDO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • COESA, 1976: U.S. Standard Atmosphere, 1976. NOAA, 227 pp.

  • Comstock, J. M., Ackerman T. P. , and Mace G. G. , 2002: Ground-based lidar and radar remote sensing of tropical cirrus clouds at Nauru Island: Cloud statistics and radiative impacts. J. Geophys. Res., 107, 4714, doi:10.1029/2002JD002203.

    • Search Google Scholar
    • Export Citation
  • Congalton, R. G., and Mead R. A. , 1986: A review of three discrete multivariate analysis techniques used in assessing the accuracy of remotely sensed data from error matrices. IEEE Trans. Geosci. Remote Sens., GE-24, 169174, doi:10.1109/TGRS.1986.289546.

    • Search Google Scholar
    • Export Citation
  • Garnier, A., Pelon J. , Vaughan M. A. , Winker D. M. , Trepte C. R. , and Dubuisson P. , 2015: Lidar multiple scattering factors inferred from CALIPSO lidar and IIR retrievals of semi-transparent cirrus cloud optical depths over oceans. Atmos. Meas. Tech., 8, 27592774, doi:10.5194/amt-8-2759-2015.

    • Search Google Scholar
    • Export Citation
  • Hahn, C. J., and Warren S. G. , 1999: Extended edited synoptic cloud reports from ships and land stations over the globe, 1952–1996. ORNL/CDIAC-123, NDP-026C, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, 76 pp., doi:10.3334/CDIAC/cli.ndp026c.

  • Hahn, C. J., Warren S. G. , and London J. , 1996: Edited synoptic cloud reports from ships and land stations over the globe, 1982-1991. ORNL/CDIAC-77, NDP-026B, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, 45 pp., doi:10.3334/CDIAC/cli.ndp026b.

  • Hildebrand, F. B., 1987: Introduction to Numerical Analysis. 2nd ed. Dover Books, 704 pp.

  • Holz, R. E., Ackerman S. A. , Nagle F. W. , Frey R. , Dutcher S. , Kuehn R. E. , Vaughan M. A. , and Baum B. , 2008: Global Moderate Resolution Imaging Spectroradiometer (MODIS) cloud detection and height evaluation using CALIOP. J. Geophys. Res., 113, D00A19, doi:10.1029/2008JD009837.

    • Search Google Scholar
    • Export Citation
  • Huang, J., Hsu N. , Tsay S.-C. , Jeong M.-J. , Holben B. N. , Berkoff T. A. , and Welton E. J. , 2011: Susceptibility of aerosol optical thickness retrievals to thin cirrus contamination during the BASE-ASIA campaign. J. Geophys. Res., 116, D08214, doi:10.1029/2010JD014910.

    • Search Google Scholar
    • Export Citation
  • Huang, J., and Coauthors, 2012: Evaluations of cirrus contamination and screening in ground aerosol observations using collocated lidar systems. J. Geophys. Res., 117, D15204, doi:10.1029/2012JD017757.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437471, doi:10.1175/1520-0477(1996)077<0437:TNYRP>2.0.CO;2.