Abstract

To improve understanding of the convective processes key to the Madden–Julian oscillation (MJO) initiation, the Dynamics of the MJO (DYNAMO) and the Atmospheric Radiation Measurement Program (ARM) MJO Investigation Experiment (AMIE) collected 4 months of observations from three radars—the S-band dual-polarization Doppler radar (S-Pol), the C-band Shared Mobile Atmospheric Research and Teaching Radar (SMART-R), and Ka-band ARM zenith radar (KAZR)—along with radiosonde and comprehensive surface meteorological instruments on Addu Atoll, Maldives, in the tropical Indian Ocean. One DYNAMO/AMIE hypothesis suggests that the evolution of shallow and congestus cloud populations is essential to the initiation of the MJO. This study focuses on evaluating the ability of these three radars to document the full spectrum of cloud populations and to construct a merged cloud–precipitation radar dataset that can be used to test this hypothesis. Comparisons between collocated observations from the three radars show that KAZR provides the only reliable estimate of shallow clouds, while S-Pol/SMART-R can reasonably detect congestus within the 30–50-km range in addition to precipitating deep clouds. On the other hand, KAZR underestimates cloud-top heights due to rainfall attenuation in ~34% of the precipitating clouds, and an empirical method to correct KAZR cloud-top height bias is proposed. Finally, a merged KAZR–S-Pol dataset is produced to provide improved cloud-top height estimates, total hydrometeor microphysics, and radiative heating rate retrievals. With this dataset the full spectrum of tropical convective clouds during DYNAMO/AMIE can be reliably constructed and, together with complimentary radiosonde data, it can be used to study the role of shallow and congestus clouds in the initiation of the MJO.

1. Introduction

The Madden–Julian oscillation (MJO; Madden and Julian 1971, 1972) is the most prominent intraseasonal (20–100 days) oscillation in the tropics. As it propagates eastward at about 5 m s−1 (Weickmann et al. 1985) from the Indian Ocean to the western and central Pacific during the boreal winter, the MJO interacts with many weather and climate systems both in the tropics and midlatitudes. For example, it affects the strength and intraseasonal variability of the circulation and rainfall associated with the Asian summer monsoon (Annamalai and Slingo 2001), Australian summer monsoon (Hendon and Liebmann 1990), and North American monsoon (Lorenz and Hartmann 2006). The MJO also interacts with the North Atlantic Oscillation (Lin et al. 2009), the Arctic Oscillation (Zhou and Miller 2005), and the Antarctic Oscillation (Carvalho et al. 2005).

Global climate models generally underestimate the strength and intraseasonal variability of the MJO (Zhang et al. 2006), while some models fail to produce even half of the observed MJO variance (Lin et al. 2006). The prediction skill is particularly low over the Indian Ocean during the MJO initiation phase, which is partly due to the lack of detailed in situ observations of key physical processes in this region (Schott and McCreary 2001; Zhang et al. 2013). To address this issue, a set of coordinated field experiments was conducted in the equatorial Indian Ocean in late 2011–early 2012 to collect in situ and remote sensing observations for the purpose of improving understanding of the MJO initiation processes and improving MJO prediction [see Yoneyama et al. (2013) for an overview of the experiment]. These experiments included the overarching international Cooperative Indian Ocean Experiment on Intraseasonal Variability in Year 2011 (CINDY2011) and the U.S.-sponsored component Dynamics of the MJO (DYNAMO) and Atmospheric Radiation Measurement Program (ARM) MJO Investigation Experiment (AMIE) campaigns. In this study, we focus on observations from the DYNAMO/AMIE campaigns.

One of the key hypotheses proposed in the DYNAMO/AMIE field campaign is that different phases of the MJO are characterized by different cloud populations and that a specific cloud population is essential to the initiation of the MJO (Yoneyama et al. 2013). It is important to determine how the statistics of the cloud population evolve from suppressed MJO phases, consisting of shallow to moderate isolated convection, to active conditions, characterized by ubiquitous, deep, and moist convection. Testing this hypothesis requires documenting the full spectrum of convective clouds, from shallow and congestus clouds to deep convection. To address this need, three radar systems with different wavelengths (S, C, and Ka bands) were deployed at a “supersite” on Addu Atoll, Maldives. Of particular importance is understanding the respective roles of shallow and congestus clouds in moistening the lower troposphere, which has been suggested to be important to the initiation of the MJO (Johnson et al. 1999; Kemball-Cook and Weare 2001), and in inducing low-level large-scale moisture convergence (by providing low-level latent heating when they precipitate), which helps convective development of the MJO (Zhang and Hagos 2009; Yoneyama et al. 2013).

The main objectives of this paper are to evaluate the ability of the cloud and precipitation radars deployed at Addu Atoll to document the full spectrum of cloud populations and to take advantage of the multiwavelength radar platforms to produce a merged cloud–precipitation dataset (Feng et al. 2009) with continuous hydrometeor microphysics and radiative heating profile retrievals that can be used to test the DYNAMO science hypothesis and to evaluate numerical model simulations. The three radar systems that were deployed at Addu Atoll are the ARM Mobile Facility (AMF) vertically pointing Ka ARM zenith radar (KAZR; Ka band), the Texas A&M University Shared Mobile Atmospheric Research and Teaching Radar (SMART-R; C band), and the National Center for Atmospheric Research (NCAR) S-band dual-polarization Doppler radar (S-Pol; S and Ka bands). The two scanning precipitation radars (SMART-R and S-Pol) provide the capability to detect three-dimensional precipitating cloud structures that are also essential to document the degree of convective organization (Yoneyama et al. 2013). However, detection of nonprecipitating clouds, particularly shallow clouds, is challenging for scanning radars due to many factors, such as ground clutter, low-level moisture gradient, and nonweather signals (e.g., Miller et al. 1998). Therefore, systematic evaluation of the accuracy in detecting shallow to congestus clouds from scanning radars is necessary before using these data to examine the roles of shallow and congestus clouds in different life cycles of the MJO.

During DYNAMO/AMIE, both the S-Pol and SMART-R performed regular range–height indicator (RHI) scans over the AMF site where the KAZR was located. Therefore, collocated measurements from all three radars at the AMF site are available during the field campaign. Comparisons between statistics of various types of clouds, particularly shallow, congestus, and deep convective clouds that are potentially important to the initiation of MJO, will be performed. Through this comparison, the ability of the two scanning precipitation radars in detecting both precipitating and nonprecipitating clouds will be evaluated, and recommendations of how to use the comprehensive suite of radar data collectively to address the DYNAMO science goal related to the initiation of the MJO will be provided.

The paper is organized as follows. Section 2 describes the three radar systems and other ancillary data used in this study, and the collocation of data and quality control; section 3 provides cloud statistics and comparison results; section 4 describes a procedure to produce a merged cloud–precipitation dataset for cloud microphysics and radiative heating rate retrievals; and a summary and conclusions are given in section 5.

2. Data and methodology

DYNAMO/AMIE successfully completed operations from October 2011 to February 2012 as an integrated part of the overarching multinational coordinated CINDY2011 field program. The period of study in this paper is from 11 October 2011 to 15 January 2012, during which the AMF KAZR, NCAR S-Pol, and Texas A&M SMART-R were simultaneously operated.

Figure 1 shows the geographic locations of the three radars on Addu Atoll. The AMF KAZR is located at the Gan Island airport next to the Gan Maldives Meteorological Service office (00°41.251′S, 73°09.00′E), S-Pol is located on the Wharf site (00°37.826′S, 73°06.175′E), and SMART-R is located on the Spit site (00°36.453′S, 73°05.748′E). The distance between S-Pol (SMART-R) and the AMF KAZR is approximately 8.5 km (11 km). A brief description of the radar systems and their operating characteristics during the DYNAMO/AMIE field campaign is provided in the following sections.

Fig. 1.

Geographic locations of the two scanning precipitation radars with respect to the AMF site at Gan Island, Maldives, during the DYNAMO/AMIE experiment. The distance from S-Pol (SMART-R) to AMF is approximately 9 km (11 km), with an azimuth angle of 140° (147°). (Image © 2012 Google and © 2012 DigitalGlobe.)

Fig. 1.

Geographic locations of the two scanning precipitation radars with respect to the AMF site at Gan Island, Maldives, during the DYNAMO/AMIE experiment. The distance from S-Pol (SMART-R) to AMF is approximately 9 km (11 km), with an azimuth angle of 140° (147°). (Image © 2012 Google and © 2012 DigitalGlobe.)

In addition to the three radar systems deployed at Addu Atoll, radiosondes were launched at Gan Island every 3 h as part of the larger sounding array for DYNAMO. A suite of surface instruments, including a micropulse lidar, wind profiler, microwave radiometer, ceilometer, rain gauge, total sky imager, surface meteorology instruments, and suite of up- and downwelling radiation measurements were also deployed at the AMF on Gan Island (see Yoneyama et al. 2013 for more details).

a. AMF KAZR

The AMF KAZR is a zenith-pointing Doppler cloud radar operated at 35 GHz (8-mm wavelength). The main purpose of the KAZR is to provide vertical profiles of clouds by measuring the first three Doppler moments: reflectivity, radial Doppler velocity, and spectral width (Bharadwaj et al. 2013) as well as the full Doppler spectrum though that is not used in this analysis. The KAZR used during DYNAMO/AMIE has an antenna diameter of 2 m, with a 0.32° beamwidth. Two simultaneous operating modes (Bharadwaj and Chandrasekar 2012) were used during the field campaign: general mode (detects full range but less sensitive) and cirrus mode (more sensitive but does not detect clouds below 2-km height due to pulse compression techniques), with a maximum range of about 18 km. The single-pulse minimum observable reflectivity is −19.7 dBZ at 1 km and −15.7 dBZ at 10 km for general and cirrus modes, respectively. Spectral processing is used to enhance the sensitivity of the radar by performing an equivalent coherent integration in the spectral domain, which adds approximately 20 dBZ sensitivity to KAZR (i.e., KAZR data have a sensitivity of approximately −40 dBZ at 1 km and approximately −36 dBZ at 10 km for general and cirrus modes, respectively).

The KAZR data used in this study are the KAZR Active Remote Sensing of Clouds (ARSCL) product produced by ARM (www.arm.gov). KAZR-ARSCL corrects for water vapor attenuation and velocity aliasing and produces a significant detection mask. By selecting the mode with the highest signal-to-noise ratio at a given point, data from the two simultaneous operating modes are combined for each profile to obtain the best-estimate time–height fields of the three radar moments (i.e., reflectivity, Doppler velocity, and spectral width). The KAZR-ARSCL product has vertical and temporal resolution of 30 m and 4 s, respectively. Although KAZR-ARSCL provides cloud boundaries that are derived from a combination of KAZR measurements and observations from the micropulse lidar and ceilometer, they are not used for comparison purpose in this study because reflectivity measurements from S-Pol and SMART-R are essentially “hydrometeor” echoes that do not separate between cloud and rain droplets. Instead, hydrometeor echo boundaries are derived using KAZR-ARSCL reflectivity measurements that are designated as “significant detection,” which are defined as signal-to-noise ratio above −18 dB and reflectivity values greater than −40 dBZ.

Several other auxiliary datasets at the AMF site are also used in this study. Surface precipitation rate is measured from the Vaisala acoustic rain sensor as part of the Present Weather Detector. Column-integrated liquid water path (LWP) is retrieved from a two-channel microwave radiometer (MWR) (Turner et al. 2007). Temperature and relative humidity profiles are obtained from sounding data. Attenuated backscatter profiles are obtained from the micropulse lidar data.

b. NCAR S-Pol

The NCAR S-Pol radar is a dual-polarimetric, dual-frequency (10 cm: S band, and 8 mm: Ka band) weather research radar (Keeler et al. 2000). Becaue of its dual-polarimetric capability, the main purpose of the S-Pol during DYNAMO/AMIE was to obtain high-quality precipitation maps and to provide detailed microphysical measurements in clouds and precipitation. The NCAR Ka-band radar system that scanned concurrently with the S-band radar was designed to match the S-band beamwidth of 1° to enable dual-wavelength (Ka and S bands) retrievals of water vapor (Ellis and Vivekanandan 2010) and liquid water content (Ellis and Vivekanandan 2011). However, the 1° beam design results in lower antenna gain and reduced sensitivity compared to the narrower KAZR beam. Therefore, we do not utilize the Ka-band measurements from the S-Pol system for comparing cloud detection with KAZR in this study. The pulse repetition frequency (PRF) and peak transmitter power of the S-Pol are 0–1300 Hz and 1 MW, respectively. The S-Pol has an 8.5-m antenna with a 0.91° beamwidth and a minimum detectable reflectivity of about −24 and −10 dBZ at a distance of 10 and 50 km, respectively, for a single pulse. The S-Pol was operated with a range resolution of 150 m, and a maximum observed range of 150 km. S-Pol is capable of transmitting the horizontal and vertical polarization waves in fast alternating or simultaneous mode while receiving both. During DYNAMO S-Pol operated in fast alternating transmit mode.

During the DYNAMO/AMIE field campaign, S-Pol operated on a 15-min scanning cycle, starting at minute 0, 15, 30, and 45 of every hour. The scanning sequence started with two RHI scans at 141° and 142° azimuth over the AMF site, with oversampled elevation angles of 0.5° up to 65° (Fig. 2a), followed by a 360° plan position indicator (PPI) surveillance volume scan consisting of eight elevation angles between 0.5° and 11°; and finally, two RHI sector scans were performed between the azimuth angles of 4°–82° and 114°–140° with a 2° azimuthal resolution up to an elevation angle of 40°. These two sectors were selected because of the low-level blockage that occurred from the deep-water port at the azimuths in between the sectors. The S-Pol data quality control includes automated ground clutter removal (Hubbert et al. 2009a,b) and correction of noise based on a beam-by-beam varying noise power algorithm (Dixon and Hubbert 2012). The calibration of S-Pol was verified and monitored continuously using daily solar scans, injecting a known power source into the receiver for every azimuth of data collected. Additionally, the differential reflectivity was calibrated using vertically pointing data in light rain, which is the most accurate and reliable method. The dual-polarimetric self-consistency calibration procedure described by Vivekanandan et al. (2003) was performed on data from various times throughout the project. The accuracy of the Vivekanandan et al. (2003) calibration method is about ±0.25 dB; thus, the S-Pol reflectivity is considered to be calibrated to better than 1 dB. There were also routine checks to ensure the pedestal was level and the pointing and ranging were accurate.

Fig. 2.

Beam elevation angles as a function of ground range for the RHI scans of (a) S-Pol and (b) SMART-R. The relative location of AMF KAZR is marked by the thick black line. The actual elevation angle resolution from S-Pol and SMART-R are approximately 0.5° and 0.2°, respectively, and only the 0.5° lines are plotted and the 1° lines are labeled for clarity.

Fig. 2.

Beam elevation angles as a function of ground range for the RHI scans of (a) S-Pol and (b) SMART-R. The relative location of AMF KAZR is marked by the thick black line. The actual elevation angle resolution from S-Pol and SMART-R are approximately 0.5° and 0.2°, respectively, and only the 0.5° lines are plotted and the 1° lines are labeled for clarity.

c. Texas A&M SMART-R

The Texas A&M SMART-R is a 5-cm (C band), single-polarimetric mobile Doppler radar (Biggerstaff et al. 2005) mounted on a diesel flatbed truck. The main purpose of SMART-R was to document the three-dimensional structure of precipitation echoes, detailed rainfall patterns, and Doppler measurements of air motions on the convective and mesoscales. The PRF and peak transmitter power of SMART-R are 300–3000 Hz and 250 kW, respectively. The SMART-R has a 2.54-m antenna with a 1.5° beamwidth and a minimum detectable reflectivity of about −21 and −7 dBZ at a distance of 10 and 50 km, respectively, for a single pulse.

During the DYNAMO/AMIE field campaign, the SMART-R operated on a 10-min scan cycle, starting at minute 0, 10, 20, … of every hour, for six total scan cycles per hour. Three RHI scans out to 100 km were directed over the AMF site at the beginning of each 10-min cycle. At the start of the campaign, these RHIs were at azimuth angles 146°, 147°, and 148°, with elevation angles up to 60° (Fig. 2b). At 1030 UTC 21 November 2011, the set of three RHI scans were changed to 0°, 90°, and 147° (directed at AMF KAZR) azimuth angles. The reasons for the change are such that high-resolution vertical snapshots through the precipitating systems in other directions besides the AMF can be obtained, and that the center azimuth angle of 147° RHI scan was determined to be appropriate for comparisons between the two radars. Following the RHI scans, a surveillance scan was performed at a 0.5° elevation out to 300 km. At about 2 and 6 min after the beginning of each 10-min cycle, a full volume scan of 13 elevation angles (0.5°–33°) was performed out to 150 km; these two scans were then interleaved to create a single merged volume scan with 25 elevation angles for each 10-min period. SMART-R calibration and quality control procedures are further described in Fliegel and Schumacher (2012).

d. Collocating measurements at AMF

To compare reflectivity measurements between the three radars, the column above the AMF site was extracted from the S-Pol and SMART-R RHI data for each time they performed an RHI scan over the AMF. Time–height series of the S- and C-band radar measurements were then reconstructed to match the KAZR data format. The KAZR-ARSCL data were averaged to a 90-m (three KAZR gates) and 30-s (about every seven profiles) temporal and vertical resolution grid, respectively, such that comparisons with the coarser vertical resolution and the RHI scan time from the two scanning radars are more reasonable than using their respective native resolutions. S-Pol and SMART-R data were linearly interpolated to match the 90-m KAZR vertical grid.

A hydrometeor echo layer is defined in this study such that at least three continuous vertical grids (i.e., 270-m echo thickness) have significant echo detection (using a minimum reflectivity threshold of −40, −30, and −30 dBZ for KAZR, S-Pol, and SMART-R, respectively). The top and bottom heights of the hydrometeor layer are then defined as the echo top and base height, respectively. Because the main focus of this paper is to evaluate the ability of the three radar systems to document the cloud populations that are key to MJO initiation—namely, shallow, congestus, and deep convective clouds—cloud echoes that are thinner than 270 m are not included in the comparison in the next section. We note that excluding these echoes in the comparison will likely remove thin tropical shallow clouds, cirrus clouds (Riihimaki and McFarlane 2010), and midlevel clouds (Riihimaki et al. 2012), but those clouds are not the focus of the radar comparison. The merged radar dataset described in section 4 will include all clouds, including those thin ones that are excluded in the comparison.

Figure 3 shows an example of the collocated KAZR, S-Pol, and SMART-R data on 28 October 2011 at the AMF site. The quality-controlled radar reflectivities from S-Pol and SMART-R are shown in Figs. 3b,c, respectively, and their raw reflectivities are shown in Figs. 3e,f, respectively. Several issues are apparent in the S-Pol and SMART-R time–height series reconstructed data and they are discussed in the following.

Fig. 3.

Example of collocated time–height radar reflectivity from KAZR, S-Pol, and SMART-R on 28 Oct 2011 at the AMF site: (a) KAZR-ARSCL reflectivity, (b) S-Pol quality-controlled reflectivity, (c) SMART-R quality-controlled reflectivity, (d) surface rain rate from rain gauge and retrieved column liquid water path from microwave radiometer, (e) S-Pol raw reflectivity, and (f) SMART-R raw reflectivity.

Fig. 3.

Example of collocated time–height radar reflectivity from KAZR, S-Pol, and SMART-R on 28 Oct 2011 at the AMF site: (a) KAZR-ARSCL reflectivity, (b) S-Pol quality-controlled reflectivity, (c) SMART-R quality-controlled reflectivity, (d) surface rain rate from rain gauge and retrieved column liquid water path from microwave radiometer, (e) S-Pol raw reflectivity, and (f) SMART-R raw reflectivity.

Data from S-Pol (Figs. 3b,e) show excellent sensitivity of the radar in detecting a wide range of hydrometeors from precipitating to nonprecipitating clouds. For example, most of the cirrus and anvil clouds between 0000 and 1000 UTC were detected by S-Pol, and the cloud-top heights of the deep convective clouds between 1100 and 1700 UTC agree well with those from the KAZR. However, many of the raw low-level reflectivities below 4 km from S-Pol (Fig. 3e) do not correspond to any echo from KAZR and an echo striation at around 7-km height at the AMF site is commonly seen in the data (e.g., between 0000 and 1000 UTC). These features are likely due to ground clutter that has not been completely removed and/or Bragg scattering (i.e., clear-air return). For quality control, the raw S-Pol data were removed if the signal-to-noise ratio of the horizontal copolar reflectivity channel (SNRHC) was below 8, 0, and 15 dB for 0–6-, 6–9.5-, 9.5–20-km height, respectively. These values are determined through subjective comparison with collocated KAZR data to remove as much nonmeteorological echoes and retain as much real hydrometeor echoes as possible. Echo boundaries were then defined using the filtered reflectivity. To remove the nonmeteorological striation returns, a higher SNRHC of 15 dB was then applied to the height level between 6.5 and 7.5 km, if no contiguous layer with echo base/top below/above these two heights was defined. The resulting quality-controlled S-Pol data (Fig. 3b) show much better agreement with KAZR, except for echoes below 2 km (0200–1100 UTC), which are probably caused by clear-air echoes from Bragg scattering and/or insects.

Data from the SMART-R (Figs. 3c,f) show considerably less sensitivity to nonprecipitating clouds compared to S-Pol and KAZR. This is expected due to the much lower transmitting power and wider beamwidth of the SMART-R. Ground clutter is strongest below 1.5 km and significant echo striations are seen below 4 km and at around 8 km. Data were removed below a threshold value of a signal quality index (SQI) of 0.2 (0.05) below (above) 9.5 km to quality control the reflectivity data. SQI is a measure of coherence of Doppler power in the linear channel and higher SQI thresholds help remove second-trip echoes, since the signal is less coherent. Similar to the S-Pol, these height values were determined through subjective visual comparison between SMART-R and KAZR data to remove as much nonmeteorological echoes and retain as much real hydrometeor echoes as possible. Additionally, reflectivities below 1.5 km that have Doppler velocity values between −0.5 and 0.5 m s−1 were filtered to remove ground clutter. Echo striations below 9.5 km that remained after this filtering were removed if their thickness was less than 0.5 km. The resulting quality-controlled SMART-R data (Fig. 3c) show that most of the reflectivities below 1.5 km and low-level echo striations have been removed.

In the next section, we compare these collocated, quality-controlled S-Pol and SMART-R reflectivity data with measurements from the KAZR to characterize the hydrometeor detection capabilities for various types of clouds during the DYNAMO/AMIE field campaign.

3. Cloud statistics results

a. Overall hydrometeor statistics

Three MJO events were observed at Addu Atoll during the 4-month-long DYNAMO/AMIE field campaign. The time–height series of vertical hydrometeor frequency observed by the KAZR is shown in Fig. 4a. The frequency is defined by the ratio of the number of times with significant hydrometeor detection to the total number of observed times (12 h) at each 90-m vertical level. The first two MJO events (10 October–1 November 2011, 10 November–1 December 2011) were slightly stronger than the third event (7–22 December 2011) with higher frequency of deep convective clouds. The period after 1 January 2012 until the end of the field campaign was during the suppressed phase of the MJO in the Indian Ocean, where most of the observed clouds were nonprecipitating shallow cumulus and cirrus clouds.

Fig. 4.

(a) KAZR-observed time series of vertical hydrometeor frequency during the DYNAMO/AMIE experiment (10 Oct 2011–8 Feb 2012) and (b) averaged profiles of hydrometeor and echo boundary frequencies. Frequency is defined as number of cloudy instances at each height within a time period. The temporal resolution shown in (a) is 12 h, and the vertical resolution shown in both panels is 90 m.

Fig. 4.

(a) KAZR-observed time series of vertical hydrometeor frequency during the DYNAMO/AMIE experiment (10 Oct 2011–8 Feb 2012) and (b) averaged profiles of hydrometeor and echo boundary frequencies. Frequency is defined as number of cloudy instances at each height within a time period. The temporal resolution shown in (a) is 12 h, and the vertical resolution shown in both panels is 90 m.

Figure 4b shows the averaged vertical frequency of hydrometeors at each 90-m level and their base and top heights. The mean frequency of hydrometeors (black line) is 0.13 with a prominent peak of high clouds between 10 and 12 km. Three distinct peaks are seen in the cloud-top frequency (red line), corresponding to shallow clouds (1 km), midlevel clouds (5–6 km, approximately at 0°C), and high clouds (13 km). This vertical distribution of cloud-top height frequency is consistent with previous studies in the tropical western Pacific region (Johnson et al. 1999; McFarlane et al. 2007; Okamoto et al. 2008). The hydrometeor base frequency (green line) includes both cloud and precipitation, with corresponding peaks around or below the three cloud-top heights mentioned above. The peak near the surface is a result of precipitating clouds with echo extending down to the lowest KAZR range gate. The sharp peak at 2 km, however, is due to the operating mode switch of the KAZR from general to cirrus mode at that level. As mentioned in section 2, the cirrus mode provides higher sensitivity than the general mode; thus, more clouds can be detected at and above this height.

b. Hydrometeor detection from all radars by cloud type

To evaluate the hydrometeor detection capabilities from the two scanning precipitating radars, hydrometeors are separated into precipitating and nonprecipitating clouds, each of which are subdivided into three categories (precipitating: shallow, congestus, deep clouds; nonprecipitating: midlevel, cirrus, anvil clouds) based on their echo base and top heights as defined in Table 1. The low- and midlevel heights are selected where KAZR-observed cloud-top frequencies are at their local minimum (Fig. 4b). An example of each cloud type observed by KAZR is given in Fig. 5.

Table 1.

Definition of cloud types and average frequency from the three radars. The differences are normalized by the KAZR frequency.

Definition of cloud types and average frequency from the three radars. The differences are normalized by the KAZR frequency.
Definition of cloud types and average frequency from the three radars. The differences are normalized by the KAZR frequency.
Fig. 5.

Example of KAZR-observed six cloud types defined in the study. Precipitating clouds are (a) shallow cloud, (b) congestus cloud, and (c) deep cloud; and nonprecipitating clouds are (d) midlevel cloud, (e) cirrus cloud, and (f) anvil cloud. (top) KAZR-observed reflectivity and (bottom) cloud phase classification.

Fig. 5.

Example of KAZR-observed six cloud types defined in the study. Precipitating clouds are (a) shallow cloud, (b) congestus cloud, and (c) deep cloud; and nonprecipitating clouds are (d) midlevel cloud, (e) cirrus cloud, and (f) anvil cloud. (top) KAZR-observed reflectivity and (bottom) cloud phase classification.

A cloud phase classification based on Shupe (2007) with parameters tuned for tropical clouds (Comstock et al. 2013) was applied to the KAZR data. The main purpose of the classification is to identify the presence of precipitation in the hydrometeor echo layers observed by KAZR. This technique uses radar Doppler moments (reflectivity, Doppler velocity, and spectrum width), lidar backscatter, microwave radiometer, and temperature profiles to identify cloud (liquid, ice), drizzle, and rain. In general, hydrometeors with high (low) lidar backscatter (>0.1) or radar reflectivity (>5 dBZ)/Doppler velocity (>2.5 m s−1) are classified as precipitating rain/snow (nonprecipitating cloud/ice) particles, which are further separated by the sounding temperature of 0°C. No liquid cloud water is allowed to exist colder than −12°C because aircraft measurements reported in Stith et al. (2002) found that liquid water is rarely observed at colder temperatures in tropical stratiform clouds. Further, for radar reflectivity of −17 ~ 5 dBZ and Doppler velocity of 1 ~ 2.5 m s−1, with temperature > 0°C, it is classified as drizzle. Although mixed-phase cloud likely exists in this dataset, hydrometeors are classified by their radiatively dominate phase. We expect that most clouds will be dominated by either liquid or ice, and therefore small amounts of either will not contribute significantly to the cloud radiative forcing or heating rates. The bottom panels in Fig. 5 show the cloud phase classification for each of the six cloud types defined in this study. As expected, precipitation (drizzle/rain, orange/green colors) appears increasingly frequent from shallow and congestus to deep clouds, while midlevel, cirrus, and anvil clouds mostly consist of nonprecipitating cloud particles (liquid/ice).

Table 1 shows the percentage of cloud profiles with precipitation (drizzle or rain) identified within the hydrometeor layer and those identified near surface (lowest range gate of KAZR) for the six cloud types, along with their respective average frequency of occurrence during DYNAMO/AMIE. Note that the definition of cloud types and cloud phase classification are two independent processes. The cloud phase classification is used as an independent evaluation of the precipitating and nonprecipitating cloud types defined using echo boundaries alone. The resulting percentages show that while more than 90% of the shallow, congestus, and deep clouds have precipitation identified within the hydrometeor layer, about 32%, 58%, and 84% of these clouds, respectively, have precipitation actually reaching the surface. While some midlevel, cirrus, and anvil clouds have precipitation inside the cloud, by definition there is no precipitation that reaches the surface. This result suggests that the definition of cloud types using hydrometeor-layer top/base heights (section 2d) can separate precipitating and nonprecipitating clouds reasonably well.

The averaged frequency of occurrence for the precipitating and nonprecipitating clouds from the three radars during DYNAMO/AMIE (values in Table 1) is plotted in Fig. 6. Note that the sum of all cloud types exceeds 100% because at a given time, multilayer clouds (e.g., shallow clouds and cirrus) could be present and would be counted in both categories. For precipitating clouds, the largest differences in frequency occur for shallow clouds. Both S-Pol and SMART-R overestimate shallow clouds frequency compared to KAZR, particularly for S-Pol, where its frequency is over sevenfold more than KAZR. Ground clutter from Gan Island, where the AMF KAZR was located, may be affecting shallow cloud detection by the two scanning radars. However, constructing the same time–height series from S-Pol RHI scans at other azimuths over the ocean (Fig. 1) and computing the frequency for different cloud types result in similarly overestimated shallow clouds (not shown). This suggests that Bragg scattering is the main reason that limits the use of S-Pol data in correctly detecting shallow clouds. Frequencies of the other two types of precipitating clouds (congestus and deep clouds) show much better agreement, with differences within 13% for both S-Pol and SMART-R. For nonprecipitating clouds, the S-Pol-observed frequency also agrees very well with KAZR, while SMART-R detects significantly less mid- and upper-level clouds due to its lower sensitivity.

Fig. 6.

Mean frequency of observed cloud types from three radar platforms. Refer to Table 1 for details of classifying the cloud types.

Fig. 6.

Mean frequency of observed cloud types from three radar platforms. Refer to Table 1 for details of classifying the cloud types.

c. Cloud detection accuracy by scanning radars

While comparison of frequency of occurrence describes general agreement in cloud detection between the three radars, collocated measurements allow more detail comparisons for clouds with coincident detection by different radars, providing useful insights into the characteristics of clouds, both precipitating and nonprecipitating, that can be accurately detected by the scanning radars within a certain range. Moreover, accurate detection of the evolution of cloud populations during different life cycles of the MJO is critical to test the DYNAMO/AMIE hypothesis, rather than the mean frequency of cloud occurrence.

To quantify the accuracy of the scanning radars in detecting various types of cloud, a 2 × 2 contingency table is used (Fig. 7). The KAZR-observed cloud types at each instance when S-Pol or SMART-R performs an RHI scan over the AMF site are treated as “true” values, and the S-Pol/SMART-R-detected cloud types are considered “predicted” values. Therefore, when both KAZR and S-Pol (or SMART-R) detect the same type of cloud at an instance, it is considered as true positive (TP); when S-Pol (or SMART-R) detects a cloud type that KAZR does not detect, it is considered as false positive (FP); when S-Pol (or SMART-R) miss the type of cloud detected by KAZR, it is considered false negative (FN); and finally, when both KAZR and S-Pol (or SMART-R) do not detect the cloud, it is considered true negative (TN). Therefore, the hit rate, accuracy rate, and false discovery rate (Schaefer 1990) can be computed as follows:

 
formula
 
formula
 
formula
Fig. 7.

A 2 × 2 contingency table where KAZR-detected clouds are treated as true value (p, n) and S-Pol/SMART-R-detected clouds are treated as predicted value (p′, n′). Total values are given by the sum of column or row, e.g., P = p′ + n′.

Fig. 7.

A 2 × 2 contingency table where KAZR-detected clouds are treated as true value (p, n) and S-Pol/SMART-R-detected clouds are treated as predicted value (p′, n′). Total values are given by the sum of column or row, e.g., P = p′ + n′.

A perfect score for hit rate and accuracy rate will be 1, and a perfect score for false discovery rate will be 0. The results of the scores are shown in Table 2. It is apparent that both S-Pol and SMART-R have relatively high hit rates (more than 0.72) and low false discovery rates (less than 0.25) for congestus and deep clouds, suggesting detection of these clouds agrees well with KAZR. For shallow clouds both S-Pol and SMART-R have high false discovery rates, especially S-Pol (0.91), which is consistent with the overestimated frequency of occurrence. The hit rate and accuracy rate for deep clouds can be underestimated due to inaccurate KAZR cloud-top estimates in heavy rainfall when its signals are severely attenuated, which will be further addressed in section 3f. For nonprecipitating clouds, both S-Pol and SMART-R have relatively high hit rates (>0.66) for anvil clouds, suggesting that their upper-level reflectivity data can be used to map the 3D structure of convective anvils. The S-Pol also performs well in detecting cirrus clouds, providing they are thicker than 270 m (section 2d). However, neither of the scanning radars is adequate for detecting tropical midlevel clouds, as they are oftentimes thin and requires high sensitivity for detection (Riihimaki et al. 2012).

Table 2.

Contingency table for S-Pol and SMART-R hydrometeor detection. KAZR hydrometeor types are used as true values.

Contingency table for S-Pol and SMART-R hydrometeor detection. KAZR hydrometeor types are used as true values.
Contingency table for S-Pol and SMART-R hydrometeor detection. KAZR hydrometeor types are used as true values.

d. Cloud thickness comparison with coincident detection

To provide guidance on the physical characteristics of clouds that can be reliably detected by the scanning radars, particularly those precipitating clouds that are important to the initiation of MJO, we compare the cloud thickness in coincident detection of clouds between the three radars. Through this comparison, guidance of using the scanning radar data to map the three-dimensional cloud fields will be provided.

Cloud thickness is defined as the distance between echo top and base height, and therefore for precipitating clouds, the thickness includes the precipitation below the actual cloud base. Figure 8 shows the distribution of cloud thickness with coincident detection from the three radars; that is, at a given instance, if S-Pol (or SMART-R) detects the same type of cloud (Table 1) as KAZR, the cloud thickness is included in the statistics. For shallow clouds, cloud thickness from KAZR shows a roughly exponential decrease. Most of the coincidently detected shallow clouds are less than 2 km thick. Congestus clouds thickness ranges between 2.5 and 7 km for both KAZR and S-Pol, while the SMART-R thickness distribution shows an ~1-km shift. This shift is likely because some real low-level echoes from SMART-R data (e.g., Fig. 3) were removed by the clutter filtering (section 2d). Most deep clouds are thicker than 7 km across all three radars; however, S-Pol reports a higher frequency of clouds that are 14 km or thicker compared to KAZR, likely due to signal attenuation of KAZR in these thick clouds; while SMART-R reports much lower frequency of deep cloud thickness above 10 km, possibly due to 1) too much low-level echo removal during data quality control, and 2) lower cloud-top height estimate because SMART-R is not as sensitive to small cloud drops near cloud top as KAZR and S-Pol. For nonprecipitating midlevel and cirrus clouds, thickness also follows an exponential distribution with a quick falloff above 2 km, while for anvil clouds the thickness ranges from 1 to 8 km.

Fig. 8.

Cloud thickness distribution of coincident cloud detection by the three radars. The cloud thickness bin values are calculated such that there are 21 equally sized bins for each cloud type. The panels are organized in the same order as Fig. 5.

Fig. 8.

Cloud thickness distribution of coincident cloud detection by the three radars. The cloud thickness bin values are calculated such that there are 21 equally sized bins for each cloud type. The panels are organized in the same order as Fig. 5.

This comparison of cloud thickness with coincident detection suggests that minimum cloud thicknesses of 2.5, 7, and 1 km can be added as additional constraints when defining congestus, deep, and anvil clouds, respectively, to increase reliability of detecting these clouds when using scanning radar data at locations outside of the AMF KAZR. While RHI scans by the scanning radars can detect 2.5-km-thick clouds, it is more challenging for PPI scans, as the gap between two typical PPI elevation angles (0.5°–1°) quickly increases with distance from the radar. For example, at the 50-km range the vertical distance between two PPI scans 1° apart is ~1 km, so that a congestus cloud of 3 km thick only contains three vertical data points. Therefore, RHI data from scanning radars should be used whenever possible for accurate detection of congestus and high clouds. Range is also an important factor to consider for scanning radars, which will be addressed in the next subsection.

e. Impact of decreasing sensitivity with distance to cloud detection

The comparisons between S-Pol, SMART-R, and KAZR so far were performed at a distance of about 10 km from the two scanning precipitation radars. As mentioned in section 2, the sensitivity decreases with range from the radar and therefore reducing the detectability of clouds, as they are farther away. Considering S-Pol has much higher sensitivity to clouds than SMART-R, we investigate the impact of S-Pol’s decreasing sensitivity with range in detecting various types of clouds. The minimum detectable reflectivity from S-Pol can be estimated by

 
formula

where C is the S-Pol radar constant (68.9 dBZ), is the S-Pol minimum detectable signal (−113.34 dBm), A is one-way atmospheric attenuation, and r is the range from radar (km). The actual value of A varies with humidity and therefore with height, but it is approximately 0.005 dB km−1 in the lowest levels. For example, at distances of 10, 20, 30, 50, 100, and 150 km, the minimum detectable reflectivity by the S-Pol are approximately −24, −18, −15, −10, −3, and 1 dBZ, respectively. To determine the impact of decreasing sensitivity with range to the detection of various cloud types, Fig. 9 shows the cumulative frequency of KAZR radar reflectivity (corrected for droplet attenuation; more details in section 4) by altitude for the six cloud types, overlaid with the minimum sensitivity of S-Pol at several typical distances. The percentages to the left of the vertical lines indicate clouds that are potentially missed by the S-Pol due to limitations of its sensitivity.

Fig. 9.

Cumulative frequency of radar reflectivity by altitude from KAZR for the six cloud types, organized in the same order as Fig. 5. Cumulative frequency is calculated starting from −40 dBZ. The vertical lines show minimum sensitivity of S-Pol as a function of distance from the radar, such that the percentage of clouds to the left of the lines is potentially undetected by S-Pol due to reflectivities below its sensitivity.

Fig. 9.

Cumulative frequency of radar reflectivity by altitude from KAZR for the six cloud types, organized in the same order as Fig. 5. Cumulative frequency is calculated starting from −40 dBZ. The vertical lines show minimum sensitivity of S-Pol as a function of distance from the radar, such that the percentage of clouds to the left of the lines is potentially undetected by S-Pol due to reflectivities below its sensitivity.

For precipitating clouds, decreasing S-Pol sensitivity has relatively less impact on cloud detection compared to nonprecipitating clouds due to higher reflectivities from rain- and drizzle-sized particles. For shallow clouds, issues with Bragg scattering likely have a larger effect on S-Pol than sensitivity. For congestus clouds, the S-Pol can potentially detect up to 80% (60%) with a range out to 30 km (50 km), making its data useful for mapping three-dimensional volumes of congestus clouds (precipitating or not) and for investigating their role in preconditioning deep convection. Similarly for anvil clouds between 6- and 12-km height, S-Pol can detect up to 70% (50%) at a range out to 30 km (50 km). Nonprecipitating midlevel and cirrus clouds require much higher sensitivity for detection. For example, at 10 km, between 30% and 50% of these clouds could be undetected by S-Pol. Users of S-Pol radar data for studying various types of clouds should be aware of the impact of distance on cloud detection.

f. KAZR attenuation by rainfall

Finally, to address the KAZR attenuation issues in precipitating clouds mentioned in section 3c, cloud-top height differences between both the S- and C-band radars and the KAZR are shown as a function of surface rainfall rate measured by rain gauge and retrieved by S-Pol/SMART-R radar, respectively (Fig. 10). Since the S-Pol (SMART-R) performed RHI scans every 15 min (10 min) over the KAZR, both S- and C-band cloud-top heights are used for this analysis to increase the frequency of detection of KAZR attenuation, particularly for heavy rainfall periods with short duration. For time periods when both S-Pol and SMART-R are available (0, 30 min), the S-Pol cloud tops are preferentially used due to its higher sensitivity.

Fig. 10.

(left) Cloud-top height differences between S-/C-band radars and KAZR as a function of surface rain gauge, and (right) near-surface radar derived rain rates. Only positive differences are plotted. The dots are median cloud-top height differences for each rain-rate bin (bin size given by , where X = 1, 2, 3, …), and the solid lines are fit lines for the median values.

Fig. 10.

(left) Cloud-top height differences between S-/C-band radars and KAZR as a function of surface rain gauge, and (right) near-surface radar derived rain rates. Only positive differences are plotted. The dots are median cloud-top height differences for each rain-rate bin (bin size given by , where X = 1, 2, 3, …), and the solid lines are fit lines for the median values.

We identify instances of KAZR attenuation by requiring cloud-top height from the S-/C-band radars to be more than 100 m (1 vertical grid) higher than that from KAZR in precipitating clouds (i.e., shallow, congestus, deep cloud; see Table 1), and the maximum reflectivity from S-/C-band radars in the lowest 6 km must exceed 10 dBZ (i.e., indicating the presence of raindrops). Therefore, Fig. 10 shows a subsample of all the collocated KAZR/S-Pol/SMART-R precipitating profiles where KAZR signals are significantly attenuated. The resulting frequency of attenuation occurrence during the entire DYNAMO/AMIE field campaign is ~5% (~34% in precipitating clouds), and the averaged KAZR cloud-top underestimation is 1.15 km. Note that this is a conservative estimate because the C-band SMART-R has lower sensitivity in detecting cloud-top heights than the S-Pol.

As shown in Fig. 10, when KAZR suffers from rainfall attenuation, cloud-top height differences during drizzle to moderate rain rate (0.1–5 mm h−1) are most frequently between 0.2 and 0.5 km, and only slightly increase with rain rate. However, attenuation sharply increases when the surface rain rate is above 5 mm h−1, such that in heavy rainfall (>20 mm h−1) KAZR can underestimate cloud-top height by more than 2 km. An empirical fit between surface rain-rate and cloud-top height differences are calculated for rain-gauge-measured and radar-derived rain rates separately. The cloud-top height difference is assumed to have a power-law relationship with the surface rain rate, such that

 
formula

where R is the rain rate (mm h−1), and ΔH is the KAZR cloud-top height underestimation (km). The empirical a and b values are indicated in each panel in Fig. 10. Because of sampling limitations, rainfall events are not separated into convective and stratiform rain types, although higher rain rates (>10 mm h−1) are typically associated with convective rain (Tokay and Short 1996). This relatively simple method provides first-order correction of Ka-band radar cloud-top height estimates in the presence of precipitating clouds and can potentially be applied to other tropical sites (i.e., the ARM tropical western Pacific sites at Darwin, Australia, and Manus, Papua New Guinea) to improve the quality of the ARSCL cloud boundary data product, which is widely used by the climate research community (e.g., Kollias et al. 2007; Xie et al. 2010; Long et al. 2013; Mather and Voyles 2013).

4. Producing merged dataset

As discussed in the previous section, while KAZR provides observations with high temporal and vertical resolution for nonprecipitating and lightly precipitating clouds, its data are questionable during moderate to heavier precipitation (i.e., rain rate > 10 mm h−1), when the signals are severely attenuated and fail to correctly detect the precipitating cloud tops (Fig. 10). Therefore, using KAZR data alone to study the evolution of tropical convective clouds could result in biases, particularly in moderate to heavily precipitating clouds.

Fortunately, the collocated S-Pol and SMART-R radar reflectivity profiles with KAZR during the DYNAMO/AMIE field campaign provide an excellent opportunity to improve the quality of the KAZR dataset in the presence of precipitating convective clouds. Compared to Ka-band radar, S-band radar data are much less affected by attenuation in precipitating clouds, and the comparisons in section 3 show that the S-Pol radar provides reliable hydrometeor detection in precipitating congestus and deep clouds. Therefore, the S-Pol data can be combined with the KAZR data to obtain total hydrometeor profile estimates (both nonprecipitating and precipitating clouds) at Gan Island during the DYNAMO/AMIE field campaign. Using this merged KAZR–S-Pol dataset, profiles of hydrometeor microphysics and cloud radiative effect (Mather et al. 2007; McFarlane et al. 2007) will be retrieved using the Pacific Northwest National Laboratory combined remote sensor retrieval algorithm (PNNL COMBRET; Zhao et al. 2012; Comstock et al. 2013). This merged retrieval dataset can be used to better address the DYNAMO science hypotheses and numerical model evaluation than the original KAZR dataset.

To produce a seamless merged radar reflectivity and PNNL COMBRET dataset between KAZR and S-Pol, an improved method based on Feng et al. (2009) is used. Figure 11 shows the flowchart of the method and dataset used for each step. First, PNNL COMBRET is applied to the original KAZR dataset. This algorithm combines radar Doppler moments, a lidar attenuated backscatter coefficient, sounding, a microwave radiometer, and a surface rain gauge to retrieve cloud water content, effective particle size, and visible extinction coefficient for both liquid and ice clouds (Comstock et al. 2013). Compared to the method used by Feng et al. (2009) that assumes linear liquid water content profiles scaled from microwave-radiometer-retrieved liquid water path, this study uses the liquid water content profiles retrieved by COMBRET instead.

Fig. 11.

Flowchart of creating a merged KAZR–S-Pol data for the retrieval. Terms ZKa, ZS, and Htop are KAZR reflectivity, S-Pol reflectivity, and cloud-top heights, respectively.

Fig. 11.

Flowchart of creating a merged KAZR–S-Pol data for the retrieval. Terms ZKa, ZS, and Htop are KAZR reflectivity, S-Pol reflectivity, and cloud-top heights, respectively.

As discussed in section 3b, cloud phase is classified following Shupe (2007) with parameters tuned for tropical clouds (Comstock et al. 2013). For a volume classified as one of the three liquid hydrometeors (cloud, drizzle, or rain), the liquid water content associated with that specific type is retrieved by COMBRET. For example, only the rainwater content is retrieved if a volume is classified as rain and all other water content is neglected. For liquid clouds, COMBRET follows the ARM baseline cloud retrieval value-added product (MICROBASE) algorithm; for drizzle/rain, COMBRET retrieves rainwater content using

 
formula

The raindrop size distribution follows an exponential form of , where the intercept parameter (Marshall and Palmer 1948), and the slope parameter . Rain rate R is derived from the following ZR relationship:

 
formula

where Z is radar reflectivity in linear units of mm6 m−3 and R is in millimeters per hour. Both the ZR relationship and the slope parameter Λ are obtained from surface disdrometer data on Addu Atoll (E. Thompson 2014, personal communication). The KAZR reflectivity profiles are corrected for liquid attenuation following the approach described by Lhermitte (1990) and Feng et al. (2009), respectively. The drizzle/rain attenuation correction (Feng et al. 2009) method finds Ka-band attenuation from lookup tables (produced using Mie theoretical calculations) as functions of liquid water content, effective liquid droplet radius, and sounding temperature. The attenuation in each range gate containing liquid water is then integrated with height to produce an attenuation correction profile, which is then applied to the original KAZR reflectivity profiles to obtain the attenuation-corrected KAZR reflectivity data.

In the second step, the S-Pol reflectivity data are first converted to linear units (mm6 m−3), then linearly interpolated in height to match the KAZR vertical grid, and finally converted back to log unit (dBZ). An important factor that must be considered when merging S-band and Ka-band radar observations is the Mie scattering effect (Lhermitte 1990). When particle diameter D exceeds about 10% of the radar wavelength, λ (i.e., D/λ > 0.1), the Rayleigh scattering assumption no longer holds. Therefore, reflectivity differences between S band and Ka band are much larger for relatively large hydrometeor particles (i.e., reflectivity > 10 dBZ). Previous studies have proposed adjustments for reflectivity differences between wavelengths using in situ measured microphysics (e.g., Protat et al. 2009; Matrosov 2009). In this study, collocated radar reflectivity measurements between S band and Ka band are used to derive relationships between different wavelengths. Figure 12 shows the relationship of radar reflectivities between the S-Pol and KAZR, before (Figs. 12a,b) and after (Figs. 12c,d) attenuation correction, for both liquid cloud and ice clouds, respectively. Reflectivity values above (below) the sounding temperature height of 0°C are assumed to be liquid (ice). Although mixed-phase hydrometeors are possible in the ice category, it is not specifically defined in this study. The results indicate that the attenuation-corrected Ka-band reflectivity in both liquid and ice clouds agree much better with S-Pol than the original Ka-band reflectivity, as shown by the fitting function being much closer to the 1-to-1 line. The S-band reflectivity data are then converted to equivalent Ka band (ZS-Ka) using the derived fit function (Table 3) in the third step.

Fig. 12.

Scatterplot of collocated radar reflectivity between S-Pol and KAZR for the entire DYNAMO/AMIE campaign. (a),(b) KAZR original uncorrected reflectivity and (c),(d) attenuation-corrected KAZR reflectivity. Black dots are the most frequent KAZR reflectivity values for each S-Pol reflectivity bin (1 dB), and red lines are least squares fits to the most frequent KAZR values. Fit equations and regression coefficients are also provided.

Fig. 12.

Scatterplot of collocated radar reflectivity between S-Pol and KAZR for the entire DYNAMO/AMIE campaign. (a),(b) KAZR original uncorrected reflectivity and (c),(d) attenuation-corrected KAZR reflectivity. Black dots are the most frequent KAZR reflectivity values for each S-Pol reflectivity bin (1 dB), and red lines are least squares fits to the most frequent KAZR values. Fit equations and regression coefficients are also provided.

Table 3.

Coefficients for the conversion functions of radar reflectivity between Ka and S band derived from the collocated radar measurements during AMIE/DYNAMO.

Coefficients for the conversion functions of radar reflectivity between Ka and S band derived from the collocated radar measurements during AMIE/DYNAMO.
Coefficients for the conversion functions of radar reflectivity between Ka and S band derived from the collocated radar measurements during AMIE/DYNAMO.

In the fourth step, “precipitation events” are defined using surface rain gauge data at the AMF site. An event is defined as a continuous time period when the surface rain rate is above 1 mm h−1. Within each precipitation event, if the S-Pol detects precipitating cloud-top (Table 1) heights more than 200 m (two vertical grids) higher than that from the KAZR for more than 10% of the time, KAZR reflectivity profiles within this period are replaced by the S-Pol data. The cloud-top height criteria is intended to capture periods when KAZR data are truly affected by attenuation from precipitation, as shown in the cloud-top height differences in precipitating clouds (Fig. 10). Figure 13 shows two example days of the merged dataset. It is clear that during the heavy precipitation events (indicated by red bars in Figs. 13e,f), the merged dataset provides improved reflectivity profile estimates rather than using KAZR alone, as well as better cloud-top height estimates (purple dots). During the DYNAMO/AMIE field campaign, about 28% of the KAZR profiles with precipitation reaching the surface were replaced by the S-Pol profiles in the merged dataset.

Fig. 13.

Examples of the S-Pol–KAZR dataset: (a),(b) KAZR original uncorrected reflectivity; (c),(d) S-Pol reflectivity; (e),(f) merged S-Pol and attenuation-corrected KAZR reflectivity and flags indicating radar data sources (color bars); and (g),(h) surface rain rate (black line) and cloud-top height difference between merged and original KAZR observation. The purple dots in (a)–(f) indicate precipitating cloud-top heights, and the black dots in (a),(b) and (e),(f) denote KAZR-ARSCL best estimate cloud-base heights for precipitating clouds.

Fig. 13.

Examples of the S-Pol–KAZR dataset: (a),(b) KAZR original uncorrected reflectivity; (c),(d) S-Pol reflectivity; (e),(f) merged S-Pol and attenuation-corrected KAZR reflectivity and flags indicating radar data sources (color bars); and (g),(h) surface rain rate (black line) and cloud-top height difference between merged and original KAZR observation. The purple dots in (a)–(f) indicate precipitating cloud-top heights, and the black dots in (a),(b) and (e),(f) denote KAZR-ARSCL best estimate cloud-base heights for precipitating clouds.

In the final step, the PNNL COMBRET cloud microphysics retrieval and radiative heating rate retrieval is applied to the merged attenuation-corrected KAZR–S-Pol data, and compared to that retrieved using the original uncorrected KAZR data. Figure 14 shows an example of radar reflectivity, retrieved water content, cloud radiative effects, and the differences between the merged attenuation-corrected and original uncorrected KAZR product. During the “precipitation event” when the KAZR signal is severely attenuated by rainfall (e.g., 0100–0600 UTC), retrievals from the merged attenuation-corrected product show much higher water content and cloud radiative effects than from the original uncorrected KAZR product due to using the S-Pol data (red bar in Fig. 14c). The differences during other periods when KAZR data are used (green bar in Fig. 14c) are resulting from attenuation correction for the KAZR reflectivity.

Fig. 14.

Examples of the cloud microphysics retrieval and cloud radiative effect for the same case in Fig. 14a from (left) the attenuation-corrected KAZR–S-Pol data, (middle) original uncorrected KAZR, and (right) their difference (merged minus original). (a)–(c) reflectivity, with color bar on top indicating radar data source; (d)–(f) water content; (g)–(i) particle diameter; (j)–(l) longwave cloud radiative effect; (m)–(o) shortwave cloud radiative effect, and (p)–(r) MWR liquid water path (black line) and retrieved liquid (red)/ice (green) water path. The black dots in (a),(b) and (d),(e) indicate KAZR-ARSCL best estimate cloud-base heights for precipitating clouds. Water contents in (d)–(f) correspond to retrieval for the particular phase (i.e., cloud/rain/snow/ice) identified in each range gate; see text for more details.

Fig. 14.

Examples of the cloud microphysics retrieval and cloud radiative effect for the same case in Fig. 14a from (left) the attenuation-corrected KAZR–S-Pol data, (middle) original uncorrected KAZR, and (right) their difference (merged minus original). (a)–(c) reflectivity, with color bar on top indicating radar data source; (d)–(f) water content; (g)–(i) particle diameter; (j)–(l) longwave cloud radiative effect; (m)–(o) shortwave cloud radiative effect, and (p)–(r) MWR liquid water path (black line) and retrieved liquid (red)/ice (green) water path. The black dots in (a),(b) and (d),(e) indicate KAZR-ARSCL best estimate cloud-base heights for precipitating clouds. Water contents in (d)–(f) correspond to retrieval for the particular phase (i.e., cloud/rain/snow/ice) identified in each range gate; see text for more details.

Note that a sharp increase of retrieved water content around 5-km height occurs because COMBRET does not explicitly retrieve mixed-phase cloud properties. When snow/ice particles fall below 0°C isotherm, they begin to melt and are coated with a liquid shell, which greatly enhances the radar reflectivity (known as the bright band with ~500-m thickness), resulting in a substantial increase in the retrieved rainwater content around that level. There are additional cases (not shown) when a cloud or a cloud with virga below crosses the melting level, which also have a discontinuity due to the type of retrieval applied. Since there is no straightforward way to retrieve simultaneously the ice and liquid properties in this region, we choose to apply the appropriate retrieval based on the phase classification. Although one past study has suggested a method to better represent attenuation of radar signal by these mixed-phase hydrometeors (Matrosov 2008), more research in this area with in situ measurements of melting snow/ice particles for validation is still needed before implementing such a technique into COMBRET.

To further examine the impact to cloud radiative heating rate retrieval from the merged KAZR–S-Pol product and the original KAZR product, we calculated the averaged cloud radiative effect during the DYNAMO/AMIE field campaign separately for the two products (Fig. 15). For precipitating clouds, the merged product produces 22% stronger longwave cooling (21% stronger heating) above (below) 10 km, and similarly 18% stronger shortwave heating (73% weaker cooling) above (below) 7 km (Fig. 15c). The net radiative effect is about 11% stronger (absolute value) from the merged product. For nonprecipitating clouds, there is negligible difference in their radiative effects (Fig. 15f) as expected due to negligible KAZR attenuation, resulting in identical microphysics retrievals. For all clouds, the impact from the merged product is similar to that of the precipitating clouds, except the net radiative effect is about 7% stronger.

Fig. 15.

Averaged cloud radiative effect during the DYNAMO/AMIE field campaign from (top) merged attenuation-corrected KAZR–S-Pol data, (middle) original uncorrected KAZR data, and (bottom) their difference (merged minus original). (a)–(c) Precipitating clouds, (d)–(f) nonprecipitating clouds, and (g)–(i) all clouds.

Fig. 15.

Averaged cloud radiative effect during the DYNAMO/AMIE field campaign from (top) merged attenuation-corrected KAZR–S-Pol data, (middle) original uncorrected KAZR data, and (bottom) their difference (merged minus original). (a)–(c) Precipitating clouds, (d)–(f) nonprecipitating clouds, and (g)–(i) all clouds.

This merged KAZR–S-Pol data product provides more accurate cloud-top height estimates and attenuation-corrected radar reflectivity for precipitating clouds compared to the KAZR-ARSCL product. Moreover, the cloud water content, the effective particle size, and the radiative heating rate for both precipitating and nonprecipitating clouds were retrieved by PNNL COMBRET. This data product has been provided to the research community (available in the ARM data archive as a principal investigator product) as best estimates of the total hydrometeor microphysics and their radiative effects at Gan Island during the DYNAMO/AMIE field campaign.

From the comparison results of precipitating clouds between the three radars performed in section 3, this new dataset provides the only reliable estimates of shallow clouds at Addu Atoll. These shallow clouds along with congestus have been suggested as contributing to moistening and preconditioning the atmosphere for deep convection (Johnson et al. 1999; Benedict and Randall 2007). The trimodal tropical convective cloud population (Johnson et al. 1999) and its evolution with the life cycle of the MJO can be reliably constructed with this merged dataset, because of the improved cloud-top height estimates for precipitating clouds. Together with the DYNAMO/AMIE intensive 3-hourly soundings, the interaction and feedback from shallow and congestus clouds and their large-scale environments can be thoroughly investigated to advance our understanding of their roles in various phases of the MJO. The retrieved cloud microphysics and radiative heating rate can also be used to evaluate model simulations.

5. Summary and conclusions

The DYNAMO/AMIE field campaign successfully completed operations from October 2011 to February 2012. One of the key hypotheses proposed in the campaign is that different phases of the MJO are characterized by different cloud populations and that a specific cloud population is essential to the initiation of the MJO. To test this hypothesis, three radar systems (KAZR, SMART-R, and S-Pol) covering the Ka, C, and S bands were deployed at a “supersite” on Addu Atoll, Maldives, to document the full spectrum of convective clouds, from shallow and congestus clouds to deep convection. This paper focuses on evaluating the ability of the vertically pointing and scanning radars deployed at Addu Atoll to document these cloud populations and to take advantage of the multiwavelength radar platforms to produce a merged cloud–precipitation radar dataset, along with continuous total hydrometeor microphysics and radiative heating profile retrievals that can be used to address the DYNAMO/AMIE science hypothesis and numerical model evaluation.

Comparisons of cloud statistics observed by collocated KAZR, S-Pol, and SMART-R radar observations were performed at Gan Island to quantify the ability of the two scanning radars in detecting both precipitating and nonprecipitating clouds, which are subdivided into three categories (precipitating: shallow, congestus, deep clouds; nonprecipitating: midlevel, cirrus, and anvil clouds) based on their cloud boundary base and top heights (Table 1). While all cloud types are compared, particular attention is focused on shallow, congestus, and deep clouds because of their importance to the DYNAMO/AMIE science goal. The results are summarized as follows:

  1. Statistics from KAZR show that while more than 90% of the shallow, congestus, and deep clouds have precipitation identified within the hydrometeor layer, about 32%, 58%, and 84%, respectively, have precipitation actually reaching the surface. For precipitating clouds, the largest difference in the frequency of occurrence between S-Pol/SMART-R and KAZR is for shallow cloud, where both scanning radars overestimate its occurrence, possibly due to low-level Bragg scatter and/or ground clutter. Frequency of congestus and deep clouds agree much better between both scanning radars and KAZR (within 13%, Table 1), along with high coincident detection rates (>72%, Table 2). For nonprecipitating clouds, both S-Pol and SMART-R reported relatively high coincident detection rates (80% and 66%, respectively) for anvil clouds, while S-Pol also detects up to 79% of cirrus clouds (thicker than 270 m) within a 8.5-km range.

  2. Comparisons of cloud thickness with coincident detection between the three radars suggest that minimum cloud thicknesses of 2.5, 7, and 1 km can be used as constraints when identifying congestus, deep, and cirrus/anvil clouds, respectively, using scanning radar data (Fig. 8), in addition to cloud boundary heights. The impact of decreasing sensitivity with range to S-Pol’s cloud detection is investigated. At 30–50-km radius, S-Pol can potentially detect up to 80%–60% and 70%–50% of congestus and anvil clouds, respectively (Fig. 9). Detection of deep clouds by S-Pol is much less affected by sensitivity due to high reflectivity from precipitation size particles.

  3. Cloud-top height comparison in precipitating clouds between KAZR and S-Pol/SMART-R reveals that KAZR underestimates cloud-top heights due to rainfall attenuation in ~34% of the precipitating clouds during the DYNAMO/AMIE field campaign, with an average cloud-top underestimate of 1.15 km. An empirical method of correcting cloud-top height bias for KAZR using surface rainfall rate has been proposed (Fig. 10). This relatively simple method can potentially be applied to other Ka-band vertically pointing cloud radars in the tropics to improve the quality of the ARSCL cloud boundary data product in the presence of precipitation.

  4. A merged KAZR–S-Pol dataset is produced to obtain total hydrometeor profile estimates at Gan Island during the DYNAMO/AMIE field campaign. This merged data product provides improved cloud-top height estimates and attenuation-corrected radar reflectivity for precipitating clouds compared to the KAZR-ARSCL product. Moreover, hydrometeor water content, effective particle size, and cloud radiative heating rate are retrieved by the PNNL COMBRET algorithm for both precipitating and nonprecipitating clouds. The retrieved cloud radiative effects using the merged KAZR–S-Pol dataset are substantially stronger than that using KAZR alone in precipitating clouds. This data product has been provided to the research community (ARM principal investigator product) as best estimates of the total hydrometeor microphysics and their radiative effects at Gan Island for the field campaign.

Comparisons between KAZR, S-Pol, and SMART-R performed in this study indicate that KAZR data are the only reliable estimates of shallow clouds at Addu Atoll, while the scanning radars can reasonably detect congestus and anvil clouds within a certain range in addition to precipitating deep clouds. To take advantage of the three-dimensional cloud-detecting capability of the scanning radars that is essential to document the degree of convective organization, and to understand the role of both precipitating and nonprecipitating congestus clouds that are important for preconditioning deep convection and the initiation of MJO, we provide the following recommendations in identifying congestus clouds from the scanning radar data: 1) define cloud-top/cloud-base height using a continuous layer of reflectivity > −20 dBZ; 2) use cloud-top height between 3 and 8 km, cloud-base height below 3 km, and cloud thickness greater than 2.5 km; 3) limit the use of the data within 30–50-km radius from the radar to keep reasonable detection of various congestus clouds; and 4) use data from RHI scans instead of PPI scans if possible to have higher vertical resolution.

As new research involving the use of DYNAMO/AMIE radar data continues to emerge, this study hopes to provide quantitative guidance in using these radar data for various cloud and precipitation studies. Previous studies have used millimeter-wavelength cloud radar data to study MJO processes (e.g., Deng et al. 2013), but they are limited in some ways in not being able to fully examine deep convective clouds due to attenuation by heavy rainfall and the lack of high-frequency large-scale soundings. The merged KAZR–S-Pol dataset produced in this study alleviates this issue and can be reliably used to construct the trimodal tropical cloud population (shallow, congestus, and deep cloud) because of the improved cloud-top height estimates for precipitating clouds. Together with the DYNAMO/AMIE intensive 3-hourly sounding arrays, the interaction and feedback from shallow and congestus clouds and their large-scale environments can be investigated to address some of the key DYNAMO/AMIE science hypothesis. The retrieved cloud microphysics and radiative heating rate provide a unique dataset at this remote oceanic region to study the radiative impact of tropical clouds and to evaluate various model simulations. Some of these research activities using this dataset are already underway by the authors of this paper and will be presented in future studies.

Acknowledgments

The authors thank Drs. Jim Mather and Laura Riihimaki for their suggestions, three anonymous reviewers in their thorough comments and constructive criticisms, Dr. Robert Houze and Stacy Brodzik for their help in processing the S-Pol data, and Elizabeth Thompson for providing the relationship in raindrop size distribution from the 2D-video disdrometer. This work is supported by the U.S. Department of Energy under the Atmospheric System Research Program. Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy under Contract DE-AC06-76RLO1830.

REFERENCES

REFERENCES
Annamalai
,
H.
, and
J. M.
Slingo
,
2001
:
Active/break cycles: Diagnosis of the intraseasonal variability of the Asian summer monsoon
.
Climate Dyn.
,
18
,
85
102
, doi:.
Benedict
,
J. J.
, and
D. A.
Randall
,
2007
:
Observed characteristics of the MJO relative to maximum rainfall
.
J. Atmos. Sci.
,
64
,
2332
2354
, doi:.
Bharadwaj
,
N.
, and
V.
Chandrasekar
,
2012
:
Wideband waveform design principles for solid-state weather radars
.
J. Atmos. Oceanic Technol.
,
29
,
14
31
, doi:.
Bharadwaj
,
N.
,
A.
Lindenmaier
,
K. B.
Widener
,
K. L.
Johnson
, and
V.
Venkatesh
,
2013
: Ka-band ARM zenith profiling radar (KAZR) network for climate study. 36th Conf. on Radar Meteorology, Breckenridge, CO, Amer. Meteor. Soc., 14A.8. [Available online at https://ams.confex.com/ams/36Radar/webprogram/Manuscript/Paper228620/14A8_ams_radconf_kazr.pdf.]
Biggerstaff
,
M. I.
, and Coauthors
,
2005
:
The Shared Mobile Atmospheric Research and Teaching radar: A collaboration to enhance research and teaching
.
Bull. Amer. Meteor. Soc.
,
86
,
1263
1274
, doi:.
Carvalho
,
L. M. V.
,
C.
Jones
, and
T.
Ambrizzi
,
2005
:
Opposite phases of the Antarctic Oscillation and relationships with intraseasonal to interannual activity in the tropics during the austral summer
.
J. Climate
,
18
,
702
718
, doi:.
Comstock
,
J. M.
,
A.
Protat
,
S. A.
McFarlane
,
J.
Delanoe
, and
M.
Deng
,
2013
:
Assessment of uncertainty in cloud radiative effects and heating rates through retrieval algorithm differences: Analysis using 3 years of ARM data at Darwin, Australia
.
J. Geophys. Res.
, 118, 4549–4571, doi:.
Deng
,
L.
,
S. A.
McFarlane
, and
J. E.
Flaherty
,
2013
:
Characteristics associated with the Madden–Julian oscillation at Manus Island
.
J. Climate
,
26
,
3342
3356
, doi:.
Dixon
,
M.
, and
J. C.
Hubbert
,
2012
: The separation of noise and signal components in Doppler radar returns. Proc. Seventh European Conf. on Radar in Meteorology and Hydrology, Toulouse, France, Météo-France, 13B.1. [Available online at http://www.meteo.fr/cic/meetings/2012/ERAD/short_abs/SP_078_sh_abs.pdf.]
Ellis
,
S. M.
, and
J.
Vivekanandan
,
2010
:
Water vapor estimates using simultaneous dual-wavelength radar observations
.
Radio Sci.
,
45
, RS5002, doi:.
Ellis
,
S. M.
, and
J.
Vivekanandan
,
2011
:
Liquid water content estimates using simultaneous S and Ka band radar measurements
.
Radio Sci.
,
46
,
RS2021
, doi:.
Feng
,
Z.
,
X.
Dong
, and
B.
Xi
,
2009
:
A method to merge WSR-88D data with ARM SGP millimeter cloud radar data by studying deep convective systems
.
J. Atmos. Oceanic Technol.
,
26
,
958
971
, doi:.
Fliegel
,
J. M.
, and
C.
Schumacher
,
2012
: Quality control and census of SMART-R observations from the DYNAMO/CINDY2011 field campaign. M.S. thesis, Dept. of Atmospheric Sciences, Texas A&M University, 82 pp.
Hendon
,
H. H.
, and
B.
Liebmann
,
1990
:
The intraseasonal (30–50 day) oscillation of the Australian summer monsoon
.
J. Atmos. Sci.
,
47
,
2909
2924
, doi:.
Hubbert
,
J. C.
,
M.
Dixon
,
S. M.
Ellis
, and
G.
Meymaris
,
2009a
:
Weather radar ground clutter. Part I: Identification, modeling, and simulation
.
J. Atmos. Oceanic Technol.
,
26
,
1165
1180
, doi:.
Hubbert
,
J. C.
,
M.
Dixon
, and
S. M.
Ellis
,
2009b
:
Weather radar ground clutter. Part II: Real-time identification and filtering
.
J. Atmos. Oceanic Technol.
,
26
,
1181
1197
, doi:.
Johnson
,
R. H.
,
T. M.
Rickenbach
,
S. A.
Rutledge
,
P. E.
Ciesielski
, and
W. H.
Schubert
,
1999
:
Trimodal characteristics of tropical convection
.
J. Climate
,
12
,
2397
2418
, doi:.
Keeler
,
R. J.
,
J.
Lutz
, and
J.
Vivekanandan
,
2000
: S-Pol: NCAR’s polarimetric Doppler research radar. IGARS 2000: IEEE 2000 International Geoscience and Remote Sensing Symposium, T. I. Stein, Ed., Vol. 4, 1570–1573, doi:.
Kemball-Cook
,
S. R.
, and
B. C.
Weare
,
2001
:
The onset of convection in the Madden–Julian oscillation
.
J. Climate
,
14
,
780
793
, doi:.
Kollias
,
P.
,
M. A.
Miller
,
E. P.
Luke
,
K. L.
Johnson
,
E. E.
Clothiaux
,
K. P.
Moran
,
K. B.
Widener
, and
B. A.
Albrecht
,
2007
:
The Atmospheric Radiation Measurement Program Cloud Profiling Radars: Second-generation sampling strategies, processing, and cloud data products
.
J. Atmos. Oceanic Technol.
,
24
,
1199
1214
, doi:.
Lhermitte
,
R.
,
1990
:
Attenuation and scattering of millimeter wavelength radiation by clouds and precipitation
.
J. Atmos. Oceanic Technol.
,
7
,
464
479
, doi:.
Lin
,
H.
,
G.
Brunet
, and
J.
Derome
,
2009
:
An observed connection between the North Atlantic Oscillation and the Madden–Julian oscillation
.
J. Climate
,
22
,
364
380
, doi:.
Lin
,
J.-L.
, and Coauthors
,
2006
:
Tropical intraseasonal variability in 14 IPCC AR4 climate models. Part I: Convective signals
.
J. Climate
,
19
,
2665
2690
, doi:.
Long
,
C. N.
, and Coauthors
,
2013
:
ARM research in the equatorial western Pacific: A Decade and counting
.
Bull. Amer. Meteor. Soc.
,
94
,
695
708
, doi:.
Lorenz
,
D. J.
, and
D. L.
Hartmann
,
2006
:
The effect of the MJO on the North American monsoon
.
J. Climate
,
19
,
333
343
, doi:.
Madden
,
R. A.
, and
P. R.
Julian
,
1971
:
Detection of a 40–50 day oscillation in the zonal wind in the tropical Pacific
.
J. Atmos. Sci.
,
28
,
702
708
, doi:.
Madden
,
R. A.
, and
P. R.
Julian
,
1972
:
Description of global-scale circulation cells in the tropics with a 40–50 day period
.
J. Atmos. Sci.
,
29
,
1109
1123
, doi:.
Marshall
,
J.
, and
W. M.
Palmer
,
1948
:
The distribution of raindrops with size
.
J. Meteor.
,
5
,
165
166
, doi:.
Mather
,
J. H.
, and
J. W.
Voyles
,
2013
:
The ARM Climate Research Facility: A review of structure and capabilities
.
Bull. Amer. Meteor. Soc.
,
94
,
377
392
, doi:.
Mather
,
J. H.
,
S. A.
McFarlane
,
M. A.
Miller
, and
K. L.
Johnson
,
2007
:
Cloud properties and associated radiative heating rates in the tropical western Pacific
.
J. Geophys. Res.
,
112
,
D05201
, doi:.
Matrosov
,
S. Y.
,
2008
:
Assessment of radar signal attenuation caused by the melting hydrometeor layer
.
IEEE Trans. Geosci. Remote Sens.
,
46
,
1039
1047
, doi:.
Matrosov
,
S. Y.
,
2009
:
A method to estimate vertically integrated amounts of cloud ice and liquid and mean rain rate in stratiform precipitation from radar and auxiliary data
.
J. Appl. Meteor. Climatol.
,
48
,
1398
1410
, doi:.
McFarlane
,
S. A.
,
J. H.
Mather
, and
T. P.
Ackerman
,
2007
:
Analysis of tropical radiative heating profiles: A comparison of models and observations
.
J. Geophys. Res.
,
112
,
D14218
, doi:.
Miller
,
M. A.
,
J.
Verlinde
,
C. V.
Gilbert
,
G. J.
Lehenbauer
,
J. S.
Tongue
, and
E. E.
Clothiaux
,
1998
:
Detection of nonprecipitating clouds with the WSR-88D: A theoretical and experimental survey of capabilities and limitations
.
Wea. Forecasting
,
13
,
1046
1062
, doi:.
Okamoto
,
H.
, and Coauthors
,
2008
:
Vertical cloud properties in the tropical western Pacific Ocean: Validation of the CCSR/NIES/FRCGC GCM by shipborne radar and lidar
.
J. Geophys. Res.
,
113
,
D24213
, doi:.
Protat
,
A.
, and Coauthors
,
2009
:
Assessment of Cloudsat reflectivity measurements and ice cloud properties using ground-based and airborne cloud radar observations
.
J. Atmos. Oceanic Technol.
,
26
,
1717
1741
, doi:.
Riihimaki
,
L. D.
, and
S. A.
McFarlane
,
2010
:
Frequency and morphology of tropical tropopause layer cirrus from CALIPSO observations: Are isolated cirrus different from those connected to deep convection?
J. Geophys. Res.
,
115
,
D18201
, doi:.
Riihimaki
,
L. D.
,
S. A.
McFarlane
, and
J. M.
Comstock
,
2012
:
Climatology and formation of tropical midlevel clouds at the Darwin ARM site
.
J. Climate
,
25
,
6835
6850
, doi:.
Schaefer
,
J. T.
,
1990
:
The critical success index as an indicator of warning skill
.
Wea. Forecasting
,
5
,
570
575
, doi:.
Schott
,
F. A.
, and
J. P.
McCreary
Jr
.,
2001
:
The monsoon circulation of the Indian Ocean
.
Prog. Oceanogr.
,
51
,
1
123
, doi:.
Shupe
,
M. D.
,
2007
:
A ground-based multisensor cloud phase classifier
.
Geophys. Res. Lett.
,
34
,
L22809
, doi:.
Stith
,
J. L.
,
J. E.
Dye
,
A.
Bansemer
,
A. J.
Heymsfield
,
C. A.
Grainger
,
W. A.
Petersen
, and
R.
Cifelli
,
2002
:
Microphysical observations of tropical clouds
.
J. Appl. Meteor.
,
41
,
97
117
, doi:.
Tokay
,
A.
, and
D. A.
Short
,
1996
:
Evidence from tropical raindrop spectra of the origin of rain from stratiform versus convective clouds
.
J. Appl. Meteor.
,
35
,
355
371
, doi:.
Turner
,
D. D.
,
S. A.
Clough
,
J. C.
Liljegren
,
E. E.
Clothiaux
,
K. E.
Cady-Pereira
, and
K. L.
Gaustad
,
2007
: Retrieving liquid water path and precipitable water vapor from the Atmospheric Radiation Measurement (ARM) microwave radiometers. IEEE Trans. Geosci. Remote Sens.,45, 3680–3690.
Vivekanandan
,
J.
,
G.
Zhang
,
S. M.
Ellis
,
D.
Rajopadhyaya
, and
S. K.
Avery
,
2003
:
Radar reflectivity calibration using differential propagation phase measurement
.
Radio Sci.
,
38
,
8049
, doi:.
Weickmann
,
K. M.
,
G. R.
Lussky
, and
J. E.
Kutzbach
,
1985
:
Intraseasonal (30–60 day) fluctuations of outgoing longwave radiation and 250 mb streamfunction during northern winter
.
Mon. Wea. Rev.
,
113
,
941
961
, doi:.
Xie
,
S.
, and Coauthors
,
2010
:
Clouds and more: ARM climate modeling best estimate data
.
Bull. Amer. Meteor. Soc.
,
91
,
13
20
, doi:.
Yoneyama
,
K.
,
C.
Zhang
, and
C. N.
Long
,
2013
:
Tracking pulses of the Madden–Julian oscillation
.
Bull. Amer. Meteor. Soc.
,
94
,
1871
1891
, doi:.
Zhang
,
C.
, and
S. M.
Hagos
,
2009
:
Bi-modal structure and variability of large-scale diabatic heating in the tropics
.
J. Atmos. Sci.
,
66
,
3621
3640
, doi:.
Zhang
,
C.
,
M.
Dong
,
S.
Gualdi
,
H. H.
Hendon
,
E. D.
Maloney
,
A.
Marshall
,
K. R.
Sperber
, and
W.
Wang
,
2006
:
Simulations of the Madden–Julian oscillation in four pairs of coupled and uncoupled global models
.
Climate Dyn.
,
27
,
573
592
, doi:.
Zhang
,
C.
,
J.
Gottschalck
,
E. D.
Maloney
,
M. W.
Moncrieff
,
F.
Vitart
,
D. E.
Waliser
,
B. Wang
, and
M. C.
Wheeler
,
2013
:
Cracking the MJO nut
.
Geophys. Res. Lett.
,
40
,
1223
1230
, doi:.
Zhao
,
C.
, and Coauthors
,
2012
:
Toward understanding of differences in current cloud retrievals of ARM ground-based measurements
.
J. Geophys. Res.
,
117
,
D10206
, doi:.
Zhou
,
S.
, and
A. J.
Miller
,
2005
:
The interaction of the Madden–Julian Oscillation and the Arctic Oscillation
.
J. Climate
,
18
,
143
159
, doi:.

Footnotes

*

Current affiliation: Climate and Environmental Sciences Division, U.S. Department of Energy, Washington, D.C.

This article is included in the DYNAMO/CINDY/AMIE/LASP: Processes, Dynamics, and Prediction of MJO Initiation special collection.