Multiyear Summertime Observations of Daytime Fair-Weather Cumuli at the ARM Southern Great Plains Facility

Transcription

1 15 DECEMBER 2013 C H A N D R A E T A L Multiyear Summertime Observations of Daytime Fair-Weather Cumuli at the ARM Southern Great Plains Facility ARUNCHANDRA S. CHANDRA AND PAVLOS KOLLIAS Department of Atmospheric and Oceanic Sciences, McGill University, Montreal, Quebec, Canada BRUCE A. ALBRECHT Division of Meteorology and Physical Oceanography, University of Miami, Miami, Florida (Manuscript received 10 April 2012, in final form 25 March 2013) ABSTRACT A long data record (14 yr) of ground-based observations at the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) site is analyzed to document the macroscopic and dynamical properties of daytime fair-weather cumulus clouds during summer months. First, a fuzzy logic based algorithm is developed to eliminate insect radar echoes in the boundary layer that hinder the ability to develop representative cloud statistics. The refined dataset is used to document the daytime composites of fairweather cumulus clouds properties. Doppler velocities are processed for lower reflectivity thresholds that contain small cloud droplets having insignificant terminal velocities; thus, Doppler velocities are used as tracers of air motion. The algorithm is implemented to process the entire 14-yr dataset of cloud radar vertical velocity data. Composite diurnal variations of the cloud vertical velocity statistics, surface parameters, and profiles of updraft and downdraft fractions, bulk velocity of updrafts and downdrafts, and updraft and downdraft mass flux are calculated. Statistics on the cloud geometrical properties such as cloud thickness, cloud chord length, cloud spacing, and aspect ratios are calculated on the cloud scale. The present dataset provides a unique insight into the daytime evolution and statistical description of the turbulent structure inside fair-weather cumuli over land. 1. Introduction Summertime fair-weather cumulus (FWC) clouds over land have typically limited vertical and horizontal extent (average cloud thickness and cloud horizontal length of ;1 km). FWC clouds are the visible manifestation of the vertical transport of momentum, heat, moisture, and pollutants from the surface into the free troposphere; thus, their life cycle is strongly coupled to the surface fluxes and important in modulating the diurnal cycle (e.g., Brown et al. 2002; Lemone and Pennell 1976; Betts and Viterbo 2005). FWC plays an important role in the atmospheric processes on various temporal and spatial scales. Locally, the presence of FWC can cause a substantial change in turbulence intensity and the internal structure of the convective boundary layer Corresponding author address: Arunchandra S. Chandra, Department of Atmospheric and Oceanic Sciences, 805 Sherbrooke Street West, Montreal QC H3A 2K6, Canada. arunchandra.chandra@mail.mcgill.ca (CBL; Stull and Eloranta 1984). Such a change in the CBL may further lead to subsequent variations in other processes or properties related to dispersion rates, chemical reaction rates, and the removal of air pollutants from the CBL (Isaac et al. 1984; Shipley and Browell 1984). Further, shallow cumulus convection provides one of the important mechanisms for the vertical redistribution of momentum and sensible and latent heat into the atmosphere by preconditioning the deep convection, thereby moistening the lower and middle troposphere (e.g., Tiedtke 1989; Chaboureau et al. 2004; Khairoutdinov and Randall 2006). The fractional cloud cover of fairweather cumulus clouds over land covers about 10% 30% (Weilicki and Welch 1986) of total surface area and it has a significant impact on the earth s radiation budget and the radiative fluxes in the boundary layer (e.g., Albrecht 1981; Stull 1985). Recent sensitivity studies have demonstrated that different shallow clouds parameterizations used in different models leads to different responses of these clouds in the climate system (Bony et al. 2006). The first step toward addressing these issues DOI: /JCLI-D Ó 2013 American Meteorological Society

2 10032 J O U R N A L O F C L I M A T E VOLUME 26 is to document long-term statistics of fair-weather cumuli to build a robust observational test bed. Several challenges remain in both modeling and observing FWC. Starting with large-scale models, the typical resolution of global climate models (GCMs) or even regional models is not adequate to resolve cloudrelated process (e.g., Zhu and Albrecht 2002) and thereby hinders our ability to accurately predict lowlevel cloud feedback in a changing climate (e.g., Tiedtke 1989; Bony et al. 2006). Parameterizations are used to quantify the cumulative effects of shallow clouds in unresolved scales. Efforts have been made to improve fair-weather cumuli parameterizations in models (e.g., Betts 1973; Albrecht 1979, 1981; Neggers et al. 2004; Berg and Stull 2005; de Rooy and Siebesma 2008). These efforts have increased both our observational basis to understand FWC processes and modeling capabilities for better understanding the nature of these clouds and their interaction with large-scale dynamics (Zhu and Albrecht 2003). Yet, some of the key issues such as subcloud cloud layer coupling, factors controlling the mass flux at the cloud base, and in-cloud turbulent structure remain poorly understood. Recent model studies have focused on increasing our understanding of the physics and dynamics of fairweather cumulus clouds over land (e.g., Zhu and Albrecht 2002; Ek and Holtslag 2004) and studies using large-eddy simulation (LES) and single-column models (SCMs) have been performed (e.g., Brown et al. 2002; Siebesma et al. 2003; Zhu and Albrecht 2003; Lenderink et al. 2004) with particular emphasis on the diurnal cycle and basic physics underlying FWC. The results from model comparison show a wide scatter, particular in the case of unsteady boundary layer (over land), which suggests that there is further need for improvement and development of parameterization schemes. Studies over land to understand the diurnal cycle of precipitation have emphasized the role of representing the processes within the convective boundary layer during dry and nonprecipitating conditions in order to understand shallow to deep convection transitions (e.g., Guichard et al. 2004; Zhang and Klein 2010). The total number of observational studies on shallow cumuli over land compared to ocean is significantly smaller. So there is a necessity for long-term observational studies on fairweather cumuli over land to serve as a robust dataset to address the above issues related to FWC. There have been several early observational aircraft studies (e.g., Warner 1977; Squires 1958; Blyth and Latham 1985; Stull et al. 1997; Vogelmann et al. 2012) on continental FWC clouds that provided invaluable information about the microphysical and dynamical structure of both CBL and FWC clouds. However, aircraft observations without on-board radar system are dimensionally challenged (Warner 1977). The aircraft penetrations give detailed horizontal structure, but the vertical structure needs multipass penetrations at many heights, which is limited because of the short lifetime of fair-weather cumulus clouds. Furthermore, the high cost associated with routine aircraft observations make their use challenging. Addressing these problems demands technologies to sample clouds with high-resolution capabilities beyond one dimension. In the last 20 years there has been substantial progress in cloud remote sensing with the development of sophisticated cloud radars, lidars, and microwave radiometers (e.g., Spinhirne 1993; Moran et al. 1998; Liljegren et al. 2001). The strength of cloud radar is its ability to detect cloud and its resolution capabilities, and its synergy with lidars, radiometers, soundings, etc., for comprehensive study of clouds. The potential to use cloud radars to study small cumulus has been demonstrated (e.g., Lhermitte 1987; French et al. 1999; Kollias and Albrecht 2000; Kollias et al. 2001). Here, 14 yr of observations from the millimeter wavelength cloud radar (MMCR) and other active and passive remote sensors at the Atmospheric Radiation Measurement Program (ARM) Southern Great Plains (SGP) site are used to compute the fair-weather cumuli macroscopic (horizontal and vertical extent) and dynamical properties of FWC. The MMCR Doppler velocity measurements are treated as representative of the average vertical air motion of the MMCR sampling volume based on the assumption that cloud droplets have negligible fall velocities and thus can be treated as tracers of a vertical air motion (e.g., Kollias and Albrecht 2000). The presented analysis is distinctly different from previous recent studies that utilized data from the same ARM site (e.g., Berg and Kassianov 2008; Vogelmann et al. 2012). Berg and Kassianov (2008) focused on the FWC cloud fraction, associated surface fluxes, and their diurnal and interannual variability using 5 yr of observations ( ). Recently, the Routine ARM Aerial Facility (AAF) Clouds with Low Optical Water Depths (CLOWD) Optical Radiative Observations (RACORO) field campaign provided extensive aircraft sampling of FWC at the SGP site (Vogelmann et al. 2012). RACORO focused on issues such as radiative impacts of boundary layer clouds, aerosol cloud interaction, and surface-based retrievals using aircraft measurements of cloud microphysics, aerosols, and state parameters. The present study covers 14 summertime periods with statistics of cloud thickness and emphasis on the turbulent structure (e.g., profiles of updraft and downdraft properties such as fraction, bulk velocity, and mass flux) inside fair-weather cumuli and its diurnal variability. The climatology

3 15 DECEMBER 2013 C H A N D R A E T A L TABLE 1. List of instruments used in this study along with their specifications. Instrument Resolution Measurements Accuracy Remarks MMCR 10 s, 45 m Reflectivity 0.5 db Vertically pointing Doppler velocity 0.1 m s 21 Spectrum width 0.1 m s 21 WACR ;2 s, 45 m Reflectivity 0.5 db Vertically pointing Doppler velocity 0.1 m s 21 Spectrum width 0.1 m s 21 MWR ;20 s Brightness temperature 0.3 K Vertically pointing Cloud water path ;20 g m 22 Radiosonde Every 6 h Temperature 0.58C Relative humidity ;5% Pressure 0.05 kpa Wind speed ;0.5 m s 21 Wind direction 108 EBBR Every 30 min Sensible heat flux 10% uncertainty Latent heat flux 10% uncertainty Net radiation 5% uncertainty Average soil surface heat flux 10% uncertainty TSI 30 s Hemispheric sky image Reliable when solar elev.108 Fractional sky cover documented in this study will serve as a complementary dataset along with RACORO and Berg and Kassianov (2008) studies for evaluating parameterizations and to address different issues (entrainment, cloud life cycle, cloud subcloud layer interactions, cloud aerosol interactions, etc.) related to FWC. Furthermore, the present study also attempts to overcome the challenge in detecting FWC using the MMCR because of the presence of strong insect echoes in the cloud layer (e.g., Clothiaux et al. 2000; Luke et al. 2008; Chandra et al. 2010). A fuzzy logic based algorithm that removes false misclassifications of insect radar returns as clouds is presented and applied to the 14 yr of MMCR observations. The improved MMCR dataset and in particular the Doppler velocities are used to develop diurnal composite profiles of hourly turbulent statistics. Fair-weather cumuli geometrical properties such as chord length, cloud thickness, aspect ratio, and cloud spacing are computed. 2. Measurements The data used in the present study are taken from the ARM archive (available online at data). Unless otherwise stated, the measurements are taken at SGP central facility. The instruments used in this study and their specifications are listed in Table 1. Specific data information is described below. The MMCR and W-band ARM cloud radar (WACR) operate at short wavelengths (MMCR at 8.66 mm and WACR at 3 mm) that have potential for probing cloud and ice particles ranging from a few micrometers in diameter to precipitating drops. These systems have capabilities to resolve approximately 45 m in the vertical and about 10 m for 2 km in the horizontal. The output radar moments are reflectivity, Doppler velocity, and spectrum width, which are useful to infer the dynamical and microphysical structures of clouds (e.g., Lhermitte 1987; Frisch et al. 1994; Kollias and Albrecht 2000). The data used for the cloud boundaries and hydrometeor detections are from the Active Remote Sensing of Clouds (ARSCL) value-added product (VAP). The ARSCL VAP combines the data from active remote sensors (laser ceilometer, micropulse lidar, and MMCR) to produce an objective determination of hydrometeor height distributions and estimates of radar moments (Clothiaux et al. 2000). ARSCL VAP is available with 10-s temporal resolution and 45-m vertical resolution. The ARSCL instruments are collocated at the field site within 50 m from each other, and all other instruments are within 100 m from each other. So the synergetic observations used in this study are fairly reasonable for time averages greater than 10 s (assuming advective winds ;10 m s 21 within the boundary layer). A collocated microwave radiometer (MWR) and an energy balance Bowen ratio (EBBR) system provide supplemental observations. The EBBR flux system at the SGP ( central facility provides 30-min estimates of sensible and latent heat at the local surface. The surface buoyancy flux is estimated from the heat and moisture fluxes at the surface along with mixing ratio and temperature data using Eq. (4.4.5d) from Stull (1988). This technique determines the Bowen ratio (ratio of sensible to latent heat

4 10034 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 1. Yearly variability of (a) number of FWC hours classified based on ceilometer detections during summertime (May August) and (b) summertime CWP distributions. The value at the box notch indicates the median value of CWP, and the top and bottom ends of the box indicate the 25th and 75th percentile bounds. The crosses denote data points larger than the 75th percentile bound. flux) assuming the same transfer coefficients for both heat and water vapor. The uncertainty in the surface flux estimation from this method is about 10 W m 22 and is mainly due to the errors associated with transfer coefficients. The Bowen ratio is then used in conjunction with net radiation and soil heat flux measurements to calculate sensible and latent heat flux based on a budget approach (e.g., Brutsaert 1982). MWR is a passive remote sensing instrument ( mwr) that provides measurements of column-integrated amounts of water vapor and liquid water at 20-s resolution. The noise threshold for cloud water path (CWP) measurements from MWR is approximately 20 g m 22. The uncertainty (noise threshold) for MWR is defined based on the signal threshold [equivalent to liquid water path (LWP) of ;20 g m 22 ] where one can reliably distinguish the mean radiation temperatures of clear air from the cloudy portion. Soundings (available online at provide the measurements of temperature, relative humidity, and wind speed and direction with height for every 6 h. Only summertime observations (defined as the 4-month period from May to August; 123 days per season) from the 14-yr period from 1997 to 2010 are used in this study. Furthermore, the analysis is limited to the daytime period defined as the 9-h period from 0900 to 1800 LST. Every hourly block of ARM observations is screened for the occurrence of fair-weather cumulus clouds (objectively defined as the cloud tops lower than 4 km) following the methodology used in Chandra et al. (2010), both by using extensive visual inspection of the total sky imager (TSI) images and by screening out periods with hourly surface buoyancy fluxes less than 50 W m 22. Rain gauge measurements and MMCR first detection height and intensity information are used to sample only the nonprecipitating clouds. Overcast conditions (stratus or stratocumulus) are filtered out using the TSI and ceilometer. The interannual variability of the occurrence of hourly periods with fair-weather cumuli and CWP during the summertime period is shown in Fig. 1. The total number of hours with FWC that can be observed during the sampled daytime summertime period is The fraction of hourly periods with FWC is based on the laser ceilometer measurements and thus is not subject to limitations in instrument sensitivity due to small amounts of liquid or particle size. Noticeable variability (from a 2% 4% minimum to a 12% 17% maximum) is observed from year to year with the maximum occurrence observed almost with a 4-yr cycle (2002, 2006, and 2010). In addition, the season-averaged CWP observed by the MWR during FWC conditions is shown (Fig. 1b). The value at the box notch in Fig. 1b indicates the median value, and the bottom and top edges of the box

5 15 DECEMBER 2013 C H A N D R A E T A L FIG. 2. (a) CWP distributions inside FWC for the periods when clouds are detected from the ceilometer and from MMCR. (b) Hourly cloud fractions from the ceilometer detections and the maximum cloud fraction detected from the MMCR. The vertical bars in (b) indicate 1-standard-deviation values. show the 25th and 75th percentile values. The twochannel microwave radiometer has a detection threshold of g m 22. For every hour of FWC observations, only time periods when the ceilometer detects a cloud are used to estimate the hourly CWP value. It is clear from the figure that the summertime shallow FWC contains small amounts of liquid water content (LWC). Although visual inspection of seasonal CWP distributions seems to have a trend, statistical test shows that the trend is insignificant. 3. Challenges in shallow cumuli detection a. Low radar reflectivities of continental shallow cumulus clouds The small amounts of LWC combined with typical continental aerosol loading conditions (e.g., Feingold et al. 2003; Vogelmann et al. 2012) result in clouds with small liquid droplet radius. This was confirmed during a recent 5-month (February June 2011) systematic aircraftbased field experiment: RACORO. The analysis of the aircraft in-site probes revealed that the cloud effective radius is between 3 and 7 mm withpreference toward the lower limit (e.g., Vogelmann et al. 2012). However, the small particle sizes also result in a larger fraction of FWC detected by the ceilometer rather than by the MMCR. The distribution of ceilometer-only and radar-only detections of FWC as a function of their observed CWP is shown in Fig. 2. Assuming that the ceilometer is capable of detecting all overpassing clouds, we can treat the ceilometer-only distribution as the true distribution of shallow cumuli detections as a function of their CWP. As discussed, the CWP distribution peak at g m 22 suggests clouds with very limited vertical development. Despite MMCR s excellent sensitivity (245 dbz at 2 km; Moran et al. 1998; Kollias et al. 2007), it misses the majority of shallow clouds with CWP, 50 g m 22 (Fig. 2a). Thus, the MMCR-based statistics presented here are representative only for a portion of the population of FWC developed over the ARM SGP site. Figure 2b shows the hourly comparison of ceilometeronly cloud fraction and cloud fraction. The ceilometeronly cloud fraction is based on the detection of cloud edge (either cloud base or tilted lateral boundary). The highest cloud fraction (;15%) is observed early in the day and the cloud fraction gradually decreases below 10% in late afternoon. The radar-only cloud fraction is the maximum cloud fraction observed by the radar with in the cloud layer defined by its ceilometer-defined cloud base and radar-defined cloud top. Earlier studies (e.g., Neggers et al. 2003) reported the significant difference in one-dimensional (1D) measurements (e.g., ceilometer or airplane) of cloud size versus 2D measurements (e.g., satellite or LES). Despite these differences, the radar-only cloud fraction is substantially

6 10036 J O U R N A L O F C L I M A T E VOLUME 26 lower (;5%). The observed differences in the cloud fraction by these two collocated active sensors are consistent with previous studies that have highlighted the challenge in objectively determining cloud fraction (e.g., Wu et al. 2011). Nevertheless, the MMCR misses a significant portion of the FWC clouds at the ARM SGP site. b. Identification of insect echoes that have been misclassified as cloud echoes In addition to the MMCR sensitivity, another challenge that hinders our ability to document FWC clouds at the ARM SGP is insect echoes that overwhelm the MMCR returns in the lowest 2 3 km (e.g., Clothiaux et al. 2000; Geerts and Miao 2005; Luke et al. 2008; Chandra et al. 2010) during the warm season. The ARSCL VAP is the most popular data product in the ARM archive and it is often used as the starting point for deriving cloud macroscopic (e.g., cloud climatology) as well as cloud microphysical and dynamical properties. As part of the ARSCL standard quality control, insect contamination is addressed by classifying MMCR echoes as either insects or mixture of cloud and insect returns. No automated algorithm is available, and often manual editing is required. In spite of the extensive quality control efforts, there is considerable misclassification of insect echoes as cloud echoes and the opposite. This is often the case when the ceilometer detects an FWC base and there is little information to enable classification of the radar echoes above the ceilometer-derived cloud base. The MMCR echoes above the cloud (base) can be (i) all cloud-related returns (in the absence of insects and if the radar reflectivity of the hydrometeors is above the sensitivity of the MMCR, (ii) cloud returns near the cloud base if the cloud is shallow and insect returns above, and (iii) insect returns only if the radar reflectivity of the hydrometeors is below the sensitivity of the MMCR. Thus, additional screening is required before the MMCR data can be used for the estimation of FWC properties (e.g., cloud-top height and updraft and downdraft properties). Two representative examples of insect contamination in the ARSCL VAP are shown in Fig. 3. The top panel shows the time height plot of MMCR reflectivity as reported in ARSCL. The first example [Fig. 3a(1)] shows broken shallow cumulus with the ceilometerdefined cloud base and MMCR-defined cloud top shown in black. The second example [Fig. 3a(2)] is from a shallow, continuous MMCR echo. All MMCR echoes below the ceilometer-defined cloud base have been removed and classified as insects. The corresponding WACR copolar channel observations are shown in Figs. 3b(1) and 3b(2). The WACR detects almost the same echoes as the MMCR in the case of the broken cumulus [Fig. 3b(1)] and no echoes in the cases of the low stratiform-like cloud [Fig. 3b(2)]. The small offset in the radar reflectivity values of the MMCR and WACR is due to water vapor and liquid attenuation and/or small calibration offset. At W band, non-rayleigh scattering suppresses the intensity of the insect radar returns by 20 db on average compare to their respective value at Ka band [see Fig. 13 in Luke et al. (2008)]. Thus, the WACR detects a shallower insect layer and lower insect radar reflectivity, and in some cases, depending on the insect layer characteristics, the WACR detects no insect echoes if their intensity is below the sensitivity of the radar (see Fig. 4) The WACR echoes in the first case [Fig. 3b(1)] are from hydrometeors and this is verified by the absence of WACR cross-polar signal [Figs. 3c(1) and 3c(2)]. The CWP time series [Figs. 3d(1) and 3d(2)] further support the provided interpretation of the MMCR and WACR echoes. Significant values of CWP values are reported during the FWC periods compared to the intermitted clear-sky periods [Fig. 3d(1)] whereas no such variability is observed during the second case [Fig. 3d(2)]. The easiest approach to separate insect echoes from cloud is to use the WACR capability to detect insect from cloud echoes. However, the WACR data were very limited (only a couple of years of intermittent operations) compared to the long-term (14 yr) observations from MMCR. Luke et al. (2008) developed a technique to remove insects based on radar Doppler spectra. However, the lack of MMCR Doppler spectra before 2005 and the significant computational resources limited the application of the aforementioned technique to the present dataset. Therefore, a new approach is required. A new method based on a fuzzy logic (FL) technique is developed to separate insects from cloud echoes. A flowchart showing the sequence of steps used in the FL algorithm to separate insects from cloud echoes is shown in Fig. 4. The WACR dataset is used as a reference to choose the test cases containing insect and cloud echoes. Careful analysis of the MMCR observations indicates that there are several variables that have the potential to assist in the discrimination of insect and cloud echoes above the ceilometer-derived cloud bases. Such variables are (i) the observed CWP time series from the MWR, (ii) the ratio of the CWP to the square of the reported cloud thickness from the ceilometer-derived cloud-base height and MMCR-derived cloud-top height, (iii) the local variability of the MMCR reflectivity, and (iv) the observed Doppler spectrum width. For insectcontaminated MMCR echoes above the ceilometer base, the CWP values are either at the MWR sensitivity level (30 50 g m 22 ) or exhibit a very small increase that indicates that the cloud echoes are not detectable by the MMCR. Similarly, the ratio of the observed CWP to the

7 15 DECEMBER 2013 C H A N D R A E T A L FIG. 3. Two sample cases showing (left) dominant cloud and (right) insect echoes for (a) ARSCL reported reflectivity (dbz), (b) WACR copoloralization reflectivity (dbz), and (c) CWP values from MWR. The black dots in the panels indicate the cloud boundaries (cloud bases and tops). (d) ARSCL flags (1 for pure cloud echoes and 2 for insect and cloud echoes) and (e) Fuzzy logic flags (1 for pure cloud and 2 for insects). square of the MMCR echo thickness is very small in insect-contaminated layers and higher for layers that contain significant liquid. Furthermore, the reflectivity texture of insect echoes is very spotty compared to the cloud echoes, which results in a higher variability in the reflectivity field, and the observed MMCR Doppler spectrum width values from insects are very small (;0.1 m s 21 ) compared to those reported in cloud layers (e.g., Luke et al. 2008). Using test cases from WACR dataset, the range of variability in each variable for insect and cloud

8 10038 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 4. Flowchart showing the sequence of steps used in the FL algorithm for separating insect echoes from cloud echoes. cases are used to formulate the membership functions (MFs: an MF is a curve that defines how each value in the input space is mapped between 0 and 1 depending on its value in insect/cloud regime) as shown in Fig. 5. Figures 5a d correspond to standard deviation of reflectivity (MF1) values, median of spectral width values (MF2), cloud water path (MF3), and the ratio of cloud water path (from MWR) to the cloud thickness (MF4) at each gate with the surrounding nine grid points. Here, the FL technique is applied to the radar echoes only when there is a valid cloud base from the ceilometer for a more conservative approach. Different weights are assigned to the membership functions based on their individual performance to separate insects from cloud echoes. The higher the weight, the more effective the parameter in separating insects from cloud echoes. The combination of these parameters is used in FL algorithm to flag insect/cloud regimes by assigning different weights (0.8 for MF1, 1.0 for MF2, 1.2 for MF3, and 1.3 for MF4) for the membership functions. The output decision flag (insect/cloud) is calculated based on the weighted average of the membership values. The generalization of FL algorithm may be restricted to particular frequency radar (in our case, Ka band), but the methodology adopted here can be applicable to other systems. Figures 3d and 3e show the comparison of output flags (cloud/insects) reported in ARSCL and FL for the cases showcased in Fig. 3b. It can be observed that there is a misclassification of pure cloud echoes being reported as cloud and insects [as in Fig. 3d(1); cf. Fig. 3e(1)] and insect echoes being reported as pure cloud echoes [as in Fig. 3d(2); cf. Fig. 3e(2)] from the ARSCL quality control. Thus this sets a necessity for an additional screening to process MMCR data. It is to be noted that there were no flags reported for most of the grid points in Fig. 3e(2). This is because the FL algorithm needs a finite number of continuous MMCR profiles (three profiles) in order to estimate the membership parameters and thus assign an insect/ hydrometeor flag. When we actually zoom in the right panel, the cloud bases are very sparse and wherever there are continuous cloud bases, the algorithm computed the flags. On applying the FL algorithm, the percentage of data containing insect echoes is quantified as a function of daytime hours and summer months. The volumetric insect fraction (defined as the ratio of the number of radar gates with insect contamination to the total number of radar gates inside the cloud in an hour) values are high between morning and noon hours with a mean and 1s values of 18% and 16%, respectively. On a monthly basis, the volumetric insect fraction values are high during the month of June with mean and 1s values of 15% and 12%, respectively. The FL algorithm is applied to the entire 14 yr of the MMCR dataset for calculating the vertical velocity statistics. Figure 6 shows the hourly variability in the original ARSCL reported cloud tops (before applying the FL algorithm) as a function of hourly volumetric insect fraction. It suggests that the echo layers with more insect contamination have lesser cloud-top variability than the pure cloud echoes. Although this parameter can be used to discriminate the echo layers with dominant insects versus cloud echoes, the extent of insect contamination inside clouds cannot be quantified. This parameter can be used to screen the insect layers from the cloud layers. Additional verification for the lack of large drops in the shallow cumuli is provided by the statistics of the observed radar reflectivities during FWC conditions. Figure 7 shows the frequency distribution of insectremoved MMCR reflectivity values observed inside FWC clouds as a function of time of day. The plot suggests that the fraction of drizzle particles present in the clouds is negligible and that almost all MMCR observations can be used for the derivation of in-cloud vertical velocity statistics. Furthermore, a decreasing trend in the FWC

9 15 DECEMBER 2013 C H A N D R A E T A L FIG. 5. Graphs showing the membership functions for each decision parameter. (a) Standard deviation of reflectivity values, (b) median of spectral width values, (c) CWP, and (d) ratio of cloud water path to the cloud thickness. radar reflectivities is observed, with higher values during morning hours and lower values during late afternoon. The higher values of reflectivities during morning hours may be as a result of the larger cloud droplet sizes. A total of 1040 h of observations of fair-weather cumuli over the 14-yr summer period ( ) are used for the analysis. The hourly statistics calculated from the insect-screened MMCR Doppler moments are classified with respect to their local standard time (LST) from 0900 to 1800 and used to provide summary daytime composites. The number of FWC hours available for each LST hourly period and the height distribution of data availability are shown in Fig. 8. Figure 8a shows the fractional radar echoes available at each height for every hour considering ceilometer cloud-base detections as a reference. Figure 8b shows the fraction of cloud echoes available out of the total radar echoes after applying the fuzzy logic algorithm. So the missing data fraction is attributed to echoes as follows: (i) insects, (ii) the fraction of cloud echoes the ceilometer misses when smaller clouds (chord length,; m) move over the radar only, (iii) drizzle fraction, and (iv) the portion of missed echoes resulting from algorithm sampling limitations. Among these, the contribution from the last two factors is quite small. The missing fraction because of insect echoes is significant as discussed above. The missing contribution due to smaller clouds cannot be quantified because of instrument limitations and observational constraints. The distribution of MMCR FWC echoes FIG. 6. Plot of hourly volumetric insect fraction vs hourly cloudtop standard deviations. Each point on the graph shows the median value of hourly cloud-top standard deviations at particular volumetric insect fraction bin (bin size of 0.2).

10 10040 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 7. Frequency distributions of insect-clutter removed MMCR reflectivity values (bin size of 2 dbz) observed inside FWC for different time of the day. screened for insects using the fuzzy logic technique is skewed toward the base of the cloud layer during LST, suggesting very shallow cloud layers (Fig. 8a). Such a clustering of the MMCR FWC echoes is not observed during the afternoon hours. This could be due to the source of cloudiness being comprised of clouds with different stages of their life cycle. The majority of the cumulus events are observed from 1200 to 1600 LST (Fig. 8c) with significantly lower hourly occurrences during the late morning hours ( LST). This is in agreement with the deepening of the CBL during the day to levels above the lifting condensation level (LCL) of the mixed layer (Figs. 8a,b). The ceilometer-derived cloud-base heights can be used to derive unbiased statistics about the frequency of occurrence of FWC clouds. However, this is not the case with the FWC cloud-top height. The use of the ARSCL dataset can lead to significant biases with respect to FWC cloud-top height variability (Fig. 9b). The ARSCLderived daytime variability of FWC cloud-top height overestimates the FWC layer thickness from 1000 to 1100 LST and underestimates throughout the rest of the day compared to the FL (Fig. 9a). Early in the day, the insect layer is deeper than the FWC layer. As the day progresses, the clouds grow deeper with the solar insolation compared to the relatively flat insect layers, resulting in an underestimate of the cloud-top heights. The cloud thickness versus CWP relation is extracted using the ARSCL and the FL dataset cloud-top height and the MWR CWP measurements provides additional support for the inferred cloud-top biases in ARSCL. Figure 9d shows the relationship between the median of the square of the FWC cloud thickness versus their corresponding CWP bins (bin size of 50 g m 22 ). It is clear FIG. 8. (a) Time height composite of hourly hydrometeor echoes from MMCR available during FWC conditions. (b) Time height composite of the fraction of the MMCR echoes shown in (a) that are from hydrometeors after applying the FL algorithm. (c) Time height composite and (d) total number of available hours with FWC observations as a function of daytime hours based on the ceilometer detections.

11 15 DECEMBER 2013 C H A N D R A E T A L FIG. 9. Daytime composite variation of (a) corrected cloud tops (after applying FL algorithm), cloud base, and LCL values. (b) Original (ARSCL) reported cloud top, cloud base, and LCL values. (c) Daytime evolution of CWP and number of FWC hours available for different times of the day. (d) Plot of CWP (bin size of 50 g m 22 ) vs the median of the square of the FWC cloud thickness (m 2 ) before and after insect removal. Black circles in (a) and (b) indicate hourly mean values and gray shaded areas represent hourly standard deviation values. The vertical bars in (a) and (b) indicate hourly standard deviation values for cloud base and LCL values respectively. that the use of the FL dataset results in an improved near-linear relationship that is consistent with clouds that experience only condensational growth and evaporation without considering the effect of entrainment. The ceilometer-reported cloud bases and LCL values show strong diurnal variation with a minimum value occurring during the morning and a maximum during late afternoon ( LST). The daytime variation of CWP values (Fig. 9c) shows a maximum value of 120 g m 22 between 1300 and 1600 LST consistent with the deepening of the cloud layer. The ARSCL-derived data produce unreasonably deep FWC layers for low CWP values (,50 g m 22 ) and slightly underestimates for higher CWP values. In addition to this, insect contamination does have considerable impact on the cloud fraction and updraft mass flux profiles with daily averages being more affected compared to the monthly and seasonally averages. On daily basis, the bias in the cloud fraction profiles varies from 0% to 60% and for the mass flux profiles it varies between 0% and 30% for various insect fractions. Thus, the insect contaminations have considerable impact on both ARSCL reported cloud tops and kinematic properties, specifically for fair-weather cumulus clouds. 4. Results a. Fair-weather cumuli geometrical properties The cloud geometrical properties such as cloud chord length, cloud spacing, aspect ratio, and cloud thickness are very important for radiative transfer calculations in

12 10042 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 10. Histograms of (a) cloud chord diameter (bin size of 200 m), (b) cloud aspect ratio (bin size of 0.25), (c) wind speed averaged over the cloud depth (bin size of 1 m s 21 ), (d) cloud thickness (bin size of 150 m), and (e) cloud spacing (bin size of 250 m). shortwave cloud radiation parameterization (e.g., Lane et al. 2002; Jiang et al. 2008). The statistics on cloud geometrical properties are calculated on cloud scale (on individual clouds; total of 5311 clouds) using all fairweather cumuli hours. Here, geometrical properties are estimated for clouds (total of 5311 cloud samples) that produced at least five consecutive ceilometer detections and have FL-derived cloud thickness of 200 m minimum. Since the ceilometer does not differentiate between actual cloud base and cloud edges, it is necessary to conditionally define the mean cloud base, defined as the lowest 25% of the ceilometer cloud detections. The average of all the reported cloud-top heights with in a cloud event is used to derive the cloud mean cloud top. Chord lengths are calculated by transforming the continuous cloud detections from the ceilometer to the horizontal length scale using the winds (from the nearest sounding data) averaged over the cloud depth. The horizontal spacing between clouds is calculated similarly by using the mean winds averaged (winds from the soundings at the closest time) over the cloud depth. The cloud thickness values are calculated from the mean cloudbase and cloud-top heights. Using all available data, summary distributions of the observed cloud chord length, cloud thickness, aspect ratio, cloud spacing, and horizontal wind speeds averaged over cloud layer depth are estimated (Fig. 10). The bin spacing of the cloud chord diameter L is 200 m (Fig. 10a), and 60% of the observed clouds have a chord length less than 500 m. This is consistent with observed horizontal scales of CBL eddies (e.g., Chandra et al. 2010). The maximum cloud chord length observed in this study is approximately 5 km. The distribution of observed cloud aspect ratio (L/D: ratio of cloud chord length to the cloud thickness) in Fig. 10b indicates that 60% of the clouds have an aspect ratio less than 1. These reports are very intriguing as cumulus clouds are generally assumed to have aspect ratios of 1 or above. Detailed inspection of

13 15 DECEMBER 2013 C H A N D R A E T A L cloud shapes shows that the clouds tend to resemble towers with lower aspect ratios preferably between the early and late morning hours. Although this explanation offers a plausible reason for the present reports, it is very difficult to verify because of the assumptions and observational limitations in estimating aspect ratios. It could be possible that the lower aspect ratios may be due to the underestimation of cloud diameters or/and bias in transforming time to length scale due to the lack of highfrequency horizontal winds. The MMCR sensitivity can also offer a plausible explanation for the observed low aspect ratio. Because of its limited sensitivity, our sample does not include the numerous shallow clouds with large aspect ratio and it is weighted by the fewer thicker cumulus that are detected by the MMCR and have lower aspect ratio. At this point, we can only offer plausible reasons. The future scope is to verify the validity of these reporting using models or observations that offer highfrequency horizontal observations. The distribution of horizontal winds averaged over cloud depth is shown in Fig. 10c. The mean winds have peaks between 7 and 10 m s 21. The distribution of the FL-derived cloud thickness D is shown in Fig. 10d. The bin spacing of the cloud thickness is 150 m. The distribution peaks at m. Finally, the distribution of horizontal spacing between clouds is shown in Fig. 10e. The most frequent occurrence of the cloud spacing values is m and drops exponentially. Figure 11 shows a plot of normalized cloud chord diameter density (defined as the probability of occurrence of clouds at the range of the cloud chord sizes) versus cloud chord diameter values. Past studies used aircraft observations and LES model output (e.g., Benner and Curry 1989; Cahalan and Joseph 1989; Neggers et al. 2003; Berg and Kassianov 2008) to develop a functional relationships to express the normalized cloud chord density as a function of cloud chord diameter. The large cloud samples (5311 clouds over 14 years) from this study provided an opportunity to examine previous functional relationships. The observed distribution (circles) agrees well with the previously suggested power-law fit, and there is a robust feature of clear scale break at about 1000 m, which compares well with the derived values of Neggers et al. (2003) (number of clouds ) using LES of cumulus over land, and aircraft observations from Jiang et al. (2008) (number of clouds 5 92). The exponent of the power law is lower ( ; the mean value is the daytime average and the value after the 6 sign indicates the standard deviation) compared with values from the previous studies [22.3 and 21.9 for the aircraft and model study of Jiang et al. (2008); 21.7 for the LES study by Neggers et al. (2003); and 22.0 for Benner and Curry (1989)]. This may be as a result of large sampling of FIG. 11. Plot of normalized cloud chord diameter density vs cloud chord diameter (bin size of 150 m). Circles denote the cloud chords calculated from ceilometer data. Solid and dashed lines denote exponential and power-law fits to the data. (inset) The mean values indicate the average of exponents b for the exponential and power-law fits to the data when the cloud chord length is decomposed for different time of the day, and the value after the 6 sign indicates standard deviation. The logarithmic representation of the distribution is shown by inset in the top-right corner. clouds covering different meteorological conditions over 14 years. In general, comparing cloud chord diameters from different platforms (aircraft, ceilometer, and satellite) involves bias due to errors in sampling (Neggers et al. 2003). b. Turbulence statistics Using the 1040 hours of FWC identified over 14 years, daytime hourly composites of surface and FWC parameters are generated (Fig. 12). A minimum hourly data fraction of 5% is used to calculate the hourly vertical velocity statistics. The FWC cloud-base updraft mass flux is estimated using the method outlined in Kollias and Albrecht (2010). Significantly higher updraft mass flux (updraft mass flux 5 air density 3 updraft fractional area 3 bulk updraft velocity) values are observed between 1200 and 1400 LST with a magnitude of kg m 22 s 21 (Fig. 12a). Lower values ( kg m 22 s 21 ) are observed during the morning and late afternoon periods. The magnitude and the timing are comparable to the recent FWC study by Kollias and Albrecht (2010), who studied island-induced FWC at the ARM tropical western Pacific site at Nauru. The updraft mass flux maximum coincides in time with the daytime maximum in the updraft fraction maximum and the daytime surface buoyancy flux maximum as shown in Figs. 12d and 12e. Figure 12b shows the net mass flux at the cloud base. The daytime variability in median values

14 10044 J O U R N A L O F C L I M A T E VOLUME 26 FIG. 12. Hourly daytime composites during FWC condition for the period for (a) cloud-base updraft mass flux (kg m 22 s 21 ); (b) cloud-base net mass flux (kg m 22 s 21 ); (c) cloud-base bulk upward velocity (m s 21 ); (d) cloud-base updraft fraction; and (e) surface sensible heat (SHF), latent heat (LHF), and buoyancy fluxes (BF) (W m 22 ). Circles with black center in (a) (d) indicate median values and the corresponding gray vertical bars depict the 25th and 75th percentile values. The isolated open gray circles indicate data points outside 25th and 75th percentile limits. of net mass flux follows cloud-base updraft mass flux (Fig. 12a). The net mass flux value shows positive (upward) around 1300 LST and net negative (downward) value around 1500 LST; for all other daytime hours, the net flux values are close to zero. The error bars associated with the net mass flux (Fig. 12b) for different hours clearly suggest that the clouds sampled at different hours are mixture of clouds from all stages of their life cycle. Figure 12c shows the hourly composite daytime variation of the mean updraft velocity at the cloud base. The maximum cloud-base updraft velocity ( m s 21 )is observed at LST, which is comparable to the study by Kollias and Albrecht (2010). Figure 12d shows the daytime variation of the updraft fraction (fraction of updraft portion of the cloud to the total clear and cloudy area in an hour). The maximum in the mean updraft fraction (;6%) is also observed between 1300 and 1400 LST. Figure 12e shows the daytime variation of the surface sensible heat flux (SHF), latent heat flux (LHF), and buoyancy flux [BF; calculated using Eq. (4.4.5d) from Stull (1988)]. The buoyancy flux reaches a maximum of about 300 W m 22 between 1300 and 1400 LST. The latent heat flux exceeds the sensible heat flux with an average Bowen ratio of Figure 13 shows hourly cloud-base and cloud-top vertical velocity statistics over the entire observing period. Every hour, velocities are sampled inside the cloud at the average locations of cloud base and top. The normalized

15 15 DECEMBER 2013 C H A N D R A E T A L FIG. 13. Histograms of hourly mean cloud-base (CB) and cloud-top (CT) statistics from all the FWC hours. (a) Mean vertical velocity (bin size of 0.2 m s 21 ), (b) vertical velocity standard deviation (bin size of 0.2 m s 21 ), and (c) vertical velocity skewness (bin size of 0.3). distributions of the hourly-averaged mean vertical air motions at the cloud-base and cloud-top heights are centered on zero (Fig. 13a). The vertical air motion standard deviation values ranges between 0.4 and 1.3 m s 21 at the cloud base and the cloud top with maximum occurrences between 0.6 and 0.9 m s 21 (Fig. 13b). To avoid sampling errors for computing statistics, a minimum cloud fraction of 5% is specified. Assuming there is no bias resulting from sampling, the normalized distribution of the vertical air motion skewness at the cloud base and tops (Fig. 13c) indicates both positive and negative values, with a preference for positive skewed vertical air motion distributions at the cloud top and negative skewed vertical air motion distributions at the cloud base. The entire dataset composites of hourly-averaged profiles of updraft and downdraft mean properties are shown in Fig. 14. Each composite profile is a result of averaging the hourly profiles over cloud depth where there are significant hourly profiles (at least 20 profiles) available for computing the statistics. The top and bottom locations in each profile indicate the average locations of cloud-base and cloud-top heights in that hour. At each hour, there is a significant variability of cloud base and top (as shown in Fig. 9), indicating that clouds are sampled from different height locations. The updraft fraction is computed as the fraction of the hour with observed in-cloud updraft. All the profiles shown in Fig. 14 are mean averaged profiles, and they show a weak daytime variability. The maxima in updraft and downdraft fraction profiles are observed just above the cloudbase height (Figs. 14a,d), and their magnitude varies considerably during the daytime with the maximum (0.045) observed between 1200 and 1400 LST that coincides with the maximum cloud fraction (Fig. 8b). The shifting of the maximum in profiles of updraft and downdraft fractions (Figs. 14a,d) is due to the evolution of average cloud-base and cloud-top locations (Fig. 9). Overall, the profiles of updraft and downdraft fractions are similar in shape and comparable in magnitude. The corresponding profiles of updraft and downdraft air motion magnitudes are shown in Figs. 14b and 14e. The weakest magnitudes for both updrafts and downdrafts are observed near the cloud base ( m s 21 ) with the exception of the period LST. The bulk updraft and downdraft velocities shows an increasing trend in the cloud layer from cloud base until near to the cloud top, and thereafter a decreasing trend up to the cloud top, which may be a result of cloud-top entrainment. The profiles of updraft and downdraft mass flux (Figs. 14c,f) show no clear trend. It is not straightforward to compare the results from previous LES studies (case studies) with the present results (composites) as there are some significant differences between them. For example, in previous model studies (e.g., Lenderink et al. 2004; Brown et al. 2002; Siebesma et al. 2003), the modeled cloud elements do have roughly the same cloud-base height at particular hour, whereas in the present study there is significant variability of cloud bases at each hour (as shown in Fig. 9). These differences influence the mass flux profiles from sampling the clouds at different heights. The Bowen ratio for the case study considered over ARM SGP for previous LES and SCM simulations of shallow cumuli (e.g., Brown et al. 2002; Lenderink et al. 2004) was very low (;0.2) compared to an average value (0.72) reported from this study. Lower Bowen ratios influence the subcloud layer through moistening, which in turn influences