Water vapor mixing ratio data from the Raman lidar and Atmospheric Emitted Radiance Interferometer (AERI) are analyzed toward inferring the presence of organized, three-dimensional, turbulent features within convective boundary layers (CBLs). The Raman lidar and AERI instruments provide unique, high-quality data sets, from 200-m to ∼3-km altitude at ≤1-min temporal resolution profiles of water vapor. CBLs under the influence of solar heating for two fair-weather days over the Atmospheric Radiation Measurement (ARM) Central Facility near Lamont, Oklahoma, during the International H20 Project 2002 (IHOP_2002), and 1 day during the Cirrus Regional Study of Tropical Anvils and Cirrus Layers (CRYSTAL)-Florida Area Cirrus Experiment (FACE) 2002 experiment, are evaluated. This research provides a means of inferring the presence of coherent, convectively driven circulation (momentum) signatures exclusively from high-temporal resolution water vapor profiles. Spectral and statistical analysis of Raman lidar water vapor-time series reveal the passage of coherent turbulent structures, including horizontal convective rolls (HCRs), within the CBL over the ARM site at ∼8 and 12 min intervals for 2 days during IHOP_2002. AERI data collected during CRYSTAL-FACE also indicate similar structures, with meteorological conditions on this day also dictating the presence of cloud streets over the AERI (every 12 and 28 min). Radiosonde, satellite, and ceilometer analyses provide information about the favored stability and CBL cloud patterns, as well as a check on our spectral analysis. Evidence within the perturbation water vapor time series from these two instruments is well correlated with HCRs and other cumulus patterns seen in 1 km resolution GOES visible satellite imagery.
 A significant number of micrometeorological measurements of water vapor (q), temperature (T) and wind are collected each year worldwide. Many of these data have been shown to be of high quality and are available at high-temporal (at frequencies of ≤10 min) and high-spatial (tens of meters) resolutions over deep layers of the troposphere (from the surface to 15 km). The recent International H2O Project 2002 (IHOP_2002) field campaign highlighted the need for gathering q information: One of this campaign's main goals was to obtain an improved understanding of q distribution in the atmosphere, especially within the atmospheric boundary layer (ABL) [Weckwerth et al., 2004], the layer above ground level (AGL) most influenced by Earth's surface [Stull, 1988]. Knowledge of the vertical and temporal distribution of q, as well as heat and momentum, has important implications for understanding many physical and dynamical processes in the atmosphere (e.g., convective storms, cloud cover, surface winds).
 The ground-based Raman lidar instrument [Goldsmith et al., 1998] is located at the Southern Great Plains (SGP) Central Facility of the Atmospheric Radiation Measurement (ARM) program site near Lamont, Oklahoma, and was the first operational Raman lidar. Ten-minute retrievals of q from 200 m to ∼3 km are available over the diurnal cycle from this instrument, but the data can be postprocessed to yield ≤1-min profiles for specific studies [Turner and Goldsmith, 1999; Turner et al., 2000]. In addition to lidar, the ground-based AERI instrument [Smith et al., 1999; Feltz and Mecikalski, 2002; Feltz et al., 2003] was recently configured in an experimental mode to collect data at 20- to 40-s time resolution (rather than the normal 8-min resolution; [Turner et al., 2006]). Data at this temporal resolution from AERI was first demonstrated during the Cirrus Regional Study of Tropical Anvils and Cirrus Layers (CRYSTAL)-Florida Area Cirrus Experiment (FACE) 2002.
 The impetus for this study is to demonstrate the ability of high-temporal resolution micrometeorological measurements of q as obtained from the “active” Raman lidar and “passive” AERI for identifying the passage of coherent turbulent structures over these instruments. Both instruments provide highly accurate, well-validated measurements of q in the CBL when compared against colocated radiosonde soundings [Turner and Goldsmith, 1999; Feltz et al., 2003]. Specifically, we will assess the potential of these data for inferring CBL turbulence patterns (e.g., HCRs, cellular convection), assuming that subtle but robust q fluctuations are indeed measured by these instruments across microscales to mesoscales (200 m to 20 km [see Orlanski, 1975]). This study relies only on q information collected by these stationary, ground-based, vertically pointing instruments during daytime conditions over land to infer such turbulence. No direct wind data are used. Given the documented 5–15% variation in q observed across upward and downward circulation branches within the CBL (Weckwerth ), and the documented accuracy of Raman lidars and AERI for estimating q, it seems likely that these instruments would detect some signal of CBL convection at these fixed locations. Although vertical profiling instruments like the Raman lidar and AERI were designed to provide robust measurements of gross tropospheric moisture and T fields (e.g., precipitable water, CBL depth, cloud altitudes), this study shows that they provide high-quality microscale measurements as well.
 For this study, spectral and simple statistical analysis are performed on the profiler data sets to identify the presence and scales of turbulent phenomena present on the case study days. Stability and shear analysis, using radiosonde soundings collected near the instrument sites, coupled with an analysis of 1 km geostationary satellite data, will be used to obtain a form of “truth” about the atmosphere and CBL turbulence being observed by these instruments.
 The paper proceeds as follows. Section 2 provides the background for this work; section 3 presents an overview of the instruments. Section 4 describes the data and analysis methods used, and section 5 presents the case analyses with respect to boundary layer theory and previous research. Finally, section 6 discusses the analysis uncertainties, and section 7 concludes the paper.
 Observations of HCRs indicate their aspect ratio to be on the order of 2–4 but often ∼3 times wider than deep [Brown, 1970; Caughey and Palmer, 1979], with mean CBL depth determining HCR depth. The spacing between the lines in cloud street cumulus patterns (as observed above adjoining HCR updrafts) has been found to range from 2 to 8 km. Individual cloud lines have been observed to have a horizontal extent of up to 400–500 km [Kuettner, 1971]. Roll orientation is typically 10–20° to the left of the geostrophic wind (measured at the CBL top), or along the mean CBL wind vector. LeMone  provides a detailed overview of HCR vortices from both a dynamic and thermodynamic perspective.
 Several studies have found that both subjective and objective observing systems based on conventional instrumentation (i.e., radiosondes) will often unintentionally misrepresent CBL characteristics [Weckwerth, 2000]. The 1–2+ g kg−1 (5–15%) variations in absolute q mixing ratio, found on micro-α to meso-γ scales, can have a marked influence on the convective initiation (CI) process, as well as on convective available potential energy (CAPE) and convective inhibition (CIN) estimates. In fact, in situ aircraft measurements, used to modify radiosonde temperature and q profiles, were the only means of explaining the observed CI along HCRs during several days of the Convection and Precipitation/Electrification (CaPE) experiment in Florida in 1991 [Weckwerth et al., 1996; Weckwerth, 2000]. The interaction of HCRs and other convective circulations with ABL forcing (fronts, outflow boundaries, etc.) has been strongly linked to CI [Tuttle et al., 1992; Wilson et al., 1992; Laird et al., 1995; Kingsmill, 1995]. Micro-α to meso-γ scale variations in equivalent potential temperature (θe), as well as subtle mass and moisture convergences, are critical agents for dictating whether a small cumulus reaches cumulonimbus scale [Rabin et al., 1990; Weckwerth and Wakimoto, 1992; Ziegler and Rasmussen, 1998].
 Many previous studies have relied on aircraft and tower measurements to gather information on CBL moisture distributions and turbulence [e.g., Young, 1988a, 1988b, 1988c; LeMone, 1973, 1976; Mahrt, 1979], in which wind information was important to the analysis technique. Some of the earliest ABL studies using lidar observations include those that employed an aerosol backscatter lidar instrument [Kunkel et al., 1977; Boers et al., 1984]. Senff et al.  used ground-based differential absorption lidar (DIAL) measurements of q to estimate q flux [with the aid of virtual temperature data from a radar radio acoustic sounding system (RASS)]. In the process, some relationships to CBL turbulence were developed. Past lidar studies have focused on the determination of atmospheric structure, as inferred from measurements of reflectivity (i.e., backscatter) as a function of aerosol size and content [see Atlas et al., 1986; Melfi et al., 1985]. Aircraft and ground-based water vapor lidar observations have revealed variations in ABL depth and turbulence over oceanic [Melfi et al., 1985; Atlas et al., 1986] and land [Senff et al., 1994; Kiemle et al., 1995] surfaces. The work of Mayor et al.  describes the use of the volume imaging lidar (VIL) in the study of convective internal boundary layers over Lake Michigan during winter. Weckwerth et al. [1997a] uses Doppler lidar to reveal “organized streak structures” in the surface layer.
Hagelberg et al.  provides unique views of microscale closed and open Rayleigh-Bernard-like cellular convection in the surface layer of a marine ABL during the Central Equatorial pacific Experiment in 1996 (CEPEX [see also Cooper et al., 1996]). Wind measurements were also not available in the Hagelberg et al.  analysis, similar to this study, and a wavelet transform method was subsequently used to quantify the dominant convective scales; coherent microscale turbulent structures are identified initially through 2-m resolution maps of q perturbations.
 The work of Feltz and Mecikalski  is one of the first examples on the use of the AERI for studying CBL stability as a function of CAPE, CIN and time changes in CBL θe over time; AERI was able to monitor CBL destabilization every 10 min as θe increased, CIN decreased and CI occurred near the instrument site. It was shown that atmospheric destabilization (the weakening of a capping inversion) can be successfully monitored using 10-min data from AERI, as was done for this examination of the 3 May 1999 Oklahoma tornado outbreak.
3. Instrument Descriptions
 Raman lidar data from two days during IHOP_2002 for which 5–10 min GOES 11 data were available, and AERI data for one day during the CRYSTAL-FACE field campaign, were evaluated. The GOES 11 satellite was placed over 101°W for the IHOP_2002 experiment, and configured to collect data at 5- to 10-min frequencies. The availability of the high time resolution GOES data during IHOP_2002 motivated the decision to utilize ARM Raman lidar data collected during this experiment.
 Since 1992, the ARM Program has operated a suite of instruments at the SGP Central Facility site in Lamont, Oklahoma (latitude 36.61°N, longitude 97.49°W) designed to measure vertical profiles of atmospheric q, T, aerosols and wind as they pertain to improving our understanding of clouds, the radiative properties of clouds and aerosols, and the distribution of tropospheric q on small spatial and temporal scales [Ackerman and Stokes, 2003]. ARM has placed significant priority on accurate q observations due to the importance of q in radiative transfer calculations. A series of water vapor intensive observation periods (WVIOPs) were conducted at the SGP site to characterize the various q measurement techniques [Revercomb et al., 2003]. These WVIOPs, as well as long-term intercomparisons, have shown that the AERI and Raman lidar measure q with an accuracy of 1–2 g kg−1 [Turner and Goldsmith, 1999; Turner et al., 2000], and 5–10% absolute [Feltz et al., 2003], respectively. Turner and Goldsmith  and Revercomb et al.  outline the extensive validation activities that have been performed for ARM q measurements.
 The q profiles observed by the ARM Raman lidar are calibrated to agree in integrated precipitable water vapor with a colocated two-channel microwave radiometer [Turner and Goldsmith, 1999], resulting in the fractional q mixing ratio accuracies of ±10% during the day and ±5% at night [Turner et al., 2000]. Comparisons of colocated microwave radiometer and radiosonde data at the SGP ARM site with “AERI+Rapid Update Cycle” (RUC) model retrievals yield the root mean square differences of 1°K for T and 5–10% absolute for q within the ABL, as reported by Feltz et al.  and Turner et al. .
3.1. Raman Lidar
Turner and Whiteman  provide a historical overview of atmospheric Raman lidar observations. The Raman lidar is an active, vertically pointing remote sensing instrument that estimates atmospheric q through the measured ratio of the Raman shifted signals of H2O and N2 molecules. Turner and Goldsmith  describe how this ratio is converted into a q mixing ratio profile. The Raman lidar instrument at the SGP ARM Central Facility site can measure q to over a 10 km altitude during nighttime clear-sky conditions, but is restricted to 3 km during daytime conditions due to solar contamination [Goldsmith et al., 1998]. The nominal temporal resolution of the ARM Raman lidar data is 10-min with a vertical resolution of 78 m [Turner et al., 2000]. However, as noted, the data can be postprocessed to yield ≤1-min samples for use in research requiring high-temporal resolution atmospheric profiles.
3.2. Atmospheric Emitted Radiance Interferometer
 The AERI, as developed over the past 15 years at the University of Wisconsin, is designed to retrieve the q and T structure of the ABL from the ground-based, high-spectral resolution infrared measurements it collects. The AERI is a well-calibrated instrument that passively measures high-resolution downwelling emitted radiances from the atmosphere [Knuteson et al., 2004]. Because of the instrument's high-spectral resolution [less than one wave number (cm−1) between 3 and 18 μm], AERI is sensitive to vertical changes in T and q within the ABL below ∼3 km AGL. Feltz et al. [1998, 2003] and Smith et al. [1990, 1999] provide details on the AERI instrument and the physical retrieval algorithm. The vertical resolution of profiles retrieved from the AERI ranges from 50 m from the surface up to 1 km, 200 m to 2 km, to 250 m at 3 km. Above 3 km altitude, broadening of spectral weighting functions results in decreased accuracy [Feltz et al., 2003].
 During the period of study, the mobile AERI (housed at the University of Wisconsin–Madison) was deployed to a location south of Naples, Florida (latitude 25.84°N, longitude 81.39°W). AERI typically measure both T and q at 10-min resolution in real time. During CRYSTAL-FACE, the mobile AERI system was run in an experimental mode, collecting data at 40-s resolution. Data collected in the “rapid sample” mode were noise filtered to significantly reduce the random error level [Turner et al., 2006].
 Precipitation and low clouds saturate the signal of a passive remote sensing instrument that senses within the infrared spectrum, thereby providing little useful thermodynamic profiling information. The AERI algorithm has been modified to retrieve profiles to cloud base given a cloud base altitude from a ceilometer, which allows for the monitoring of ABL average mixing ratio. A long-term analysis of the AERI data at the SGP site showed that the AERI was able to retrieve T and q profiles ∼62% of the time, with the majority of the unsuccessful retrievals due to precipitation, fog and low clouds [Feltz et al., 2003]. Use of AERI's 40-s sampling data during CRYSTAL-FACE and other similar experiments allows significant information about the ABL to be obtained when sky conditions are partially or mostly clear (i.e., sampling can occur through breaks in cloud cover).
3.3. AERI and Raman Lidar Data Usage
 The channel used for sensing the Raman shift associated with q is at 408 nanometers, which receives an appreciable amount of solar radiation during the daytime. To make the daytime measurements, the system must be able to discriminate the backscattered photons (initiated by the lidar laser beam) from the solar background. This is done by only receiving radiation in a very narrow bandwidth (associated with the known wavelength of the Raman q backscatter) along with measuring the radiation in a restricted field of view (FOV). Naturally, the wider the FOV, the more solar energy is admitted, resulting in noisier q profiles. However, with a more narrow FOV, one can profile higher in the troposphere, but at the expense of compromising the quality of lower altitude data. The solution used by the Raman lidar is to use two channels with differing FOVs: a wide FOV channel for lower tropospheric profiling and a narrow FOV channel for the middle to upper troposphere.
 Within the Raman lidar profiles, as provided in Figures 1a and 1b for 11 and 20 June 2002, are relatively “noise-free” q retrievals from 200 to 500 m and from 800 to 1200 m. The lowest 60 m are obtained from tower in situ observations mounted on a tower at the surface, 25 m, and 60 m. Between 60 and 200 m, the signal-to-noise ratio (S/N) gradually increases as the laser beam completely “fills” the telescope's FOV, with data above 200 m being completely from the Raman lidar. In Figures 1a and 1b, high-quality data are present from 200 to 500 m AGL but degrades quickly from 500 to 800 m as the signals from the two lidar channels (narrow and wide FOV) are combined. During conditions of high solar insolation, the solar background greatly decreases the S/N in the wide channel, and the S/N quickly approaches unity (near 1 km). At night, in contrast, the S/N in the wide FOV is acceptable at altitudes >1 km. Therefore the two channels need to be merged at lower altitudes (500–800 m) during the day than at night (1300–1500 m); data from the two channels are merged linearly over these ranges. The 312-m and 1014-m levels have therefore been chosen for analysis due to the high S/N from the wide and narrow FOV channels, respectively.
 Although no known problems are associated with the AERI instrument, AERI data from comparable heights (315 and 928 m AGL) were selected in order to maintain consistency in the analysis and results from the two instruments. These levels were also close enough to the ground level AERI to take advantage of the relatively sharp weighting functions within the first kilometer of atmosphere (as obtained by this instrument). Clear-sky conditions are the only requirement for the AERI to provide q retrievals. Passing clouds prevent the retrievals from AERI for brief (<3 min periods), but do not diminish our ability to perform a power spectrum analysis.
4. Analysis Methodology
 On the basis of the previous validation and intercomparison of the ARM Raman lidar and AERI, high-quality data are available for this analysis. Processing of q mixing ratio data from both instruments initially involves partitioning water vapor mixing ratio variations between those of the mean basic state, and those due to microscale and mesoscale turbulent motions, as described in previous studies [e.g., LeMone, 1976; Mahrt et al., 1994]. In this application, scales greater than the organized cumulus cloud scale, ∼2 km, are regarded as mesoscale, while scales ≤2 km are then defined as “turbulent,” or micrometeorological scales. Thus the method employed decomposes q measurements from the AERI and Raman lidar into local averages [q(z, t)], a time series (running) average 〈q(z, t)〉, and the deviations from the temporal average q*(z, t). Here, z refers to a level (in meters) in the AERI or Raman profile that a time (t) series is taken. This gives
From equation (1), with [q(z, t)] being the mean portion of the time series, the entire time series including the turbulent component q′(z, t), is given as
Optimal q time series analysis relies on identifying the level above ground where the signal of this organized turbulence is maximum.
4.1. Power Spectrum Analysis
 For the selected Raman lidar days, data from 1800–2200 UTC (1200–1600 LT) on 11 and 20 June (Figures 1a and 1b) were chosen as the primary analysis period due to the presence of cumulus cloud “streets” as seen in GOES 11 visible imagery (Figures 2a and 2b). AERI data from 1700–2100 UTC (1200–1600 LT) were chosen on 29 July during CRYSTAL-FACE, as similar cumulus cloud structures were apparent in GOES 8 visible imagery. A 60-min running (temporal) mean of q mixing ratio (q*) at both levels were calculated for the entire analysis period. Use of a 60-min running temporal mean allowed for the preservation of trends within the data, as q slowly changed over the daytime period. Outlying values (e.g., <2 or >30 g kg−1) were not included in the running mean calculation. q′ values were then calculated by subtracting the 1-min (40-s) Raman lidar (AERI) q mixing ratio value from the running mean. An example of the raw perturbation q (q′) signal and the associated running mean are provided in Figures 3 and 4 for the IHOP_2002 cases. Five-minute running averages of the q′ time series (〈q(z, t)〉) were then calculated at both levels for the three selected cases. This was done to smooth out high-frequency variability produced by both instrument noise and disorganized micro-γ and -β CBL turbulence (making the time series “stationary,” i.e., the true mean of the variable and its higher-order statistical moments are independent of the particular time in question). Using q* and 〈q(z, t)〉, the q′ time series are developed (by solving equation (2) for q′), and power spectrum analyses are performed.
 The autocovariance is the covariance of a variable with itself at some other time, measured by a time lag (or lead); high autocovariances at a given time lag indicate that a significant q perturbation periodicity exists at that lag. Autocovariance coefficients, using the notation in equation (2), are calculated using the formula
where N is the total number of observations, L is the total number of “lags”, q′i is water vapor mixing ratio perturbation (q′ = q − 〈q〉 − q*) at time i, is the q mixing ratio perturbation through the entire time series, and rk is the autocovariance at lag k. Autocovariances are then converted into spectral space to evaluate their statistical significance. Raw spectral amplitudes are calculated using the formula (adapted from Panofsky and Brier )
where Bk is the raw spectral amplitude at lag kj. Equation (4) does not give the best estimate of the smoothed power spectrum function. In effect, it computes weighted means of unsmoothed spectral estimates. Unfortunately, some of the weights can be negative; this results in greatly distorted spectral estimates whenever there are rapid fluctuations in the true spectrum. To overcome this drawback, the spectral amplitudes are smoothed by the Hanning method: Si = 0.25Bi−1 + 0.50Bi + 0.25Bi+1, where Si represents the smoothed spectral amplitude. The smoothed spectra for both June 2002 days are represented by the lines in Figures 5a and 5b and Figures 6a and 6b for levels 312 m and 1014 m, respectively.
 At this point in the analysis procedure, the statistical significance of the spectral amplitude peaks are not known. Any or all of the spectral peaks could result from a periodic random event. A chi-square test was performed in order to assess which spectral peaks are indeed statistically significant. The null hypothesis for the chi-square test is red noise (i.e., a random event with persistence) through a first-order autoregressive process. The spectral amplitude of red noise is defined as
where Tf is the spectral amplitude of the red noise curve at frequency f, is the averaged raw spectral amplitude over the entire time series, and r1 is the “lag 1” autocovariance. Figures 5 and 6 provide illustrations of the null hypothesis and 99% confidence curves for the IHOP_2002 cases. We assume that spectral peaks above the 99% confidence interval are of meteorological significance.
 The method described by equations (3)–(5) has the ability to identify aliased as well as primary signals within a time series (e.g., a 4-min primary signal, and 8-, 12-, 16-min, etc., aliased signals). This lack of discrimination leaves us to prove/disprove the significant peaks within the time series spectra as real or artifacts of aliasing, for which the satellite and ceilometer data will help corroborate.
4.2. Sounding Analysis
 Sounding analyses is also performed in part to evaluate whether the clouds seen in the GOES data (Figures 2a and 2b) are what they appear to be, either HCRs, cellular convection, or some combination. Using soundings taken at the ARM Central Facility and near the CRYSTAL-FACE AERI instrument, mean wind shear, low-level stability (e.g., bulk Richardson number, RiB), and boundary layer depth (zi) are estimated.
 In particular, zi is very important for identifying the depth of rolls given that their horizontal scale is proportional to 3 times their vertical scale [Caughey and Palmer, 1979]. The zi's were determined in sounding data by subjectively locating the top altitude of nearly constant potential temperature (θ), as well as by a change in wind direction in conjunction with the θ discontinuity. The equation for RiB used here [applied from the top of the surface layer (zsfc) to zt; zi = zt − zsfc] is given as
In general, roll convection is favored when RiB is low, with more cellular convection favored when it is high [see LeMone, 1973]; the critical value of RiB is about 0.25 for “thin” layers on the order of the CBL depth [Stull, 1988].
 Last, the advective wind velocity, V, for assessing the motion of convection (e.g., the frequency of HCR passage) is the wind at height zi (estimated from the soundings of Figures 7b and 8b). It is anticipated that roll orientation will fall at about 10–20° to the “left” of the wind at the base of the inversion atop the CBL [Plank, 1966; LeMone, 1973], or to the left of V. The advective (forward and lateral) motion of the rolls is roughly that of mean V in the CBL. For the IHOP_2002 days, ARM radiosonde soundings from the Central Facility, launched within 100 m of the Raman lidar, are used. For the CRYSTAL-FACE day, soundings taken at the AERI instrument site, south of Naples, Florida, are used. For all three days, the sounding closest to the chosen analysis period is assumed to represent the conditions of the mature CBL at these locations (although comparisons to other soundings near the analysis period were made).
 The weather on all three days studied was characterized by fair, nonprecipitating conditions, in the absence of high- or middle-level cloud cover, allowing the observation of CBLs under the influence of strong solar heating. CBL moisture and sky conditions were similar on all days, despite the great difference in geography.
 Over Oklahoma, atmospheric moisture levels on the two IHOP_2002 days were quite high, with mixing ratios ranging from 15–17.5 g kg−1 at 312 m (Figures 1a and 1b). This lead to the widespread development of “fair weather” cumulus clouds, yet none of these clouds developed into deep convective storms in the vicinity of the ARM site (squares on Figure 2). No distinct surface-based convergent boundary existed across Oklahoma on these days, giving the appearance of a rather homogeneous ABL across a large section of the central Great Plains during the analysis time period (1800–2200 UTC) when the cloud patterns were pronounced in GOES 11 visible satellite imagery (Figures 2a and 2b).
 The aforementioned atmospheric parameters useful for assessing the CBL convective regime are presented in Table 1. The 2030 UTC sounding was chosen for the 11 June case (Figures 7a and 7b), being that it occurred near the middle of our analysis period. For 20 June, the 2330 UTC sounding (Figures 8a and 8b) was used given that the 2030 UTC sounding was unavailable. (Weather conditions over the ARM site changed little from 2030 to 2330 UTC, and thus we feel estimating the above quantities for the CBL on 20 June from this later sounding provides useful information.)
Table 1. Atmospheric Parameters Derived From Sounding Analysis for the Three Selected Daysa
zi (AGL), m
See text for parameter definitions.
11 June 2002
200° at 12.0 m s−1
20 June 2002
170° at 11.5 m s−1
29 July 2002
120° at 7.0 m s−1
 As presented in Figures 5a and 6a, the power spectra show that spectral amplitudes with near 99% confidence are found at ∼8.5 and 10 min at 312 m, and 6, 8, and 12 min at 1014 m on 11 June. For 20 June (Figures 5b and 6b), spectral peaks (>99% confidence) are found at 6, 8, and 17 min at 312 m and 5.5, 7, and 10.5 min at 1014 m. The satellite analysis and discussion to follow will address the possible causes of the various peaks. The fact that the spectral peaks at the different levels occur at ∼6–8 min suggests that the Raman lidar is measuring q perturbations accompanying coherent convective structures.
 The GOES 11 imagery for 11 and 20 June (Figures 2a and 2b) illustrates the presence of HCRs (cloud streets) over much of the Southern Plains. These features appeared first in the imagery during the midmorning to late morning hours (circa 1600–1700 UTC) during both cases. On 11 June, a closer view of the cloud patterns, centered over Lamont, indicates cloud-street features spaced 4–5 km apart (see Table 2). Line spacing on 20 June were less, ∼3–4 km, suggesting shallower circulations than those of 11 June, consistent with the lower zi. Cloud (HCR updraft) widths on 11 June were generally from 1–2 km; cumulus line widths on 20 June were ∼1 km. We assume that the width of one HCR (e.g., either the clockwise or counterclockwise helical circulation) is roughly one half the clear-sky distance between cloud street lines to the middle of the cloud line itself (a “wavelength” for of an HCR is defined as the distance between updrafts); the updrafts of adjoining HCRs produce the positive q′ values. For comparison, the predicted horizontal scale of the rolls based on CBL theory is approximately 4.5 km [3.0 × zi (1488 m)] on 11 June and 3.7 km (3.0 × 1232 m) on 20 June [LeMone, 1973].
Table 2. Characteristics of Rolls Visually Determined From GOES Observations on the Three Days Analyzed
11 June 2002
20 June 2002
29 July 2002
 GOES 11 image frequency varied between 5 and 10 min over the analysis periods, and although this was a much higher frequency than normally provided by GOES satellites (15 min), it was still insufficient to perfectly analyze cloud feature passages at the ARM site. However, visual analysis and time interpolation of convective cell passages were sufficient to allow for a reasonable quantitative comparison of satellite clouds to the Raman lidar q′ analysis.
 Visual inspection of animated imagery reveals that cumuli propagated over the Central Facility with periodicities that congregated between ∼9 and 13–14 min on 11 June. The magnitude of these periodicities are greater than any other periodicity (either lower than 9 or higher than 15 min), therefore matching reasonably well those found via Raman lidar data analysis.
 Satellite estimates of cloud/HCR-updraft passages at Lamont on 11 June were computed given orientations of ∼185°, motion along 200° (Table 2), an advective velocity (along 200°) of V ≃ 12 m s−1 (estimated from satellite cloud motion analysis and from the sounding), and cloud spacing of ∼4.5 km. Assuming a range of spacing, which determines the variation in q above the Raman lidar, from <1 km (i.e., from roll updraft to roll updraft) to 2 km (i.e., 2.5 km clear-sky gaps between cloudy updrafts), we can estimate cumulus cloud passages over Lamont based on simple geometry to be from 24 min for a small updraft spacing (<1 km) roll, to ∼13 min for the 2 km distance. Comparing the significant peaks seen in Figures 5a and 6a, this 13-min frequency agrees with the 10- to 12-min peaks, but not with the 6- to 8.5-min peaks. It also suggests supporting evidence that the small ∼25-min peak may indeed be HCR updrafts as seen in GOES. Errors in these satellite estimates are influenced by the maximum resolution of GOES.
 On 20 June, cumulus cloud feature passages at Lamont were somewhat more difficult to visually assess from satellite analysis, due to the fact that they were not as well organized as those of 11 June. Repeating the above analysis, provided a degree of HCR integrity, the observed orientation is ∼150°, the motion is V ≃ 11.5 m s−1 (again confirmed by satellite cloud motions) along 170° (Table 2), and the cumulus line spacing is ∼3.5 km. The range in periodicities is again based on a variation in updraft-to-updraft spacing of <1 km to a realistic 1 km (with 2.5 km gaps). This results in a predicted cloud passage frequency of 10–15 min. The 10.5-min frequency corresponds well with the spectral peak of ∼10.5 min at 1014 m. The higher-frequency time variations in q′ (5.5–8 min) are however not well described by the satellite analysis on 20 June.
 For the sub-10-min spectral peaks in q′, we turn to Vaisala ceilometer cloud base information for both June 2002 days. This ceilometer is located at the Central Facility within several hundred meters of the Raman lidar. The Vaisala ceilometer is a self-contained, ground-based, active, remote sensing device designed to measure cloud base height and potential backscatter signals from aerosols. Model CT25K has a maximum vertical range of 7800 m, with 15-m horizontal resolution. The ceilometer transmits near-infrared pulses of light, and the receiver telescope detects the light scattered back by clouds and precipitation. This instrument is described by Lonnqvist , as well as on the ARM Web site (http://www.arm.gov).
Figures 9a and 9b present the ceilometer estimates of cloud passages at the Central Facility from 1700 UTC on 11 June to 0000 UTC on 12 June (Figure 9a), and from 1700 UTC on 20 June to 0000 UTC on 21 June (Figure 9b). Counts of cloud passages on 11 June between approximately 2042 and 2206 UTC suggest ∼7–8 over this 84-min period, or about one every 10.5 to 12 min. On 20 June, between 1842 and 1954 UTC, approximately 9–10 clouds pass over the Central Facility, or about one every 7.2 to 8 min. Cloud passage frequencies on the order of one every 7 min are again seen on 20 June between 2012 and 2048 UTC. It should be noted that even a small error in these cloud counts from the ceilometer will influence these results.
 From the ceilometer measurements, there is strong evidence that cloud passages (e.g., the updraft branch of adjoining HCRs, or isolated cells) do support the high-frequency spectral peaks seen in q′ at 312 and 1014 m AGL from the Raman lidar, especially on 20 June. Our speculation then is that the ceilometer “sees” clouds that are not well observed by GOES 11, which implies that many clouds that passed over the Central Facility are sub-1 km in horizontal scale (i.e., below the optimal viewing capabilities of GOES 11 Imager's visible sensor).
 More discussion will be provided below on the apparent disagreement between the satellite and q′ time series analysis to ascertain what may be causing the <10-min signals in q′ (i.e., what atmospheric phenomena may be producing the small-scale cumulus patterns). Clearly however, none of the longest (>25 min on 11 June, and >15 min on 20 June) peaks can be explained by direct satellite analysis, suggesting that these are the result of aliasing of higher-frequency features.
 Portions of 29 July 2002 also met the sky condition and data availability criteria described above. A mesoscale cyclonic circulation passed to the west of the CRYSTAL-FACE AERI site during the morning and early afternoon hours. Cirrus clouds produced by convection associated with this circulation advected over the experiment site, thereby limiting ABL heating and cloud development until early afternoon. After the cirrus dissipated around 1730 UTC, a cumulus cloud field began to form. As deep convection developed to the north of the AERI site, an outflow boundary from the convection began to interact with the ABL features and additional storms initiated near the site at 2200 UTC, which persisted for the remainder of the afternoon and early evening. Therefore the time period chosen for the AERI data analysis was 1700–2100 UTC (Figure 10), when the organized cumulus patterns became most pronounced in satellite imagery. Figure 11 presents the temperature, moisture, and wind profiles for this CRYSTAL-FACE case day.
 As noted, AERI-retrieved profiles of q are not available when clouds below ∼2 km altitude AGL are present. Thus the data used for this study contain 10 missing segments of 1–10 min in duration; q mixing ratios are interpolated during these missing segments so that the time series remains continuous. Because the time series remains continuous (and nonvarying), these missing segments will not be reflected in the final time series analysis.
 As presented in Figures 12a and 12b, the power spectrum of the q′ time series shows that spectral amplitudes at 315 m with >99% confidence are found over a broad time interval, from ∼16 to 28 min. At 928 m, just above the CBL top, two distinct peaks were found at 12 and ∼26 min. These longer time frequencies are believed to be the result of lighter winds in the CBL.
 The GOES 8 image at 1925 UTC 29 July is shown in Figure 13. Close visual inspection shows east-southeast to west-northwest (along a 110–115° axis) oriented cloud streets (clusters of cumulus) extending from far southeast Florida to the AERI location (square). Examination of Figure 13 shows cloud line spacing to be ∼4–5 km, combined with many smaller-scale cumuli. Cloud widths are generally from 1 to 2 km.
 The predicted horizontal scale of the individual rolls is ∼2.1 km (3.0 × 700 m), consistent with that observed. Their orientation obtained from satellite loops were from 110 to 115°, which is ∼10–15° to the left of V = 120° on this day. Visual estimates of cloud line/feature passages at the AERI site were determined using 15-min satellite imagery. Periodicities were again calculated based on simple geometry, with orientations along ∼110–115°, motions from 120°, and advective velocities (along 120°) of V ≃ 7 m s−1. Assuming again a range of line widths from <1–2 km (with 2.5 km clear-sky gaps), estimated cumulus cloud passage frequencies over the AERI site are from ∼42 min for a <1-km-wide cloud line, to ∼23 min for a 2-km-wide cloud line. These frequencies support the observed longer time lags seen in Figures 12a and 12b spectra of 26–28 min at the 315- and 928-m levels. The shorter time peaks (i.e., at 12 and 16 min), although not generally supported by this satellite analysis, are seen in ceilometer observations suggesting cumulus cell passages.
 Vaisala ceilometer data (Figure 14) are shown for the period 1700–2100 UTC 29 July. Although most cloud bases were above the 928-m level, we use these data to count the number of clouds (or moist updraft plumes) that passed over AERI. From 1900 UTC onward, with the exception of the 1900–1930 UTC time, (cumulus) clouds were observed to pass over AERI every 8 to 17 min, with this frequency increasing toward approximately 2045 UTC. Between 1900 and 2045 UTC, the mean cloud passage frequency was 15.3 min, which increased to 12.8 min between 1930 and 2045 UTC. This 12.8-min time supports the significant spectral peak seen in Figure 12 at ∼13 min.
6. Analysis Uncertainties
 One issue requiring further explanation is q variability at sub-10-min time frequencies on the IHOP_2002 days, and the 12-min peak on the CRYSTAL day. In the absence of “rapid scan” satellite imagery (1 min), we return to the ceilometer observations. These data on 20 June show strong evidence supporting the 7- to 8-min spectral peaks in the q′ time series. Ceilometer data on 11 June supports cloud passages every 10–12 min.
 During IHOP_2002, cloud lines are ∼1–2 km in width and up to 50 km in continuous length, and within the (animated) satellite imagery, subtle evidence of convectively induced or generated waves, presumably traveling atop the CBL, are seen, yet not necessarily directly over the Raman lidar instruments. Convection waves are not as well understood as HCRs. Kuettner et al.  and Hauf  have analyzed these waves, which could account for the “pearls on a string” cloud patterns when they appear together with rolls (i.e., perpendicular cloud lines with enhanced cumuli at the wave-roll intersections). Unlike HCRs, waves can be expected to pass over a stationary ground point with much higher frequency because their orientation is perpendicular to the mean flow as opposed to quasi-parallel [see LeMone, 1973, Figure 1].
 On 20 June 2002, waves are observed in satellite (loops) ∼150–200 km east and southeast of the Central Facility, oriented perpendicular to V with average wavelengths (spacing between adjacent wave clouds) of 5–7.5 km. These waves propagated along a 170–350° trajectory, and were noted to pass over ground points east and southeast of Lamont at 7- to 9-min frequencies. Curiously, this closely matches the 7- to 10.5-min peaks seen in Figures 5b (312 m) and 6b (1014 m). Questions remain, however, as it is unclear whether these waves would have a reflection in q′ data at the 312-m level; Kuettner et al.  suggests that these features may not be accompanied by full ABL overturning in the manner that HCRs produce.
 Speculation of the presence of smaller-scale convective features influencing CBL q can be extended to 11 June and 29 July, namely, the 6- to 8.5-min periodicity in q′ on 11 June and the 12-min peak on 29 July. In addition, the updrafts (of adjoining HCRs) are seen to vary in width from 2 to 8 km on any given day. Thus q′ changes due to cumulus cloud passage are likely to vary about the mean passage frequency by a factor up to 20% (e.g., by 2–3 min for a lag of 15 min). This process is likely leading to the broadened peaks seen in some of the autocorrelation results (e.g., Figure 12a for AERI at 315 m).
 There is high confidence that instrument-related effects on the retrieval of q are not resulting in spurious q′ variations. Proof of this comes as a result of Raman lidar data analysis on 9 June 2002 (not shown), a day when the air conditioner within the Raman lidar instrument enclosure was cycling too rapidly for optimal lidar retrievals. Such high cycling was shown to produce a spurious, very repetitive, consistent 15-min signal in q′ as a result of the Raman lidar instrument itself undergoing internal temperature fluctuations. (The air conditioning problem was fixed by 10 June to provide very limited temperature variation within the lidar enclosure and thus nearly undetectable effects on (i.e., optimal conditions for) the retrieval of q; Raman lidar instrument temperature variations could not be correlated with q′.)
 For AERI, environmental conditions surrounding the instrument had no adverse affect on the retrieved profiles. However, the AERI retrieval algorithm requires periodic “first guess” updates [Feltz et al., 2003], with the potential for spurious periodicities in q to enter the analysis within the upper ABL (2–3 km). During CRYSTAL-FACE these first guess updates are time interpolated and thus are not felt to have resulted in false spectra, especially below 2 km AGL.
 Finally, the time series analysis method above is subject to the identification of aliased signals in the q′ series. Efforts to compare the IHOP_2002 time series analysis against 5- to 10-min satellite data suggest that the longest, significant lags (e.g., 17 min, Figure 5b; ≫25 min, Figure 6b) are likely spurious, aliasing of shorter lag signals. These longer-frequency signals are not correlated with any cloud pattern in satellite data (i.e., are nonphysical q′ periodicities). The ∼56–57 min peak (with <99% confidence) seen in Figure 12a is an example of an aliased feature; recall, the 26- to 28-min peaks during CRYSTAL-FACE were related to cloud passages seen in the GOES 8 data.
 This study sought to assess the ability of 1-min Raman lidar and 40-s AERI profiles of q mixing ratio to identify micro-α to meso-γ convective circulation structures within the CBL due to the 5–15% q variability across these circulations. It is understood that neither the AERI or Raman lidar are perhaps ideal for measuring high-frequency ABL q variation, compared to various measurement inferred from radar (e.g., W and S band), but the high-accuracy of the retrieved q from these instruments allows for such analysis.
 Data from three afternoons (11 and 20 June 2002 during IHOP_2002, and 29 July 2002 during CRYSTAL-FACE), in which cumulus and HCRs were observed in GOES imagery, were analyzed. Comparisons to Vaisala ceilometer, in light of stability analysis, strongly suggests that indeed the Raman lidar and AERI systems are measuring these forms of organized CBL turbulence via q variations. Results of time series analysis show spectral amplitudes of ≥99% confidence for unique CBL features. Significant (≥99%) amplitudes in q′ were found in the 6–10 min time interval for the two IHOP_2002 days, and the 12 and ∼26–28 min time intervals for the CRYSTAL-FACE case.
 Vaisala ceilometer data provides evidence that cumuli with scales less than the 1 km GOES resolution passed over and were measured by the Raman lidar, and support for q perturbation frequencies ≤10 min on both June days. It is speculated that these higher-frequency q perturbations are likely caused by thermals/cells, or perhaps waves traveling atop the CBL, especially on 20 June. Satellite analysis on 29 July near the AERI revealed the ∼12–18 and ∼25 min peaks to be highly correlated with cumulus and HCRs.
 The larger scope of this research is multifaceted. In light of this study's results, q and T profiling systems, placed within pre-CI environments, may be configured to address several less well understood aspects of CBL growth and issues of short-term weather prediction. In particular, clues to the spacing of CI along boundaries such as dry lines and convergent zones (e.g., cold fronts) should exist in a preconvective environment with HCRs intersecting such boundaries. Since advanced sounding systems such as AERI and Raman lidar measure changes in dry and moist static energies with high time resolution in growing CBLs, they should support applications that require knowledge of CBL deepening and surface heating (i.e., sensible heating) rates. Applications that require knowledge of CBL turbulence for aviation-related activities stand to benefit from this research as it describes how Raman lidar and AERI data may be interpreted for these purposes.
 The authors wish to thank several IHOP_2002 science team members for valuable contributions to this work, especially Tammy Weckwerth (National Center for Atmospheric Research). The authors would also like to thank Jim Kossin (UW CIMSS) for his assistance in the time series analysis. This research was partially funded by NASA grant NAG5-12536, Department of Defense grant N00014-01-1-0850, Department of Energy Atmospheric Radiation Measurements Program grant DE-FG-02-92ER61365, and National Science Foundation grant ATM-0136158. The Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy under contract DE-AC06-76RL01830. This paper's quality was significantly improved with the help of three anonymous reviewers.