a. The site

Located on the west coast of Ireland, the Atmospheric Research Station at Mace Head, Carna, County Galway, is unique in Europe (O’Connor et al. 2008). Its position offers westerly exposure to the North Atlantic Ocean through the “clean sector” (Fig. 1, dashed lines edging the 180°–300°N sector) and the opportunity to study atmospheric composition under Northern Hemispheric background conditions as well as European continental emissions when the winds favor transport from that region. The site location, at 53.20°N, 9.54°W, is in the path of the midlatitude cyclones that frequently traverse the North Atlantic. The instruments are located 300 m from the shoreline on a gently sloping hill (4° incline).

b. Airmass and cloud classifications

The Mace Head meteorological records show that, on average, over 60% of the air masses arrive at the station via the clean sector. These air masses are ideal for carrying out background aerosol and trace gas measurements. Significant pollution events also occur at the site when European continental air masses, generally originating from an easterly direction, reach Mace Head. The Mace Head Atmospheric Research Station is uniquely positioned for resolving these different air masses and for comparative studies of their constituents and characteristics. For the presented study, 12 cases of single to multilayer clouds have been selected based on parallel information about (i) the origin of the air masses [obtained from 7-day backward trajectories calculated by the National Oceanic and Atmospheric Administration (NOAA) Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model]; (ii) the actual cloud pattern [as observed by the thermal infrared (IR) channel (10.3–11.3 μm) of the Advanced Very High Resolution Radiometer (AVHRR) satellite and manually from synoptic stations located at Shannon (52.71°N, 8.87°W) and Belmullet (54.22°N, 9.99°W)]; and (iii) the meteorological conditions, with no selected cases showing precipitation events longer than 30 min. Table 1 shows the origin of the air masses determined by the 7-day backward trajectories.

The airmass arrival directions are within the clean sector in most of the cases, except for three exceptions for southwest-stagnant air masses associated with a high pressure system centered over the United Kingdom. These three cases show a single developed layer of stratus (St) cloud persisting throughout several days. In contrast, the other nine case studies represent a more complex pattern, with multilayer structures of different cloud types. The attenuated backscatter profile time series measured by the two ceilometers and the aforementioned satellite and synoptic information are used for cloud-type classification (see Table 2).

The cases can be sorted in four groups: (i) a single-layer stratus cloud deck with relatively constant base height; (ii) a single-layer cloud type resulting from lowering of altostratus (As) into precipitating nimbostratus (Ns); (iii) a two-layer structure comprising a higher layer of altostratus clouds and a lower layer of stratus or nimbostratus; (iv) a three-layer structure comprising a higher layer of altostratus clouds, an intermediate layer of convective (open or closed) cumuliform cells, and a lower layer of boundary layer cumuli. In Fig. 2, four examples of different cloud patterns, as observed by the infrared channel of the AVHRR satellite, are shown. The relatively wide range of case study conditions was chosen to test the robustness of the retrieval algorithms under differing cloud fields. Figure 2a shows a case, 22 October 2008, of three cloud layers with a complex middle layer of convective open cells (north and south of Ireland at the moment of satellite pass). Such a situation, such as the case in Fig. 2c, is ideal for testing an algorithm’s efficiency, especially when two consecutive cloud-base detections are separated by several kilometers and associated with different cloud layers.

a. Vaisala CL31

The enhanced single lens technology applied in the CL31 ensures realistic data recording over the nominal range 0–7.5 km with full overlap at about 70 m (see Fig. 3). The good quality of the received signal is made possible by the strong and stable signal over the whole measurement range. The single lens technology is meant to provide reliability during precipitation, although the receiving system becomes very quickly saturated during rain events. The laser is an InGasAs diode emitting at the 910-nm wavelength with a manufacturing estimated accuracy of ±5 m (against hard target) equal to the highest vertical resolution Δz = 5 m (see Table 3 for technical data).

b. Jenoptik CHM15K

The CHM15k ceilometer measures atmospheric target backscatter profiles over the nominal range 0.03–15 km with full overlap height at about 100 m. The operating range is 15 km where it can reliably detect lower cloud layers as well as cirrus clouds, although the latter could be significantly hidden in the noisy component of the signal at these high ranges. The highest vertical resolution at which the instrument can work is 15 m with measured full vertical profiles of aerosol backscatter and detected cloud height, boundary layer height, and visibility values. The measuring principle is lidar based with a photon counting detection system and a solid-state Nd:YAG laser source emitting at the 1064-nm wavelength with undeclared manufacturing accuracy (see Table 3 for technical data).

a. The THT algorithm

In the signal processing, the attenuated atmospheric volume backscatter coefficient β^att is computed as

View Expanded

The THT scheme is based on the information about the mutual positions of the local maxima in the β^att vertical profile and its vertical gradient (GS, where the gradient applies to the natural logarithm of β^att). Each β^att profile has temporal and vertical resolutions of 15 s and 5 m for the Vaisala CL31 and of 30 s and 15 m for the Jenoptik CHM15K. In the postprocessing procedure, the Vaisala CL31 resolutions have been reduced to 30 s and 15 m to make it fully comparable with the CHM15K. The choice of the altitude range and temporal resolution is determined by the range and time as a trade-off between noise and feature discrimination.

For each case study with duration of measurements within the range of time (0000–2359 UTC), the index i in Eq. (3) indicates the single successive measurements in the selected period with time resolution 30 s. The single GS_i profiles provide single evaluations of the CBH and, combined in a sequence, give its temporal evolution,

View Expanded

The algorithm computes the mean GS and values from single GS_i and profiles averaged over 10 min. The choice to set the time of averaging equal to 10 min was taken to ensure relatively nonvariable cloud base at the start of each algorithm block. On its first step, the algorithm detects, starting from ground level, the heights of the largest GS and maxima, respectively. The mean value between the heights of the two maxima is the reference height h_ref. The related uncertainty is σ_ref = σ(2/GS) + σ(2/β^att), where the standard deviations σ_GS and answer for the variability of the single GS and β^att calculated heights during 10 min, on average, and are assumed to be statistically independent. The σ_ref accounts then for the overall temporal variability of the single detected heights and represents the h_ref related uncertainty. The assumption of statistical independence of σ_GS and is made possible by the fact that, despite the analytic relation between and its gradient GS_i, the heights of the two maxima are determined using different thresholds on the β^att and GS profiles [threshold on the signal-to-noise ratio (SNR) set to 1.5 for β^att and 2 for GS], making the selection of the β and GS maxima two independent processes. Once determined h_ref (i = 1), a range of altitudes h_j ∈ [0.85(h_ref − σ_ref) − 1.15(h_ref + σ_ref)] limits the vertical region in which to look for the principal maxima of GS₁ and . The entire vertical length of each profile is scanned for maxima, and these are iteratively rejected until they enter the interval [0.85(h_ref − σ_ref) − 1.15(h_ref + σ_ref)]. Every time the algorithm selects a GS maximum GS^max along the profile, this is checked in value against the GS maximum corresponding to the reference height . The ratio between the two values is used as a threshold (to be exceeded) to accept the current GS^max selected by the algorithm as a valid CBH, despite its position along the profile. This procedure ensures the correct determination of real cloud bases, even when they are located far (outside the current vertical interval) from the previous CBH determination. Once the GS₁ and maxima are determined, the mean value between the two heights is the first cloud-base height CBH₁. The CBH-related uncertainty is the standard deviation of all the rejected heights after each algorithm’s step. For 2 ≤ i ≤ 20, the algorithm determines, in the new range of altitudes [0.85(CBH_i−1 − σ_CBHi−1) − 1.15(CBH_i−1 + σ_CBHi−1)], a new value of CBH at each i step. For i = 21, the algorithm computes two new GS and profiles averaged over the time interval specified by i ∈ [11–30]. The algorithm performs the same process as for the previous 20 i steps starting by determining, for i = 21, a new h_ref value to be used for the calculation of CBH₂₁. The algorithm is built in blocks of 20 steps each until the end of the dataset is reached. An outline of the algorithm’s logic is shown in Fig. 4. In the left side of the figure, the β^att time–height cross section is shown with a highlighted time interval showing three cloud layers. In the right side of the figure, arrows indicate the rejected maxima lying outside the interval [0.85(h_ref − σ_ref) − 1.15(h_ref + σ_ref)] and the selected one within it. The algorithm can distinguish between different cloud layers, even when they are very close, simply by selecting different h_ref values on the GS and profiles. Once all the CBH determinations are obtained for each instrument, the values are compared for same temporal interval and vertical range (the vertical range goes from 0 to 7 km because of the Vaisala CL31’s limited range). From the detailed description of the algorithm rationale, it emerges that the tracking system of the single cloud-base detections along with the threshold value are two outstanding features allowing smooth CBH temporal evolution and realistic jumps between contiguous (in time) layers at different heights, respectively.

a. Methodology

The intercomparison was carried out through three steps: first, the VAIS and JEN manufacturer’s operational CBH outputs are compared and tested for the linearity between the two outputs (correlation coefficient for linear fit on scatterplot of the two outputs), the relative gain between the two outputs (slope of the scatterplot), and any bias or offset between the two (from the linear fit intercept on the y axis). The value of the slope also provides information about the SNR of the backscatter profiles. Second, the THT algorithm is applied to both instruments’ backscatter data and gives qualitative interpretation about the laser alignment of the two instruments (intercept of the linear fit) and the level of the SNR (slope of the linear fit). A low SNR means a higher number of spurious peaks in the β^att profiles, generating larger dispersion of the CBH determinations and affecting the value of the angular coefficient of the linear fit. Third, a graphic comparison between THT and JEN and THT and VAIS outputs allows extraction of additional information about under-/overestimations of the CBH determinations using the two algorithms.

b. VAIS versus JEN

The values of the correlation coefficient R², the slope a, and y-axis intercept b are shown in Fig. 5 for each case study. Cases in the figure are sorted columnwise by increasing cloud-base height from left to right. The average value of the slope was a = 0.91 and, within the range of the statistical variance (σ_a = ±0.15), it is comparable to the unity slope; nevertheless, the statistical variance is quite large, on the order of 16%. The average intercept was b = 192.57 m and illustrates a large offset between the two manufacturer’s proprietary algorithms. The statistical variance (σ_b = ±177.52 m) suggests that the VAIS retrievals provide a systematic higher cloud-base estimate than that from the JEN.

Figures 5c,e show the best and the worst cases (based on the R² values) for the VAIS versus JEN intercomparison. The worst case exhibits R² = 0.52, slope of 0.62, and intercept of 570 m, whereas the best case exhibits R² = 0.99, slope of 1.01, and intercept of 75 m. The cases presented in the other panels show, beyond the value of R², how cloud-base heights can be widely scattered over the whole range of altitudes between 0 and 7 km. Figure 5c clearly shows the existence of a higher CBH VAIS retrieval than that from JEN. The cases covering the period 23–25 December 2008 are examples of sustained and stable stratus clouds throughout an extended period. For these cases, the values of the intercepts vary very little, being approximately 80 m through the three cases. The case in Fig. 5n reports cloud-base data from a single layer of altostratus lowering and thickening to become nimbostrati during the late evening (Table 2, 18 September 2008 case). Data are widely scattered in both domains of altitudes, in the CHM15K’s and CL31’s, showing in particular the existence of three false layers in the CL31’s retrievals situated between the ground level and the actual cloud base (this artifact is discussed later).

All of the VAIS and JEN individual case study data are grouped in a single scatterplot and shown in Fig. 6, where R² = 0.788, slope = 0.925, and an intercept corresponding to an average instrument CBH offset of 160 m is seen. The outcome of the comparison can be summarized as follows: (i) the two algorithms provide in several cases very different estimates of the same cloud base; (ii) the distribution of the retrieved CBH values become more scattered as the number of cloud layers increases; and (iii) spurious cloud-base echoes exist especially in the CL31 β^att profiles. The results of the comparison suggested investigating further the relation between the β^att profiles from the two ceilometers and derivation of CBH using an independent algorithm.

c. Application of THT algorithm

CBHs resulting from the application of the THT algorithm to the β^att profiles are shown in Fig. 7 for the 12 cases, along with the associated linear regression data. As with the previous comparison, cases are sorted columnwise by increasing cloud-base height from left to right in Fig. 7. Data points have been linearly fitted to extract information about the slope a and the intercept b. The average slope value was a = 0.99; within the statistical variance (σ_a = ±0.02), it is effectively unity slope. The average value of the intercept was b = 13.56 m; within the statistical variance (σ_b = ±37 m), it is comparable to an effective zero intercept.

Figures 7n,i are shown the best and the worst cases, respectively, based on the values of the coefficient of determination R² ranging from 0.95 to 0.996. Figures 7h,c show two other examples of linear correlation with values of R² between the two extremes. The case in Fig. 7h is an example of the robustness of the algorithm that determines very close values of cloud base for the two instruments, even in conditions of complex cloud pattern (see Table 2, 22 October 2008 case). Error bars in the graphs show the uncertainty of each CBH estimate based on the standard deviation of the CBH values rejected during each algorithm step. All the THT_VAIS and THT_CHM15K outputs are grouped in a single scatterplot shown in Fig. 8. For the full dataset, R² has increased to 0.997, whereas the slope is 0.998 and the intercept is −3 m (effectively zero).

d. THT versus VAIS and THT versus JEN

The third step of the intercomparison is a visual comparison between THT and JEN outputs and, separately, between THT and VAIS outputs. Figure 9 shows the time–height β^att cross section with superimposed CBH values as determined by the THT and JEN algorithms (Fig. 9a) and THT and VAIS algorithms (Fig. 9b). The case on 24 December 2008 is an example of well-developed stratus layer establishing in a vast area of high pressure centered over the United Kingdom. These two examples clearly show the differences between the three algorithms. The THT detects the cloud base slightly above the JEN retrievals and definitely below the VAIS. The percentage difference between THT and JEN retrievals and between THT and VAIS retrievals is 6% of the average THT retrievals for both cases.

The previously mentioned false layer detection is evident in both panels of Fig. 10, where several spurious layers of cloud base are seen as “detected” by JEN and VAIS algorithms below the actual cloud layer of altostrati. The effect is likely because neither JEN nor VAIS algorithms seem to track the cloud layer, allowing the unrealistic jumps between contiguous CBH detections.

Figures 11a,b show again the scattered CBH values determined by the two built-in algorithms above and below the real cloud targets. On the contrary, the THT algorithm performs an efficient detection of the cloud base, even in such a complicated cloud pattern as in the 22 October 2008 case.

It is the authors’ belief that the determination of the cloud base should occur at the level where the β^att value starts to increase firmly (Eberhard 1986): namely, at that level where the β^att maximum rate of growth occurs (Pal et al. 1992; Campbell et al. 1998; Gaumet et al. 1995). Other algorithms have been created, and they purely rely on thresholds of the backscatter signal used to discriminate clear- from cloudy-sky profiles (Clothiaux et al. 2000), but they fail in detecting low scattering targets like high clouds. On the contrary, the THT algorithm is based on the analysis of changes in the slope of the backscattered powers with height and can then detect all types of clouds, including high clouds such as cirrus.