PathfinderTURB : an automatic boundary layer algorithm . Development , validation and application to study the impact on in-situ measurements at the Jungfraujoch

Continuous observations of the vertical structure of the planetary boundary layer are invaluable for the 9 validation of atmospheric transport models on the micro and meso scale. Lidar and ceilometer backscatter 10 observations offer a robust technique with growing spatial coverage, but the obtained backscatter profiles need to 11 be carefully translated into boundary layer parameters. Here we present the development of the PathfinderTURB 12 algorithm for the analysis of ceilometer backscatter data and the real-time detection of the vertical structure of 13 the planetary boundary layer. Two typical aerosol layer heights are retrieved by PathfinderTURB: the 14 Convective Boundary Layer (CBL) and the Continuous Aerosol Layer (CAL). PathfinderTURB combines the 15 strengths of gradientand variance-based methods and addresses the layer attribution problem by adopting a 16 geodesic approach. The algorithm has been applied to one year of data measured by two CHM15k ceilometers 17 operated at the Aerological Observatory of Payerne (491 m, a.s.l.) on the Swiss plateau, and at the Kleine 18 Scheidegg (2061 m, a.s.l.) in the Swiss Alps. The retrieval of the CBL has been validated at Payerne using two 19 reference methods: (1) manual detections of the CBL height performed by independent human experts using the 20 ceilometer backscatter data of the year 2014; (2) values of CBL heights calculated using the Richardson’s 21 method from co-located radio sounding data. We found average biases as small as 27 m (53 m) with respect to 22 reference method 1 (2). Based on the excellent agreement with the two reference methods, PathfinderTURB has 23 been applied to the ceilometer data at the mountainous site of the Kleine Scheidegg for the period September 24 2014 till November 2015. At this site, the CHM15k is operated in a novel, tilted configuration at 71° zenith 25 angle to probe the air next to the Sphinx Observatory (3580 m, a.s.l.) on the Jungfraujoch (JFJ). The analysis of 26 the retrieved layers led to the following results: the CAL reaches the JFJ during 41% of the time in summer and 27 during 21% of the time in winter for a total of 97 days during the two seasons. The season-averaged daily cycles 28 show that the CBL height reaches the JFJ only during short periods (4% of the time) on 20 individual days in 29 summer and never during winter. Especially during summer the CBL and the CAL modify the air sampled in30 situ at JFJ, resulting in an unequivocal dependence of the measured absorption coefficient on the height of both 31 layers. This highlights the relevance of retrieving the height of CAL and CBL in mountainous regions. 32


30
During convective periods, particles and gases are mixed homogeneously within the convective boundary layer 31 (CBL). The upper limit of the CBL corresponds to the interface between the well-mixed region and the free 32 troposphere (FT) above it. This interface, also called entrainment zone (EZ), is a turbulent transition of few tens 33 to few hundreds of meters thick characterized by negative buoyancy flux. The study of the EZ and the way the 34 CBL air is mixed through it has drawn particular attention during the last decades. There are various methods to 35 study the CBL and the EZ, based on profiles of temperature, backscatter or turbulence measured either by radio-2 sounding or by passive and active remote sensing or calculated by numerical models. Amongst the different 1 observational methods, the remote sensing technique ensures the largest amount of profile data. Active remote 2 sensing (acoustic or laser-based) provides the best vertical resolution allowing to resolve the multiple transitions 3 (including the EZ) between different layers in the CBL and the FT. Probably the best-suited instrument to study 4 these dynamics at high temporal and vertical resolution is the ceilometer, a low-power, compact and cost-5 effective version of a research LIDAR. A ceilometer is a laser-based instrument normally emitting in the near-          Close to the ground, for most of the industrial bistatic ceilometers, the overlap between the transmitter and 34 receiver is close to zero. In this region, called blind region, the returned signal is extremely weak, dominated by 35 the noise and it oscillates around zero. It is thus not possible to retrieve the CBL height (CBLH) in this region 8 (low clouds or fog detections are however possible). Above this region, the overlap increases until becoming 1 complete and the noise component becomes negligible compared to the signal at least within aerosol layers. A 2 positive gradient is then expected at the transition between the blind region and the region above. We thus define 3 the lower altitude limit, minH, as the first range where the transition from a zero to a positive gradient occurs and 4 we impose minH not to be higher than 350 m (where the overlap of the ceilometer is normally sufficiently large 5 to allow physical measurements).

6
During the morning and until the end of the afternoon, the CBLH exceeds the height minH due to its convective 7 growth. An additional lower limit for the altitude is minH TURB , which marks the onset of turbulence starting from 8 the ground. Turbulence is calculated based on the temporal variation of the LIDAR signal for each z-level due to 9 the atmospheric variability. The lower altitude limit minH is replaced by minH TURB whenever the latter is higher 10 than the former. The selected minimum limit is called liminf in Figure 1. cloud detected by the ceilometer in the first (lower) layer, whose vertical depth is less than 500 m and whose top 31 (cloud base + depth) is lower than the site-specific climatological CBLH limit set beforehand. This criterion is 32 purely mathematical as the cloud depth provided by the ceilometer just gives the depth of the not-totally 33 attenuated part of the signal and not the real depth.

34
Strong negative and positive gradients 1 Strong positive or negative gradients indicate discontinuities in the vertical aerosol distribution and can then 2 correspond to the CBLH. Strong positive gradients indicate normally a change from an FT region to an aerosol 3 layer or a cloud base or from a CBL region to a cloud base. Strong negative gradients correspond to a signal drop 4 between two adjacent gradient points of 25% (only 15% during the early morning period due to the still present 5 RL above the forming CBL), whereas strong positive gradients correspond to a signal gain between two adjacent 6 gradient points of 15% (only 5% during the early morning period due to the optically thin fog layer often lifted 7 above the forming CBL).
The offset is calculated taking the absolute minimum of Ω over the whole day and at all altitudes. The value of 10 For the same reason, at KSE the SNR condition for calculating the TCAL is not used, since the TCAL could be 1 biased towards the maximum detected range. The shortest path in a graph (the geodesic in the metric space defined by the weights) is calculated using the 4 Dijkstra's algorithm (Dijkstra, 1959) and is based on the original method described by de Bruine et al. (2016).

5
The Ω-weighted graph is constructed using the signal profiles starting from sunrise (midnight at KSE) over

19
In Figure 2c, the weights        3 The comparison shows that PathfinderTURB is robust and that can address the attribution problem adequately.

4
Although PathfinderTURB combines both gradient and variance methods to improve the correctness of the 5 retrieval in different atmospheric conditions, the retrieval's uncertainty grows larger during the afternoon due to 6 the decay of convection before sunset, the weak turbulence and the lack of well-marked aerosol gradients.

7
During this period, temperature or vertical wind variability profiles may provide more valuable information than 8 ceilometer profiles.  Figure   1 3). The distribution of the differences in Figure 4, bottom panel, has a Gaussian shape with slight positive offset 2 values. About the 98% of the data have an error smaller than 500 m, and the 82% have an error smaller than the 3 10% (plus 100 m) of the CBLH retrieved by bR. In general, the correlation between PathfinderTURB 4 (ceilometer-based) and the bR retrievals (RS-based) is not as good as the one between PathfinderTURB and the 5 manual retrievals (both ceilometer-based). For the comparison shown in Figure 4, it should be remembered that   principle during day and night, and so it looks for the first discontinuity in the uninterrupted aerosol region. For 25 this reason and for simplicity we will refer to the retrieved nocturnal boundary layer as to LCBL even when the 26 mixing is not due to convection, but rather to mechanical mixing from the surface and katabatic winds. can say that the LCBLH peak occurs later at KSE than at PAY. During the night, the LCBLH drops, due to the 36 concurrent effects of aerosol gravitational settling, subsidence and katabatic winds, which result from radiative 20 cooling of the surface triggering katabatic drainage flows. A likely explanation of the delay in the onset of the 1 LCBL and of the afternoon peak at KSE is that due to the nighttime katabatic winds driving FT air down into the 2 valley underneath. Depending on the season, these winds can continue to blow for few hours after sunrise 3 (especially from the shaded mountain side) and work against the formation of the LCBL. The LCBLH temporal 4 evolution follows the classical shape of a growing convective boundary layer like over flat terrain, but the 5 growth and the duration of the LCBL occur over a shorter period. This is consistent with the delayed onset of the 6 LCBL due to the persisting katabatic winds in the first hours of the morning and the earlier weakening of 7 convection due to the shading effect of the surrounding mountains and the afternoon onset of the katabatic 8 winds. This phenomenon is particularly enhanced during winter when the solar irradiance is at its minimum and 9 the katabatic winds tend to suppress LCBL during most of the time.

10
In autumn (September-November), the LCBLH shows a less pronounced daily cycle than in spring and summer, 11 this is probably due to the fact that the vertical transport of aerosol-rich air is reduced by the stabilization within 12 the lower troposphere during this period (Lugauer et al., 1998).

19
Although it is impossible to establish the exact origin of the air in the AL (i.e., the injection layer), we can 20 speculate that winter AL is composed of aerosols originating from long-range transport and synoptic-scale 21 lifting, rather than LCBL injections.

22
During summer (dark grey rows in Table 1

12
In the same way as in Figure 7 for the LCBLH, Figure 8 shows the relation between α and the TCAL. TCAL-α pairs is linear, but with different slopes for altitudes below and above the JFJ, with a clear modification 3 of α due to the injections of the LCBL air into the aerosol layer (AL) reaching the in-situ instrumentation at the 4 JFJ. In a more general way, the overall impact of the LCBL, CAL and FT on the in-situ measurements of α is 5 shown in figure 9. As expected the LCBL modifies more the in-situ measurements at the JFJ in terms of absolute 6 value of α, but it is outnumbered by a factor of 10 in terms of occurrences by the CAL. The CAL is in fact more 7 diluted than the LCBL but embeds the JFJ 10 times more frequently than the LCBL and then its impact on the