Inter-comparison of ABL height estimates from different profiling sensors and models in the framework of HyMeX-SOP1

This paper reports results from an inter-comparison effort involving different sensors/techniques used to measure the Atmospheric Boundary Layer (ABL) height. The effort took place in the framework of the first Special Observing Period of the Hydrological cycle of the Mediterranean Experiment (HyMeX-SOP1). Elastic backscatter and rotational Raman 10 signals collected by the Raman lidar system BASIL were used to determine the ABL height and characterize its internal structure. These techniques were compared with co-located measurements from a wind profiler and radiosondes and with ECMWF-ERA5 data. In the effort we consider radiosondes launched in the proximity of the lidar site, as well as radiosondes launched from the closest radiosonde station included in the Integrated Global Radiosonde archive (IGRA). The intercomparison effort considers data from October 2012. Results reveal a good agreement between the different approaches, 15 with values of the correlation coefficient R in the range 0.52 to 0.94. Results clearly reveals that the combined application of different techniques to distinct sensors’ and model datasets allow getting accurate and cross-validated estimates of the ABL height over a variety of weather conditions. Furthermore, correlations between the ABL height and other atmospheric dynamic and thermodynamic variables as CAPE, friction velocity and relative humidity are also assessed to infer possible mutual dependences. 20


Introduction
The ABL is lowest portion of the atmosphere, directly in contact and influence by the Earth's surface , which reacts to the combined action of mechanical and thermal forcing factors. In this layer, as a result of turbulent air motion and vertical mixing induced by shear and buoyancy forces (Stull, 1988), physical quantities such as flow velocity, temperature and 25 moisture are characterized by rapid fluctuations. The variability of these quantities is typically considered to estimate the ABL height.
The evolution of the ABL structure and height has an important impact on meteorology. Accurate measurements of the ABL height (ABLH) allows validating air quality and forecast models and improve specific physical schemes, among others, the boundary layer turbulence and shallow convection parameterizations. However, the complexity of the phenomena occurring 30 within the ABL and the influence of advection and local accumulation processes, in many cases prevents an unambiguous https://doi.org/10.5194/amt-2021-261 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License.
In the present paper this approach is applied to radiosounding data from one station included in the IGRA database and results are compared with independent ABLH estimates from co-located elastic backscatter lidar and wind profiler 50 measurements, from ECMWF model reanalysis, as well as from other standard methods applied to radiosoundings data. As a result of recent technological progresses, lidar systems are presently able to provide continuous measurements of atmospheric variables, as particle backscatter or water vapour concentration profiles, and thus allow providing continuous measurements of the ABLH. Wind profilers are quite effective in providing long-term ABLH measurements as a result of their unattended operation over extended observation periods and the availability of an extensive network of operational 55 wind profilers over wide areas of the globe. An effective approach to determine the ABLH from wind profiler signal-tonoise ratio (SNR) measurements was developed by Angevine et al. (1994) in the early nineties. However, the operational use of wind profilers is limited by their lack of sensitivity within the surface atmospheric layer (up to 500 m), which prevents from an adequate monitoring of the ABLH and structure at night or in the presence of shallow ABLs. Additionally, ABLH estimates from wind profilers are very sensitive and frequently affected by the presence of insects' swarms. 60 In the present research effort we consider Raman lidar measurement from BASIL collected in Southern France in the period September-November 2012 in the frame of HyMeX-SOP1. We focus our attention on the measurements carried out during October 2012. This paper does not represent the first research effort dedicated to an extensive inter-comparison of different ABLH sensors/methodologies. In a previous paper, Seibert et al. (2000) compared different methods to estimate the ABLH https://doi.org/10.5194/amt-2021-261 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. from radiosounding data and other instruments and carefully examined advantages and shortcomings of all investigated 65 approaches.
The considered approaches are shortly summarized in what follows. The turbulence method identifies the ABLH as the depth of the lowest continuous turbulence layer (Stull, 1988). The turbulent region is determined by tracking the fluctuations of the different wind components (U, V, and W), for example through a high-pass wavelet filter (Wang et al. 1999). Such fluctuations can be identified in wind lidar data, but measurements can also be performed with radiosondes and tethered 70 balloons or in-situ sensors on-board scientific aircrafts. The ascent velocity of tethered balloons can be carefully controlled to cope with the desired time-height resolution. However, in case of high wind speed or strong convective activity, tethered balloons are inapplicable. A major advantage of the use of aircraft sensors is the simultaneous measurements of both mean and turbulent quantities with high sampling rates. However, lack of observations from the lower atmospheric levels may result from limitations and restrictions to low level flights. 75 The temperature gradient method relies on the identification of temperature inversions, which are found to typically cap the ABL top and determine a maximum in the potential temperature lapse rate (Bianco and Wilczak, 2002;Martucci et al. 2007).
Potential temperature (z), which is a function of atmospheric pressure and temperature, is given by the expression: ( 1 ) with P 0 is the standard pressure, T and P are the pressure and temperature, respectively, at altitude z, with is R/Cp = 0.286. 80 (z) tends to keep nearly constant with height within the mixed layer. The level of the maximum vertical gradient in potential temperature indicates the transition from a convectively unstable region, located below this maximum, to a stable or more stable region, located above the maximum. The stable layer at the top of the mixed layer stops the turbulent eddies from further rising. Stable layers characterized by increasing temperatures with height (positive lapse rates, called capping inversions) can prevent the development of deep convection. During the day, the level at which air parcels become 85 negatively buoyant corresponds to a main temperature inversion. Such inversions can be identified in radiosonding data, with radiosondes providing very accurate information throughout the troposphere. Additionally, radiosonding data can provide a long observational record, which is particularly suited for ABLH climatological studies (Madonna et al. 2021). However, radiosonde measurements are characterized by lack of temporal resolution (only 2-4 launches per day from each station) and a poor geographical distribution, with an uneven distribution density in the two hemispheres. 90 The Richardson number method relies on the identification of Richardson number gradients, which is an important diagnostic indicator of dynamic flow stability. This method assumes the ABLH to be the level where the so called "bulk Richardson number for the entire ABL" exceeds a specific threshold value, Ri bc . Ri bc at height z can be calculated from the wind speed and the potential temperature values at z and at surface level, as originally reported in Hanna (1969) and extensively described in e.g., Stull (1988) and Garratt (1994). Such gradients can be revealed in wind lidar, wind profiler, 95 radiosonde and aircraft in-situ sensors' profile data (Sicard et al., 2006 The ABLH can also be estimated from the vertical wind profile, using e.g. the lowest wind-speed maximum height, often referred to as the low-level jet (LLJ) height (Melgarejo and Deardorff,1974). The LLJ height is again identifiable in wind lidar, wind profiler, radiosonde and aircraft in-situ sensors' profile data.
The paper outline is the following. Section 2 shortly describes the profiling sensors and model data involved in the inter-100 comparison effort. Section 3 illustrates the results from the inter-comparison effort. Finally, section 4 provides a summary, concluding remarks and indications for possible future follow-on studies. intensive observation periods (IOPs, Di Girolamo et al., 2016;Stelitano et al., 2019). BASIL is capable to perform highresolution and accurate measurements of atmospheric temperature and water vapour, both in daytime and night-time, based on the application of the rotational and vibrational Raman lidar techniques, respectively, in the UV (Di Girolamo et al., 2004, 2017. This measurement capability makes BASIL an effective tool for the characterization of water vapour inflows in Southern France, which are a key ingredient of heavy precipitation events in the North-western Mediterranean 115 basin. BASIL makes use of a Nd:YAG laser source capable of emitting pulses at 355, 532 and 1064 nm, with single pulse energies at 355 nm, i.e. the wavelength used to stimulate rotational and roto-vibrational Raman scattering from atmospheric molecules, of 500 mJ (average optical power of 10 W at a laser repetition rate of 20 Hz). The receiver includes a Newtonian telescope in (45-cm diameter primary mirror). Data are acquired with a rough vertical and temporal resolution of 30 m and 10 sec, respectively, but vertical and temporal smoothing is typically applied when processing water vapour and temperature 120 profile measurements for data assimilation purposes or process studies.

Determination of the ABLH from Raman lidar measurements
There are several methodologies to determine the ABL height from elastic lidar signals, which rely on the circumstance that aerosols are more abundant within the ABL than in the free troposphere and they can act as tracers of atmospheric motions. 125 An extensively used methodology relies on the detection of vertical gradients in elastic backscatter lidar echoes, which result from such echoes being much stronger within the ABL than in the free troposphere.
The elastic lidar equation, expressed in terms of number of collected photons as function of height, is defined as: where  0 is the emitted and received lidar wavelength, respectively, z is the vertical height, 0 is the number of laser emitted 130 photons, O(z) is the overlap function, A is the telescope collection area,  mol (z) and  par (z) is the backscatter coefficient for molecules and particles, respectively, T mol (z) and T par (z) represent the molecular and particle contribution to atmospheric transmissivity, respectively, and P bgd is the background signal associated with solar irradiance and detectors' noise. In order to compress the signal dynamical variability and define an uncalibrated quantity proportional to total (molecular + particle) attenuated backscattering coefficient, it is often preferable to make use of the range-corrected signals (RCSs), which is 135 defined as: The ABL height is estimated from the height derivative of RCSs through the expression: Transitions between different aerosol layers are identified with the minima in expression (3), with the absolute minimum 140 typically representing the ABLH. For the specific purposes of our present study, the elastic backscatter signal at 355 nm, P  (z), is considered in expression (2). The methodology is applied to signals with vertical and time resolutions of 30 m and 5 min, respectively (Summa at al ., 2013;Vivone et al., 2021). Overlap effects affect lidar signals in the lower few hundred meters have marginal effects on gradient measurements and consequently ABLH estimates.

WIND profiler
The five-beam wind profiler (WPR) considered in the study was also deployed in Candillargues. The system is manufactured by Degreane (model PCL 1300) and operates in the UHF band with a primary frequency at 1.274 GHz.. A detailed description of the WPR, its specifications and main working parameters, data processing methodologies and delivered 150 geophysical products are illustrated in Saïd et al. (2016). The WPR operated almost continuously throughout the duration of HyMeX-SOP1 (Saïd et al., 2018). For the ABL height measurements reported in this paper, the WPR was operated in low mode, with a pulse length of 1μs. This allows sampling the lower troposphere from 0.15 to 5.7 km a.g.l., with a vertical resolution of 150m. The methodology to determine the ABL height relies on the identification of a distinctive strong peak in the WPR time-height reflectivity plot (Gage et al., 1990), though the strength of this peak may depend on a variety of 155 factors. While this approach is very effective in the determination of the ABLH, its applicability can be limited by the presence of strong reflectivity peaks associated with marked temperature and humidity gradients. This is the primary uncertainty source WPR-based ABLH measurements. This aspect will be carefully accounted for when comparing the different approaches. intensive observation periods, with a launching rate of up to one launch every 1.5 h. The radiosondes were set to provide vertical profiles of atmospheric pressure, temperature, humidity, and wind direction and speed during both the ascent and 165 descent phases. The thin-wire temperature sensor, whose measurements are used for the ABLH estimate, is characterized by a very fast response time and has a hydrophobic coating protection to reduce the effects of evaporative cooling after emerging from clouds (Dirsken et al., 2014;Madonna et al., 2020). ABLH estimates are obtained from the radiosonde data based on the application of the temperature gradient method, which relies on the identification of maxima in the potential temperature vertical gradient. Small, but non-negligible uncertainties (0.05 K) may affect radiosonde temperature 170 measurements as a result of the sonde's pendulum motion and the resulting increased sensor ventilation (Dirksen al., 2014), with again marginal effects on temperature gradient measurements and consequently ABLH estimates.

IGRA DataBase
ABLH estimates from BASIL and WPR measurements carried out during HyMeX-SOP1 were compared with those 175 obtained from the radiosondes launched from nearest Integrated Global Radiosonde archive (IGRA) radiosounding station.
IGRA is the most comprehensive, authoritative collection of historical and near-real-time radiosonde and pilot balloon observations, with global coverage, maintained and distributed by the National Oceanic and Atmospheric Administration's National Centers for Environmental Information (NCEI). Data are extracted from version V2 IGRA database, which was released in 2016 (Durré et al., 2018) and includes enhanced quality with respect to the previous version (V1). ABLH 180 estimates from IGRA radiosondes are obtained based using agin the temperature gradient method. The specific IGRA station considered in our study is Nimes-Courbessac (lat: 43.8569°N, 4.4064°E, 60 m asl, WMO index= 7645), typically carrying out four radiosonde launches per day and using GPS radiosondes manufactured by Meteomodem (model: M10). The performance of M10 radiosondes was assessed and verified to be accurate during the WMO 2010 radiosonde intercomparison effort in Yangjang (Nash et al., 2011, Madonna et al. 2021, being again characterized by negligible 185 uncertainties in temperature gradient measurements and consequently ABLH estimates.

ECMWF-ERA5
ABLH estimates from the aforementioned observational sources have been also compared with ECMWF-ERA5 atmospheric reanalysis. ECMWF-ERA5 is the latest reanalysis produced by ECMWF, including hourly data on a regular latitude-190 longitude grids, with a 0.25° x 0.25° resolution (Hersbach et al., 2020) and atmospheric parameters provided at 2m and at an additional 36 pressure levels. ERA5 is publicly available through the Copernicus Climate Data Store (CDS, https://cds.climate.copernicus.eu).
In order to properly carry out the comparison, the nearest ERA5 grid point to the CV site was considered, assuming the representativeness uncertainty associated with the use of the nearest grid-point to be comparable with the uncertainty 195 affecting most interpolation approaches (e.g. kriging, bilinear interpolation, etc.). In general, reanalysis reliability can considerably vary depending on the location, time and selected atmospheric variable (Dee et al., 2016). Mixed layer https://doi.org/10.5194/amt-2021-261 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. parametrizations in atmospheric forecast models typically consider boundary layer height estimates from entraining parcel models. However, to guarantee an effective and reliable ABLH monitoring, also in neutral and statically stable atmospheric conditions, the bulk Richardson number method is frequently preferred as an independent diagnostic proxy in turbulence 200 parametrization (Seidel et al., 2012), this also being the method considered in the present paper. The algorithm used in ERA5 requires approximations, which ultimately increase uncertainties (Seidel et al., 2012): for example, the lack of accurate information on the local surface roughness affects friction velocity estimates and, consequently, surface frictional effects are is compared to the virtual dry static energy at the boundary layer top: 1 and |∆ | ( 7 ) 215 and being the are the horizontal wind speed components at the ABLH. The ABLH is obtained from both the radiosonde and model data using this algorithm, which is applied from the surface upwards. In case the ABLH falls in between two levels, a linear interpolation is applied to determine its exact position.
In the present paper we provide ABLH estimates based on the application of different approaches with the aim to assess the performance achievable with each observational or model data and verify their applicability in different atmospheric 220 scenarios. In the attempt to ensure the highest possible performance achievable with each observational or model data, deviations observed the in the comparison of the different datasets sensors and mothers and approaches are carefully analyzed and discussed, also counting for results from previous inter-comparison efforts (Seidel et al., 2012).

Results 225
In this section we illustrate and discuss the ABLH estimates as obtained from measurements/model data through the application of the five different approaches/sensors/models illustrated in the previous sections. We first provide a more climatological assessment, focusing on the evolution of the ABLH throughout the duration of the month of October 2012.  reference ABLH values, with R 2 being equal to 0.94 and A being equal to 1.04, which confirms a small positive bias (4 %) 265 affecting these radiosondes when compared with the reference estimate. Figure 3b compares ABLH values from the IGRA radiosondes with the reference ABLH values, with R 2 being equal to 0.81 and A being equal to 1.00, which confirms the very good agreement between these radiosondes and all other sensors/models. Figure 3c compares ABLH values from the wind profiler with the reference ABLH values, with R 2 being equal to 0.93 and A being equal to 1.01, which confirms a very small positive bias (1 %) affecting the wind profiler when compared with the reference estimate. Figure 3d  Additional parameters from ERA5 reanalysis data have been considered to corroborate the analysis the data and the interpretation of the observed atmospheric features. Figure  We also compared and tried to quantitative correlate ABLH estimates with the variability of specific atmospheric dynamic and thermodynamic variables, as CAPE, friction velocity and relative humidity. CAPE daily gradient values were computed 290 considering the time series of CAPE values during the time interval 1-31 October 2012. A linear fit was applied to the time series of CAPE daily gradient and the friction velocity values in the time period 1-31 October 2012 vs. the corresponding ABLH estimates. Considered ABLH estimates are those from ERA5 reanalysis and the mean of the 5 different approaches/sensors/models considered above. The correlation between ERA5 ABLH estimates and the corresponding friction velocity values is found to be 0.71, while the correlation between the mean ABLH estimates and the ERA5 friction 295 velocity values is found to be 0.75. The correlation between ERA5 ABLH estimates and the corresponding CAPE daily gradient values is 0.49, while the correlation between the mean ABLH estimates and the ERA5 CAPE daily gradient values is found to be 0.52. Finally, the correlation between ERA5 ABLH estimates and the corresponding relative humidity values https://doi.org/10.5194/amt-2021-261 Preprint. Discussion started: 14 September 2021 c Author(s) 2021. CC BY 4.0 License. is 0.32, while the correlation between the mean ABLH estimates and the ERA5 relative humidity values is 0.42. These values reveal a good correlation between ABLH estimates and corresponding friction velocity values, a very mild correlation 300 between ABLH estimates and corresponding CAPE daily gradient values and a missing correlation between ABLH estimates and corresponding relative values. These results are summarized in table 3.
For the purpose of assessing the performance in the characterization of the short-term variability, our analysis was also focused on one specific case study, covering the daytime portion of 18 October 2012. Figure 5a illustrates the time-height cross-section of the particle backscattering coefficient at 355nm,  355 (z), as measured by BASIL measurements of over the includes ABLH estimates obtained from the application of the temperature gradient method to the the IGRA radisounding 315 data (blue star), these values being in very good agreement with those simultaneously estimated from the lidar data. Figure 5b illustrates the time-height cross-section of the water vapour mixing ratio measurements carried out by BASIL on the same day, which are displayed over the same time intervals considered in figure 5a. A dry layer appears at the ABL top, probably resulting from sub-cloud low-level rain evaporation. This dry layer may ultimately have contributed to convection regeneration events observed throughout the passage of the MCS (Li et al. 2009;Morrison et al. 2009). It is to be specified 320 that rain evaporation significantly contributes to the heat and moisture budgets of clouds (Emanuel et al., 1994), but few observations of these processes are available (Gamache et al., 1993). Dry layers are frequently observed in mid-latitude convective environments as a result of air being advected from different source regions under directionally sheared vertical wind profiles (Carlson and Ludlam 1968). Deep convective precipitation events are influenced, and frequently favored, by the presence of aerosols and midlevel dry layers, and this circumstance may have played a significant role in the formation 325 and development of the observed MCS.

Conclusions
In the present paper we illustrate and discuss the results from an inter-comparison effort considering ABL height estimates from different sensors/techniques. The effort was carried out in the framework of HyMeX-SOP1. A climatological 330 assessment focusing on the evolution of the ABL height throughout the duration of the month of October 2012 is provided.
Results reveal a good agreement between the different sensors/approaches, all of the them being able to capture the major