Mind-the-gap part I: Accurately locating warm marine boundary layer clouds and precipitation using spaceborne radars

Mind-the-gap part I: Accurately locating warm marine boundary layer clouds and precipitation using spaceborne radars Katia Lamer1*,2**, Pavlos Kollias2,3,4, Alessandro Battaglia5,6 and Simon Preval5 1 City University of New York affiliation 2 Brookhaven National Laboratory 3 Stony Brook University 4 Cologne University 5 University of Leicester, Leicester, UK 6 UK National Centre for Earth Observation * Affiliation when work was conducted ** Current affiliation


1
Introduction 24 25 Because of their ubiquitous nature and of the way they interact with solar and longwave radiation, warm marine 26 boundary layer (WMBL) clouds play a crucial role in the global energy budget [Klein and Hartmann, 1993]. 27 Unfortunately, numerical models still struggle to properly represent their coverage, vertical distribution, and 28 brightness (e.g., [Nam et al., 2012]). This uncertainty ultimately affects our confidence in future climate projections 29 [Bony et al., 2015;Sherwood et al., 2014]. Climate simulations could be improved from comparisons with additional 30 observations of the macrophysical and microphysical properties of WMBL clouds, as well as from improvements in 31 our understanding of the relationships between low-level clouds and their environment. 32 33 Millimeter-wavelength radar signals, because of their ability to penetrate clouds, have long been used to document 34 the vertical distribution of WMBL clouds (e.g., [Haynes et al., 2011;Sassen and Wang, 2008]) and their internal 35 structure (e.g., [Bretherton et al., 2010;Dong and Mace, 2003;Huang et al., 2012;) as well as to 36 identify precipitation (e.g., [Ellis et al., 2009;Leon et al., 2008;Rapp et al., 2013]) and characterize its vertical 37 structure (e.g., [Burleyson et al., 2013;Comstock et al., 2005;Frisch et al., 1995;Kollias et al., 2011]). However, the 38 representativeness of radar observations largely depends on factors such as coverage, radar sensitivity, 39 vertical/horizontal resolution and on the presence of clutter. 40 41 Spaceborne radars are often preferred over ground-based and airborne ones because of their ability to cover vast areas 42 of the globe [Battaglia et al., Submitted]. The first spaceborne Cloud Precipitation Radar (CPR) designed to detail the 43 vertical structure of clouds was launched in 2006 onboard CloudSat [Stephens et al., 2002]. The CloudSat-CPR is still 44 operational; it transmits a 3.3 microsecond pulse with a 1.4 km field of view at the surface and can achieve a sensitivity 45 of -28 dBZ after its measurements are averaged in 0.32-s time intervals and sampled at 0.16-s along its nadir track 46 [Stephens et al., 2002]. However, the CloudSat-CPR's long power pulse also generates a surface clutter echo which 47 CPR was mispointing, which leads to vertical offset in the surface peak return. 137 138

ARM Ground-based Observations 139 140
The ARM program's KAZR is a 34.86 GHz (i.e., Ka-band) radar able of generating a 4 microsecond long symmetrical 141 vertical pulse creating a 0.3° wide 3-dB beamwidth. Following signal integration (1-s, 6,000-pulses), this radar 142 achieves a -44 dBZ minimum detectable signal (MDS) at 1 km. The KAZR is able to collect observations from 87 m 143 above ground to 18 km at ~30 m vertical resolution and 2 s time resolution [Lamer et al., 2019]. Because the KAZR's 144 observations are not oversampled in the vertical, they are considered more independent than that of the CloudSat-145 CPR. 146 147 We analyze the complete data record collected by the ground-based ARM sensors between October 2015 and 148 November 2017 (719 days) to 1) characterize the properties of WMBL clouds and precipitation (results in Sect. 4,0) 149 and 2) to evaluate the performance of theoretical radar architectures in detecting those clouds (results in Sect. 5.0). 150 This period also includes the 117 CloudSat overpass days, which we analyze separately to identify gaps specific to 151 the currently deployed CloudSat-CPR (results in Sect. 3.0). 152 153 For each analysis, we extract several complementary datasets from the ARM archive: i) KAZR general mode 154 (processing level a1): reflectivity, snr_copol (co-polar signal to noise ratio), ii) ceilometer: first_cloud_base_height, 155 iii) Parsivel laser disdrometer: equivalent radar reflectivity, and iv) radiosonde: temperature. 156 157 KAZR signal-to-noise ratio measurements are used as input to the Hildebrand and Sekhon [1974] algorithm to 158 distinguish significant echoes (hydrometeors and clutter) from noise. Liquid cloud base height determination from 159 collocated ceilometer is used to isolate radar echoes associated with cloud (above the first liquid cloud base height) 160 and precipitation (below the first liquid cloud base height) and to filter out clutter in the subcloud layer. Clutter filtering 161 is based on the argument that precipitation falling from cloud base should be continuous, thus any echo in the subcloud 162 layer detached from the main echo is labelled as clutter and is filtered out. All echoes thinner than 90m (3 range gates) 163 are also labelled as clutter and filtered out; comparison with the ceilometer confirms that this step does lead to the 164 removal of cloudy echoes. An example of processed radar reflectivity from KAZR is depicted in Fig. 1a to the forward-simulations. First, assuming a constant horizontal wind speed of 10 m s -1 , the KAZR time axis is 198 converted to horizontal distance. Then, to emulate the surface reflectivity which is not seen by KAZR, an artificial 199 surface echo is added to the processed KAZR reflectivity field at 0 m altitude (see Appendix I for more information 200 on how real CloudSat-CPR observations were used to construct this surface echo). Each spaceborne radar 201 configuration is simulated by first horizontally convolving the high-resolution (30 m x 20 m) KAZR reflectivity fields 202 using an along-track weighting function represented using a symmetrical gaussian distribution covering a distance 203 equivalent to 2 times the along-track field of view and then by vertically convolving the horizontally convolved 204 reflectivity field using either of the two range-weighting functions depicted in Fig. 2. The asymmetrical range 205 weighting function is modelled after that of the EarthCARE-CPR which was obtained from prelaunch testing of the 206 EarthCARE-CPR (provided by the mission's engineering team). The symmetrical range-weighting function used 207 (only) for the CloudSatf forward simulation is modelled using a gaussian distribution adjusted to produce a surface 208 https://doi.org/10.5194/amt-2019-473 Preprint. Discussion started: 21 January 2020 c Author(s) 2020. CC BY 4.0 License. clutter echo profile similar to that observed by the CloudSat-CPR post-launch (more information in Appendix I). 209 Finally, along-track integration is emulated by averaging the convolved profiles in sections dictated by the integration 210 distance of each spaceborne radar without overlap between the section. Note that these forward-simulations are two 211 dimensional and as such do not capture cross-track effects; Also note that liquid attenuation and noise are not 212 represented. 213 For cloud and precipitation characterization, the forward-simulated radar reflectivity fields are finally filtered for 214 surface clutter. To do this, forward simulations of clear sky conditions are used to estimate the vertical extent and 215 intensity of surface clutter. For each radar configuration, for all heights affected by surface clutter, the clear sky surface 216 clutter reflectivity is removed from the forward-simulated radar reflectivity and only echoes with reflectivity at least 217 3 dB above the surface clutter reflectivity are conserved and deemed reliable. Otherwise, for all heights above the 218 surface clutter, only those echoes with reflectivity below the radar MDS are filtered out. 219 220

2.4
Evaluation metrics 221 222 Radars alone do not have the capability to distinguish between clouds and precipitation. For this reason, we often refer 223 to them as hydrometeor layers. The current study aims at characterizing: 224 225 i) the base of the lowest hydrometeor layer (cloud or virga base being indistinguishable), which we take to 226 be the height of the lowest radar echo in the profile; 227 ii) the top of the highest hydrometeor layer (i.e. cloud top), which we take to be the height of the highest 228 radar echo in the profile; 229 iii) the depth covered by hydrometeor layers, which we estimate as the distance between the top of the 230 highest hydrometeor layer and the base of the lowest hydrometeor layer. 231 Note that we report hydrometeor boundary heights at the center point of each radar's vertical range gate and not as its 232 upper or lower limit. This distinction, while seemingly insignificant for radars operating at a fine range sampling (e.g., 233 KAZR 30 m), can become important for radar systems having a coarse range sampling (e.g., the CloudSat-CPR 240 234 m). 235 236 We also estimate over the entire observation periods: First, agreement between the KAZR reported cloud cover and the ceilometer reported cloud cover confirms that the 266 KAZR's sensitivity is sufficient to detect even the most tenuous clouds forming in this marine boundary layer regime; 267 this makes the KAZR an ideal sensor to document the properties of WMBL clouds and evaluate the CloudSat-CPR's 268 performance (Fig. 3a). Although not expected to perfectly match, the large hydrometeor cover discrepancy between 269 the KAZR (46.7%) and CloudSat-CPR (27.4%) suggest that the CloudSat-CPR fails to detect clouds in more than a 270 few (on the order of ~40% ) of the atmospheric columns it samples (Fig. 3a). On the other hand, the CloudSat-CPR 271 seems to capture the shape and magnitude of the hydrometeor fraction profile above 1.0 km reasonably well (Fig. 3b). 272 This suggests that the CloudSat-CPR is able to detect the bulk of the thick hydrometeor layers controlling hydrometeor 273 fraction above 1.0 km. This also leads us to believe that the CloudSat-CPR's hydrometeor cover biases results either 274 from its inability to detect clouds entirely located below 1.0 km and/or due to its inability to detect thin and narrow 275 hydrometeor layers that are negligible contributors to hydrometeor fraction. Detailed analysis of the location of 276 individual cloud tops show evidence supporting both of these postulations (Fig. 4a). Specifically: 1) The distribution 277 of KAZR-detected cloud top heights shows clouds below 0.6 which are undetected by the CloudSat-CPR. We estimate 278 that this near-surface cloud mode produces 7.5% of the total cloud cover and so its misdetection could explain nearly 279 half of the CloudSat-CPR hydrometeor cover bias. 2) The distribution of KAZR-detected cloud top heights also shows These elevated cloud tops modes are likely related to the several echo bases between 1.4 and 2.5 km that nearly all 282 went undetected by the CloudSat-CPR (Fig. 4b). A figure showing time-height observations from two additional 283 overpass days allows us to visualize that these layers are generally thin, weakly reflective, and broken ( Fig. 4i and ii). 284 We speculate that misdetection of such thin/tenuous clouds explains the remaining of the CloudSat-CPR's cloud cover Analysis of the ground-based observations suggests that WMBL cloud fraction exceeds 5% at all heights between 320 305 m and 2.09 km with cloud fraction peaking at 1.13 km ( Fig. 5a; solid black curve). On the other hand, rain tends to be 306 found in the sub cloud layer below 1.28 km altitude occupying the largest fractional area between 100 m and 1.1 km 307 ( Fig. 5a; dotted black curve). The low height at which WMBL clouds and precipitation are found is especially 308 challenging for spaceborne system which are known to suffer from contamination from the surface return. We estimate 309 that roughly 20% of the cloud echoes and 52% of the rain echoes recorded by the KAZR fall within the CloudSat-310 CPR's surface echo region which extends at best only to 0.75 km ( Fig. 5a; red curves). 311

312
The intensity (in terms of radar reflectivity) of cloud and precipitation also largely affects their ability to be detected 313 by radars. Using KAZR observations, we characterized the intensity of the hydrometeor echoes observed at each 314 height and report in Fig. 5b (colormap) the fraction of echoes with a reflectivity above a given threshold at each height. 315 Generally, cloud and precipitation producing radar reflectivity above a radar MDS can be detected. Thus, we would 316 expect that the CloudSat-CPR, with its -27dBZ MDS (depicted by the broken black line on Fig. 5b), should have the 317 capability to detect at best 80% of all cloud and/or echoes forming at any given height, de facto missing at least 20% 318 https://doi.org/10.5194/amt-2019-473 Preprint. Discussion started: 21 January 2020 c Author(s) 2020. CC BY 4.0 License. of hydrometeor echoes. Radar performance degrades within the surface clutter region. In the clutter region, only those 319 hydrometeor echoes whose intensity is larger than the surface echo intensity can be detected. To reflect this and for 320 reference, we overlaid on Fig. 5b the median reflectivity recorded by the CloudSat-CPR in clear sky days between 321 2010 and 2016 as well as its variability as quantified by the interquartile range (broken and dashed black lines 322 respectively). Over that time interval, the CloudSat-CPR's median surface echo varied from 37 dBZ at the surface 323 decreasing to -27 dBZ at 0.75km. Using this curve, we estimate that at 0.5 km height, based simply on sensitivity, the 324 CloudSat-CPR would miss at least 80% of the echoes detected by KAZR because their reflectivity is below that of the 325 surface clutter. 326

327
Adding to the challenge is the fact that boundary layer systems are shallow. Based on KAZR observations, 53% of 328 WMBL systems (cloud and rain) forming at ENA are shallower than 500 m, 33% shallower than 250 m and 16% 329 shallower than 100 m ( Fig. 5c; red line). Sampling hydrometeor layers using radar pulses longer than the hydrometeor 330 layer thickness inherently produces partial beam filling issues, which lead to a weakening of the returned power. This 331 results in an underestimation of the reflectivity of the thin echoes sampled and may even lead to their misdetection if 332 the resulting reflectivity is below the radar MDS. There is also an unfortunate relationship between hydrometeor layer 333 thickness and mean reflectivity such that thin layers not only suffer from more partial beam filling, but also have 334 We emulate the impact of these radar modifications by constructing forward-simulations for 7 radar configurations, 349 each of which has been gradually improved by the aforementioned radar modification (described in Sect. 2.3, Table 1  than the real CloudSat-CPR when compared to KAZR indicating that the forward simulator captures enough of the 364 radars characteristics to reasonably emulate its performance. In a nutshell, the CloudSatf underestimates hydrometeor 365 cover by more than 10% (Fig. 7a) likely owing to its misdetection of an important fraction of clouds with tops between 366 750 m and 1.75 km (Fig. 8a) and its inability to detect the small fraction of clouds forming entirely below 500 m. Just 367 like the real CloudSat-CPR, the CloudSatf performs well in capturing hydrometeor fraction between 750 m and 3 km 368 but poorly below that height since it suffers from contamination by surface clutter (Fig. 7b). 369 370 Prelaunch testing of the EarthCARE-CPR showed that its pulse generates an asymmetrical point target response. This 371 mean that, unlike the CloudSat-CPR, the EarthCARE-CPR has an asymmetrical range weighting function (Fig. 2). 372 The range weighting function of the EarthCARE-CPR's pulse has a rapid cut off at a factor of 0.5 time the pulse length 373 at its leading edge, and a longer taper extending off to 1.5 times the pulse at its trailing edge. To isolate performance 374 changes resulting strictly from this range weighting function, we contrast the result of forward simulations performed 375 with the CloudSat-CPR's original configuration (CloudSatf results depicted in royal blue) and with a CloudSat-like 376 configuration with the EarthCARE-CPR's asymmetrical range weighting function (CloudSata, results depicted in 377 cyan). Time-series comparison of CloudSata (Fig. 6b) and CloudSatf (Fig. 6a) reflectivity shows that the asymmetrical 378 range weighting function reduces the vertical extent of the surface clutter echo, allowing for the detection of a larger 379 fraction of hydrometeor at 500 m. Over the entire set of 719 forward simulations, this leads to improvements in the 380 representation of the hydrometeor fraction profile (Fig. 7b) and of the echo base height distribution (not shown) around 381 500 m. However, differences in the echo base height from KAZR (black dots) and from CloudSata (cyan dots) suggest 382 that changes in the shape of the pulse point target response alone are insufficient to accurately detect the base of the 383 precipitating WMBL systems found at the ENA (Fig. 6b). We also note that the change in range weighting function 384 shape alone only marginally improve CloudSatf's ability to determine hydrometeor cover (improvement from 27.9% 385 to 28.2% compared to 39.1% reported by KAZR); The reason for this is that hydrometeor cover is controlled by thin, 386 tenuous clouds and clouds located entirely below 0.5 km. As a potential drawback, the asymmetrical range weighting 387 function seems to lead to slightly more vertical stretching of cloud top signals (on average 37 m) such as visible by 388 comparing the examples in Fig. 6a and 6b, and in Fig. 8a. When compounded over the entire ensemble of forward 389 simulated clouds this leads to a 0.24% overestimation of hydrometeor fraction at all height between 0.75 and 3.00 km the pulse lengths which is accompanied by additional power being focused in that region of the pulse in contrast to a 392 symmetrical pulse such as that of the CloudSat-CPR (see Fig. 2). 393

394
Besides having an asymmetrical range weighting function, the EarthCARE-CPR will also operate with a MDS of -35 395 dBZ which is 7 dB more sensitive than the CloudSat-CPR. To isolate performance changes resulting strictly from this 396 sensitivity enhancement, we contrast the result of forward simulations performed with a CloudSat-like configuration 397 with the asymmetrical range weighting functions (CloudSata, results depicted in cyan) with that of a CloudSat-like 398 configuration with both an asymmetrical range weighting function and enhanced sensitivity (CloudSata+es, results 399 depicted in purple). Time-series comparison of CloudSata+es (Fig. 6d) and CloudSata (Fig. 6b) reflectivity shows that 400 the sensitivity enhancement allows for the detection of hydrometeors in previously undetected columns such as the 401 broken hydrometeor segments observed by KAZR around 100 km distance along the forward-simulated track. 402 Quantitatively, the more sensitive CloudSat-CPR configuration detects 8% more cloudy columns than either of the 403 other two CloudSat-CPR configurations discussed so far (i.e., with or without the asymmetrical range weighting 404 function) missing only 2.4% of the cloudy columns detected by KAZR (Fig. 7a). This implies that, if an important 405 mission objective is detecting even tenuous cloudy columns, improving the MDS is crucial. That being said, we advise 406 against accomplishing this by transmitting a longer pulse (e.g., like done in the first 4 years of operation of the GPM-407 CPR) since there are two main drawbacks to transmitting a long pulse with a higher sensitivity, both caused by partial 408 beam filling. Firstly, the enhanced sensitivity leads to additional vertical stretching of cloud boundaries, an effect 409 visible between 400 and 800 km along track when comparing Fig. 6d to 6b. This is because the signal from cloud 410 boundaries away from their location resulting from their interaction with the edges of the radar range weighing 411 function now exceeds the MDS. Secondly, the enhanced sensitivity also leads to previously undetected thin layers 412 becoming detectable, but it stretches them vertically at least to the vertical extent of the radar pulse length. From 413 changes in the location of the cloud top height distribution peak shown in Fig. 8a, we estimate that enhancing the 414 sensitivity of a 3.3 microsecond long pulse from -28 dBZ to -35dBZ would lead to a 250 m bias in detected cloud top 415 height for the types WMBL clouds forming at the ENA. Moreover, because it both vertically stretches clouds and 416 detects more real clouds, the highly sensitive CloudSata+es overestimates hydrometeor cover by up to 7% at all heights 417 between 500 m and 3.0 km (Fig. 7b). 418 419 Since EarthCARE will travel at an altitude closer to the Earth surface it will also have half the horizontal field of view 420 of CloudSat. Our results suggest that halving the CloudSat-CPR's horizontal field of view and halving its integration 421 distance would lead to a slight reduction in its estimated hydrometeor cover (1.7% less). We take this as an indication 422 that the larger horizontal field of view of the CloudSat-CPR only marginally artificially broadens broken clouds (see 423 CloudSata+es+hf, results depicted in gold in Fig. 7). That being said, note that this result, like all the others presented 424 here, is based on 2-D forward-simulation and as such it does not take into account cross-track effects which may also 425 generate biases especially in sparse broken cloud fields. Comparison of the ensemble of EarthCARE (magenta) and CloudSata+es+hf (gold) forward-simulations indicates that 438 this precision can be achieved without causing significant biases in hydrometeor cover (Fig. 7a) or hydrometeor 439 fraction (Fig. 7c). 440 441 Although the EarthCARE-CPR's performance is significantly better than that of the CloudSat-CPR when it comes to 442 detecting thin, tenuous and broken clouds as well as clouds and precipitation near 500 m, its configuration still does 443 not allow to detect all WMBL clouds and precipitation. Remaining detection limitations occur below 500 m within 444 the region of the surface clutter echo. Additional reduction of the vertical extent of the surface clutter can be achieved 445 by reducing the pulse length. This, however, comes at the expense of reduced sensitivity. Comparing EarthCARE 446 (results depicted in magenta), ACCP250 (results depicted in red) and ACCP100 (results depicted in green) simulations 447 allows us to see the gain and penalty incurred from shortening the radar vertical range resolution from 500 m, to 250 448 m to 100 m at the cost of reducing sensitivity from -35 dBZ to -26 dBZ and -17dBZ. In alignment with our previous 449 conclusion that a high sensitivity is necessary for detecting all cloudy columns, reducing the radar pulse length and 450 sensitivity reduces the fraction of cloudy columns which can be detected by the ACCP configurations (Fig. 7a). For 451 instance, the ACCP250 configuration, which is nearly as sensitive as CloudSat (-26 dB versus -28 dB), performs very 452 similarly in terms of the number of cloudy columns it is able to detect (Fig. 7a) and in terms of how well it can capture 453 the vertical distribution of hydrometeors between 500 m and 3.0 km (Fig. 7d) which we determined is influenced by 454 the deeper more reflective clouds rather than the thin and tenuous ones. The ACCP250 configuration does, however, 455 have the advantage of providing information on the base of clouds and/or precipitation down to 250 m which is much 456 more than the CloudSat-CPR can achieve (Fig. 7d). ACCP250's shorter pulse also helps mitigate the amount of cloud 457 stretching related to partial beam filling issues thus providing a more precise characterization of cloud top height (Fig.  458 8c, effects also visible in Fig. 6e). So generally speaking, reducing vertical pulse length reduces the fraction of detected 2) They are weakly reflective, with 50 % of the hydrometeors detected by KAZR having reflectivity below -22 483 dBZ. We also find that hydrometeor layer mean reflectivity is strongly related to hydrometeor layer thickness 484 such than the thinnest layers are also typically the least reflective ones, further challenging their detection. 485 3) They form at low levels, with 50% of WMBL cloud echoes being located below 1.2 km and 50 % of sub-486 cloud layer rain echoes below 0.75 km. Therefore, their backscattered power may easily overlap and be 487 masked by the strong surface return detected by spaceborne radars. [2018], our results suggest that a little over half of this bias can be attributed to the CloudSat-CPR inability to sample 492 thin, tenuous cloud while the other half results from misdetection of clouds that form entirely within the CloudSat-493 CPR surface (some of which are also thin and tenuous). Using forward simulations, we determined that mitigating the 494 vertical extent of the surface clutter by changing its range weighing function or by reducing its vertical range resolution 495 by half would only partially improve the CloudSat-CPR's ability to detect all cloudy columns, which is very much 496 limited by the CloudSat-CPR's low sensitivity. In other words, when it comes to detecting all cloudy columns, we 497 find that improving radar MDS is more important than reducing the vertical extent of the surface clutter. For this 498 reason, the 7 dB more sensitive EarthCARE-CPR is expected to detect significantly (19.7%) more cloudy columns 499 than the CloudSat-CPR, only missing < 9.0% of the simulated cloudy columns. 500 501 https://doi.org/10.5194/amt-2019-473 Preprint. Discussion started: 21 January 2020 c Author(s) 2020. CC BY 4.0 License.
On the other hand, our overpass and forward-simulation results also suggest that the CloudSat-CPR is able to capture 502 the general vertical distribution of hydrometeor (i.e., hydrometeor fraction profile) above 750 m which we find is 503 dominantly controlled by thicker more reflective clouds. Unfortunately, we estimate that because of its asymmetrical 504 range weighting function and because of the long length of his highly sensitive pulse, the EarthCARE-CPR's will 505 overestimate (by ~250 m) cloud top height and underestimate cloud base height, making hydrometeor layers appear 506 artificially thicker than they are, which will also bias the EarthCARE-CPR's hydrometeor fraction estimates. This 507 effect would need to be addressed to extract accurate information about the location of cloud boundaries and about 508 the vertical distribution of clouds and precipitation, two aspects likely to become increasingly important as we continue 509 moving towards increasingly high-resolution global modeling. Synergy with a collocated ceilometer could potentially 510 help correct cloud top height, however, such corrections would only be possible in single layer conditions and 511 alternative techniques would need to be developed to improve the EarthCARE-CPR's ability to accurately estimate 512 the vertical extent of multi-layer boundary layer clouds. 513 514 Below 1.0 km, the surface clutter echo seen by the CloudSat-CPR masks portions of clouds and virga. Based on a 515 subset of KAZR observations, we estimate that the surface echo limits the CloudSat-CPR's ability to observed true 516 cloud base in ~52% of the cloudy columns it detects and true virga base in ~80%. In other words, the CloudSat-CPR 517 often provides an incomplete view of even these cloud systems it does detect. Our analysis of real CloudSat-CPR's 518 observations shows that the clutter mask part of the GEOPROF version 4.0 product is relatively aggressive, and we 519 believe the CloudSat-CPR's performance could perhaps be somewhat improved by revising this clutter mask. In terms 520 of future spaceborne radar missions, radar architectures with finer range resolution could more precisely characterize 521 the boundaries of hydrometeor layers. For instance, the 250-m range resolution (oversampled at 125-m) radar 522 architecture presented here produces echo top height statistics comparable to that of the ground based KAZR in terms 523 of detecting the minimum, maximum and mode of the distributions. However, since a shorter pulse can currently only 524 be achieved at the expense of reduced sensitivity, this radar would suffer from the limitations similar to that of the 525 CloudSat-CPR in terms of the number of cloudy columns it could detect. This means that while improving the 526 detection of virga below 500 m might be possible, improving the detection of cloud bases below 500 m is unlikely 527 achievable with current technologies. 528

529
Overall this analysis suggests that no one single radar configuration can adequately detect all WMBL clouds while 530 simultaneously accurately determining the height of cloud top, cloud base and virga base. The alternative of deploying 531 spaceborne radars capable of operating with interlaced operation modes is thus worth considering [Kollias et al., 532 2007]. For example, a radar capable of generating both a highly sensitive long-pulse mode and a less sensitive but 533 clutter limiting short-pulse mode would likely provide a more comprehensive characterization of the boundary layer 534 by detecting both low-reflectivity clouds and low-altitude rain. 535 536 On a related note, it is likely that the partial beam filling issues identified here as affecting both the CloudSat-CPR 537 and the EarthCARE-CPR ability to locate clouds might, as hinted by Burns et al. [2016], also affect their ability to accurately measure their true reflectivity. Such radar reflectivity biases would affect water mass retrievals performed 539 using radar reflectivity measurement and follow up efforts should aim at quantifying this effect and should look into 540 alternative retrieval techniques and/or radar configurations that could address this issue [Battaglia et al.,In 541 preparation]. CPR_Echo_Top mask variable). We further ignore observations from non-significant echoes (Z < -27 dBZ) and 585 mispointing events (profiles, which have their maximum reflectivity more than 75 m from 0 m height). Over this 586 period, the median surface reflectivity profile (depicted by the broken black profile in Fig. 5c) shows a main peak at 587 surface level quickly reducing in intensity within height; the surface radar reflectivity return was observed to reduce 588 by ~34 dB at a distance of 0.5 km (i.e., half the pulse length) away from it actual location at the surface. A secondary 589 lobe whose peak intensity is ~50 dB lower than that of the main lobe was observed to spread from a distance of roughly    CloudSatnps which is CloudSat operating with the EarthCARE asymmetrical range weighting function (cyan dots), d) 783 CloudSatnps+es which additionally has an enhanced sensitivity equivalent to the EarthCARE (purple dots), c) 784 EarthCARE which additionally operates with a factor of 5 vertical oversampling (magenta dots), e) ACCP250 which 785 instead has a 250-m range resolution (red dots) and f) ACCP100 which instead has a 100-m range resolution (green 786 dots). For reference, the corresponding KAZR observed radar reflectivity are depicted in Fig. 1a