The retrieval of snow properties from SLSTR Sentinel-3 – Part 2: Results and validation

To evaluate the performance of the eXtensible Bremen Aerosol/cloud and surfacE parameters Retrieval (XBAER) algorithm, presented in the Part 1 companion paper to this paper, we apply the XBAER algorithm to the Sea and Land Surface Temperature Radiometer (SLSTR) instrument on board Sentinel-3. Snow properties – snow grain size (SGS), snow particle shape (SPS) and specific surface area (SSA) – are derived under cloud-free conditions. XBAERderived snow properties are compared to other existing satellite products and validated by ground-based and aircraft measurements. The atmospheric correction is performed on SLSTR for cloud-free scenarios using Modern-Era Retrospective Analysis for Research and Applications (MERRA) aerosol optical thickness (AOT) and the aerosol typing strategy according to the standard XBAER algorithm. The optimal SGS and SPS are estimated iteratively utilizing a lookup-table (LUT) approach, minimizing the difference between SLSTR-observed and SCIATRAN-simulated surface directional reflectances at 0.55 and 1.6 μm. The SSA is derived for a retrieved SGS and SPS pair. XBAER-derived SGS, SPS and SSA have been validated using in situ measurements from the recent campaign SnowEx17 during February 2017. The comparison shows a relative difference between the XBAER-derived SGS and SnowEx17-measured SGS of less than 4 %. The difference between the XBAER-derived SSA and SnowEx17-measured SSA is 2.7 m2/kg. XBAERderived SPS can be reasonably explained by the SnowEx17observed snow particle shapes. Intensive validation shows that (1) for SGS and SSA, XBAER-derived results show high correlation with field-based measurements, with correlation coefficients higher than 0.85. The root mean square errors (RMSEs) of SGS and SSA are around 12 μm and 6 m2/kg. (2) For SPS, aggregate SPS retrieved by XBAER algorithm is likely to be matched with rounded grains while single SPS in XBAER is possibly linked to faceted crystals. The comparison with aircraft measurements, during the Polar Airborne Measurements and Arctic Regional Climate Model Simulation Project (PAMARCMiP) campaign held in March 2018, also shows good agreement (with R = 0.82 and R = 0.81 for SGS and SSA, respectively). XBAER-derived SGS and SSA reveal the variability in the aircraft track of the PAMARCMiP campaign. The comparison between XBAERderived SGS results and the Moderate Resolution Imaging Spectroradiometer (MODIS) Snow-Covered Area and Grain size (MODSCAG) product over Greenland shows similar spatial distributions. The geographic distribution of XBAERderived SPS over Greenland and the whole Arctic can be reasonably explained by campaign-based and laboratory investigations, indicating a reasonable retrieval accuracy of the retrieved SPS. The geographic variabilities in XBAER-derived SGS and SSA both over Greenland and Arctic-wide agree with the snow metamorphism process.


42
Change of snow properties is both a consequence and a driver of climate change (Barnett et al., 43 2005). Snow cover and snow season, especially in Northern Hemisphere, are reported by 44 different models, to decrease due to climate change (Liston and Hiemstra, 2011). The reduction 45 of snow cover leads to the change of surface energy budget (Cohen and Rind, 1991     homogeneity. The TOA reflectance at 0.412 μm, the ratio of TOA reflectance at 0.76 and 0.753 256 μm, standard deviation of TOA reflectance at 0.412 μm are used to characterize cloud 257 brightness, cloud height, and cloud homogeneity, respectively. A pixel is identified as a cloud-258 free snow pixel when both SLSTR and OLCI identify it as a cloud-free snow pixel. Identified 259 clouds can be surrounded by a so-called "twilight zone" (Koren et al., 2007), which can extend 260 more than ten kilometers from a cloud pixel to a cloud-free area. The surrounding 5×5 pixels 261 of an identified cloud pixel will be marked as a cloud to avoid the "twilight zone" effect. A 262 more details description of this cloud screening method can be found in Mei et al. (2020a). 263 264

Atmospheric correction 265
Due to the low atmospheric aerosol loading over the Arctic snow covered regions (e.g. 266 Greenland), atmospheric correction using path radiance representation (Chandrasekhar, 1950;267 Kaufman et al., 1997) can provide accurate estimation of surface reflection even under 268 relatively large SZA (Lyapustin, 1999). The TOA reflectance at selected channels (0.55 and 269 1.6 μm) is described by the path radiance representation (Chandrasekhar, 1950;Kaufman et al.,   The atmospheric correction is performed based on the following equation: The atmospheric correction is based on the Look-Up- Table (  includes pre-processing of cloud screening using the synergy of OLCI and SLSTR and the 301 atmospheric correction using MERRA providing AOT and weakly absorbing aerosol type. The 302 SGS and SPS are obtained using the LUT-based minimization routine. SSA is then calculated 303 using the retrieved SGS and SPS. 304 https://doi.org/10.5194/tc-2020-270 Preprint.   and found droxtal is a reasonable assumption to take ice particle non-sphericity into account. 347 The above conclusions can be used as "qualitative reference" to understand the satellite-derived

371
In this section, we will quantitatively compare/validate XBAER derived snow properties with 372 ground-based/aircraft measurements. We emphasize that the results presented in this section is 373 considered as preliminary and the validation of our satellite-derived snow products will be 374 complemented by the MOSAiC expedition, which is now ongoing. 375

Comparison with the observations of SnowEx17 campaign 376
The above analysis shows that the XBAER is capable to derive SGS, SPS, and SSA, which 377 agrees reasonably well with existing satellite products or can be qualitatively explained by 378 campaign-based and laboratory findings. In order to have a quantitative evaluation of XBAER-379 derived SGS, SPS, and SSA, we have collocated the SLSTR observations with recent campaign 380 measurements provided by SnowEx17, as described in section 2. Due to overpass time and 381 cloud cover, only limited match-ups between XBAER retrievals and SnowEx17 measurements 382 have been obtained. 383 Table 3 summarizes match-up information. The first three columns in Table 3 show the 384 observation time and locations (longitude and latitude). The fourth and fifth columns indicate 385 the cloud conditions. Cloud conditions in Table 3 are given by three categories: cloud-free snow, 386 cloud-contaminated snow, and cloud-covered snow. These three categories are classified by the 387 XBAER cloud identification results and are illustrated by the RGB composition figures, 388 covering the SnowEx campaign area, as presented in Fig. 4. An optically thin cloud over a 389 melting snow layer, a thick cloud over snow, and snow scenarios are presented in Fig. 4 (a) Table 4 summarizes the comparison between XBAER retrieval results and SnowEx17 429 campaign measurements. The first three columns in Table 4 are the same as Table 3, Table 5, reported by campaign 438 participators), the warmer condition leads to a quicker snow metamorphism process, forming 439 large ice crystal particles. 440 An underestimation is found for the first match-up on the 9 th of February. This is explained 441 by the cirrus cloud contamination as presented in Fig. 4 and 5. According to our independent 442 XBAER cloud retrieval (Mei et al., 2018), the COT is ~0.5, cloud contamination with COT=0.5 443 introduces ~30% underestimation according to Fig. 11 in part 1 of the companion paper. So for 444 SGS=100 μm, provided by SnowEx, XBAER is expected to have a theoretically retrieved SGS 445 of ~ 70 μm while a value of 78.2 μm is obtained from the real satellite retrieval. In order to 446 further confirm this negative bias feature caused by cloud contamination, 11 th February (a 447 snowstorm at the measurement site is reported by campaign participators), although filtered by 448 the XBAER cloud screening routine, is forced to retrieve the full-cloud-covered scenario as a 449 cloud-free case. According to the theoretical investigations presented in part 1 of the companion 450 paper, for COT≥5, the XBAER algorithm retrieves cloud effective radius, rather than SGS. 451 The retrieved ice crystal size depends on the cloud effective radius of the cloud above the 452 underlying snow layer. The independent XBAER cloud retrieval provides SGS value of ~ 38 453 μm while 32.3 μm is obtained by the XBAER snow retrieval, for a reference value of 100 μm 454 as provided by SnowEx17 measurement. This is consistent with a typical ice cloud effective   Table 5 shows the same match-up information as in Table 4, but for SPS. We would like 461 to highlight again, the SPSs proposed by Yang et al (2013) are used for the radiative transfer 462 calculation. From a single ice crystal point of view, those shapes are very unlikely to occur 463 exactly in reality. This is similar to the issue in field measurements. Spherical shape assumption 464 is widely used (e.g., the measurement of SSA), however, a pure spherical shape is also very 465 unlikely to occur in natural snow. To have a reasonable comparison between satellite-derived 466 SPS and field-measured SPS, the quantitative information of "roundish" or "irregular" shapes 467 from both satellite and field measurement communities may be an option. Under this 468 comparison strategy, a "droxtal" shape derived from satellite observation is somehow identical 469 with a "spherical shape" in field measurement. 470 The second and third column in Table 5 Table 6 shows the comparison of SSA. For the three cloud-free samples, the difference of 492 XBAER-derived SSA and SnowEx17 measured SSA is 2.7 m 2 /kg, which is significantly 493 smaller than what has been reported by previous publications. For instance, the differences 494 between satellite retrievals and field measurements are reported to be 9 m 2 /kg and ~6 m 2 /kg as 495 for these two samples, which is due to the different SPSs. SnowEx shows that the SPSs are new 500 snow and facets for these two samples, respectively. XBAER derived SSAs are 24.5 and 12.9 501 m 2 /kg, which agrees well with SnowEx measurement. Since both SnowEx and XBAER provide 502 very similar SGS (250 μm vs 254.4 μm), the agreement of SSA indicates that XBAER derived 503 "aggregate of 8 columns" is comparable to "new snow" while XBAER derived "droxtal" is 504 somehow "identical" to "facets" in SnowEx. Cloud contamination introduces an overestimation 505 of SSA, especially for 11th February. According to the investigation from the companion paper, 506 for reference SSAs of 37.3 and 25.9 m 2 /kg, SSA is expected to be ~ 65 m 2 /kg and >100 m 2 /kg 507 for cloud contamination with COT ~ 0.5 and 10, respectively. The real satellite retrieval values 508 are 56.5 and 136.8 m 2 /kg, respectively.

514
The above validation for the retrieval of SGS, SPS, and SSA using the XBAER algorithm, 515 although with limited samples, indicate the consistent of the sensitivity study from the 516 companion paper in part 1 and the retrieval results in part 2, as presented in this section. 517 518

519
The optical snow grain size over Arctic sea ice was derived from airborne SMART 520 measurements as described in Sect. 2.3. Fig. 6 (a) Fig. 6 (a)) SGS-values of up to 350 µm were derived from the aircraft albedo measurements. 527 Also the XBAER algorithm reveals higher values in this region. For a direct comparison 528 XBAER data were allocated to the time series of the SMART measurements along the flight 529 track. Afterwards all successive SMART data points assigned to the same XBAER location 530 were averaged to compile a joint time series of both data sets as displayed in Fig. 6 (b). Overall Note, that the flight altitude varied for the flight section shown in Fig. 6 (a). Due to the low sun, 538 such a non-smooth surface produces a significant fraction of shadows which lowers the 539 measured albedo. Consequently, the retrieved SGS is affected in particular for the lowest flight Eq. (A1) from companion paper. The map of the SSA (Fig. 6 (c)) reflects a similar pattern than 546 observed for the SGS, showing an inverse behavior to Fig. 6 (a)  simultaneously. This is the first paper, to our best knowledge, attempting to retrieve both SGS, 617 SPS and SSA using passive remote sensing observations. 618 The new algorithm is designed within the framework of XBAER algorithm. The XBAER 619 algorithm has been applied to derive SGS, SPS and SSA using the newly launched SLSTR geometries and snow properties. The retrieval is relatively time-consuming because the 643 minimization has to be performed for each ice crystal shape and the optimal SGS and SPS are 644 selected after the 9 minimization are done. The SSA is then calculated using the retrieved SGS 645 and SPS based on another pre-calculated LUT. 646 The comparison between XBAER derived SGS, SPS and SSA show good agreement with 647 the SnowEx17 campaign measurements. The average absolute and relative difference between 648 XBAER derived SGS and SnowEx17 measured SGS is about 10 μm and 4%, respectively. Although the presented version of the XBAER retrieval algorithm shows promising results, 666 we see at least three possibilities to improve its accuracy. An intensive validation with the 667 support of MOSAiC campaign is needed. Currently only single ice crystal shape is used in the 668 retrieval, the mixture of different ice crystal shapes i.e., the snow grain habit mixture model 669 (e.g., Saito et al. 2019) will be tested in further work. Another potential improvement may be 670 linked to the usage of polydisperse ice crystals (e.g. gamma distribution). The potential impacts 671 of the vertical structure of SGS and SPS also need to be investigated in the future.