Analysis of long-term precipitation pattern over Antarctica derived from satellite-borne radar

Introduction Conclusions References


Introduction
The study and characterization of atmospheric phenomena in polar environments is a crucial aspect in solving the uncertainties of Earth's water cycle, energy budget, and climate studies.However, many limits affect the analyses of meteo-climatic parameters and of their variability at high latitudes (Lubin and Massom, 2007).Among these, precipitation is a key geophysical parameter with the highest spatial and temporal variability, and its measure and monitoring using either ground based measurements or satellite observations, is a challenging task over Antarctica, where mostly weak solid precipitation (snow and ice crystals) occurs (Bromwich, 1988).
The knowledge of precipitation characteristics over Antarctica relies on a sparse network of instruments based on different principles (snowpits, heated raingauges, snow stakes, acoustic sounders), measuring different properties of solid precipitation with different sensitivity and errors.The spatial interpolation of such sparse data is also difficult, with results depending on the data/instruments and techniques used (Arthern et al., 2006;Eisen et al., 2008).Moreover, ground based measurements are affected by severe experimental problems, such as the difficulty to measure relatively small snow deposits with an acceptable degree of accuracy and reliability (Eisen et al., 2008;Knuth et al., 2010), ensuring also effective instrument maintenance for the long time Introduction

Conclusions References
Tables Figures

Back Close
Full frames necessary to study precipitation characteristics.Recently, an experimental site devoted to the intensive study of clouds and precipitation microphysics has been established (Gorodetskaya et al., 2014).The use of numerical models, in general, shows discordant results, with large bias and relatively high errors if compared with ground measurements (Genthon and Krinner, 2001).Modeled precipitation fields, however, are still extremely valuable to assess annual trends for climatic studies (Bromwich et al., 2011).At finer scales, the polar version of the MM5 model (Guo et al., 2003), has been applied to infer spatial and temporal distribution of Antarctic precipitation (Bromwich et al., 2004), addressing also the role of blowing snow.Also the RACMO2 regional atmospheric climate model has recently updated its physical package giving some improvements on Surface Energy Balance, including precipitation (Van Wessem et al., 2014).
The detection from passive space-borne Low Earth Orbiting (LEO) sensors of light and solid precipitation is made difficult over Antarctica at any wavelength, due to the highly reflecting and emitting cold background and the relatively low ice content of the clouds.Several theoretical studies have been carried out to analyze the potential of passive microwave observations to detect and retrieve snowfall and to relate microphysical properties of iced hydrometeors to the observations (e.g., Skofronick-Jackson and Johnson, 2011;Munchak and Skofronick-Jackson, 2014).The use of high frequency channels (> 150 GHz) is promising, but it is complicated by their sensitivity to the highly variable microphysical characteristics of iced hydrometeors (shape, size and density) (e.g., Johnson et al., 2012).The use of passive microwave radiometry based on the use of water vapor absorption band channels (183 GHz) (e.g., Surussavadee and Staelin, 2009;Laviola and Levizzani, 2011), less sensitive to background signal, is a valuable choice at lower latitudes, but in Antarctica it is complicated by the extremely dry atmospheric conditions.Dedicated studies on the potential of these techniques for snowfall detection and/or retrieval in the polar regions are missing.Given the sensitivity of microwave radiation to the type of snow or ice at the ground, an alternative approach was attempted to evaluate the occurrence of snowfall by the changing radia-Introduction

Conclusions References
Tables Figures

Back Close
Full tive properties of the ground surface (Bindschadler et al., 2005) but without providing indication of the amount of snow accumulated.Great advances in the detection and retrieval of snowfall from space are foreseen with the launch of the Global Precipitation Measurement Mission core satellite (Hou et al., 2014), but only latitudes between 65 • S and 65 • N will be covered.
A new tool for clouds and precipitation remote sensing at high latitudes became available with the launch of the Cloud Profiling Radar (CPR) onboard the low earth orbit CloudSat satellite in 2006.The W-band, nadir pointing radar is designed to provide vertical profiles of reflectivity of the atmosphere, allowing a physical retrieval of the vertical structure of clouds and precipitation (Stephens et al., 2002;Liu, 2008), and, under some assumptions, the quantitative estimate of the snowfall rate near the ground (Hudak et al., 2008;Liu, 2008;Kulie and Bennartz, 2009).After the first operational year of CloudSat, some studies have been done over the Antarctic region and Liu (2008) and Hiley et al. (2011) showed the first remotely sensed precipitation maps, together with extensive discussion on the Antarctic snowfall estimate using CloudSat data.Since early 2013, the CloudSat Project releases a global product (2C-SNOW-PROFILE) where the "near surface" snowfall rate is reported for each CPR profile (Wood, 2013), and these data were used by Palerme et al. (2014) to compute a multi-year precipitation map of the Antarctica ice sheet.
Along this line, we used the snowfall rate retrieval algorithm from Kulie and Bennartz (2009) from raw CPR reflectivity data, tailored to the characteristics of the Antarctic precipitation and to account for the highly reflecting background surface, obtaining a precipitation product suited for Antarctica.Furthermore, we used the 2C-SNOW-PROFILE product to extend the Palerme et al. (2014) precipitation analysis over the ocean.We analyzed two years of data (2009 and 2010) and studied the spatial distribution of annual snow amount, also as function of cloud cover.The results were compared with ERA-Interim (Dee et al., 2011) reanalysis and with the data from 6 Acoustic Depth Gauges (ADG) available at the ground.ADG data are also compared to single snowfall events, showing consistency between the CPR snowfall retrievals and the ground-Introduction

Conclusions References
Tables Figures

Back Close
Full based measurements.Moreover, we evaluated the seasonal cycle of precipitation and the influence of the orography on snowfall rate.It has to be considered that a systematic comparison between the two snowfall products (the one obtained from Kulie and Bennartz (2009) and 2C-SNOW-PROFILE product) is not the primary focus of this study.The two different approaches for CloudSat precipitation estimates over Antarctica are shown to highlight their potentials, and possible explanations for differences between the two are discussed.As opposed to the 2C-SNOW-PROFILE product used in Palerm et al. (2014) where also wet snow and mixed phase precipitation is included, with Kulie and Bennartz (2009) algorithm we focus on the dry snow only.From the analysis of the results and from the comparison with ERA-Interim reanalysis it is evident that the impact of wet snow (and mixed phase precipitation) is significant over ocean.Depending on the application, the 2C-SNOW-PROFILE product can be used to study the climatology of the total precipitation, while the estimates from Kulie and Bennartz (2009) can be used to study the contribution to the precipitation due only to dry snow.Further studies dedicated to systematic comparisons between different snowfall retrieval algorithms are currently being undertaken.

CloudSat products and algorithms
The CloudSat is an experimental satellite equipped with a W-band (94 GHz) nadirpointing radar (CPR) designed to provide the observations necessary to advance the understanding of the role of clouds in climate (Stephens et al., 2008).CloudSat, launched in 2006, orbits in formation as part of the A-Train constellation of satellites and collects profiles of reflectivity factor of the lower 30 km of the atmosphere, with vertical resolution of 240 m and a footprint 1.7 km × 1.4 km.Radiation at CPR frequency can suffer from attenuation by gases (O 2 and H 2 O) and hydrometeors.Attenuation effects can be significant in environments with high water vapor amounts and under heav-Introduction

Conclusions References
Tables Figures

Back Close
Full ily precipitating conditions, but attenuation effects are generally limited in drier, colder environments like Antarctica.The CPR's minimum detectable signal (around −26 dBZ) allows it to be very sensitive to light precipitation (Haynes et al., 2009), making the CPR the most reliable orbiting instrument covering all latitudes, including the polar regions, to effectively detect and retrieve snowfall distribution and intensity (Liu, 2008), with an accuracy comparable with ground based observations (Matrosov et al., 2008;Hudak et al., 2008).Two years of CloudSat CPR data from January 2009 to December 2010, hereafter referred to as Observing Period (OP), are considered in this study.A significant gap, due to CPR battery problems, occurred from 7 December 2009 to 16 January 2010 reducing the number of satellite tracks to 9541 (about 560 tracks less than the expected number).We selected the area south of 60 • S. Figure 1 shows the map of the daily frequency of CloudSat overpasses, averaged over the OP, and sampled over 1 • latitude by 1.5 • longitude wide grid boxes.Over continental Antarctica the daily frequency ranges between 0.3 to 1.3 overpasses per day, while there is an average of 1 overpass every 3 to 10 days over oceanic regions.
The CPR data used in this work are the radar reflectivity factor vertical profiles, hereafter referred to as reflectivity (Z), as released in the 2B-GEOPROF product (Mace, 2007), where the complete data navigation and geolocation are also present.The CPR_echo_top data (also present in the 2B-GEOPROF product), providing information on the cloud top height and on cloud type (low, middle, high or multi layer), is also used to analyze cloud cover and properties related to snowfall episodes.A data quality flag is also delivered in 2B-GEOPROF and we considered only profiles with the highest quality.Together with 2B-GEOPROF product, surface snowfall rate retrievals from the 2C-SNOW-PROFILE product (Wood, 2013) are also used.It has to be mentioned that these products are "global", i.e. no particular adaptation to the Antarctic environment is applied to the products.In addition, the CloudSat ECMWF-AUX product is used, where the two-meter temperature is provided together with other ancillary ECMWF state variable data interpolated to each CloudSat CPR footprint.Introduction

Conclusions References
Tables Figures

Back Close
Full The algorithm here considered to retrieve snowfall properties over Antarctica is based on the technique developed by Kulie and Bennartz (2009) (hereafter referred as KB09) that makes use of CPR reflectivity profiles to select clouds with precipitation and then assigns a snowfall rate to the profiles.The algorithm first selects profiles with nearsurface reflectivity (Z) (i.e., the reflectivity of the 6th bin from the ground, around 1200 m above the surface) above −15 dBZ, and a continuous vertical layer of reflectivity exceeding −15 dBZ in at least 5 contiguous bins.After a comparison with ECMWF-AUX temperature field, profiles with 2 m temperature higher than 0 • C are screened out, in order to consider only profiles with potentially dry snow.The selected profiles are candidate snowfall profiles and the snowfall rate (S) is computed by applying a suitable Z-S relationship to the reflectivity of the near-surface bin.A more detailed description of the algorithm can be found in Kulie and Bennartz (2009).
A further study (Hiley et al., 2011) addressed the sensitivity of the algorithm to the choice of some parameters, such as the depth of the cloud vertical continuity, the reflectivity thresholds, and the parameters of the Z-S relationship, among others.
Several values of the a and b coefficients in the Z-S formula (in the form of Z = a S b , where Z is in mm 6 m −3 and S is the water equivalent snowfall rate in mm h −1 ) are proposed in the literature and a summary can be found in Hiley et al. (2011).Some experimental studies reported on the microphysical properties of precipitating ice particles finding average effective radius (area-weighted mean radius) around 24 µm (Walden et al., 2003) and recognizing the bulk of precipitation as due to columns and bullets and aggregates of them, with sizes varying at different experimental sites (Bromwich, 1988).Given the little information available on shape and size distribution of ice cloud and precipitation particles over Antarctica, we used a and b averaged values, computed as a best fit by considering 20 different Z-S relationships found in the literature, following Hiley et al. (2011).Two other pairs of coefficients are introduced in order to Introduction

Conclusions References
Tables Figures

Back Close
Full define the plus and minus one SD (σ) of the coefficients distribution.The a and b values used here are reported in Table 1.
The definition of a correct value of the near-surface bin, i.e. the lowest bin not affected by ground clutter signal, is particularly relevant over Antarctica for two reasons: (1) the coastal areas are very steep in Eastern Antarctica and thus ground clutter could affect higher bins in the CPR vertical profile; (2) very often snow could be drifting by strong low-level winds resulting in the local deposit of snow not related to snowfall (Li and Pomeroy, 1997).Following Kulie and Bennartz (2009), a further test is carried out on the reflectivity of the lowest CPR bins.In particular, we assumed the value of 20 dBZ as the theoretical upper limit for clutter-free reflectivity: when reflectivity at the 6th bin exceeds the threshold, it means that it is very likely affected by ground clutter, and the retrieval algorithm is applied starting at the 8th bin of the CPR, checking again for continuity of 5 vertical layers with reflectivity larger than −15 dBZ.For these profiles, the snowfall rate computed from the Z at the 8th bin is assigned.In Fig. 2 the Probability Density Function (PDF) of the reflectivity used for the retrieval (i.e., at the first nearsurface bin not affected by ground-clutter and used in the Z-S relationship) is shown (black line), while the red line shows the PDF of the 6th bin reflectivity exceeding the 20 dBZ threshold.The number of such contaminated bins is very low (around 10 −5 of the total), but screening them out is important because they could result in very high snowfall rate, able to impact on mean precipitation values (within the OP, the correction is just of the order of 0.37 % over the whole Antarctic region, below 60 • S, but it increases to 3.53 % if we consider only the grounded region below 2000 m where the ground clutter is expected to influence more the reflectivities values).

CloudSat 2C-SNOW-PROFILE product
The CloudSat snow profile product (2C-SNOW-PROFILE, hereafter 2C-SNOW) provides estimates of vertical profiles along with surface snowfall rate (Wood, 2013;Palerme et al., 2014).As KB09, this CloudSat product does not consider the lower bins close to the surface in order to avoid ground clutter effects.The 2C-SNOW prod-Introduction

Conclusions References
Tables Figures

Back Close
Full uct does not consider the bins below the 3rd bin from the surface if the profile is over ocean without sea ice, or over inland water, and below the 5th bin if the profile is over land, sea ice, or unknown surface.Also in this case a reflectivity threshold of −15 dBZ is applied to the near surface bin to separate snowing from not snowing profiles.Together with geolocation and snowfall rate variables, the product provides a retrieval status flag (snow_retrieval_status, SRS flag).This flag needs to be taken into account when using the snowfall rate data.The bit 3 of the SRS binary value is active when the snowfall rate at the base of the snow layer is substantially larger than that in the profile bin immediately above and this could be an indication of the effects of surface clutter, shallow precipitation, or partial melting of the snow.The developers suggest to use this profiles with caution and in cases where users conclude that the profile may be affected by ground clutter, the retrieval results for the bin above the near-surface bin may be used instead.
After a deeper investigation of the SRS flag behavior over Antarctica, we found that profiles with SRS bit 3 active are affected by ground clutter and are then corrected using the snowrate value of the bin immediately above the near surface bin.In Fig. 3 the PDF of the reflectivity used in the 2C-SNOW product retrieval is shown (black line), together with the PDF of profiles marked as affected by ground clutter (red line) from the 2C-SNOW product.As for the KB09 algorithm, the number of bins affected by ground clutter is not very high (also in this case only 10 −5 of total profiles within the OP), but enough to affect mean snowfall rate values (the annual mean snowfall rate decreases 1.3 % over the total Antarctic region, but more than 9.6 % if we consider only the grounded ice sheet below 2000 m where the ground clutter contamination is expected to be higher).
The optimal Z-S relationship is in this case dynamic and is calculated by minimizing a cost function which represents differences between simulated and observed reflectivities and also between estimated and a priori values for the snow microphysical properties (further details can be found in Wood, 2013).Introduction

Conclusions References
Tables Figures

Back Close
Full

Ground data: the Acoustic Depth Gauges
Long term ground based precipitation measurements over Antarctica rely on a sparse number of stations equipped with heated raingauges and/or Acoustic Depth Gauges (ADG), the latter measuring the distance from the snow covered surface to the instrument.Both instruments have strong limitations when a reliable measure of snowfall rate is needed (Eisen et al., 2008).For raingauge measurements in polar regions most of the uncertainties are related to wetting, evaporation losses, and strong winds (Sugiura et al., 2003): in particular winds induce undercatch, but also an overestimation bias due to blowing snow.The relative impact of these two effects depends on the type of windshield used.The measure of accumulated snow by ADG is influenced by many other factors than precipitation itself, since other different processes can change the height of the snow at the ground: blowing snow, deposition of hoar frost, surface sublimation, snowdrift sublimation, snow settling and compaction over time, meltwater (Knuth et al., 2010;Li and Pomeroy, 1997).
In this work we make use of the data from 6 ADG stations, whose location is shown in Fig. 4, managed by the University of Wisconsin-Madison Automatic Weather Station Program during the OP (Knuth et al., 2010).We processed the ADG 10 min data in order to evaluate the snow amount accumulated at the ground at two different time scales, monthly and at the single snowfall event time scale.The monthly distance variation between instrument and snow covered ground is computed by subtracting the ADG distance measurement at the end of the month by the one at the beginning of the month.The distance at the beginning (and at the end) of each month is computed as the mean distance over the first (last) five days of the month and the last (first) five days of the previous (following) month.After the screening of the outliers (i.e.very large distance variations over 10 min) we obtained the monthly distance variation which corresponds to a variation of the snow accumulation between the end and the beginning of each month of the OP: a positive ADG distance variation for one month indicates that a decrease of the accumulated snow occurred (negative snow accumulation variation,

Conclusions References
Tables Figures

Back Close
Full i.e., average snow accumulation at the end of the month less than average accumulation at the beginning of the month), while a negative ADG distance value is related to an increase of the accumulation (positive snow accumulation variation, i.e., average snow accumulation at the end of the month more than average accumulation at the beginning of the month).
CloudSat has overpasses within a 10 km radius of 5 of the 6 ADG stations, for which a comparison at the single snowfall event scale is possible.The distance variation at the time of a CPR overpass for a given station (hereafter, instantaneous distance variation) is computed as the difference between the average distance computed over the 5 h following the time of the overpass, and the average distance over the 5 h preceding the overpass.This relatively long time slot was necessary in order to have a minimum number of observations to compute the average, considering the high number of missing data and outliers.

Results
The analysis of solid precipitation over Antarctica and surrounding ocean is carried out at two different spatial and temporal resolutions.The CPR retrievals (both KB09 and 2C-SNOW) are analyzed at the CPR footprint scale, discussing statistical properties of snowrate at the highest resolution available.A parallel analysis is performed, sampling the CPR data on 1 • (in latitude) by 1.5 • (in longitude) wide grid boxes, where monthly and annual cumulated snow amounts are computed.

Analysis of annual gridded snowfall pattern
As first result, we computed the mean snowfall rate over the OP, expressed in mm of water equivalent per year, estimated in each 1 • × 1.5 • grid box: the sum of all instantaneous snowfall rate estimates (in mm h −1 ) divided by the total number of observations (i.e.number of CPR profiles) over the whole OP within each grid box.The mean snow-

Conclusions References
Tables Figures

Back Close
Full fall rate in mm h −1 is then reported in mm yr −1 .Figure 5a shows the result obtained using the AVE values for the coefficients a and b in the Z-S relationship for KB09 (see Table 1).The same map is shown for 2C-SNOW in Fig. 5b.In Fig. 5c the correspondent quantity as computed by the ECMWF ERA-Interim (ERA-I) reanalysis is shown, considering the original 12 hourly snowfall field data (provided in m of equivalent water), at 0.25 • × 0.25 • spatial resolution (Dee et al., 2011), interpolated onto the 1 • × 1.5 • grid, and summed over the entire OP.The total cumulated snowfall for the entire OP has been divided by 2 (years) to obtain the mean snowfall per year to be compared to the mean snowfall rate derived from CPR (all the three expressed in mm yr −1 ).We considered here ERA-I as a reference to assess the order of magnitude of our estimates as the only available spatially distributed snowfall fields over Antarctica.As for KB09 retrievals, the comparison indicates that the choice of the AVE values for the a and b coefficients is appropriate, while the comparison with estimated maps obtained by using AVE+σ and AVE−σ (not shown here) presents marked under-and over-estimation, respectively, with respect to the ERA-I map.
Comparing the estimate maps with the forecast reanalysis map, the first distinguishing feature is that the order of magnitude of the mean annual snowfall rate is very similar over the grounded ice sheet for both CloudSat-based methodologies.The KB09 estimates, however, exhibit a larger extension of the very dry area (snowfall rate below 50 mm yr −1 ) that approaches the coastlines.The 2C-SNOW estimates better describe the precipitation pattern showing, similarly to the ERA-I reanalysis, more precipitation along the coasts, especially along the western side of the Antarctic Peninsula.However, it is worth noting that both the satellite estimates locate a clear maximum over the western tip of the Peninsula (yellow areas), also shown in the reanalysis.Over the ocean the KB09 shows a precipitation pattern very similar to the ERA-I one, with some discrepancies in the spatial distribution and extension of local minima and maxima.A relative maxima is found in both maps over the eastern ocean, around 60 primarily due to the different approach used to determine mixed precipitation in ERA-I and in 2C-SNOW.As a matter of fact, if we compare the 2C-SNOW map and the "Total Precipitation" field of ERA-I the two become very similar (see Fig. 5d).On the other hand, it is worth noting that only the contribution of dry snow is considered in the KB09 product.Further investigation is necessary to assess the effective contribution of wet snowfall or mixed phase to the precipitation in the 2C-SNOW product and this will be the focus of future work.
These results indicate that the CPR observation, even subject to low spatial and temporal sampling, are able to reconstruct the main features of the mean annual snowfall patterns, due also to the expected low variability of snowfall fields.The patchy snowfall pattern over the oceans, where more homogeneous fields are expected given the lack of surface forcing, could be due to the poor sampling of the CPR (see Fig. 1).
The scatterplot between ERA-I and CPR retrieved mean annual snowfall, all remapped on the 1 • × 1.5 • grid are shown in Fig. 6.The KB09 estimate displays a fairly good correlation (0.89) for snowfall rate values below 400 mm yr −1 , with an underestimation of the CPR retrieved snowfall amount with respect to the ERA-I values.The correlation decreases for larger snowfall amounts, with a significant overestimation of the satellite values with respect to the ERA-I.The maximum value (around 1240 mm yr −1 ) is found by the satellite technique on the Antarctic Peninsula.Small areas with a marked overestimation are mainly over the ocean at lower latitudes (especially on the west side), where the CPR sampling is particularly poor (see Fig. 1).2C-SNOW estimates show a good global correlation due to a fairly good correlation for snowfall rate values below 400 mm yr −1 (0.88) and to an even acceptable correlation for larger snowfall values that however show a general overestimation with respect to ERA-I.Again maximum values are found over the Antarctic Peninsula.

Cloudiness influence on precipitation pattern
Cloudiness over Antarctica has been studied from both in situ and remote observations (Spinhirne et al., 2005;Lachlan-Cope, 2010;Bromwich et al., 2012), providing Introduction

Conclusions References
Tables Figures

Back Close
Full information on cloud horizontal and vertical structures, and cloud microphysical characteristics.In this work, we use the cloud top height information of the CPR_echo_top product to compute a cloud cover map for the OP. Figure 7a shows the percentage of profiles classified as cloudy by the CPR_echo_top product (hereafter "cloud" profiles) with respect to the total number of profiles in each grid box.The cloud cover is highly correlated with topography: it is higher over the ocean and low elevation land, while is low over the Plateau.Over the ocean cloud cover patterns are more related to atmospheric circulation with a significant dependence on longitude, with wide areas, such as the Weddell Sea (around 30-50 • W) and Ross Sea (around 170 • W), showing lower cloud cover.Figure 7b and c show the precipitating cloud fraction, i.e., the fraction of cloudy profiles with snowfall rate larger than a minimum threshold of 0.1 mm h −1 (hereafter "snow" profiles), as retrieved from KB09 and from 2C-SNOW respectively.For each grid box the number of "snow" profiles is divided by the number of "cloud" profiles.Even with different percentage of "snow" profiles (due to the different constraints applied for the retrieval) the effect of topography on the cloud efficiency in producing snowfall is evident: the Plateau inhibits cloud formation and the production of significant snowfall, while the red spot on the Peninsula coastline is consistent with the maximum in the mean annual snowfall shown in Fig. 5a and b, and shows a higher potential of having precipitation in that region.The lack of precipitating clouds over the peripheral ocean area for the KB09 algorithm estimate is mainly due to the temperature restrictions applied to the algorithm and the exclusion of wet snow or mixed precipitation from the dataset.

Qualitative comparison of monthly ground based and CPR snowfall values
A comparison between satellite estimated monthly mean snowfall rate (computed in the same way as the annual mean snowfall rate, but on a monthly basis) and the measurements from the 6 ADG is performed in order to provide a qualitative measure of the accuracy of the CPR ability to detect snowfall.Introduction

Conclusions References
Tables Figures

Back Close
Full The monthly distance variation from ADG, computed as described in Sect.2.2, is compared to the mean snowfall rate (mm month −1 ) in each grid box containing the ADG stations, and results are reported in Fig. 8 for each month of the OP.
Results from KB09 and from 2C-SNOW estimates present high qualitative similarities.The large majority of the months with snowfall (as derived from CPR) have a negative monthly distance variation measured by the ADG (positive snow accumulation variation, i.e., average accumulation at the end of the month more than average accumulation at the beginning of the month), especially for snowfall rates higher than 25 mm month −1 .Many months present significant negative distance variation values in correspondence with moderate or low mean snowfall rates, and here it is expected a significant contribution of blowing snow episodes (Li and Pomeroy, 1997;Knuth et al., 2010).Points with positive distance variation (i.e., negative snow accumulation variation, i.e., average accumulation at the end of the month less than average accumulation at the beginning of the month) are mainly related to low or negligible estimated mean snowfall rates.A significant exception is found for the maximum snowfall rate (around 180 mm month −1 as estimated by KB09 and around 320 mm month −1 as estimated by 2C-SNOW), corresponding to a positive distance variation.These values are registered for November 2010, a warm month, when ablation and snow compaction are likely to occur (Radok and Lilie, 1977;Knuth et al., 2010).
In the validation of satellite precipitation retrievals (i.e., Puca et al., 2014), often a contingency table is constructed by matching estimates with ground reference measurements, where the number of hits (when estimates and observation both detect precipitation), misses (when only observations measure precipitation), false alarms (when only the estimate detects precipitation), and correct negatives (when both do not detect precipitation) are reported.In this case, for the reasons explained in Sect.2.2, we can not assume the distance variation as "true" representation of the precipitated snow amount.However, an inspection of the resulting contingency tables (reported in Tables 2 and 3) could help in evaluating the overall matching between the two compared fields.Introduction

Conclusions References
Tables Figures

Back Close
Full We define a hit when a month with a negative distance variation (increase of the snow accumulation) corresponds to a monthly mean snowfall rate above zero, a miss when it corresponds to estimated snowfall equal to zero, a false alarm when an estimated snowfall above zero corresponds to a distance variation below or equal to zero (decrease or no variation of the snow accumulation), and the correct negatives accordingly.
From the total number of 140 couples of monthly estimates and observations (computed in the 6 grid boxes containing ADG stations for each month, with the exception of four months during the OP when ground based data are missing), few indicators can be computed to summarize the results.For both estimate algorithms, most of the grid boxes where the CPR estimate detects precipitation corresponds to an increase of the snow accumulation (78 out of 114, i.e., 68 % as estimated by KB09 and 89 out of 132, i.e. 67 % as estimated by 2C-SNOW), while 86 % (78 out of 91) of grid boxes with an increase in the snow accumulation have an estimated snowfall amount above zero as estimated by KB09 and 98 % (89 out of 91) as estimated by 2C-SNOW.The 13 KB09 and the 2 2C-SNOW cases where the increase of snow accumulation does not correspond to estimated snowfall amount above zero could be due to blowing snow episodes, carrying snow from other sites.The main error is associated to the 36 (KB09) and to the 43 cases (2C-SNOW) where snowfall amount above zero is estimated, but no increase of the snow accumulation is measured at the ground: again blowing snow (in this case removing the snow) could be an important factor, but as mentioned, many other factors contribute to the decrease (or no variation) of the snow accumulation between the end and the beginning of the month, and these mechanisms may be active in the same months where snowfall occurs.The larger occurrence of this cases for the 2C-SNOW (43) with respect to KB09 (36) could be due to the absence of a well defined temperature threshold for the 2C-SNOW snowfall estimates, including cases of mixed precipitation that are not well or not at all measured by ADG instruments.
To study the relationship between the snow accumulation as measured by the ADG instruments and the estimated snowfall rate from CPR, we performed the compari-

Conclusions References
Tables Figures

Back Close
Full son also at the single event temporal scale, and the results are presented in the next Section.

Analysis of instantaneous snowfall rates at CPR profile spatial scale
A comparison between instantaneous snowfall rate estimated from CPR and snow accumulation variation at the time of the CPR overpass as measured by the 5 ADG close to the CloudSat overpasses (station n. 08915 is out of the CPR coverage) has been carried out.A total of 195 CloudSat overpasses occurred at a distance lower than 10 km from the considered stations, and 19 events with snowfall retrieved from CPR data were collected.The distance variation measured by the ADG stations around the CPR overpass is estimated as described in Sect.2.2 and the results are summarized in Fig. 9.
As found for the monthly values, instantaneous snowfall detected by the CPR (for both algorithms) corresponds to negative variations of the distance difference for most of the cases (increase of the snow accumulation).The relatively small number of misses, i.e. cases when the satellite does not detect snowfall but the snow accumulation increases, can be attributed to blowing snow episodes.Most of the significant snowfall episodes are related to negative distance differences, with one exception (station n. 30374) when relatively high snowfall rate corresponds to slightly positive distance variation.A closer inspection of the corresponding CPR profiles (not shown) shows a marked tilting of the snow column just above the ADG, indicating a likely occurrence of strong low level winds, able to weaken the local relationship between measurements at the ground from the ADG and the retrieval from CPR from the atmospheric layers aloft.
From the analysis of annual snowfall maps it is evident a clear signal of the influence of the topography on the precipitation patterns (see Fig. 5a and b and Bromwich, 1988).
For a deeper discussion of the impact of topography on snow spatial and temporal distribution we divided the domain in three different background surface classes: (1) ocean and sea ice, (2) coastal areas and lowlands and (3) plateau (hereafter ocean, land, and Introduction

Conclusions References
Tables Figures

Back Close
Full plateau, respectively), according to the height above the sea level.The separation is based on the CPR elevation flag, choosing 2000 m a.s.l. as threshold to separate between land and plateau classes.The normalized PDF of instantaneous snowfall rate, as estimated by KB09 and by 2C-SNOW, is shown in Fig. 10 for the three classes, with steps of 0.1 mm h −1 .The normalization is made with respect to the total number of "snow" profiles, i.e., profiles with snowfall rate larger than 0.1 mm h −1 .The snowfall rate over the plateau is very low (more than 99 % of the "snow" profiles has values below 1 mm h −1 ), but it has to be remarked that in this area snow accumulation is mainly due to the precipitation of small crystals in clear sky conditions (Bromwich, 1988;Knuth et al., 2010), likely associated to very low rates, not detected by CPR.Ocean and land profiles have similar behavior, with maximum snowfall rate reaching the same value of 3.5 mm h −1 as estimated by KB09, but with a relative abundance of higher snowrates over land (above 1.4 mm h −1 ) with respect to ocean profiles, indicating the effect of the coastline and low elevation land on the enhancement of precipitation intensity.The 2C-SNOW PDFs show a behavior similar to the KB09 PDFs but with higher snowfall rate values exceeding 5 mm h −1 and with the abundance of land profiles with respect to ocean profiles starting around 2 mm h −1 .As estimated by KB09, the mean snowrates are 3.6 × 10 −2 mm h −1 for ocean, 1.7 × 10 −2 mm h −1 for land, and 2.0 × 10 −3 mm h −1 for the plateau, while the values become 6.2 × 10 −2 mm h −1 for ocean, 3.3 × 10 −2 mm h −1 for land, and 4.6 × 10 −3 mm h −1 for the plateau as estimated by 2C-SNOW product.
A further analysis on CPR snowrate at single profile scale was attempted to highlight seasonal signals in the snowfall rate distribution: we computed, for each of the three surface type classes (land, ocean, and plateau) the average snowfall rate for three-month periods (MAM, JJA, SON and DJF) of the OP and the results are reported in Fig. 11.We used here the common months grouping used to define seasons (as in Bromwich, 1988), considering though that the number of daylight hours vary substantially across the months, and thus the signal due to the change of daily insolation is embedded in the seasonal cycle.Over ocean and land both snowfall products have the highest average snowfall rate in the austral autumn (MAM), in agreement with Figures

Back Close
Full  Bromwich (1988), while the driest season is the summer (DJF).For the plateau region, the behavior is different: higher averaged snowfall rate is estimated during DJF and lower in SON.
A further investigation was focused on the estimation of the spatial extension of precipitation pattern over Antarctica.The use of point-like ground measurements allows the estimation of the duration of snowfall events (Knuth et al., 2010), but no information can be derived on the spatial extension of snowfall patterns, given the sparse distribution of ground instruments.On the other side, the CPR does not provide information about temporal features, but, from the analysis of the lengths of the track segments with snowfall, it is possible to infer the 1-D horizontal extension of the snowfall events.
Within the dataset of CloudSat orbits with detected snowfall the 1-D horizontal extension (hereafter "length") of the event was defined as the number of contiguous CPR footprints with snowfall rate above 0.1 mm h −1 .Only precipitation patterns that started and ended within the same orbit are considered.The lengths are grouped into three classes: class 1 (below 10 km), class 2 (between 10 and 100 km), and class 3 (beyond 100 km) and the number of lengths in each class are normalized with respect to the total number of lengths for each of the four seasons previously defined.The resulting histograms are reported in Fig. 12.Of course, the different algorithm used to estimate the precipitation influences the length of snowfall patterns, but the overall behavior shows common results.The austral winter (JJA) reports the highest percentage of the longest snowfall patterns (average length around 53 km for KB09 and around 46 km for 2C-SNOW).This results can be associated to the findings of Knuth et al. (2010) who found in JJA the highest percentage of long lasting events (duration longer than 1 day), to establish that JJA is the season when snowfall events with longer duration and larger spatial extension are more likely to occur.Shorter lengths (< 10 km) show a slightly different behavior between the two algorithms: higher percentages are found in DJF for KB09 (average length around 36 km) and in MAM for 2C-SNOW (average length around 39 km, but very close to the value for DJF).Finally, the overall higher percentage of longer snowfall patterns (10-100 km) relative to shorter patterns (< 10 km) Introduction

Conclusions References
Tables Figures

Back Close
Full for KB09 with respect to 2C-SNOW can be explained by the different methodology used to define the near surface bin reflectivity in the two algorithms: the higher near surface bins used in KB09 may cause missed shallow precipitation events, mainly over the ocean, leading to a loss of shorter precipitation pattern in favour of longer patterns (10-100 km).

Summary and conclusions
Two different approaches for snowfall rate estimate from the CloudSat W-band spaceborne Cloud Profiling Radar reflectivity profiles were applied to the Antarctic region to study solid precipitation characteristics at different spatial and temporal scales.The algorithm of Kulie and Bennartz (2009) has been used to convert CPR reflectivity profiles for the years 2009 and 2010 to instantaneous water equivalent snowfall rates.The algorithm has been tailored to the characteristics of precipitation in Antarctica.In addition, the official CloudSat 2C-SNOW-PROFILE product with different characteristics and based on different assumptions has also been considered.For both products the significant impact of ground clutter contamination in this region has been taken into account.
The instantaneous, footprint scale, retrieved snowfall rates are first up-scaled and integrated in time to obtain spatially continuous mean annual snowfall rate maps, to be compared with ERA-I reanalysis of snowfall.The KB09 map shows general agreement with the ERA-I mean annual snowfall rate map, both in terms of numerical values and spatial distribution, while the 2C-SNOW shows good agreement with the ERA-I mean annual rate map of the total precipitation.Because of the similar behavior of 2C-SNOW and ERA-I over the grounded continent (where the surface temperature is always below 0 • C), we can assert that the different amount of precipitation over the ocean is mainly due to the different approach used to determine mixed precipitation in ERA-I and in 2C-SNOW.The relation between cloud cover and snowfall frequency highlights the impact of coastal orography in enhancing the snowfall occurrence.A comparison Introduction

Conclusions References
Tables Figures

Back Close
Full of the mean monthly snowfall rate with ground ADG data indicates a general correspondence between the estimated occurrence of snowfall with the mean monthly snow accumulation variation increase, also considering the different mechanisms, other than precipitation, that influence the snow accumulation process (Knuth et al., 2010).
A deeper analysis at finer scale (considering instantaneous, footprint-scale CPR estimates) shows that for KB09, higher snowfall rates (always above 1.5 mm h −1 ) are more frequent over coast and lowlands, with respect to surrounding ocean, while on the plateau the snowfall rate never exceeds 1.8 mm h −1 .2C-SNOW shows higher snowfall rates with respect to KB09, also over the plateau, with maximum values exceeding 5 mm h −1 .A comparison between estimated instantaneous snowfall rate and snow accumulation variation at the ground (as measured by ADG) during the CPR overpass was attempted for 19 cases.Results show that significant snowfall episodes are very often related with an increase of snow accumulation, while the opposite is not true likely due to the contribution of blowing snow occurrence.This comparison is affected by several shortcomings, related to the different processes affecting snow accumulation variation and not related to precipitation (Gorodetskaya et al., 2014).The use of a larger dataset would allow to include other variables (such as wind speed, temperature, relative humidity) in the analysis, to study more in detail the involved processes, and to provide more accurate validation results.Finally, the influence of orography on the precipitation intensity, in relation to the seasonal variation, was also analyzed, showing that mean snowfall rate is higher in MAM over ocean and land, while on the Plateau higher mean snowfall rate is found in DJF.The results showed also that the seasonal cycle affects the spatial extension of snowfall pattern: relatively larger snowfall fields are found in JJA, while shorter paths are found in DJF and MAM.
This study is not intended for suggesting the best algorithm for precipitation estimation over the Antarctic region, but for evaluating the potentials of CPR for snowfall retrieval in Antarctica independently of the approach used, relating the differences in Introduction

Conclusions References
Tables Figures

Back Close
Full the results to the different assumptions and methodologies (i.e., temperature threshold used, height of near surface bin, and to the Z-S relationship).This work shows that, in spite of its limited temporal and spatial sampling capabilities, CPR can be used as a valuable source of snowfall rate data in Antarctica that can be analyzed at different temporal scales, providing support to the sparse network of ground-based instruments in this region, as well as to numerical models used to simulate single snowfall events (at high temporal and spatial resolution), or used for climatological studies.CPR snowfall products tailored to the characteristic of the Antarctic region could then be used in the analysis and interpretation of passive measurements from the large number of radiometers on board polar satellites orbiting around the globe providing full spatial coverage of the whole region.Introduction

Conclusions References
Tables Figures

Back Close
Full  Full Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | • E and between 110 and 150 • E. On the other hand, 2C-SNOW, over the ocean, shows a very high precipitation amount if compared to ERA-I values.This behavior could be Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Table 1 .
Values of the Z = a S b parameters used in this work (Z is in mm 6 m −3 and S is the water equivalent snowfall rate in mm h −1 ).