The Arctic has changed dramatically in recent decades, and causes of these changes and their feedbacks to the global climate system are under intense investigation. The Arctic is warming at a higher rate than that of the global average, a phenomenon known as Arctic amplification (Solomon et al., 2007; Serreze and Francis, 2006); Arctic sea ice extent has been decreasing dramatically (Serreze et al., 2015), and this trend is expected to continue (Holland and Bitz, 2003; Overland and Wang, 2013). Changes in the Arctic have likely led to changes in the weather and climate in the midlatitudes through teleconnections in the large-scale circulation (Francis and Vavrus, 2012). By studying the factors influencing the Arctic climate system and its changes, we will improve understanding of the Arctic climate and its relationship to the global climate system. The largest uncertainty in predicting the Arctic climate arises from our lack of understanding of the role clouds play in the Arctic climate system (Solomon et al., 2007; Boucher et al., 2013). A complete, accurate description of three-dimensional cloud properties is critical to determine the radiation flux both at the surface and at the top of atmosphere (TOA), as well as the radiative heating rate in the atmosphere. Examining and understanding changes in these vertical distributions are key to studying the recent Arctic changes.

Cloud products from space-based combined radar–lidar observations have the potential to provide comprehensive information on the vertical distribution of cloud properties. These observations have been used to describe global cloud spatial distributions and their temporal changes (Li et al., 2015; Naud et al., 2015). However, space-based low-cloud observations are limited by radar ground clutter and strong attenuation of lidar signals, especially by liquid and mixed-phase clouds (Marchand et al., 2008; Blanchard et al., 2014). Radar reflectivity from CloudSat has been used to generate high-vertical-resolution longwave and shortwave radiative flux profiles and corresponding heating rates (L'Ecuyer et al., 2008); assessing the product's accuracy shows that CloudSat's weakness in detecting low clouds introduces the largest uncertainty. This product has been improved by the inclusion of complementary cloud and aerosol information mainly from space-based lidar observations (Henderson et al., 2013). Complementing the space-based observations, surface observations have superior performance near the surface (Shupe et al., 2011; Shupe, 2011; Zhao and Wang, 2010) and in resolving the diurnal cycle at a specific location, with a relatively weaker performance in the middle and upper levels.

Efforts have been made to investigate the differences in cloud fraction/frequency from surface-based and space-based radar–lidar combined observations and their impact on the radiative fluxes at multiple surface stations. Using such observations, Protat et al. (2014) studied the cloud occurrence frequency around Darwin, Australia, and found that space-based observations underestimated the cloud occurrence frequency below 2 km above mean sea level (a.m.s.l.; hereafter, all heights are in km a.m.s.l.), while surface observations do not detect most of the cirrus clouds above 10 km. Blanchard et al. (2014) investigated the difference in cloud fraction and vertical distribution at Eureka, Canada, in the Arctic from surface and space-based combined radar–lidar observations from 2006 to 2010. Among many valuable findings, they found that space-based radar–lidar measurements can depict a complete picture of the cloud vertical profile down to 2 km. Mioche et al. (2015) compared vertical profiles of cloud occurrences from surface lidar and space-based lidar, radar, and combined lidar and radar over the Ny-Ålesund station during March and April 2007, and showed similar results above 2 km as those in Blanchard et al. (2014). The strengths and limitations of these observations are also discussed in other papers, e.g., Kay et al. (2008), Kay and Gettleman (2009), and Huang et al. (2012).

This study focuses on further examining and comparing the performance of space-based and surface-based radar–lidar observations and retrievals to capture the vertical distribution of cloud properties, including cloud fraction, cloud phase, and cloud water content, at two Arctic atmospheric observatories, Barrow, Alaska (USA), and Eureka, Canada. Since cloud phase has been shown to have a particularly strong impact on Arctic cloud radiative effects on the surface (Shupe and Intrieri, 2004), it is particularly important to understand how differences in viewing geometry impact observations of different cloud phases. Differences between space-based and surface-based clouds (ice clouds, liquid clouds, and mixed-phase clouds) amounts, and cloud ice and liquid water contents are shown in terms of monthly means. Based on the comparison performed here, this study also proposes blended products of cloud property vertical distributions from surface and space-based cloud observations at those two Arctic sites to serve as a best estimate cloud product for model and reanalysis evaluation.

Space-based radar and lidar in this paper refer to existing instruments, i.e., the Cloud Profiling Radar (CPR) onboard the CloudSat and the Cloud–Aerosol LIdar with Orthogonal Polarization (CALIOP) onboard the Cloud–Aerosol lidar and Infrared Pathfinder Satellite Observation (CALIPSO). However, the conclusions will likely be valid for other space-based radar and lidar instruments, e.g., the ATmospheric backscatter LIDar (ATLID) and the CPR onboard the EarthCARE mission (Hélière et al., 2007).

From the possible Arctic atmospheric observation sites, we have selected Barrow (71∘19′ N, 156∘37′ W) and Eureka (80∘80′ N, 85∘57′ W) because of the availability of daily cloud vertical profiles from surface observations from 2006 to 2010 when space-based observations are available. The combined radar–lidar cloud fraction best estimation, cloud fraction vertical profiles, cloud phase vertical profiles, and cloud water content vertical profiles, from surface observations at these two sites, are described in detail in Shupe et al. (2011, 2015) and Shupe (2011). These products are based on coincident measurements from the Ka-band cloud radar, depolarization lidars including the micropulse lidar (MPL) at Barrow, and the high-spectral-resolution lidar (HSRL) at Eureka, microwave radiometer, and radiosondes, which are combined to determine cloud phase (Shupe, 2007) and microphysical properties at 1 min temporal and 100 m vertical resolutions.

Observations from CloudSat and CALIPSO provide an unprecedented opportunity for a spatially extensive picture of cloud cover in the Arctic (Stephens et al., 2002; Winker et al., 2003). The Vertical Feature Mask (VFM) version 3.01 from CALIPSO's CALIOP provides cloud vertical distribution in up to 10 vertical layers at 5 and 1 km horizontal resolutions, and up to 5 vertical layers at 1/3 km horizontal resolution (Vaughan et al., 2009). The vertical resolution is 30 m below 8.2 km, and 60 m between 8.2 and 20.2 km. A Selective Iterated BoundarY Location (SIBYL) scheme is applied to detect all features within a given scene. Strongly scattering features, e.g., stratus clouds, can be identified in a single laser pulse, with the 1/3 km horizontal resolution, and these features are then removed in order to detect any surrounding aerosol layers. Weakly scattering features, e.g., thin cirrus clouds, are detected with the average of several laser pulses, e.g., 5 km horizontal resolution, for higher signal-to-noise ratio (Vaughan et al., 2005). Compared to the 1 km resolution data, the 5 km resolution product can identify weaker cloud features (Vaughan et al., 2009). Combining the cloud layer products at 5 and 1/3 km provides a complete vertical distribution of clouds from CALIPSO (Vaughan et al., 2005, 2009). The newly available VFM version 4.10 reports the spatial and optical properties all cloud layers detected at 5 km averaging resolution, and combination of VFM at 5 and 1/3 km is no longer needed for a complete cloud vertical distribution. In this study, the CALIPSO products (version 3.01) from June 2006 to December 2010 were obtained from the Atmospheric Science Data Center at NASA Langley Research Center.

The CPR onboard CloudSat also provides an echo mask, in the variable “CPR_Cloud_mask” at 125 vertical range bins, with a bin size of 240 m, in a product known as the Level 2 geometrical profiling product (2B-GEOPROF; Marchand et al., 2008). The latest CloudSat cloud mask (R04) has negligible surface contamination from about 0.96 km above the surface. Due to the surface clutter, only strong cloud or precipitation signals can be detected in the lowest approximately 0.7 km, while weaker cloud signals are missed. In this study, a range bin is defined as clouds when the CPR_Cloud_mask is equal to or larger than 20, which includes weak echo, good echo, and strong echo. Very weak echo and echo with likely surface clutter are not included. This threshold is the same as that used in the Radar–Lidar Geometrical Profile Product 2B-GEOPROF-lidar (Mace et al., 2009; Mace and Zhang, 2014), and a false positive detection of 5 % is estimated with this threshold in the 2B-GEOPROF-lidar (Mace et al., 2009). The 2B-GEOPROF-lidar merges the CloudSat GEOPROF (Marchand et al., 2008) and the CALIPSO VFM (Vaughan et al., 2009). The 2B-GEOPROF-lidar contains parameters for up to five hydrometeor layers, including the cloud base and cloud top heights above mean sea level for each hydrometeor layer in one radar footprint along with the longitude and latitude.

A Level 2 combined product, 2B-CLDCLASS-lidar, combines CPR and CALIOP measurements for cloud phase determination into eight basic cloud types (Sassen and Wang, 2012). Ice, water/liquid, and mixed-phase clouds are identified for up to 10 layers. The 2B-CLDCLASS-lidar collocates CALIOP L1 measurements to CPR footprints, then determines cloud vertical structures (Wang et al., 2008) and cloud phase. The microphysical property differences between water and ice particles, including size, location, falling speed, and number concentrations, result in large differences in their radiative properties, and in turn large differences in the CALIPSO lidar and CloudSat CPR signals. Cloud phase is effectively determined using the different sensitivities of CloudSat radar and CALIPSO lidar to ice crystals and water droplets, together with the cloud top and cloud base temperatures.

Based on the measured CPR radar reflectivity factor, another Level 2 product, the CloudSat Radar-Only Cloud Water Content product (2B-CWC-RO), estimates cloud liquid and ice water content, as well as effective radius. Effective radius and water content are retrieved based on the assumption that the radar profile is due to a single phase of water, either liquid or ice. Using a simple scheme based on a model temperature profile, this product combines separate liquid and ice profiles into a mixture of ice and liquid phases over a portion of the vertical profile within the proper temperature range. The temperature profile is obtained from European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data that have been collocated in space and time to the CloudSat radar profile and interpolated to the CloudSat vertical resolution. It should be noted that the retrieval is not designed to determine mixed-phase cloud properties directly.

In this study, vertical profiles of cloud fraction from CALIPSO at 1/3, 1, and 5 km horizontal resolution, 2B-GEOPROF and 2B-GEOPROF-lidar, vertical profiles of cloud phase (ice, liquid, and mixed phase) from 2B-CLDCLASS-lidar, and vertical profiles of cloud effective radius and water content from 2B-CWC-RO are calculated and examined. Vertical profiles of all these products within 50 km of the two Arctic atmospheric observation sites, Barrow and Eureka, are extracted and archived. The cloud fraction vertical distribution at a resolution of 30 m is calculated as follows. The mean cloud fraction at each vertical level is calculated as the ratio of the number of profiles with clouds detected at a particular vertical level to the total number of profiles. The cloud vertical distributions from CALIPSO at 1/3 and 5 km are calculated first, then combined as the mean of the cloud fractions at each vertical level. This combined product, referred as CALIPSO 5 km, provides a complete vertical distribution of clouds from CALIPSO and is shown in Sect. 3. For comparison, the vertical profiles of cloud fractions from CALIPSO at 1/3 and 1 km are also combined (referred to as CALIPSO 1 km) and shown in Sect. 3. For cloud microphysical property vertical distribution, the mean cloud phase frequency at each vertical level is calculated as the ratio of the number of profiles with each phase to the total number of profiles. Mean cloud water content for the ice (liquid) phase at each vertical level is calculated as the mean value of water content from all available ice (liquid) cloud retrievals at that level. To derive these statistics, ice in any type of clouds (ice and mixed phase) is included, while liquid in any type of clouds (liquid and mixed phase) is included, respectively. After this step, the vertical resolution of all products is 30 m. Total cloud (ice clouds, liquid clouds, mixed-phase clouds) amounts are also calculated as the ratio of the number of profiles with clouds (ice clouds, liquid clouds, mixed-phase clouds) detected in any layer to the total number of profiles.

Surface-based radar, lidar, and radar–lidar combined products are available from June 2006 to December 2010. Details of the collection and processing of the data can be found in Shupe (2011) and Shupe et al. (2011, 2015). Surface observations of good quality are available at Eureka for most of this time period and at Barrow from mid-February 2008 to December 2010. Hereafter, “observations at Barrow and Eureka from 2006 to 2010” refers to observations at Barrow from June 2006 to December 2010 and observations at Eureka from mid-February 2008 to December 2010. For consistency, the space-based results are considered over the same time periods as the available surface observations at each site. Monthly means are calculated for both surface observations and for the space-based sensors. All heights are above the mean sea level. All surface profiles in a month are accumulated to calculate monthly means. The monthly mean sample number of the satellite sensors is a function of latitude in the Arctic, with the fewest at 60∘ N, gradually increasing to a maximum around 80∘ N (Liu, 2015). Both factors are reflected in the large number of samples at Eureka, with over 6000 total samples per month from June 2006 to December 2010 in contrast to around 1500 total samples at Barrow per month from mid-February 2008 to December 2010. The vertical resolution of the calculated space-based monthly means is interpolated to 100 m to be consistent with and compared to those from surface observations.

This study compares the annual cycles of cloud vertical distributions of total cloud, ice cloud, liquid cloud, and mixed-phase cloud occurrence fractions from combined surface active lidar–radar observations and from multiple space-based active lidar–radar products at two Arctic atmospheric observation stations, Barrow and Eureka. The primary conclusions are as follows.

All space-based active radar–lidar cloud observations have limitations in the lowest 1 km a.m.s.l.; the surface measurements have superior performance near the surface, and thereby complement the space-based observations. Surface observations show that the highest total cloud fractions of all clouds, ice clouds, liquid clouds, and mixed-phase clouds appear between the surface and 1 km. All space-based observations show lower total cloud fractions below 1 km, with the lowest from 2B-GEOPROF, then CALIPSO 1 km, CALIPSO 5 km, and 2B-GEOPROF-lidar. The annual mean total cloud fractions from space-based observations show 25–40 % fewer clouds below 0.5 km than those from surface-based observations. Compared to surface-based observations, space-based observations show much fewer ice clouds and mixed-phase clouds, and slightly more liquid clouds from the surface to 1 km. These results are generally consistent with conclusions from previous studies (Protat et al., 2014; Blanchard et al., 2014; Mioche et al., 2015).

Surface observations perform well in describing the cloud vertical distribution at these observation sites. Above 1 km, space-based observations show similar patterns to surface observations, but different magnitudes for total clouds, ice clouds, liquid clouds, and mixed-phase clouds. For satellite-based total cloud fractions, CALIPSO 1 km shows the lowest values, with higher values from CALIPSO 5 km especially above 6 km, and the highest values from 2B-GEOPROF mainly in the middle level. The 2B-GEOPROF-lidar, which merges CALIPSO and CloudSat, provides the vertical distribution closest to that from surface observations. While the surface observations generally show cloud fractions that are comparable to, or higher than, the satellite-based fractions at most heights, the space observations show greater ice cloud fractions above 9 km, greater liquid cloud fractions in general, and greater mixed-phase cloud fractions above 1 km.

For the annual cycle of the total cloud fraction, monthly means from space-based observations are generally lower than those from surface observations. Each perspective has its limitations, with the surface observations missing some high-level clouds and the space-based sensors missing many low-level clouds. Both estimates are likely lower than the true cloud fraction, if those missed clouds do not overlap with other clouds. Because low clouds are more prevalent at these locations, the surface-based estimate is likely closer to the true total cloud fraction. Annual cycles of monthly mean cloud occurrence by phase show fewer ice and mixed-phase clouds, and greater liquid clouds from space-based observations. This result suggests that active sensor satellite-based estimates of cloud fraction across the Arctic are likely lower than the true cloud fraction, particularly at lower levels and at times of year when low clouds are frequent.

Annual cycles of the total cloud fraction at Barrow and Eureka show a similar evolution, with highest values in autumn, e.g., September and October, and local minimum values in summer, e.g., June and July, and with generally higher monthly cloud fractions at Barrow, except in January and February. Annual cycles of ice clouds at both sites also have a similar evolution, with a relative decrease in summer, and show similar magnitude; liquid-containing clouds at Eureka show lower values than those at Barrow, and its maximum generally shifts to the autumn relative to that at Barrow. These similarities and differences in annual cycles explain the key differences in the total cloud fractions and can be attributed to the generally colder and drier conditions at Eureka relative to Barrow (e.g., Shupe, 2011).

A blended cloud fraction vertical distribution using the larger value of surface and space-based observations can provide a more complete description of the cloud vertical distribution of total clouds, and ice, liquid, and mixed-phase clouds from the surface to 11 km. Such a blended product would be important when considering net atmospheric heating rates above these sites. Such an approach can also be useful in the tropics for a complete depiction of the cloud fraction vertical distribution.

Existing space-based cloud distributions in the lowest 1 km do not capture all clouds, especially ice and mixed-phase clouds. How these missed clouds in the lowest 1 km affect the radiation flux calculations at the surface and at the top of the atmosphere is a topic of future work and may impact past studies that examine Arctic surface radiative fluxes, as suggested by L'Ecuyer et al. (2008). The blended cloud property vertical distribution can be used as an input to a Monte Carlo radiative transfer model for a more accurate surface radiation flux calculation at these sites. A blended cloud property vertical distribution can also be used to evaluate cloud parameterizations in both weather and climate models (Klaus et al., 2016), to study Arctic atmosphere–sea ice–ocean interactions (Kay et al., 2008; Kay and Gettleman, 2009; Taylor et al., 2015; Liu et al., 2012a), and in other Arctic cloud studies (Devasthale et al., 2011; Liu et al., 2012b; Liu and Key, 2016).

Low-level clouds are frequent in the Arctic and important for the surface radiation balance. While space-based cloud observations from active radar–lidar sensors have been critical for improving our understanding of Arctic clouds and their interactions with other climate components, challenges remain in depicting Arctic low-level clouds from space. Surface observations of clouds at existing atmospheric observatories and a few field campaigns have provided valuable information on Arctic clouds, especially for studying low-level clouds (Tjernström et al., 2014; Uttal et al., 2002). However, such observations are limited in spatial extent and may not represent pan-Arctic cloudiness. Thus, it is critical to combine key information from both space-based and surface-based cloud measurements to provide the most comprehensive characterization of Arctic clouds possible and to facilitate further understanding of the Arctic climate system.

Cloud frequency from the surface is calculated in the temporal domain, while the cloud fraction from space-based observations is calculated in the spatial domain although near the surface sites. Differences in spatial resolution, viewing angles, vertical resolution, instrument sensitivity to clouds, and retrieval algorithms may all contribute to the differences in the cloud vertical distributions from different instruments. Long-term averages of products may mitigate the impacts of some of these factors. Causes of the remaining differences are worth further investigation.