Arctic low cloud changes as observed by MISR and CALIOP: Implication for the enhanced autumnal warming and sea ice loss



[1] Retreat of Arctic sea ice extent has led to more evaporation over open water in summer and subsequent cloud changes in autumn. Studying recent satellite cloud data over the Arctic Ocean, we find that low (0.5–2 km) cloud cover in October has been increasing significantly during 2000–2010 over the Beaufort and East Siberian Sea (BESS). This change is consistent with the expected boundary layer cloud response to the increasing Arctic evaporation accumulated during summer. Because low clouds have a net warming effect at the surface, October cloud increases may be responsible for the enhanced autumnal warming in surface air temperature, which effectively prolong the melt season and lead to a positive feedback to Arctic sea ice loss. Thus, the new satellite observations provide a critical support for the hypothesized positive feedback involving interactions between boundary layer cloud, water vapor, temperature, and sea ice in the Arctic Ocean.

1. Introduction

[2] Summer sea ice extent in the Arctic is shrinking at a pace faster than most of the climate model predictions [Stroeve et al., 2007]. Arctic warming, nearly twice as large as the global average [Intergovernmental Panel on Climate Change, 2007; Graversen et al., 2008; Gillett et al., 2008], has been the fundamental driving force of the rapid sea ice loss. Because perennial ice is increasingly replaced by thinner first-year ice [Nghiem et al., 2007; Kwok, 2007], the ice pack becomes more vulnerable to annual warming and wind-driven export, leading to the expectation that the summer Arctic Ocean would be ice-free in 20–30 years [Serreze et al., 2007].

[3] The Arctic warming reported in surface air temperature (SAT) occurs nonuniformly with season with the strongest increase in autumn [Serreze et al., 2009]. Over the Arctic Ocean, the Beaufort and East Siberian Sea (BESS) show the largest SAT increases in September–November, resembling the pattern of sea ice reduction in September. An immediate impact of the rising autumn Arctic temperature is to lengthen the melt season and reduce the possibility of perennial ice pack formation. As illustrated in Figure 1, it was not until 2000–2009 that the autumn temperature starts to increase significantly above the envelope defined by previous decades. Interestingly, the temperature increases are most pronounced during the period of late September to December, but with only moderate in spring and summer. This nonuniform SAT change is puzzling because it cannot be explained by direct solar radiation. Current research has been focused on coupled atmospheric-oceanic-cryospheric processes.

Figure 1.

(a) Daily mean SAT from National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis over the Beaufort-East Siberian Sea (110°E–220°E and 70°N–80°N) for each decade since 1950. (b) The SAT perturbations with respect to the 1950–1980 mean.

[4] In a quest for the cause of the intensified Arctic warming, several mechanisms have been explored, including oceanic heat transport [Shimada et al., 2006], influence of the upper air [Graversen et al., 2008], ocean-to-atmosphere heat transfer [Serreze et al., 2009], and water vapor feedback [Curry et al., 1995; Screen and Simmonds, 2010]. In these proposed mechanisms, clouds are still an enigma of the coupled Arctic climate system, because they are related intimately to regional radiation and dynamics [Curry et al., 1996; Intrieri et al., 2002]. Advanced from the positive ice-albedo feedback concept [Lindsay and Zhang, 2005], a cloud-temperature-ice feedback is suggested as an effective mechanism to warm autumnal SAT [e.g.,Kay and Gettelman, 2009; Vavrus et al., 2010; Eastman and Warren, 2010a]. This mechanism is based on increased absorption of solar radiation during summer, since more open water and hence more evaporation occurs over the Arctic Ocean [Perovich et al., 2007]. As a result, more clouds are likely to form in autumn when air temperature drops. In autumn, cloud longwave (LW) radiation, especially from liquid clouds [Shupe and Intrieri, 2004], dominates the surface heat budget by trapping LW between the surface and cloud layers. The increased cloudiness prevents or delays the accumulated heat from releasing back to space. Thus, the LW trapped by clouds could effectively increase SAT and lengthen the melt season. As shown by Belchansky et al. [2004], the magnitude of the summer melt is closely related to changes in the duration of the melt season, and the longer the season, the less chance of forming perennial ice, and the more thinning of sea ice. Because the cloud warming effect is proportional to low cloud fraction (CF) [Intrieri et al., 2002], it is hypothesized that more open water from the summer-melt Arctic Ocean could yield more cloud amount in autumn, and more trapped sensible heat.

[5] Cloud observations over the Arctic Ocean have been difficult and controversial [e.g., Schweiger, 2004; Wang and Key, 2005; Eastman and Warren, 2010b]. Most of ground-based observations are limited to islands [e.g.,Eastman and Warren, 2010a], sea ice regions [e.g., Uttal et al., 2002; Shupe et al., 2011], and short-period field/airborne campaigns [Curry et al., 2000; Verlinde et al., 2007; McFarquhar et al., 2011]. Because of the poor contrast between clouds and icy/snowy surfaces, cloud observation in the Arctic is a great challenge from space, and the reported trends on Arctic cloud cover have been controversial, sometimes contradict to each other. In this study, we analyzed the low-cloud observations from the Multiangle Imaging SpectroRadiometer (MISR) instrument on NASA's Terra satellite since 2000. Although it is a short period, as seen inFigure 2, MISR observations cover a critical period during which the rate of Arctic summer sea loss is tripled, compared to one in the previous two decades. The severe sea ice loss reached a record in summer 2007. We also analyzed the cloud data from Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) on the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellite, to verify the cloud changes observed by MISR. With a superb vertical resolution, CALIOP is considered as one of the most accurate cloud remote sensing instruments in terms of unambiguously distinguishing between clouds and surfaces. Both MISR and CALIOP observations are made from sun-synchronous orbits (with 10:30 descending and 13:30 ascending equator-crossing time, respectively). In the Arctic the two data sets are closer or even overlapped in local time sampling. Together, MISR swath coverage and CALIOP vertical resolution provide a more robust picture of recent cloud changes over the Arctic Ocean.

Figure 2.

Sea ice extent (SIE) data from National Snow and Ice Data Center (NSIDC) [Fetterer et al., 2002], showing the accelerating SIE loss in September. The dashed lines are the fits to the SIE before and after 2000, showing rates of −0.041 × 106 km2/yr and −0.14 × 106 km2/yr, respectively. The Terra/MISR observations since 2000, although short, cover the period with the rapid SIE reduction. CALIPSO was launched just in time (2006) to observe the record low SIE in 2007.

2. Data and Methods

[6] With nine view-angles and four visible wavelengths, MISR employs a stereoscopic technique to determine cloud top heights at an accuracy higher than most of the passive sensors from space [Horváth and Davies, 2001; Muller et al., 2002; Moroney et al., 2002; Davies et al., 2007]. As shown in Figure 3, the three near-nadir cameras, separated by 26° in view angle and 46 s in time along track, can measure the parallax of a cloud feature to subpixel (<275 m) accuracy. In the absence of along-track cloud motion, the parallax is related to cloud top height (CTH) through the simple geometry. Different from other passive cloud imaging techniques, MISR cloud detection is not much limited by snowy/icy surfaces nor by radiometric calibration, because it depends primarily on brightness textures in a scene (relative differences in reflectance) to determine the feature's positions. From these positions, parallaxes are derived and used to compute the feature's geometric height. MISR stereo technique is sensitive to features with optical depth >0.3 [Marchand et al., 2007]. Thus, the MISR stereo cloud detection is similar to lidar/radar methods using feature height to discriminate clouds from surfaces. MISR cloud height product is generated at 1.1 km horizontal resolution with a swath of ∼350 km. This swath is important for obtaining reliable cloud statistics in a short period of time because sufficient density of sampling is needed to average down large cloud variability.

Figure 3.

Schematic diagram to show the stereo technique with three MISR angles for CTH measurements. The multiangle images are separated by 46 s in time, during which cloud features might move. If there is no along-track motion, the parallax is simply proportional to CTH.

[7] MISR measurements are particularly valuable for cloud observations over the Arctic Ocean where icy and snowy surfaces are problematic to other sensors that are sensitive to radiometric/thermal calibration and the contrast between surface and cloud radiances. These conditions are often violated in the Arctic because cloud IR radiances are close to the surface radiation and cloud visible reflectance is close to those from snowy/icy surfaces [e.g., Rossow et al., 1993]. Therefore, using the IR method to detect clouds and determine cloud heights has severe limitations in the polar atmosphere as the temperature lapse rate becomes so shallow or even reversed.

[8] Since February 2000 MISR has been continuously observing the Arctic region during daytime. MISR 14 bit imagers allow acquisition of useful measurements even when solar zenith angle is low, which extends polar observations to early spring and late fall. As seen in an MISR image (Figure 4), a cold-air outbreak event occurred over the Beaufort Sea on 17 October 2009, triggering a vast blanket of low-level roll clouds near the marginal ice zone (MIZ). Over the open water where air over sea surface is relatively warm and humid, the cold airflowing from sea ice to open water, creating a unstable and saturated marine planetary boundary layer (PBL) for these roll clouds to form. As the Arctic sea ice continues to retreat, air-sea interactions with a fast cloud response to atmospheric disturbances becomes increasingly important in the Arctic basin. Thus, PBL cloud feedback on surface radiation, atmospheric circulation and hydrological cycle is an interesting process to explore.

Figure 4.

Roll clouds from a cold-air outbreak event over the Beaufort Sea on 17 October 2009. The image was acquired from the 46° forward camera of the MISR instrument on NASA's Terra satellite. The top of cloud rolls/streets rises from 200 m near the sea ice edge to ∼1 km in the ocean water. Some transparent high cirrus is evident in the middle of the image.

[9] In this study, the standard MISR Level-2 stereo cloud product, namely, zero-wind cloud top height (CTH0), is analyzed. This stereo product is retrieved from three (±26° and nadir) MISR cameras with the assumption of no cloud motion. The pattern-matching algorithm for producing CTH0needs to be fast enough to process all the MISR data. Thus, it only calculates the parallax to accuracy of the pixel size (275 m), which corresponds to 562 m in cloud height for the three-angle stereo height retrieval. In the current algorithm, a pattern is defined in an 11 × 11 pixel area, but the CTH0 retrieval is reported at 1.1 km horizontal resolution. The data are limited to daytime only, and therefore coverage for the Arctic is poor in winter months. The months with useful data for the Arctic are generally from March to October. Cloud wind effects on CTH0 can be significant and the correction method is discussed below.

[10] Two postprocessing steps are required in order for scientific uses of the CTH0 product. The first processing is to correct CTH0for the along-track wind effect. Roughly speaking, a wind speed error of 1 m/s in the along-track direction (nearly meridional at low and middle latitudes) is equivalent to a ∼100 m error in CTH. To correct the wind-induced component in CTH0, we compute the along-track wind speed from the NCEP reanalysis and subtract the wind-induced parallax from CTH0 [Wu et al., 2009]. The resulting CTH is termed CTHNCEP. The second step processing is to determine the height threshold for cloud detection. In this study a feature is classified as cloud if CTHNCEP is greater than mean surface elevation +1 standard deviation of topography +562 m. The 562 m is the discretization error associated with the standard CTH0 retrieval, which is the dominant error source in the CTH0 measurement. NCEP PBL winds may also have uncertainties, and we expect them to be similar to those near the surface [Meissner et al., 2001], with ∼0.5 m/s accuracy and 2.4 m/s standard deviation. The surface height variation is negligible for oceans. The MISR team has also developed an approach of using MISR-retrieved cloud winds for the CTH0 correction. However, MISR cloud wind retrievals are too sparse over the Arctic Ocean, not suitable for this study.

[11] The flagged CTHNCEPdata are further composited into a standard grid with 0.5 km height and 2.5° × 2.5° latitude-longitude bins to calculate monthly cloud fraction (CF), namely CFbA. MISR has been operating almost continuously since March 2000 except in October 2008 when the instrument went off for more than 16 days. The CFbA provides a measure of horizontal cloud cover at each height bin. In addition to cloud fraction, another useful quantity that can be derived from CFbA is the mean CTH. It may be calculated for each gridbox or for a region of interest. The mean CTH can be used as an indicator of cloud vertical extent if cloud bases are unchanged. Although the total CF may be same, cloud tops may move to a higher altitude and produce a higher mean CTH. As seen inFigure 5, there is a considerable amount of variability in CFbA, which can be characterized by the mean CTH in the PBL. Most of the PBL clouds reside at altitudes <3 km, and both cloud amount and CFbA peak may vary largely from month to month and from year to year. In this study we define a mean PBL CTH, namely CTHPBL, as the average CTHNCEP weighted by CFbA between 0 km and 3 km.

Figure 5.

Monthly CFbA over Beaufort and East Siberian Sea (BESS) for June, July, August, September and October in 2006–2010 as observed by MISR (solid) and CALIOP (dashed). Because MISR and CALIOP CF data have different height bin sizes (0.5 km and 0.2 km, respectively), the CF profile is normalized by height bin size and reported in %/km, i.e., percent per unit height bin, such that the integral over all height bins (= total cloud fraction) + clear-sky fraction is unity. Only daytime CALIOP data are used here and MISR CF in October 2008 has large sampling error due to 16 day off in instrument operation.

[12] In addition to MISR CTHNCEP, we also analyzed monthly CALIOP daytime cloud statistics for the BESS region, using version 3 Level 2 cloud layer product (05km_CLAY). Launched in 2006, CALIOP is a dual-frequency (532 and 1064 nm) lidar with high vertical resolutions for cloud and aerosol measurements. CALIOP can penetrate into optically thin clouds to measure cloud thickness and multilayer clouds. To mimic the MISR CTH measurement, we pick only one cloud height per CALIOP profile, which is the lowest cloud top height in a profile. This MISR-like CALIOP CTH is further sorted into 0.2 km height bins and 2.5° × 2.5° latitude-longitude bins for monthly CFbA statistics. Because of the narrow curtain sampling, the monthly lidar CFbA maps are subject to larger meteorological variability than MISR maps.

[13] To compare MISR and CALIOP CFbA statistics, we choose a large domain to ensure a sufficient number of samples. The BESS region is of most interest because of the severe sea ice loss in recent years. Figure 5 shows the vertical distribution of MISR and CALIOP CFbA over the BESS region for five months (June, July, August, September, and October) in 5 years (2006–10). Remarkable agreement is found between the two CFbA data sets, especially at altitudes above ∼1 km. In the PBL, MISR sometimes overestimates CFbA at 1 km but always underestimate it at 0.5 km in general. The underestimation at 0.5 km creates an artificial peak in MISR CFbA that appears to be higher than CALIOP in some months, which is a manifestation of MISR's limited ability of detecting clouds below 0.5 km. Measurement uncertainties, such as MISR CTH0quantization error (562 m) and NCEP wind errors in the parallax correction, can misclassify these clouds as surface features. The quantization error is correctable with a subpixel matching algorithm, but wind errors would require more accurate MISR cloud wind measurements. In the new version algorithm, which is under development by the MISR team, the stereo products will be improved in these aspects. The limited MISR ability for near-surface cloud detection is evident inFigure 5, showing large 0.5 km differences in June–September when most clouds are below 1 km but small differences in October when CTH is elevated. The CFbA missed by MISR at altitudes <1 km also affect the total CF substantially, which explains the large differences seen in total CFs between MISR and the lidar. Nevertheless, compared to other passive techniques, MISR has demonstrated a great potential with better accuracy and capability for PBL clouds in terms of height registration and cloud detection over snowy/icy surfaces [Garay et al., 2008; Wu et al., 2009].

[14] Rising PBL CTH in the BESS region is evident from June to October in the MISR and CALIOP CFbA statistics, which is likely associated with the increasing water vapor abundance in the BESS region and the increased amount of heat accumulated/trapped in the PBL. Like the process of stratocumulus-to-cumulus transition over the subtropical ocean, cloud top entrainment, which changes with surface temperature, leads to a deepening PBL, weaker or more diffused inversion, and decoupled cloud cover [Bretherton and Wyant, 1997]. In the Arctic, PBL CTH is generally higher over water than over ice [Brümmer, 1996; Brümmer and Thiemann, 2002; Palm et al., 2010]. Of particular interest is the 2007 case when Arctic summer sea ice extent hit a record low. Reduction of total cloudiness over Beaufort and Chukchi Sea was found during June–August (JJA) in 2007, but the 2007 autumn low cloud amount had a higher percentage [Kay and Gettelman, 2009; Eastman and Warren, 2010a]. Their finding for a high CF in autumn is generally consistent with the observations from MISR and CALIOP CFbA profiles in the BESS region. Rising CTH in the PBL, which is probably associated with increasing cloud optical depth, makes these clouds more readily detectable by MISR, as the stereo technique is currently limited to clouds with CTH > ∼0.5 km. Figure 5 shows significant enhancement in low cloud cover during September and October 2007, and the peak of CFbA for October 2007 is elevated to ∼1.5 km, compared to 1 km in 2006 and 2008. The CFbA for October 2010 also peaks at a slightly higher altitude.

[15] Seasonal variations between March–November from multiyear averages over the BESS region are summarized in Figure 6for MISR (2000–2010) and CALIOP (2006–2010) CFbA. Again, the two data sets reveal a consistent seasonal variation with more PBL cloudiness during May and October, but less in the middle and upper troposphere. The high-level clouds have a peak in the summer months. The satellite cloud climatology for the BESS region is also consistent with the ground-based observations for PBL clouds at Barrow and SHEBA [Shupe et al., 2011], although the latter shows enhanced cloudiness in slightly different months (April and September/October) at the two specific sites. It is worth noting that in these observations, autumn has more PBL clouds with CTH extended to a higher altitude compared to spring. The enhanced autumnal cloudiness is expected for the increasing evaporation over open water and the accumulated heat from absorption of solar radiation during summer. However, the cause for springtime enhancement in PBL cloudiness is still puzzling.

Figure 6.

Monthly mean CFbA from March to November for MISR (2000–2010) and daytime CALIOP (2006–2010). The CALIOP data are for the lowest CTH in each profile to mimic the MISR measurement. CALIOP CTH exhibits a similar climatology even without selecting the lowest CTH.

3. Arctic Low Cloud Changes

[16] Because of sufficient sampling in the Arctic region, MISR can provide maps of the linear trends for 2000–2010 from the monthly CFbA data. We regress the monthly gridded data independently for three altitude regions: low (0–3 km), middle (3–6 km) and high (6+ km) levels, for the mean, linear trend and the standard deviation (σ) about the trend. The maps of mean CF and trend for low clouds (0–3 km) are shown in Figure 7 for months of June to October.

Figure 7.

Maps of MISR low CF (0–3 km), its 10 year trend, and 5 year CALIOP mean CF (0–3 km) over the Arctic Ocean. Sea ice extent is indicated by the black contour line, which marks 20% monthly ice coverage. The box indicates the Beaufort-East Siberian Sea (BESS) region of interest in this study, and there is no valid data in the white circle (84°N poleward). At right are MISR monthly CF (diamonds) and trend (red line) over the BESS region. Cyan lines are the CALIOP data from the same region. The estimated trend or slope of MISR CF in BESS is indicated in the title, showing 2-σ uncertainty in the parentheses.

[17] A striking feature is the increasing trend in October low-cloud cover seen by MISR over the Arctic Ocean, showing an average of 2.12% per year in the BESS region. Although Arctic low-cloud CF is highest in September, most of the CF trends are insignificant, except for months of June and October. In the case of October, as expected from accumulated evaporation from summer, its increase in low-cloud cover occurs mostly near the MIZ where air above open water is moist and nearly saturated such that PBL clouds readily form as cold airflows over warmer water. Within the BESS region, the mean CFs over the Eastern Siberian and Laptev Sea, appear to be lower than the Beaufort and Chukchi Sea. But the latter are experiencing a more rapid increase during recent years, which warrants close monitoring.

[18] MISR observed the greatest amount of low cloud in the BESS region in October 2007, following the record-breaking summer sea ice loss in that year. Because more open water and evaporation became available in the 2007 summer, it is anticipated that the Arctic PBL would respond to the change with more low clouds in autumn when temperature drops.Kay and Gettelman [2009] reported the similar PBL cloud increases in early fall over the Beaufort, Chukchi and E. Siberian Sea, and a higher cloud fraction over open water.

[19] The CALIOP data in the BESS region, although only available since 2006, support the increasing trend seen by MISR for October. The CALIOP low-cloud (0–3 km) CFs would be slightly higher than the MISR CF if integrated from 200 m to 3 km. To make it comparable to the MISR data, here we integrate the CALIOP data from 500 m to 3 km. Despite two very different cloud observing techniques both MISR and CALIOP low-cloud cover data show a convincing trend of an increase since 2006, regardless of whether it is integrated from 200 or 500 m. Studying a longer record from ICESat and CALIOP data,Palm et al. [2010] reported a ∼7% increase in total CF during 2004–2008 over a region slightly larger than BESS. Since most of the CF increase is associated primarily with low clouds, the trend from total CF would be weaker than with only low clouds.

[20] The mean and trends (with 2σuncertainty) of MISR low, middle and high-level CFs over the BESS region are summarized inTable 1 for March–October. Most of the regressed trends are not significantly above the 2σconfidence because of large interannual variability in the Arctic. The middle-level clouds in September and October show a significant decreasing trend, while for the low-level clouds only June and October are identified with a significant increasing trend. Despite moist air in summer, low-cloud cover in the summer months does not show any significant trends. CFs from middle-to-high-level clouds may have opposite trends to low clouds (e.g., September and October), which make it more difficult to detect a significant trend in total CF. In October, the increase trend in low-cloud cover is so strong that the total CF still exhibits an increasing trend even though there is a decreasing trend in middle-to-high-level cloud cover.

Table 1. MISR Mean CF and Trends for the BESS Region During 2000–2010a
  • a

    Mean CF is in percent; Trends are in percent per year.

  • b

    Trends with 2σ significance.

>6 km
Trend (2σ)−0.07 (0.29)−0.05 (0.18)0.02 (0.07)0.01 (0.16)−0.01 (0.26)0.10 (0.21)0.01 (0.35)−0.17 (0.47)
3–6 km
Trend (2σ)0.09 (0.31)−006 (0.26)0.14 (0.15)0.05 (0.36)0.08 (0.39)0.19 (0.25)−0.36 (0.31)b−0.49 (0.40)b
0–3 km
Trend (2σ)0.12 (1.35)0.44 (1.11)0.55 (1.03)0.45 (0.31)b0.17 (0.71)0.04 (0.68)0.57 (0.76)2.12 (1.02)b
Trend (2σ)0.14 (1.38)0.33 (1.00)0.70 (0.92)0.51 (0.64)0.24 (0.68)0.32 (0.66)0.22 (0.69)1.46 (0.73)b

4. Discussions

4.1. Sampling and Resolution Issues

[21] Density of spatial sampling is critical for mapping regional cloud cover and its trends, especially in the Arctic basin where dynamical variability is often high. The MISR 350 km swath proves to be quite valuable, yielding a full coverage of the Arctic (70° poleward) region every three days, but it is still insufficient for resolving many short-scale variabilities. Hence, in this study we make monthly maps to further average down the short-term variabilities.

[22] Vertical resolution is also critical for separating between PBL clouds and the surface. Because atmospheric responses to the sea ice loss occur mostly in a layer near the surface, the ability of detecting and monitoring cloud changes in the PBL is more desirable than measuring total column cloud cover when understanding different processes is needed. Given complex forcing and feedback processes in the Arctic climate system, atmosphere and clouds may vary differently in each height domain [e.g., Graversen et al., 2008; Serreze et al., 2009; Vavrus et al., 2010]. In fact, the decreasing trend in October CF from middle-to-high clouds is not related directly to the sea ice loss. Instead, it is more likely affected by large-scale circulation in the free troposphere. Therefore, by separating PBL clouds from variations at other altitudes, it helps to disentangle dynamical and cloud processes over the Arctic Ocean. On the other hand, MISR coarse vertical resolution may miss some clouds if CTH is too close (<500 m) to the surface. Near the MIZ, cloud tops tend to elevate to a higher altitude over water during cold off-ice airflows [Brümmer, 1996] than over ice during warm on-ice airflows [Brümmer and Thiemann, 2002]. This shortcoming may explain why MISR observes less PBL clouds over ice than CALIOP (Figure 7).

[23] Surface albedo differences can be a factor to affect MISR cloud detection because clouds have higher contrast over water than over ice. However, in Figure 7, although the MISR CF distribution seems to be divided by the sea-ice border in October, there is no obvious division in the July map. Thus, MISR low (0.5–3 km) cloud cover is not likely to be affected much by the albedo effect. Furthermore, if it were the albedo effects that cause the October trend, we would observe a more prominent trend in September as it has the largest sea ice loss trend in summer.

[24] In addition, high-cloud blocking can affect the observed PBL cloud cover and its variations. High optically thick clouds prevent satellite sensors from seeing clouds beneath it. This is a fundamental limitation for all remote sensing techniques from space, and must be carefully considered when interpreting the results.Marchand et al. [2007] showed that MISR stereo technique is sensitive to features with optical depth >0.3. It means that high clouds with optical depth of 0.3 can play a role in blocking low cloud detection. However, in multilayer cases, the stereo result will come out of the winning brightness contrast generated by high and low cloud layers in a scene. If the reflectance of high clouds is relatively uniform and one of low clouds is more inhomogeneous, the stereo technique will yield a CTH for the low clouds, and vice versa.

[25] Cloud blocking affects the “apparent” variations in the observed low cloud amount as the amount of high clouds may change as well. In the case of the October low-cloud increase in the BESS region (Figure 7), the decrease in high-cloud amount may contribute partially to the low-cloud trend. In other words, the reduced high-cloud cover could be the cause of the observed trend in low clouds. However, the high-cloud decreasing trend inTable 1does not appear to be large enough to offset the observed low-cloud increasing trend in October.

[26] Given the aforementioned caveats and limitations, nevertheless, it is still encouraging to see the level of agreement between the MISR and CALIOP CFbA profiles (Figures 5 and 6). There are some robust cloud statistics revealed from two very different remote sensing techniques: passive stereo pattern matching (MISR) versus active lidar cloud backscattering (CALIOP). It remains challenging to measure PBL clouds with passive techniques from space, but the stereo technique demonstrates great potential of probing this low, shallow cloud layer. As a promising approach for future cloud remote sensing, the stereo cloud detection does not require knowledge of atmospheric thermal structure, nor assumptions about the atmospheric temperature lapse rate and surface emissivity that may be a problem for other passive techniques [e.g., Naud et al., 2005; Garay et al., 2008; Holz et al., 2008]. Finally, image matching has relatively loose requirements on radiometric calibration and therefore it is insensitive to cloud/surface emissivity, making it equally viable for water and ice clouds.

[27] MISR observations of Arctic clouds are consistent with multidecadal cloud measurements from ground-based visual data, which also show sharp increases of low-cloud amount in recent years in the Arctic [Eastman and Warren, 2010a]. However, sampling from ground-based stations is limited mostly to landmasses, and the statistics over the Arctic Ocean are based on a few sparse stations. The increasing trends observed by ground-based stations appear to be more prominent in the DJF and SON seasons, showing the strongest in SON. Over the Beaufort-Laptev Sea, the increase of low-cloud amount started to accelerate after ∼1998, coherent with the decreasing trend of September SIE. The consistency in low-cloud observations among MISR, CALIOP and ground-based techniques help to reconcile some of the reported differences in Arctic cloud measurements [Schweiger, 2004; Wang and Key, 2005; Eastman and Warren, 2010b].

4.2. Implications for Arctic Sea Ice and Air-Sea Interactions

[28] Low clouds have a net warming effect on SAT through trapping LW and sensible heat within the PBL for most of the seasons except summer [Screen and Simmonds, 2010]. The LW warming effect dominates heat budget in autumn and winter when shortwave (SW) forcing is reduced. Tendency of SAT warming with increasing cloudiness was observed on Arctic ground stations [Vihma and Pirazzini, 2005].

[29] The low-cloud increases observed by MISR and CALIOP support the positive cloud-temperature-ice feedback in autumn, which has been hypothesized as a cause of the accelerated SAT warming in the recent decade [e.g.,Kay and Gettelman, 2009; Vavrus et al., 2010]. Increases of low-cloud cover in autumn and winter may effectively lengthen the melt season by trapping more sensible heat near the surface over water for a longer period of time, which would lead to a reduced probability of forming perennial ice, and subsequently thinner sea ice and more loss in the coming years [Belchansky et al., 2004; Kwok and Rothrock, 2009]. Thinner first-year ice, vulnerable to the next melt season, likely continues to retreat and open up more water in the Arctic Ocean.

[30] The increasing presence of low clouds would enhance not only cloud radiative forcing but also hydrological cycle in the Arctic PBL. During cold-air outbreak events (off-ice flow to open water), surface heat fluxes (turbulent sensible and latent heat) are enhanced over open water with gradually rising PBL top [Hartmann et al., 1997; Brümmer and Pohlmann, 2000; Liu et al., 2006]. As a good insulator to prevent efficient exchanges of heat, momentum and moisture between the surface and atmosphere, sea ice provides a relatively stable PBL and inhibiting formation of convective clouds. Thus, more open water in the Arctic Ocean means more moisture supply with unstable PBL structures and more heat/momentum/moisture exchanges in terms of air-sea interactions. A reverse process to cold-air advection is warm-air outbreaks, producing on-ice flow from open water [Brümmer and Thiemann, 2002]. Warm-air advection of moisture can also increase cloudiness near the MIZ and affect surface temperatures [Vihma and Pirazzini, 2005]. As a fast process, clouds tend to respond to synoptic atmospheric disturbances quickly in the regions with water vapor abundance by increasing cloud cover and cloud thickness. In other words, Arctic clouds are playing an increasingly important role in balancing regional radiation budgets in spring and autumn, which is a feedback process in the Arctic climate system. However, a more quantitative assessment of such a feedback requires further investigations and understanding.

[31] Changes in the Arctic air-sea interactions are evident in the NCEP/NCAR reanalysis data for October SAT. To illustrate it with NCEP interannual SAT variability, we derive the empirical orthogonal functions (EOF) and their principal components (PC) from the monthly data from 1948 to 2010 for the Arctic region 60° poleward. In addition, we analyze short-term SAT variances by removing a 7 day running mean in the time series. Since low clouds have a typical lifetime less than 2–3 days [Brümmer and Pohlmann, 2000], the short-term variances serve as an indicator for the air-sea interactions over the Arctic Ocean.

[32] As shown in Figure 8 (top) and 8 (middle), the mean October SAT started a rapid increase since 1990, and so does the index of the first EOF pattern. The year of 1988 has been postulated to be a tipping point in Arctic climate changes in terms of sea ice thinning and loss [Lindsay and Zhang, 2005]. The leading EOF pattern reveals striking responses near the border of sea ice extent, or MIZ, where SAT might be effectively regulated by PBL clouds. In other words, clouds in the MIZ might have acted to stabilize the atmosphere by reducing surface-atmosphere temperature gradient in response to synoptic disturbances. Although latent/sensible heat fluxes between air and surface increase, SAT differences would be reduced in the cold-air and warm-air outbreaks as a result of the cloud presence.

Figure 8.

(top) October mean SAT map and trend for the 60° poleward region, (middle) the first EOF of October SAT anomalies and its PC index, and (bottom) short-term (<7 days) SAT variance and its time series. The first EOF count for 32% of the variance of October SAT variations in 1948–2010.

[33] A signature of the cloud stabilization effect is perhaps seen in the short-term NCEP/NCAR SAT variances for October. As shown inFigure 8 (bottom), the variances generally have low values outside the mean sea ice border, which is the region where most of low clouds form (Figure 7). The time series, showing a reduced short-term SAT variance since 1990s, seems to correlate with the increasing SAT trend. Thus, it would not be surprising that the increasing low-cloud cover in October had acted to stabilize SAT during the recent years. We also analyzed the NCEP/NCAR SAT data for other months and found similar behaviors with an increasing trend in SAT anomaly and the first EOF index. However, the decreasing trend of the short-term SAT variance is most evident in September and October.

5. Conclusions

[34] We analyzed 11 year MISR and 5 year CALIOP cloud observations over the Arctic Ocean and found a significant increasing trend in October low (0.5–2 km) cloud cover in the Beaufort and East Siberian Sea (BESS) during 2000–2010. The observed autumnal cloud increase is anticipated for PBL cloud response to the expanded open water in recent years and accumulated summer evaporation in the Arctic Ocean. An important implication of the low cloud increase in autumn is to warm regional surface air temperature (SAT), which has been accelerating in the recent decade. Furthermore, the observed cloud increase seems to support the theorized positive cloud-temperature-ice feedback to the Arctic sea ice loss, because a lengthened melt season from the cloud effect would weaken perennial ice formation. The presence of PBL clouds tends to stabilize SAT in the events of synoptic atmospheric disturbances. As Arctic sea ice continues to retreat, air-sea interactions are expected to play an increasingly important role in distributing sensible/latent heat and driving polar atmospheric circulations. Therefore, September–November remains to be a dynamic season to watch in the future Arctic changes. For reliable model predictions of future sea ice loss, it requires a better understanding and dedicated studies on how atmospheric-oceanic-cryospheric processes interplay over the Arctic Ocean.


[35] This work was funded by NASA Terra project, part of which was performed at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration (NASA). We are grateful for constructive comments and suggestions from anonymous reviewers, which helped to improve the manuscript. The data processing by the NASA Langley Research Center Atmospheric Sciences Data Center are gratefully acknowledged.