Geophysical Research Letters

Changes in seasonal cloud cover over the Arctic seas from satellite and surface observations

Authors


Abstract

[1] Winter and spring changes in cloudiness are compared over the arctic seas (ocean areas north of 60°N) from the TOVS (TIROS Operational Vertical Sounder) Polar Pathfinder retrievals and two separate datasets derived from the Advanced Very High Resolution Radiometer (AVHRR). All satellite products exhibit significant decreases in cloud fraction over the arctic seas during winter (December, January, February) on the order of 5 %/decade. An equally striking increase in spring (March, April, May) cloudiness is evident from the TOVS Pathfinder (TPP) and the extended AVHRR Polar Pathfinder (APP-x) projects. In the Central Arctic these positive trends can be as large as 15 %/decade. Surface observations from the Russian drifting meteorological stations are consistent with satellite- observed changes during the 1980s. Negative trends in spring cloudiness reported by Comiso [2003] are in conflict with these findings. Spring changes in cloudiness are associated with changes in the atmospheric circulation. These dramatic, large-scale changes may have substantial impacts on the surface energy balance.

1. Introduction

[2] The Arctic is believed to play a pivotal role in the global climate system. Models predict that change will occur earlier and more strongly in snow- and sea ice- covered areas, but sorting out the cloud-radiation interactions and feedbacks that are thought to play an important role in this behavior is a challenge. Consequently, observing changes in cloud properties is an essential step toward understanding why the Arctic has recently changed so rapidly [e.g., Serreze et al., 2000] and for assessing how these patterns of change may evolve.

[3] Two recent publications Wang and Key [2003] (hereafter referred to as WK) and Comiso [2003] (hereafter referred to as JC) report conflicting trends in Arctic cloudiness from two separate AVHRR-derived data sets. One shows an increase in spring cloud fraction [WK], the other a decrease [JC]. The present paper reports changes in Arctic cloud fraction from a third data-set -the TOVS Polar Pathfinder data set (hereafter TPP) [Schweiger et al., 2002] and compares temporal trends in these data sets with surface observations.

[4] The role of clouds in the transfer of radiative energy is complex and is controlled by many variables. Cloud fraction, height, microphysical properties and temperature profiles as well as surface properties all are important factors. Among these, cloud fraction is the one most readily available from satellite and surface observations. Moreover, the detection of clouds and determination of cloud fraction are the first steps in the retrieval of surface (albedo, surface temperature) and cloud properties (temperature or optical properties) and errors in the identification of clouds will affect the accuracies of those parameters. Therefore clouds not only are an important climate parameter but they are linked to other climate parameters as well. As a first step this paper focuses on the commonalities and differences in temporal changes in cloud fraction in these data sets.

[5] This investigation is motivated by the larger question of whether changes in cloud cover, and therefore radiation, may have played a role in driving observed sea-ice changes, such as thinning [Rothrock et al., 1999] and reduction in extent [Cavalieri et al., 2003]. I therefore focus on the arctic seas and exclude discussion of changes over land areas.

2. Retrievals From TOVS

[6] The TPP data set [Schweiger et al., 2002] contains atmospheric temperature and humidity profiles as well as retrievals of cloud fraction and height for the period 1980–2001. This information is derived from the TOVS instrument, which includes the High Resolution Infrared Radiation Sounder (HIRS) and the Microwave Sounding Unit (MSU), processed with the Improved Initialization Inversion (3I) algorithm [Scott et al., 1999]. The cloud detection scheme used by the 3I algorithm is based on a series of spectral tests and has been adapted specifically for applications over sea ice [Francis, 1994]. TPP cloud fractions have been extensively validated against meteorological surface observations and LIDAR retrievals [Schweiger et al., 1999, 2002]. Satellite-derived cloud fractions are within 20% (RMS difference) of surface observations.

3. Retrievals From AVHRR

[7] WK reported trends in cloud fraction from the extended AVHRR Polar Pathfinder data set (APP-x) for the period 1982–1999. JC presents an analysis of AVHRR retrievals using a different algorithm [Comiso, 2000] for the period 1981–2000. JC results discussed here are for sea ice areas (ice concentrations >80%). APP-x results show a statistically significant decrease (5 %/decade) in cloud amount over the Arctic during winter (December, January, February, hereafter DJF) and increases (3 %/decade) in spring (March, April, May, hereafter MAM) and an increase (1.5 %/decade) in summer (June, July, August, hereafter JJA). JC finds a decrease in cloud fraction in all seasons. Because the results in WK are for the entire Arctic (N > 60°N, including land), APP-x changes presented below are recomputed and limited to the arctic seas defined here as ocean areas north of 60°N.

4. Results

[8] Trends in the time series have been calculated by fitting a linear model to seasonal means. These trends are a convenient descriptor to compare the long term temporal variability in these time series over the period. The calculation of linear trends does not assume that an underlying linear process causes observed changes.

[9] Figure 1 shows seasonal trends in cloud fraction over the arctic seas from the TPP and APP-x data sets (see Table 1 for detailed statistics). During winter (DJF) both trends are negative: TPP is −4.1 %/decade and APP-x is −7.1 %/decade and are statistically significant at the 95% level. JC calculates a similar decrease of 5.4 %/decade over sea ice. Consequently, the decrease in wintertime cloudiness appears to be consistent among these three analyses. TPP and APP-x spring (MAM) trends (Figure 1b) are strongly positive, with trends of 4.8 %/decade and 3.4 %/decade. In contrast JC finds a small decrease in cloud fraction (1.3 %/decade). Though his results are averaged over different areas (sea ice with concentration >80%), a decrease in cloud fraction is present over all areas studied by JC thus excluding the averaging domain as a cause for this opposite trend.

Figure 1.

Time series of seasonally averaged cloud fraction over the arctic seas from the APP-x and TPP data set. Cloud fractions are averaged over ocean areas for a) winter, b) spring, c) summer and d) fall.

Table 1. Trendsa in Seasonal Cloud Fraction From TPP, APP-x and JC
 WinterSpringSummerFall
  • a

    Trends are given in %/decade with 95% confidence intervals and R2 values- providing information the variance explained by the linear fit-given in parentheses. Bold faced numbers are significant at the 95% level.

TPP−4.1 (±3.4, 0.18)4.8 (±1.6, 0.68)0.4 (±2.1, 0)0.4 (±1.4, 0)
APP-x−7.1 (±4.3, 0.38)3.4 (±3.0, 0.33)0.9 (±1.5, 0.13)−1.6 (±2.7, 0.1)
JC (sea ice)−5.4−1.3−1.5−3.4

[10] Linear trends for the summer and fall seasons in TPP and APP-x data sets (Figures 1c and 1d) are not statistically significant and cannot be considered meaningful descriptors for the change in cloudiness over this period. The 95% confidence intervals shown in Table 1 show the substantial uncertainty in these trend estimates, but also indicate that TPP and APP-x agree within their range of uncertainty in winter and spring. APP-x and TPP trends during spring are clearly positive, thereby contradicting the finding of JC. TPP and APP-x seasonal estimates are well correlated in spring (R = 0.45) and summer (R = 0.68). Differences in mean winter and summer cloud fractions are within 2% (Table 2). The TPP cloud estimates exceed APP-x estimates on average by 11% in spring, but the difference is nearly constant from year to year and does not affect the trend calculations. Mean cloud fractions over sea ice retrieved by JC, however, are substantially lower than those for APP and TPP, except during spring where they more closely match APP-x values, and they exhibit virtually no annual cycle.

Table 2. Seasonal Mean of Cloud Fraction From the TPP, APP-x Data Sets and JC
 WinterSpringSummerFall
PATHP76778184
APPX74668180
JC (sea ice)67.966.065.266.6

[11] Although decadal trends in winter cloud amounts in TPP and APP-x are very similar and show a significant decrease, it should be noted that shorter term variability between the two is not well correlated. Winter cloud retrievals are more uncertain because frequent near-surface temperature inversions, the uselessness of visible channels, and frequent occurrence of optically thin clouds complicate cloud detection. The difference in spring cloud trends between JC and WK is difficult to explain. Both employ a series of similar spectral and temporal tests involving AVHRR thermal (3–5) and visible (1,2) channels though algorithm details and thresholds differ. One possible explanation is that WK use AVHRR composites centered on a 14:00 local time to maximize the use of more reliable daylight retrievals. JC in contrast uses all orbits. The fact that spring cloud trends from the APP-x based on 04:00 time composites are slightly negative supports this explanation. More detailed comparisons are needed to identify the source(s) of the differences.

5. Influence of Changes in Observing System

[12] Satellite sensors providing data for this study were designed for operational purposes, and were not necessarily calibrated to the degree required for climate research. Climate applications are complicated by the possibility that variations introduced by changes in the observing system, for example owing to changes in sensor design, calibration, or orbital parameters, may be interpreted as climate variations. Calibration issues related to AVHRR were addressed by WK and JC. Evidence of calibration problems in the TPP data set have been discussed by Overland et al. [2002] and Chen et al. [2002]. To test the sensitivity of our results to possible changes in the observing system, we conducted the following experiment. First, seasonally averaged cloud fractions were normalized to the NOAA-10 period (1987 to 1991) by subtracting the mean annual cloud fraction over the lifetime of each satellite from the mean annual cloud fraction for the lifetime of NOAA-10. Annual means over the entire TPP domain (>60°N) were used for normalization, as calibration differences are expected to be seasonally and spatially invariant. These differences were then applied to the seasonal means corresponding to each satellite period. Seasonal trends were then recomputed. Normalized TPP trends for winter and summer are −4.3 %/decade and 4.4 %/decade, respectively, and within 0.3 %/decade of the unnormalized estimates. This is not surprising, as the retrieval of cloud fractions in the TPP algorithm is based on a variety of channel differences [Francis, 1994] and most calibration issues are correlated across channels. Trends in cloud fraction, consequently, are less sensitive to calibration errors than are variables whose retrieval is based on single channels or combinations of channels. Based on this sensitivity study, we conclude that calibration uncertainty does not contribute significantly to the uncertainty in the observed cloud trends.

6. Spatial Trends

[13] An analysis of TPP cloud fractions suggest that significant changes have occurred over the past two decades. Regionally these changes are even more striking. Figure 2 presents a map of spring trends in TPP cloud fraction over the Arctic seas. Trends are largest in the central Arctic near the North Pole, where trends reach 16 %/decade. Over the Beaufort, Chukchi and East Siberian Seas cloud cover has increased as much as 10 %/decade. Winter trends are shown in Figure 3. Substantial, large-scale decreases are apparent, particularly in the Barents and Kara Seas, where the reduction exceeds 12 %/decade. Increased cloud fractions occurred over the Beaufort and Chukchi Seas. Patterns derived from APP-x are similar (not shown), although increases over the central Arctic are somewhat smaller (less than 10%) during spring and decreases are more pronounced and widespread during winter.

Figure 2.

Linear trend in cloud amount in units of %/decade during spring from the TPP for the period 1980–1998. Contours indicate the 99% significance levels for individual grid points based on t-statistics.

Figure 3.

Linear trend in cloud fraction from the TPP during winter (DJF) for the period 1980–2001.

[14] Processes involved in cloud formation are complex, and a thorough investigation into the many possible causes of the observed change is beyond the scope of this study. Additional corroborating evidence, however, is suggestive of plausible explanations. Figure 4 displays the change in sea level pressure (NCEP Reanalysis) over the same time period. A substantial decrease is apparent over the arctic seas, consistent with the observed shift in the atmospheric circulation to a more positive state of the Northern Annual Mode (NAM) [Thompson et al., 2000]. In fact, spatial patterns of spring cloudiness from TPP are strongly correlated with the NAM (not shown). Cyclonic activity in the Arctic also has increased during the 1990s [Zhang et al., 2004], providing further evidence of strong synoptic influence on spring cloud changes. An analysis of TOVS-derived, total-column precipitable water suggests that the atmospheric moisture content has also increased markedly (not shown).

Figure 4.

Linear trend in surface pressure in units of mb/decade during Spring from the NCEP reanalysis project for the period 1980–1998. Contours indicate 95 and 99% significance levels for individual grid points.

[15] The causes of winter decreases in cloud amount are less intuitive than for spring. A stronger polar vortex during the 90ies with a dominant positive phase of the NAM may have prevented the entry of storms into the Central Arctic. However, spatial patterns in cloud amount are not significantly correlated with the NAM, suggesting more complex mechanisms for the winter decrease in clouds. Another cause maybe a decrease in temperature [WK], leading to more efficient precipitation of cloud particles. This process is thought to be one of the principal drivers of the seasonal cycle of Arctic clouds [Beesley and Moritz, 1999].

7. Trends in Surface Observations

[16] The North Pole (NP) drifting stations operated by the former Soviet Union provide a substantial record of surface meteorological observations [Lindsay, 1998]. Unfortunately, the NP data record ends in 1991 and only partially overlaps with the satellite record. However, analyses of these measurements during the overlap supports satellite-observed increases in spring cloud fraction. Figure 5 compares seasonally averaged, spring cloud fraction from NP stations during 1955–1991 with TOVS retrievals for the same times and locations. The time series shows a dramatic increase of nearly 15% in cloudiness over the satellite overlap period from 1980 to 1991. The start of the satellite period also coincides with the minimum in the time series, which changes relatively little before 1978. This increase in spring time Arctic cloudiness was previously noted by Makshtas et al. [1999], who speculated that the increase in cloud condensation nuclei through pollution may have contributed to this increase. Thus, NP observations further corroborate evidence of a substantial increase in spring cloudiness revealed by two independent satellite data sets, but also clearly suggest that the trend is critically dependent on the interval analyzed.

Figure 5.

Time series of seasonally averaged spring cloud fraction from the NP drifting stations and corresponding TPP retrievals. Seasonal averages are computed from daily averages for all stations in a given time period (usually 1 or 2). TPP averages are computed from daily retrievals for the grid cell closest to the location of the surface station.

8. Summary and Conclusions

[17] Retrievals from APP and TPP show a strong increase (∼5 %/decade) in cloudiness during spring over the past two decades. A similar increase is found in meteorological surface observations from the NP drifting stations. The increase in spring cloudiness appears associated with a drop in surface pressure and is consistent with an increase in cyclonic activity reported over the same period. The agreement of these satellite data sets based on two independent and different satellite instruments and algorithms, as well as patterns of change in other physically related fields, leads us to believe that these increases are indeed real and constitute a significant climate variation that began near 1980. For the winter season, all three satellite data sets show a decrease in cloudiness of 5 %/decade.

[18] The identification of the causes of these variations presents an opportunity for future analysis, particularly with respect to modeling. Our analysis shows that the ERA-40 reanalysis does not reproduce the observed spring increase in total cloud amount, though the winter decrease is present but somewhat weaker. Further, vast differences exist among global climate models in their projected response of Arctic cloud fraction to doubling CO2: they disagree in sign as well as magnitude [Holland and Bitz, 2003].

[19] Cloud changes may have a strong influence on the surface radiation balance, which in turn affects sea ice melt and growth. During winter cloud forcing is always positive (i.e., cloud warm the surface via increased longwave emission) so the decrease in cloudiness would contribute to greater sea ice growth. During spring short wave and longwave effects on the net radiation balance have opposite signs and changes tend to cancel each other [WK], thus spring changes may have little effect on sea ice. WK also find that on an annual basis, Arctic clouds have a net cooling effect.

Acknowledgments

[20] This work was supported my NASA grants NAG5-11800 and NNG04GH52G and NOAA grant NA17RJ1232, A21. Ron Lindsay and Jennifer Francis are thanked for helpful suggestions. Xuanji Wang for providing APP-x data.

Ancillary