Journal of Geophysical Research: Atmospheres

Implications for Arctic amplification of changes in the strength of the water vapor feedback

Authors

  • Debjani Ghatak,

    Corresponding author
    1. Institute of Marine and Coastal Sciences, Rutgers, The State University of New Jersey, New Brunswick, New Jersey, USA
    • Corresponding author: D. Ghatak, Institute of Marine and Coastal Sciences, Rutgers, The State University of New Jersey, New Brunswick, NJ, USA. (ghatak@marine.rutgers.edu)

    Search for more papers by this author
  • James Miller

    1. Institute of Marine and Coastal Sciences, Rutgers, The State University of New Jersey, New Brunswick, New Jersey, USA
    Search for more papers by this author

Abstract

[1] One of the major climatic changes apparent over the Arctic Ocean has been the amplified rate at which air temperature has been increasing relative to the global mean. There are multiple factors which play roles in this amplification, including changes in sea ice/albedo, atmospheric circulation, clouds, and water vapor. We investigate the positive feedback on temperature caused by increasing downward longwave radiation flux (DLF) associated with increasing atmospheric precipitable water (PW). The Japanese 25-year Reanalysis and ERA-Interim reanalysis are used to examine the role of the DLF/PW component of the water vapor feedback loop on the enhanced warming in the Arctic between 1979 and 2011. We find a nonlinear relationship between DLF and PW, which suggests that the sensitivity of DLF to changes in PW varies by season, with the highest in winter and the lowest in summer. The positive trends in DLF and PW are widespread over the Arctic during autumn and spring but are centered mainly over the Atlantic sector in winter. The strength of the PW feedback loop depends on both the sensitivity of DLF to changes in PW and the change in PW during 1979–2011. If, in the future, PW were to increase significantly during winter in the central and Pacific sectors of the Arctic, there could be an expansion of Arctic amplification during winter. We also examine the effect of changes in cloud cover and find that such changes account for a much smaller proportion of the changes in DLF than does PW.

1 Introduction

[2] During recent decades, amplified warming rates have been observed over the Arctic Ocean [Solomon et al., 2007; Serreze et al., 2009]. This amplification, in conjunction with the relatively rapid recent loss of summer sea ice, appears to be occurring in response to several factors, including sea ice/albedo, water vapor/radiation, and cloud/radiation feedback loops. This amplification of the warming rate in the Arctic is particularly pronounced during seasons when solar radiation is small or absent [Serreze et al., 2009; Screen and Simmonds, 2010; Orsolini et al., 2011]. The net heat gain within the ice-ocean column is enhanced during summer months owing to the increasing open water areas associated with the sea ice loss, which leads to enhanced heat fluxes from ocean to atmosphere during autumn; this plays a major role in the amplification [Serreze et al., 2009; Screen and Simmonds, 2010]. Serreze et al. [2011] concluded that the recent positive lower tropospheric air temperature anomalies over the Arctic are caused by a combination of many factors (e.g., general warming associated with greenhouse gas increases, changes in atmospheric circulation, reduced sea ice extent, and higher sea surface temperatures), and they also acknowledged the potential influence of feedbacks associated with changes in clouds and water vapor.

[3] Several studies have emphasized the role of water vapor and clouds in warming the Arctic atmosphere [e.g., Winton, 2006; Miller et al., 2007; Lu and Cai, 2009; Graversen and Wang, 2009; Screen and Simmonds, 2010; Chen et al., 2011]. Francis and Hunter [2007] used satellite retrievals in the Arctic and identified water vapor and cloud cover as the two primary factors that influence the downward longwave radiation flux (DLF). The DLF increases in response to increases in atmospheric water vapor. There is a tendency for the same response to increasing clouds or cloud optical thickness. Changes in cloud base height can produce mixed responses in DLF. For low clouds during winter when inversions are common in the Arctic, increasing cloud base height leads to increased DLF [Chen et al., 2006]. Changes in precipitable water (PW) contribute to the trend in DLF more than changes in the cloud fraction when clouds are composed of ice particles, whereas changes in both precipitable water and clouds play important roles in driving the trend in DLF when clouds are composed of liquid water droplets [Francis and Hunter, 2007].

[4] Chen et al. [2011] used global climate model simulations to examine the seasonal sensitivity of DLF to changes in PW in the Arctic, and they found that for the present climate, the sensitivity is highest in winter and smallest in summer. Their analysis and that of Miller et al. [2007] also showed that during winter in the Arctic, the sensitivity of DLF to changes in PW will decrease during the 21st century. Both of these studies focused on the central Arctic at the Surface Heat Budget of the Arctic project (SHEBA) site in the Beaufort Sea where extensive observations were available. Both the short-term (seasonal) and long-term changes in sensitivity occur because the response of DLF to changes in PW is nonlinear, with high sensitivities when PW is low and lower sensitivity when PW is higher.

[5] The positive feedback loop on temperature associated with water vapor is shown in Figure 1. An increase in surface air temperature produces an increase in atmospheric water vapor which, in turn, increases the DLF and a subsequent increase in the surface air temperature. In this study, we do not examine the first leg of the feedback loop (Figure 1) because apportioning the increase in water vapor to local effects or larger-scale effects is a difficult problem in its own right [Trenberth, 1999]. The third leg (Figure 1) should have a double-pointed arrow because increases in temperature will enhance the DLF, providing an additional positive feedback on DLF. Our focus here is on the second leg of the feedback loop (Figure 1), which is the response of DLF to changes in PW. The focus on PW is a simplification because the vertical distribution of specific humidity can also be very important.

Figure 1.

Feedback diagram shows the positive feedback loop between temperature, PW, and DLF. Positive sign means that two variables change in the same direction.

[6] The focus of this paper is on the role of the PW feedback on the recent enhanced atmospheric warming in the Arctic region during non-summer seasons. We extend the works of Miller et al. [2007] and Chen et al. [2011] to the whole Arctic Ocean by using the Japanese 25-year Reanalysis (JRA 25) and ERA-Interim (ERA-I) reanalyses to examine how the sensitivity of DLF to changes in PW varies across the Arctic Ocean by region and by season. We also provide an estimate of the strength of the DLF/PW component of the water vapor feedback loop between 1979 and 2011 and discuss the significance of the water vapor feedback in the Arctic atmosphere. In addition, we examine the potential impact of changes in cloud cover. We emphasize that this study is based on climate variables from reanalysis products that may be mostly model based in data sparse regions such as the Arctic. Serreze et al. [2012] suggest that the results from any reanalysis product should be considered with caution as there is considerable variability in trends in precipitable water as well as in vertical structure of specific humidity among these products during recent years. There is, however, internal consistency in the data that allows us to examine relationships among climate variables. We discuss our methodology in section 2, present our results in sections 3 and 4, and summarize our conclusions in section 5.

2 Methodology

[7] The Japanese reanalysis, known as JRA 25, is a global reanalysis product which has 40 vertical layers and is available up to the present time. It is compiled using the Japan Meteorological Agency numerical assimilation and forecast system [Onogi et al., 2007] and has been used to study both global and Arctic climates [e.g., Simmonds and Keay, 2009; Serreze et al., 2009; Stroeve et al., 2011; Simmons et al., 2010]. A three-dimensional variational method has been used to assimilate the precipitable water from orbital satellite microwave radiometer radiance and other satellite data [Onogi et al., 2007]. Serreze et al. [2011] used multiple reanalyses products including JRA 25 as well as radiosonde data to examine the changes in tropospheric water vapor and showed that relative to other reanalyses, specific humidity from JRA 25 has one of the lowest biases at 1000 hPa when compared to radiosonde data.

[8] ERA-I is an advanced version of the ERA 40 reanalysis produced at the European Centre for Medium-Range Weather Forecasts [Dee et al., 2011]. This reanalysis made significant advancements over the ERA 40 reanalysis by improving stratospheric circulation, the hydrologic cycle, and temporal consistency [Dee et al., 2011]. ERA-I uses four-dimensional variational analysis in the atmospheric analysis and hence leads to more effective use of observations [Dee et al., 2011]. Another important advancement in ERA-I is the introduction of Variational Bias Correction as part of the analysis cycle to estimate the parameters [Dee et al., 2011]. A study over the Mississippi River shows that ERA-I cloud cover matches observations better than ERA 40 [Betts et al., 2009]. Furthermore, the ERA-I reanalysis improves the representation of the Arctic temperature trend [Screen and Simmonds, 2010] and of the global hydrological cycle [Simmons et al., 2006, 2010].

[9] In this study, we present results based on our analyses of three variables at 2.5° horizontal resolution, i.e., DLF (W/m2), PW (kg/m2) for the whole atmospheric column, and cloud cover (percentage) between 1979 and 2011. We investigate the sensitivity of DLF to changes in PW for autumn (September, October, and November, SON), winter (December, January, and February, DJF), and spring (March, April, and May, MAM). We employ linear trend analysis of DLF, PW, and cloud cover at each grid cell and assess statistical significance of the regression coefficients at 5% level of significance, using Student's t test [Helsel and Hirsch, 1992].

3 Analysis of Changes in and Relationships Between DLF and PW

[10] In this section, we first examine the sensitivity of DLF to changes in PW, then discuss the trends in PW and DLF during the 3 decade period, and then combine them to estimate what fraction of the DLF trend can be accounted for by changes in PW. As noted before, there are several global reanalyses that are available for such studies. To check the consistency between two different reanalyses, we compare the DLF and PW fields from the JRA 25 and ERA-I reanalyses in this section.

3.1 Sensitivity of DLF to Changes in PW

[11] The relationship between DLF and PW is clearly nonlinear for the Arctic region, as shown in Figure 2. As noted in other studies [e.g., Zhang et al., 2001; Naud et al., 2012], this nonlinearity occurs outside the Arctic, too. However, in regions that are cold and dry, such as the Arctic, points on the curve move toward low values of q where the slope is steepest and the nonlinearity becomes more prominent. Figure 2 shows monthly values of DLF and PW averaged over the area north of 70°N for both the JRA 25 and ERA-I reanalyses as well as the average of the monthly values. Winter points are found on the lower left part of the curve where the slope (sensitivity) is highest, and the slope is steeper for ERA-I than for JRA 25. Summer points, when PW is highest, are found on the upper right part of the curve where the slope (sensitivity) is smallest. This occurs because the DLF-water vapor relationship saturates, with any additional water vapor having little effect on the longwave radiation emitted. As expected, spring and autumn points on the curve are found between those for summer and winter. Mean seasonal sensitivities for the region north of 70°N based on the slopes of the curves in Figure 2 are 18 W/kg, 22 W/kg, and 17 W/kg for autumn, winter, and spring, respectively, for JRA 25 and 18 W/kg, 31 W/kg, and 21 W/kg for ERA-I. Except for winter, the sensitivities are in good agreement between the two reanalyses. The sensitivities that we calculate here are likely upper bounds on the actual sensitivity because we are not considering the third leg of the feedback loop in Figure 1, and in particular, we are not considering the impact of the enhanced temperature on DLF.

Figure 2.

Relationship between DLF (W/m2) and PW (kg/m2) based on their monthly averages north of 70°N for the 33 year record. Black diamonds denote each month of the record. Colored diamonds are seasonal averages with red diamonds for winter (DJF), yellow for spring (MAM), green for summer (JJA), and blue for autumn (SON). The curve on the top is based on ERA-I and the bottom one is based on JRA 25 reanalysis.

[12] In addition to the seasonal variability of the DLF/PW sensitivity, there is also spatial variability. To calculate the sensitivity of DLF to changes in PW in Figure 3, we calculate the daily differences for DLF and PW for each month of the entire 33 year record. Here the daily difference of a variable indicates the change in the variable from one day to the next. This methodology is consistent with Chen et al. [2003] and provides an opportunity for comparison. Furthermore, by using consecutive daily differences, we remove local linear trends that may exist in the data sets. We obtain the sensitivity at each grid cell for a particular month by calculating ordinary linear regression coefficients between the daily differences of DLF and PW using all the daily differences in the entire 33 year record. The seasonal sensitivity is obtained by then averaging the monthly sensitivities.

Figure 3.

Sensitivity of DLF to changes in PW for the Arctic basin for autumn, winter, and spring based on (a–c) JRA 25 and (d–f) ERA-I reanalyses. Please see section 3 for the details of the methodology.

[13] Figure 3 shows the sensitivity of DLF to PW at each grid cell for three seasons, autumn, winter, and spring based on both JRA 25 and ERA-I. The magnitudes are consistent with those from the previous studies of Miller et al. [2007] and Chen et al. [2011] for the SHEBA site in the central Arctic. As expected from Figure 2, the sensitivities are highest during winter and somewhat lower in autumn and spring. Although not shown, they are also very low in summer when PW reaches its maximum because the nonlinearity of Figure 2 means that summer points are found on the curve where slopes are smallest and the DLF-water vapor relationship saturates. The very high cloud cover in summer is likely an additional factor for the low sensitivity to PW in summer [Chen et al., 2011]. Chen et al. [2011] also found that the sensitivity is lowest during summer; therefore, we do not include an analysis of the summer season here since we are primarily concerned with the feedback loop during the seasons when solar radiation is small and the observed warming is greatest. Therefore, the impact of the DLF/PW feedback is likely to be more important during non-summer months. There is considerable spatial variability in sensitivity during autumn, winter, and spring with the highest sensitivities centered mainly over the central Arctic Ocean, the Canadian Archipelago, and Greenland both in JRA 25 and in ERA-I. High-sensitivity areas extend over the Eurasian landmasses during winter. The DLF/PW sensitivity is higher during winter and spring in ERA-I than in JRA 25. During all seasons, regions of high sensitivity tend to coincide with regions where the atmosphere is dry, which is consistent with Figure 2. Although not shown, we also examined the seasonal climatology of PW for the same 33 year time period from both of these reanalyses and found a similar result (i.e., a tendency for higher sensitivities in regions where mean PW was lower).

3.2 Changes in DLF and PW

[14] Next, we examine how DLF changes over the Arctic during the 33 year period based on both JRA 25 and ERA-I reanalyses. Figure 4 shows Hovmoller plots of the average DLF anomaly north of 70°N, where anomalously high DLF is apparent in recent years particularly during autumn, winter, and spring with an exception of spring 2008–2009. Although not shown, the time evolution of the monthly anomaly pattern in PW is quite similar to that of DLF. However, there is high spatial and inter-seasonal variability in the DLF trend, with both positive and negative trends. The trends are obtained using linear regression analyses for the 33 year period for both the PW and DLF fields, and they exhibit similar patterns (Figures 5 and 6). Significant positive trends in PW are centered over the North Atlantic and Chukchi, East Siberian, Greenland, and Norwegian Seas during autumn (Figures 5a and 5d). The patterns of DLF and PW trends are similar during autumn, although the significant positive trend in DLF extends over a much broader region including the Laptev and Beaufort Seas as well as the central Arctic Ocean (Figures 6a and 6d). During winter, there are significant positive trends in PW and DLF over the Atlantic sector of the Arctic, particularly over the Greenland Sea and north of the Barents Sea, but trends are negative over much of the Pacific sector, with significant negative trends over East Asia and over the East Siberian Sea (Figures 5b, 5e, 6b, and 6e). During spring, significant positive trends in PW and DLF spread over most of the Arctic basin with the strongest positive trends in DLF over the marginal seas along the Eurasian coast (Figures 5c, 5f, 6c, and 6f).

Figure 4.

Hovmoller plots showing the time evolution of the monthly anomaly of DLF (W/m2) for the Arctic north of 70°N based on (a) JRA 25 and (b) ERA-I reanalyses. Anomalies are calculated with respect to the monthly average DLF for the 1979–2011 period.

Figure 5.

Seasonal trends in PW for the Arctic basin for the 33 year record of the (a–c) JRA 25 and (d–f) ERA-I reanalyses. Trends significant at the 95% confidence level are shown by the white contours and shading.

Figure 6.

Seasonal trends in DLF for the Arctic basin for the 33 year record of the (a–c) JRA 25 and (d–f) ERA-I reanalyses. Trends significant at the 95% confidence level are shown by the white contours and shading.

3.3 Contribution of PW to the Changes in DLF

[15] The magnitude of the DLF/PW component of the temperature/water vapor feedback loop during a period of time is the product of two factors: the sensitivity of DLF to changes in PW and the change in PW during the period. Hence, even if the sensitivity of DLF to changes in PW is high, the strength of this leg of the feedback loop would be zero if PW is unchanged during the period of interest. We next calculate “estimated trends” in DLF by multiplying the sensitivity of DLF to PW (as in Figure 3) by the change in PW during the 33 year period. The change in PW is calculated as in Figure 5 for 1979–2011. Figure 7 shows the percentage of the trend in DLF (original field) explained by the estimated trend but only where the trend in the original DLF field is significant at the 95% confidence level. Furthermore, we show the region where the estimated trend is of opposite sign from that of the trend in the original DLF field. There is some circularity in obtaining the estimated trend because we first use the reanalysis to calculate sensitivities and then use the reanalysis to obtain the long-term trends in PW. However, the use of about 1000 daily differences to calculate regression slopes for each month not only makes the long-term derived estimates of sensitivity more reliable but also reduces some of the noise caused by simultaneous variations in other climate variables.

Figure 7.

Ratio of the estimated trend (see text) in DLF based on changes in PW and the trend in the original DLF field based on (a–c) JRA 25 and (d–f) ERA-I reanalyses. Ratios are plotted only where the trend in the original DLF field is significant at the 95% confidence level. Stippled gray contour indicates where the trend in DLF estimated from PW is of opposite sign to the trend in the original DLF field.

[16] Figure 7 indicates that the role of PW in explaining the trends in DLF varies spatially and seasonally in both JRA 25 and ERA-I. It is important to note that the magnitudes of the DLF/PW sensitivities are likely to be upper bounds because they also include a component of the DLF increase owing to increased temperature. Thus, the percentages in Figure 7 relating the estimated trend to the “original trend” are likely to be upper bounds as well. During autumn, increases in PW contribute to less than half of the increase in DLF over much of the central Arctic and Pacific sector, whereas increases in PW explain about twice as much of the increase in DLF over much of the Atlantic sector (Figures 7a and 7d). As the seasons change from autumn to spring, there is a tendency for the percentage of DLF trends accounted for by changes in PW to increase from the North Atlantic region in autumn toward the Pacific sector of the central Arctic in spring (Figure 7).

4 Potential Impact of Changes in Cloud Cover

[17] In this section, we examine the potential impact of changes in clouds. Although changes in various cloud properties could have an impact (e.g., cloud cover, cloud base height, optical depth, and ice water content), the focus here is on changes in cloud cover. Since the changes in cloud cover are very small for the ERA-I reanalysis throughout the Arctic during our period of interest, the discussion in this section refers to only the JRA 25 reanalysis. Changes in the JRA 25 cloud cover are generally small over the Arctic during all non-summer seasons between 1979 and 2011 (Figures 8a–8c). However, there are significant positive trends in cloud cover near the Eurasian coast during autumn (Figure 8a) and over the Barents Sea, Kara Sea, and central Arctic Ocean during spring (Figure 8c). There are negative trends in cloud cover over the Canadian archipelagos during winter and spring and also over western Siberia during spring. These results indicate that the spatial and temporal variability of DLF and PW is similar (compare Figures 5 and 6); this is not generally the case for DLF and cloud cover (Figures 6 and 8), although their changes are similar in some places in autumn and spring, particularly near the Eurasian coast.

Figure 8.

(a–c) Seasonal trends as in Figures 5 and 6 but for cloud cover from JRA 25 reanalysis. Sensitivities of DLF to changes in PW for (d) autumn, (e) winter, and (f) spring but only using pairs of days when cloud cover changes by less than 25% from one day to the next based on JRA 25 reanalysis. (g–i) Similar to Figure 7 but for the ratio between the estimated trend in DLF based on changes in cloud cover and the trend in the original DLF field.

[18] One way to examine the potential role of cloud cover on the DLF/PW feedback is to investigate how changes in cloud cover alter the sensitivity of DLF to changes in PW shown in Figure 3. Figures 8d–8f show the sensitivities of DLF to changes in PW as in Figure 3 except only using pairs of days when cloud cover changes by less than 25% from one day to the next based on JRA 25 reanalysis. Although this reduces the number of data points used in the regression analysis, it does remove the effect of large day-to-day changes in cloud cover on the DLF/PW sensitivities. Removing the effect of clouds reduces the sensitivity of DLF to PW. When averaged over the region north of 70°N, this reduces the sensitivities by relatively small amounts, 21%, 12%, and 18% in autumn, winter, and spring, respectively, with the highest sensitivities still occurring in winter. Although not shown, we did the same analysis with the ERA-I reanalysis and found even smaller impacts of clouds on the DLF/PW sensitivities.

[19] We next examine the potential contributions of cloud cover changes to the trends in DLF based on the JRA 25 reanalysis. We use the same methodology as in section 3.3 to calculate the sensitivity of DLF to changes in cloud cover by replacing the daily changes in PW with the daily changes in cloud cover. Then we calculate the estimated trends in DLF caused by changes in cloud cover by multiplying the sensitivity of DLF to cloud cover by the change in cloud cover during the 33 year period. The change in cloud cover is calculated as in Figures 8a–8c for 1979–2011. The percentage of the trend in DLF (original field) explained by the estimated trend is shown in Figures 8g–8i. Increases in cloud cover generally contribute to less than 30% of the increase in DLF in both fall and spring; they contribute almost nothing in winter. There are also several large regions where cloud cover fails to explain the sign of the trend in DLF (original field) with this occurring to the maximum extent in winter (Figures 8g–8i). This means that for such regions, clouds have been increasing (decreasing) while DLF has been decreasing (increasing), and other factors must be responsible for the bulk of the DLF changes.

5 Discussion and Conclusions

[20] In this paper, we investigate the potential role and significance of the DLF/PW feedback on the amplified warming over the Arctic Ocean during the last 3 decades. We have used the JRA 25 and ERA-I reanalyses to analyze the spatial and seasonal variability of the sensitivity of DLF to PW, as well as the long-term trends in PW and DLF, to better understand the role and strength of the PW feedback. We also examine the potential impact of changes in cloud cover on the results.

[21] Consistent with previous studies [e.g., Miller et al., 2007; Chen et al., 2011], there is a nonlinear relationship between DLF and PW in the Arctic north of 70°N. Zhang et al. [2001] found that the DLF/water vapor feedback is important over the snow surface in Alaska. Other recent studies have suggested that this same nonlinear relationship between DLF and water vapor is important for winter warming in high-elevation regions where the sensitivity of DLF to water vapor is similar to that found in polar regions [Ruckstuhl et al., 2007; Rangwala, 2012; Rangwala et al., 2009; Naud et al., 2012].

[22] Since the DLF/PW sensitivity is highest during winter and lowest during summer and since Arctic amplification is most prominent during non-summer seasons, the focus here is on the autumn, winter, and spring seasons when solar radiation is absent or small. High-sensitivity regions correspond well with areas where the atmosphere is climatologically dry and cold, such as over the central Arctic Ocean, the Canadian Archipelago, and Greenland. The magnitudes of the sensitivities are consistent with those from previous studies at the SHEBA site in the central Arctic [e.g., Miller et al., 2007; Chen et al., 2011].

[23] The JRA 25 and ERA-I reanalyses indicate that DLF has been anomalously high over the Arctic during the last decade. However, there is considerable regional and seasonal variability in the DLF trends. For example, there are significant positive trends over the Beaufort, Chukchi, Laptev, and East Siberian Seas during autumn. These are the same regions where the recent sea ice anomalies have been most negative [Serreze et al., 2007]. During winter, these regions show mostly negative and insignificant trends. During both autumn and winter, there is a large positive trend in DLF in the Atlantic sector of the Arctic. Thus, the anomalously high DLF over the Arctic in winter during recent years (see Figure 4) occurs because the large positive trend in DLF over the Atlantic sector dominates the smaller decreasing trends in DLF that occur in the Pacific sector of the Arctic.

[24] During autumn, the sensitivity of DLF to PW is large in the central Arctic and low to the east of Greenland. However, DLF is increasing in both regions, and increases in PW account for most of the increase in DLF in the low-sensitivity region to the east of Greenland and less than half of the increase in DLF in the higher-sensitivity region north of 75°N, except over Greenland. These results from JRA 25 and ERA-I reanalyses suggest that there are variables other than PW, such as clouds, atmospheric circulation, surface air temperature, and local sensible and latent heat fluxes, that affect the changes in DLF. For example, amplified sensible and latent heat fluxes over the open water area in the Pacific sector can increase DLF by increasing the atmospheric temperature and by increasing the amount of water vapor near the surface relatively more than PW.

[25] These results from JRA 25 and ERA-I reanalyses for autumn are generally consistent with those of Screen and Simmonds [2010] who used the ERA-Interim reanalysis and found evidence that increasing PW in autumn was correlated with diminishing sea ice cover. Increased PW based on JRA 25 and ERA-I reanalyses during winter over the Atlantic sector of the Arctic coincides with the strong winter warming identified by Serreze et al. [2011]. Increased atmospheric water vapor over this region may be caused by a greater atmospheric influx of water vapor from the Atlantic or from enhanced vertical fluxes locally [Serreze et al., 2011]. The spring trend in DLF based on both JRA 25 and ERA-I reanalyses is consistent with that derived from The Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder [Francis and Hunter, 2007].

[26] We examined the potential role of cloud cover changes in contributing to the changes in DLF during the 33 year period based on JRA 25 reanalysis and found that they contributed to a relatively small part of the changes in DLF. We also found that the sensitivities of DLF to changes in PW were reduced when the simultaneous effects of cloud cover changes were removed. Changes in cloud cover based on ERA-I reanalysis are generally negligible and hence make almost no contribution to the enhanced Arctic warming during the 33 year period, a result that is consistent with Screen and Simmonds [2010] who also used the ERA-I reanalysis.

[27] Although the winter sensitivity of DLF to PW is highest over the central Arctic and over the adjacent landmasses, the trends in DLF during winter are positive over the Atlantic sector of the Arctic and negative over part of the Pacific sector and over East Asia based on both JRA 25 and ERA-I reanalyses. The difference between the spatial pattern of sensitivity and the spatial pattern of DLF trends emphasizes the need to examine both factors in the PW component of the feedback loop, namely, the sensitivity of DLF to changes in PW as well as the changes in PW during the 3 decade period. Miller et al. [2007] and Chen et al. [2011] have shown that during Arctic winter, the projected sensitivity of DLF to changes in PW will become smaller during this century as PW increases, thus moving points on Figure 2 to the right where the slopes/sensitivities are lower. However, one of the implications of our study here is that increases in DLF could amplify winter warming in the high-sensitivity Arctic regions if PW begins to increase fast enough to counteract the corresponding decreases in sensitivity. Furthermore, climate model projections indicate that the downward trend in sea ice will continue through the 21st century [Holland et al., 2006; Wang and Overland, 2009]. The delay in ice formation and thickening means that there will be more open water areas during autumn and early winter, a condition that is favorable for adding more water vapor and heat directly into the atmosphere locally in addition to potentially modifying atmospheric circulation to transport in more heat and water vapor from lower latitudes. Thus, these projected surface and atmospheric conditions, in conjunction with our sensitivity analyses, suggest that the impact of the water vapor feedback on Arctic amplification during winter, already present in the Atlantic sector of the Arctic, could spread into the central and Pacific sectors in the future. A similar case could be made for increasing cloud cover during colder months.

[28] The focus here is on the spatial and seasonal variability of the role and strength of the DLF/PW feedback loop; however, there are other variables that cause changes in DLF, such as advection of heat and moisture, cloud properties, and temperature [Francis and Hunter, 2007; Serreze et al., 2011]. Circulation changes may bring in warm, moist air from lower latitudes, which can trigger the positive enhancement associated with DLF through both increased temperature and increased water vapor. Further investigation is needed to examine the role of other variables on changes in DLF. The most difficult variable to separate from changes in water vapor is temperature. Higher atmospheric temperatures are generally correlated with higher PW, and temperature and water vapor are closely coupled in the Arctic [Zhang et al., 2001]. Hence, the sensitivities of DLF to changes in PW that we have calculated here are likely upper bounds on these sensitivities. The sensitivities would likely be smaller if we could remove the effects of changes in temperature, although Zhang et al. [2001] did find that water vapor has a greater impact than temperature on snowmelt in Alaska. In contrast, Sedlar and Devasthale [2012] suggest that between 2003 and 2010, the temperature anomalies have a greater impact on the downwelling longwave anomalies than do the water vapor anomalies in the East Siberian and Laptev Seas.

[29] The results here are based on recent reanalysis products, JRA 25 and ERA-I. We have emphasized that the results are model dependent because observations are sparse over much of the Arctic Ocean. Both the JRA 25 and ERA-I reanalyses show increasing trends in DLF and PW during the 33 year period and DLF/PW sensitivities that are high during winter and low during summer. The spatial patterns of changes in DLF and PW are generally consistent between the two reanalyses which gives us additional confidence in the results here, although the ERA-I sensitivities are somewhat larger than those for JRA 25, particularly in winter. Beyond that, the strength of the DLF/PW component of the water vapor feedback loop for a particular reanalysis will depend on both the magnitude of the sensitivity of DLF to PW and how PW varies spatially and seasonally during a particular period of time.

Acknowledgments

[30] Debjani Ghatak is grateful to the Rutgers Institute of Marine and Coastal Sciences for the postdoctoral support, as well as to NASA MEaSUREs award NNX08AP34A. James Miller has received partial support from project 32103 of the New Jersey Agricultural Experiment Station. We thank Imtiaz Rangwala for helping in the acquisition of the JRA 25 data. We also thank Maike Ahlgrimm, Yonghua Chen, Dick Dee, Jennifer Francis, Elias Hunter, Imtiaz Rangwala, David Robinson, Julienne Stroeve, Wei Wu, and three anonymous reviewers for the helpful discussions and comments.