Comment on “Solar influences on cosmic rays and cloud formation: A reassessment” by Bomin Sun and Raymond S. Bradley

Authors


1. Introduction

[1] Svensmark and Friis-Christensen [1997] (hereinafter referred to as SFC) proposed a cosmic ray–cloud cover link on the basis of five different satellite data sets (including the International Satellite Cloud Climatology Project (ISCCP)) of total cloud cover between 1979 and 1996. Marsh and Svensmark [2000] (hereinafter referred to as MS00), using the updated ISCCP satellite observations, found this to be limited to low clouds between 1983 and 1994 and suggested that this was consistent with a mechanism involving ionization and aerosol nucleation in the lower troposphere.

[2] Sun and Bradley [2002] (hereinafter referred to as SB02) present their interpretation of the same ISCCP satellite observations and compare them with surface- and ship-based observations, the latter covering the more extensive period 1953–1995. They find no evidence to support the hypothesis of a link between cosmic ray and total cloud cover over land or oceans, nor do they accept that there is a solid basis for a relationship between cosmic rays and low cloud. Their findings are based on the following: (1) surface-based observations of total cloud over land are not correlated with cosmic rays, (2) the correlation between cosmic rays and total cloud (SFC) and low cloud (MS00) are limited to the Atlantic Ocean and restricted to the period before 1991, (3) there is a lack of evidence to suggest that the cosmic ray–low cloud correlation is physically based, (4) the increasing long-term trend of ship observations between 1953 and 1995 is opposite to that expected if a cosmic ray–cloud link exists, and (5) changes in higher cloud cause apparent variability in low cloud as seen by satellite. We will show in the following comment that findings 2, 3, and 5 are unfounded and that the evidence summarized above in findings 1–5, as presented by SB02, does not rule out a link between cosmic rays and clouds.

[3] Since the publication of SB02, ISCCP cloud data have been extended to cover the period July 1983 to September 2001. We restrict our discussion in section 2 to the ISCCP cloud data used by SB02, namely, the period July 1983 to August 1994. Results relevant for the cosmic ray–cloud link obtained from all available ISCCP cloud data are given by Marsh and Svensmark [2003] (hereinafter referred to as MS03). In their response to this comment, Sun and Bradley [2004] (hereinafter referred to as SB04) have taken the opportunity to make some additional remarks regarding these extended data. On the basis of surface observations of low cloud cover (LCC) over the United States, SB04 question the quality of the ISCCP D2 LCC restricted to infrared observations (LCC-IR) as used by MS00 and MS03 to establish the cosmic ray–low cloud link. In section 3 we show that the new evidence presented by SB04 is only valid over the United States and does not invalidate the global response between LCC-IR and cosmic rays previously found by MS00 and MS03.

2. Discussion of ISCCP Period July 1983 to August 1994

2.1. Lack of Correlation in Surface-Based Observations of Total Cloud Cover Over Land

[4] SB02 found no correlation between cosmic rays and surface-based observations of total cloud cover over land areas for (1) the last 45 years over the United States, China, and former USSR, and (2) the period 1982–1996 for tropical and Northern Hemisphere extratropical land areas. These regions represent limited areas of the globe; those described in point 1 cover just 7% of the Earth's surface; thus a negative result does not rule out a global signal. Further, the lack of correlation over land areas is consistent with the suggestion by MS00 that any cosmic ray–low cloud effect may only be significant under maritime conditions where aerosol nucleation and growth can be limited by ionization.

2.2. Correlation Limited to the Atlantic Ocean and Restricted to Period Before 1991

[5] On the basis of their global correlation map of cosmic rays and total cloud cover (TCC) (SB02, Figure 4), SB02 argue that the significant correlation found by SFC for the global ocean average originates mainly from the Atlantic Ocean. However, global trends do not necessarily reflect the properties of a limited region of the globe. This is evident for the global ocean TCC average where a good correlation with cosmic rays is also present when excluding the Atlantic region, as seen in Figure 1. This indicates that the correlation found with ISCCP observations between 1983 and 1991 is really of global origin and not restricted to the Atlantic region alone.

Figure 1.

Normalized 12-month running means of ocean averages for total cloud cover (TCC) obtained from ISCCP D2 between 1983 and 1991. Averages are made globally (solid thick), over the Atlantic (dotted), and globally excluding the Atlantic (dashed). Cosmic rays (solid thin) are represented by normalized monthly neutron counts from Climax, Colorado, with cutoff rigidity 3 GeV.

[6] While the correlation with globally averaged TCC breaks down after 1991, LCC-IR displays a good correlation for the period July 1983 to August 1994 (Figure 2). SB02 dispute the robustness of this latter finding on the basis of the lack of correlation with average low cloud amount over the North Atlantic for the same period (Figure 2 and SB02's Figure 10). In addition, the global correlation map by MS00 (Figure 2) shows little correlation with cosmic rays over this region, which covers just 8% of the Earth's surface. This would appear to contradict the strong correlation previously found over the North Atlantic with total cloud amount (Figure 1). However, regional climate change can produce simultaneous regions of positive and negative cloud responses both vertically and horizontally. By taking a reduced area average of LCC-IR over the North Atlantic, rather than a global average of TCC, reduction of monthly fluctuations due to both instrument and climate noise will be limited. As a result, local climate features may dominate over any global signal that might also be present, e.g., the North Atlantic Oscillation. In these circumstances, only by taking a more globally representative average of LCC-IR is it possible to recover any global signal that may be present.

Figure 2.

Normalized 12-month running means for average low cloud cover (LCC-IR) obtained from ISCCP D2. Averages are made globally (solid thick), over the North Atlantic (dotted (adapted from SB02)), and globally excluding the North Atlantic (dashed). Cosmic rays (solid thin) are represented by normalized monthly neutron counts from Huancayo with cutoff rigidity 12.9 GeV (adapted from MS00). Also shown is the low cloud cover average (thick grey) over regions where the satellite view of low cloud is not obscured by high or middle level clouds.

[7] The results presented in Figure 2 indicate that comparing global averages with and without the North Atlantic region has little effect on the overall correlation with cosmic rays. The lack of correlation over the North Atlantic region should not be used to deflect from the overall result, i.e., that there is a correlation between LCC-IR and cosmic rays.

2.3. Lack of Evidence to Suggest That the Cosmic Ray–Low Cloud Correlation is Physically Based

[8] SB02 (p. 10) observed, from MS00 (Figure 2), that no correlation with cosmic rays was apparent “over the continental United States and most of the other middle to high-latitude land areas.” SB02 then concluded that since Yu and Turco's [2000] mechanism for an ion-aerosol influence was based on studies at the field site in Idaho Hill, Colorado, there is no evidence to suggest that the cosmic ray and low cloudiness relationship is physically based. However, a more recent publication (Yu and Turco [2001] and reference 17 in MS00), which did appear in print before the publication of SB02's contribution, indicated a similar effect was found from observations over the Pacific Ocean. SB02 are right when they state that the ion-mediated nucleation process involves complicated chemical/physical interactions. However, the theoretical considerations of Yu and Turco [2001] included the known interactions in a comprehensive model, which led them to conclude that the ambient nucleation rate was limited by the availability of ions in and above the marine boundary layer. This finding is supported by aerosol observations over the Pacific and concurs with the correlation maps of cosmic rays and cloud properties reported in MS00.

2.4. Increase in the Long-Term Trend of Ship Observations Between 1952 and 1995

[9] SB02 reported the finding of Norris [1999] that ship-based observations reveal an increase in total cloud of 1.9% and low cloud of 3.6% over the global oceans between 1952 and 1995. This is opposite to that expected if a cosmic ray–cloud link exists. However, they neglect to point out the possibility of numerous inhomogeneities both temporally and spatially that may be present in the ship-based observations of clouds. In fact, Norris [1999, p. 1864] stated that it “remains uncertain whether the observed increases in global mean ocean total and low-cloud cover between 1952 and 1995 are spurious. Corroboration by related meteorological parameters and satellite-based cloud data sets should be required before the trends are accepted as real.” When comparing the trends for these ship-based observations of global and low clouds with the ISCCP data where they overlap, it is quite clear that they do not agree, a point which SB02 (in their Figure 9) acknowledge. Norris [1999, p. 1869] adds “the uniform upward nature of the trends is consistent with a possible observational artifact produced by ships travelling over most of the global ocean.” So it is difficult to draw conclusions regarding trends in past ocean cloud cover changes from these ship-based observations.

2.5. Satellite View of Low Cloud Obscured by Overlaying Higher Clouds

[10] SB02 raise an important issue by suggesting that high clouds obstruct a satellite's view of clouds at a lower altitude. This could introduce artificial trends in low clouds that are simply the inverse trends of clouds above, a point that has now been explicitly addressed by MS03. MS03 identified regions of LCC-IR that are significantly (negatively) correlated with high or middle level clouds. A LCC-IR average excluding these obscured regions (thick grey line, Figure 2) is seen to be in good agreement with the full global average, confirming that the correlation with cosmic rays is a robust feature of LCC-IR.

3. Discussion of ISCCP Period July 1983 to September 2001

[11] SB04 claim that the globally averaged ISCCP D2 LCC restricted to infrared observations used by MS00 and MS03 is unreliable and calls into question the existence of a possible cosmic ray–low cloud link. This is based on their finding that surface-based (SB) observations of LCC over the United States agree well with a surface-based definition of LCC obtained from ISCCP D2 using both infrared and visible (VIS) observations (LCC-IR/VIS-SB) but not with LCC-IR (see SB04's Figure 1). Below we argue that the cosmic ray–low cloud link based on globally averaged LCC-IR is reliable, and we will show the following: (1) it is a generalization to assume that results obtained from a limited area automatically apply globally; (2) the poor agreement between LCC-IR/VIS-SB and LCC-IR appears to be restricted to the United States; and (3) assumptions used by SB04 to define LCC-IR/VIS-SB are known to be unrealistic. We begin with a description of the total cloud cover and low-cloud cover definitions LCC-IR and LCC-IR/VIS-SB that have been obtained from the ISCCP cloud data.

[12] ISCCP satellites detect a cloud if the infrared (IR, 10.5 μm observed both day and night) or visible (VIS, 0.7 μm observed daytime only) radiance received by a satellite differs from the clear-sky value by a precalculated threshold. The IR detection of a cloud is limited when its temperature is close to that at the surface. However, a better contrast is often obtained from VIS radiances. As a result a greater amount of TCC is detected when combining both IR/VIS threshold tests (TCC-IR/VIS) than when using the IR threshold test alone (TCC-IR). Globally averaged TCC-IR/VIS detects roughly 5.7% more cloud than TCC-IR, although the correlation coefficients reported in Table 1 indicate that the monthly fluctuations and long-term trends appear to be similar. However, differences between IR and IR/VIS definitions do occur when TCC is divided up into different cloud types.

Table 1. Correlation Coefficients Between Monthly Averages of TCC and LCC for Daytime Observationsa
 GlobalOceanLandUnited States
  • a

    Global, land, ocean, and U.S. averages are obtained from the blue shaded region in Figure 3b. Coefficients calculated from anomalies are in parentheses. Significance levels, taking into account any serial correlation (autocorrelation) that may be present, are indicated as bold (>95%) and italic (>99%).

TCC-IR versus TCC-IR/VIS0.93 (0.94)0.93 (0.94)0.90 (0.91)0.97 (0.95)
LCC-IR versus LCC-IR/VIS0.48 (0.53)0.57 (0.46)0.67 (0.57)0.20 (0.06)
LCC-IR versus LCC-IR/VIS-SB0.40 (0.45)0.55 (0.41)0.49 (0.42)−0.24 (−0.16)
LCC-IR/VIS versus LCC-IR/VIS-SB0.95 (0.95)0.97 (0.97)0.88 (0.94)0.77 (0.86)

[13] ISCCP distinguishes between different cloud types based on cloud altitude estimated from IR and optical depth estimated from VIS. Cloud altitude is initially found where a cloud is assumed to be an opaque blackbody, i.e., no cloud transmittance, and is then adjusted by using the optical depth to estimate cloud transmission. LCC-IR is based on the opaque black body assumption only, however, when VIS radiances are available clouds at each altitude level (low, middle, and high) are additionally divided into three cloud types based on optical depth. In the following, LCC-IR/VIS provides a VIS adjusted estimate of LCC-IR and is obtained from the summation over the three ISCCP-defined low cloud types: cumulus, stratocumulus, and stratus. Both LCC-IR and LCC-IR/VIS are restricted to clouds with tops below 3.2 km. However, surface-based observations of LCC are restricted to clouds with bases below 2 km and therefore also include clouds whose tops are above 3.2 km. SB04 try to account for this in the ISCCP data by estimating LCC-IR/VIS-SB from a summation over the following cloud types: low (cumulus, stratocumulus, and stratus); middle (nimbostratus); and high (deep convective). This is a fundamentally different measure of low cloud cover than LCC-IR. Comparing LCC-IR/VIS-SB with LCC-IR, as SB04 do, is not comparing like with like. The lack of correlation between these two quantities over the United States does not provide a convincing argument to claim that the LCC-IR is unreliable. In fact, as is described below, a significant correlation is found between LCC-IR/VIS-SB and LCC-IR when extending the spatial average over the globe, ocean, and land, which does not support SB04's claim.

[14] In Figure 3a, ISCCP monthly means of LCC-IR (red), LCC-IR/VIS (green), and LCC-IR/VIS-SB (blue) are plotted for daytime observations, with spatial averages taken over the area indicated in Figure 3b. Correlation coefficients between all three definitions of LCC (Table 1) over the United States indicate that the monthly fluctuations of LCC-IR are not correlated with either LCC-IR/VIS or LCC-IR/VIS-SB, which agrees with the finding of SB04. However, LCC-IR is significantly correlated with LCC-IR/VIS and LCC-IR/VIS-SB for global, land, and ocean averages. Using SB04's own criteria for reliability, this latter result implies that LCC-IR is in fact reliable at the global scale. It is an overgeneralization to assume that results obtained from a limited area, such as the United States, apply globally, and (again) this should not be used to deflect from the overall result.

Figure 3.

(a) Monthly averages of LCC-IR (red), LCC-IR/VIS (green), and LCC-IR/VIS-SB (blue) for daytime observations. Cosmic rays (black) are represented by normalized monthly neutron counts from Huancayo with cutoff rigidity 12.9 GeV. A step increase of 1.5% in LCC-IR after October 1994 (red dotted) improves the correlation with cosmic rays. This step may be a result of the gap in available ISCCP intercalibration satellites (thick grey vertical line). (b) Area over which spatial averages of all three LCC definitions in Figure 3a have been taken (blue shaded region). Grid boxes with an incomplete monthly time series over the period July 1983 to September 2001 are also indicated (white regions).

[15] Figure 3a also includes a plot of cosmic rays (black), with the corresponding correlation coefficients reported in Table 2. Over the period July 1983 to August 1994 a correlation with cosmic rays is found to be restricted to LCC-IR, which vanishes when including the more recently released data up to September 2001. It is interesting to note that between September 1994 and January 1995, ISCCP were unable to maintain a continuous intercalibration of the satellite observations (Figure 3a, thick grey vertical line), a period during which an unusual step of ∼1.5% is seen in LCC-IR (Figure 3a, red). Adjusting LCC-IR with a step increase of 1.5% after October 1994 (Figure 3a, red dotted) enables the correlation with galactic cosmic rays (GCR) to return (Table 2). A similar result has also been obtained when comparing with independent satellite observations of cloud cover (MS03) or by subtracting a linear trend [Usoskin et al., 2004]. The continued correlation between cosmic rays and LCC-IR, but lack of correlation with LCC-IR/VIS or LCC-IR/VIS-SB, raises the question of whether the long-term variation in LCC-IR is a physical property of LCC or an artifact. The uncertainties in LCC-IR/VIS and LCC-IR/VIS-SB, which rely on the VIS radiances to estimate optical depth, are much larger than in LCC-IR. Relative uncertainties in VIS radiances are <5% and IR radiances are <2%. Further, the estimation of optical depth relies on a number of assumptions regarding the cloud droplet distribution. In particular, the effective droplet radius is held constant, which has been shown to be unrealistic [Han et al., 2002]. This will affect the accuracy at interannual timescales, and it is unlikely that the correlation between cosmic rays and LCC-IR could be resolved in LCC-IR/VIS or LCC-IR/VIS-SB with the relatively large uncertainties that are present in the VIS adjustments.

Table 2. Correlation Coefficients Between Galactic Cosmic Rays and LCC-IR, LCC-IR/VIS, and LCC-IR/VIS-SB for Daytime Observations as Plotted in Figure 3aa
 July 1983–Aug. 1994July 1983–Sept. 2001
  • a

    Coefficients are calculated from monthly data averaged over the blue shaded region in Figure 3b, having first removed the seasonal cycle. The coefficient in parentheses includes a step increase of 1.5% in LCC-IR after October 1994. Significance levels, taking into account any serial correlation (autocorrelation) that may be present, are indicated as bold (>95%).

LCC-IR versus GCR0.600.16 (0.55)
LCC-IR/VIS versus GCR−0.08−0.21
LCC-IR/VIS-SB versus GCR0.01−0.17

4. Summary

[16] This comment highlights why SB02's results do not rule out a link between cosmic rays and clouds. SB02's (p. 11) claim that there is “no solid evidence for the existence of the galactic cosmic ray flux-low cloud relationship” is unfounded. In particular, we have demonstrated the danger of extrapolating results obtained from a local region to the whole globe. On the basis of satellite observations, there continue to be strong indications of a globally distributed correlation between cosmic rays and low cloud cover. Finally, in contrast to SB02's claim, theoretical mechanisms have been developed that may be relevant for a microphysical explanation of the cosmic ray–low cloud link.

[17] Observations have raised the tantalizing possibility of a relationship between galactic processes and Earth's climate. It is important not to discard such a link without a thorough investigation; this should include both theoretical and experimental work.

Ancillary

Advertisement