The Intergovernmental Panel on Climate Change (1) estimates the global climate forcing by fossil fuel black carbon (BC) aerosols as 0.2 W/m². Jacobson (2) suggests that the fossil fuel BC forcing is larger, ∼0.5 W/m². J.H. and colleagues (3–5) have argued that the total anthropogenic BC forcing, including BC from fossil fuels, biofuels, and outdoor biomass burning, and also including the indirect effects of BC on snow/ice albedo, is still larger, 0.8 ± 0.4 W/m². Here we estimate the magnitude of one component of the BC climate forcing: its effect on snow/ice albedo.

Several factors complicate evaluation of the BC snow/albedo climate forcing and dictate the approach we use to estimate the forcing.

BC Amount in Snow. BC is highly variable in space and time. A classic study of Arctic sites (6) found average BC (excluding Greenland) of 30 ppbw (parts per billion by weight; equivalent to ng/g or μg/liter meltwater) in 1983–1984. In contrast, only 4 ppbw was found in 1998 (7) upwind of the drifting SHEBA (Surface Heat Budget of the Arctic Ocean) site on Arctic sea ice (∼76°N, 165°E), although 35 ppbw and more was found in a limited region downwind of the SHEBA site. Differences among measured amounts exceed estimated errors and imply real variability.

BC Optical Effects. Classic theory calculates snow/ice albedo as multiple scattering by BC and ice spheres (8, 9). Empirical data (8) revealed that BC was two to five times more effective in reducing snow albedo than the model indicated. A factor of two was accounted for by use of a more realistic density and absorption cross-section for BC. Still greater absorption per unit BC mass, by perhaps another factor of two, can be obtained with realistic shapes, voids, and degrees of internal mixing of the BC within ice particles (10, 11), but there remains uncertainty and variability of the appropriate absorption cross-section for BC dispersed in snow.

BC vs. Soot. Soot is produced by incomplete combustion of carbonaceous material, mainly fossil fuels and biomass. The carbon combustion products are usually classified as BC and organic carbon (OC), but distinctions within the complex carbonaceous mixtures can be ambiguous (12, 13). BC is commonly defined in an operational sense as the absorbing component of carbonaceous aerosols, which may result in some humic-like or other organic material contributing to estimated BC absorption. We employ this operational definition, because the practical question concerns the impact of soot on snow/ice albedo, and for this it matters little whether the carbon is elemental or in other carbonaceous aerosols.

We compile empirical data on BC amount in snow and compare calculated effects on snow albedo with field data. We then use these data to specify plausible albedo changes for climate simulations. We calculate the forcings due to specified albedo changes, carry out equilibrium climate simulations for these forcings, calculate the efficacy of snow/ice albedo forcing relative to CO₂, and carry out transient climate simulations for 1880–2000. Finally, we discuss potential implications.

Soot in Snow. Be thou as chaste as ice, as pure as snow, thou shalt not escape calumny (Shakespeare, Hamlet). Perceptions persist about the purity of fresh snow, but measurements tell another story. Optical and electron microscopes showed (14) the typical snow crystal in Sapporo, Japan in the 1970s to contain thousands of aerosols, including soot. It was inferred (14) that the snowflakes, falling from ∼500 m, had a collection efficiency of order unity (>0.2) in sweeping up aerosols in their path. In central Antarctica, snow crystals in 1969 were found to commonly contain 25–50 aerosols (15), two orders of magnitude less than in Japan, but again implying a collection efficiency of order unity. Mechanisms for high collection efficiency may include (15) electrostatic attraction, thermophoresis (temperature gradient between snowflake and environment), and diffusiophoresis (vapor pressure gradient between snowflake and environment).

High collection efficiency is not general, but it is believed that wet deposition (via snow and rain) is the primary removal mechanism for aerosols as small as BC (16, 17), which has a typical dimension 0.1 μm (12, 13). Dry deposition also can be significant, accounting for several tens of percent of deposition in some situations (18).

We use observed BC amounts in surface snow and precipitation, together with measured snow albedos, to estimate BC albedo effects. The distribution of BC in global aerosol transport models helps extrapolate site measurements to the global scale.

BC Concentration in Snow. Table 1 summarizes BC amounts in surface snow and precipitation based on data in Table 4, which is published as supporting information on the PNAS web site. Snow samples in the 1980s (6), including sites in Alaska, Canada, Greenland, Sweden, and Spitzbergen, and on sea ice in the central Arctic, yielded typical BC amounts of 10–50 ppbw (excluding Greenland). “Arctic haze” studies (19) showed that most of the aerosols originated in Europe and the Former Soviet Union (FSU) in winter/spring driven by circulation around the Icelandic Low and Siberian High. BC emissions from Eurasia probably declined sharply in the 1990s as, e.g., FSU BC emissions fell by a factor of four with the collapse of the FSU economy (20). Reduced BC emissions are not necessarily permanent in the face of possible economic recovery, increased shipping in the opening Northwest and Northeast Passages, regional hydrocarbon resource development, and increased use of diesel-powered vehicles.

BC amounts found in snow at lower latitudes in the Northern Hemisphere are highly variable, usually in the range 5–100 ppbw. Larger amounts found in the French Alps, ∼100–300 ppbw (21), may be related to the high proportion of diesel engines in European surface transportation. It seems likely that East Asia snow has large BC amounts, because China and India are now the largest sources of BC emissions, and photographs reveal a thick brown haze filled with BC that butts against the Himalayas (22), but measurements are lacking. Greenland BC measurements yield mainly 2–6 ppbw.

Pristine Antarctic regions have been found to contain 0.1–0.3 ppbw (23, 24), two orders of magnitude less than in the Arctic. BC amounts of 3 ppbw were found 1 km downwind of the South Pole station (23), where the station's power plant and aircraft operations were a suspected source. BC of 3 ppbw was reported at Siple Dome above the Ross Ice Shelf (25), an amount that might be related to the location being closer to the coast, but the amount seems large for Antarctica unless it was influenced by local pollution.

Clarke and Noone (6) found a consistent “scavenging ratio,” relating BC amount in snow to BC concentration in air, in polar regions of both hemispheres. This relation allows estimates of BC in snow for other locations (see Supporting Text, which is published as supporting information on the PNAS web site).

BC Effect on Snow Albedo. Theory relating snow albedo to BC amount (8, 9), which combines Mie scattering and a multiple scattering approximation, accounts well for measured albedo as a function of ice crystal size at wavelengths where absorption by ice is dominant (λ > 0.7 μm). The presentation of this theory in figure 2 of ref. 9 accounts for externally mixed soot and ice particles but leaves open the possible need to increase the BC effective absorption by perhaps a factor of two for best agreement with empirical data. A critique (11) concludes that there are inherent uncertainties in the effect of BC on snow albedo; specifically, (i) if BC particles are within ice crystals, rather than externally mixed, the absorption power of BC particles increases by a factor 1.4 or more; (ii) the soot particle shape affects its absorption power, with randomly oriented needles or disks being more absorbing then spheres of the same mass, by factors as much as two or more; (iii) voids in the soot particles increase the absorption power per unit mass nearly in proportion to the void fraction; and (iv) there is large uncertainty in the optical constants of soot, as much as a factor of two or more in the imaginary part of the refractive index.

Given the uncertainty in the effect of a given BC mass on snow albedo, we consider two cases in Table 1. External mixing is the standard relation (figure 2 of ref. 9), whereas internal mixing increases the BC absorption coefficient by a factor of two, for better agreement with empirical data. Table 1 gives results for both fine-grained fresh snow and snow in which the ice particles are larger because of aging or melting effects. Snow grain size tends to be larger in late winter and spring, when there is enough sunlight that BC absorption is most important. Average conditions in nature should fall between fresh and old snow, and empirical data suggests that internal mixing is a better approximation than external mixing, but substantial uncertainty must be admitted.

Specified Albedo Changes. We carry out climate simulations for specified snow/ice albedo changes of Table 2. The multiple cases help determine the contributions of different geographical regions. The response is linear for these small forcings, so results for alternative forcings with the same geographical distribution are implicit. In cases 1 and 2, the albedo is decreased by the specified amount only at λ < 0.77 μm and half that amount at 0.77 μm < λ < 0.86 μm. The spectrally integrated albedo change is ∼60% of that at λ < 0.77 μm. Cases 3 and 4 change the albedo individually for Arctic sea ice and Northern Hemisphere land, with the change made at all wavelengths solely to assure good signal/noise response.

The snow albedo at λ < 0.77 μm is decreased 2.5% (1.5% spectrally averaged) in the Arctic, a realistic and perhaps conservative estimate of BC effects in the 1980s, by 1% at λ < 0.77 μm in Greenland, which contributes negligibly to the hemispheric mean response, and by 5% at λ < 0.77 μm (3% spectrally averaged) in other snow-covered land areas in the Northern Hemisphere, because BC in the air and in precipitation is about twice as large in these regions compared to mean Arctic conditions in the limited measurements available and in tracer transport models. All cases have no albedo change in Antarctica. Case 1 has 1% albedo change in snow regions of the Southern Hemisphere other than Antarctica. We consider case 1 to be the most realistic.

Climate Model. Our simulations use the current Goddard Institute for Space Studies climate model. Model physics is similar to that in the SI2000 model (5), but with the current “model E” version of the code as modularized, modified for parallel computations, documented, and otherwise improved by G. Schmidt and R. Ruedy. Specifically, we use model version E037 that existed in the summer of 2003, which we dub SI2003. Resolution is 4° × 5° with 18 layers, the added 6 layers, compared with the 12-layer SI2000 model, being in the stratosphere. The model top is raised to 0.1 hPa and stratospheric drag is reduced, compared with SI2000, to allow more realistic stratospheric climatology (J. Perlwitz and J.H., unpublished work). The climate sensitivity of the SI2003 model is ∼2.6°C for doubled CO₂, somewhat less than the ∼3°C sensitivity of the SI2000 model.

Climate Forcing. Climate forcing caused by the specified snow/ice albedo changes (Table 2) is shown in Fig. 1. This adjusted forcing, F _a, is obtained as the flux change at the top of the atmosphere after the stratospheric temperature has adjusted to the presence of the perturbation with the troposphere and surface temperatures fixed. F _a is a commonly used forcing (1, 26). An alternative forcing definition is compared below.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Climate forcing in W/m² (Upper) and equilibrium annual-mean T _s response in °C (Lower) for changes of snow and ice albedos specified in Table 2. Numbers on the upper right are global means.

The climate forcing due to snow/ice albedo change is of the order of 1 W/m² at middle- and high-latitude land areas in the Northern Hemisphere and over the Arctic Ocean for cases 1 and 2, which we suggest have a realistic magnitude for the soot effect (3% spectral-mean snow albedo change over land, 1.5% over the Arctic, and 0.6% over Greenland). This compares with a global mean forcing by present anthropogenic CO₂ (compared to preindustrial times) of ∼1.5 W/m², which is relatively uniform over the globe (1, 26). The mean Northern Hemisphere forcing is ∼0.3 W/m² in the more realistic cases 1 and 2.

Equilibrium Response. We calculate the equilibrium response to these forcings in the Goddard Institute for Space Studies climate model with a Q-flux ocean in which horizontal heat transports are specified (27). The annual mean response of surface air temperature (T _s) is shown in Fig. 1. The calculated warming is unexpectedly large, relative to the forcings, if one uses the sensitivity to standard forcings such as CO₂ to judge the expected response. The reason for the greater “efficacy” of the surface albedo forcing is discussed below.

Fig. 2 shows the calculated equilibrium T _s response as a function of month and latitude. This, and the map of temperature change (Fig. 1), can be compared with observed temperature change since 1880 (Fig. 3). The model results are an equilibrium response, but transient results are similar, as shown below. The observed change is based on an analysis (28) that screens and adjusts meteorological station data from the Global Historical Climatology Network for urban and other inhomogeneities and uses sea surface temperature data (29, 30) for ocean regions. It is apparent that warming due to snow/ice albedo change potentially could account for a significant fraction of observed warming. The location and season of observed warming in the Northern Hemisphere, with warming at middle and high latitudes that is large in the winter, extends well into the spring, and is minimal in summer, is consistent with a significant fraction of the warming being due to a decreased snow/ice albedo.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Equilibrium T _s (°C) response to the snow/ice albedo forcings of Fig. 1 as a function of month and latitude.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Observed 1880–2002 T _s change based on adjusted meteorological station data over land (28) and sea T _s data for the ocean (29, 30).

These geographical and seasonal features do not confirm the existence of a snow/ice albedo forcing. High-latitude and winter amplifications of the climate response are a natural response of the climate system and climate models to most forcings. Careful study of the transient climate response to all forcings is needed, and this requires information on the time-dependence of forcings, some of which may be difficult to establish.

Nevertheless, it is worth noting that the anthropogenic soot snow/ice albedo forcing and response must be mainly in the Northern Hemisphere. This is the same region in which the negative direct and indirect forcings by anthropogenic sulfate and nitrate aerosols are concentrated. If these negative forcings are as large as has been estimated (1, 3), it raises the question of why observed warming is as large in the Northern Hemisphere as in the Southern Hemisphere. A suggested explanation (5) is that there are large positive air pollution forcings in the same region, primarily BC and tropospheric O₃. Figs. 2 and 3 suggest that the BC indirect snow/ice albedo forcing could provide a substantial part of the reason for large Northern Hemisphere warming.

Efficacy. When climate forcings are used to compare different climate change mechanisms, there is an implicit assumption that the climate response is expected to be proportional to the forcing. We (26) tested this assumption for a large number of forcings and found that it was good for many forcings, in the sense that the global mean T _s response was consistent among these forcings within ∼20%. Absorbing aerosols and O₃ were identified as exceptions, for which the climate response depended sensitively on the altitude at which the forcing was introduced. We did not explicitly test the effectiveness of changes in snow albedo, but we showed that an arbitrary (“ghost”) forcing at the surface was twice as effective at high latitudes as at low latitudes. Reasons for the large response to high-latitude surface forcings were identified as (i) the ability of a high-latitude forcing to stimulate snow/ice albedo feedbacks and (ii) the relative stability of the atmospheric temperature profile at high latitudes, which tends to confine the thermal response to the surface.

We define the “efficacy” of a climate forcing, E, as the ratio of the equilibrium T _s change for that forcing to the temperature change for the same magnitude of forcing by CO₂. Comparison with CO₂ has practical merit, because CO₂ is expected to be the largest anthropogenic climate forcing in the foreseeable future.

Table 3 gives the efficacies for the snow/ice albedo forcings of Fig. 1 for two definitions of climate forcing. F _i and F _a are the standard “instantaneous” and “adjusted” climate forcings (1, 26). For these forcings, the efficacy of the snow/ice albedo forcing is ∼2, i.e., the equilibrium T _s response to the forcing is about twice as large as for a CO₂ forcing of the same magnitude. (We did not run the climate model for small CO₂ forcings equivalent to the ice/snow albedo forcings but rather scaled the doubled CO₂ response in proportion to the forcing. For positive forcings within the present range that is a good approximation. For larger forcings, and especially for negative forcings, high accuracy requires that the climate model be run for several CO₂ changes to define the nonlinear relationship.)

The variation of efficacy from one case to another is due mainly to the latitude of the forcings. For the reasons mentioned above, the response to polar forcings is larger than the response to lower latitude forcings. The efficacy for changes of Arctic sea ice albedo is >3. In additional runs not shown here, we found that the efficacy of albedo changes in Antarctica is also >3.

We define the effective forcing as F _e = F _a E _a. F _e provides a better measure of expected climate impact of a climate forcing mechanism than does either F _i or F _a. The effective forcing for the assigned snow albedo change in the most realistic cases 1 and 2 is F _e ∼ 0.6 W/m² in the Northern Hemisphere or F _e ∼ 0.3 W/m² globally.

Transient Simulations. The above results suggest that a significant fraction of global warming in the industrial era might be due to soot's effect on snow albedo. Thorough investigation of this issue requires knowledge of soot deposition on snow as a function of geography and time, and the study should include other known forcings, so that results can be compared with observations.

Here we make transient calculations for 1880–2000 by using a simple soot snow/ice albedo forcing, with the limited aim of obtaining a better indication of the possible magnitude of global warming due to soot snow/ice albedo forcing. The time dependence of the forcing is taken proportional to global mean BC amount in the transport model of Koch (17) with the BC emissions history including a “technology factor” (20). The technology factor accounts for reductions in BC emissions due to increasing use of low-emission power plants and improved diesel technology. Despite the technology factor, global BC emissions increase with time, as shown by figure 3 of ref. 20.

We take the geographical distribution of the forcing as in case 1. In the Koch/Novakov BC scenario, 83% of the 1850–2000 BC increase occurs during 1880–2000. So the forcing in the transient simulation increases by 0.83 × 0.16 ∼ 0.13 W/m² between 1880 and 2000. The fixed pattern of the forcing is probably not a serious flaw for our purposes, because we only examine the total change from 1880 to the present, which tends to average over fluctuations of sources.

Fig. 4 is the five-member ensemble mean T _s change, obtained with the Q-flux ocean with diffusive mixing of heat into the 4-km deep ocean (5). The exact pattern of simulated warming has limited meaning, given the crude spatial distribution of the forcing and the simple representation of the ocean. However, the mean warming provides an indication of what should be expected for the assumed albedo change. This global warming, 0.17°C, is a substantial fraction of observed warming (Fig. 3). Even if our snow/ice albedo forcing were too large by a factor two, the impact is not negligible.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Simulated 1880–2002 T _s change for the transient BC snow/ice albedo forcing that peaks in the 1990s with 83% of the case 1 forcing of Fig. 1.

The simulated warming is largest at high latitudes in winter, because of reduced sea ice cover induced by the warming, and least in summer. These spatial and temporal characteristics are consistent with observations. This does not provide confirmation of the soot albedo effect, because other forcings may have a similar impact on the pattern of climate change. All forcings need to be investigated one-by-one in systematic simulations and compared with observations.