Ice cores simultaneously record many proxies of the Earth environment over long periods of time, providing a window into the atmospheric, oceanic, land, solar and volcanic variables that determine our climate. Core sites include the polar ice caps - in Greenland and Antarctica, and high mountain glaciers at a range of latitudes. The records cover a range of time scales, from 400kyr at Vostok (East Antarctica) (see Barnola, 1999) to 0.2yr at Siple (Antarctica). Snow accumulation and ice-sheet depth determine the resolution and length of the record. An accurate ice chronology has been established by counting of annual layers which have recognisable physical, chemical, electrical, and isotopic characteristics (Lorius, 1985 and Legrand and Mayewski 1997). After calibration to historically known volcanic eruptions, ice dating errors are only 1% (Alley, 2000). Here I review the chemistry of the temperature and atmospheric gas (mainly CO2) reconstructions in detail, since these are major indicators in the glacial/interglacial cycle. There are a wide range of other important proxies including: 10Be as a measure of solar activity (see Raisbeck et al., 1990); sulfate for volcanic activity (see Zielinski et al., 1995); Al for the extent of continental deserts; black carbon for forest fires; and other soluble ions which help reconstruct the sulphur, nitrogen and carbon cycles (see Legrand and Mayewski, 1997).
The stable isotope content of water molecules in ice acts as a palaeothermometer. There are three major isotopic forms of water: H216O is the major form; whereas HDO and H218O are heavier and comparatively rare. Heavy water condenses more readily, so on condensation or evaporation the liquid (or solid) phase becomes enriched compared to the vapour phase (here δD or δ18O are used for the change in D/H or 18O/16O ratio; δ refers to both isotopes) (Jouzel et al., 1997). The fractionation is temperature-dependent, so low temperatures initially give high d. A colder air mass loses its heavy water more quickly, so its precipitation becomes light (Alley, 2000). So, low δ in ancient snow implies a low condensation temperature (Tc) in the air mass, and vice-versa. But is Tc representative of the surface temperature Ts? Studies show a strong correlation between δ and Ts over the past 50 years, and the regional δ/Ts variations have been comprehensively mapped out to give a ‘spatial slope’ for any given site (Jouzel et al., 1997). Initially, these were used to calibrate the Vostok δ record (extending over 160kyr) to mean annual surface temperature, showing an 11° mean difference between glacial and interglacial periods. This interpretation however assumes that the modern δ/Tc and Tc/Ts relationships have remained constant, even during glacial conditions. In reality, shifting atmsopheric circulation changes could alter the vapour source, and with it the initial δ which is dependent on the water temperature (TW) where it evaporated. Whether such shifts actually occurred is uncertain (White et al. 1997). The status of the cold air ‘inversion’ layer above polar ice where condensation takes place is also uncertain during the glaciations. A plausible glacial strengthening of the inversion would mean cooling could be understated by up to 10°(Jouzel et al. 1997). Finally, δ can only record the temperature during actual snowfall evenTs. Since snowfall is predominantly seasonal, the record is biased towards the snowfall-dominated season. Climate models imply a glacial/interglacial change from summer to winter dominated snowfall, indicating δ would sample the glacial summer and the interglacial winter, and underestimating Ts differences.
Do these three effects obscure the Ts signal in the past? This has been tested with climate models covering the last glaciation. They do suggest a correlation between δ and Ts, but that the δ/Ts relationship would be lower during glaciation. There are however large inconsistencies between models (Jouzel et al. 1997). A more direct approach is to use other empirical palaeothermometers to recalibrate δ. Borehole thermometry measures the physical temperature of the ice at depth, where it has not yet re-adjusted to present climate (Alley, 2000). The δ temperature record can be used as the upper boundary to reconstruct the borehole record, and they are compared and fine-tuned until a good fit is reached. In two separate Greenland ice cores (Cuffey et al., 1995 and Johnsen et al. 1995) calibration with borehole thermometry concurs with the models, finding that the glacial δ/Ts relation is around half the interglacial and thus a glacial/interglacial increase of between 20 and 25°. Another, ‘gas isotope’ palaeothermometer (Severinghaus et al 1998) also supports this finding. At times of rapid climate change a temperature gradient develops between the surface and the bubble-trapping depth, and thermal diffusion of the air means that heavier isotopes become enriched in colder areas. This isotope signal then becomes trapped in air bubbles (see below). During a rapid deglaciation, for example, the surface becomes warmer than ice at depth, which is then enriched with heavy isotopes. The degree of enrichment depends on the temperature gradient, so the magnitude of the climate change can be inferred (Alley, 2000). Since air diffuses faster than heat, this technique gives a better resolution than borehole thermometry.
Although direct analysis of the atmosphere has only been possible in recent times, ice sheets sample it naturally over millennia. Air diffuses within the firn (the upper layer of unconsolidated snow (Battle et al. 1996)) and is trapped at depth where the porous snow is compressed to ice. Samples from the core are ground at -20°C without melting, to prevent contamination. 80 to 90% of the air escapes and is condensed at 14K, then analysed either via gas chromatography (for gas concentration) or mass spectrometry (for isotopic composition) (Friedli et al. 1986). Two important trace gases are CO2 (a greenhouse gas) and methane (exclusively ‘swamp gas’ in pre-industrial times, and thus a proxy of global wetland area). Trace gases are generally well mixed in the atmosphere, so the core effectively records the global scenario. The isotopic composition of gases can provide information about their sources. For example, CO2 from fossil fuel and biomass combustion is depleted in 13C whereas atmospheric and oceanic is relatively enriched (Friedli et al. 1986).
The firn-ice transition occurs between densities of between 0.8 and 0.83 gcm-3, and at a variable depth depending on snow accumulation and temperature at the site (e.g. 90m depth at Vostok,). Therefore not only is the air is younger than the ice it is trapped in (Barnola et al. 1987), but the trapping depth and thus the age difference (ΔT) is variable with climate. This makes dating the air with respect to the ice chronology difficult (see above, Lorius et al. 1995). Diffusion within the firn-ice transition zone also means that each sample represents an average value over hundreds of years, and that this age-distribution also varies between different climates. Models of firn-densification have simulated both these factors, showing that in interglacial/glacial periods ΔT lies between 2500 and 6000yr, and the age distribution between 300 and 900 yr (Barnola et al. 1991). The latter means that short-term changes are smoothed out. After calibration there is an estimated 5% uncertainty in ΔT (Barnola et al. 1991). Here however we assume that air mixes rapidly within the firn, and is thus in equilibrium with the atmosphere (Barnola et al. 1991). However, Battle et al. (1996) conclude that mixing is slow, so air age does increase with depth. ΔT may be less than previously imagined.
Is the extracted air representative of the atmosphere at the time of close-off? Gravity fractionation, which occurs because CO2 is heavier than air (i.e. the major constituents - O2 and N2), enriches CO2 at the firn base by up to 2ppmv (Barnola, 1999). The oxidation of organic material within the ice can increase CO2 concentrations, especially in the more impure Greenland cores (Barnola, 1999). Acid-base reactions are another possible contaminant (Delmas, 1993). Furthermore, there are artifacts associated with the extraction process that become significant because of the small volume of air that can be analysed (Battle et al. 1996). All of these effects are well understood (Alley et al. 2000) and introduce small errors which, together with the gas ‘age distribution’ (above) obscure short-term variability in the record. Uncertainties in ΔT are the most significant problem. Within these limitations the reconstruction over the recent period is supported by Francey et al. (1997) (in Barnola, 1999) who measured CO2 and δ13C in air, firn and ice simultaneously. Over the long term we rely on firn-densification models (above) which are more error-prone. Records from distinct cores in the two ice caps are consistent (Alley, 2000): for example 4 distinct records over the past millenium lie within a 5ppmv [CO2] envelope (Barnola 1999). This raises confidence but does not rule out any common systemic errors.
The unparalleled advantage of ice cores is their ability to record many different proxies in a variety of ways. This has not only given a large scope, but a good way of testing our assumptions about how the record was laid down in variable climates. There is intrinsic uncertainty and noise in the record, but these are generally well understood, and can be narrowed further by correlation between cores from a wide range of sites. But what has the ice core record revealed? There is a strong correlation between CO2 and temperature over the last 4 glacial/interglacial cycles (Barnola, 1999). CO2 concentrations during interglacial/glacial periods have been high (c. 260-280 ppmv) then low (c. 180-220 ppmv) (Barnola et al. 1991). Comparison with the δD profile shows that during deglaciations CO2 and temperature are in phase, and that CO2 lags temperature by 4.5kyr during deglaciations (Barnola 1999). The pre-industrial CO2 concentration has been pinpointed at between 260 and 280 ppmv. Increasing CO2 concentrations, correlated with the depletion of its 13C content, has also highlighted the impact of anthropogenic greenhouse gas emissions (Friedli et al 1986).