Northern Hemisphere the Last Million Years

Variations in global ice volume during the Pleistocene, and probably the Pliocene, have been dominated by growth and decay of Northern Hemispheric ice, and most ice-sheet modelling of this period has been concerned with the Laurentide, Cordilleran, Greenland and Eurasian ice sheets. Weertman (1976) was the first to include ice dynamics in model simulations of long-term ice-sheet evolution. He used a perfectly plastic icesheet profile representing a north-south cross section of the Laurentide Ice Sheet, and forced it from 400 ka to the present with a prescribed pattern of net annual accumulation minus ablation versus latitude and elevation (termed a snowline pattern below), shifted vertically in proportion to summer insolation at 65°N to represent actual Milankovitch orbital variations. This study was the first to capture the several thousand-year lag of ice volume behind climate forcing due to the mass inertia of the ice sheet. It produced reasonable ~ 23-41 kyr cycles in direct response to precessional and obliquity orbital variations, but did not realistically simulate the dominant observed 100 kyr ice cycles.

Subsequent work extended this study to (i) use energy-balance climate models (EBMs) instead of a prescribed snowline pattern (Pollard, 1978; Berger et al., 1990; Deblonde and Peltier, 1993) and (ii) use ice-sheet flow models instead of plastic profiles, allowing bedrock topography and lagged depression below the ice (Birchfield et al., 1981; Pollard, 1982; Hyde and Peltier, 1987). For the most part, the addition of climate models did not change Weertmans' results significantly, and just showed that the smooth summer-temperature variations in EBMs have much the same effect as a shifted snowline pattern. The additions of high-latitude topography and lagged bedrock depression and rebound were found to be more important: high-latitude plateaus allow nascent ice caps to initiate more easily following interglacials, and deep bedrock depressions at glacial maxima amplify subsequent retreat due to warmer air temperatures at lower elevations. Although these mechanisms produced more non-linear response, they were still not enough to produce realistic 100 kyr response and full G-I cycles.

Retreats are triggered in these models by hot northern summer orbital configurations that are coeval with the start of major observed terminations, but the modelled retreats are not complete, allowing substantial ice cover to persist through interglacials.

It was found that some additional physics were needed that come into play during the major deglaciations to terminate each 100 kyr cycle, and a large variety of mechanisms have been proposed. These range from cyclic freezing versus melting and surging at the ice-sheet base (Oerlemans, 1982b; MacAyeal, 1993), calving by proglacial lakes (Pollard, 1982), prescribed climatic effects of atmospheric CO2 and/or North Atlantic heat transport and thermohaline circulation (Tarasov and Peltier, 1997), and atmospheric dust concentrations (Peltier and Marshall, 1995). These mechanisms involve or respond to some other long-term component (besides ice volume) with intrinsic time scales of thousands to tens of thousands of years, which is logically necessary because mechanisms that only involve fast-time-scale components (atmosphere, snow, sea ice, upper ocean) would have been triggered not just during major deglaciations but at other times with similar orbits and mid-range ice extents. However, there are two exceptions: snow aging (Gallee et al., 1992) and sea-ice switching (Gildor and Tziperman, 2001). Other aspects of the 100 kyr cycles have been studied recently, such as phase locking by eccentricity or obliquity/precession (e.g. Tziperman et al., 2006) and whether these cycles are deterministic at all (Wunsch, 2003). Although many of the models mentioned above produce realistic Pleistocene ice volume time series including ~23, 41, and 100 kyr cycles, there is still no consensus on what processes are responsible for the rapid inceptions and complete deglaciations that are key to the dominant 100 kyr cycles of the last million years.

An entirely different approach is to use zero-dimensional 'box' models comprised of a few coupled non-linear ordinary differential equations for quantities such as global ice volume, mean ocean temperature and atmospheric CO2 level (Matteucci, 1989; Saltzman and Verbitsky, 1993; Paillard, 1998). These can be tuned to yield fairly realistic 'ice volume' time series forced by Pleistocene orbital variations. Hargreaves and Annan (2002) used Saltzman's model and Monte Carlo Markov Chain (MCMC) techniques to find the optimal parameter set and probability density functions yielding the most realistic results; similar approaches have used three-dimensional ice sheet models (Tarasov and Peltier, 2004). Paillard's zero-dimensional results basically support the general conclusion from more physical models that nonlinear thresholds are needed to successfully produce 100 kyr cycles from the higher-frequency orbital forcing. However, the highly conceptual formulation of the zero-dimensional models make it hard to translate their results into specific action items for improvements in more physically explicit models.

A unique approach was taken by Bintanja et al. (2005) to address the ambiguity between ice volume and local sea-water temperature inherent in deep-sea core d18O records. They used a NH ice-sheet flowline model, driven through multiple G-I cycles over the last million years with simplified climate-orbital linkages as in many previous studies. But instead of simply running the model forward in time, (i) a standard empirical formula was used to predict foraminiferal 818O at any time from the model's current deep-sea temperature and ice volume, and (ii) at each time step the climate temperature was determined not from a physical model, but from the requirement that, when the ice model was stepped forward, the computed foraminiferal d18O yielded the observed deep-sea-core value for the next time step. This produced a self-consistent estimate of the relative contributions of Northern Hemispheric ice versus deep-ocean temperature in 818O core records, with the main conclusion that the ice-sheet contribution varied from ~10% in the beginning of glacial cycles to ~60% at glacial maxima; the latter is consistent with relict pore water measurements (Schragg et al., 1996).

Some attempts have been made to use General Circulation Models (GCMs) in conjunction with long-term ice-sheet variations. Computer time is prohibitive for direct GCM integrations longer than a thousand years or so, and several strategies have evolved to circumvent this limitation. Several studies have driven three-dimensional ice sheet models through the last glacial cycle (last ~ 125 kyr), with climates obtained by interpolating between two stored GCM snapshots at the LGM (21 ka) and modern. The weighting of the two GCM climates is proportional to an observed ice core record (usually Greenland GRIP) representing NH climate of the last 125 kyr (Marshall and Clarke, 1999b; Zweck and Huybrechts, 2005; Charbit et al., 2007). However, this procedure heavily constrains the predicted variations of ice volume (on millennial, orbital and 105-year time scales) to those in the ice core record, so the approach is useful mainly for diagnostic assessment of processes in the ice model. For instance, one striking diagnostic result is the greatly expanded extent of basal melting under the Laurentide as LGM is approached, which is consistent with the basal freeze-thaw-surge hypothesis for amplification of 100 kyr cycles (Marshall and Clark, 2002; cf. Johnson and Fastook, 2002).

To couple relatively coarse-grid GCM climates (~100'skm) to the finer ice-sheet model grids and topographies (~10'skm) in these experiments, straightforward downscaling techniques have been developed (e.g. Ramstein et al., 1997; Thompson and Pollard, 1997; Marshall and Clarke, 1999a). First, monthly mean air temperatures and precipitation are interpolated both horizontally and vertically to the ice surface locations. Degree-day parameterisations are used to convert monthly temperatures to snow or ice melt, which are able to capture subtle effects of orbital changes in seasonal insolation (e.g. Huybers, 2006). Some calculations explicitly account for refreezing of meltwater. Also, 'anomaly' techniques are often used to reduce biases in GCM climate, by only using GCM predicted changes in climate from the present, combined with modern observed climatology. This is usually done additively for temperature, and by ratio for precipitation to avoid negative precipitation values.

A few studies have used asynchronous coupling between GCMs and ice sheets, whereby the GCM is run for a few decades every few thousand years and its climate is used to drive the ice sheet model alone for the next few thousand years. Then the new ice-sheet extent is used to update the GCM's surface conditions, and the sequence is repeated. For the Pleistocene, this has been done only over the last 21 kyr through the last deglaciation (Charbit et al., 2002). More recently, Earth Models of Intermediate Complexity (EMICs) have made it possible to run coupled climate-ice sheet models through long periods (Charbit et al., 2005). Realistic Pleistocene cycles have been obtained in this way (Calov and Ganopolski, 2007), but, as found in some of the earlier studies above, prescribed variations of atmospheric CO2 and North Atlantic ocean heat transport were needed to obtain complete 100 kyr cycles. A new approach using climate parameterisa-tions based on past GCM snapshots has recently been taken by Abe-Ouchi et al. (2007).

Many Global Climate Model (GCMs) and Regional Climate Model (RCMs) have performed 'snapshot' simulations of Pleistocene climates at individual times, usually with prescribed ice sheets. One of the first such applications was Gates (1976) at the LGM, using prescribed CLIMAP SSTs and ice sheet reconstructions; recently, fully coupled A/OGCMs have been used for LGM (Hewitt et al., 2003; Kim et al., 2003), and GCMs and EMICs have been applied to other times such as the end of the last interglacial, sometimes driving nascent ice sheets (deNoblet et al., 1996; Vettoretti and Peltier, 2004; Kubatzki et al., 2006). A few studies have used RCMs or zoomed-region GCMs to simulate the mass balance over NH ice sheets at LGM and other times (Hostetler et al., 2000; Krinner et al., 2004). A detailed survey of these simulations is outside the scope of this chapter, but two general points emerge concerning ice sheets: (i) there is wide scatter between GCM-derived mass balances of ice sheets, due to the strong sensitivity of summer melt to air temperatures over the ablation zones (Pollard and PMIP, 2000), and (ii) although reasonable mass balance patterns on major NH ice sheets can be attained, several characteristics consistently disagree with geologic data, such as too much ice buildup over Alaska and Siberia (Charbit et al., 2007). It is relatively easy to adjust the downscaling of climate or other parameters to achieve a reasonable mass balance for a given GCM, time and ice sheet, but much more challenging to achieve good results for other times and other ice sheets using exactly the same model and parameter values.

0 0

Post a comment