by Donald Rapp
In a recent paper, Ellis and Palmer (2016) proposed that deposition of dust on giant ice sheets, thus reducing their albedo, was a principal factor in the termination of Ice Ages over the past 800 kyrs.
Background
The origin and causes of quasi-periodic Ice Ages over the past ~800,000 years is an intriguing topic that has fascinated many scientists. Many hundreds of papers have been published, and a consensus has grown that a key factor is the so-called Milankovitch model of variable solar input to high northern latitudes dues to wobbles in the Earth’s orbit. While this model has significant resemblance in outline to the ice core and sediment data, there are also some notable exceptions and deviations. The most difficult thing to explain is why Ice Ages end at all, and particularly why they end so suddenly.
In reviewing the literature on understanding Ice Ages, I find a great deal of concentration on variable solar input to high northern latitudes, and the role of changing CO2 concentration, but rather less on the role of dust. Models that did include dust usually were limited to the effect of suspended dust in the atmosphere, rather than dust deposited on ice sheets. Compendiums reviewing the topic of Ice Ages typically downplay the role of dust. For example, the review by Berger and Qin (2012) does not contain the word “dust”. Even my own book (Rapp, 2012) contains only a minimal discussion of glacial dust. Yet a few papers, especially those of Ganopolski et al. mentioned the importance of dust deposition on ice sheets as an important factor in termination of Ice Ages.
In a recent paper, Ellis and Palmer (2016) proposed that deposition of dust on giant ice sheets, thus reducing their albedo, was a principal factor in the termination of Ice Ages over the past 800 kyrs. In their model, terminations occur when extreme dust buildup occurs in an aging Ice Age simultaneously with a sharp rise in solar intensity, They presented arguments in favor of this hypothesis based on previously available data from ice cores showing a large buildup of dust prior to termination of Ice Ages, as well as some dubious reworking of Milankovitch-type solar variability. However, Ellis and Palmer (2016) only reported briefly on previous work in regard to the connection between dust deposition and Ice Age terminations, and it seems worthwhile to review the data and models related to dust deposition developed prior to their paper, to establish a foundation before reviewing the thesis of Ellis and Palmer (2016).
Review of Previous Work
Starting around the 1990s, a number of investigators attempted to analyze glacial cycles using global climate models. In such work, they attempted to account for all of the factors that produced “forcings” that induced climate change in the glacial cycle, including the effects of dust. Many of the earlier papers dealt only with the effect of suspended dust in the atmosphere – producing a net cooling effect by reflection of sunlight. A few however, did include with deposition of dust on ice sheets as an important factor in termination of Ice Ages.
Peltier and Marshall (1995) said:
… our analyses suggest that the albedo variations in the ice-sheet ablation zone caused by dust loading may represent an extremely important ablation mechanism. Using our parameterization of “dirty” snow in the ablation zone we find glacial retreat to be strongly accelerated, such that complete collapse of the otherwise stable Laurentide ice sheet ensues.
Overpeck et al. (1996) said:
Our results point to dust aerosols as a potential source of episodic warming during the last glacial period, and suggest that this warming might be the trigger mechanism needed to account for previously unexplained major abrupt climate events.
Thus, mineral dust appears to have been the most globally distributed aerosol, with the largest radiative effect over snow- and ice-covered regions.
Except for regions of northern Canada and Alaska, the dust-induced average annual warming was greater at progressively higher latitudes, and was greatest (up to 4.4°C) in regions with dust over high-albedo snow- and ice-covered areas.
Calov et al. (2005) said:
Dust affects climate dynamics by reducing the albedo of snow and ice. This feedback presumably is a negative one, at least at some stages of the glacial cycle, because an increase of ice volume leads to a dryer climate, to an increase of dust generation and hence, deposition, thereby lowering snow albedo and amplifying snowmelt.
Krinner et al. (2006) used a general circulation model with high regional resolution and a parameterization of snow albedo to show that
…ice-free conditions in northern Asia during the LGM were favored by strong glacial dust deposition on the seasonal snow cover. Our climate model simulations indicate that mineral dust deposition on the snow surface leads to low snow albedo during the melt season. This, in turn, caused enhanced snow melt and therefore favored snow-free peak summer conditions over almost the entire Asian continent during the LGM, whereas perennial snow cover is simulated over a large part of eastern Siberia when glacial dust deposition is not taken into account.
Ganopolski et al. (2010) used a high-resolution climate model that “directly accounts for the effect of dust deposition on snow albedo” to analyze the evolution of the most recent Ice Age from about 130 kya to the present. They assumed some rather high dust deposition rates. They went on to say:
In reality, most of the dust associated with glacial erosion is deposited during summer; hence, the concentration of dust in the upper snow layer during snowmelt is expected to be much higher than it would be in case of uniform mixing over the year. Moreover, when snow melts, only a fraction of dust is removed by melt water and therefore the concentration of dust in snow increases with time. Although the accurate modeling of all of these processes is problematic, related uncertainties are not crucial, because a saturation effect occurs for dust concentration in snow of more than 1000 ppmw and the albedo of snow reaches a value comparable to that for dirty ice.
They included both globally transferred dust as well as glaciogenic dust that “originates from the southern flanks of the ice sheets and this source is significant only for mature ice sheets, which reach well into areas covered by thick terrestrial sediments.” Their model produced an Ice Age that increased in extent from 130 kya to 20 kya, although there were some vacillations in the growth of ice sheets. “Deglaciation began soon after 20 kya and accelerated significantly after 16 kya.” Their estimate of the dust deposition rate during the most recent Ice Age peaked at the LGM about 20 kya at 40 to 60 g/m2-yr. Their model indicated that without inclusion of glaciogenic dust, the ice sheets would have shrunk by 60% over the past 20,000 years but would not disappear. With inclusion of glaciogenic dust, the ice sheets would totally disappear about 8,000 years ago. They concluded:
Hence, at least in our model, accounting for the additional source of dust related to the glacial erosion is crucial for simulating of a complete termination of the glacial cycle …
When they changed the amount of dust deposition, the effect was not great. A small amount of dust was necessary to terminate the Ice Age, but adding more dust had a smaller effect.
Ganopolski et al. (2011) extended the work of Ganopolski et al. (2010) to cover the Ice Ages over the past 800,000 years, using the same model. Their results indicated:
Each glacial termination is associated with a large increase in dust deposition, which reduces surface albedo and enhances ablation. There are two main reasons for the increase in the dust deposition rate during glacial termination: (i) a considerable portion of the North American ice sheet at the glacial maxima spreads over the area covered by thick terrestrial sediments and (ii) most of the ice sheet base over this area is at the pressure melting point. Both of these factors enable fast sliding of the ice sheets and a large sediment transport towards the ice margins which, in turn, lead to enhanced glaciogenic dust production and dust deposition over the ice sheets. This amplifies the direct effect of rising summer insolation and GHG concentration.
It is not very clear from reading Ganopolski et al. (2011) how they set the dust loading as a function of year across 800,000 years. Ganopolski and Calov (2012) mentioned:
The dust deposition is computed as the sum of the background dust deposition, taken from GCM simulations, and the deposition of glaciogenic dust, which is interactively computed in the CLIMBER- 2 model (Ganopolski et al. 2010).
Nevertheless, if we take their results at face value, we can examine the relation between dust levels and termination of Ice Ages as provided by their model. Figure 1 shows their results comparing the rate of dust deposition with the ice volume.
Figure 1. Comparison of modeled dust loading (g/m2-yr) (brown curve) with ice volume (meters of global eustatic sea level equivalent) (blue curve).
Figure 2. Comparison of modeled dust loading (g/m2-yr) (red curve) with ice volume (meters of global eustatic sea level equivalent) (blue curve) showing peaks of dust loading as vertical dotted lines.
It is interesting to compare the modeled dust loading with ice volume in greater detail as shown in Figure 2. The Ice Ages are outlined by the blue lines, while the rate of dust deposition is outlined by the red lines. It is immediately obvious that terminations of Ice Ages are typically preceded by spikes in the dust deposition rate. Furthermore, there is typically a short delay between the dust spike and the rapid termination of the Ice Age. In Figure 2, a vertical brown dotted line is drawn through each spike in dust loading, and the spikes are labeled A through U. The relationship of each dust spike to the ice volume curve is summarized in Table 1.
We note the following:
(1) Not all Ice Ages were of equal duration and were not perfectly regularly spaced. Yet, there is similarity between most of the Ice Ages, and the periodicity, though far from perfect, is suggestive of roughly 100,000 years.
(2) Occasionally, significant increases in ice volume occur between Ice Ages, but these are typically short-lived (G and Q; no highlight).
(3) Textbook cases where a high dust spike immediately precedes a sudden termination of an Ice Age included six cases: A, C, F, L, S and U (yellow highlight). Three other terminations followed medium dust spikes (I, N and P)(orange highlight).
(4) In seven cases (B, D, J, K, M, O and R) a minor dip in the dust loading produced a minor dip in ice volume – as expected (green highlight).
(5) There were several anomalous cases where the magnitude of the change in ice volume was disproportionate to the change in dust loading: E, G, H, Q and T (no highlight).
Table 1. Relationship of the dust spikes to the ice volume curve.
Thus the model of Ganopolski et al. (2011) indicated a primary relationship between a high spike in dust loading followed shortly by a precipitous drop in ice volume. There are some anomalies in this picture, as we have shown, but by and large, the relationship between dust loading and termination of Ice Ages seems to be firmly predicted by their model.
As Ganopolski et al. (2011) pointed out:
As was shown by Ganopolski et al. (2010), simulation of a complete glacial termination, even with the prescribed GHG forcing, is only possible when the deposition of glaciogenic dust is taken into account.
The explanation of glacial termination requires an additional strong nonlinear mechanism, which, in our case, is the dust feedback. This feedback is activated after the ice sheets spread well into the area covered by thick terrestrial sediments. High rates of dust deposition over the ice sheets reduce their albedo, which enhances ablation and thus amplifies the ice sheet response to rising insolation. Note that this mechanism was already proposed by Peltier and Marshall (1995) …
Ganopolski and Calov (2012) reviewed their previous work and concluded:
Switching off the effect of the dust deposition on snow albedo leads to a rapid development of unrealistically large ice sheets, which cannot be melted even during periods of high CO2 concentration and summer insolation. This confirms our earlier speculation (Calov et al. 2005) about the importance of eolian dust in restriction of growth of the ice sheets and their rapid terminations
The model explicitly accounts for the direct radiative forcing of the atmospheric dust and the effect of dust deposition on snow albedo. The latter, as shown in Calov et al. (2005) and Ganopolski et al. (2010), plays an important role in controlling the spatial extent of the ice sheets and the rate of deglaciation.
The dust deposition is computed as the sum of the background dust deposition, taken from GCM simulations, and the deposition of glaciogenic dust, which is interactively computed in the CLIMBER- 2 model (Ganopolski et al. 2010).
Switching off the effect of the dust deposition on snow albedo leads to a rapid development of unrealistically large ice sheets, which cannot be melted even during periods of high CO2 concentration and summer insolation. This confirms our earlier speculation (Calov et al. 2005) about the importance of eolian dust in restriction of growth of the ice sheets and their rapid terminations.
Bauer and Ganopolski (2014) investigated further into the role of dust in glacial cycles. Among the many factors they discussed, the forcing due to dust and the source of dust in their model was presented. Their modeled forcing due to dust deposition on ice sheets is shown in Figure 3. This dust forcing is compared to a measure of relative temperature at EPICA-Dome in Figure 4.
Figure 3. Modeled forcing due to dust deposition on ice sheets by Bauer and Ganopolski (2014).
Figure 4. Comparison of modeled dust forcing by Bauer and Ganopolski (2014) with relative EPIC-Dome temperatures.
Ganopolski and Brovkin (2015) analyzed the most recent four Ice Ages with their model. However, instead of using the actual Antarctic dust data as input to the model, they used their own dust model that was strangely different with generally lower dust levels than the measured values. Although their paper said:
… the dust cycle model simulates atmospheric dust loading and dust deposition rate. The latter affects surface albedo of snow and iron fertilization effect in the South Ocean.
They did not mention further the role of dust deposition.
Throughout the work of Ganopolski and co-workers from 2010 through 2014, they included dust deposition as one factor, along with other inputs to the model, but did not generally single out the relative importance of dust loading in terminations, although they did emphasize that terminations could not occur without dust deposition.
Evidently these models imply a strong connection between dust deposition on ice sheets and termination of Ice Ages. However, other studies have not necessarily placed emphasis on dust in terminations. For example, Claquin et al. (2003) took the opposite view:
Although the net effect of dust over ice sheets is a positive forcing (warming), much of the simulated high-latitude dust was not over the ice sheets, but over unglaciated regions close to the expanded dust source region in central Asia.
In 2012, a book was published entitled “Climate Change Inferences from Paleoclimate and Regional Aspects” with articles by a number of prominent paleoclimatologists (Berger et al., 2012). Several articles in this book did not acknowledge an important role of dust deposition in terminations.
In this same volume, an article by Hansen and Sato attempted to estimate the climate sensitivity of the Earth by comparing conditions at the last glacial maximum (LGM) with modern pre-industrial times. Using known and estimated differences in temperature, CO2 concentration, and other parameters, they calculated various forcings involved in the transition from the LGM to modern times. They did not seem to include deposition of dust on ice sheets as a major forcing; yet as Figures 1 and 2 show, there was a remarkable rise in modeled atmospheric dust prior to the termination of the LGM.
A scan of Zweck and Huybrechts (2005) and Roche, et al. (2012) reveals that the words “dust” and “termination” did not appear once in their papers. Clark et al. (2012) does not contain the word “dust” yet its title is “Global climate evolution during the last deglaciation”.
Heinemann et al. (2014) attempted to model the termination of the last Ice Age. They said:
Paleoclimate data and model studies indicated that the atmosphere during the LGM transported more dust than at present (Mahowald et al., 1999, and references therein). Dust, while it is in the atmosphere, can increase the planetary albedo, which causes a cooling. Dust deposition on snow reduces their albedo, which causes a warming. None of these processes are accounted for in the present study. (Emphasis added).
Hence the strong emphasis placed on dust deposition in terminations by Ganopolski and co-workers was not universally shared by other investigators.
Ellis and Palmer (2016) Concept
In 2016, Ellis and Palmer published a paper that could prove to be important. In this paper they emphasized the importance of dust deposition as a trigger to initiate terminations of Ice Ages.
As Ellis put it (private communication):
Almost everyone agreed that Milankovitch cycles controlled the glacial cycle. But they were unable to explain why some cycles failed to produce an interglacial while others did, and during subsequent research it became apparent that there was no accepted answer to this troubling but central question. A theory is not a theory, if it has a thumping great lacuna in the middle of it. This led me into a detailed study of the glacial cycle, and the revelation dust was at a peak just before each interglacial.
It was only when Michael Palmer sought to refine my rough and rugged draft paper, that the prior research of Mahowald and Galopolski and many others was discovered. And it was surprising that all of these papers danced around what I saw as the central agent of ice age modulation, without identifying and explaining it as such. Ganopolski, for instance, identified a link between ice sheet volume and dust and presumed that the volume of ice was causing the dust – in other words this must have been glaciogenic dust caused by ice-rock erosion. But previous papers had already identified the source of the dust as the Gobi and Taklamakan deserts, excluding the possibility that the dust was glaciogenic.
A key observation made by Ellis and Palmer (2016) was:
When CO2 reaches a minimum and albedo reaches a maximum, the world rapidly warms into an interglacial. A similar effect can be seen at the peak of an interglacial, where high CO2 and low albedo results in cooling. This counterintuitive response of the climate system also remains unexplained, and so a hitherto unaccounted for agent must exist that is strong enough to counter and reverse the classical feedback mechanisms.
They proposed:
The answer to both of these conundrums lies in glacial dust, which was deposited upon the ice sheets towards the end of each glacial maximum… during the glacial maximum, CO2 depletion starves terrestrial plant life of a vital nutrient and causes a die-back of upland forests and savannahs, resulting in widespread desertification and soil erosion. The resulting dust storms deposit large amounts of dust upon the ice sheets and thereby reduce their albedo, allowing a much greater absorption of insolation.
They asserted their proposal:
… explains each and every facet of the glacial cycle, and all of the many underlying mechanisms that control its periodicity and temperature excursions and limitations.
which of course is not as absolute as all that; yet the proposal does have considerable potential merit.
The paper by Ellis and Palmer then goes off on a tangent regarding variability of solar input to high latitudes, which is very heavily traveled ground, and we need not discuss this here. But one point they raised is worth emphasizing: One cannot invoke rising solar input to high latitudes as the sole cause of terminations of Ice Ages since many such increases in solar input do not produce terminations. Increased solar input might be necessary for terminations but is clearly not sufficient.
The essential basis for the hypothesis advanced by Ellis and Palmer is illustrated in Figure 5. The vertical scales are not specified since they are not essential to the argument at this point. Part (C) shows the dust loading in the ice core at Antarctica (using binned data to reduced scatter). Vertical red dashed lines are drawn at each of the major peaks in the dust loading. It can be seen that these red dashed lines align with the sharp minima in Antarctic temperature in Part (A) of the figure. These minima immediately precede rapid increases in Antarctic temperature with a time lag between the peak in dust load and the rise in temperature of several thousand years. The times associated with the rapid increases in temperature are drawn as vertical black dashed lines in Part (B), the solar intensity on June 21 at 65°N latitude. These black dashed lines coincide with rising solar intensity in every case. Therefore the inference made by Ellis and Palmer is that two situations are necessary precursors to a termination of an Ice Age:
(1) There must be a sharp maximum in dust loading.
(2) The sharp dust maximum must coincide with sharply rising solar intensity.
Note particularly that rising solar intensity by itself does not necessarily lead to a termination. Many such rising lobes of solar intensity do not lead to termination. Furthermore, some very high rising lobes of solar intensity have no effect at all on continuity of Ice Ages. Therefore, one might infer that peak dust is even more important than solar intensity in termination of Ice Ages. While these inferences do not in themselves prove a cause–effect relationship, they are highly suggestive.
At this point in their paper, Ellis and Palmer developed a “side bar” discussion of the role of CO2 feedback in glacial-interglacial cycles. This discussion was not germane to their main thesis and is not considered here.
Figure 5. (A) Antarctic temperature. (B) Solar intensity at 65°N on June 21. (C) Dust loading in Antarctica ice core. (All three graphs shown with arbitrary scale)
Validation of the Hypothesis Of Ellis and Palmer
Having established a connection between dust levels and terminations of Ice Ages, the next step is to attempt to quantitatively validate the hypothesis that dust deposition is a principal cause of termination of Ice Ages, by estimating the amount of dust deposited on the ice sheets, and showing that such levels of dust deposition would exert sufficient warming forces to initiate terminations of Ice Ages. Unfortunately, the data are sparse, and one must be content with limited data and approximate models to show support for the hypothesis. Nevertheless, it is important to examine the limited data that are available to assess the validity of the hypothesis to the extent possible.
In the present review, it is desired to keep two aspects of the dust deposition hypothesis separate: (1) levels of dust deposition, and (2) changes in ice albedo due to dust deposition at any level. Ellis and Palmer (2016) tended to discuss both aspects in together, but I will attempt to extract their discussions into two separate sections.
Levels of Dust Deposition on Ice sheets
Ellis and Palmer (2016) pointed out that dust levels within Antarctic ice cores average only 0.8 ppm at peak dustiness, yet the effects of low levels of dust have been demonstrated to be significant. Dust levels at Ngrip in Greenland peaked at about ten times that level shortly before the LGM, as shown by Ruth et al. (2007). The peak dust loading was 8 ppm. The data of Ruth et al. (2007) are shown in Figure 6.
Ellis and Palmer made the assumption that the dust concentration in the ice at the southern flank of the ice sheets during the LGM was about three times that in Greenland at 75°N. That is a reasonable guess. On the other hand, precipitation of snow was likely to be much greater at the lower latitudes, thereby somewhat masking the increased dust load. The claim regarding relative dust deposition in the Great Lakes area vs. Central Greenland depends to some extent upon how one reads the blue colors in Plate 5b of Mahowald et al. (1999). With this assumption, dust loading on the ice sheets might have reached as high as about 25 ppm.
Figure 6. Dust loading from Greenland ice core.
Relation Between Dust Deposition Rate and Dust Loading in Ice Cores
Ganopolski et al. (2010) indicated an annual precipitation rate of 500 mm/yr, while Gildor et al. 2014 estimated 400 mm/yr on the ice sheets during the LGM. The density of snow can vary widely depending on how it packs. Assuming a snow density of roughly 500,000 g/m3, the deposition of snow was roughly 0.5 ´ 500,000 = 250,000 g/m2-yr.
For every g/m2-yr of dust deposited, the ice cores would contain 1/250,000 or 4 ppm.
Ganopolski et al. (2010) modeled the peak dust deposition rate at the LGM to be about 50 g/m2-yr (see Figure 7). That would correspond to ice cores containing 200 ppm, which seems to be far in excess of reality.
If the peak dust loading were about 25 ppm as estimated by Ellis and Palmer, that would correspond to a dust deposition rate of roughly 6 g/m2-yr.
Ganopolski et al. (2010) suggested that the dust is not uniformly mixed with snow over the whole year. They said:
In reality, most of the dust associated with glacial erosion is deposited during summer; hence, the concentration of dust in the upper snow layer during snowmelt is expected to be much higher than it would be in case of uniform mixing over the year. Moreover, when snow melts, only a fraction of dust is removed by melt water and therefore the concentration of dust in snow increases with time.
It is noteworthy that Mahowald et al. (1999) estimated the dust deposition rate for both Greenland and a “miscellaneous ice core” (Plate 6b of their paper) to be about 10 g/m2-yr during the LGM. This is a factor of five lower than the estimate of Ganopolski et al. (2010) and closer to the estimate by Ellis and Palmer. Figure (1c) of Albani et al. (2016) indicated that the LGM dust deposition rate was about
10 g/m2-yr at the lower end of the Laurentide ice sheet, decreasing gradually to about 1 g/m2-yr at the northern end of the ice sheet. At the other end of the scale, Figure 4 of Takemura et al. (2009) indicates “Arctic” deposition as about 0.2 g/m2-yr during the LGM, but this is only about a factor of two higher than at present, and seems unlikely.
Figure 7. Simulated temporal evolution of annual dust deposition rates in the grid cell south of the Laurentide Ice sheet (blue) and south of Fennoscandian ice sheet (red). (Ganopolski et al., 2010).
Most of the dust was presumably deposited during summer. It is possible that there was relatively little snow deposition during summer. Therefore during the dust might have been deposited with little or no admixture of snow, increasing the concentration on the surface during the summer months.
As Palmer put it (private communication):
For glacial termination to go all the way, it seems important that the ice sheets have become saturated, or mostly saturated, with dust from top to bottom. Deposition of fresh dust drops off steeply in the early stages of deglaciation; the continuing melt-off is thus not sustained by fresh dust but must instead be carried by the old dust that is exposed layer by layer by the melt-off itself. Of course, the old dust gets covered by new snow in the winter, but the strong high latitude insolation may be enough to break through this fresh snow cover even without strong fresh dust deposition. (Maybe this is the crucial role of the strong insolation – to break through each year’s fresh snow cover and get back to work on the old dirty ice underneath.)
In this context, it would be important to understand not only how much dust was dropped on the ice sheets during a glacial maximum, but also how many millennia it took for a freshly deposited layer of dirty snow to reach the bottom of the ice sheets.
It seems reasonable to assume that this occurred faster in the big glacial ice sheets than it does now in Greenland, but this is difficult to model. The same goes for modeled dust deposition rates; my impression is that there is altogether too much modeling and too little evidence collection going on in paleoclimatology.
Blocking Area of Deposited Dust
Following Mahowald et al. (1999), if it is assumed that the average dust particle diameter is 2.5 microns, the blocking area of a dust particle is estimated to be
3.14 ´ (1.25 ´ 10-6) m2 ~ 5 ´ 10-12 m2.
And the mass of a dust particle is estimated to be
4/3 ´ 3.14 ´ (1.25 ´ 10-6)3 m3 ´ 1.5 ´ 106 g/m3} = 1.2 ´ 10-11 g
If the dust deposition rate were as high as estimated by Ganopolski et al. (2010), namely 50 g/m2-yr, the number of dust particles per m2 would be 50/(1.2 ´ 10-11) = 4 ´ 1012. The total blocking area of this number of particles would be
4 ´ 1012 ´ 5 ´ 10-12 m2 = 20 m2 per square meter. The surface of the ice sheet would be covered with multiple layers of dust. Even for a dust deposition rate of 6 g/m2-yr, as estimated by Ellis and Palmer (2016) the blocking area of annual dust would be 2.5 m2 per square meter. Thus, the estimated levels of dust on the ice sheets at LGM were enough to make them appear very dusty.
Source of the Dust
Ellis and Palmer (2016) reviewed data and models on the source of dust and albedo effects of dust during Ice Ages. I will be content with a very brief mention of a few points. There seems little doubt that the combination of low temperature and low CO2 was detrimental to plant life. They claimed the principal impact was on high altitude regions, normally arid regions, and northern regions during summer. The source determined by isotopic analysis was attributed to the Gobi and Taklamakan deserts, and there was little evidence of glaciogenic dust. As Ellis put it (private communication):
The question is how these peaks in dust flux were generated during the latter millennia of each ice age. If it was not the grinding of ice sheets that caused the dust, then what did? The likely answer was that the low CO2 concentrations at each glacial maximum led to plant extinction in high altitude arid regions, turning them into new CO2 deserts. And this was a scenario that fitted the source region for Greenland dust very well – the Gobi. The Gobi is mostly steppe grassland, but under the low CO2 conditions of a glacial maximum the entire Gobi became a shifting-sand desert that created vast dust clouds, as is proven by the massive dust deposits upon the Loess Plateau in China.
And so the beauty of this theory and paper is that it is a simple thought-experiment that can be followed by anyone, and yet in my view it explains every facet the glacial-interglacial cycle. And it even explains the previously inexplicable – the reason why some strong precessional insolation maxima failed to produce much in the way of melting and warming, let alone an interglacial.
In short:
Ice ages cause ice sheet extension –> oceanic cooling –> oceanic CO2 absorption = plant asphyxiation on the Gobi plateau –> new CO2 deserts –> dust generation –> ice sheet contamination for 10 kyrs.
Then:
Rising NH Milankovitch insolation –> ice sheet surface ablation and melting –> ice sheet dust exposure and concentration –> ice sheet albedo reduction –> increased insolation absorption –> increased ablation and melting –> interglacial.
It is a simple feedback system that is very powerful and operates regionally, unlike CO2 which is a global feedback agent of indeterminate strength. And yet we know that interglacials are regional phenomena rather than global phenomena, because they only coincide with increased Milankovitch insolation in the northern hemisphere, and never with increased insolation in the south. Ergo, the primary feedback controlling interglacial inception must be regional to the NH, rather than global.
Summary
The hypothesis put forth by Ellis and Palmer (2016) has considerable potential merit. The role of dust in terminations of Ice Ages is probably far more important than many realize. It appears likely that unusually high dust levels coupled to sharply rising solar intensity at high latitudes was a major factor in initiating termination of Ice Ages. As Palmer put it, layers of buried dust probably sustained the evolution of the interglacial period.
Moderation note: As with all guest posts, please keep your comments civil and relevant.