| |

Received: 5 November 2019    Revised: 11 May 2020    Accepted: 18 May 2020

DOI: 10.1111/jbi.13930

RE SE ARCH PAPER

Global vegetation patterns of the past 140,000 years

Judy R. M. Allen1 | Matthew Forrest2 | Thomas Hickler2 | Joy S. Singarayer3 |

Paul J. Valdes4 | Brian Huntley1

1
Department of Biosciences, Durham
University, Durham, UK Abstract
2
Senckenberg Biodiversity and Climate Aim: Insight into global biome responses to climatic and other environmental changes
Research Centre, Frankfurt am Main,
is essential to address key questions about past and future impacts of such changes.
Germany
3
Department of Meteorology, University of
By simulating global biome patterns 140 ka to present, we aimed to address impor-
Reading, Reading, UK tant questions about biome changes during this interval.
4
School of Geographical Sciences, University Location: Global.
of Bristol, Bristol, BS8 1SS, UK
Taxon: Spermatophyta.
Correspondence Methods: Using the LPJ-GUESS dynamic global vegetation model, we made 89 simu-
Brian Huntley, Department of Biosciences,
Durham University, South Road, Durham lations driven using ice-core atmospheric CO2 concentrations, Earth's obliquity, and
DH1 3LE, UK. outputs from a pre-industrial and 88 palaeoclimate experiments run using HadCM3.
Email: [email protected]
Experiments were run for 81 time slices between 1 and 140 ka, seven ‘hosing’ ex-
Funding information periments also being run, using a 1-Sv freshwater flux to the North Atlantic, for time
Leverhulme Trust, Grant/Award Number:
RPG-2014-338 slices corresponding to Heinrich Events H0–H6. Using a rule-based approach, based
on carbon mass and leaf area index of the LPJ-GUESS plant functional types, the
biome was inferred for each grid cell. Biomes were mapped, and the extent and total
Handling Editor: Jack Williams vegetation biomass of each biome, and total global vegetation biomass, estimated.
Results: Substantial changes in biome extents and locations were found on all vege-
tated continents. Although the largest magnitude changes were in Eurasia, important
changes were seen in tropical latitudes and the Southern Hemisphere. Total global
extent of most biomes varied on multi-millennial (orbital) time scales, although some
(e.g. Tropical Raingreen Forest) responded principally to the c. 100-kyr glacial–inter-
glacial cycle and others (e.g. Temperate Broad-leaved Evergreen Forest) mainly to
the c. 20-kyr precession cycle. Many also responded to millennial contrasts between
stadial (‘hosing’) and interstadial climates, with some (e.g. Tropical Evergreen Forest)
showing stronger responses than to the multi-millennial changes.
Main conclusions: No two time slices had identical biome patterns. Even equiva-
lent Holocene and last interglacial time slices, and the last and penultimate glacial
maxima, showed important differences. Only a small proportion of global land area
experienced no biome change since 140 ka; many places experienced multiple biome
changes. These modelling experiments provided little evidence for long-term biome
stability.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2020 The Authors. Journal of Biogeography published by John Wiley & Sons Ltd

Journal of Biogeography. 2020;00:1–18.  |

wileyonlinelibrary.com/journal/jbi     1
 |
2       ALLEN et al.

KEYWORDS

biomes, carbon mass, glacial, global palaeovegetation maps, HadCM3, Heinrich Events,
interglacial cycle, LPJ-GUESS, millennial climatic fluctuations

1 |  I NTRO D U C TI O N global palaeovegetation maps to be drawn, then analysed to address

these questions.
A recent study of almost 600 palaeovegetation records concluded
that climatic changes since the last glacial maximum (LGM) drove
moderate-to-large changes in both the composition and struc- 2 | M ATE R I A L S A N D M E TH O DS
ture of vegetation at most locations across the globe (Nolan
et  al.,  2018). Vegetation modelling studies have reached similar 2.1 | Palaeoclimate experiments
conclusions, albeit they also emphasize the role of the increase
in atmospheric carbon dioxide concentrations ([CO2]atm) since the An internally consistent series of 88 palaeoclimate experiments,
LGM in driving observed vegetation changes (Prentice, Harrison, including seven ‘hosing’ experiments designed to mimic Heinrich
& Bartlein,  2011). Although such studies contrasting the LGM Events H0–H6, in which a 1-Sv fresh water flux was added to the
and recent past provide valuable insights into global vegetation surface layer in the North Atlantic for 200 years, plus a pre-indus-
changes across the past 21  kyr, both global climate and [CO2]atm trial experiment, were performed using the HadCM3 fully coupled
have fluctuated substantially, on time scales of 10 4–105  years, atmosphere–ocean general circulation model (AOGCM). The design
throughout at least the past 800  kyr (EPICA community mem- of the experiments followed Singarayer and Valdes (2010) with re-
bers, 2004; Jouzel et al., 2007; Luthi et al., 2008), whilst climate spect to the boundary conditions and forcings applied, but with two
has also exhibited strong centennial to millennial fluctuations important exceptions. Firstly, rather than using a static pre-industrial
(Rasmussen et  al.,  2014). Using the LPJ-GUESS dynamic vegeta- land surface for all experiments, the AOGCM was coupled to Triffid
tion model (Smith, Prentice, & Sykes,  2001; Smith et  al.,  2014), (Cox, 2001), a simple DGVM with five plant functional types (PFTs),
Huntley, Allen, Collingham, et al. (2013) showed the sensitivity enabling the vegetation and climatically important landsurface prop-
of vegetation in higher latitudes of the Northern Hemisphere to erties (e.g. forest vs. grassland) to interact dynamically with the at-
the contrasts between simulated stadial and interstadial climates mosphere. The precise model configuration is HadCM3B-M2.1aD,
during the last glacial stage (Singarayer & Valdes, 2010). Other re- as defined in Valdes et  al.  (2017). Secondly, rather than using ice
search has shown a strong correlation in southern Africa between sheets of LGM extent but reduced elevation for pre-LGM time slices
high present diversity of endemic bird species and areas where the of the last glacial stage, and modern ice sheets not only for LIG time
same biome was simulated to have persisted throughout the past slices but also for pre-LIG time slices, more realistic representations
140 kyr (Huntley et al., 2016). of pre-LGM ice sheets were used. For pre-LGM time slices of the last
The objective of this study was to address a number of important glacial stage, ice-sheet configuration used the ICE-5G (Peltier, 2004)
questions relating to the global impacts upon vegetation of both the ice sheet for the post-LGM time slice with the closest matching eus-
climatic changes of various magnitudes and the changes in [CO2]atm tatic sea-level depression. For LIG and pre-LIG time slices, ice sheets
that have taken place during the past 140  kyr, a period extending were specified following de Boer, van de Wal, Lourens, Bintanja, and
from the penultimate glacial maximum (PGM) through the last inter- Reerink (2013). More details of the model set-up can be found in
glacial (LIG), the last glacial stage and the Holocene, to the present. Davies-Barnard, Ridgwell, Singarayer, and Valdes (2017).
Specific questions to be addressed were as follows: What was the For input to LPJ-GUESS, anomalies were calculated for each
magnitude of changes in the extent of forest biomes, especially in palaeoclimate experiment with respect to the pre-industrial exper-
the tropics, across the last glacial–interglacial cycle? How extensive iment, in each case using the climatology of the last 30 years of the
and persistent were the temperate and boreal non-forest biomes experiment (Singarayer & Valdes,  2010). Anomalies were required
considered favourable for Pleistocene mega-herbivores? What was for the twelve monthly mean temperatures (°C), monthly precipita-
the nature and extent of vegetation responses in tropical latitudes tion totals (mm) and monthly mean cloud cover (%).
and the Southern Hemisphere to last glacial stage millennial cli-
matic fluctuations? How similar were LIG global vegetation patterns
to those that characterize the Holocene? How similar were global 2.2 | LPJ-GUESS simulations
vegetation patterns of the PGM and LGM? In order to achieve our
objective, we used the LPJ-GUESS dynamic global vegetation model The LPJ-GUESS DGVM simulates the dynamics of stands of vegeta-
(DGVM; Smith et al., 2001, 2014) to simulate global vegetation for a tion comprised of a series of PFTs defined by the user. PFTs that
series of time slices, 0–140 ka BP, for which climatic conditions had can coexist in a given climate compete for resources, their functional
been simulated. The LPJ-GUESS simulations were used to infer the traits determining their competitive balance and, hence, their rela-
extent and location of a series of biomes for each time slice, allowing tive biomass and fractional cover. The fire module used here (Smith
ALLEN et al. |
      3

et al., 2014; Thonicke, Venevsky, Sitch, & Cramer, 2001) simulates fire thin-plate spline surfaces fitted to data for land grid cells in the CRU
disturbance and its effects based on soil moisture (as a proxy for fuel TS 3.0 dataset (Harris, Jones, Osborn, & Lister, 2014). Palaeoclimatic
moisture), fuel load and a PFT-specific fire resistance. In addition, a conditions for each grid cell, including those on shelf areas, were
stand-replacing stochastic disturbance with pre-defined average re- obtained following the approach of Miller et  al.  (2008). Anomalies
turn interval represents other disturbances, such as wind storms and derived from the palaeoclimate experiments were applied to the
pest attacks. As nitrogen deposition for past periods is uncertain, CRU TS 3.0 annual time series of climatic data for 1901–30, monthly
the model was run without the nitrogen cycle and nitrogen limita- mean temperature time series first being detrended to remove the
tion (i.e. the C-only version of Smith et al., 2014). The vegetation of warming trend seen over those three decades. This time series was
each cell of a 0.5 × 0.5° longitude × latitude grid, comprising all grid then cycled through repeatedly as required. Atmospheric carbon di-
cells that had non-zero land area for at least one time slice, was simu- oxide concentration for each time slice was determined according to
lated. Following Huntley, Allen, Barnard, Collingham, and Holliday Vostok ice-core data (Loulergue et al., 2008; Petit et al., 1999) using
(2013), the potential 1901–30 monthly mean temperature, monthly the EDC3 chronology (Parrenin et  al.,  2007). Obliquity was speci-
precipitation and monthly mean cloudiness of grid cells on shelf fied following Laskar et al. (2004). Twenty PFTs were used (Table 1).
areas exposed by lowered glacial sea levels were estimated using Eleven tree and two grass PFTs were specified following Forrest

TA B L E 1   Plant functional types

Acronym Plant functional type

BNE Boreal Needle-leaved Evergreen Tree

BINE Shade-intolerant Boreal Needle-leaved Evergreen Tree
BNS Boreal Needle-leaved Summergreen Tree
BIBS Shade-intolerant Boreal Broad-leaved Summergreen Tree
TeNE Temperate Needle-leaved Evergreen Tree
TeBS Temperate Broad-leaved Summergreen Tree
TeIBS Shade-intolerant Temperate Broad-leaved Summergreen Tree
TeBE Temperate Broad-leaved Evergreen Tree
TrBE Tropical Broad-leaved Evergreen Tree
TrIBE Shade-intolerant Tropical Broad-leaved Evergreen Tree
TrBR Tropical Broad-leaved Raingreen Tree
BESh Boreal Evergreen Shrub
BSSh Boreal Summergreen Shrub
TeESh Temperate Evergreen Shrub
TeRSh Temperate Raingreen Shrub
TeSSh Temperate Summergreen Shrub
TrESh Tropical Evergreen Shrub
TrRSh Tropical Raingreen Shrub
C3G C3 Grass
C4G C4 Grass
PFT Aggregates used in biome inference
B-tree Sum of Boreal Trees (BNE + BINE  + BNS + BIBS)
Te-tree Sum of Temperate Trees (TeNE + TeBS +TeIBS + TeBE)
Tr-tree Sum of Tropical Trees (TrBE + TrIBE +TrBR)
B-shrub Sum of Boreal Shrubs (BESh + BSSh)
Te-shrub Sum of Temperate Shrubs (TeESh + TeRSh +TeSSh)
Tr-shrub Sum of Tropical Shrubs (TrESh + TrRSh)
BE_tree Sum of Boreal Evergreen Trees (BNE + BINE)
BS_tree Sum of Boreal Summergreen Trees (BNS + BIBS)
TeE_tree Sum of Temperate Evergreen Trees (TeNE + TeBE)
TeS_tree Sum of Temperate Summergreen Trees (TeBS + TeIBS)
TrE_tree Sum of Tropical Evergreen Trees (TrBE + TrIBE)
 |
4       ALLEN et al.

et al. (2015), whilst seven shrub PFTs were newly defined. Attribute TA B L E 2   Simulated biomes
values for the shrub PFTs were estimated on the basis of information
Acronym Biome
from the literature. Attribute specifications for all PFTs are given in
Des Desert
Appendix S1.
Se-des Semi-desert
For each grid cell, 25 replicate 0.1-ha stands were simulated. For
each stand, the model was run for 590 simulated years; average re- TrG Tropical Grassland

turn time for stand-destroying disturbances was 100 years. Grid cell Sav Savanna

values of vegetation attributes were computed as the overall mean TrRF Tropical Raingreen Forest
of the mean simulated values for the last 90 years for the replicate TrEF Tropical Evergreen Forest
stands. Simulations were made for each of the 88 time slices for TeSh Temperate Shrubland
which palaeoclimate experiment results were available, as well as for WTeWo Warm Temperate Woodland
the ‘present’ (i.e. the pre-industrial experiment). For the latter, cli- TeBEF Temperate Broad-leaved Evergreen Forest
mate data for 1901–30 were used without application of any anom- TeSF Temperate Summergreen Forest
alies, atmospheric carbon dioxide concentration was 280 ppmv and
TeNEF Temperate Needle-leaved Evergreen Forest
obliquity specified at its contemporary value. Two variables com-
TeMxF Temperate Mixed Forest
puted by LPJ-GUESS were analysed for each PFT and for each 0.5º
TePk Temperate Parkland
grid cell: vegetation biomass (carbon mass) and leaf area index (LAI).
St Steppe
For carbon mass, grid cell totals for each PFT were computed using
BPk Boreal Parkland
the estimated ice-free land area of the grid cell at that time (see
Appendix S2). BENF Boreal Evergreen Needle-leaved Forest
BSNF Boreal Summergreen Needle-leaved Forest
BSBF Boreal Summergreen Broad-leaved Forest
2.3 | Biome inference BWo Boreal Woodland
ShT Shrub Tundra
The biome for each 0.5° grid cell and during each time slice was in- Tun Tundra
ferred from the simulated values of LAI and carbon mass for the 20 UNC Unclassified
PFTs. A total of 21 biomes was recognized (Table 2). Inference was
performed using a series of rules applied sequentially (expressed in
the form of a dichotomous key in Appendix 1) and implemented in a (e.g. Olson, Watts, & Allison, 1983; Olson et al., 2001). Discrepancies
Fortran program written for the purpose (see Appendix S3). In addi- between the simulated and published biome distributions were then
tion to reporting the inferred biome for each grid cell, this program progressively minimized by iteratively tuning the criteria used in
also reported for each time slice: the extent of each biome, both as the rule set, re-applying the rules and mapping the resulting biome
a total area (km2) and as percentage of ice-free land area; the num- distribution. This process resulted in the addition of two further
ber and percentage of grid cells with ice-free land assigned to each biomes to the set being inferred, and to the final carbon mass and
biome; and the total carbon mass (kg) for each biome. The global LAI criteria used for assigning the 21 biomes (see Appendix 1). No
total of ice-free land area, number of grid cells with ice-free land and quantitative assessment of the match to any particular published
global total carbon mass were also calculated and reported for each biome map was attempted, not only because none of these define
time slice. For each biome, the total carbon mass of each PFT, and or portray quite the same set of biomes, but, in addition, as Scheiter
the percentage of the biome's total carbon mass contributed by each and Higgins (2009) showed, the accuracy of the match is strongly
PFT, were also calculated and reported for each time slice. influenced by the choice of map with which to make the comparison.
Using maps of the simulated carbon mass and LAI for the present Although all grid cells had a biome inferred for the ‘present’ sim-
for the 20 PFTs and 11 PFT aggregates, as well as a map of total ulation, a small number of grid cells (median 3, maximum 14, out of a
simulated carbon mass, and a proposed set of 19 biomes, the ini- median overall number of grid cells with ice-free land of 62,919) was
tial rule set was developed, guided principally by expectations as to unclassified for 81 of the 88 other time slices. The median areal ex-
which PFTs would characterize each biome, and applied. Whereas tent of unclassified grid cells was 0.005% of global ice-free land area,
most biomes could be distinguished using LAI criteria for associated with a maximum extent of 0.026% at 126 ka. Inspection of the sim-
PFTs, a carbon mass threshold was needed to distinguish desert and ulated PFT profiles of unclassified grid cells revealed no consistent
tundra biomes from the remainder, the initial threshold value being pattern with respect to their PFT composition; thus we concluded
assigned by inspection of the map of total simulated carbon mass. that they represent neither an additional no-analogue biome nor a
The resulting global distribution of simulated biomes was mapped consistent location in climatic space.
and compared visually with published maps of simulated potential Global biome distributions were mapped for all time slices, with
pre-industrial biomes (e.g. Prentice et al., 1992, 2011) and of present ice-sheet and ocean masks overlain onto each map, and the maps
biomes inferred from remaining natural and semi-natural vegetation compiled as a slide show (see Supplementary Material Slideshow).
ALLEN et al. |
      5

In addition, difference maps were drawn and transition matrices 3.2 | Impacts of orbitally forced climatic changes on
calculated for time slices compared when discussing the questions global biome patterns and carbon mass
we sought to address [see Appendix S4 for description of method,
difference maps (Figures  S4.1–S4.26) and transition matrices Biome distributions at 22, 10, 114 and 126 ka provide an opportu-
(Tables S4.4–S4.29)]. nity to explore the effects of orbital-scale changes in insolation and
climate on biome distributions and terrestrial carbon mass (Figure 2;
see also Figures S4.1–S4.5 and Tables S4.4–S4.8 for difference maps
3 | R E S U LT S and transition matrices). The first two time slices represent the
Northern Hemisphere (55°N) last glacial minimum (22.4  ka) and
3.1 | Simulated present biomes Holocene maximum (9.7 ka) of summer (June 21st–July 20th) insola-
tion, whilst the latter two represent the extreme minimum (114.7 ka)
Tropical Evergreen Forest (TrEF– see Table 2 for all biome acronyms and maximum (125.4  ka), respectively, of Northern Hemisphere
used below) is the most spatially extensive biome at present (Table 3; summer insolation since 140 ka.
Figure 1), followed closely by Des, TrRF and TeSh. TrEF, however, Although global ice-free land area at 22  ka is almost identi-
makes substantially the largest contribution to global carbon mass, cal to that today, ice-covered areas being almost exactly com-
accounting for 37.76%, with the next largest contributor, TrRF, ac- pensated for by exposed shelf areas, global total carbon mass
counting for just 14.96% and only three other biomes exceeding 5% at 22  ka is only 61.2% of that simulated for the present. This
(TeBEF – 8.27%; BENF – 6.94%; TeSF – 6.14%). reflects more or less doubled extents of Des and ShT, an almost
7-fold increase in the extent of Tun, and substantial increases in
the extent of St, TrG, Sav and WTeWo, all biomes with relatively
low carbon mass per unit area, along with reduced extents of
several biomes with relatively high carbon mass per unit area.
TA B L E 3   Ice-free land area and carbon mass of simulated
present biomes TeMxF and TrRF both have c.  3-fold reductions in extent, TeSF
is reduced to 38.5% and BENF is reduced 4-fold, whilst TrEF, the
Ice-free land
biome with the highest present carbon mass per unit area, is re-
area Carbon mass
duced by nearly 10%.
(% global At 10 ka, despite the residual ice sheets, especially the Laurentide
Biome acronym (Mm2) (Pg) total)
Ice Sheet in North America, global ice-free land area is almost 5%
Des 15.5 1.88 0.28 greater than today. More strikingly, global total carbon mass is 97.5%
Se-des 4.9 4.21 0.62 of that simulated for the present, and hence almost 60% greater than
TrG 4.6 3.08 0.45 at 22 ka. This primarily reflects greater extents of TrEF (+6.1%) and
Sav 5.7 8.87 1.30 TrRF (+5.4%), as well as slightly increased carbon mass per unit area
TrRF 14.0 102.00 14.96 of TeEF (+1.2%), compared to present.

TrEF 15.7 258.00 37.76

Notwithstanding 114 ka corresponding to the extreme Northern
Hemisphere summer insolation minimum, estimated ice-sheet ex-
TeSh 13.4 25.10 3.68
tent is similar to that at 10 ka, and ice-free land area is c. 4% greater
WTeWo 2.6 6.56 0.96
than today. Somewhat unexpectedly, global carbon mass is 101.2%
TeBEF 7.9 56.50 8.27
of that simulated for the present, despite marginally smaller extents
TeSF 5.4 41.90 6.14
of TrEF (−0.9%) and TrRF (−1.9%) than present, and decreased carbon
TeNEF 0.4 3.14 0.46
mass per unit area of both compared to present (TrEF −1.4%, TrRF
TeMxF 3.9 31.10 4.55 −10.0%). Four temperate and boreal forest biomes (TeMxF, TeSF,
TePk 3.6 17.40 2.54 TeBEF and BENF), that have higher carbon mass per unit area than
St 1.9 1.51 0.22 at present, make important contributions to the higher global carbon
BPk 6.6 33.40 4.89 mass, with TeMxF making a 56% greater contribution to global car-
BENF 9.1 47.40 6.94 bon mass than at present as a result of a combination of 38% greater
BSNF 0.5 1.49 0.22 extent and 76% higher carbon mass per unit area.

BSBF 3.3 13.60 1.99 Ice-free land area at 126  ka also is greater than at present, by
about 2.7%, whereas global carbon mass is only 90.8% of that simu-
BWo 4.0 11.60 1.70
lated for today, principally due to reduced extents (TrEF −3.6%, TrRF
ShT 6.2 14.00 2.05
−2.4%) and decreased carbon mass per unit area (TrEF −6.8%, TrRF
Tun 0.9 0.18 0.03
−9.4%) of tropical forest biomes.
UNC 0.0 0.00 0.00
Comparing biome distributions for the four time slices
Total 130.1 683 100.00
(Figure 2) reveals striking differences; none are closely similar. The
 |
6       ALLEN et al.

180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N 80° N

Arctic Circle
60° N 60° N

40° N 40° N

Atlantic Ocean
Tropic of Cancer

20° N 20° N

Pacific Ocean
Pacific Ocean
Equator
0° 0°

Indian Ocean

20° S 20° S
Tropic of Capricorn

40° S 40° S

60° S Antarctic Circle 60° S

80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

Legend
Desert Warm Temperate Woodland Boreal Parkland
Semi-desert Temperate Broad-leaved Evergreen Forest Boreal Evergreen Needle-leaved Forest
Tropical Grassland Temperate Summergreen Forest Boreal Summergreen Needle-leaved Forest
Savanna Temperate Needle-leaved Evergreen Forest Boreal Summergreen Broad-leaved Forest
Tropical Raingreen Forest Temperate Mixed Forest Boreal Woodland
Tropical Evergreen Forest Temperate Parkland Shrub Tundra
Temperate Shrubland Steppe Tundra
Unclassified Ice sheet Ocean or lake

F I G U R E 1   Simulated potential ‘present’ biome distribution. Global distribution of biomes inferred from an LPJ-GUESS simulation
driven by early twentieth century (1901–30) climatic conditions, pre-industrial atmospheric carbon dioxide concentration of 280 ppmv and
contemporary obliquity

most eye-catching differences are on the northern continents, es- 126 ka the extent of WTeWo is markedly reduced and those of Sav
pecially Eurasia being dominated by Des, Tun and ShT at 22 ka, with and TrG are increased.
very restricted extents of temperate and boreal forest biomes and a At 22 ka, areas in Central and northern South America occupied
78% increase in the extent of St. In contrast, the northern continents today by TrEF are instead occupied by TrRF or Sav, whereas much
at the 114 ka insolation minimum are much more similar to present of the area of TrRF to the south of Amazonia today is replaced by
(Figure 1), except for northern North America, where the influence TrEF. In addition, the western desert area is increased in extent, as
of the ice sheet is apparent, and north-eastern Eurasia, where forest is the area of WTeWo east of the Andes, whilst southernmost parts
biomes extend much further north than today. The patterns at 10 ka of the continent today supporting principally TeSh and St are occu-
on the northern continents are again broadly similar to those today, pied instead by ShT and Des. In contrast, at the 114 ka insolation
although the extent of BENF is much reduced. The most similar pat- minimum biome patterns in Central and South America are much
terns to today on the northern continents are at 126 ka, despite this more similar to those today, although with the extent of TrEF some-
being the peak insolation maximum and the present a modest inso- what reduced, being replaced by TrRF in a belt along the Atlantic
lation minimum. coast of Amazonia. At the 10  ka Northern Hemisphere insolation
Biome differences in tropical and southern latitudes are also im- maximum, the extent of TrEF is increased, especially to the south of
portant. In sub-Saharan Africa, Sav and TrRF extend substantially Amazonia where this biome occupies areas today supporting TrRF
further north at the insolation maxima of 10 and 126 ka than they or Sav. The extent of WTeWo is also increased south of the Tropic
do at the minima of 22, 114 ka and today. South of the equator in of Capricorn, mostly into areas today occupied by TeBEF. Again,
Africa, areas occupied today by TrRF and Sav are at 22 ka simulated biome patterns at the 126 ka insolation maximum show a contrast-
to be largely replaced by TeBEF and WTeWo, whereas at 114 ka dif- ing pattern of differences from those today. However, as at 10 ka,
ferences from the present are much less striking. A similar contrast there is a greater extent of WTeWo and a reduced extent of TeBEF,
is seen between the insolation maxima, with 10 ka showing similar and the extent of TrEF is decreased, with TrRF and Sav occupying
biome patterns in southern Africa to those at present, whereas at increased areas.
ALLEN et al.       7 |
(a) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N
(b) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Circle Arctic Circle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tropic of Cancer Tropic of Cancer

20° N 20° N 20° N 20° N

Pacific Ocean Pacific Ocean

Pacific Ocean Pacific Ocean
Equator Equator
0° 0° 0° 0°

Indian Ocean Indian Ocean

20° S 20° S 20° S 20° S

Tropic of Capricorn Tropic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Circle 60° S 60° S Antarctic Circle 60° S

22 ka BP 10 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

(c) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N (d) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Circle Arctic Circle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tropic of Cancer Tropic of Cancer

20° N 20° N 20° N 20° N

Pacific Ocean Pacific Ocean

Pacific Ocean Pacific Ocean
Equator Equator
0° 0° 0° 0°

Indian Ocean Indian Ocean

20° S 20° S 20° S 20° S

Tropic of Capricorn Tropic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Circle 60° S 60° S Antarctic Circle 60° S

114 ka BP 126 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

F I G U R E 2   Simulated biomes for Northern Hemisphere summer insolation minima and maxima. Biomes inferred from LPJ-GUESS
simulations driven by climate scenarios generated from palaeoclimate experiments for 22 ka (a), which is the time slice closest to the last
glacial minimum of Northern Hemisphere summer insolation; 10 ka (b), which is the time slice closest to the early Holocene maximum of
Northern Hemisphere summer insolation; 114 ka (c), which is the time slice closest to the most extreme minimum of Northern Hemisphere
summer insolation since 140 ka; and 126 ka (d), which is the time slice closest to the most extreme maximum of Northern Hemisphere
summer insolation since 140 ka. (Legend as in Figure 1)

In south-east Asia and Australasia, aside from striking changes in are particularly clearly established for Greenland from ice-core data
land–ocean configuration resulting from changes in global eustatic (Wolff, Chappellaz, Blunier, Rasmussen, & Svensson,  2010), with
sea level, the most striking contrast with today is at 22  ka, when the most rapid and large magnitude warming episodes being as-
Sav is simulated to occupy large areas today supporting TrRF, TrG sociated with Heinrich Events (Andrews,  1998). Figure 3 (see also
occupies much of the Indian sub-continent and the extent of Des in Figures  S4.6–S4.11 and Tables  S4.9–S4.14) presents biome maps
central Australia is very much greater than today. In contrast, biome for two time slices, 24 and 32 ka, corresponding to Heinrich Events
patterns in the region at 114  ka are much more similar to those H2 and H3. In each case biomes were simulated using the results
today, apart from on the Indian sub-continent where TrG occupies from each of two palaeoclimatic experiments, the first, referred to
large areas today supporting Sav or TrRF. At 10 ka biome patterns hereafter as the ‘normal’ experiment, driven only by ‘slow’ forcings
are most similar to those today, the principal exception being the (orbital configuration, atmospheric greenhouse gas concentrations,
reduced extent of TrEF in eastern parts of the south-east Asian ice-sheet configuration and land–ocean palaeogeography), and the
mainland that instead are occupied by an extension of the area of second, designed to mimic the Heinrich Event and referred to as the
TrRF. Biome patterns at 126 ka are even more similar to those of the ‘hosing’ experiment, additionally forced with a 1-Sv freshwater flux
present. into the North Atlantic (Singarayer & Valdes, 2010).
For H2, which occurred at 24 ka close to the LGM minimum of
Northern Hemisphere summer insolation, the contrast in biome
3.3 | Impacts of last glacial stage millennial climatic patterns between the two experiments (Figure 3a,b) is striking, es-
fluctuations on global biomes pecially in Eurasia, where the extent of forest biomes (principally
BENF, BSNF, BSBF and TeBEF) simulated for the normal experiment
The last glacial stage is characterized by repeated alternations be- is very much greater than for the hosing experiment, forest biomes
tween relatively warm interstadials and severely cold stadials, as being almost entirely absent from western Europe for the latter (c.f.
demonstrated by a range of Northern Hemisphere palaeoclimatic Huntley, Allen, Collingham, et al., 2013). Striking contrasts are also
records spanning the last glacial stage and that have sufficient tem- seen elsewhere around the globe, notably in Central America and
poral resolution to record these centennial to millennial fluctuations northern South America, where large areas of TrEF and TrRF sim-
(e.g. Allen et al., 1999; Allen, Watts, & Huntley, 2000). The magnitude ulated for the normal experiment are replaced by Sav, TrG or Des
and rate of climatic warming from stadials to succeeding interstadials in the hosing experiment. In sub-Saharan Africa, the extent of TrEF
 |
8       ALLEN et al.

(a) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N (b) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Ci rcle
Arctic Ci rcle
60° N 60° N
60° N 60° N

40° N 40° N
40° N 40° N

Atlantic Ocean Atlantic Ocean

Tr opic of Cancer
Tr opic of Cancer
20° N 20° N 20° N 20° N

Pa cific Ocean Pa cific Ocean

Pa cific Ocean Pa cific Ocean
Equator Equator
0° 0° 0° 0°

In dian Ocean In dian Ocean

20° S 20° S 20° S 20° S

Tr opic of Capricorn Tr opic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Cir cle 60° S 60° S Antarctic Cir cle 60° S

24 ka BP H2 - 24 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

(c) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N (d) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Ci rcle Arctic Ci rcle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tr opic of Cancer Tr opic of Cancer
20° N 20° N 20° N 20° N

Pa cific Ocean Pa cific Ocean

Pa cific Ocean Pa cific Ocean
Equator Equator
0° 0° 0° 0°

In dian Ocean In dian Ocean

20° S 20° S 20° S 20° S

Tr opic of Capricorn Tr opic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Cir cle 60° S 60° S 60° S

Antarctic Cir cle

32 ka BP H3 - 32 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

F I G U R E 3   Simulated biomes for inter-stadial and stadial intervals of the last glacial stage. Biomes inferred from LPJ-GUESS simulations
driven by climate scenarios generated from the ‘normal’ (a & c) and ‘hosing’ (b & d) palaeoclimate experiments for 24 ka (a & b) and 32 ka (c
& d). Whereas the ‘normal’ experiments give climatic conditions corresponding to those of inter-stadials, the ‘hosing’ experiments, which are
designed to mimic Heinrich Events H2 and H3, respectively, give stadial climatic conditions. (Legend as in Figure 1)

is less for the hosing experiment, as is the extent of Sav, whilst in 3.4 | Holocene and last interglacial global
the Indian sub-continent much of the area occupied by TrG for the biome patterns
normal experiment is Des for the hosing experiment. In South Africa,
the continuous belt of TeBEF that extends from the Western Cape Comparisons of the current and last interglacials test the degree
eastwards to the Mozambique border and beyond for the normal ex- to which biome distributions are similar or different between in-
periment is, for the hosing experiment, fragmented and replaced by terglacial periods (Figure 4; see also Figures  S4.12–S4.23 and
WTeWo, Se-des and TePk. Similar more localized changes are seen Tables S4.15–S4.26). Whilst there are some clear parallels, there are
in other regions, for example in Japan where ShT extends further also some striking differences. These differences partially reflect the
south in the hosing experiment, mostly replacing BENF that in turn longer duration of the LIG, for which we follow Brauer et al. (2007)
is shifted southwards, largely replacing TeMxF, the area of which is who estimate the duration as 17.7 ± 0.2 kyr (127.2 to 109.5 ka), and
consequently reduced as it is squeezed against the southern coast. thus c. 6 kyr longer than the Holocene. This longer duration means
In contrast to H2, H3 occurred close to a peak in Northern that by the end of the LIG, at 109.5 ka, Northern Hemisphere sum-
Hemisphere summer insolation at 33 ka. The 32 ka/H3 biome pat- mer insolation had passed its minimum at 114.7 ka and had already
terns (Figure 3c,d) differ much less than for 24 ka/H2, especially in increased by 34% of the difference between that minimum and
Eurasia. Nonetheless, in the normal experiment, much of western the subsequent maximum at 103.1  ka. However, ice sheets are in-
Europe is simulated to have supported forest biomes, principally ferred to have been present in North America by the middle, and in
BSBF, with St towards the western fringes, whilst for the hosing Fennoscandia by the later millennia, of the LIG (Figure 4e,f), whereas
experiment ShT dominates the region, with Tun and Des in the ex- no ice sheet was present in either area during later millennia of the
treme west and north-west. Elsewhere, marked contrasts in biome Holocene.
patterns are seen in Central America and northern South America, Although there are obvious differences among all six maps pre-
and in sub-Saharan Africa, largely paralleling those seen for 24 ka/ sented in Figure 4, three are more similar, and also more similar to
H2. In Australia, the normal experiment has a much smaller and frag- that for the present (Figure 1), namely those for 7  ka (Figure 4b),
mented area of Des in the centre of the continent than does either 3 ka (Figure 4c) and 126 ka (Figure 4d). Of these, the first two repre-
the normal or hosing experiment for 24 ka, whereas the area of Des sent the mid- and late-Holocene (Walker et al., 2012), the first falling
is much increased in the H3 hosing experiment. The overall biome some 4,700 yr after the Holocene onset at 11.7 ka, whereas 126 ka
pattern in Australia at 32 ka also markedly differs from 24 ka, with a is only 1,200 yr after the LIG onset, as dated by Brauer et al. (2007),
much reduced extent of TeSh and increased extents of TrG and Sav and thus closest in timing relative to interglacial onset to the early
in the north. Holocene time slice at 11 ka (Figure 4a). Notable differences include:
ALLEN et al.       9 |
(a) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N
(d) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Ci rcle Arctic Ci rcle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tr opic of Cancer Tr opic of Cancer

20° N 20° N 20° N 20° N

Pa cific Ocean Pa cific Ocean

Pa cific Ocean Pa cific Ocean
Equator Equator
0° 0° 0° 0°

In dian Ocean In dian Ocean

20° S 20° S 20° S 20° S

Tr opic of Capricorn Tr opic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Cir cle 60° S 60° S Antarctic Cir cle 60° S

11 ka BP 126 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

(b) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N
(e) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Ci rcle Arctic Ci rcle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tr opic of Cancer Tr opic of Cancer

20° N 20° N 20° N 20° N

Pa cific Ocean Pa cific Ocean

Pa cific Ocean Pa cific Ocean
Equator Equator
0° 0° 0° 0°

In dian Ocean In dian Ocean

20° S 20° S 20° S 20° S

Tr opic of Capricorn Tr opic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Cir cle 60° S 60° S Antarctic Cir cle 60° S

7 ka BP 118 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

(c) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N
(f) 80° N
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N

Arctic Ci rcle Arctic Ci rcle

60° N 60° N 60° N 60° N

40° N 40° N 40° N 40° N

Atlantic Ocean Atlantic Ocean

Tr opic of Cancer Tr opic of Cancer

20° N 20° N 20° N 20° N

Pa cific Ocean Pa cific Ocean

Pa cific Ocean Pa cific Ocean
Equator Equator
0° 0° 0° 0°

In dian Ocean In dian Ocean

20° S 20° S 20° S 20° S

Tr opic of Capricorn Tr opic of Capricorn

40° S 40° S 40° S 40° S

60° S Antarctic Cir cle 60° S 60° S Antarctic Cir cle 60° S

3 ka BP 112 ka BP
80° S 80° S 80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180° 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

F I G U R E 4   Simulated biomes for the Holocene and the last interglacial stage. Biomes inferred from LPJ-GUESS simulations driven by
climate scenarios generated from palaeoclimate experiments for time slices representing early, middle and late in each of the Holocene and
last interglacial: (a) 11 ka; (b) 7 ka; (c) 3 ka; (d) 126 ka; (e) 118 ka; and (f) 112 ka. (Legend as in Figure 1)

greater extent of the belt of TeBEF in eastern North America at extents of Des, St and TeSh are less, and of Se-des greater, than
126  ka; more northerly location of the boundary between TeMxF today.
and BENF in Europe at 126  ka; greater extent of ShT in western • At 118 ka (Figure 4e), ShT is more extensive in northern Eurasia
Siberia at 126  ka; smaller extent of Sav and greater extent of TrG than it is today, whereas BWo is much less extensive. BENF forms
south of the Sahara at 3 ka; and greater extent of TrG in Australia a more continuous belt across Eurasia, especially in the east, than
at 126 ka. today, as well as extending further north in northern Europe and
The other three biome distributions have substantial differences western Siberia. In North America, the extent of BENF is greater
from the present (Figure 1), including the following: than at present in the northern conterminous United States and
Canada, mostly at the expense of BPk which is of much lesser
• At 11  ka (Figure 4a), the extent of ShT in Siberia is much less extent. In the Indian sub-continent, TrRF is absent except in the
than today, whereas BWo is more extensive and areas of BSNF, south west, and the extent of Sav is reduced, and of TrG increased,
BSBF and BPk are simulated in northern Siberia well north of compared to present. Sub-Saharan Africa and South America
their northernmost areas simulated for the present. A similar exhibit few striking differences from the present, whereas in
more northerly extension of BSNF, BSBF, BENF and BPk is seen Australia the extent of Des is much greater, and of TrG and Sav
in Alaska. In sub-Saharan Africa, Sav extends further north than much less, than today.
today, and TrG is less extensive south of the Sahara. WTeWo is • Global biome patterns at 112  ka  BP (Figure 4f) are strikingly
more extensive in the north of the Indian sub-continent, whereas different from those today, and more similar to those at 11  ka
Sav and TrRF are both less extensive than today. In Australia, than to those of any other LIG or Holocene time slice illustrated.
 |
10       ALLEN et al.

Notwithstanding this general similarity, there are nonetheless no- and troughs in Northern Hemisphere summer insolation (Figure 7),
table differences between biome patterns at 112 ka and at 11 ka. notably TeBEF, BPk, BENF and BSNF. However, some, for example
In Europe, BENF extends to the British Isles at 112  ka; the ex- TeBEF, have greater extents during insolation minima and others, for
tent of TeSF is much less, and of TeMxF greater, than at 11 ka. In example BSNF, during insolation maxima.
sub-Saharan Africa Sav is less extensive, with the belt south of Simulated global total carbon mass is highest during the
the Sahara extending less far north and much reduced, whereas Holocene and LIG, and reaches minimum values during the LGM
TrG is more extensive in that region, but again extends less far and PGM (Figure 7). Overall changes in carbon mass are very sim-
north, compared both to 11 ka and to today. In Central America ilar to those calculated from the TRIFFID model by Davies-Barnard
and northern South America biome patterns are relatively similar et al. (2017), although that study did not include the Heinrich events.
at 112 ka to those at 11 ka, with a greater extent of Sav and less Values are intermediate during most of the last glacial stage, but
of TrEF than today, whereas further south in South America, and with strongly reduced values in simulations made for the Heinrich
in Australia, 112 ka patterns are more similar overall to those at experiments. Several peak values correspond to minima in Northern
present. In the Indian sub-continent, the strongest similarity is to Hemisphere summer insolation, although this pattern is not seen at
the biome patterns at 118 ka. the LGM and PGM, nor at the subdued insolation minimum centred
on 41.8  ka. There is also a general correspondence between the
changing [CO2]atm and simulated global total carbon mass (Figure 7),
3.5 | Last and penultimate glacial maximum global although detailed comparison reveals that peak [CO2]atm values do
biome patterns not generally correspond closely to peak global carbon mass. Thus,
whilst changing [CO2]atm almost certainly contributes to the ‘first-or-
Biome patterns for 21 ka, the time slice most frequently used to rep- der’ pattern of changes in simulated carbon mass between glacial
resent the LGM, and for 140  ka, used here to represent the PGM and interglacial conditions, the ‘second-order’ variations in carbon
(Figure 5; see also Figures  S4.24–S4.26 and Tables  S4.27–S4.29)), mass are more closely linked to insolation variation, and hence to
both show marked differences from present biome patterns (Figure 1), consequent changes in global climatic conditions, perhaps espe-
as expected. Less expected, however, are marked differences between cially in regions with strong climatic seasonality. Marked reductions
LGM and PGM biome patterns in many parts of the world. These in- in global carbon mass in simulations for Heinrich Event mimicking
clude: temperate (TeSF, TeBEF, WTeWo, TeNEF) and Boreal (BSBF, experiments represent an intriguing ‘third-order’ pattern of inverse
BSNF, BENF, BWo) forest and woodland biomes are less extensive in relationship, the ice-core record showing [CO2]atm rising to peak val-
Europe and western Siberia at the PGM than at the LGM, whereas ues coincident with Heinrich Events. This raises the possibility that
Des, ShT and St are more extensive, especially in Europe; the PGM these peaks in [CO2]atm were driven at least in part, if not primarily,
has greater extents of TeBEF and TeSh in the south-east and south- by releases of carbon from, and reduced carbon uptake by, global
west, respectively, of North America, and of TrEF and TrRF in Central vegetation during Heinrich Events. The simulated median reduction
America and northern South America; in southern Amazonia and sub- in global carbon mass of 68.5 Pg (range 28–91 Pg) during Heinrich
Saharan Africa, TrEF is less, and TrRF more extensive at the PGM; and Events H0–H6 is more than sufficient to account for the increased
in southern Africa, TeBEF is less extensive and WTeWo and TeSh more [CO2]atm recorded in ice-core data if this carbon was returned to
extensive at the PGM. the atmosphere, and such a source is also consistent with observed
shifts in δ13C of atmospheric CO2 (Bauska et al., 2018).

3.6 | Global biome extents and carbon mass 140 ka

to present 4 | D I S CU S S I O N

Over the last 140 ky, notable features of global biome distributions Visual comparisons showed general concurrence between our pre-
include the generally large extent of Des (Figure 6), with peak extents sent biome distribution pattern and published global biome maps
during the maxima of the penultimate and last glacial stages, and fur- (e.g. Haxeltine & Prentice, 1996; Prentice et al., 2011). Similarly, vis-
ther increases in the Heinrich hosing simulations. Unsurprisingly, Tun ual comparisons of our past biome distributions with published pal-
and ShT, and to a lesser extent St, are also more extensive during gla- aeovegetation maps, whether derived from modelling (e.g. Guetter
cials, and further increase in extent in Heinrich Event simulations. In & Kutzbach,  1990; Prentice et  al.,  2011) or palaeovegetation data
contrast, TrRG, TeSF, TeMxF, BPk and BENF are most extensive dur- (e.g. Jolly et  al.,  1998; Marchant et  al.,  2009; Tarasov et  al.,  1998;
ing the Holocene and LIG, but show limited changes in global extent Williams, Summers, & Webb, 1998), showed general concurrence of
for Heinrich Event simulations. Several biomes show limited changes overall global patterns. Exceptions occurred mainly when compar-
in global extent, most notably TrEF, although this biome exhibits ing with biomes inferred from palynological data or derived from
marked decreases in extent under Heinrich simulations matched older palaeoclimate simulations using earlier vegetation models (e.g.
by marked increases in TrG. Several biomes tend to increase and BIOME 1, Prentice et  al.,  1992). Closest matches were with more
decrease in extent associated with the precessionally driven peaks recent DGVM studies (e.g. Prentice et al., 2011).
ALLEN et al. |
      11

(a) 180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° N 80° N

Arctic Ci rcle
60° N 60° N

40° N 40° N

Atlantic Ocean
Tr opic of Cancer

20° N 20° N

Pa cific Ocean
Pa cific Ocean
Equator
0° 0°

In dian Ocean

20° S 20° S
Tr opic of Capricorn

40° S 40° S

60° S Antarctic Cir cle 60° S

21 ka BP 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
80° S

180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°
(b) 80° N 80° N

Arctic Ci rcle
60° N 60° N

40° N 40° N

Atlantic Ocean
Tr opic of Cancer

20° N 20° N

Pa cific Ocean
Pa cific Ocean
Equator
0° 0°

In dian Ocean

20° S 20° S
Tr opic of Capricorn

40° S 40° S

60° S Antarctic Cir cle 60° S

140 ka BP
80° S 80° S
180° 160° W 140° W 120° W 100° W 80° W 60° W 40° W 20° W 0° 20° E 40° E 60° E 80° E 100° E 120° E 140° E 160° E 180°

F I G U R E 5   Simulated biomes for the last and penultimate glacial maxima. Biomes inferred from LPJ-GUESS simulations driven by climate
scenarios generated from palaeoclimate experiments for 21 ka (a), generally taken as the last glacial maximum, and 140 ka (b), a time slice
close to the penultimate glacial maximum. (Legend as in Figure 1)

As stated in Section 1, our study aimed to address several ques- cycle, including those for climatic conditions produced by climate
tions. The first related to the magnitude of changes in the extent model experiments designed to mimic Heinrich Events, give a ratio
of forest biomes, especially in the tropics, over the last glacial–in- of 2.1 between the maximum extent, at 11  ka, and minimum ex-
terglacial cycle. Prentice et  al.  (2011) simulated a ratio of 1.2 be- tent, for the H1-mimicking simulation at 17 ka, of TrEF + TrRF. For
tween the present and LGM extents of Tropical Forest, whereas we temperate forests (TeMxF + TeSF + TeBEF), our simulations give a
simulate a ratio of 1.6 between the present and 21  ka extents of maximum-to-minimum extent ratio of 1.5, and for Boreal forests
TrEF + TrRF. Our simulations spanning the entire glacial–interglacial (BENF  +  BSNF  +  BSBF) of 3.9. The minimum extent of the former
12       | ALLEN et al.

st

Fo t
ed res
t

st
es

re

re
Fo

o
es For

st

-le d F
re
n

Fo
en

ee

ve
Te rat edl een gre

av
ed
t
rg

le
pe Par ore Eve

av
er

e-
pe Mi -lea For

ad
Ev

od ee edl
av d

le
-le lan

e-
st

st

o
Sh Wo rgr Ne
d

ve

an st

Br
re

re

dl
d

e
oo

ee
Fo

Fo

gr
d

n
an

e
W

er

N
F
en

en

e
nd

e
bl

gr
n
Te rate ad

St rate ed

Tu land
Te rate ate
re

re

kl
la

ee
ru

d
m

er

e
pe Bro

n
ss

ng

e
Sh

er

pe Su

gr

ra
m
la
er

m te N
ra
t

ed
p
ai

rk

er

um

nd
op ser

Ev

Te Tem
te

e
G

Pa

Bo Su
Ev

To sifi
Tr na

ra

S
e

al

al

al

r
i-d

ra

s
pe

pe

pe
t

al

al

al

al

al
n
ic

ic

ic
er

b
m

la
va

nd
re

re

re

re

re

ru

l
m

ep
op

op

ta
ar
es

nc
Se

Sa

Bo

Bo

Bo

Bo

Tu
Te

Te
W
Tr

Tr
D

U
0 0

10 10

20 20

30 30

40 40

50 50
Age (ka BP)

60 60

70 70

80 80

90 90

100 100

110 110

120 120

130 130

140 140
10 20 30 40 10 10 10 10 20 10 20 10 20 10 10 10 1 10 10 10 10 20 10 10 10 10 10 10 1 100 105
Areal extent (103 km2) Extent relative
to 0 ka BP (%)

F I G U R E 6   History of global extents of simulated biomes since 140 ka. Time series, 140 ka to present, of the simulated global areal extent
of each biome, and the total extent of ice-free land expressed relative to that at present, are shown by the series of graphs. (Note: The
horizontal scales of the shaded graphs for Temperate Needle-leaved Evergreen Forest and Unclassified differ from those of the remainder,
the former by a factor of 0.1 and the latter by 0.01)

700
21 June–20 July (W m )

520
Global C-Mass (Pg)

Insolation – 55°N

600 300
concentration (ppmv)

500
Atmospheric CO2

500 480 260

460
400 220
-2

440

180
0 20 40 60 80 100 120 140
Age (ka BP)

F I G U R E 7   Simulated global carbon mass 140 ka to present. Total global carbon mass (heavy black line) simulated by LPJ-GUESS for the
88 palaeoclimate time slices and for the ‘present’. For comparison, the summer (21st June–20th July) insolation at 55° N (smooth light grey
line), computed following Laskar (2004), and atmospheric carbon dioxide concentration (dark grey line), following Bereiter et al. (2015; data
downloaded 3rd October 2019 from http://ncdc.noaa.gov/paleo​/study​/17975), are also shown

was simulated for 60 ka and of the latter for H2 (24 ka); maximum the more diverse TrEF (maximum-to-minimum extent ratio 1.65).
extents of both were simulated for 118 ka. Thus, our results indicate Similarly, the three categories of temperate forest show markedly
that forests everywhere have experienced moderate-to-substantial different maximum-to-minimum ratios, the highest being for TeMxF
changes in global extent over the last 140 kyr. In terms of their bio- (6.10). The ratio of 3.95 for TeSF may have a significant biodiversity
diversity legacy, changing extents of tropical forests may be of the legacy, whilst that of 1.54 for TeBEF, the most globally extensive
greatest importance; it is noteworthy that TrRF experienced greater temperate forest biome, may at least in part account for its generally
changes in extent (maximum-to-minimum extent ratio 3.53) than did higher species richness than that of TeSF. The most extreme ratio
ALLEN et al. |
      13

of maximum-to-minimum global extent, a value of 206, is seen for (Axmanová et al., 2020), furthermore, indicates that they consumed
the generally least extensive of the Boreal forest biomes, BSNF, its woody as well as herbaceous material, leading to an inference that
minimum extent being reached at 120 ka and its maximum at 86 ka. they occupied landscapes with mosaics of grasslands, shrublands
BENF, the Boreal forest biome with the greatest median extent, and woodlands. This suggests that ShT which, despite being domi-
reached its minimum extent at H2 (24 ka) and its maximum at 118 ka, nated by the BESh PFT, has a C3G carbon mass per unit area >50%
with a ratio of 7.46, whilst BSBF, with a ratio of 3.07, reached its min- higher than Tun, and a total carbon mass per unit area two orders
imum extent at 130 ka and its maximum at 108 ka. Thus, although of magnitude greater than Tun and an order of magnitude greater
forests everywhere experienced changes in extent since 140 ka, the than St, may have provided suitable habitat for these creatures.
magnitude of these changes varies considerably and there are strong Furthermore, ShT was extensive during the glacials, especially in
indications that biomes that experienced smaller past changes in ex- Eurasia, and much more extensive than either Tun or St (Figures 5
tent support greater biodiversity today. & 6). As illustrated in Figure 4, vegetation patterns in eastern Siberia
Our second question concerned the extent and persistence of differ between the LIG and Holocene, reflecting the quantitatively
temperate and boreal non-forest biomes with which Pleistocene me- different climatic forcing that in turn leads to differences in simu-
ga-herbivores are generally considered to have been associated. Of lated climatic conditions and patterns. Strikingly, there is a generally
temperate and boreal biomes, Tun and St are generally dominated by greater regional extent of ShT during the LIG than the Holocene.
the C3 Grass (C3G – see Table 1 for PFT acronyms) PFT, with lim- Whether or not the climatic and resulting vegetation differences
ited components of tree and/or shrub PFTs. Both have greater simu- alone are sufficient to explain persistence of Pleistocene mega-her-
lated last glacial than Holocene or LIG extents (Figure 6); Tun had its bivores during the LIG but not the Holocene remains contentious,
greatest extent under the H2 (24 ka) climate and St under the 32 ka but our simulations provide additional support for a key role of veg-
climate, whilst both had minimum extent under the 122 ka climate. etation differences.
However, extent alone tells only part of the story; simulated C3G Our third question addressed the nature and extent of vegeta-
carbon mass per unit area also shows systematic patterns of change tion responses in tropical latitudes and the Southern Hemisphere
in these biomes since 140  ka. For Tun this was markedly greater to last glacial stage millennial climatic fluctuations. As illustrated in
during the penultimate glacial and most of the last glacial stage, the Figure 3, and described above, although the most striking vegeta-
exception being the millennia around the LGM, whereas for St it was tion responses were simulated in Europe and Eurasia, substantial
consistently less during glacials than during either the Holocene or responses of tropical and Southern Hemisphere vegetation were
LIG. Together with the changes in extent, this results in Tun holding also simulated. Our results also show that both the magnitude and
>20% of the C3G carbon mass present in the two biomes for most spatial patterns of these responses differ depending upon the state
of the last glacial, rising to >25% in simulations for most time slices of the orbital forcing when the shift between stadial and inter-stadial
between 16 and 64 ka, and to peaks of >35% in those for H1, H2, conditions occurs. Overall, our simulations indicate that vegetation
H3 and 140  ka. In contrast, Tun generally holds <15% of the C3G patterns in most tropical and Southern Hemisphere land areas were
carbon mass of the two biomes during the Holocene and LIG, falling likely markedly affected by millennial climatic fluctuations during the
to minima of c. 10%–11% during both interglacials. Overall, C3G car- last glacial, although there were localized exceptions to this gener-
bon mass simulated for these biomes was generally markedly greater alization, including New Zealand, south-east Australia and Tasmania
during the glacials, and had minimum values during earlier millennia (Figure 3).
of the LIG. Thus, whilst the marked reduction in herbaceous plant The extent to which LIG vegetation patterns paralleled those of
material biomass in these biomes simulated after the LGM corre- the Holocene was the subject of our fourth question. As discussed
sponds to the principal period of decline, and in some cases eventual above and illustrated in Figure 4, whilst there are, as expected, broad
extinction, of grazing specialists among the Pleistocene mega-her- similarities, there are also numerous differences both regionally and
bivores of the northern continents, their persistence during the LIG in terms of the temporal evolution of the vegetation patterns during
when simulated herbaceous plant biomass in these biomes reached the two interglacials. Furthermore, these differences are neither
minimum values appears anomalous. However, fossil evidence of restricted to nor most striking in higher latitudes of the Northern
LIG presence of Pleistocene mega-herbivores has been reported Hemisphere, but are obvious on all vegetated continents. Once
from north-eastern Siberia, where evidence from fossil beetle as- again, it is the climatic consequences of the quantitatively different
semblages, as well as palaeovegetation evidence, indicates the veg- forcing that underlie the differences in simulated spatial patterns and
etation remained relatively open, with a smaller proportion of tall temporal evolution of the vegetation. Both the greater magnitude of
woody components than during the Holocene (Sher,  1997). Our the Northern Hemisphere insolation peak, and the higher [CO2]atm,
simulations parallel this pattern; both total carbon mass and LAI of at the onset of the LIG than at the onset of the Holocene (Figure 7)
boreal tree PFTs (Sum of Boreal Trees – B-tree) are much less over resulted not just in globally warmer conditions (Fischer et al., 2018),
eastern Siberia during the early LIG, when C3G carbon mass of Tun but in regional differences in climatic conditions and climatic evolu-
and St was at a minimum, than in the present simulation. Recent ev- tion that are reflected by the simulated vegetation differences.
idence from analyses of gut contents of frozen mega-herbivore indi- Finally, we asked how similar were global vegetation patterns at
viduals recovered from permafrost in Northern Siberia and Beringia the PGM and LGM. Figure 5 illustrates our biome simulations for
 |
14       ALLEN et al.

the LGM (21 ka) and PGM (140 ka), showing marked differences in Primary exceptions are the core areas of TrEF, although the extent
many regions; these differences were described above. Once again of TrEF in each of its principal centres, South America, sub-Saharan
the quantitatively different forcing, principally the more extreme Africa and south-east Asia, is nonetheless simulated to have varied
Northern Hemisphere summer insolation minimum at 140  ka than substantially, with marked reductions in extent at some times in all
at 21  ka, leads to differences in simulated climatic conditions that three regions.
in turn result in the differences in biome patterns. Thus, as with the Despite being in general accord with palaeovegetation ev-
Holocene and LIG, climate, and the resulting extent and distribution idence, one striking feature that our simulations do not show is
of biomes, differed between the LGM and PGM. a vegetated Sahara during either the early Holocene or LIG, thus
conflicting with inferences of a ‘green’ Sahara at these times on
the basis of geomorphic and fossil evidence (Larrasoaña, Roberts,
5 |  CO N C LU S I O N S & Rohling, 2013). Although all modelling studies show an enhanced
West African monsoon during the early Holocene, they struggle
Firstly, as palaeovegetation records from various parts of the world to simulate the magnitude of change; this is a general challenge
show, regional vegetation patterns during the LIG differed from in palaeoclimatic modelling (Valdes,  2011). Early modelling stud-
those of the Holocene (e.g. Fréchette, Wolfe, Miller, Richard, & de ies suggested that without incorporation of the strong land sur-
Vernal,  2006; Herring & Gavin,  2015; Kershaw et  al.,  2007; van face to atmosphere feedbacks as the vegetation zones expand in
der Hammen & Hooghiemstra,  2003; Whitlock, Sarna-Wojcicki, response to enhanced monsoon precipitation, climate models are
Bartlein, & Nickmann, 2000). However, the timing of some key re- unable to generate sufficient precipitation in the Sahara to support
gional contrasts between the two interglacials differs in our simu- the early Holocene ‘greening’ (Claussen & Gayler, 1997; de Noblet-
lations, drawing attention to the difficulty of independently dating Ducoudre, Claussen, & Prentice,  2000). However, this result was
terrestrial records of the LIG, and hence of demonstrating such dia- not replicated by fully coupled atmosphere–ocean–vegetation
chronous patterns using palaeovegetation records. Thus, for exam- models participating in PMIP2 and PMIP3, including HadCM3 with
ple, whereas the northernmost extension of BENF in western Siberia TRIFFID (Braconnot et al., 2012; Harrison et al., 2014). Recent cli-
is simulated to occur around the middle of the LIG (see Figure 4d, mate model simulations suggest that changes in dust could also be
118 ka), the greatest reduction in the extent of Des in the Sahara, important (Pausata, Messori, & Zhang, 2016), although this result
Arabia and the Indian sub-continent is simulated early in the LIG may have been an artefact of that particular model (Hopcroft &
(see Figure 4b, 126 ka). Testing these model-generated hypotheses Valdes, 2019; Thompson, Skinner, Poulsen, & Zhu, 2019). A lack of
awaits advances in dating and new records from these regions. sensitivity of modelled vegetation to rainfall in sub-Saharan Africa
Secondly, simulated biome patterns also differ between the last may also contribute to the problem (Hopcroft, Valdes, Harper, &
and penultimate glacial maxima, emphasizing the more general con- Beerling, 2017).
clusion that, although orbital forcing has a series of characteristic Notwithstanding this limitation, this series of palaeovegetation
frequencies, the forcing is not strictly cyclic. Thus, no two time slices maps simulated for the past 140  kyr (Supplementary Information
have identical forcing; hence, no two have identical climate or biome – Slideshow) provides valuable insights into likely changes in the
patterns. Although fewer palaeovegetation records are available extent and location of biomes worldwide. As such, it represents a
spanning glacials than interglacials, where such records are available valuable new resource for investigations into the role of environ-
they accord with this conclusion, each warmer or colder interval hav- mental history, whether in the origins of present biodiversity pat-
ing similar but not identical vegetation, see, for example Ioannina terns (Colville et al., in press), in relation to pre-historic megafaunal
(Roucoux, Tzedakis, Lawson, & Margari,  2011; Tzedakis,  1994) and extinctions, or in relation to the biogeographical history of anatom-
Lago Grande di Monticchio (Allen & Huntley, 2009; Allen et al., 2000; ically modern humans.
Allen, Watts, McGee, & Huntley, 2002).
Thirdly, last glacial stage millennial climatic fluctuations, previ- AC K N OW L E D G E M E N T S
ously shown to be likely to have strongly influenced vegetation in BH, JRMA, JSS, TH and MF were supported by a Leverhulme Trust
Eurasia (Huntley, Allen, Collingham, et al., 2013) and biome patterns Research Grant to BH (RPG-2014-338). Henk Slim provided invalu-
in southern Africa (Huntley et al., 2016), had widespread effects on able support to JRMA and BH in their use of the Durham University
global vegetation patterns in the tropics and Southern Hemisphere, High Performance Computing Facility that was used to perform
as well as on Northern Hemisphere vegetation, and on global carbon the LPJ-GUESS simulations and associated computation. Yvonne
budgets. For some biomes, most notably TrEF, climatic conditions Collingham provided key support with respect to generating the cli-
simulated by climate model experiments designed to mimic Heinrich matic anomalies required to run LPJ-GUESS from the HadCM3 out-
Events had a larger impact upon the simulated global extent of the puts, and also generated the estimates of ‘present' climate for shelf
biome than did the overall orbitally forced glacial–interglacial cycle grid cells exposed at the LGM. HadCM3 simulations were performed
of climatic conditions (Figure 6). on the facilities of the Advanced Computing Research Centre at
Fourthly, most of Earth's land surface has experienced at least the University of Bristol (http://www.bris.ac.uk/acrc/). We are ex-
one, and in many regions multiple, biome changes since 140  ka. tremely grateful for very helpful comments from Jack Williams, and
ALLEN et al. |
      15

for those of an anonymous referee, on our initial submission that Brauer, A., Allen, J. R. M., Mingram, J., Dulski, P., Wulf, S., & Huntley, B.
(2007). Evidence for last interglacial chronology and environmental
enabled us greatly to improve the paper.
change from Southern Europe. Proceedings of the National Academy
of Sciences of the United States of America, 104, 450–455. https://doi.
DATA AVA I L A B I L I T Y S TAT E M E N T org/10.1073/pnas.06033​21104
The following files have been deposited in the Data Dryad reposi- Claussen, M., & Gayler, V. (1997). The greening of the Sahara during the
tory: (a) 89 pairs of files in.csv format, one pair for each time slice, mid-Holocene: Results of an interactive atmosphere-biome model.
Global Ecology and Biogeography Letters, 6, 369–377. https://doi.
including the ‘present’, each pair comprising one file each for the
org/10.2307/2997337
LPJ-GUESS simulated LAI and C-Mass values for all half-degree grid Colville, J. F., Beale, C. M., Forest, F., Altwegg, R., Huntley, B., & Cowling,
cells. (b) One.csv format file giving the biome assignments for all 89 R. M. (in press). Plant richness, turnover and evolutionary diversity
time slices for all half-degree grid cells. (c) One ascii fixed format file track gradients of stability and ecological opportunity in a megadi-
versity centre. Proceedings of the National Academy of Sciences of the
giving, for all half-degree grid cells, the total area of each grid cell
United States of America.
and the % ice-free land area of that grid cell for each of the 89 time Cox, P. M. (2001). Description of the "TRIFFID" dynamic global vege-
slices. (d) Four sets of ArcGIS ascii grid files for a 6-min grid. The tation model. Hadley Centre, Met Office. Hadley Centre Technical
first three sets are binary masks for: (i) ice; (ii) ocean; and (iii) land; Note, 24. 16 pp. M.O. Hadley Centre, Bracknell, Berks.
Davies-Barnard, T., Ridgwell, A., Singarayer, J., & Valdes, P. (2017).
and the fourth set is quantitative ice-fraction (% ice cover). Each set
Quantifying the influence of the terrestrial biosphere on glacial-in-
comprises one file for each of the 89 time slices. and (e) A metadata terglacial climate dynamics. Climate of the Past, 13, 1381–1401.
file describing the contents and formats of these data files. https://doi.org/10.5194/cp-13-1381-2017
de Boer, B., van de Wal, R. S. W., Lourens, L. J., Bintanja, R., & Reerink,
T. J. (2013). A continuous simulation of global ice volume over the
ORCID
past 1 million years with 3-D ice-sheet models. Climate Dynamics, 41,
Brian Huntley  https://orcid.org/0000-0002-3926-2257 1365–1384. https://doi.org/10.1007/s0038​2-012-1562-2
de Noblet-Ducoudre, N., Claussen, R., & Prentice, C. (2000). Mid-Holocene
REFERENCES greening of the Sahara: First results of the GAIM 6000 year BP Experiment
with two asynchronously coupled atmosphere/biome models. Climate
Allen, J. R. M., Brandt, U., Brauer, A., Hubberten, H., Huntley, B., Keller,
Dynamics, 16, 643–659. https://doi.org/10.1007/s0038​20000074
J., … Zolitschka, B. (1999). Rapid environmental changes in southern
EPICA Community Members. (2004). Eight glacial cycles from an
Europe during the last glacial period. Nature, 400, 740–743. https://
Antarctic ice core. Nature, 429, 623–628.
doi.org/10.1038/23432
Fischer, H., Meissner, K. J., Mix, A. C., Abram, N. J., Austermann, J.,
Allen, J. R. M., & Huntley, B. (2009). Last interglacial palaeovegetation,
Brovkin, V., … Zhou, L. (2018). Palaeoclimate constraints on the im-
palaeoenvironments and chronology: A new record from Lago
pact of 2 °C anthropogenic warming and beyond. Nature Geoscience,
Grande di Monticchio, southern Italy. Quaternary Science Reviews, 28,
11, 474–485. https://doi.org/10.1038/s4156​1-018-0146-0
1521–1538. https://doi.org/10.1016/j.quasc​irev.2009.02.013
Forrest, M., Eronen, J. T., Utescher, T., Knorr, G., Stepanek, C., Lohmann,
Allen, J. R. M., Watts, W. A., & Huntley, B. (2000). Weichselian pal-
G., & Hickler, T. (2015). Climate-vegetation modelling and fossil plant
ynostratigraphy, palaeovegetation and palaeoenvironment: The
data suggest low atmospheric CO2 in the late Miocene. Climate of the
record from Lago Grande di Monticchio, southern Italy. Quaternary
Past, 11, 1701–1732.
International, 73(74), 91–110. https://doi.org/10.1016/S1040​
Fréchette, B., Wolfe, A. P., Miller, G. H., Richard, P. J. H., & de Vernal, A.
-6182(00)00067​-7
(2006). Vegetation and climate of the last interglacial on Baffin Island,
Allen, J. R. M., Watts, W. A., McGee, E., & Huntley, B. (2002). Holocene
Arctic Canada. Palaeogeography Palaeoclimatology Palaeoecology,
environmental variability – The record from Lago Grande di
236, 91–106. https://doi.org/10.1016/j.palaeo.2005.11.034
Monticchio, Italy. Quaternary International, 88, 69–80. https://doi.
Guetter, P. J., & Kutzbach, J. E. (1990). A modified Köppen classifica-
org/10.1016/S1040​-6182(01)00074​-X
tion applied to model simulations of glacial and interglacial climates.
Andrews, J. T. (1998). Abrupt changes (Heinrich events) in late
Climatic Change, 16, 193–215.
Quaternary North Atlantic marine environments: A history and re-
Harris, I., Jones, P. D., Osborn, T. J., & Lister, D. H. (2014). Updated
view of data and concepts. Journal of Quaternary Science, 13, 3–16.
high-resolution grids of monthly climatic observations – The CRU
https://doi.org/10.1002/(SICI)1099-1417(19980​1/02)13:1<3:AID-
TS3.10 Dataset. International Journal of Climatology, 34, 623–642.
JQS36​1>3.0.CO;2-0
https://doi.org/10.1002/joc.3711
Axmanová, I., Robovský, J., Tichý, L., Danihelka, J., Troeva, E., Protopopov,
Harrison, S. P., Bartlein, P. J., Brewer, S., Prentice, I. C., Boyd, M., Hessler,
A., & Chytrý, M. (2020). Habitats of Pleistocene megaherbivores re-
I., … Willis, K. (2014). Climate model benchmarking with glacial and
constructed from the frozen fauna remains. Ecography, 43(5), 703–
mid-Holocene climates. Climate Dynamics, 43, 671–688. https://doi.
713. https://doi.org/10.1111/ecog.04940
org/10.1007/s0038​2-013-1922-6
Bauska, T. K., Brook, E. J., Marcott, S. A., Baggenstos, D., Shackleton,
Haxeltine, A., & Prentice, I. C. (1996). BIOME3: An equilibrium terrestrial bio-
S., Severinghaus, J. P., & Petrenko, V. V. (2018). Controls on millen-
sphere model based on ecophysiological constraints, resource availabil-
nial-scale atmospheric CO2 variability during the last glacial period.
ity and competition among plant functional types. Global Biogeochemical
Geophysical Research Letters, 45, 7731–7740.
Cycles, 10, 693–710. https://doi.org/10.1029/96GB0​2344
Bereiter, B., Eggleston, S., Schmitt, J., Nehrbass-Ahles, C., Stocker, T. F.,
Herring, E. M., & Gavin, D. G. (2015). Climate and vegetation since the
Fischer, H., … Chappellaz, J. (2015). Revision of the EPICA Dome C
Last Interglacial (MIS 5e) in a putative glacial refugium, northern
CO2 record from 800 to 600kyr before present. Geophysical Research
Idaho, USA. Quaternary Science Reviews, 117, 82–95. https://doi.
Letters, 42, 542–549.
org/10.1016/j.quasc​irev.2015.03.028
Braconnot, P., Harrison, S. P., Kageyama, M., Bartlein, P. J., Masson-
Hopcroft, P. O., & Valdes, P. J. (2019). On the role of dust-climate feed-
Delmotte, V., Abe-Ouchi, A., … Zhao, Y. (2012). Evaluation of climate
backs during the mid-Holocene. Geophysical Research Letters, 46,
models using palaeoclimatic data. Nature Climate Change, 2, 417–424.
1612–1621. https://doi.org/10.1029/2018G​L080483
https://doi.org/10.1038/nclim​ate1456
 |
16       ALLEN et al.

Hopcroft, P. O., Valdes, P. J., Harper, A. B., & Beerling, D. J. (2017). Parrenin, F., Barnola, J. M., Beer, J., Blunier, T., Castellano, E., Chappellaz,
Multi vegetation model evaluation of the Green Sahara climate re- J., … Wolff, E. (2007). The EDC3 chronology for the EPICA dome C
gime. Geophysical Research Letters, 44, 6804–6813. https://doi. ice core. Climate of the Past, 3, 485–497. https://doi.org/10.5194/
org/10.1002/2017G​L073740 cp-3-485-2007
Huntley, B., Allen, J. R. M., Barnard, P., Collingham, Y. C., & Holliday, P. R. Pausata, F. S. R., Messori, G., & Zhang, Q. (2016). Impacts of dust re-
(2013). Species’ distribution models indicate contrasting late-Qua- duction on the northward expansion of the African monsoon during
ternary histories for southern and northern hemisphere bird spe- the Green Sahara period. Earth and Planetary Science Letters, 434,
cies. Global Ecology and Biogeography, 22, 277–288. https://doi. 298–307. https://doi.org/10.1016/j.epsl.2015.11.049
org/10.1111/j.1466-8238.2011.00751.x Peltier, W. R. (2004). Global glacial isostasy and the surface of the ice-age
Huntley, B., Allen, J. R. M., Collingham, Y. C., Hickler, T., Lister, A. M., earth: The ICE-5G (VM2) model and GRACE. Annual Review of Earth
Singarayer, J., … Valdes, P. J. (2013). Millennial climatic fluctuations and Planetary Sciences, 32, 111–149. https://doi.org/10.1146/annur​
are key to the structure of last glacial ecosystems. PLoS One, 8, ev.earth.32.082503.144359
e61963. https://doi.org/10.1371/journ​al.pone.0061963 Petit, J. R., Jouzel, J., Raynaud, D., Barkov, N. I., Barnola, J.-M., Basile,
Huntley, B., Collingham, Y. C., Singarayer, J. S., Valdes, P. J., Barnard, P., I., … Stievenard, M. (1999). Climate and atmospheric history of the
Midgley, G. F., … Ohlemüller, R. (2016). Explaining patterns of avian past 420,000 years from the Vostok ice core, Antarctica. Nature, 399,
diversity and endemicity: Climate and biomes of southern Africa 429–436. https://doi.org/10.1038/20859
over the last 140,000 years. Journal of Biogeography, 43, 874–886. Prentice, I. C., Cramer, W., Harrison, S. P., Leemans, R., Monserud, R.
Jolly, D., Prentice, I. C., Bonnefille, R., Ballouche, A., Bengo, M., Brenac, A., & Solomon, A. M. (1992). A global biome model based on plant
P., … Waller, M. (1998). Biome reconstruction from pollen and plant physiology and dominance, soil properties and climate. Journal of
macrofossil data for Africa and the Arabian peninsula at 0 and Biogeography, 19, 117–134.
6000 years. Journal of Biogeography, 25, 1007–1027. https://doi. Prentice, I. C., Harrison, S. P., & Bartlein, P. J. (2011). Global vegeta-
org/10.1046/j.1365-2699.1998.00238.x tion and terrestrial carbon cycle changes after the last ice age. New
Jouzel, J., Masson-Delmotte, V., Cattani, O., Dreyfus, G., Falourd, S., Phytologist, 189, 988–998. https://doi.org/10.1111/j.1469-8137.
Hoffmann, G., … Wolff, E. W. (2007). Orbital and millennial Antarctic 2010.03620.x
climate variability over the past 800,000 years. Science, 317, 793– Rasmussen, S. O., Bigler, M., Blockley, S. P., Blunier, T., Buchardt, S. L.,
796. https://doi.org/10.1126/scien​ce.1141038 Clausen, H. B., … Winstrup, M. (2014). A stratigraphic framework for
Kershaw, A. P., McKenzie, G. M., Porch, N., Roberts, R. G., Brown, J., abrupt climatic changes during the Last Glacial period based on three
Heijnis, H., … Newallt, P. R. (2007). A high-resolution record of vege- synchronized Greenland ice-core records: Refining and extending
tation and climate through the last glacial cycle from Caledonia Fen, the INTIMATE event stratigraphy. Quaternary Science Reviews, 106,
southeastern highlands of Australia. Journal of Quaternary Science, 14–28. https://doi.org/10.1016/j.quasc​irev.2014.09.007
22, 481–500. https://doi.org/10.1002/jqs.1127 Roucoux, K. H., Tzedakis, P. C., Lawson, I. T., & Margari, V. (2011).
Larrasoaña, J. C., Roberts, A. P., & Rohling, E. J. (2013). Dynamics of Vegetation history of the penultimate glacial period (Marine Isotope
green Sahara periods and their role in hominin evolution. PLoS One, Stage 6) at Ioannina, north-west Greece. Journal of Quaternary
8, e76514. https://doi.org/10.1371/journ​al.pone.0076514 Science, 26, 616–626. https://doi.org/10.1002/jqs.1483
Laskar, J., Robutel, P., Joutel, F., Gastineau, M., Correia, A. C. M., & Scheiter, S., & Higgins, S. I. (2009). Impacts of climate change on the
Levrard, B. (2004). A long-term numerical solution for the insolation vegetation of Africa: An adaptive dynamic vegetation model-
quantities of the Earth. Astronomy & Astrophysics, 428, 261–285. ling approach. Global Change Biology, 15, 2224–2246. https://doi.
Loulergue, L., Schilt, A., Spahni, R., Masson-Delmotte, V., Blunier, T., org/10.1111/j.1365-2486.2008.01838.x
Lemieux, B., … Chappellaz, J. (2008). Orbital and millennial-scale fea- Sher, A. (1997). Late-Quaternary extinction of large mammals in north-
tures of atmospheric CH4 over the past 800,000 years. Nature, 453, ern Eurasia: A new look at the Siberian contribution. In B. Huntley,
383–386. W. Cramer, A. V. Morgan, H. C. Prentice, & J. R. M. Allen (Eds.), Past
Luthi, D., Le Floch, M., Bereiter, B., Blunier, T., Barnola, J. M., Siegenthaler, and future rapid environmental changes: The spatial and evolutionary
U., … Stocker, T. F. (2008). High-resolution carbon dioxide concen- responses of terrestrial biota (pp. 319–339). Berlin: Springer-Verlag.
tration record 650,000–800,000 years before present. Nature, 453, Singarayer, J. S., & Valdes, P. J. (2010). High-latitude climate sensitivity to
379–382. ice-sheet forcing over the last 120 kyr. Quaternary Science Reviews,
Marchant, R., Cleef, A., Harrison, S. P., Hooghiemstra, H., Markgraf, V., van 29, 43–55.
Boxel, J., … Wille, M. (2009). Pollen-based biome reconstructions for Smith, B., Prentice, I. C., & Sykes, M. T. (2001). Representation of
Latin America at 0, 6000 and 18 000 radiocarbon years ago. Climate vegetation dynamics in the modelling of terrestrial ecosystems:
of the Past, 5, 725–767. https://doi.org/10.5194/cp-5-725-2009 Comparing two contrasting approaches within European climate
Miller, P. A., Giesecke, T., Hickler, T., Bradshaw, R. H. W., Smith, B., Seppa, space. Global Ecology and Biogeography, 10, 621–637. https://doi.
H., … Sykes, M. T. (2008). Exploring climatic and biotic controls on org/10.1046/j.1466-822X.2001.00256.x
Holocene vegetation change in Fennoscandia. Journal of Ecology, 96, Smith, B., Warlind, D., Arneth, A., Hickler, T., Leadley, P., Siltberg, J.,
247–259. https://doi.org/10.1111/j.1365-2745.2007.01342.x & Zaehle, S. (2014). Implications of incorporating N cycling and N
Nolan, C., Overpeck, J. T., Allen, J. R. M., Anderson, P. M., Betancourt, J. limitations on primary production in an individual-based dynamic
L., Binney, H. A., … Jackson, S. T. (2018). Past and future global trans- vegetation model. Biogeosciences, 11, 2027–2054. https://doi.
formation of terrestrial ecosystems under climate change. Science, org/10.5194/bg-11-2027-2014
361, 920–923. https://doi.org/10.1126/scien​ce.aan5360 Tarasov, P. E., Webb III, T., Andreev, A. A., Afanas’eva, N. B., Berezina, N. A.,
Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Bezusko, L. G., … Zernitskaya, V. P. (1998). Present-day and mid-Ho-
Powell, G. V. N., Underwood, E. C., … Kassem, K. R. (2001). Terrestrial locene biomes reconstructed from pollen and plant macrofossil data
ecoregions of the world: A new map of life on Earth. BioScience, 51, from the former Soviet Union and Mongolia. Journal of Biogeography,
933–938. https://doi.org/10.1641/0006-3568(2001)051[0933:- 25, 1029–1053. https://doi.org/10.1046/j.1365-2699.1998.00236.x
TEOTW​A]2.0.CO;2 Thompson, A. J., Skinner, C. B., Poulsen, C. J., & Zhu, J. (2019). Modulation
Olson, J., Watts, J. A., & Allison, L. J. (1983). Carbon in live vegeta- of mid-Holocene African rainfall by dust aerosol direct and indirect
tion of major world ecosystems. Oak Ridge National Laboratory, effects. Geophysical Research Letters, 46, 3917–3926. https://doi.
ORNL-5862. org/10.1029/2018G​L081225
ALLEN et al. |
      17

Thonicke, K., Venevsky, S., Sitch, S., & Cramer, W. (2001). The role of performed the palaeovegetation simulations and prepared
fire disturbance for global vegetation dynamics: Coupling fire into a
Figure 6; B.H. developed the biome inference approach and
Dynamic Global Vegetation Model. Global Ecology and Biogeography,
10, 661–677. https://doi.org/10.1046/j.1466-822X.2001.00175.x
wrote the program to implement this, mapped the simulated bi-
Tzedakis, P. C. (1994). Vegetation change through glacial-interglacial cy- omes, analysed the simulation results, prepared Figures 1–5 and
cles - a long pollen sequence perspective. Philosophical Transactions 7, prepared the Supplementary Material and drafted the manu-
of the Royal Society of London: Series B - Biological Sciences, 345, script. All authors commented upon and contributed to the final
403–432.
version of the manuscript.
Valdes, P. (2011). Built for stability. Nature Geoscience, 4, 414–416.
https://doi.org/10.1038/ngeo1200
Valdes, P. J., Armstrong, E., Badger, M. P. S., Bradshaw, C. D., Bragg, S U P P O R T I N G I N FO R M AT I O N
F., Crucifix, M., … Williams, J. H. T. (2017). The BRIDGE HadCM3 Additional supporting information may be found online in the
family of climate models: HadCM3@Bristol v1.0. Geoscientific Model
Supporting Information section.
Development, 10, 3715–3743.
van der Hammen, T., & Hooghiemstra, H. (2003). Interglacial-glacial
Fuquene-3 pollen record from Colombia: An Eemian to Holocene cli-
mate record. Global and Planetary Change, 36, 181–199. https://doi. How to cite this article: Allen JRM, Forrest M, Hickler T,
org/10.1016/S0921​-8181(02)00184​- 4 Singarayer JS, Valdes PJ, Huntley B. Global vegetation
Walker, M. J. C., Berkelhammer, M., Björck, S., Cwynar, L. C., Fisher,
patterns of the past 140,000 years. J Biogeogr. 2020;00:1–18.
D. A., Long, A. J., … Weiss, H. (2012). Formal subdivision of the
Holocene Series/Epoch: A discussion paper by a working group of
https://doi.org/10.1111/jbi.13930
INTIMATE (Integration of ice-core, marine and terrestrial records)
and the Subcommission on Quaternary Stratigraphy (International
Commission on Stratigraphy). Journal of Quaternary Science, 27, 649–
659. https://doi.org/10.1002/jqs.2565 APPENDIX 1
Whitlock, C., Sarna-Wojcicki, A. M., Bartlein, P. J., & Nickmann, R. J.
(2000). Environmental history and tephrostratigraphy at Carp lake, Rule-based biome inference
southwestern Columbia basin, Washington, U.S.A. Palaeogeography,
The biome occupying each grid cell was inferred using the following
Palaeoclimatology, Palaeoecology, 155, 7–29. https://doi.org/10.1016/
S0031​- 0182(99)00092​-9
rules, applied in the order given below where they are expressed
Williams, J. W., Summers, R. L., & Webb, T. (1998). Applying plant in the form of a dichotomous key. The majority of the rules applied
functional types to construct biome maps from eastern North relate to the simulated leaf area index (LAI) of one or more plant
American pollen data: Comparisons with model results. Quaternary functional types (PFTs – see Table 1). The initial rule, however,
Science Reviews, 17, 607–627. https://doi.org/10.1016/S0277​
considers the simulated total carbon mass (C-Mass) for the grid cell,
-3791(98)00014​-6
Wolff, E. W., Chappellaz, J., Blunier, T., Rasmussen, S. O., & Svensson, relative to a threshold value of 1.0 × 109 kg for a half-degree grid cell
A. (2010). Millennial-scale variability during the last glacial: The ice at the equator, area 3,087.57 km2, all of which is ice-free land. The
core record. Quaternary Science Reviews, 29, 2828–2838. https://doi. threshold applied to a given grid cell is adjusted to account for its
org/10.1016/j.quasc​irev.2009.10.013
ice-free land area, thus taking into account the reduced areas of grid
cells at higher latitudes, of coastal grid cells and of grid cells partially
B I O S K E TC H covered by ice sheets.
Brian Huntley is a palaeoecologist, ecologist and biogeographer 1. C-Mass < threshold value† 
with research interests in interactions between species, ecosys- 2. LAI BESh = 0.0 1. Desert
tems and their changing environment. His work encompasses 2. LAI BESh > 0.0 2. Tundra
studies of the palaeoecology and palaeoenvironments of the
1. C-Mass ≥ threshold value
Quaternary, of present ecosystems and biogeographical patterns,
3. LAI TrBE > 2.0 3. Tropical
and of the potential impacts of anthropogenic global change on Evergreen
species and ecosystems. His research has considered a range of Forest
taxonomic groups, from plants to extinct Quaternary mammals, 3. LAI TrBE ≤ 2.0
and various ecosystems, from Arctic tundra to fynbos. He has 4. LAI TrBR > 0.6 4. Tropical
particular interest in birds and climatic change, and in the de- Raingreen
velopment of conservation strategies informed by research into Forest

how species and ecosystems respond to environmental changes. 4. LAI TrBR ≤ 0.6

5. LAI C4G > 0.3
Author contributions: B.H. conceived the study; P.J.V. and J.S.S. 6. LAI Tr-shrub > 0.3 OR LAI 5. Savanna
performed the palaeoclimate simulations; M.F. and T.H. advised TrBR > 0.3

on running the palaeovegetation simulations; MF developed and 6. LAI Tr-shrub ≤ 0.3 AND LAI 6. Tropical
implemented code to incorporate the climatic anomalies, extend TrBR ≤ 0.3 Grassland

simulations to the shelf areas and specify the varying obliq- 5. LAI C4G ≤ 0.3

uity; M.F., T.H. and J.R.M.A. parameterized the PFTs; J.R.M.A.

 |
18       ALLEN et al.

7. LAI TeBS > 1.0 16. LAI BNE ≤ 0.3

8. LAI B-tree > 0.1 OR LAI TeNE > 0.1 7. Temperate 17. LAI BNS > 0.1
Mixed Forest 18. LAI BESh > 1.0 15. Boreal
8. LAI B-tree ≤ 0.1 AND LAI 8. Temperate Woodland
TeNE ≤ 0.1 Summergreen 18. LAI BESh ≤ 1.0 16. Boreal
Forest Needle-leaved
7. LAI TeBS ≤ 1.0 Summergreen
9. LAI TeNE > 1.0 20. Temperate Forest
Needle-leaved 17. LAI BNS ≤ 0.1
Evergreen 19. LAI BIBS > 0.1 17. Boreal
Forest Broad-leaved
9. LAI TeNE ≤ 1.0 Summergreen
10. LAI TeBE > 1.0 9. Temperate Forest
Broad-leaved 19. LAI BIBS ≤ 0.1
Evergreen 20. LAI BESh > 0.3 18. Shrub Tundra
Forest
20. LAI BESh ≤ 0.3
10. LAI TeBE ≤ 1.0
21. LAI C3G > 0.3 19. Steppe
11. LAI TeRSh plus LAI TeESh > 0.3
21. LAI C3G ≤ 0.3
12. LAI B-tree > 0.02 10. Boreal
Parkland 22. [ ( Σ LAI minus ( LAI C3G plus LAI 2. Tundra
BESh))] ≤ 0.01
12. LAI B-tree ≤ 0.02
22. [ ( Σ LAI minus ( LAI C3G plus LAI
13. LAI TeBS > 0.02 11. Temperate BESh))]> 0.01
Parkland
23. [ ( Σ LAI minus ( LAI TeBE plus 13. Temperate
13. LAI TeBS ≤ 0.02 LAI C3G plus LAI TeESh plus LAI Shrubland
14. LAI TeBE > 0.3 12. Warm TeRSh))] ≤ 0.01
Temperate 23. [ ( Σ LAI minus ( LAI TeBE plus
Woodland LAI C3G plus LAI TeESh plus LAI
14. LAI TeBE ≤ 0.3 TeRSh))]> 0.01
15. LAI C3G> ( LAI TeRSh plus LAI 21. Semi-desert 24. LAI TrBE plus LAI TrIBE > 1.7 3. Tropical
TeESh) Evergreen
15. LAI C3G ≤ ( LAI TeRSh plus LAI 13. Temperate Forest
TeESh) Shrubland 24. LAI TrBE plus LAI TrIBE ≤ 1.7 22. Unclassified
†
11. LAI TeRSh plus LAI TeESh ≤ 0.3 ((( grid cell area × % ice-free land )/100)/3087·57 ) x (1·0 × 109) kg.
16. LAI BNE > 0.3 14. Boreal
Needle-leaved
Evergreen
Forest