1
Supplementary Information
A. Materials and Methods
A1. Sample description and preparation
A2. Stable isotope analysis
A3. Calculations to derive stable isotope values and their errors
A4. PMIP2 simulations
A5. Description of the high
-
resolution and standa
rd
-
resolution simulations with
the LMD
Z model
B. Analysis of results
B1. Systematics of clumped isotope paleothermometry in soil carbonates and
terrestrial gastropods
B
2
. Isotope flux models of terrestrial gastropod body water
B3. Analysis of PMIP2 mode
l output
B
4
. Comparisons of high
-
resolution simulations with clumped isotope data
B
5
. Causes of temperature changes
B
6
. Causes of water
δ
18
O
changes
B
7
. Possible implications of results for
surface
relative humidity change
Supplementary Figures
Fig. S1: Map of the central Chinese
L
oess
P
lateau
Fig. S2: Images of gastropod specim
ens
Fig. S3: Stationary waves in PMIP2 models
Fig. S4: Soil carbonate clumped isotope calibration data
Fig. S5: Climatological data for site on loess plateau
Fig. S6: Terrestrial gastropod carbonate clumped isotope calibration data
Fig. S7: Distribution o
f water isotope and temperature data from Puxian
gastropods
Fig. S8: Relationship between the calculated change in local precipitation
δ
18
O
and relative humidity of snail habitats
Fig. S9: Diagnostic simulation with LMDZ
Fig. S10: Resolution of LMDZ simulations
Fig. S11: Simulation
of changes in temperature and feedbacks resulting in
simulated regional climate sensitivity
Fig. S12: Simula
ted changes from present to LGM of temperature, precipitation,
and
δ
18
O
Fig. S13: Simulated
vertical profile of
δ
18
O at site
Supplementary Tables
Table S1:
Average stable isotope measurements for all glacial and modern
specimens
Table S2: Average stable isotope measurements for individual specimens
Table S3:
Raw
δ
47
and
Δ
47
data with corrections in analytical sequence
2
Table S4:
Elevation of GCM grid cells fo
r Puxian from PMIP2 GCMs
Table S5:
LGM
-
Modern temperature differences at Puxian from PMIP2 GCMs
Table S6:
LGM
-
Modern differences in hydrologic parameters from PMIP2 GCMs
Table S7: Calculated surface water
δ
18
O
(i.e., precipitation
δ
18
O
) based on model
of s
nail body water
δ
18
O
Table S8: Hydrologic data used for snail body water model
Table S9: Comparison of LMDZ model output with other proxy records
3
A. Materials and Methods
A1. Sample description and preparation
3
2
modern terrestrial gastropod snail shells were collected from
the Chinese loess plateau
at Puxian in Shanxi Province (
Fig. S1
-
S2;
36.421°N, 111.147°E,
1148 m
).
Modern snails
were collected on the surface of the Loess Plateau. Individuals that recently died can be
identified by the colorful organic coating (banding) o
n the shells. Specimens were
identified as
Cathaica
sp. as shown in Fig. S2 and have an aragonitic mineralogy. Field
observations show that snails are active only during the rainy season.
37 shells from the same site were selected from loess strat
igraphic
unit
L1.LL1
which has been extensively described and is known to
correspond to the Last Glacial
Maximum (LGM
)
(1
-
3)
. The stratigraphic unit was identified by pedogenic observations
in field and measurements of magnetic susceptibility. To obtain specimens
of fossil snails
and concretions, about 50 kg of loess sample was
immerse
d in water until fully
disperse
d.
Then the sample was passed through a 150
-
mesh sieve (150 μm) in the field. The >150
μm fraction that is dominated by authigenic grains, from which r
hizoliths, carbonate
concretion, pseudomyce
lia and fossil snail shells could be picked. Due to low abunda
nce
of snail shells, specimens were also hand picked from the LGM layer in the field. The
resistance of snail shells to erosion compared to the loess matrix made snails quite
obvious on the surface of the cliff. Previous
work on shell preservation has indi
cated that
shells of this age exhibit excellent preservation in the central Chinese loess, and that
aragonite/calcite transformations are only seen in much older sediments
(4)
. Soil
carbonates were recovered from the same glacial loess strata at Puxian, an
d are typically
brown nodules of less than 1cm diameter.
The controls on the time of soil carbonate
formation are purely stratigraphic at our site, however we note that sites proximal to ours
coherent glacial
-
interglacial changes in soil carbonate isotopic
composition have been
observed
(5)
, supporting the notion that these carbonates do in fact record information
about the environment at the time their stratigraphic position suggests they should.
The
modern soil carbonate from the Badain Jaran Desert analyz
ed here, was a larger white
nodule of greater than 2 cm diameter.
Individual gastropod shells were first cleaned with a course
-
grained paintbrush to
remove large fragments of soil and or organic material. Shells were gently crushed to
produce large fragmen
ts and then sonicated and washed 6
-
10 times in deionized water, a
procedure that produced very clean shell material free from any macroscopic
contaminants. Shell material was then dried at 50
°C
overnight, and then crushed into a
fine powder with a pestle a
nd mortar.
A2. Stable isotope analysis
Approximately 8
-
12 mg samples of calcium carbonate were reacted for 20
minutes on a 90
°C
online common phosphoric acid bath system described previously
(6)
.
CO
2
gas was immediately frozen by liquid nitrogen, after pas
sing through a dry
ice/ethanol trap. Cryogenic purification of CO
2
was achieved using an automated online
vacuum line, as described previously
(6)
. Additional automated sample cleanup steps
included passing sample gas through a
Porapak Q
TM
120/80 mesh GC co
lumn at
-
20
°
C to
remove potential organic contaminants and silver wool (Sigma
-
Aldrich) to remove sulfur
compounds.
δ
13
C,
δ
18
O,
Δ
47
,
and
Δ
48
,
in CO
2
derived from the phosphoric acid digestion
4
of carbonates was determined using two different Thermo
Scientifi
c
MAT 253 gas
-
source mass spectrometers at Caltech using published configuration and methods
(6
-
8)
.
Raw data and averages for each specimen data are reported in Tables S1
-
S3. Carbonate
standards of know isotopic compositions were run every 4
-
5 analyses (Tab
le S3).
In order to monitor cleaned CO
2
gas samples for the presence of contaminating
molecules such as hydrocarbons and sulfur compounds at the important molecular masses
Δ
48
values were calculated in the same way as
Δ
47
values, by references to the
Δ
48
stochastic distribution as defined by the analysis of heated gases
(7
-
9)
. Large deviations
of over 1
‰
from the
Δ
48
heated gas line are considered potentially indicative of
the
presence of contaminants and so these measurements are excluded from further analysis.
A3. Calculations to derive stable isotope values and their errors
Since the publication of the initial inorganic calibration of the carbonate clumped
isotope thermo
meter
(10)
, alternative calibrations based of experimental and theoretical
data have been reported
(11, 12)
. However, as the calibration line of Ghosh et al.
, 2006,
appears to adhere most closely to
most
published calibration
data derived from
biologically p
recipitated calcite, aragonite, and apatite we will continue to use it in this
study
(13
-
15)
. These published biogenic calibration datasets do produce calibration lines
with marginally different slopes and intercepts to the inorganic calibration of Ghosh et
al., but these differences are small, particularly in the temperature range 20
-
35
°
C where
most of the data in this study fall and the 95% confidence intervals over calibration lines
overlap with each other
(13)
.
Ghosh et al. derived the following relations
hip between
measured
Δ
47
values and
carbonate growth temperature:
Δ
47
= 0.0592 (10
6
.T
-
2
)
–
0.02
Where T is the temperature in Kelvin
. As sample reactions were carried out at 90
°
C,
rather than 25
°
C in Ghosh et al., 2006, we applied the empirically derived
acid digestion
fractionation correction of 0.08
‰
for
Δ
47
values as
reported by Passey et al., 2010
(6, 10)
,
Errors in reported
Δ
47
values and calculated temperatures include the propagated
uncertainty in heated gas determination and in sample measurement.
The vast majority of data for this study was collected before the proposition of an
“absolute reference frame” for clumped isotope studies of CO
2
based on the analysis of
water equilibrated CO
2
gases
(16)
, We therefore present data in Table S1
-
S3 both re
lative
to the stochastic value (i.e., the nomenclature used in most previous studies) and in the
absolute reference frame. Data in the absolute reference frame are generated using an
empirical transfer function developed using data for heated gases, an in
-
house Carrara
Marble standard, and the vein calcite 102
-
GC
-
AZ01 using values in Table S3 following
the procedure described in Dennis et al. (2011)
(16)
. Nevertheless as no water equilibrated
gases were analyzed we considered it most appropriate to continue
to use the equations
described in Ghosh et al., 2006 to convert our measurements into temperatures, as
described above.
Average measured values on Mass Spectrometer 1 for Carrara Marble (n = 10)
relative to the stochastic distribution are 0.366
‰
and on Mas
s Spectrometer 2 (n = 7) are
0.348
‰
, whilst accepted values from our laboratory based on a large (>60) number of
measurements are 0.352
‰.
Average measured values for Carrara Marble on both
machines given in Table S3 on the absolute reference frame are 0.39
2
‰
, identical to the
value of Dennis et al., 2011
(16)
due to the nature of the transfer function used.
For NBS
-
5
19 average measured values on Mass Spectrometer 1 on the absolute reference frame was
0.382
‰ (n=2) compared to an accepted value of 0.392‰. Rela
tive to the stochastic
distribution our measured value was 0.356‰ compared to an accepted value of 0.352‰.
This standard was not run of Mass Spectrometer 2. 102
-
GC
-
AZ01 was run (n = 4) on
Mass Spectrometer 2, yielding a value of 0.665‰ on the relative to
the stochastic
distribution, compared to an accepted value of 0.644‰. See Table S3 for original data.
For aragonite
δ
18
O calculations an acid digestion fractionation factor of
1.00854126 was used, calculated by extrapolation from a published calibration
(11, 17)
.
For calcite a value of 1.00821000 was used
(18)
.
Temperatures estimated using sample
Δ
47
values were used to c
alculate the
δ
18
O of water from which the mineral precipitated using
paired measurements of carbonate
δ
18
O and the following published equations
(17, 19)
.
1000
ln
α
(
Calcite
-
H
2
O) = (18.03.10
3
)/T
–
32.42
1000
ln
α
(
Aragonite
-
H
2
O) = (17.88.10
3
)/T
–
31.14
Pr
opagated errors
in water
δ
18
O and in Modern
-
LGM
Δ
47
temperature differences were
calculated as follows:
E
δ
18
Owater
=
√
(
E
2
δ
18
Ocarbonate
+
E
2
Τ
)
‰
Where E is one standard error. To derive the
E
2
Τ
term for calculating the propagated error
in
δ
18
O
, the uncer
tainty in the
Δ
47
derived temperature is converted into a per mil value
using the appropriate equation.
Table S1 reports average stable isotope data for individuals and Table S2 reports
individual analyses.
A4. PMIP2 simulations
In order to compare our
proxy data to model results we utilized output from both
ocean
-
atmosphere and ocean
-
atmosphere
-
vegetation coupled global circulation models
(GCMs) from the PMIP2 (Paleoclimate Modeling Intercomparison Project, Phase II)
database
(20)
. Results from LGM and c
ontrol simulations from the PMIP2 GCMs are
reported in Figure 2 in the main
text, Fig. S3 and Tables S4
-
S6.
We
acknowledge the
international modeling groups participating in PMIP2 for providing their results for
analysis, the Laboratoire des Sciences du Cl
imat et de l'Environnement (LSCE) for
collecting and archiving the model results
.
The PMIP 2 Data Archive is supported by
CEA, CNRS and the Programme National d'Etude de la Dynamique du Climat (PNEDC).
More information is available on
http://pmip2.lsce.ips
l.fr/
.
A5. Description of the high
-
resolution and standard
-
resolution simulations with
the
LMDZ model
The numerical simulations presented here were performed on the NEC
-
SX8 of the
IDRIS/CNRS computer centre.
This work represents the first time that a high
-
resolution
simulation with isotopic diagnostics has been applied to LGM climate.
The LMDZ4 (Laboratoire de Météorologie Dynamique
-
Zoom) general circulation
model
(21)
is the atmospheric component of the Institut Pierre
-
Simon Laplace coupled
6
model (IPSL
-
C
M4)
(22)
used in CMIP3 (Coupled Model Intercomparison Project)
(23)
.
It is a grid point model. Water vapor and condensate are advected using a second order
monotonic finite volume advection scheme
(24, 25)
.
The physical package includes the
Emanuel convecti
ve scheme
(26, 27)
and a statistical cloud scheme
(28)
.
The isotopic
version of LMDZ is described in detail elsewhere
(29, 30)
.
The standard resolution of LMDZ is 2.5° in latitude, 3.75° in longitude and 19
vertical levels. The present
-
day (PD) simulation is
forced by monthly
-
mean sea surface
temperatures (SST) and sea ice calculated as the long
-
term average between 1979 and
2007 of the Atmospheric Model Intercomparison Project (AMIP)
(30)
SST and sea ice.
The CO
2
concentration is set to 348 ppm. This simulatio
n is very similar to the AMIP
simulation forced by inter
-
annually varying SSTs described in detail in Risi et al.,
2010
(29)
.
The isotopic composition of the precipitation and water vapor in the latter
simulation has been comprehensively evaluated at daily,
monthly and inter
-
annual time
scales
(29, 31
-
33)
.
The SST and sea ice forcing for the LGM simulation are based on
simulations by
a coupled model: IPSL
-
CM4
(34, 35)
. The forcing is
calculated by adding monthly
-
mean
climatological LGM anomalies to the SST
an
d sea ice forcing used in the PD simulation.
The LGM anomalies are
calculated as the difference between LGM and pre
-
industrial
period
simulated by the IPSL coupled model. Given the significant SST biases for
PD in
coupled
models, this method allows us to u
se a simulation with
realistic SSTs as a control
simulation. To ensure that the SSTs are
representative of a average climate, pre
-
industrial
and LGM SSTs and
sea
-
ice were averaged over 50 years. Greenhouse concentration
,
orbital setting and topography are
set following PMIP2. The CO
2
concentration is set to
180 ppm. Orbital parameters are set following Berger, 1978
(36)
.
We use the ICE5G ice
sheet reconstruction of Peltier, 1994
(37)
.
We increase the sea surface
δ
18
O by 1.2
‰
compared to present day based on t
he reconstruction by Labeyrie et al., 1987
(38)
.
In one
additional experiment, the LGM simulation was run with no ice
-
sheets, i.e. with present
-
day topography and land ice fraction.
Both PD and LGM simulations are run for 6 years and the 3 last years are
an
alyzed. The isotopic composition from this LGM simulation has been evaluated against
some ice
-
core, cellulose, groundwater and speleothem records throughout the globe in
Risi et al., 2010
(29)
.
Whereas the
δ
18
O
change in mid and high latitude was satisfactorily
captured by LMDZ, the observed depletion observed at LGM in low latitude ice cores
(Tibet and Andes) was systematically underestimated.
The specificity of the LMDZ model is to offer a zoom functionali
ty to enable
high
-
resolution simulation over a specific region. The grid can be stretched in such a way
as to increase the resolution in a region of interest down to a few tens of kilometers, and
decrease the resolution everywhere else. LMDZ can thus be us
ed as a regional model.
The major difference from a regional model is that the simulations remain global, so that
the benefits of the high resolution on the large
-
scale circulation and the vapor isotopic
composition can feedback on the global simulation. I
n mountainous regions such as the
Andes
(39)
and the Tibetan region
(32)
, it has been shown that the isotopic simulation is
dramatically improved when using the zoom functionality. Based on the hypothesis that
the difficulties of LMDZ to capture the depletio
n observed at LGM in low
-
latitude ice
cores could be due to a coarse resolution
(29)
, we thus performed zoomed simulations for
the present day and LGM. This is the first time that a high
-
resolution simulation by an
7
isotopic model for a past climate is docum
ented.
The zoom configuration used here is described in
Gao et al., 2011
(32)
.
The
resolution is
about 50 km in a wide region from about 10°N to 50°N and
70°E to 120°
E
(Fig. S10). In the standard
-
resolution and zoomed simulations, the altitudes of the grid
box containing Puxian are 833 m and 1117 m respectively. The zoomed simulation thus
captures the actual altitude of Puxian (1148 m) much more accurately than the standard
-
resolution simulation. More generally, the topographic features of the Tibetan platea
u and
surrounding ranges are much better resolved
(Fig. S10).
For numerical stability reasons in zoomed simulations, the winds need to
be
nudged (i.e. relaxed towards a prescribed state) outside the zoom. In
Gao et al
.,
2011,
winds wer
e nudged towards re
-
a
nalyses
(32)
. Since there are
no reanalysis for the LGM,
we nudged the winds towards the 6
-
hourly wind
fields simulated by the standard
-
resolution simulations described above.
The relaxation time is 30 minutes outside the
zoom and 6h inside the zoom.
The lo
nger time scale inside the zoom allows the model to
create its own
circulation resulting from the finely
-
resolved topography.
Our simulations are relatively short due to computational limitations. The
standard
-
resolution simulations were run for 6 years an
d the last 3 years
were analyzed.
Year
-
to
-
year variability stem from intrinsic atmospheric
variability only, since SST and
sea
-
ice forcing’s are monthly mean
climatologies with no inter
-
annual variability. The
standard deviations
among the 3 analyzed years
are 0.9K for
temperature and 0.5
‰
for
precipitation
δ
18
O at Puxian in annual average. These values are lower
than the LGM
-
PD
differences discussed in this paper. In zoomed
simulations, winds are nudged towards
year 3 of the standard
-
resolution
simulation
during 3 years and the 2 last years were
analyzed. The
differences between the 2 last years are only 0.04K for temperature and
0.07
‰
for precipitation
δ
18
O at Puxian in annual average, due to
the nudging.
B. Analysis of results
B1.
Systematics of clumpe
d isotope thermometry in soil carbonates and terrestrial
gastropods.
Pedogenic carbonates are thought to preferentially form in the warmer months,
and so their stable isotope composition is thought to reflect environmental conditions
during the summer mon
ths (eg. Breecker et al., 2009)
(40)
. This idea was confirmed by a
recent in depth calibration study of modern/recent soil carbonates of a range of
morphologies, and from different multiple different localities (including China) showed
that
Δ
47
derived temperature from soil carbonates were generally hotter than mean annual
temperatures, but correlate well with warm summer month average temperatures
(6)
. A
modern soil carbonate from the Badan Jarain Desert in China run as part of this study
als
o conforms to this
relationship (Table S2
-
S3; Fig. S4).
We used data from the Yan An (Shaanxi Province) climate station (36.6
°N
,
109.5
°
E, 959 m) for comparison due to its relatively close proximity to Puxian and
similar altitude. These data show that summ
er time (JJA) daily maximum temperatures in
this region peak at over 30°
C and average 28.8
±
1.4°
C (one standard deviation) for the
years 2007
-
2010, whereas mean summertime daily temperatures are in the order of 23
-
25
°
C and average 23.3
±
1.1°
C. MJJAS are
the months with a highest rainfall, as
8
expected in a monsoonal region, with July to October the most humid (Fig. S3). In
contrast the coldest
five months of the year generally have average daily temperature of
≤
6
°
C and very little rainfall (Fig S5) Therefo
re
given the strong seasonality in both
temperature and rainfall it appears a robust assumption that the predominant, if not
exclusive, period of growth of both soil carbonates and gastropods at Puxian is the
warmest and wettest summer months.
In a recent
study Zaarur et al., 2011 presented the first investigation of the
relationship between
Δ
47
measurements from terrestrial gastropod shells and
environmental temperatures
(41)
. They studied 10 different taxa from 12 different
locations, including tropical, desert, and temperate environments. Perhaps unsurprisingly
they did not find a universal
relationship between
Δ
47
-
derived
gastropod calcification
temperatures and environmental temperature that held across all taxa and environments,
and often calcification temperature were hotter than average environmental temperatures
during the expected grow
ing season. This was not attributed to a kinetic isotope effect, or
“vital effect” on
13
C
-
18
O bond abundance in terrestrial gastropods, but the authors
concluded that it was most likely a results of species and location specific differences in
the timing o
f calcification and between the optimum growth temperatures of different
taxa. However, when excluding data from
their study from samples that are from arid
environments (where winter is wettest season and the warmest months are dry) a
correlation with env
ironmental temperatures is observed (Fig S6). Therefore in
environments most similar to monsoonal China (ie. Where the warmest months are also
wet) it appears that
gastropod calcification temperatures are correlated with
environmental temperatures (Fig S6)
. However calcification temperatures are generally
hotter than both mean annual temperatures and warm month average temperature,
correlating most closely with average daily high temperatures (Fig S6). This is an
indication that terrestrial gastropod
s
must
have a warm affinity for optimum growth, an
observation that has been born out by culturing experiments on both shell forming and
no
n
-
shell forming gastropod taxa (eg.
(42, 43)
).
Therefore despite uncertainties on the precise relationship between gastropod
calcification temperatures and climatological data its clear that they have the potential to
archive climate information and in particular data on temperature change over time when
considering the same taxa and locality as this reduces the potential for s
pecies specific or
inter
-
location effects. From our
Δ
47
measurements from modern
Cathaica
shells from
Puxian we derive a
temperature of 31.2
±
1.5°
C (2 s.e.) which correlate with daily
summertime maximum temperatures (Fig S
5
-
S
7
). Crucially we find that gla
cial
specimens of
Cathaica
record a significantly colder
temperature of 24.2
±
1.9°
C
. It seems
unlikely that this could reflect a lengthening of the growing season of
Cathaica
at the
LGM as given its apparent affinity for warm conditions for optimum growth
and if
anything it might be expected to have a shorter summer growing season at the LGM.
As gastropod shells are likely to integrate climate information over a much shorter
time period and than nodular soil carbonates we took the approach of measuring a
large
number of individual shells (>32). The
necessity of large numbers of numbers of replicate
measurements to gain significant constraints of temperature is illustrated by the relatively
higher variability of gastropod derived
Δ
47
(Table S1
-
S3; Fig. S7). The
high
inter
-
individual variability seen in
our
gastropod
Δ
47
data
may also go some way to explaining
the scatter in the modern day calibration study. As least squared linear regression lines
9
through the modern gastropod calib
ration dataset have relatively large uncertainties in
slope (Fig. S7) we did not attempt to develop a transfer function to relate calcification
temperatures to environmental temperatures
and instead chose to simply quote the data as
“
gastropod calcificatio
n temperatures
” (Table 1), a change in which from LGM to present
would clearly relate to a change in warm summer month temperatures.
Ultimately a crucial test of our assumptions on the systematics of interpreting
temperatures from both soil carbonate and g
astropod
Δ
47
measurements is whether they
give mutually consistent results. The fact that they both give the same answer, within
error, suggests that our approach is robust enough for our main objective which is
distinguishing between lower magnitudes of c
ooling (2
-
3°
C)
°
C and greater than 5
°
C of
temperature change that allows us to critically evaluate model outputs and previous
interpretations of proxy data
.
B2. Isotope flux models of terrestrial gastropod body water
We used the land snail isotope flux mo
del of Balakrishnan and Yapp, 2004 to
constrain changes in precipitation
δ
18
O
using our reconstructed values of gastropod body
water
δ
18
O
(44)
. The water input is from surface water that the snail body comes in
contact with. Thus, their main water source is surface water collected on the ground. As
land snails are active during
and following rain events, this surface water should be
isotopically close to rainwater. The aragonite shells of snails are precipitated from body
water assumed to be at an isotopic steady state
(44)
. The main water loss is through the
evaporation of body w
ater into the atmosphere. Assuming that the direct loss of liquid
water is negligible compared to the loss due to evaporation, the snail isotope model is
equivalent to the Craig
-
Gordon model
(45)
commonly applied to evaporation from lakes
and leaves. At a s
teady state, the isotopic composition of snail body water (
R
sbw
) is
calculated as:
R
sbw
=
α
wv
(
α
k
(1
-
h
surf
)
R
surf
+
h
surf
R
vap
)
where
h
surf
is the relative humidity of the snail habitat,
α
k
is the kinetic fractionation
during evaporation (1.0285
)
(46)
,
and
α
wv
is the temperature
-
dependent equilibrium
fractionation between liquid water and vapor
(47)
.
R
surf
and
R
vap
are the isotopic
composition of surface water and local water vapor, respectively. For the high humidity
surface habitat of snails, we assume
that atmospheric vapor is in isotopic equilibrium
with the surface water (
R
vap
=
R
surf
/
α
wv
) so that snail body water can be approximated as:
R
sbw
= (
α
wv
α
k
(1
-
h
surf
) +
h
surf
)
R
surf
Using the mean
δ
18
O of summer (JJA) precipitation, approx.
-
7.1
‰,
as
δ
18
O
surf
for the
sample site at present
-
day conditions yields a local relative humidity of 0.82 to explain
the
δ
18
O
sbw
values of modern snails. This is close to the nighttime humidity range of 0.85
to 0.95 considered optimal for snail farming
(48)
.
The LGM
-
modern shift in
δ
18
O
sbw
reflects three factors: (1) changes in
α
wv
caused
by changes in temperature, (2) changes in the local relative humidity
h
surf
of the snail
habitat, and/or (3) changes in
δ
18
O
surf
related to mean summer precipitation. Temperature
re
lated changes in
α
wv
are usually small, only about 0.5
‰ (
Table S7). If the only change
10
in
δ
18
O
surf
at the LGM was the 1.2
‰
enrichment due to the increase in global ice volume,
to explain the observed
δ
18
O
sbw
difference would require a shift to substantiall
y higher
h
surf
(
Table S7). However, a higher local relative humidity at the LGM is unlikely given
the generally drier climate compared to present
-
day conditions.
For the more plausible assumption that the local relative humidity of the snail
habitat was t
he same (or slightly drier) at the LGM, the observed
δ
18
O
sbw
difference
requires large changes in the isotopic composition of
surface water, and hence
precipitation. For constant
h
surf
, the resulting
δ
18
O
surf
was approx
imately
5.2
‰
lighter at
the LGM
(also including ice
-
volume correction)
, which we assume is
the most likely
outcome. If we assume a reduction in relative humidity by 5%, then we calculate an
LGM
δ
18
O
surf
that was
7.1
‰ lighter at the LGM (Table S7)
which is a significant
mismatch to soil carbonate data and model results and so is probably an indi
cation that
humidity in the snail microenvironment was buffered and stable from LGM to present
.
Using
δ
18
O
rain
data from nearby GNIP (Global Network of Isotopes in Precipitation) sites
yields a range of possible
δ
18
O values for precipitation during the LGM
, however, the
LGM
-
modern
δ
18
O difference is constant irrespective of the site (Table S8). Thus, the
change in body water
δ
18
O of snails indicates a substantial shift in the
δ
18
O of mean
summer precipitation towards LGM values that were depleted by at leas
t ~
5.2
‰, relative
to present
according to this model of body waters
(Fig. S8).
B3. Analysis of PMIP2 model output
We have analyzed the northern
-
hemisphere summer (JJA) stationary wave
response in PMIP2 model simulations of the LGM in order to give context
to our
simulations with the LMD models. The JJA stationary wave response in each model is
demonstrated by differencing LGM simulations and the same model under present
-
day
(PD) conditions. Fig. S3 displays LGM
-
PD of the 500
-
hPa wind and surface temperat
ure
anomalies for JJA. Circulating wind vectors indicate the enhancement of stationary
waves in the LGM versus PD. Regions
with enhanced cyclonic (CCW) winds roughly
coincide with regions of enhanced cooling, demonstrating the importance of stationary
wa
ves for regional climate sensitivity in these models.
The HadCen (oav) and FGOALS models stand out as having enhanced high
-
latitude cooling relative to the others. The HadCen model, however, shows more than
double the
cooling at Puxian (
-
5.9
o
C) as compa
red to FGOALS (
-
2.65
o
C). In the 500
-
hPa
wind anomalies, the HadCen (oav) has a much more pronounced and southward
-
shifted
cyclonic (CCW) circulation north of Puxian. This dynamic feature advects cool air from
the north towards Puxian. In addition, the I
PSL simulation has enhanced cooling at
Puxian (
-
5.74
o
C), and also shows evidence for an enhanced cyclonic circulation north of
Puxian
(Figure 2)
. This feature north of Puxian is present in the simulation in our study
(see Figure
1
in main text), and is re
sponsible for the enhanced northwesterlies at Puxian.
B
4
. Comparisons of
LMDZ
simulations with clumped isotope data
The zoomed model simulations predict a mean annual temperature change of
-
6.4°C from the present to the LGM, and a mean summer temperat
ure change of
-
6.3°C.
This is consistent with the reconstructed temperature change of
-
7+/
-
2°C from terrestrial
gastropods and
-
5.5 ± 2.0°C from soil carbonates
. The zoomed simulations predict a
11
decrease in mean annual precipitation
δ
18
O
of
-
3
.
3
‰
(with
ice
-
volume correction
)
, which
is comparable
to the reconstructed value of
-
2
‰
from soil carbonates and
-
5.2
‰
from
snails
.
The
standard simulations with the LMDZ model predicts changes in mean annual
temperature and precipitation
δ
18
O
of
-
8.2°C and
-
2
‰ after ice
-
volume correction)
respectively. As discussed in the main text, within error temperature estimates from
gastropods and soil carbonate are consistent with both LMD
Z
simulations, and suggest a
greater magnitude of cooling than mos
t PMIP2 models suggest. Both model simulations
and proxy archives support a shift towards lighter isotopic composition
of precipitation at
the LGM, but differ in the precise value. Therefore we can validate this fundamental
result in our model, and go on t
o interrogate the causes of isotope shifts in our model
output, but cannot constrain which of the two LMDZ simulations is closest to the true
value.
B5. Causes of temperature changes in LMDZ simulations
The temperature change at the LGM compared to prese
nt is amplified with
altitude and latitude (Fig. S5). The amplification with altitude has already been
documented in syntheses of tropical terrestrial records such as that of Farrera et al.,
1999
(49)
.
It can be explained by the weakening of the atmospheric
lapse rate, which
follows a moist adiabat in the tropics
42
. Iso
-
contours of temperature changes are closed to
horizontal in the tropics, and their intersection with the topography explain the
amplification of temperature changes at higher elevation (Fig.
S11). The amplification
with latitude has also been documented by data syntheses such
as that of Shakun and
Carlson, 2010
(50)
.
LMDZ simulates a cooling of of
-
2 to
-
3°C in the Western Pacific, in
agreement with the data synthesized by Shakun and Carlson
,
2
010
(50)
.
At the longitude
of Puxian, as for the rest of the globe, the LGM cooling increases with latitude up to
-
12°C at 60°N.
To understand this amplification with altitude, we performed a radiative kernel
decomposition of the climate sensitivity
(
51
)
.
This method allows us to quantify the
radiative feedbacks of surface temperature changes on top
-
of
-
the
-
atmosphere radiation
associated with different processes: water vapor feedback, lapse rate feedback, cloud
feedbacks and surface albedo feedbacks. This
method had already been used successfully
to analyze the climate sensitivity for both future
(
51
)
and LGM
(
52
)
changes.
Results show
that the surface albedo
feedback increases with latitude and exhibits a secondary peak on
the Tibetan plateau (Fig. S11). Th
erefore, the albedo feedback is likely the major
contribution to the amplification of
temperature changes with latitude and altitude. It is
associated with enhanced snow cover at LGM. Note that this decomposition method does
not take into account the effec
ts of large
-
scale circulation.
In addition, the changes in large
-
scale circulation amplify the LGM cooling. To
support this we compare PMIP2 models and LMDZ and relate their LGM
-
PD cooling at
Puxian to the magnitude of northerly wind to the Northwest of Pu
xian.
Figure
2 in the
main text
shows a clear correlation between 500
-
hPa, JJA northerly winds averaged over
40N
-
50N
latitude and 80E
-
110E longitude and simulated cooling at Puxian, LGM
-
PD.
Enhanced northerly flow results in enhanced advective cooling fro
m the region north of
Puxian. From inspection of Fig. S3, we conclude these northerly winds are associated
with continental
-
scale stationary waves,
thereby directly linking the regional climate
12
sensitivity of East Asia to large
-
scale dynamics.
A simulati
on without ice
-
sheets (ie.
with
present
-
day topography and land ice fraction)
exhibited suppressed stationary waves at
the LGM. This simulation exhibited
~30%
less cooling than the standard LGM LMDZ
simulation, an indication that
stationary waves enhanced
by continental ice
-
sheets are a
significant factor in cooling at the site (Fig. 2; Fig. S9)
Somewhat counter
-
intuitively, LGM with PD
land albedo gives the strongest
northwesterly winds at
Puxian of any model, and interactive land albedo reduces the
north
westerlies (and associated Puxian cooling) in the LMDZ model. The LGM
-
PD
anomalies for a set of sensitivity runs (Fig. S9
) help to clarify. In particular, the "LGM
w/PD albedo" (
top right
panel) has very clean
-
looking mid
-
latitude stationary waves with
o
bvious correlations in surface temperature anomalies, especially over East Asia. In
comparison, stationary waves in the standard LGM case with interactive land albedo (top
left
panel) are
clearly apparent but
not as organized over mid
-
latitude Asia, and i
n
particular the cooled region over East Asia is more compact in the PD
-
albedo case.
Removing the ice sheets (bottom panel)
very substantially reduced the amplitude of the
stationary wave response
.
Taken together, our analyses suggests the LGM cooling at
Puxian is dominated by stationary waves, and interactive land feedback can either
enhance or reduce this cooling.
B6. Causes of water
δ
18
O
changes in high
-
resolution simulation
Proposed major controls on precipitation
δ
18
O (
18
O
p
) in this monsoon regio
n (e.g
.
(35,
53
)
)
include either local effects (e.g
.
(
54
)
),
or through isotopic depletion along air mass
trajectories (e.
g.
(35)
).
Here we focus on understanding the LGM
-
PD difference of JJA
mean precipitation, which is the time recorded by proxies. The high
-
resolution
simulations indicate decreased LGM water
δ
18
O
may originate due to the (1) decrease in
local temperature associated with the altitudal and latitudal amplification of LGM
cooling, (2) decrease in last condensation temperature associated with the weakening
lapse
-
rate in the tropical troposphere, (3
) circulation changes that bring more dry and
depleted air from the North
-
West, and (4)
δ
18
O
p
recording vapor
δ
18
O
(
δ
18
O
v
) changes at
the condensation level where
δ
18
O
v
changes are amplified due to (1) and (2).
The change in mean annual
δ
18
O
p
can be decom
posed into changes of
δ
18
O
p in
the different seasons and changes in the precipitation seasonality. Changes in the
seasonality of precipitation on the Loess Plateau explain only 11% of the simulated
annual
-
mean
δ
18
O
p
change. Most of the change in mean annu
al
δ
18
O
p
is explained by
changes in
δ
18
O
p
in June
-
July
-
August (JJA, 23%) and in September
-
October
-
November
(SON, 67%). This is because half of the precipitation in Puxian in the model falls in JJA,
and the other half in SON, and changes in
δ
18
O
p
are larges
t in SON.
The simulated
-
3.3‰ change in JJA
-
mean
δ
18
O
p
in the zoomed simulation (after
correction for sea
-
water change) can be decomposed into 2 components: a component
related to the change in the (
δ
18
O
v
), which is controlled by large
-
scale processes, a
nd a
change in the difference between
δ
18
O
p
and
δ
18
O
v
. The change in
δ
18
O
v
explains 87% (
-
2.9‰) of the
δ
18
O
p
change. In addition, spatial patterns of
δ
18
O
p
changes are similar to
those in
δ
18
O
v
(Fig S12). Therefore, most of the following paragraphs will f
ocus on
understanding the change in
δ
18
O
v
in JJASON.
13
In LMDZ at LGM, precipitation increases over Northern India and over Burma
(Fig. S12). This leads to a depletion of
δ
18
O
p
in these regions, but the effect on
δ
18
O
v
is
very limited (Fig S12). The
δ
18
O
p
d
epletion is not due to the depletion of the vapor, but to
the decrease of rain reevaporation in a moister atmosphere leading to more depleted
precipitation relatively to the vapor
(29,
55
)
. In LMDZ, the effect of precipitation on
δ
18
O
p
is thus restricted to
convective regions, and is not exported up to higher latitude
sites such as Puxian. This suggests that precipitation changes are not responsible for the
observed
δ
18
O
p change. However, LMDZ has been suspected to overestimate the local
controls of convecti
on on
δ
18
O
p
and underestimate its remote effects through
δ
18
O
v
(39)
.
Therefore, we cannot rule out a small contribution of the precipitation increase over India
to the
δ
18
O
p
decrease at Puxian, as suggested by Pausata et al., 2011
(35)
.
The pattern of
δ
18
O
v
change exhibits similarities with that of temperature changes
(Fig. S12). Cooling and depletion are both largest over the Tibetan plateau and at high
latitude. This suggests that the temperature change is a factor responsible for the
δ
18
O
v
change. However
, the most striking similarity of patterns in between
δ
18
O
v
changes and
relative changes in specific humidity (dq/q, where q is the specific humidity) (Fig. S12).
The spatial correlations in the Puxian region between
δ
18
O
v
changes and dq/q are 0.79,
while
the spatial correlation between
δ
18
O
v
changes and temperature changes are 0.35 in
JJA. The good relationship between
δ
18
O
v changes and
Δ
q/q is predicted by Rayleigh
distillation:
R
v
=R
v
0*(q/q0)**(alpha
-
1)
(Eq 1)
where R
v
and q are the isotopic ratio a
nd specific humidity of the distilled air vapor, R
v
0
and q0 are the isotopic ratio and specific humidity of the initial vapor, and alpha is the
fractionation coefficient. For small changes in q and R
v
, we get:
Δ
δ
18
O
v
~
Δ
R
v
/R
v
*1000 = (alpha
-
1)*1000*
Δ
q/q
(Eq 2)
The slope of the simulated spatial relationship
Δ
q/q is about 8‰ in JJA This is thus
consistent with the order of magnitudes of the predicted slope of (alpha
-
1)*1000, which
is 9.5+/
-
3 permil for Puxian PD and LGM temperatures.
Simulated
δ
18
O
v
chan
ges correlate much better with dq/q than with temperature
changes because both specific humidity and
δ
18
O
v
depend on the temperature during the
last condensation event
(
56
, 5
7
)
, which can be different from the local temperature. For
example, the amplificati
on of temperature changes with altitude is restricted to low
latitude (equatorward of about 35°C, Fig. S11). In contrast, the amplification of dq/q and
δ
18
O
v
changes with altitude holds at all latitudes including that of Puxian. This is because
the dry and
depleted anomalies associated with colder temperature temperatures in the
tropical
-
subtropical upper
-
troposphere, where condensation occurs, are propagated
poleward by the mean flow.
Also, the circulation changes discussed in section B5 likely contribut
e to the
drying and to the depletion of the water vapor at Puxian at LGM, due to the stronger
northwesterly component of the winds at LGM (Figure 3 of main text).
Finally, we mentioned that the decrease in
δ
18
O
v
explains only 87% of the
14
decrease in
δ
18
O
p
. The remaining 13% are due to the altitude amplification of
δ
18
O
v
changes (Fig. S13). Most of the condensation over Puxian occurs between about 5 km
and 10 km above ground level. Therefore, the precipitation
at Puxian records a LGM
-
recent anomaly of the vapor at about 6 km (Fig. S13). This leads to additional 0.4‰
depletion in
δ
18
O
p
compared to
δ
18
O
v
.
Supplementary references
1.
Kukla G (1987) Loess stratigraphy in central China
Quatern. Sci. Rev.
6(3
-
4):
191
-
207.
2.
Kukla G & An Z (1989) Loess stratigraphy in central China
Palaeogr.
Palaeoclimatol. Palaeoecol.
72:203
-
225.
3.
Ding ZL
, et al.
(2002) Stacked 2.6
-
Ma grain size record from the Chinese loess
based on five sections and correlation with the deep
-
s
ea
δ
18
O record.
Paleoceanography
17(3):10.1029/2001PA000725.
4.
Sheng X, Chen J, Cai Y, Chen Y, & Ji J (2005) Aragonite
-
calcite transformation
in fossil snail shells of loess sequences in Loess Plateau, Central China.
Chinese
Science Bulletin
50:891
-
895.
5
.
Rao Z, Zhu Z, Chen F, & Zhang J (2006) Does delta
-
13
Ccarb of the Chinese loess
indicate past C
3
/C
4
abundance? A review of research on stable carbon isotopes of
the Chinese loess.
Quatern. Sci. Rev.
25:2251
-
2257.
6.
Passey BH, Levin NE, Cerling TE, Brown
FH, & Eiler JM (2010) High
-
temperature environments of human evolution in East Africa based on bond
ordering in paleosol carbonates.
Proc. Natl. Acad. Sci. USA
107(25):11245
-
11249.
7.
Affek H & Eiler JM (2006) Abundance of mass 47 CO2 in urban air, car exh
aust,
and human breath
Geochim. Cosmochim. Acta
70(1):1
-
12.
8.
Huntington KW
, et al.
(2009) Methods and limitations of 'clumped' CO
2
isotope
(Delta47) analysis by gas
-
source isotope ratio mass spectrometry.
J. Mass
Spectrom.
44(9):1318
-
1329.
9.
Eiler JM &
Schauble E (2004)
18
O
13
C
16
O in Earth’s atmosphere.
Geochim.
Cosmochim. Acta
68(23):4767
-
4777.
10.
Ghosh P
, et al.
(2006)
13
C
–
18
O bonds in carbonate minerals: a new kind of
paleothermometer.
Geochim. Cosmochim. Acta
70(6):1439
-
1456.
11.
Guo W, Mosenfelder J
L, Goddard III WA, & Eiler JM (2009) Isotopic
fractionations associated with phosphoric acid digestion of carbonate minerals:
insights from first
-
principles theoretical modeling and clumped isotope
measurements.
Geochim. Cosmochim. Acta
73(24):7203
-
7225.
1
2.
Dennis KJ & Schrag DP (2010) Clumped isotope thermometry of carbonatites as
an indicator of diagenetic alteration.
Geochim. Cosmochim. Acta
74(14):4110
-
4122.
13.
Tripati AK
, et al.
(2010)
13
C
–
18
O isotope signatures and ‘clumped isotope’
thermometry in f
oraminifera and coccoliths
Geochim. Cosmochim. Acta
74(20):5697
-
5717.
15
14.
Eagle RA
, et al.
(2010) Body temperatures of modern and extinct vertebrates
from
13
C
-
18
O bond abundances in bioapatite.
Proc. Natl. Acad. Sci. USA
107(23):10377
-
10382.
15.
Thiagaraj
an N, Adkins J, & Eiler J (2011) Carbonate clumped isotope
thermometry of deep
-
sea corals and implications for vital effects.
Geochim.
Cosmochim. Acta
75:4416
-
4425.
16.
Dennis KJ, Affek HP, Passey BH, Schrag DP, & Eiler JM (2011) Defining an
absolute refer
ence frame for 'clumped' isotope studies of CO
2
.
Geochim.
Cosmochim. Acta
75:7117
-
7131.
17.
Kim S
-
T, Mucci A, & Taylor B (2007) Phosphoric acid fractionation factors for
calcite and aragonite between 25 and 75
deg C.
Chem. Geol.
246:135
-
146.
18.
Swart PK,
Burns SJ, & Leder JJ (1991) Fractionation of the stable isotopes of
oxygen and carbon in carbon
-
dioxide during the reaction of calcite with
phosphoric
-
acid as a function of temperature and technique.
Chem. Geol.
86(2):89
-
96.
19.
Kim S
-
T & O'Neil JR (1997)
Equilibrium and nonequilibrium oxygen isotope
effects in synthetic carbonates.
Geochim. Cosmochim. Acta
61(16):3461
-
3475.
20.
Braconnot P
, et al.
(2007) Results of PMIP2 coupled simulations of the Mid
-
Holocene and Last Glacial Maximum
-
Part 1: Experiment
s and large
-
scale
features.
Clim. Past
3:261
-
277.
21.
Hourdin F
, et al.
(2006) The LMDZ4 general circulation model: Climate
performance and sensitivity to parametrized physics with emphasis on tropical
convection.
Clim. Dyn.
27:787
-
813.
22.
Marti O
, et al
.
(2005) The New IPSL Climate System Model.
Tech. Rep.,
IPSLCM4, IPSL
.
23.
Meehl GA
, et al.
(2007) The WCRP CMIP3 multimodel dataset:.
Bull. Amer.
Met. Soc.
88:1383
-
1394.
24.
Van Leer B (1977) Towards the ultimate conservative difference scheme. IV. A
new
approach to numerical convection.
J. Comput. Phys.
23:276
-
299.
25.
Hourdin F & Armengaud A (1999) The use of finite volume methods for
atmospheric advection of trace species. Part I: Test of various formulations in a
general circulation model.
Mon. Wea. Re
v.
127:822
-
837.
26.
Emanuel KA (1991) A scheme for representing cumulus convection in large
-
scale
models.
J. Atmos. Sci.
48:2313
-
2329.
27.
Emanuel KA & Zivkovic
-
Rothman M (1999) Development and evaluation of a
convection scheme for use in climate models.
J
. Atmos. Sci.
56:1766
-
1782.
28.
Bony S & Emanuel KA (2001) A parameterization of the cloudiness associated
with cumulus convection; Evaluation using TOGA COARE data.
J. Atmos. Sci.
58(3158
-
3183).
29.
Risi C, Bony S, Vimeux F, & Jouzel J (2010) Water stable
isotopes in the
LMDZ4 General Circulation Model: model evaluation for present day and past
climates and applications to climatic interpretation of tropical isotopic records.
J.
Geophys. Res.
115:D12118, doi:12110.11029/12009JD013255.
30.
Gates WL (1992)
AMIP: The Atmospheric Model Intercomparison Project.
Bull.
Am. Meteorol. Soc.
73:1962
-
1970.
31.
Steen
-
Larsen HC
, et al.
(2011) Understanding the climatic signal in the water
16
stable isotope records from the NEEM shallow firn/ice cores in northwest
Greenland
.
J. Geophys. Res.
116:D06108; doi:06110.01029/02010JD014311
32.
Gao J
, et al.
(2011) Precipitation Water Stable Isotopes in the South Tibetan
Plateau: Observations and Modeling.
J. Climate
24:3161
-
3178.
33.
Risi C (
201
2
) Process
-
evaluation of tropospheri
c humidity simulated by general
circulation models using water vapor isotopologues. Part 1: comparison between
models and observations.
J. Geophys. Res.
117:
DOI:
10.1029/2011JD016621
34.
Lee JE, Fung I, DePaolo DJ, & Otto
-
Bliesner B (2008) Water isotopes
during the
Last Glacial Maximum: New general circulation model calculations.
J. Geophys.
Res.
113:D19109, doi:19110.11029/12008JD009859.
35.
Pausata FR, Battisti DS, Nisancioglu KH, & Bitz CM (2011) Chinese stalagmite
δ
18
O controlled by changes in the Indi
an monsoon during a simulated Heinrich
event
Nat. Geosci.
4:474
-
480.
36.
Berger AL (1978) Long
-
term variations of caloric insolation resulting from the
earth's orbital elements.
Quatern. Res.
9:139
-
167.
37.
Peltier WR (1994) Ice age paleotopography.
Scienc
e
265:195
-
201.
38.
Labeyrie LD, Duplessy JC, & Blanc PL (1987) Variations in mode of formation
and temperature of oceanic deep waters over the past 125,000 years.
Nature
327:477
-
482.
39.
Vimeux F, Tremoy G, Risi C, & Gallaire R (2011) A strong control of
the South
American SeeSaw on the intra
-
seasonal variability of the isotopic composition of
precipitation in the Bolivian Andes.
Earth Planet. Sci. Lett.
307:47
-
58.
40.
Breecker D, Sharp ZD, & McFadden L (2009) Seasonal bias in the formation and
stable isot
ope composition of pedogenic carbonate in modern soils from central
New Mexico. .
GSA Bulletin
121:630
-
640.
41.
Zaarur S, Olack G, & Affek HP (2011) Paleo
-
environmental implication of
clumped isotopes in land snail shells
Geochim. Cosmochim. Acta
75(22):6859
-
6869.
42.
Aufderheide J
, et al.
(2006) Effects of husbandry parameters on the life
-
history
traits of the apple snail, Marisa cornuarietis: effects of temperature, photoperiod,
and population density.
Invertebrate Biology
125(1):9
-
20.
43.
Risin
g TL & Armitage KB (1969) Acclimation to temperature by the terrestrial
gastropods,
Limax maximus
and
Philomycus carolinianus
: oxygen consumption
and temperaature preference.
Comp. Biochem. Physiol.
30:1091
-
1114.
44.
Balakrishnan M & Yapp CJ (2004) Flux ba
lance models for the oxygen and
carbon isotope compositions of land snail shells.
Geochim. Cosmochim. Acta
68(9):2007
-
2024.
45.
Craig H & Gordon LI (1965) Deuterium and oxygen 18 variations in the ocean
and marine atmosphere.
proc. Stable Isotopes in Ocean
ographic Studies and
Paleotemperatures
, (V. Lishi e F., Pisa., Spoleto, Italy), pp 9
-
130.
46.
Merlivat L (1978) Molecular diffusivities of H
2
16
O, HD
16
O, and H
2
18
O in gases. .
J. Chem. Phys.
69:2864
–
2871.
47.
Majoube M (1971) Fractionnement en oxygen
-
18 et
en deuterium entre l'eau et sa
vapeur.
J. Chem. Phys.
68:1423
-
1436.
48.
Thompson R & Cheny S (1996)
Raising snails
(National Agriculture Library
Special reference briefs. , NAL SRB 96
-
05).
17
49.
Farrera I
, et al.
(1999) Tropical climates at the Last Glacial
Maximum: a new
synthesis of terrestrial palaeoclimate data. I. Vegetation, lake
-
levels and
geochemistry.
Clim. Dyn.
15(823
-
856).
50.
Shakun JD & Carlson AE (2010) A global perspective on Last Glacial Maximum
to Holocene climate change
Quatern. Sci. Rev.
2
9:1801
-
1816.
5
1
.
Soden BJ
, et al.
(2008) Quantifying Climate Feedbacks Using Radiative Kernels.
J. Climate
21:3504
-
3520.
5
2
.
Yoshimori M, Yokohata T, & Abe
-
Ouchi A (2009) A Comparison of Climate
Feedback Strength between CO2 Doubling and LGM Experiments.
J
. Climate
22:3374
-
3395.
5
3
.
Wang Y
, et al.
(2008) Millennial
-
and orbital
-
scale changes in the East Asian
monsoon over the past 224,000 years.
Nature
451:1090
-
1093.
5
4
.
Dansgaard W (1964) Stable isotopes in precipitation.
Tellus
16:436
-
468.
5
5
.
Risi C, Bon
y S, Vimeux F, Chong M, & Descroix L (2010) Evolution of the water
stable isotopic composition of the rain sampled along Sahelian squall lines.
Q. J.
R. Meteorol. Soc.
136:227
-
242.
5
6
.
Galewsky JA, Sobel A, & Held I (2005) Diagnosis of subtropical humidity
dynamics using tracers of last saturation.
J. Atmos. Sci.
62:3353
-
3367.
5
7
.
Galewsky J & Hurley JV (2010) An advection condensation model for subtropical
water
vapor
isotopic
ratios.
J.
Geophys.
Res.
115(D16116):doi:10.1029/2009JD013651.
5
8
.
Thompson LG
,
et al.
(1997) Tropical Climate Instability: The Last Glacial Cycle
from a Qinghai
-
Tibetan Ice Core.
Science
276:1821
-
1825.
59
.
Thompson LG
, et al.
(1989) Holocene
-
Late Pleistocene Climatic Ice Core Record
from Qinghai
-
Tibetan Plataeu.
Science
246:474
-
477.
6
0
.
Wang YJ
, et al.
(2001) A high
-
resolution absolute dated Late Pleistocene
Monsoon Record from Hulu Cave, China.
Science
294:2345
-
2348.
6
1
.
Dykoski CA
, et al.
(2005) A high
-
resolution, absolute
-
dated Holocene and
deglacial Asian monsoon record from Dongg
e Cave, China.
Earth Planet. Sci.
Lett.
233:71
-
86.
Fig. S1:
Map
showing
the central Chinese
L
oess
P
lateau
.
Fig. S2:
Pictures of
gastropod
specimens.
Scale b
ars represent 1 cm.