Water-stable isotopes in the LMDZ4 general circulation model: Model evaluation for present-day and past climates and applications to climatic interpretations of tropical isotopic records

TL;DR: In this article, the LMDZ-iso general circulation model was used to simulate water-stable isotopes from a midlatitude station and evaluated at different time scales (synoptic to interannual).

Abstract: We present simulations of water-stable isotopes from the LMDZ general circulation model (the LMDZ-iso GCM) and evaluate them at different time scales (synoptic to interannual). LMDZ-iso reproduces reasonably well the spatial and seasonal variations of both delta O-18 and deuterium excess. When nudged with reanalyses, LMDZ-iso is able to capture the synoptic variability of isotopes in winter at a midlatitude station, and the interannual variability in mid and high latitudes is strongly improved. The degree of equilibration between the vapor and the precipitation is strongly sensitive to kinetic effects during rain reevaporation, calling for more synchronous vapor and precipitation measurements. We then evaluate the simulations of two past climates: Last Glacial Maximum (21 ka) and Mid-Holocene (6 ka). A particularity of LMDZ-iso compared to other isotopic GCMs is that it simulates a lower d excess during the LGM over most high-latitude regions, consistent with observations. Finally, we use LMDZ-iso to explore the relationship between precipitation and delta O-18 in the tropics, and we discuss its paleoclimatic implications. We show that the imprint of uniform temperature changes on tropical delta O-18 is weak. Large regional changes in delta O-18 can, however, be associated with dynamical changes of precipitation. Using LMDZ as a test bed for reconstructing past precipitation changes through local delta O-18 records, we show that past tropical precipitation changes can be well reconstructed qualitatively but not quantitatively. Over continents, nonlocal effects make the local reconstruction even less accurate.

Summary

1. Introduction

• Because of differences in mass and symmetry of the main isotopic forms of the water molecule (H2 16O, HDO, H2 18O), an isotopic fractionation occurs during phase changes depending on atmospheric conditions.
• It is the atmospheric component of the Institut Pierre Simon Laplace (IPSL) ocean‐land‐atmosphere coupled model [Marti et al., 2005] that participated in CMIP3 [Meehl et al., 2007].
• In section 2, the authors describe the LMDZ4 model, the implementation of water‐stable isotopes and the various simulations performed.
• In section 3, the authors evaluate the simulation of the isotopic composition for the present‐day climatology, synoptic variability, interannual variability and past climates.

2.1. LMDZ4 GCM

• The dynamical equations are discretized in a latitude‐ longitude grid, with a standard resolution of 2.5° × 3.75° and 19 vertical levels.
• Water in its vapor and condensed forms is advected by the Van Leer advection scheme [Van Leer, 1977], which is a monotonic second‐order finite volume scheme.
• It includes in particular the Emanuel convective parameterization [Emanuel, 1991; Grandpeix et al., 2004] coupled to the Bony and Emanuel [2001] cloud scheme.
• Each grid cell is divided into four subsurfaces: ocean, land, ice sheet and sea ice.
• In the stand‐alone version of LMDZ4 used here, the land surface is represented as a simple bucket model, and land surface evaporation is calculated as a single flux: no distinction is made between transpiration, bare soil evaporation, or evaporation of intercepted water by the canopy.

2.2. Isotopic Processes

• Water isotopic species (H2 16O, H2 18O and HDO) are transported and mixed passively by the large‐scale advection and various air mass fluxes.
• In the Van Leer advection scheme, it is assumed that the water content advected from one box to the next is a linear combination of the water contents in the two grid boxes involved.
• While the proportion of the drop that reequilibrates isotopically is prescribed in many GCMs [e.g., Hoffmann et al., 1998], here the relative proportion of evaporative enrichment and diffusive equilibration is calculated depending on relative humidity following Stewart [1975].
• In addition, the model takes into account the evolution of the compositions of both the rain and the surrounding vapor as the rain drops reevaporate [Bony et al., 2008].

2.3.1. AMIP Simulations

• A first 1979–2007 simulation has been performed following the AMIP protocol [Gates, 1992], using prescribed monthly and interannually varying SST and sea ice and a constant CO2 value of 348 ppm.
• Another simulation, named “nudged,” uses the same protocol but was nudged by the three‐dimensional horizontal winds from ERA‐40 reanalyses [Uppala et al., 2005] until 2002 and operational analyses thereafter.
• A first LGM simulation was performed following a protocol similar to PMIP1 [Joussaume and Taylor, 1995], using the Climate: Long‐Range Investigation, Mapping, and Prediction [ Project Members, 1981] SST and sea ice, a CO2 concentration of 180 ppm, orbital parameters following Berger [1978].
• The authors use here the climatological SST from an LGM simulation performed under the PMIP2 protocol (Braconnot et al. [2007], with LGM orbital configuration and a CO2 concentration of 185 ppm), averaged over 50 years.
• The authors force their additional LGM simulation with T′LGM/IPSL = TLGM/IPSL − TPI + TPD.

3. Evaluation and Sensitivity Tests

• The present‐day climate simulated by LMDZ4 has been extensively evaluated by Hourdin et al. [2006].
• The mean annual temperature and precipitation maps in the nudged simulation are given in Figure 1 for reference.
• The authors present an evaluation of d18O, expressed in permil, defined as ¼ Rsample RSMOW 1 1000; where Rsample and RSMOW are the ratio of HDO or H2 18O over H2 16O in the sample and the Standard Mean Ocean Water (SMOW) reference, respectively.
• At first order, variations in dD follow the same patterns as d18O but are 8 times larger.
• The authors thus present an evaluation of this parameter as well, which is expected to provide stronger constraints on the simulated hydrological and isotopic processes.

3.1. Evaluation of the Spatial and Seasonal Distributions

• The authors use in this section the whole AMIP simulations averaged over the period 1979–2007 to produce average seasonal cycles.
• Note that since the authors compare point data with simulated values averaged over a GCM grid box, the scale mismatch may contribute to the model data difference.
• 1.1. Annual Mean Spatial Distribution of Isotopes in Precipitation [27].
• In LMDZ‐iso, setting l to 0.002 leads to very strong dp values over central Antarctica (up to 28‰), whereas setting l to 0.004 gives results more consistent with the data .
• During the dry season, simulated dp values increase with continentality as one goes inland, in agreement with observations [Salati et al., 1979; Gat and Matsui, 1991; F. Vimeux et al., manuscript in preparation, 2010].

3.1.3. Evaluation of the Vapor‐Precipitation Equilibrium

• A large uncertainty in the representation of water‐ stable isotopes in GCMs is the representation of isotopic exchanges between vapor and rain droplets as the rain falls and partially reevaporates [Lee and Fung, 2008].
• Caution is necessary for two sources of uncertainties in the model‐ data comparison, in addition to possible uncertainties in the data.
• First, vapor samples were not collected every day and thus may not be representative of monthly averages, given the significant variability observed in the vapor at the daily time scale [e.g., Angert et al., 2008].
• In the data over all stations, the rain is more enriched than the low‐level vapor (d18Op − d18Ov ranges from +7 to +20‰), but over Vienna and Manaus the rain is more depleted (by up to 6‰) than would be expected if the rain was in complete equilibrium with the vapor.
• This behavior is qualitatively well captured by LMDZ‐iso, but the noisiness of the data limits any deeper analysis. [37].

3.2. Evaluation of the Isotopic Variability at the Synoptic Scale

• The authors evaluate the ability of the nudged simulation to simulate the variability at the daily and weekly scale.
• Nudging with reanalyzed winds enables a more rigorous evaluation of the isotopic variability at the synoptic scale [Yoshimura et al., 2008].
• Here the authors present an evaluation using unpublished daily data of both vapor and precipitation collected at the surface at the station of Saclay (near Paris, France, 48.73°N, 2.17°E) from September 1982 to September 1984.
• The temporal slope of d18Ov versus temperature at the daily scale in winter is underestimated by the model (0.2‰/K in LMDZ‐iso and 0.6‰/K in the data).
• LMDZ‐iso, when nudged by reanalyses, can thus satisfactorily simulate the day‐to‐day variability in temperature that is related to large‐scale atmospheric disturbances, and the associated d18O variability in vapor and precipitation (at least qualitatively).

3.3. Evaluation of the Isotopic Variability at the Interannual Scale

• Water isotopes have been shown to record interannual to decadal variability of the precipitation in the tropics [Hoffmann, 2003; Ramirez et al., 2003] and modes of variability in the extratropics such as the North Atlantic Oscillation [Baldini et al., 2008; Sodemann et al., 2008] or the Southern Annular Mode [Noone and Simmonds, 2002b].
• The simulation nudged by reanalyzed winds simulates better than the free simulation the interannual variability in temperature, precipitation, and isotopes, as can be shown by time series over Vienna and Bangkok.
• The improvement is particularly strong in midlatitudes.
• The correlation is 0.80 between model and data monthly anomalies of temperatures (filtered with a 6 month running mean) over Vienna in the nudged simulation compared to 0.05 in the free simulation.
• Therefore, LMDZ‐iso, when forced by observed SST and nudged by reanalyzed winds, simulates relatively well the interannual variability in d18Op, though it has more difficulties in simulating dp.

3.4. Evaluation of Isotopic Variations at Paleoclimatic Scales

• The authors have seen that LMDZ‐iso reproduces reasonably well the present‐day climate and its variability from the synoptic, regional scale to the interannual, large scale.
• The authors evaluate the capacity of LMDZ‐iso to simulate the isotopic changes associated with two past climates (described in section 2.3.3): the Last Glacial Maximum (LGM) and the Mid‐Holocene (MH).

3.4.1. Last Glacial Maximum

• Comparing the model results to the data for the LGM is not straightforward.
• On the other hand, if the SST change has a lower equator to pole gradient, as simulated by the IPSL or CAM coupled models [Lee et al., 2008], then using the spatial slope for temperature reconstruction leads to an underestimation of past temperature changes (by about 40% in the LMDZ‐iso simulation forced by IPSL SSTs).
• Another typical failure of isotopic GCMs for the LGM is their inability to simulate the lower dp measured in ice cores at high latitudes during LGM [Werner et al., 16 of 27 2001], or more generally to simulate d18O and d variations of the same sign on climatic time scales [Noone, 2008].
• Therefore, LMDZ‐iso is not able to reproduce the d18Op changes in monsoon regions that are out of phase between hemispheres [Cruz et al., 2009], but erroneously produces more negative d18Op throughout the entire tropical belt .
• The corresponding P − d18Op slopes are much higher than at the interannual or seasonal scales.

4. Climatic Information Recorded by Water Isotopes in the Tropics

• At longer time scales, the interpretation of isotopic records from tropical ice cores has been the subject of debate.
• This could suggest a large‐scale control of the isotopic signal, which was first interpreted as temperature variations [Thompson et al., 2000].
• Given that the main process controlling low‐latitude d18O variations at present day is the precipitation amount, these variations have subsequently been interpreted as wetter conditions upstream of ice cores [Vimeux et al., 2009]. [59].
• Given the ability of LMDZ‐iso to reproduce the main features of the observed water isotopic distributions, the authors now use it to investigate issues related to the interpretation of isotopic records as proxies for past changes in temperature and precipitation.

4.1. How Much Do Global Temperature Changes

• A global temperature change is likely to imprint d18Op over the whole planet.
• Therefore, the sensitivity of d18Op to mean SST in the tropics simulated by LMDZ cannot explain by itself the strong depletion in d18Op measured in the tropics for the LGM.
• As the tropical d18Op is closely related to the precipitation amount, the ability of GCMs to reproduce past d18Op changes might help to assess, indirectly, the ability of GCMs to simulate the precipitation response to a global climate change.
• The probability that the relative error in reconstructed DP be smaller than 50% at locations where |d18Op| changes are larger than 2‰ is only 20% for the LGM with IPSL SSTs and 17% for the MH (Table 4). [67].
• Where measured changes of d18Op are high (>2‰), it is very likely that the reconstructed DP has the right sign (92%): this means that interannual and climatic controls on d18Op are similar.

5.1. Evaluation of LMDZ‐iso

• The authors present the implementation of water‐stable isotopes in the LMDZ‐iso GCM, and evaluate the present‐day isotopic distribution simulated at different time scales: synoptic, seasonal, and interannual, as well as for past climate changes.
• LMDZ‐iso forced by observed SSTs reproduces the annual mean and the seasonal distribution of d18Op reasonably well, as well as its interannual variability in the tropics.
• Numerous sensitivity tests were performed on both isotopic and nonisotopic parameters of the model.
• More measurements are certainly needed to better constrain these processes.
• In particular, the degree of equilibration of the rain drops with the vapor can be parametrized in many ways [e.g., Stewart, 1975; Hoffmann et al., 1998;Mathieu et al., 2002; Lee and Fung, 2008; Bony et al., 2008] and is difficult to evaluate owing to the scarcity of isotopic measurements in the vapor.

5.2. Interpretation of Paleoclimatic Proxies

• The accuracy of this reconstruction in Antarctica strongly depends on the equator‐pole SST gradients of the reconstructed past climate.
• If the equator‐pole at LGM was weaker than reconstructed by CLIMAP, then past temperature reconstructions in Antarctica would be underestimated, in agreement with Lee et al. [2008]. [78].
• In the tropics changes in d18Op may result from global‐scale changes in SSTs, and/or from regional precipitation changes associated with changes in SSTs that are not spatially uniform.
• The authors analysis suggests that past local precipitation changes can be reconstructed from d18Op records, but only in cases where the signal to noise ratio for d18Op is the largest.
• This overestimates the magnitude of precipitation changes.

5.3. Perspectives

• [80] LMDZ‐iso, like other GCMs, does not simulate the large isotopic depletion measured in tropical ice cores, questioning whether all processes affecting d18Op in the tropics are well represented.
• Tropical ice cores are located in mountainous regions, characterized by a complex topography which can only be resolved with high‐resolution modeling [Sturm et al., 2005].
• Besides, transpiration does not fractionate relatively to the soil water [Washburn and Smith, 1934; Barnes and Allison, 1988] whereas bare soil evaporation does and is thus depleted relative to the soil water [Moreira et al., 1997; Yepez et al., 2003; Williams et al., 2004].
• Moreover, processes by which precipitation is recycled (transpiration or evaporation from open water or soil) are suggested to strongly affect d excess gradients over the Amazon [Salati et al., 1979; Gat and Matsui, 1991; Henderson‐Sellers et al., 2004] and thus possibly the Andean ice core d excess.
• In LMDZ‐iso as in most other GCMs, the authors have assumed no fractionation when recycling precipitation over land, owing to the simplicity of the land surface model.

Figures

Figure 2. (a, b) Annual mean d18O and (c, d) d excess in precipitation, in the data (Figures 2a and 2c) and the LMDZ‐iso nudged simulation (Figures 2b and 2d). The data are the GNIP data [Rozanski et al., 1993], the Antarctica data from Masson‐Delmotte et al. [2008], and Greenland surface data [Masson‐ Delmotte et al., 2005b]. The data are gridded over a coarse 7.5 × 6.5° grid for visualization purpose. The simulation is the nudged simulation averaged over the 1979–2007 period. Zonal means of the annual averages of (e) d18O and (f) d excess for the data (red crosses), the nudged simulation (solid black curve), and the free simulation (dashed green curve). The lines are zonal averages for LMDZ‐iso simulations, and the points are the values at data stations.

Figure 8. (a) Monthly d18O in precipitation (solid curves) and vapor at the lowest level (dashed curves) over Vienna, Manaus, and Ankara for the GNIP‐vapor data (red curves) and the LMDZ‐iso nudged simulation (green curves). For the simulated vapor composition, the dotted green curves show the LMDZ results sampled on the same days as GNIP vapor measurements, whereas the dashed lines show the monthly means. (b) Difference between d18O in precipitation and the d18O of a liquid that would be in equilibrium with the vapor, for the GNIP‐vapor data (red curve) and the LMDZ‐iso nudged simulation (green curve). For LMDZ, the monthly means were used. (c) Same as Figure 8a but for d excess. (d) Same as Figure 8b but for d excess. The error bars for the GNIP vapor data correspond to the standard deviation for the different measurements available for a given month.

Figure 12. Annual average change between LGM and present day (PD) for (a) temperature, (b) precipitation, (c) d18O, and (d) d excess. The LGM simulation used the CLIMAP SST [CLIMAP Project Members, 1981). Stippling indicates where the LGM‐PD difference is stronger than the standard deviation of the interannual variability.

Figure 16. (a) LGM–present‐day change in P simulated by LMDZ‐iso. (b) LGM–present‐day change in P reconstructed from the interannual d18Op − P slope simulated by LMDZ‐iso. Areas where the correlation coefficient is higher than −0.25 are white. (c) Annually averaged precipitation change between LGM and PD as a function of the annually averaged change in d18Op, simulated by LMDZ‐iso, over all tropical locations. The curve shows the average while the bars show the median and quartiles. (d) Error in the reconstructed precipitation change (reconstructed minus simulated by LMDZ‐iso) as a function of change in d18Op for different methods: local slope between annual mean precipitation and annual mean d 18O (dashed blue curve); constant relationship of −0.7‰/(mm/d) (dash‐dotted green curve); and local slope between monthly anomalies of d18Op and precipitation, applied on monthly values (dashed brown curve).

Figure 4. Annual mean d excess as a function of annual mean dD in precipitation in Antarctica, in the data from Masson‐Delmotte et al. [2008] (green) and in LMDZ‐iso free simulations (blue and red). In blue, the supersaturation l was set to its standard value of 0.004, whereas in red, it was set to 0.002.

Figure 3. (left) Annual mean dD as a function of annual mean air temperature at first level and (right) annual mean d excess as a function of annual mean dD in precipitation, in the data fromMasson‐Delmotte et al. [2008] (green) and in the LMDZ‐iso nudged simulation (blue).

Table 4. Performance of the Reconstruction of Past Changes in Precipitation Using the Present‐Day Interannual Slope of d18Op Versus P in the Perfect Model Experimentsa

Figure 7. Average relationship between precipitation rate and (left) d18O and (right) d excess of tropical oceanic precipitation, at the monthly scale, for GNIP data (dash‐dotted green curve), and 3 year LMDZ‐ iso simulations with set to 0.9 (solid red curve) and to 0 (dashed blue curve). Error bars indicate the standard deviation in each precipitation bin.

Figure 11. (a, c, e, and g) Correlation coefficients and (b, d, f, and h) slopes of the interannual relationship between monthly anomalies (seasonal cycle subtracted) of d18O in precipitation and temperature (Figures 11a–11d) and precipitation rate (Figures 11e–11h). A 6 month running mean was applied to the time series of monthly anomalies before performing the linear regression. The regression is performed for the GNIP data (Figures 11a, 11b, 11e, and 11f) and the nudged 1979–2007 LMDZ‐iso simulation (Figures 11c, 11d, 11g, and 11h). White zones correspond to correlations lower than 0.4 in absolute value.

Figure 10. Monthly interannual anomaly of air temperature, precipitation rate, d18O, and d excess in precipitation over the (left) Vienna and (right) Bangkok GNIP stations, comparing GNIP data and the LMDZ‐iso nudged simulation.

Table 3. Data Used for Comparison With the MH Simulationa

Figure 14. Same as for Figure 12 but for Mid‐Holocene minus present day.

Table 2. Data Used for Comparison With the LGM Simulationa

Figure 9. Comparison of the synoptic variability of water isotopes simulated by LMDZ‐iso with that observed at the surface in Saclay (near Paris, France, 48.73°N, 2.17°E) from September 1982 to September 1984. Daily evolution of (a) temperature, (b) d18Op, (c) d 18Ov, (d) dp, and (e) dv. Probability density function of the daily (f) d18O and (g) d for all months, in both vapor and precipitation. Probability density function of the daily difference (h) between d18O in the precipitation and the vapor and (i) between d in the precipitation and the vapor. Observed temperatures are solid purple curves, measured d18Ov and dv are dashed purple curves, d 18Op and dp are dashed green curves, and LMDZ‐iso d 18O is solid red curve and d is solid blue curve.

Table 1. Name, Location, and Active Years of the GNIP Stations Chosen to Assess Seasonal Cycles and Interannual Variability of the Isotopic Composition of the Precipitationa

Figure 6. Monthly d18O in precipitation as a function of precipitation rate for the nine ocean tropical GNIP stations selected by Bony et al. [2008] for the GNIP data (dash‐ dotted green curve), the LMDZ‐iso nudged simulation (solid red curve), and the single‐columnmodel of Bony et al. [2008] (dashed blue curve). Curves are second‐order polynomial fits.

Figure 15. (a) Frequency distribution of the d18O of precipitation over the tropics (−30°S–30°N) for the different simulations: present day (red), idealized simulation of uniform SST changes, and the two LGM simulations. For each simulation, monthly d18Op values (averaged over the different years of simulation) for all the tropical grid boxes were used. (b) Annually and tropically average change in d18Op (compared to present day) as a function of the annually and tropically averaged change in sea surface temperature. (c) Mean relationship of d18Op over tropical oceans as a function of the large‐scale vertical velocity at 500 hPa w500. For each simulation and in each bin of w500, we calculate the average d 18Op using monthly w500 and d 18Op values in all tropical grid boxes. (d) Same as Figure 15c but as a function of the precipitation rate P.

Figure 1. Annual mean (left) temperature and (right) precipitation in the LMDZ‐iso nudged simulation.

Figure 13. Same as for Figure 12 but for the LGM simulation using the SSTs from an LGM simulation with the coupled IPSL model. The setup of this simulation is detailed in section 2.3.3.

Full text

HAL Id: hal-01142300
https://hal.archives-ouvertes.fr/hal-01142300
Submitted on 15 Apr 2015
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Water-stable isotopes in the LMDZ4 general circulation
model: Model evaluation for present-day and past
climates and applications to climatic interpretations of
tropical isotopic records
Camille Risi, Sandrine Bony, Françoise Vimeux, Jean Jouzel
To cite this version:
Camille Risi, Sandrine Bony, Françoise Vimeux, Jean Jouzel. Water-stable isotopes in the LMDZ4
general circulation model: Model evaluation for present-day and past climates and applications to
climatic interpretations of tropical isotopic records. Journal of Geophysical Research: Atmospheres,
American Geophysical Union, 2010, 115 (D12), pp.D12118. �10.1029/2009jd013255�. �hal-01142300�

Click
Here
for
Full
A
rticl e
Water‐stable isotopes in the LMDZ4 general circulation model:
Model evaluation for present‐day and past climates and
applications to climatic interpretations of tropical isotopic records
Camille Risi,
1
Sandrine Bony,
1
Françoise Vimeux,
2
and Jean Jouzel
3
Received 27 September 2009; revised 18 December 2009; accepted 13 January 2010; published 23 June 2010.
[1] We present simulations of water‐stable isotopes from the LMDZ general circulation
model (the LMDZ‐iso GCM) and evaluate them at different time scales (synoptic to
interannual). LMDZ‐iso reproduces reasonably well the spatial and seasonal variations of
both d
18
O and deuterium excess. When nudged with reanalyses, LMDZ‐iso is able to
capture the synoptic variability of isotopes in winter at a midlatitude station, and the
interannual variability in mid and high latitudes is strongly improved. The degree of
equilibration between the vapor and the precipitation is strongly sensitive to kinetic effects
during rain reevaporation, calling for more synchronous vapor and precipitation
measurements. We then evaluate the simulations of two past climates: Last Glacial
Maximum (21 ka) and Mid‐Holocene (6 ka). A particularity of LMDZ‐iso compared to
other isotopic GCMs is that it simulates a lower d excess during the LGM over most
high‐latitude regions, consistent with observations. Finally, we use LMDZ‐iso to explore
the relationship between precipitation and d
18
O in the tropics, and we discuss its
paleoclimatic implications. We show that the imprint of uniform temperature changes on
tropical d
18
O is weak. Large regional changes in d
18
O can, however, be associated with
dynamical changes of precipitation. Using LMDZ as a test bed for reconstructing past
precipitation changes through local d
18
O records, we show that past tropical precipitation
changes can be well reconstructed qualitatively but not quantitatively. Over continents,
nonlocal effects make the local reconstruction even less accurate.
Citation: Risi, C., S. Bony, F. Vimeux, and J. Jouzel (2010), Water‐stable isotopes in the LMDZ4 general circulation model:
Model evaluation for present‐day and past climates and applications to climatic interpretations of tropical isotopic records,
J. Geophys. Res., 115, D12118, doi:10.1029/2009JD013255.
1. Introduction
[2] Because of differences in mass and symmetry of the
main isotopic forms of the water molecule (H
2
16
O, HDO,
H
2
18
O), an isotopic fractionation occurs during phase changes
depending on atmospheric conditions. As a consequence,
water‐stable isotopes are widely used as a tracer of past
climate variations and of the present‐day water cycle. In
particular, the isotopic composition recorded in polar ice
cores have long been used to reconstruct past temperatures
[Dansgaard, 1953; Jouzel, 2003]. More recently, the isoto-
pic composition recorded in low‐latitude ice cores
[Thompson et al., 2000; Ramirez et al., 2003] or speleothems
[Wang et al., 2008; Cruz et al., 2005a] have also been used to
infer past temperatures [Thompson et al., 2000] or precipi-
tation rates [Hoffmann et al., 2003].
[
3] However, processes that control the water isotopic
composition are numerous and complex. For the Greenland
ice cores, for example, using the spatial slope as a surrogate
for the temporal slope to evaluate past local temperature
changes leads to a large uncertainty of a factor of 2 [Jouzel,
1999; Jouzel, 2003]. This could be due to a change in air
mass origins [Werner et al., 2001] or in precipitation sea-
sonality [Krinner et al., 1997b; Krinner and Werner, 2003],
or to a dampening of isotopic changes by ocean evaporation
[Lee et al. , 2008]. At low latitudes, the paleoclimatic
interpretation of isotopic records is even less quantitative.
Most of the tropical precipitation arises from convective
processes, which strongly affect the isotopic composition of
both vapor and precipitation [Lawrence et al., 2004; Bony
et al., 2008; Risi et al., 2008a, 2008b]. While the earliest
interpretation of Andean ice cores had linked isotopes to
temperatures [Thompson et al., 2000], more recent studies
have stressed the importance of the precipitation intensity
upstream the air mass trajector ies [Hoffmann, 2003; Vimeux
et al., 2005] and the role of tropical Pacific sea surface
temperatures (SSTs) on the isotopic variability in Andean
ice core records [Bradley et al., 2003]. As a consequence,
while Rayleigh distillation models (representing the loss of
1
LMD, IPSL, UPMC, CNRS, Paris, France.
2
UR Great Ice, IRD, LSCE, IPSL, (CEA, CNRS, UVSQ), Gif‐ sur‐
Yvette, France.
3
LSCE, IPSL (CEA, CNRS, UVSQ), Gif‐sur‐Yvette, France.
Copyright 2010 by the American Geophysical Union.
0148‐0227/10/2009JD013255
JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, D12118, doi:10.1029/2009JD013255, 2010
D12118 1of27

heavier isotopes during condensation and precipitation) are
useful to study at first order the evolution of air masses as
they are transported from a moisture source region to higher
latitudes [Ciais and Jouzel, 1994], more complex models are
necessary to take into account the numerous processes
affecting the isotopic composition of precipitation.
[
4] Atmospheric general circulation models (GCM) are
now frequently used for isotopic studies. They represent the
three‐dimensional transport of air masses and isotopes as
well as large‐scale condensation and atmospheric convec-
tion, albeit in a parameterized way. Since the pioneering
work of Joussaume et al. [1984], water isotopes have been
implemented in at least a half‐dozen GCMs: GISS [Jouzel
et al., 1987; Schmidt et al., 2007], ECHAM [Hoffmann et al.,
1998], MUGCM [Noone and Simmonds, 2002a], GENESIS
[Mathieu et al., 2002], CAM [Lee et al., 2007], GSM
[Yoshimura et al., 2008], Hadley GCM [Tindall et al., 2009]
as well as in regional models (REMO [Sturm et al. , 2005]).
They have been used, for example, to better understand how
the climatic signal is recorded by isotopes in ice cores, at the
interannual to decadal time scales [Vuille et al., 2003; Vuille
and Werner, 2005] and at paleoclimatic time scales [Werner
et al., 2001].
[
5] In this paper, we present the implementation of water‐
stable isotopes in the LMDZ4 model (whose isotopic ver-
sion is hereafter named LMDZ‐iso). The LMDZ4 model is
the GCM developed at the Laboratoire de Mtorologie
Dynamique (LMD) [Hourdin et al., 2006]. It is the atmo-
spheric component of the Institut Pierre Simon Laplace
(IPSL) ocean‐land‐atmosphere coupled model [Marti et al.,
2005] that participated in CMIP3 [Meehl et al., 2007]. Its
dynamical and physical packages have completely changed
since the pioneering work of Joussaume et al. [1984]. An
interesting particularity of this GCM now is the possibility of
using stretched grids [Hourdin et al., 2006], allowing studies
at both global and regional scales [e.g., Krinner et al.,
1997a].
[
6] The first goal of this paper is to evaluate the simula-
tion of water‐stable isotopes by LMDZ‐iso at different time
scales. We evaluate the present‐day isotopic spatial and
seasonal distribution and the isotopic variability at time
scales ranging from synoptic to interannual. For this pur-
pose, we have performed an Atmospheric Model Inter-
comparison Project (AMIP) [Gates, 1992] simulation forced
by monthly observed SSTs from 1979 to 2007. To evaluate
the isotopic simulation in a more rigorous way, we have also
performed an AMIP simulation over the same period with
the large ‐scale atmospheric dynamics nudged by meteoro-
logical reanalyses. Since a particular effort has been in-
vested in the representation of the droplet reevaporation in
the model [Bony et al., 2008], we pay particular attention to
evaluating the equilibrium between droplets and water vapor
using simultaneous vapor and precipitation data available at
some stations. We also pay a lot of attention to evaluating
the d excess, which is sensitive to kinetic fractionation
notably during rain reevaporation. Finally, we evaluate the
isotopic distribution for two past climates for which isotopic
data are available: Last Glacial Maximum (21,000 years
ago, 21 ka) and Mid‐Holocene (6 ka).
[
7] The second goal is to use LMDZ‐iso to investigate the
controls of the isotopic composition of precipitation in the
tropics, where the paleoclimatic interpretation is the most
uncertain. In particular, what are the relative influences of
temperature and precipitation changes on the isotopic
composition of tropical precipitation? How useful may d
18
O
records be for reconstructing past local precipitation changes
in the tropics?
[
8] In section 2, we describe the LMDZ4 model, the
implementation of water‐stable isotopes and the various
simulations performed. In section 3, we evaluate the simu-
lation of the isotopic composition for the present‐day cli-
matology, synoptic variability, interannual variability and
past climates. In section 4, we use LMDZ‐iso to explore what
paleoclimatic information is recorded in tropical isotopic
records. We conclude and give perspectives in section 5.
2. Model and Simulations Description
[9] In this section we briefly describe the LMDZ4 GCM,
the implementation of water‐stable isotopes and the dif-
ferent simulations performed.
2.1. LMDZ4 GCM
[
10] The dynamical equations are discretized in a latitude‐
longitude grid, with a standard resolution of 2.5° × 3.75°
and 19 vertical levels. Water in its vapor and condensed
forms is advected by the Van Leer advection scheme [Van
Leer, 1977], which is a monotonic second‐order finite vol-
ume scheme. The physical package is described in detail by
Hourdin et al. [2006]. It includes in particular the Emanuel
convective parameterization [Emanuel, 1991; Grandpeix et
al., 2004] coupled to the Bony and Emanuel [2001] cloud
scheme. Each grid cell is divided into four subsurfaces:
ocean, land, ice sheet and sea ice. In the stand‐alone version
of LMDZ4 used here, the land surface is represented as a
simple bucket model, and land surface evaporation is cal-
culated as a single flux: no distinction is made between
transpiration, bare soil evaporation, or evaporation of in-
tercepted water by the canopy.
2.2. Isotopic Processes
[
11] Water isotopic species (H
2
16
O, H
2
18
O and HDO) are
transported and mixed passively by the large‐ scale advec-
tion and various air mass fluxes. In the Van Leer advection
scheme, it is assumed that the water content advected from
one box to the next is a linear combination of the water
contents in the two grid boxes involved. For numerical
reasons, we assume similarly that the isotopic ratio of the
water advected from one box to the next (rather than the
isotopic content) is a linear combination of the isotopic
ratios in the two grid boxes involved (Appendix A).
[
12] Equilibrium fractionation coefficients between vapor
and liquid water or ice are calculated after Merlivat and Nief
[1967] and Majoube [1971a, 1971b]. We take into account
kinetic effects during the evaporation from the sea surface
following Merlivat and Jouzel [1979] and during snow
formation following Jouzel and Merlivat [1984], with the
supersaturation parameter l set to 0.004 to optimize the
simulation of d excess over Antarctica (section 3.1.1).
[
13] Given the simplicity of the land surface parameteri-
zation in LMDZ4, no information is available about the
fraction of the evapo,transpiration flux arising from frac-
tionating evaporation (e.g., evaporation of bare soil [Barnes
and Allison, 1988]). We thus assume no fractionation during
RISI ET AL.: WATER ISOTOPES IN LMDZ D12118D12118
2of27

the evapotranspiration over land, as done in most other
GCMs [e.g., Hoffmann et al., 1998; Lee et al., 2007]. The
coupling with the more detailed land surface scheme
ORCHIDEE [Ducoudré et al., 1993; Rosnay and Polcher,
1998; Krinner et al., 2005] will be reported in a subsequent
paper.
[
14] The implementation of water‐stable isotopes in the
convective scheme has been extensively described by Bony
et al. [2008]. We pay particular attention to the represen-
tation of the reevaporation and diffusive exchanges as the
rain falls, which is significantly different compared to other
GCMs. While the proportion of the drop that reequilibrates
isotopically is prescribed in many GCMs [e.g., Hoffmann
et al., 1998], here the relative proportion of evaporative
enrichment and diffusive equilibration is calculated depending
on relative humidity following Stewart [1975]. In addition,
the model takes into account the evolution of the compo-
sitions of both the rain and the surrounding vapor as the
rain drops reevaporate [Bony et al., 2008]. However, when
the relative humidity is 100% we simply assume total
reequilibration between raindrops and vapor, contrary to
Stewart [1975] and Lee and Fung [2008], who take into
account the raindrop size distribution in this particular case.
2.3. Simulations
2.3.1. AMIP Simulations
[
15] A first 1979–2007 simulation has been performed
following the AMIP protocol [Gates, 1992], using pre-
scribed monthly and interannually varying SST and sea ice
and a constant CO
2
value of 348 ppm. We allowed a spin‐up
time of 17 months before January 1979. This simulation is
named “ free.” Another simulation, named “nudged,” uses
the same protocol but was nudged by the three‐dimensional
horizontal winds from ERA‐ 40 reanalyses [Uppala et al.,
2005] until 2002 and operational analyses thereafter. We
did not notice any discontinuity associated with this change
in the nudging data set. The simulated wind fields are
relaxed toward the reanalyzed winds with a time constant
t = 1 h, so that each component of the horizontal wind field
u verifies the following differential equation:
@u
@t
¼
X
n
i¼1
U
i
þ
u
obs
 u

where u
obs
is the reanalysis wind and U
i
are the temporal
tendencies of each of the n dynamical and physical packages
in the model.
[
16] The 17 month spin‐up time seems to be enough to
reach an equilibrium: for example, in the nudged simulation,
the globally and annually average d
18
O in precipitation for
1979 is −7.56‰, very close to the average value over 1979–
2007 of −7.55‰ compared to the range of interannual
variability of 0.06‰ over 1979– 2007.
2.3.2. Sensitivity Tests
[
17] Sensitivity tests to tunable parameters in the physical
or isotopic parameterization have been performed on 3 year
simulations with climatological SST, with a spin‐up of
17 months. The sensitivities to parameters discussed in this
paper are much larger than the interannual variability,
justifying shorter simulations that are computationally less
expensive than the AMIP 1979–2007 simulations.
[
18] Additional 6 year simulations have been performed
using the same protocol, but with uniform SST perturbations
(as suggested by Cess and Potter [1988]): −4K,−2 K, and
+2 K. The sea ice distribution is not modified consistently
with the SST in these simulations, but we restrict their
analysis to tropical regions in this paper.
2.3.3. Past Climate Simulations
[
19] As suggested by the Paleoclimate Model Intercom-
parison Project (PMIP) project [Jous saume and Tay lor,
1995; Braconnot et al., 2007] and as in other isotopic
modeling studies [e.g., Jouzel et al., 2000], we perform past
climate simulations for two periods: the Last Glacial Max-
imum (LGM, 21 ka) and the Mid‐Holocene (MH, 6 ka). For
both these periods, a large amount of data is available for
model evaluation. These simulations are 5 years long, with a
spin‐up of 17 months.
[
20] A first LGM simulation was performed following a
protocol similar to PMIP1 [Joussaume and Taylor, 1995],
using the Climate: Long‐Range Investigation, Mapping, and
Prediction (CLIMAP) [CLIMAP Project Members, 1981]
SST and sea ice, a CO
2
concentration of 180 ppm, orbital
parameters following Berger [1978]. We set the sea surface
d
18
O to 1.2‰ [ Labeyrie et al. , 1987] and d to 0‰. Contrary
to the PMIP1 protocol, we use the Peltier [1994] ICE‐5G
ice sheet reconstruction (as in the work by Lee et al. [2008]
and PMIP2 [Braconnot et al., 2007]), which differs from the
ICE‐4G reconstruction (from Peltier [1994], used by
Joussaume and Jouzel [1993] and Jouzel et al. [2000]) in
the spatial extent and height of Northern Hemisphere ice
sheets. Except for the different ice sheet topography, our
simulation is similar to that performed by Joussaume and
Jouzel [1993], Jouzel et al. [2000], and Werner et al. [2001].
[
21] The MH simulations were performed following the
PMIP1 protocol, as in work by Jouzel et al. [2000]. The
only changes compared to present day are the orbital con-
figuration [Berger, 1978] and atmospheric gas concentra-
tions (CO
2
concentration of 280 ppm).
[
22] The LGM and MH simulations are compared to the
free AMIP simulation, considered as a reference for present
day (PD).
[
23] The warm tropical SSTs and the extensive sea ice of
the CLIMAP reconstruction have been questioned [e.g.,
MARGO Project Members, 2009]. Therefore, as in work by
Lee et al. [2008], we perform an additional simulation using
the SST and sea ice simulated by a coupled model (here the
IPSL model [Marti et al., 2005] for LGM conditions. We
use here the climatological SST from an LGM simulation
performed under the PMIP2 protocol (Braconnot et al.
[2007], with LGM orbital configuration and a CO
2
con-
centration of 185 ppm), averaged over 50 years. However,
significant SST biases in the IPSL model are common to all
climate conditions, including LGM and PD. Therefore the
direct comparison between SSTs simulated for LGM by the
IPSL model (T
LGM/IPSL
) and SSTs observed at PD (T
PD
)is
misleading: SST biases in the IPSL model could be con-
fused with LGM‐PD signals. To circumvent this problem,
we use the SSTs from an IPSL model pre‐industrial simu-
lation (PI) simulation, performed following the PMIP2
protocol (with present‐day orbital configuration and a CO
2
concentration of 280 ppm). We force our additional LGM
simulation with T′
LGM/IPSL
= T
LGM/IPSL
− T
PI
+ T
PD
. This
way, the biases in the IPSL model common to both the
RISI ET AL.: WATER ISOTOPES IN LMDZ D12118D12118
3of27

LGM and PI simulations are canceled out. We can thus
compare our LGM isotopic simulations from both CLIMAP
and IPSL SSTs in a consistent way.
3. Evaluation and Sensitivity Tests
[24] The present‐day climate simulated by LMDZ4 has
been extensively evaluated by Hourdin et al. [2006]. The
mean annual temperature and precipitation maps in the
nudged simulation are given in Figure 1 for reference. We
focus here on the isotopic simulation. First, we examine the
isotopic spatial and seasonal distribution, then its variability
at synoptic to interannual time scales in the present‐day
climate, and finally we evaluate isotopic variations associ-
ated with past climates.
[
25] We present an evaluation of d
18
O, expressed in per-
mil, defined as
 ¼
R
sample
R
SMOW
 1

 1000;
where R
sample
and R
SMOW
are the ratio of HDO or H
2
18
O
over H
2
16
O in the sample and the Standard Mean Ocean
Water (SMOW) reference, respectively. At first order, var-
iations in dD follow the same patterns as d
18
O but are
8 times larger. The deviation to this behavior is quantified
by the deuterium excess : d = dD − 8·d
18
O[Dansgaard,
1964]. This second‐order parameter is known to be more
difficult to simulate by GCMs [Lee et al., 2007; Mathieu
et al., 2002]. We thus present an evaluation of this parameter
as well, which is expected to provide stronger constraints on
the simulated hydrological and isotopic processes.
3.1. Evaluation of the Spatial and Seasonal
Distributions
[
26] We use in this section the whole AMIP simulations
averaged over the period 1979–2007 to produce average sea-
sonal cycles. We compare the spatial distribution and seasonal
cycle with the Global Network of Isotopes in Precipitation
(GNIP) data [Rozanski et al., 1993], to which we add data
from Antarctica (compiled by Masson‐Delmotte et al. [2008])
and Greenland (compiled by Masson‐ Delmotte et al.
[2005b]). Note that since we compare point data with simu-
lated values averaged over a GCM grid box, the scale mis-
match may contribute to the model data difference.
3.1.1. Annual Mean Spatial Distribution of Isotopes in
Precipitation
[
27] The spatial distribution of annual mean d
18
O in pre-
cipitation (d
18
O
p
) is well simulated in the model, featuring the
well‐known “effects” [Rozanski et al., 1993]: enhanced
depletion with decreasing temperature (“temperature effect”),
increasing altitude (“altitude effect”) or continentality
(“continental effect”), or precipitation intensity (“amount
effect”) (Figure 2). The d
18
O
p
over central Greenland and
Antarctica is however overestimated, owing to an over-
estimated temperature over Antarctica (minimum annual
temperature over Antarctica of −42°C in nudged LMDZ‐iso
and −60°C in the work by Masson‐Delmotte et al. [2008]).
This warm bias is frequent in GCMs [Masson‐Delmotte et al.,
2006] and is worsened when nudging the model with mete-
orological reanalyses. The temperature effect in Antarctica is,
however, relatively well reproduced, with a spatial slope of
0.73‰/K (r = 0.95) in the data and 0.65‰/K in the model
(r = 0.97) over the temperature range simulated by LMDZ‐
iso (Figure 3).
[
28] The deuterium excess in precipitation (d
p
) is of the
right order of magnitude over most regions except on
tropical continents (simulated d
p
too high by up to 10‰
over equatorial Africa and northern South America).
LMDZ‐iso reproduces the d minimum over high‐latitude
oceans [Uemura et al., 2008] and features a relationship
with dD consistent with observations over Antarctica
(Figures 3 and 4). LMDZ‐iso also captures the d
p
maximum
over the Middle East, as was also the case in other GCMs
[Hoffmann et al., 1998; Schmidt et al., 2007]. LMDZ‐iso
simulates the spatial distribution reasonably well compared
to other GCMs: Yoshimura et al. [2008] report spatial cor-
relations between 0.39 and 0.52 between observed and
simulated annual d
p
values in four isotopic GCMs; LMDZ‐
iso value is 0.45. However, d
p
is slightly underestimated by
Figure 1. Annual mean (left) temperature and (right) precipitation in the LMDZ‐iso nudged simulation.
RISI ET AL.: WATER ISOTOPES IN LMDZ D12118D12118
4of27

Frequently asked questions

Q1. Why does an isotopic fractionation occur during phase changes?

Because of differences in mass and symmetry of the main isotopic forms of the water molecule (H216O, HDO, H2 18O), an isotopic fractionation occurs during phase changes depending on atmospheric conditions. 

Q2. How is the reequilibration between vapor and precipitation reproduced?

The reequilibration between precipitation and vapor for d is well reproduced, with d most frequently 2‰ lower in average in precipitation than in the vapor in LMDZ‐iso and about 5‰ lower in the data. 

Q3. How much uncertainty does the spatial slope have?

For the Greenland ice cores, for example, using the spatial slope as a surrogate for the temporal slope to evaluate past local temperature changes leads to a large uncertainty of a factor of 2 [Jouzel, 1999; Jouzel, 2003]. 

Q4. What is the way to improve the reconstructions?

Considering the seasonal cycle of P and d18Op both for the calibration and reconstruction would improve the reconstructions quantitatively, in particular for past climates associated with strong changes in precipitation seasonality (e.g., MH). 

Q5. What is the sensitivity of d18Op to mean SST?

The probability distributions of d18O in the vapor and in the evaporation are equally shifted (not shown), suggesting that this small sensitivity to mean SST is mainly due to a change in fractionation during evaporation at the sea surface (a sensitivity of 0.08‰/K is predicted by the Merlivat and Jouzel [1979] simple closure assumption). 

Q6. How is the decrease in d18Op in the tropics?

Even when using SSTs from the IPSL coupled model, which are about −2.9 K colder than PD in the tropics, the decrease in d18Op is small (less than 2‰). 

Q7. Why have the isotopic compositions been used to reconstruct past temperatures?

In particular, the isotopic composition recorded in polar ice cores have long been used to reconstruct past temperatures [Dansgaard, 1953; Jouzel, 2003]. 

Q8. What is the effect of the LMDZiso model on the d18O?

LMDZ‐iso is not able to reproduce the d18Op changes in monsoon regions that are out of phase between hemispheres [Cruz et al., 2009], but erroneously produces more negative d18Op throughout the entire tropical belt (Figure 14). 

Q9. What is the probability of the reconstructed DP relative error?

The uncertainty in the reconstruction due to the seasonality of the precipitation is the largest for the MH simulation: the probability that the reconstructed DP relative error is smaller than 50% rises to 31% (compared to 16%) when considering seasonal information (Table 4). 

Q10. What is the effect of the SST on the accuracy of reconstructions?

This suggests that the controls of d18Op over Antarctica, and thus the accuracy of reconstructions based on present‐day spatial slopes, strongly depend on the pattern of SST change.