Journal of Geophysical Research: Atmospheres

Stable water isotopes in the ECHAM5 general circulation model: Toward high-resolution isotope modeling on a global scale



[1] In this study, a first set of four present-day global experiments with the ECHAM5 atmospheric general circulation model enhanced by stable water isotope diagnostics (ECHAM5-wiso) is presented. Model resolution varies from a typical coarse horizontal grid of 3.8° × 3.8° (T31) to a fine grid of 0.75° × 0.75° (T159). Vertical resolution varies from 19 to 31 model levels. On a global scale, the ECHAM5-wiso simulation results are in good agreement with available observations of the isotopic composition of precipitation from the Global Network of Isotopes in Precipitation (GNIP), on an annual as well as a seasonal time scale. In many instances, the isotope simulation results clearly benefit from an increased horizontal and vertical model resolution. The exemplary relevance of this model resolution dependence is demonstrated for the simulation of the isotopic composition of Antarctic precipitation. Here, the simulation with the fine T159L31 model resolution not only yields a better agreement with observational data sets but also allows for a more realistic retuning of the supersaturation function leading to improved deuterium excess performance over the Antarctic continent, which is important for the interpretation of polar ice cores. Finally, the ECHAM5-wiso simulation results are compared to newly available measurements of the isotopic composition of atmospheric water vapor. Model and data agree well, with differences in the range of ±10‰ for near-surface atmospheric values at several GNIP stations. A comparison of the ECHAM5-wiso simulations with total column averaged HDO data from the Scanning Imaging Absorption Spectrometer for Atmospheric Cartography (SCIAMACHY) instrument on board the environmental satellite Envisat shows the same latitudinal gradients but an offset between 20‰ and 50‰ of unknown origin.

1. Introduction

[2] During the past two decades, several atmospheric and oceanic general circulation models (GCMs) have been enhanced by the capability to explicitly simulate the hydrological cycle of the two stable water isotopologues (hereafter designated by the term “water isotopes”) HDO and H218O [e.g., Joussaume et al., 1984; Jouzel et al., 1987; Hoffmann et al., 1998; Schmidt, 1998; Mathieu et al., 2002; Schmidt et al., 2007; Lee and Fung, 2008; Tindall et al., 2009; Risi et al., 2010]. In a closed model system all relevant parameters determining the strength and evolution of isotopic fractionation are known. Global modeling of the isotopic composition of water may therefore help to interpret observed isotopic changes in various archives [Hoffmann et al., 2000; Jouzel et al., 2000]. A number of studies have already clearly demonstrated this possibility of an improved interpretation of detected water isotope variability in terms of climate and environmental change by appropriate GCM simulation results [e.g., Mathieu et al., 2002; Noone and Simmonds, 2002; Werner and Heimann, 2002; Vuille and Werner, 2005; Lee and Fung, 2008; Herold and Lohmann, 2009; Tindall et al., 2009; Risi et al., 2010]. At present, about a dozen different state-of-the-art GCMs with explicit isotope diagnostics exist [see Sturm et al., 2010, for a detailed model overview].

[3] We have now implemented the water isotope module into the atmospheric GCM ECHAM5 [Roeckner et al., 2003] and report first isotope simulation results. Compared to the previous model release, ECHAM4, a number of substantial general changes have been introduced in both the numerics and physics of the ECHAM5 model. These changes include a flux-form semi-Lagrangian transport scheme for positive definite variables like water components and chemical tracers [Lin and Rood, 1996], separate prognostic equations for cloud liquid water and cloud ice, a prognostic-statistical cloud cover parameterization [Tompkins, 2002], and a new cloud microphysical scheme [Lohmann and Roeckner, 1996]. For tropical convection processes, a mass flux scheme by Tiedtke [1989] with modifications for deep convection according to Nordeng [1994] is applied. Changes have also been made in the representation of land surface processes, including the representation of orographic drag forces [Lott and Miller, 1997]. In addition, a new data set of land surface parameters has been compiled for the ECHAM5 model [Hagemann, 2002]. On the other hand, horizontal and vertical diffusion, as well as the spectral dynamics of the model have remained essentially unchanged as compared to the previous ECHAM4 model.

[4] The overall ECHAM5 performance for different horizontal and vertical model resolution has been evaluated in detail by Roeckner et al. [2006]. It was found that for a low vertical resolution (L19 = 19 model levels) the model does not converge to a more realistic climate state for horizontal grid sizes smaller than approximately 2.8°. At higher vertical resolution (L31 = 31 levels), however, root-mean-squared errors (RMSE), calculated as the difference between ECHAM5 simulations and respective ERA-15 [Gibson et al., 1997] and ERA-40 [Uppala et al., 2005] data for the period 1979–1993, decrease monotonously with increasing horizontal resolution. Furthermore, in general, the L31 versions are superior to their L19 counterparts, and the improvements become more evident at increasingly higher horizontal resolutions. This applies, in particular, to the zonal mean climate state and also to stationary wave patterns in boreal winter. Roeckner et al. [2006] concluded that the substantial benefits in refining both horizontal and vertical resolution in ECHAM5 are in accord with scaling arguments deduced from quasi-geostrophic theory implying that horizontal and vertical resolution ought to be chosen consistently. For the isotope-relevant simulation of atmospheric temperatures, largest differences of up to 4°C between model values and the respective ERA data are detected in the polar upper troposphere and lower stratosphere [Roeckner et al., 2006].

[5] Hagemann et al. [2006] evaluated in detail the hydrological cycle in the ECHAM5 model. Over land, both simulated precipitation amount as well as integrated atmospheric water vapor agree in general well with the observational values, except for a 10–15% overestimation in the Tropics. Over ocean, there exists a general slight overestimation of both precipitation and water vapor by ECHAM5. Further model deficiencies reported Hagemann et al. [2006] are an overestimation of precipitation along steep mountain slopes and during the Asian summer monsoon season, as well as a dry bias over Australia. Analyzing the impact of the ECHAM5 model resolution on the simulated hydrological cycle, a beneficial effect of an increased vertical resolution on simulated precipitation with respect to both the annual mean and the annual cycle was found. On the other hand, the influence of increased horizontal resolution from T63 to even higher resolution is comparatively small. Most of the improvements at higher vertical resolution are due to large-scale moisture transport, whereas the impact of local water recycling through evapotranspiration is somewhat smaller.

[6] Our simulations with the water isotope-enabled ECHAM5 GCM (hereafter referred to as ECHAM5-wiso) are customary climate control simulations with fixed present-day boundary conditions. Three of these new ECHAM5-wiso simulations have been run with a horizontal model resolution between 3.8° (T31 spectral resolution) and 1.8° (T63), comparable to previous published similar isotope GCM studies [e.g., Mathieu et al., 2002; Lee et al., 2007; Herold and Lohmann, 2009; Tindall et al., 2009; Risi et al., 2010]. However, an additional ECHAM5-wiso simulation in T159 spectral mode employed a fine horizontal resolution of 0.75° by 0.75°, which is to our knowledge much finer than any other atmospheric isotope GCM study, so far. Similar spatial resolutions have been achieved in the past in isotope modeling studies using the regional circulation model REMOiso [e.g., Sturm et al., 2005] and are envisaged for future simulations of the LMDZ-iso model with stretched grid functionality [Risi et al., 2010]. In contrast to these approaches, our ECHAM5-wiso T159 model setup yields simulation results with an identical fine spatial scale on the whole globe within one single model experiment. We will examine the new simulation results of H218O and HDO on a global scale and address the question whether there is a convergence toward more realistic simulation results for an increased horizontal and vertical model resolution. This is also important for future climate applications, which will allow us choosing a suitable, computationally efficient model resolution for ECHAM5-wiso.

[7] For the validation of ECHAM5-wiso, we compare model values to the available global observational data set on the isotopic composition of precipitation provided by the Global Network of Isotopes in Precipitation (GNIP) [IAEA/WMO, 2006], as have previous studies with water isotope-enabled GCMs [e.g., Hoffmann et al., 1998; Schmidt et al., 2005; Lee et al., 2007; Risi et al., 2010]. As the most depleted values of H218O and HDO in precipitation are observed in East Antarctica, where ice cores provide a unique archive of past environmental changes [e.g., Petit et al., 1999; Jouzel et al., 2007; Fischer et al., 2008] this polar region is of high interest and constitutes an extreme test for isotope-enabled GCMs. For analyzing the ECHAM5-wiso model performance over Antarctica, we make use of the Antarctic observational database compiled by Masson-Delmotte et al. [2008].

[8] To further evaluate the model performance, the modeled isotopic composition of atmospheric water vapor is compared to near-surface vapor measurements available from several GNIP stations plus an only recently published global data set for the mean total column averaged HDO values of atmospheric water vapor [Frankenberg et al., 2009]. The latter is based on measurements using the Scanning Imaging Absorption Spectrometer for Atmospheric Cartography (SCIAMACHY) [Bovensmann et al., 1999] on board the European Space Agency's environmental research satellite Envisat.

2. Model Description and Simulation Setup

2.1. Model Description

[9] Both stable water isotopes H218O and HDO have been explicitly implemented into the hydrological cycle of ECHAM5 analogous to the isotope modeling approach used in the previous model releases ECHAM3 [Hoffmann et al., 1998] and ECHAM4 [e.g., Werner et al., 2001]. For each phase of ''normal” water (vapor, cloud liquid, cloud ice) being transported independently in ECHAM5 a corresponding isotopic counterpart is implemented in the model code. The isotopes and the ''normal” water are described identically in the GCM as long as no phase transitions are concerned. Additional fractionation processes are defined for the water isotope variables whenever a phase change of the “normal” water occurs in ECHAM5.

[10] For the water isotope fractionation during evaporation from oceanic surfaces, we use the bulk formula given in Hoffmann et al. [1998] which includes both equilibrium and nonequilibrium fractionation effects. The isotopic composition of the evaporative flux depends on the related sea surface temperature, the relative humidity, the near-surface wind speed, and the isotopic composition of the vapor above the ocean surface [Hoffmann et al., 1998].

[11] During evaporation from land surfaces, water in ECHAM5 evaporates either from a thin skin layer intercepting a certain fraction of the precipitation, a snow layer or a soil water pool. From the latter, soil water can either evaporate from bare soil, which includes in reality an isotopic fractionation effect [Gat, 1996], or be transpired through plants to the atmosphere, with no fractionation occurring within the plants [Bariac et al., 1994]. As the description of the combined evapotranspiration of water from land surfaces is not considered in sufficient detail in the ECHAM5 land surface model, we neglect any isotope fractionation during such evapotranspiration processes from land surfaces in ECHAM5-wiso. This simplification is applied in most other state-of-the-art atmospheric isotope GCMs [e.g., Lee et al., 2007; Tindall et al., 2009; Risi et al., 2010], too, but should be kept in mind when interpreting isotope model results.

[12] Substantial changes of coding the cycling of the water isotopes were necessary for the separate treatment of cloud water and cloud ice in the new stratiform cloud scheme. The scheme includes phase changes between the three water components and precipitation processes (autoconversion, accretion, aggregation). Furthermore, evaporation of rain and melting of snow are considered, as well as sedimentation of cloud ice. While this stratiform cloud scheme differs substantially from the one used in ECHAM4 [Roeckner et al., 1996] the principle formulation approach of fractionation processes during cloud formation in ECHAM5-wiso has remained the same and no ECHAM5-specific adjustments of isotope-related cloud parameters were required. Condensation either to ice or to liquid water within the cloud is primarily described as an equilibrium fractionation process. For liquid cloud water, it is assumed that the condensate stays in isotopic equilibrium with the surrounding during the condensation processes (“closed system”), while for ice crystal formation an instantaneous extraction of the condensate is specified (“open system”). The latter choice is based on the fact that the frozen condensate does not exchange with the surrounding vapor because of the low diffusivities of H218O and HD16O in ice. In the case of ice crystal formation, we furthermore replace the equilibrium fractionation factor by an effective fractionation factor that accounts for additional kinetic effects occurring at very low temperatures [Jouzel and Merlivat, 1984]. The required oversaturation function S has been set to S = 1.01 − 0.0045·Tcond (see section 4.2 for details on this formulation choice). For in-cloud water phase changes by melting, freezing or sublimation processes, no isotopic fractionation occurs [Jouzel, 1986].

[13] Like in the previous ECHAM4 model, an additional fractionation is applied when a raindrop falls through the undersaturated air below the cloud base. As ECHAM5-wiso does not contain any estimation of the speed of falling raindrops, we use the same assumptions as in ECHAM4: convective showers produce primarily large raindrops equilibrating isotopically to only 45% with the surrounding vapor, while large-scale clouds produce smaller drops equilibrating nearly completely to 95% [Hoffmann et al., 1998].

2.2. Simulation Setup

[14] For this study, four different ECHAM5-wiso simulations with average grid box sizes of 3.75° (T31), 1.88° (T63) and 0.75° (T159) have been performed. The simulations differ in both their horizontal and vertical model resolution (19 or 31 vertical layers between surface and 10 hPa) as well as in the internal time step of the model (Table 1). After a spin-up period of 1 year, which has been disregarded for the analyses, all simulations have been run for 10 years. Identical AMIP-conform [Gates et al., 1999] present-day boundary conditions, including monthly climatological sea surface temperatures and sea ice cover for the period 1979–1999 have been prescribed in all simulations.

Table 1. Overview of Different ECHAM5 Horizontal and Vertical Model Resolutions Used in This Study
Horizontal ResolutionVertical ResolutionModel Grid Size (degree)Model Time Step (min)

[15] For the lower boundary condition of the atmospheric isotopes, a global gridded data set based on the 18O isotopic composition in the seawater data set of LeGrande and Schmidt [2006] is specified for the isotopic composition of sea surface water and sea ice. As no equivalent data set of the HDO composition of seawater exists, the deuterium isotopic composition of the seawater in any grid cell has been set equal to the related 18O composition, multiplied by a factor of 8, in accordance with the observed relation for meteoric water on a global scale [Craig and Gordon, 1965].

[16] For all analyses, we express the H218O and HDO composition of different water bodies in the usual δ-notation as the deviation from the Vienna Standard Mean Ocean Water (V-SMOW). For H218O, the δ18O value of a water sample is calculated as δ18O = (([H218O]/[H216O])Sample/([H218O]/[H216O])V-SMOW − 1) · 1000. For HDO, δD is determined in an analog manner. Differences between both isotopes are expressed as the deuterium excess dex = δD − 8 · δ18O. Long-time mean values of δ18O and δD are then calculated as precipitation-weighted means, if not stated otherwise. For all quantitative data-model comparisons and calculations of various isotope relationships, we interpolate the ECHAM5-wiso simulation results from the various model grids to the same geographical locations as the observational data sets, if not stated otherwise.

3. Observational Data Sets

3.1. Isotopic Composition of Precipitation

[17] Since 1961, the International Atomic Energy Agency (IAEA) and the World Meteorology Organization (WMO) have operated the Global Network of Isotopes in Precipitation (GNIP). Since the start of GNIP, more than 800 meteorological stations in 101 countries have been collecting monthly precipitation samples [IAEA/WMO, 2006]. While several stations have continuously collected samples for two or more decades (e.g., GNIP stations in Krakow, Ottawa, Reykjavik, and Vienna), many other GNIP stations have only been in operation for a much shorter period. For this study, we select from the GNIP database those 231 stations that recorded for at least 1 complete calendar year monthly values of 2 m air temperature (T2m), precipitation amount (P), as well as both δ18O and δD in precipitation (hereafter designated as δ18Op and δDp, respectively). From this set of 231 GNIP stations, we selected for some more quantitative analyses a subset of 70 stations that contained a minimum of 5 calendar years of T2m, P, δ18Op, and δDp, any time within the period 1961 to 2007. We are aware that a 5 year record at a specific GNIP station might still be too short for representing an adequate climatological mean value at this station, comparable to the model values of the performed ECHAM5-wiso climate control simulations. However, there are too few GNIP locations with a 10 year (35 stations) or even a 20 year isotope record (8 stations) for a reliable reconstruction of the global δ18Op and δDp pattern from these two subsets.

[18] The strongest depletion of H218O and HDO in precipitation is observed for Antarctic snowfall and is related to both the relative long transport pathways of atmospheric water to Antarctica as well as the very low temperatures in this region [Dansgaard, 1964; Lorius et al., 1969]. Numerous annual mean Antarctic surface temperatures, accumulation rates and present-day isotopic values of Antarctic snowfall are documented [e.g., Morgan, 1982; Dahe et al., 1994; Giovinetto and Zwally, 1997], and Masson-Delmotte et al. [2008] compiled in the year 2008 all available observational Antarctic data sets. From this data set we have chosen those 176 records, which contain observational data of all four investigated variables (Tsurf, accumulation, δDp, dexp) at the same location.

3.2. Isotopic Composition of Vapor

[19] For the isotopic composition of near-surface atmospheric vapor (named δ18Ov and δDv), in the past, only very few, sporadic measurements existed [e.g., Jacob and Sonntag, 1991; Araguás-Araguás et al., 2000], and therefore most isotope modeling studies focused on comparison to various observational δ-values of precipitation. However, recently the GNIP database was expanded by first systematic samples of the isotopic composition of water vapor. For this study, we have selected those 5 GNIP stations, which contain at least 1 complete calendar year of monthly measurements (Vienna, Ankara, Belem, Manaus, and Rehovot).

[20] Besides the near-surface GNIP data of δ18Ov and δDv, two global data sets of HDO in water vapor, based on different satellite measurements, have been released during the past few years [Worden et al., 2007; Frankenberg et al., 2009], too. Further upper tropospheric and stratospheric vapor isotope data has become available from both aircraft measurements [Webster and Heymsfield, 2003], and from remote sensing [Moyer et al., 1996; Kuang et al., 2003]. For this study, we decided to focus on the recently published SCIAMACHY data set of mass weighted total column averaged δDv values of atmospheric water vapor [Frankenberg et al., 2009]. In contrast to the other data sets [e.g., Worden et al., 2007], the SCIAMACHY measurements cover a full 3 year period (2003–2005) as well as the total atmospheric water column. We are aware that even the SCIAMACHY values are still far from representing climatology mean values due to their limited observational period as well as the anomalous European summer heat wave in 2003 and the exceptional warm global surface temperature record in 2005.

[21] As the SCIAMACHY instrument on board the Envisat satellite does not fly over all areas of the Earth with the same frequency, the long-term mean SCIAMACHY values at different locations do not equally represent all seasons [see Frankenberg et al. [2009], Supplement, for details]. For a comparison of the ECHAM5-wiso δDv values to the satellite data, we have therefore calculated for every model grid box the relative contribution of SCIAMACHY measurements for each specific month and afterward calculated the accordingly weighted mean of simulated monthly δDv values. For example, if for a specific ECHAM5 model grid box in total 20 SCIAMACHY measurements existed, with 10 measurements in January, 5 in February, 5 in March, and no measurement during the rest of the year, the related weighted mean ECHAM5-wiso δDv value is calculated from the simulated monthly δDv values of January, February, and March only, with the January value weighted twice the February and March value. Furthermore, as SCIAMACHY measurements over the ocean depend on the existence of reflecting low-level clouds and therefore might not always represent the δDv value of the total atmospheric vapor column [Frankenberg et al., 2009], we limit this model-data comparison to land surface values.

4. Results and Discussion

4.1. Global Isotopic Composition of Precipitation

[22] Figure 1a shows the global distribution of δ18Op based on the selected 231 GNIP stations. Figures 1b1d show simulation results of three ECHAM5-wiso experiments with the following model resolution: T31L19 (Figure 1b), T63L31 (Figure 1c), and T159L31 (Figure 1d). All major characteristics of the global H218O pattern in precipitation as already described by Dansgaard [1964] can be detected in the global map of the selected GNIP stations (Figure 1a). Depletion of δ18Op is generally higher in midlatitude to high-latitude regions as compared to values in the low latitudes (temperature effect), and the highest depletion of δ18Op occurs over the polar ice sheets of Greenland and Antarctica. A gradient of isotopic depletion can be detected for water masses transported from the Atlantic toward Europe and Eurasia and toward eastern North America (continental effect), and highly depleted δ18Op values are also observed over alpine mountain regions like the Tibetan Plateau or the Andes (altitude effect). All these major characteristics of the global δ18Op distribution are well simulated by all three ECHAM5-wiso experiments. Furthermore, a comparison of the different spatial model resolutions in Figure 1 reveals that even for a rather coarse spatial model grid of 3.75° × 3.75° and 19 vertical levels (Figure 1b, T31L19 experiment), ECHAM5-wiso correctly simulates the global distribution of δ18Op. By decreasing the ECHAM5 model grid box size from 3.75° to 1.8° or 0.75° and simultaneously increasing the number of vertical levels from 19 to 31 (T63L31 and T159L31 simulations), the simulated global pattern of δ18Op does not substantially change (Figures 1c and 1d). Hence, on a global scale the ECHAM5-wiso results appear robust and do not depend on the chosen model resolution. However, in very dry desert regions, such as the Sahara, the Taklamakan, and central Australia, the year-to-year variability of the simulated annual mean δ18Op values slightly increases with a higher spatial resolution.

Figure 1.

Global maps of observed and simulated present-day mean annual H218O in precipitation (δ18Op) based on data from the Global Network of Isotopes in Precipitation (GNIP) and three different ECHAM5-wiso simulations. Figure 1a shows a spatial data interpolation of 231 GNIP stations that recorded at least 1 complete calendar year of monthly δ18Op values (open circles). For 70 out of these 231 stations, a monthly δ18Op record of at least 5 years exists (solid circles). Figures 1b–1d show results of three ECHAM5-wiso simulations with different horizontal and vertical model resolutions: (b) T31L19, (c) T63L31, and (d) T159L31. In each of these plots, the simulated mean δ18Op values are plotted on the raw model grid without additional interpolation. The hatched areas in Figures 1b–1d mark those model grid boxes where the simulated year-to-year variability of δ18Op is larger than ±2‰.

[23] While all four ECHAM5-wiso simulations result in a very comparable global δ18Op pattern, a finer spatial grid does allow a more realistic representation of the isotopic composition of precipitation on a continental or even smaller spatial scale. As an example, the different ECHAM5-wiso simulation results of δ18Op over Europe are shown in Figure 2. It can clearly be seen that the finer T159L31 model resolution (Figure 2d) results in a more realistic δ18Op pattern over many regions of Europe, e.g., the European Alps and the Scandinavian mountain range, as compared to the simulations with the coarse T31L19 (Figure 2b) or medium T63L31 (Figure 2c) resolution.

Figure 2.

As in Figure 1 but for Europe.

[24] For a more quantitative analyses of the different ECHAM5-wiso simulations, we refer to the global subset of 70 GNIP stations with at least 5 full years of measurements. The RMSE between the observed GNIP mean δ18Op values and the model results at these GNIP station locations varies between 2.2‰ (T31L19) and 1.3‰ (T159L31). For the T2m values, an almost identical RMSE of 2.1°C is calculated for both the T31L19 and T159L31 resolution. Thus, the simulated δ18Op values benefit more from the finer ECHAM5 model resolution than the T2m values near the various GNIP stations due to improved large-scale moisture transport [Hagemann et al., 2006]. This result is in agreement with the findings of Schmidt et al. [2007] that the δ18Op value at a specific location is in many cases a regionally integrated signal of several climate variables rather than a unique proxy for local changes of one particular climate variable.

[25] Next, we analyze the spatial relation of mean annual δ18Op versus local temperature and precipitation amount on a global scale (Figure 3). For the isotope-temperature relation, we select only those 61 GNIP stations (out of our subset of 70 stations) where the annual mean 2 m temperature (T2m) is below 20°C. The calculated linear spatial δ18Op − T2m relation for these GNIP stations is δ18Op = 0.46·T2m − 13.2, r2 = 0.67. This global relationship is similar in all four ECHAM5-wiso experiments (Table 2), and both GNIP and ECHAM5-wiso correlation coefficients are significant (p-value ≪ 10−3). Similarly, the difference between the global δ18Op-P relation for all 9 low-latitude GNIP stations with an annual mean temperature equal or above 20°C and the four ECHAM5-wiso experiments is rather small (Figure 3, bottom). While we calculate a δ18Op-P gradient of −0.15‰/(cm/month) for the 9 GNIP stations, the ECHAM5 results vary between −0.21‰/(cm/month) for the T31L19 simulation, and −0.10‰/(cm/month) for the T159L31 simulation. However, the significance of the correlation coefficients is rather low (GNIP: p-value is 0.025; ECHAM5-wiso: p-values are between 0.003 and 0.04), and the relative uncertainties of the calculated slopes are large (Table 2). Because of the few available tropical GNIP station records, it is impossible to determine whether the underestimation of the δ18Op-P gradient in the T159L31 simulation represents an essential model deficit or just occurred by chance. Regarding second-order isotopic effects, the ECHAM5-wiso simulated δDp-δ18Op relations are all in good agreement with the GNIP-based relation δDp = 8.2·δ18Op − 10.4, and the simulated differences in the regression lines are within the 95% confidence intervals (Table 2).

Figure 3.

Comparison of the spatial δ18Op-T2m relation (top), and δ18Op-P relation (bottom), as derived from the GNIP observational data (black and blue crosses) and four different ECHAM5-wiso simulations (red and green circles). The relation between δ18Op and temperature T2m (precipitation P) is analyzed for all GNIP station locations with an annual temperature below (above) 20°C and at least 5 complete calendar years of monthly measurements any time between 1961 and 2006. Colored lines indicate the calculated linear regression lines of δ18Op in relation to T2m and P for the different data sets.

Table 2. Linear Correlation Gradient m, y-intercept Value b, and Pearson Correlation Coefficient r2 for Both Observed and Simulated δ18Op-T2m, δ18Op-P, and δDp-δ18Op Relationshipsa
RelationshipData SetResolutionmbr2
  • a

    The observed relationships are based on data from the Global Network of Isotopes in Precipitation (GNIP), while simulated relationships are calculated for four different ECHAM5-wiso simulations. The uncertainties of m and b represent the 95% confidence interval of the applied simple linear regression calculations.

δ18Op-T2mGNIP 0.46 ± 0.07−13.2 ± 0.80.67
 ECHAM5-wisoT31L190.46 ± 0.04−11.7 ± 0.50.83
  T63L190.48 ± 0.04−12.4 ± 0.50.86
  T63L310.52 ± 0.04−12.8 ± 0.50.88
  T159L310.45 ± 0.05−12.0 ± 0.50.81
δ18Op-PGNIP −0.15 ± 0.10−1.4 ± 1.90.54
 ECHAM5-wisoT31L19−0.21 ± 0.09−1.7 ± 1.30.74
  T63L19−0.16 ± 0.12−2.5 ± 1.80.48
  T63L31−0.15 ± 0.09−2.7 ± 1.50.57
  T159L31−0.10 ± 0.07−3.1 ± 1.30.50
δDp-δ18OpGNIP 8.18 ± 0.1610.4 ± 1.60.99
 ECHAM5-wisoT31L198.13 ± 0.088.6 ± 0.71.00
  T63L198.18 ± 0.0710.3 ± 0.61.00
  T63L318.16 ± 0.0710.1 ± 0.61.00
  T159L318.29 ± 0.1012.1 ± 0.91.00

[26] To further evaluate the ECHAM5-wiso simulation results on a global scale, we analyze the model's capability to correctly simulate the seasonal cycle of both δ18Op and δDp (Figure 4). For this task we chose 5 GNIP stations from very different precipitation regimes which all have at least 5 complete years of monthly recorded δ18Op and δDp values. For Vienna, central Europe, all ECHAM5-wiso simulations show a correct seasonal timing and amplitude of T2m and δ18Op (Figure 4a). The δ18Op value is overestimated by approximately 2‰ in the T31L19 simulation, and the results improve for the winter when the ECHAM5 resolution is increased. Similarly, the amplitude of the deuterium excess (defined as dexp = δDp − 8·δ18Op) is overestimated in the coarse T31L19 simulation but becomes smaller and more realistic in both the T63L31 and T159L31 simulation. In contrast to these findings, the amplitude of the simulated precipitation amount cycle in Vienna is too large in all simulations and does not improve with a finer spatial model resolution. δ18Op values at Reykjavik are clearly influenced by the nearby Atlantic water source and thus less depleted than δ18Op values at Vienna. Furthermore the amplitude of the seasonal cycle is reduced (Figure 4b). Similar to the results for Vienna, we detect too high (too low) values of the simulated δ18Op (dexp) cycle in the T31L19 simulation, and the isotope model results improve for a finer model resolution. In agreement with the findings of Hagemann et al. [2006], the simulated precipitation amount near Reykjavik increases for the finer model resolutions, leading to unusually high precipitation values in late fall and early winter. For Ankara, the largest model deviations from the GNIP observations are detected for the seasonal cycle of dexp (Figure 4c). Again, the simulation results continuously improve when going from T31L19 to T159L31 resolution. While the seasonal cycle of the surface temperature T2m has been reasonably well simulated in all ECHAM5-wiso simulations for Vienna, Reykjavik and Ankara, the opposite is true for the GNIP stations Belem and Apia (Figures 4d and 4e). For both low-latitude locations, the ECHAM5-wiso model clearly overestimates the amplitude of the seasonal temperature cycle. While there is a constant model offset of approx. +1.5°C for the location of Apia, the ECHAM5-wiso simulations mainly deviate from the GNIP record during the period September to December for Belem. Simulated precipitation amounts near Belem are underestimated by a factor 2–3, while precipitation amounts near Apia fit the observations. Despite these clear model deficits for T2m and P, the seasonal cycle of δ18Op is remarkably well simulated both for Belem and Apia. For dexp, a clear offset of several ‰ is found in the model values at these locations. Both GNIP stations are strongly influenced by surrounding tropical sea surface conditions, and this marine influence is apparently underestimated in the ECHAM5-wiso simulation of near-surface temperatures. Furthermore, Apia is located on the relatively small island of Upolu (Samoa), which is not resolved as a land surface point for any of the applied ECHAM5 model resolutions.

Figure 4.

Seasonal cycles of temperature T2m, precipitation amount P, isotopic composition of precipitation δ18Op, and deuterium excess values dexp at five locations: (a) Vienna, (b) Reykjavik, (c) Ankara, (d) Belem, and (e) Apia. The bold red lines represent the observational GNIP values, while the thin colored lines indicate model results of four different ECHAM5-wiso simulations: T31L19 (black), T63L19 (green), T63L31 (yellow), and T159L31 (blue). For all 5 GNIP stations, at least 5 complete calendar years of monthly measurements of T2m, P, δ18Op, and dexp exist.

4.2. Isotopic Composition of Antarctic Precipitation

[27] In Figure 5 we compare the simulated mean annual δDp values for the spatial model resolution T159L31 with the chosen subset of observations compiled by Masson-Delmotte et al. [2008]. We find a general good agreement between ECHAM5-wiso and the observational data. Most depleted isotopic values occur in the dry inner region of East Antarctica with simulated minimum δDp values below −400‰. Less depleted precipitation values are found in coastal regions and West Antarctica. The RMSE of simulated annual mean temperature Tsurf, δ18Op and δDp values as compared to the observations amount to 6.3°C (Tsurf), 4.6‰ (δ18Op), and 36.1‰ (δDp). Thus, all RMSE are on the order of 10%–15% of the related observed mean Antarctic values (Tsurf: −37.8°C, δ18Op: −39.3‰, δDp: −307.9‰). Our analyses reveal that ECHAM5 fails to reproduce the very low surface temperatures of Antarctica, and consequently the isotopic values of precipitation in ECHAM5-wiso are less depleted as compared to the observations. Such a warm bias over the Antarctic continent has also been reported for other isotope-enabled GCMs [e.g., Lee et al., 2007; Risi et al., 2010] and is frequent in GCM simulations [Masson-Delmotte et al., 2006]. It might be linked to the general poor representation of the polar atmospheric boundary layer and related atmospheric inversion temperatures in GCMs [e.g., Krinner et al., 1997].

Figure 5.

Observed and simulated annual mean HDO values in precipitation (δDp) for Antarctica. Observational values (colored circles) stem from a compilation by Masson-Delmotte et al. [2008]. The underlying color pattern represents ECHAM5-wiso model results derived from a fine spatial T159L31 simulation. As in Figure 1, the raw model grid values are plotted without additional interpolation.

[28] Next, we compare the spatial δDp-Tsurf relation as well as the accumulation-δDp relation of the four different ECHAM5-wiso simulations to the data of Masson-Delmotte et al. [2008]. For this analysis, the simulated accumulation rate has been estimated as precipitation minus evaporation. For the coarse T31L19 resolution (Figure 6a), the spatial δDp-Tsurf gradient m is clearly underestimated in the simulation (m = 4.8‰/°C) as compared to the observations (m = 6.6‰/°C). However, increasing the horizontal resolution to T63 (Figures 6b and 6c) or even to T159 (Figure 6d) leads to a steeper δDp-Tsurf gradient, which is in much better agreement with the observations (T63L19: m = 5.4‰/°C; T63L31: m = 5.7‰/°C; T159L31: m = 6.8‰/°C). A comparison of the T63L19 (Figure 6b) and T63L31 (Figure 6c) results clearly shows that a combined increase of both the horizontal and vertical model resolution, in contrast to a pure refinement of the horizontal resolution, yields a further improvement of the ECHAM5-wiso values. This finding is consistent with that of Roeckner et al. [2006] in a more general context for ECHAM5. Similar findings as for the δDp-Tsurf gradient are valid for the different model simulations of the Antarctic accumulation-δDp relation (Figure 6, bottom). Especially for the high-resolution T159 simulation (Figure 6d), the agreement between the simulated and observed accumulation-δDp relation is remarkably good and a clear improvement of previously published isotope-enabled AGCM results [Masson-Delmotte et al., 2008]. However, even the T159L31 simulation does not capture rather high accumulation rates corresponding to δDp values in the range of −300‰ to −400‰. It has been proposed that the large range of accumulation values in this δDp range probably results from wind-driven postdepositional effects on the slopes leading to the Antarctic Plateau [Masson-Delmotte et al., 2008]. Wind erosion and drifting of snow, which may be a significant contributor to the Antarctic surface mass balance [Gallée et al., 2001; Genthon and Krinner, 2001], are not implemented in the ECHAM5 GCM.

Figure 6.

Comparison of the δDp-Tsurf relation (top) and the accumulation-δDp relation (bottom) for Antarctica as derived from the compilation of observational data by Masson-Delmotte et al. [2008] (black and blue crosses) and four different ECHAM5-wiso simulation (red and orange circles). Both relations are shown only for those records of the Masson-Delmotte et al. [2008] database that contain values of all four analyzed variables (temperature, accumulation, δDp, dexp) at the same location.

[29] The deuterium excess dexp in the Antarctic Snow is strongly influenced by kinetic fractionation effects occurring during the formation of ice crystals at very low temperatures [Jouzel and Merlivat, 1984]. One crucial but unknown parameter controlling the strength of these kinetic effects is the formulation of the supersaturation function S. In a first step, we have chosen an identical formulation of S as for the previous ECHAM4-wiso model release: S = 1. − 0.002·Tcond (with Tcond as the condensation temperature, given in °C). This formulation resulted in fairly correct maximum absolute dexp values of approximately +20‰ in a previous ECHAM4 T31L19 simulation [Werner et al., 2001] and the same is true for the ECHAM5-wiso T31L19 simulation. However, the simulated dexp-δDp relation in this T31L19 simulation strongly deviates from the observed relation (Figure 7a). Furthermore, if the same supersaturation function S is applied for the fine T159L31 resolution, erroneously high dexp values of up to +40‰ are simulated for Antarctica (Figure 7b). Thus we applied in ECHAM5-wiso a modified supersaturation function S = 1.01 − 0.0045·Tcond. This formulation yields for both the T31L19 and T159L31 simulations a dexp-δDp relation, which is in good agreement with the observations. It is noteworthy that our new formulation of S is very close, but not identical to the formulation S = 1. − 0.004·Tcond used in the latest isotope version of the LDMZ4 [Risi et al., 2010], CAM2 [Lee et al., 2007], and GISS ModelE [Schmidt et al., 2005] as well as the formulation S = 1. − 0.005·Tcond used for HadCM3 [Tindall et al., 2009].

Figure 7.

Comparison of the deuterium excess dexp-δDp relation for Antarctica as derived from the compilation of observational data by Masson-Delmotte et al. [2008] (black crosses) and two ECHAM5-wiso simulations with different formulations of the supersaturation function S (blue, S = 1. − 0.002·Tcond; red, S = 1.01 − 0.0045·Tcond, with Tcond as the condensation temperature during ice crystal formation). Figure 7a shows simulation results for the coarse T31L19 model resolution. Figure 7b shows simulation results for the fine T1591L31 model resolution. All three relations are shown only for those locations of the Masson-Delmotte et al. [2008] database that contain values of all four analyzed variables (temperature, accumulation, δDp, dexp) at the same location.

4.3. Isotopic Composition of Atmospheric Vapor

[30] To evaluate the simulated isotopic processes during evaporation, we compare the modeled composition of atmospheric vapor to available near-surface vapor measurements of 5 GNIP stations. For the sake of clarity, we restrict our analyses to the medium-resolution T63L31 simulation results. The comparison of the simulated δDv values with the GNIP data shows a general reasonably accurate agreement between ECHAM5-wiso results and the observational data (Figure 8). For Vienna, Ankara, and Belem, ECHAM5-wiso simulates the seasonality of δDv in a correct manner but slightly overestimates and underestimates some individual monthly values. However, the δDv GNIP data series are very short and contain for any month δDv measurement of 1–3 consecutive years only. The measured year-to-year variability in δDv is rather high and often on the same order of magnitude as the difference between ECHAM5-wiso and GNIP δDv values.

Figure 8.

Seasonal cycles of the isotopic composition of vapor δDv at the locations of (a) Vienna, (b) Ankara, (c) Belem, (d) Manaus, and (e) Rehovot. The bold red line represents the observational GNIP values, while the bold black line indicates ECHAM5-wiso T63L31 simulation results. For all 5 GNIP stations, at least 1 complete year of monthly measurements of δDv has been recorded. When multiyear monthly measurements for a specific GNIP station and month exist, the observed interannual variability (1σ standard deviation) is indicated by orange shading. Thin black lines represent the corresponding interannual variability (1σ) of the ECHAM5-wiso T63L31 simulation.

[31] For Manaus, Brazil, the simulated δDv values are in general too high by approximately 10‰–20‰, except for January and November. On the opposite, the modeled δDv values near Rehovot, Israel, are in general too low by approximately 5–10‰. This GNIP station is located near the Mediterranean Sea, where local evaporation effects can have a strong influence on the isotopic composition of different water reservoirs [Gat et al., 2003]. These effects might not be sufficiently well captured in the T63L31 simulation.

[32] Next, we analyze the simulated latitudinal variations of δDv between 60°S and 60°N (Figure 9), as obtained from the SCIAMACHY data set. We restrict the analysis to this region, as higher-latitude areas have not been well covered by SCIAMACHY. The latitudinal mean SCIAMACHY δDv values vary between −90‰ and −210‰ with the least depleted values near the equator and the most depleted values in both high-latitude regions (Figure 9a). A very similar pattern is found in the ECHAM5-wiso T63L31 simulation, there exists, however, a clear offset in the range of +20‰ to +50‰ for the modeled δDv values. Such systematic offset is detected neither in the latitudinal variations of the simulated δDp values in precipitation, as compared to the available GNIP δDp data set, nor in the analyzed near-surface δDv values of the 5 GNIP stations (Figure 9b).

Figure 9.

(a) Latitudinal mean values of mass weighted, column averaged δDv values of atmospheric water vapor as detected by the Scanning Imaging Absorption Spectrometer for Atmospheric Cartography (SCIAMACHY) instrument on board the Envisat satellite for the period 2003–2005 [Frankenberg et al., 2009] and interpolated to ECHAM5 T63 grid resolution. The absolute δDv values given by Frankenberg et al. [2009] (crosses), the SCIAMACHY data shifted by +20‰ (asterisks; see sections 4.3 and 4.4 for details), and the corresponding ECHAM5-wiso simulation values (solid red circles, T63L31 resolution) are plotted. (b) Latitudinal distribution of annual mean GNIP δDv (crosses, 5 GNIP stations; cf. Figure 8) and δDp (diamonds, 70 GNIP stations with at least 5 complete calendar years of monthly isotope data; cf. Figure 1a) and corresponding ECHAM5-wiso simulation values (δDv, solid red circles; δDp, solid blue circles).

[33] A qualitative comparison of the geographical variations of SCIAMACHY and ECHAM5-wiso δDv values reveals that typical depletion pattern seen in the isotopic composition of precipitation (Figure 1) can be detected in both SCIAMACHY and ECHAM5-wiso mean δDv values of the total vapor column, too (Figure 10). Examples include a continental depletion effect of δDv over Eurasia and North America, and less depleted δDv values over (sub)tropical regions of South America and Africa. However, as for the latitudinal means, absolute ECHAM5-wiso δDv values are substantially less depleted than the SCIAMACHY δDv values in many terrestrial regions, with the largest model-data deviations in high-latitude areas, even if we account for a potential bias in the SCIAMACHY δDv values (see next paragraph) of +20‰.

Figure 10.

Mass weighted, column averaged HDO values of atmospheric water vapor. (a) SCIAMACHY δDv data for the period 2003–2005 [Frankenberg et al., 2009], shifted by +20‰ (see section 4.4 for details) and interpolated to ECHAM5 T63 grid resolution. (b) ECHAM5-wiso simulation values (model resolution: T63L31). For the calculation of the long-term mean ECHAM5-wiso δDv values, the unequal representation of seasons at different locations in the SCIAMACHY data set has been taken into account (see section 3.2 for details).

4.4. Difference Between ECHAM5-wiso and SCIAMACHY δDv Values

[34] The systematic deviations between the ECHAM5-wiso and SCIAMACHY δDv values are unique and cannot be detected in any other data-model comparison of this study. At this point we can only speculate about possible reasons for these obvious differences.

[35] According to C. Frankenberg (personal communication, 2010), potential errors in the satellite retrieval algorithms might lead to a general bias of absolute SCIAMACHY δDv values up to 20‰. Applying a systematic shift to the SCIAMACHY δDv values of +20‰ results in a much better agreement between the SCIAMACHY and ECHAM5-wiso latitudinal mean δDv values in equatorial regions. However, for all other regions the deviations between model results and satellite-based δDv values remain in the range of 0‰ to +30‰ (Figure 9a). Deviations are in general larger in the Southern Hemisphere than in the Northern Hemisphere, with the strongest model bias between 20°S–50°S. Furthermore, first analyses on the seasonal differences (winter-summer) of SCIAMACHY δDv values have not revealed any clear indication that the whole data set contains a systematic bias in its HDO values [Carlsen, 2010].

[36] Although all presented comparisons to the SCIAMACHY data are based on one T631L31 simulation only, we rule out the possibility that the δDv differences are related to the chosen model resolution, as an analogous analysis of the T159L31 simulation gives similar results (not shown).

[37] One possible explanation for the differences might be the neglect of cloud-free versus cloudy days in a specific month of the ECHAM5-wiso simulation. While all SCIAMACHY δDv measurements are based on a cloud-free sky, the analyzed monthly mean ECHAM5-wiso δDv values are based both on cloud-free and cloudy times during a specific month. However, a first additional T63L31 sensitivity experiment over 3 simulation years, where we deliberately used only cloud-free days for the calculation of δDv, did not result in a better agreement between SCIAMACHY and ECHAM5-wiso values.

[38] Another possible reason could be related to the formulation of fractionation effects during water evapotranspiration processes from land surfaces. As explained in section 2.1, no isotope fractionation occurs during evaporation of water masses directly from bare land surfaces in ECHAM5-wiso. Such simplified evaporation isotope flux approach might affect the isotopic composition of vapor over land stronger than the related isotopic composition of precipitation, as in ECHAM5 on a global scale approximately 40% of the water raining out over the continents stems from water evaporated over ocean surfaces [Hagemann et al., 2006]. However, the reasonable agreement between ECHAM5-wiso and the limited GNIP vapor measurements contradicts this hypothesis. We are currently implementing stable water isotopes in the more sophisticated land surface and vegetation scheme JSBACH (Jena Scheme for Biosphere-Atmosphere Coupling in Hamburg [Raddatz et al., 2007]) that can be coupled to ECHAM5. This will allow us to test in more detail whether the simplified formulation of fractionation processes during evaporation and transpiration of water from land surfaces does indeed lead to the detected model deviations of δDv.

[39] Another limitation of the presented simulations is the underestimation of the full range of δDv variability due to the fixed underlying SST and isotope boundary conditions. Either a nudged integration or a coupled atmosphere-ocean-sea ice model with interactive isotope cycle could cover the full range of variability and lead to a proper estimate of the sampling error in the data. A logical next step would be to perform such further simulation studies as well as further model-data comparisons to other isotopic data sets of water vapor, e.g., from the Tropospheric Emission Spectrometer (TES) [Worden et al., 2006], to explain the reasons of the δDv differences between ECHAM5-wiso and SCIAMACHY. In addition, a simulator of the SCIAMACHY retrieval algorithm might be added to a future version of the ECHAM5-wiso model enabling a comparison of the exact same diagnoses from the model outputs and from the satellite observations.

5. Conclusions

[40] As a summary of this study, we conclude that both stable water isotopes H218O and HDO have been successfully implemented into the hydrological cycle of the ECHAM5 atmospheric GCM. Major changes to ECHAM5 [Roeckner et al., 2003] and therefore also to the implementation of stable water isotopes include a flux-form semi-Lagrangian transport scheme for positive definite variables like water components and chemical tracers, separate prognostic equations for cloud liquid water and cloud ice, a prognostic-statistical cloud cover parameterization, and a new cloud microphysical scheme.

[41] For the evaluation of the first ECHAM5-wiso simulations, we used as many as possible high-quality data from various GNIP stations, a compilation of Antarctic isotope data, and the SCIAMACHY data for the isotopic composition of the total atmospheric water column. We find that on a global scale the ECHAM5-wiso results appear robust for different horizontal and vertical resolution of the model when compared to available observational data of the isotopic composition of precipitation from the GNIP database [IAEA/WMO, 2006]. However, the simulation of water isotopes in precipitation does clearly improve for an increased horizontal and vertical model resolution. This improvement has been demonstrated for the geographical distribution of H218O in precipitation over Europe, and by analyzing the simulated seasonal cycle of the isotopic composition in precipitation at 5 selected GNIP stations.

[42] For Antarctica, the simulation of HDO in precipitation is in good agreement with observational records [Masson-Delmotte et al., 2008], especially when a finer spatial model resolution is applied. For the T159L31 simulation, the mean model-data HDO difference (RMSE) is less then 12%. The same is true for the simulated absolute values of surface temperature and accumulation. In addition, the T159L31 ECHAM5-wiso simulation results have been used for a retuning of the strength of kinetic fractionation effects occurring at low temperatures during ice crystal formation [Jouzel and Merlivat, 1984]. The new formulation of the required supersaturation function results in an improved simulation of the deuterium excess dexp values in Antarctic precipitation. Nonetheless, even for the fine T159L31 ECHAM5-wiso simulation, a warm bias is detected over Antarctica, which is typical for many GCMs [Masson-Delmotte et al., 2006] and consequently leads to less depleted isotope values, as compared to observational data.

[43] The simulated near-surface isotopic composition of atmospheric water vapor δDv is also in fairly good agreement with recent observations from 5 different GNIP stations. When we compare the ECHAM5-wiso results to recently published mass weighted total column averaged HDO values of atmospheric water vapor from the SCIAMACHY instrument on board the Envisat satellite [Frankenberg et al., 2009], however, there exists a clear offset in the range of +20‰ to +50‰ for the modeled δDv values. SCIAMACHY measurements are likely substantially better suited for a water isotope-related model-data comparison than other earlier satellite-derived data sets [e.g., Worden et al., 2007], as the SCIAMACHY measurements cover a full 3 year period (2003–2005) as well as the total atmospheric water column. However, at this point we can only speculate about the reason for this obvious mismatch. Since the detected offset can neither be seen in the latitudinal variations of the simulated and available GNIP data, we assume that this offset is most likely not related to sampling errors in the data sets or underestimated natural interannual-to-decadal variability in our simulations. Further ECHAM5-wiso simulations including the full range of interannual-to-decadal variability and further satellite data analyses covering longer time intervals are required to resolve this issue.


[44] P.M.L. is grateful for funding received from the Deutsche Forschungsgemeinschaft (DFG) within the framework of the DAPHNE Forschergruppe (“Dated speleothems archives of the palaeoenvironment,” DFG Forschergruppe 668). M.H. acknowledges support from the Center for Marine Environmental Sciences (MARUM), which is also funded by the DFG. The constructive comments and suggested changes from three anonymous reviewers helped us to improve our manuscript and are greatly appreciated.