Variations of oxygen-18 in West Siberian precipitation during the last 50 years

Abstract. Global warming is associated with large increases in surface air temperature in Siberia. Here, we apply the isotope-enabled atmospheric general circulation model ECHAM5-wiso to explore the potential of water isotope measurements at a recently opened monitoring station in Kourovka (57.04° N, 59.55° E) in order to successfully trace climate change in western Siberia. Our model is constrained to atmospheric reanalysis fields for the period 1957–2013 to facilitate the comparison with observations of δD in total column water vapour from the GOSAT satellite, and with precipitation δ18O measurements from 15 Russian stations of the Global Network of Isotopes in Precipitation. The model captures the observed Russian climate within reasonable error margins, and displays the observed isotopic gradients associated with increasing continentality and decreasing meridional temperatures. The model also reproduces the observed seasonal cycle of δ18O, which parallels the seasonal cycle of temperature and ranges from −25 ‰ in winter to −5 ‰ in summer. Investigating West Siberian climate and precipitation δ18O variability during the last 50 years, we find long-term increasing trends in temperature and δ18O, while precipitation trends are uncertain. During the last 50 years, winter temperatures have increased by 1.7 °C. The simulated long-term increase of precipitation δ18O is at the detection limit (

Abstract. Global warming is associated with large increases in surface air temperature in Siberia. Here, we apply the isotope-enabled atmospheric general circulation model ECHAM5-wiso to explore the potential of water isotope measurements at a recently opened monitoring station in Kourovka (57.04 • N,59.55 • E) in order to successfully trace climate change in western Siberia. Our model is constrained to atmospheric reanalysis fields for the period 1957-2013 to facilitate the comparison with observations of δD in total column water vapour from the GOSAT satellite, and with precipitation δ 18 O measurements from 15 Russian stations of the Global Network of Isotopes in Precipitation. The model captures the observed Russian climate within reasonable error margins, and displays the observed isotopic gradients associated with increasing continentality and decreasing meridional temperatures. The model also reproduces the observed seasonal cycle of δ 18 O, which parallels the seasonal cycle of temperature and ranges from −25 ‰ in winter to −5 ‰ in summer. Investigating West Siberian climate and precipitation δ 18 O variability during the last 50 years, we find long-term increasing trends in temperature and δ 18 O, while precipitation trends are uncertain. During the last 50 years, winter temperatures have increased by 1.7 • C. The simulated long-term increase of precipitation δ 18 O is at the detection limit (< 1 ‰ per 50 years) but significant. West Siberian climate is characterized by strong interannual variability, which in winter is strongly related to the North Atlantic Oscil-lation. In winter, regional temperature is the predominant factor controlling δ 18 O variations on interannual to decadal timescales with a slope of about 0.5 ‰ • C −1 . In summer, the interannual variability of δ 18 O can be attributed to shortterm, regional-scale processes such as evaporation and convective precipitation. This finding suggests that precipitation δ 18 O has the potential to reveal hydrometeorological regime shifts in western Siberia which are otherwise difficult to identify. Focusing on Kourovka, the simulated evolution of temperature, δ 18 O and, to a smaller extent, precipitation during the last 50 years is synchronous with model results averaged over all of western Siberia, suggesting that this site will be representative to monitor future isotopic changes in the entire region.

Introduction
For the last several decades, an unequivocal warming of the climate system has been reported, evident from observations of increasing global average air and ocean temperatures, widespread melting of snow and ice and rising globalmean sea level (IPCC, 2013). However, while the rate of global warming averaged over the last 50 years amounts to about 0.1 • C per decade, high-latitude regions of the Northern Hemisphere, such as Siberia, have been warming at considerably higher rates (e.g. Tingley and Huybers, 2013, and Published by Copernicus Publications on behalf of the European Geosciences Union. references therein). Positive feedbacks associated with snow and sea-ice albedo, water vapour, clouds, and moisture transport as well as complex land surface-atmosphere interactions have been discussed as possible reasons for the observed Arctic amplification (for an overview see Masson-Delmotte et al. (2013), and references therein). While most studies so far have been focussed on the observed present and projected future temperature increase, it is uncertain how much other components of the Arctic climate system (like the hydrological cycle) will change as a consequence of the temperature rise. Bengtsson et al. (2011) estimated that the strength of the Arctic water cycle, in terms of annual precipitation, may increase by some 25 % by the end of the 21st century.
Since the pioneering work of Dansgaard (1953Dansgaard ( , 1964, Craig (1961), Merlivat et al. (1973), Sonntag et al. (1976), and others, it is well known that changes in climate and the atmospheric water cycle leave an imprint on the isotopic composition of different water reservoirs on earth. For meteoric water, Dansgaard (1964) successfully explained (through the atmospheric distillation process) the linear relation between the isotopic composition of precipitation and the local temperature at the precipitation site (the so-called "temperature effect") observed in many mid-to high latitude regions on earth. Assuming that the observed spatial and temporal isotope-temperature relationships are equivalent and constant in time (see Jouzel et al., 2000, for a review of this assumption), it is possible to infer past temperatures from stable water isotopes. Given the magnitude of Arctic warming over the past decades, climate change should be recorded in the isotopic composition of meteoric waters (or natural archives) in boreal regions of the Northern Hemisphere. The magnitude of this isotopic response is of interest when it comes to reconstructing past regional climate changes via isotopic data retrieved from various palaeoclimate archives (e.g. Sidorova et al., 2010).
Unfortunately, isotope records of present-day boreal precipitation are sparse and discontinuous. For Russia, Kurita et al. (2004) have reviewed the modern isotope climatology. Data from 13 Russian monitoring sites sampled during the period 1996-2000 depict eastward isotopic depletion over Russia, which is explained by the gradual rain-out of moist, oceanic air masses, which are transported towards and over Russia by westerly winds. This isotopic gradient, established from earlier isotope records, is known as the "continental effect" (e.g. Araguas-Araguas et al. (2000), and further references therein). The continental effect weakens in summer due to continental moisture recycling. Altogether, Kurita et al. (2004) estimate that 55 % of the summertime isotopic variability in Russian precipitation is linked to temperature changes and variations of the recycling ratio of continental water sources, the latter effect accounts for just 20 % of the signal. Combining simulation results of the isotope-enabled atmospheric general circulation model (AGCM) LMDZiso with satellite-based estimates of the isotopic composition of water vapour, Risi et al. (2013) found that intraseasonal vari-ations in continental recycling are minor contributions to the isotopic variability of high-latitude summer precipitation.
To study the impact of climate change in western Siberia, the Russian "mega-grant" research project "Impact of climate change on water and carbon cycles of western Siberia" (WSibIso, http://wsibiso.ru) has recently started monitoring the isotopic composition of water vapour and precipitation at two high-latitude sites in western Siberia. At Kourovka Observatory (57.04 • N, 59.55 • E), located approximately 80km west of Yekaterinburg, isotope monitoring started in 2012 , while regular isotope measurements at Labytnangi (66.65 • N, 66.40 • E) began in summer 2013.
Within the WSibIso Project, the understanding of the signals recorded at Kourovka and Labytnangi is supported by state-of-the-art climate simulations with two AGCMs equipped with explicit stable water isotope diagnostics, ECHAM5-wiso  and LMDZiso (Risi et al., 2010a). Such isotope-enabled AGCMs provide a mechanistic understanding of the atmospheric processes influencing the isotopic composition of meteoric water. Since the pioneering work of Joussaume et al. (1984), Jouzel et al. (1987), Hoffmann et al. (1998 and others, about a dozen stateof-the-art GCMs have been equipped with explicit isotope diagnostics (see Sturm et al., 2010, for a detailed model overview). A number of studies have clearly demonstrated their usefulness for an improved climatic interpretation of present and past water isotope variability (e.g. Jouzel et al., 2000;Mathieu et al., 2002;Noone and Simmonds, 2002;Werner and Heimann, 2002;Vuille and Werner, 2005;Lee and Fung, 2008;Tindall et al., 2009;Risi et al., 2010b). A comparison of different models allows for the evaluation of robust features, and the ability to scrutinize each model's parametrizations.
For Kourovka Observatory, the observed variations of the surface vapour isotopic composition are very similar to the results of the ECHAM5-wiso simulation covering the period April to September 2012 . Both exhibit short-term fluctuations on timescales from a few hours to a few days. These variations can be attributed to the passage of synoptic-scale weather systems, advecting air from different source regions with different isotopic signatures to Kourovka . A detailed comparison of the Kourovka data with LMDZiso model results will be presented in an accompanying paper (Gryazin et al., 2014).
Here, we extend the isotope analyses from year 2012 to the last 50 years. As there are no Russian water vapour isotope measurements available prior to 2012, the extended time frame implies that we focus on the isotopic composition of precipitation. Our key questions are as follows. (1) How much has the isotopic composition of precipitation varied in western Siberia over the last five decades? (2) What are the main mechanisms and processes causing the variations? (3) How well can large-scale West Siberian climate and water cycle variations be observed in the isotopic composition of precipitation at Kourovka Observatory, one of the Atmos. Chem. Phys., 14, 5853-5869, 2014 www.atmos-chem-phys.net/14/5853/2014/ key monitoring sites within the WSibIso project? Our analysis is based on a so-called "nudged" ECHAM5-wiso climate simulation performed for the period 1957-2013, covering the entire period of available ECMWF reanalysis data (Uppala et al., 2005;Berrisford et al., 2011;Dee et al., 2011). The paper is arranged as follows: after a description of the model setup we test the model performance with respect to various observational data sets. In particular, we present a thorough comparison of simulated and observed precipitation δ 18 O in Russia, going beyond the previous global model assessment by Werner et al. (2011). This validation is a prerequisite for the following discussion of isotopic interannual variability and mechanisms during the last decades and of the representativeness of Kourovka in this context. We finish with conclusions regarding the potential of isotope measurements at Kourovka to trace future climate changes in western Siberia.

Model description
Atmospheric simulations were carried out using ECHAM5wiso , which is the isotope-enabled version of the atmospheric general circulation model ECHAM5 (Roeckner et al., 2003;Hagemann et al., 2006;Roeckner et al., 2006). The model considers both stable water isotopes H 18 2 O and HDO, which have been explicitly implemented into its hydrological cycle, analogous to the isotope modelling approach used in the previous model versions ECHAM3 (Hoffmann et al., 1998) and ECHAM4 (e.g. Werner et al., 2001). For each phase of "normal" water (vapour, cloud liquid, cloud ice) being transported independently in ECHAM5, a corresponding isotopic counterpart is implemented in the model code. Isotopes and "normal" water are described identically in the AGCM as long as no phase transitions take place. Therefore, the transport scheme for all water-related variables is the flux-form semi-Lagrangian transport scheme for positive definite variables implemented in ECHAM5 (Lin and Rood, 1996).
Additional fractionation processes are defined for the water isotope variables whenever a phase change of the "normal" water occurs in ECHAM5 (considering equilibrium and non-equilibrium fractionation processes). Equilibrium fractionation takes place if the corresponding phase change is slow enough to allow full isotopic equilibrium (Merlivat and Jouzel, 1979). On the other hand, non-equilibrium processes even depend on the velocity of the phase change, and therefore on the molecular diffusivity of the water isotopes (Jouzel and Merlivat, 1984). Processes involving isotopic fractionation include the evaporation from the ocean, condensation either to liquid or to ice, as well as re-evaporation of liquid precipitation within the atmosphere. For evapotranspiration from land surfaces, possible isotopic fractionation is ne-glected (see Hoffmann et al. (1998) and Haese et al. (2013), for a detailed discussion of this issue).
ECHAM5-wiso has been evaluated against observations of isotope concentrations in precipitation and water vapour, both on a global and on a European scale Werner et al., 2011). On both scales, annual and seasonal ECHAM-5-wiso simulation results are in good agreement with available observations from the Global Network of Isotopes in Precipitation, GNIP (IAEA/WMO, 2013). Werner et al. (2011) have shown that the simulation of water isotopes in precipitation clearly improves with increased horizontal and vertical model resolution. Thus, for this study, we choose a horizontal model resolution of T63 in spectral space (horizontal grid size of approximately 1.9 • × 1.9 • ), and a vertical resolution of 31 levels on hybrid sigmapressure coordinates. Local ECHAM5-wiso results for GNIP stations (discussed further below) were obtained by bilinear interpolation to the station coordinates. To ensure a most realistic simulation of present-day climate variability, the model is forced with prescribed yearly values of present-day insolation and greenhouse gas concentrations (IPCC, 2000), as well as with monthly varying fields of sea-surface temperatures and sea-ice concentrations according to ERA-40 and ERA-Interim reanalysis data (Uppala et al., 2005;Berrisford et al., 2011;Dee et al., 2011). Furthermore, the dynamicthermodynamic state of the ECHAM model is constrained to reanalysis data by an implicit nudging technique (Krishamurti et al., 1991; the implementation in ECHAM is described by Rast et al., 2013) -i.e. modelled fields of surface pressure, temperature, divergence and vorticity are relaxed to the corresponding ERA-40 and ERA-Interim reanalysis fields (Uppala et al., 2005;Berrisford et al., 2011;Dee et al., 2011). The nudging interval is 6 hours, ensuring that the simulated large-scale atmospheric flow is modelled in agreement with the ECMWF reanalysis data on all analysed timescales. In contrast to the atmospheric flow, the hydrological cycle and its isotopic variations is still fully prognostic and not nudged to any reanalysis data. Vegetation in the model is prescribed by a time-invariant set of land surface data (vegetation ratio, leaf area index, forest ratio, background albedo; Hagemann, 2002).
The performed simulation covers the period September 1957 until July 2013. Here we regard the first 28 months as model spin-up and analyse the 51-year period between 1960 and 2010. Unless stated otherwise, we focus on monthly averaged model results of the isotopic composition of precipitation (typically expressed in a delta-notation as δ 18 O or δD), covering the full period of available stable water isotope measurements in Russia.

Observations of isotopes in Russian precipitation
In western Siberia (here defined as the region ranging from 55-90 • E and 55-70 • N), monthly precipitation δ 18 O data are available from 9 GNIP stations operating for different time Table 1. GNIP δ 18 O records from Russia considered in this study. Stations located in western Siberia are highlighted in bold. For each station, we report the latitude, longitude and altitude, as well as the sampling period (t obs ) and the number of available monthly measurements (N obs ).

Name
Lat (

Satellite observations of isotopes in water vapour
Although precipitation and water vapour have a different isotopic composition due to fractionation processes, a comparison of ECHAM5-wiso results for isotopes in water vapour with available data from satellite and ground-based remote sensing techniques is valuable. It will reveal some first-order information if the model correctly simulates the spatial gradients of the isotopic signal in atmospheric water vapour (and thus consistently in precipitation) over Russia. Ground-based remote sensing of δ 18 O in water vapour has been realized recently  but so far, only a small number of measurements have been carried out. For this reason we consider global observations of deuterium (δD) in total column water vapour retrieved from the GOSAT satellite (Boesch et al., 2013;Frankenberg et al., 2013) for the period April 2009 to June 2011. We select only measurements which pass a series of quality criteria involving the absence of clouds and the retrieval precision for different species Risi et al., 2013).
To rigorously compare ECHAM5-wiso with GOSAT, the model output needs to be processed in two ways. First, we need to take into account the spatio-temporal sampling. GOSAT makes measurements only along its orbit, and not all measurements pass our quality selection. Therefore, we select only the locations and days for which GOSAT has made valid measurements. Such a selection makes sense because ECHAM5-wiso is nudged toward reanalysis, so that the atmospheric properties simulated by ECHAM5-wiso and observed by GOSAT can be compared at the daily scale.
Secondly, we need to take into account the instrument sensitivity. The column-integrated δD value retrieved by GOSAT is not exactly the column-integrated δD value which actually occurred. This is because retrievals are affected by Atmos. Chem. Phys., 14, 5853-5869, 2014 www.atmos-chem-phys.net/14/5853/2014/ atmospheric conditions such as the vertical profiles of temperature and water vapour or the presence of clouds. The "averaging kernels" describe how a given vertical profile in δD, for given atmospheric conditions, translates into the GOSAT column-integrated δD retrieval values (Rodgers and Connors, 2003). An averaging kernel is produced for each GOSAT measurement. Therefore, for each GOSAT measurement, we apply the corresponding averaging kernel to the model outputs. This allows us to compute the columnintegrated δD values that GOSAT would retrieve if it was flying in ECHAM5-wiso's simulated atmosphere (Risi et al., 2012a).

Present-day mean climate
Gribanov et al. (2014)  . Simulated surface temperatures show a small cold bias of less than 1 • C in the annual mean, with larger deviations (of up to −3 • C) in winter. Precipitation rates simulated by ECHAM are slightly above observed values in the annual mean (by about 2 mm month −1 ) and between October and June (by 2-8 mm month −1 ), but up to −16 mm month −1 lower between July and September. Figure 1 shows annual-mean patterns of simulated surface temperature (T ), total precipitation amount (P), and oxygen-18 content of precipitation (δ 18 O) for the region 0-160 • E, 40-80 • N, covering central and eastern Europe, Russia and parts of Asia.
Surface temperatures (Fig. 1a) decrease from southwest to northeast. Comparing model temperatures averaged over 1960-2010 with the CRU temperature reconstruction for the same period (University of East Anglia Climatic Research Unit (CRU), Jones and Harris, 2013), we find slightly colder values in the simulations than observed (in the range −0.5-−1.5 • C) in Eurasia and western Siberia. Conversely, simulated temperatures in central and eastern Siberia are higher by ∼ 0.5-1.5 • C than the CRU data, with maximum differences of up to 5 • C found for the Verkhoyansk Range.
Precipitation fields (Fig. 1b) are zonally aligned in the simulated annual mean pattern. Total precipitation decreases from 60 to 80 mm month −1 at the East European Plain east-wards and arrives at minimum values of 20-30 mm month −1 in a zone ranging from the southern part of the West Siberian Plain to northeast Siberia. Precipitation values peak in the Russian Far East in the Stanovoy Range. Compared to the CRU precipitation reconstruction (University of East Anglia Climatic Research Unit (CRU), Jones and Harris, 2013), simulated precipitation is higher by 10-20 mm month −1 north of 60 • N, and by up to about 30 mm month −1 in the Russian Far East mountain areas. In the southern part of the West Siberian Plain, ECHAM produces slightly less precipitation than observed. The precipitation deficit increases with increasing continentality.
The simulated annual mean δ 18 O values of precipitation over Russia are plotted in Fig. 1c. Within Russia, δ 18 O values decrease from southwest to northeast. In contrast to temperature and precipitation, no global data set of observed δ 18 O in precipitation yet exists for comparison. The geographical distribution of GNIP stations shown in Fig. 1c illustrates the large observational gaps in Russia. A more rigorous modeldata comparison of δ 18 O will be presented further below.
Total column water vapour values of δD, according to ECHAM5-wiso and to GOSAT satellite retrievals, are shown in Fig. 2. As there is no absolute calibration for columnintegrated δD of GOSAT , we subtract the global average of δD for both GOSAT and ECHAM to enable an improved comparison focussing on the spatial distributions. The model captures the pattern of total column water vapour δD variations over Russia well ( Fig. 2a  and b). Even some details such as the regional δD gradient southwest of Kourovka Observatory are resolved. Regarding the meridional δD gradient along the longitude zone including Kourovka, ECHAM5-wiso captures the northward depletion retrieved by GOSAT (Fig. 2c). Considering the zonal δD variation along the latitude zone including Kourovka (Fig. 2d), we find that ECHAM5-wiso tends to underestimate the eastward depletion associated with the continental effect. From 20 • E to 120 • E, δD decreases by about 80 ‰ in GOSAT observations and by only about 40 ‰ in ECHAM5wiso.
For more quantitative analyses, we compare the climatology from our ECHAM5-wiso simulation results to available GNIP measurements. For each GNIP location, we restrict the data-model comparison to those months within the period 1960-2010, when measurements have been reported (see Table 1 for details). Thus, mean values of T , P and δ 18 O are calculated over different periods for each GNIP station and from ECHAM5-wiso. The results of this comparison are shown in Fig. 3. The uncertainty range indicated by the error bars is ± 2 standard errors of the estimated mean values. In the following, uncertainty ranges always refer to the 95 % confidence interval.
As expected from our nudging strategy, Fig. 3a shows good agreement between modelled surface temperatures and GNIP observations. Model results and observations are highly correlated (r 2 ∼ 0.95) and lie close to a line with slope  1. A linear fit, applying an algorithm which accounts for the uncertainties in both coordinates (Krystek and Anton, 2007), yields an optimum slope of 1.05 ± 0.23 and an optimum intercept of (−0.75 ± 1.45) • C. The root mean square (RMS) difference between modelled and observed mean temperatures is 1.13 • C. A comparison of the averages over all stations (ECHAM5-wiso: 2.59 • C, GNIP: 3.30 • C) suggests that the model tends to underestimate the observed temperatures. Regarding individual stations, the largest difference is found for Perm, where the model is too cold by 2.9 • C. However, given the uncertainty ranges of observed and simulated temperatures, the conclusion of an overall cold bias of ECHAM5-wiso is not robust. The average variability of modelled and observed temperatures (i.e. the average length of the error bars) is virtually the same (± 2.83 • C vs. ± 2.88 • C). This indicates that ECHAM5-wiso captures the temporal variability of observed surface temperatures in Russia (note that the range of model values only reflects temporal variability while the observational range may also include measurement errors).
For precipitation the scatter of model results is larger (Fig. 3b), resulting in a correlation which is weaker than for temperature (r 2 ∼ 0.73). The RMS difference between simulations and observations is about 10.5 mm month −1 , with the largest deviations for Murmansk (+24 mm month −1 ) and for Rostov (−19 mm month −1 ). The model results for these stations are clearly beyond the uncertainty range of a linear fit (slope = 1.25 ± 0.16 and intercept = (−6.68 ± 5.78) mm month −1 ). The average precipitation rate according to ECHAM5-wiso is 10 % higher than observed (48 mm month −1 vs. 44 mm month −1 ), which points to a moist bias of the model. On the other hand, the model Atmos. Chem. Phys., 14, 5853-5869 slightly underestimates the variability of observed precipitation rates (± 6.55 mm month −1 vs. ± 6.99 mm month −1 ). Figure 3c indicates that ECHAM5-wiso captures annual mean δ 18 O records in Russia reasonably well. Model results and GNIP data are highly correlated (r 2 ∼ 0.96) and coalesce along a line with a slope close to 1 (0.97 ± 0.11; intercept: (1.15 ± 1.37) ‰). The RMS difference between model results and observations amounts to 1.05 ‰. ECHAM5wiso tends to overestimate observed δ 18 O. The average over all stations is −12.1 ‰ vs. −13.6 ‰ according to GNIP. Maximum differences of up to +4 ‰ are found for two stations in eastern Siberia (Cherskiy and Yakutsk) where ECHAM5-wiso does not simulate sufficient depletion. Uncertainty ranges of the simulated and observed means are rather the same (± 1.17 ‰ simulated vs. ± 1.16 ‰ observed). The model slightly underestimates the spatial gradients of δ 18 O observed in Russia. The meridional isotope gradient (observed slope: −4.2 ‰ per 10 • latitude, simulated slope: −3.6 ‰ per 10 • latitude), reflecting the meridional temperature gradient, is by about a factor of four to five larger than the zonal isotope variation (observed slope: −0.9 ‰ per 10 • longitude, simulated slope: −0.8 ‰ per 10 • longitude) associated with increasing continentality (not shown).
Studying anomalies of δ 18 O ( δ 18 O) and surface temperatures ( T), we find a linear relationship for all seasons with a typical slope of 0.5 ‰ • C −1 (Fig. 4). The correlation between GNIP T and δ 18 O is most pronounced in au-tumn (SON, r 2 = 0.69) and winter (DJF, r 2 = 0.61). In spring (MAM, r 2 = 0.50) and summer (JJA, r 2 = 0.40) the correlation is weaker, but still significant (p 0.05 applying a t test). The weaker coupling indicates that the δ 18 O signal during the warm season is significantly affected by other processes such as moisture recycling. ECHAM5-wiso simulates a similar seasonal relationship but the correlation between δ 18 O and T is higher than for the observations (r 2 = 0.50-0.79). The model overestimates the coupling between δ 18 O and T especially in spring (r 2 = 0.79).
We now compare the simulated and observed seasonal cycle of precipitation δ 18 O in western Siberia (Fig. 5). The data exhibit seasonal variations ranging from −25 ‰ in winter to −5 ‰ in summer, closely following the seasonal cycle of temperature. Peak values of up to −1 ‰ were observed in Perm during the second half of the 1980s. Despite the reported small annual-mean temperature biases, ECHAM5wiso correctly simulates the timing and magnitude of the seasonal variations of both temperature and δ 18 O (Fig. 5) in western Siberia. Observations and simulations from Pechora and Perm also show interannual variations, which will be discussed in the next section.

Interannual to decadal variations over the last five decades
The sampling period of precipitation δ 18 O in Russia is too short for a thorough investigation of the long-term variability seen in the West Siberian isotope and climate records shown in Fig. 5. To overcome this problem, we extend the time frame by considering model results for the period 1960-2010. Therefore, the following analysis of interannual to decadal variations of T , P and δ 18 O over Russia during the past decades is entirely based on model results. Unless stated otherwise, the model results are presented as anomalies from their long-term climatological mean . A zerophase bidirectional low-pass filter with a length of 24 equally weighted months is employed on the ECHAM5-wiso results to highlight long-term variability. To explore the potential influence of global warming during the period 1960-2010, we apply a t test comparing the reference period with the period 1981-2010 (significance level is 5 %). In addition, we investigate linear long-term trends derived from a regression of annual-mean model results. The numerical results of this trend analysis are listed in Table 2.
At the global scale, (Fig. 6a, blue line), surface warming has been accompanied by increasing atmospheric moisture content (not shown) while modelled precipitation over land has slightly decreased (Fig. 6b). In parallel, global water vapour and precipitation have become progressively enriched with heavy water isotopes (Fig. 6c). The simulated increase of land surface temperatures is statistically significant. The trend analysis indicates a long-term increase of annualmean global land surface temperatures by (1.23 ± 0.01) • C per 50 years which is in the range of trend estimates based on observations (for a compilation of climatological observations see Hartmann et al., 2013, and references therein). Simulated anomalies of global land precipitation peaked in the mid-1970s which is also seen in global precipitation data sets. For more recent periods, the simulation does not reproduce the observed amplitude of interannual precipitation variability. However, the changes are statistically significant, and the modelled long-term decrease of global land precipitation by (2.18 ± 0.49) mm month −1 per 50 years is in line with observations (Hartmann et al., 2013; note that global precipitation trends there relate to the different period 1951-2008).
Long-term trends in T , P, and δ 18 O are also found in western Siberia (averaged over the area 55-90 • E, 55-70 • N) as well as at Kourovka (Fig. 6, red and green lines). Regional and local warming is statistically significant and occurred at lower rates than global ones in the annual mean ((1.19 ± 0.18) • C per 50 years averaged over western Siberia and (1.08 ± 0.15) • C per 50 years in Kourovka). The long-term warming is particularly pronounced in winter (DJF), especially in western Siberia and Kourovka, where DJF warming rates are in the range 1.5-1.7 • C per 50 years. Our model also suggests a positive long-term trend of annual precipitation. . Phys., 14, 5853-5869, 2014 www.atmos-chem-phys.net/14/5853/2014/  For both western Siberia and Kourovka, annual-mean precipitation rates have been increasing by 2-3 mm month −1 during the last 50 years, with a tendency towards enhanced DJF precipitation at the expense of JJA rainfall. However, except for winter precipitation in western Siberia, the changes are statistically insignificant, and the uncertainty range of the regional and local precipitation trends is high, exceeding the projected average long-term anomaly. Long-term trends of precipitation δ 18 O are also positive. The changes are small and at the detection limit (< 1 ‰ per 50 years) but statistically significant everywhere at the annual timescale. For western Siberia we find that the long-term changes of δ 18 O are more pronounced during JJA than during DJF, which is opposite to the simulated seasonal temperature trends. The reason for this decoupling is that moisture import to western Siberia intensifies more in summer than in winter (not shown), while the opposite is simulated for precipitation. As a consequence, the isotopic signature of moisture available for precipitation is less affected by continental depletion during recent summer seasons than during recent winters. At the regional and local scale the long-term trends are superimposed by strong interannual variability, reaching values  riod: 1961-1990). The ECHAM5-wiso results are averaged globally (blue line), for the region of western Siberia (green line), and interpolated to the location of Kourovka Observatory (red line). A zero-phase bidirectional low-pass filter with a length of 24 equally weighted months has been applied to the simulated monthly mean values for filtering short-term fluctuations. Straight lines are global (blue), regional (green) and local (red) trends obtained from leastsquare fits of annual-mean values for the period 1960-2010; dashed lines are 95 % confidence intervals for the trends. See also Table 2 for a summary of numerical results.

Atmos. Chem
Atmos. Chem. Phys., 14, 5853-5869, 2014 www.atmos-chem-phys.net/14/5853/2014/ of up to ± 1.5 • C, ± 10 mm yr −1 , and ± 1 ‰, respectively. Temperature anomalies simulated for western Siberia and Kourovka have covaried since the late 1960s. During most of the simulation period, temperature differences between western Siberia and Kourovka are less than about ± 0.5 • C, with Kourovka showing larger fluctuations than western Siberia. Simulated precipitation anomalies for western Siberia and Kourovka appear to be less synchronous than is the case for temperature, particularly between 1995 and 2000 when the precipitation curves are out of phase. Moreover, in Kourovka the precipitation variability is considerably larger (by up to 10 mm month −1 ) than its average for western Siberia. The larger deviation between the mean precipitation amount in West Siberia and the values in Kourovka as compared to the surface temperatures is not surprising, as precipitation is known to strongly vary at small spatial and temporal scales. Consistent with the temperature patterns, anomalies of δ 18 O in western Siberia and Kourovka are in phase most of the time and differ within ± 0.5 ‰, with the larger variability being simulated for Kourovka. In our simulation, δ 18 O and surface temperature mostly covary, which is not the case for the precipitation amount. However, δ 18 O and temperature are not rigidly coupled. This is indicated by our model results for the years around 1990, when temperatures in Kourovka were below the western Siberian average, while the opposite is obtained for δ 18 O. The overall good agreement between mean temperature and δ 18 O changes in West Siberia and Kourovka is a key finding with respect to the objectives of the WSibIso Project. It indicates that Kourovka Observatory is a highly representative site for monitoring climate change in western Siberia. While we find that annual and seasonal-mean values of T , P and δ 18 O have been slowly changing during the last decades (at least at the global scale), we do not arrive at a significant conclusion regarding potential changes in their interannual variance. Compared with the interannual variance between 1961 and 1990 (estimated from the standard deviation of detrended model results), the last decade (2001-2010) is characterized by increased variability of winter temperatures (0.4 • C) and winter precipitation (2 mm month −1 ) in western Siberia, and by increased variance of winter and summer precipitation rates (3 mm month −1 ) in Kourovka. The last decade does not exhibit substantial changes in δ 18 O variability.
Further analyses of the temporal correlation between simulated values of δ 18 O in precipitation and surface temperatures (Fig. 7) reveal that the correlation of annual-mean values seen in Fig. 7a is mainly controlled by a strong linkage between surface temperature and δ 18 O in precipitation in western Siberia during winter (DJF). While the correlation coefficient between winter T and δ 18 O can reach maximum values of up to 0.9 (Fig. 7b), the correlation between both climate variables is substantially weaker for summer (Fig. 7c). In the WSibIso target area, only one-quarter of the observed  interannual δ 18 O variability can be explained by a linear relationship with local surface air temperature changes. We now explore the relationship between West Siberian climate and precipitation isotopic composition, and largescale atmospheric circulation. Previous studies have revealed a strong linkage between surface temperatures, δ 18 O in precipitation and the North Atlantic Oscillation (NAO) for major parts of Europe (e.g. Baldini et al., 2008;Field, 2010;Langebroek et al., 2011;Casado, et al., 2013). It is also known that the influence of the NAO on the large-scale atmospheric circulation is not bound to Europe but extends further east towards Russia (e.g. Halpert and Bell, 1997  Year Seasonal NAO index DJF 19601965197019751980198519901995  simulated δ 18 O values in Kourovka with the simulated global sea level pressure field, we find a pattern for winter which is characteristic for the NAO (Fig. 8; cf. Hurrell and Deser, 2009). Thus, as a next step we investigate the influence of NAO variations on temperature and δ 18 O variability over Russia. Figure 9 shows the observed and simulated station-based NAO seasonal winter (DJF) index for the period 1960-2010. ECHAM5-wiso model faithfully captures the observed NAO index (reference stations are Ponta Delgada, Azores, and Stykkishólmur/Reykjavik, Iceland; Hurrell et al., 2013) consistent with the reanalysis pressure fields used for nudging. Minor deviations between the observed station-based index  (Fig. 10). To a large extent, the covariation between winter NAO and δ 18 O (Fig. 10a) is controlled by air temperature. Winters are mild in years when the NAO is strong, which is indicated by the positive correlation between NAO index and DJF surface temperatures shown in Fig. 10b. Winter precipitation in northern Russia also increases when the NAO is strong (Fig. 10c). Our analyses shown in Fig. 10 reveal that in wintertime the NAO-associated atmospheric circulation changes have a slightly weaker impact on precipitation δ 18 O (r ∼ 0.6) over western Siberia than on the surface temperatures in this region (r ∼ 0.7). On the contrary, δ 18 O in precipitation is much more strongly correlated to the NAO than the precipitation amount itself (r ∼ 0.3). This exposes the potential of reconstructing past changes of the NAO strength from various δ 18 O records, for example those retrieved from lake sediments, speleothems, or tree rings (e.g. Sidorova et al., 2010)   such a teleconnection for the summer. Previous studies have shown that evapotranspiration fluxes significantly contribute to summer precipitation in Russia, and estimated that the regional moisture recycling rates can exceed 80 % (e.g. Koster et al., 1993;Numaguti, 1999;Risi et al., 2013). Accordingly, Kurita et al. (2003Kurita et al. ( , 2004 have suggested that snowmelt and subsequent evaporation of soil moisture carrying the isotopic imprint of winter precipitation could significantly influence the isotopic composition of regional precipitation, counterbalancing the positive coupling between temperature and δ 18 O. Studying one-point correlation maps for JJA, we find that δ 18 O at Kourovka is negatively correlated with the regional soil moisture reservoir (Fig. 11a) and local evaporation (see Table 3) which is in line with moisture recycling through re-evaporation. Moreover, we identify a negative correlation between δ 18 O and total precipitation (Fig. 11b), which is mainly due to the variability of convective precipitation (Fig. 11c, see also Table 3). If we consider only years in which convective precipitation is below the arithmetic longterm average, the correlation between δ 18 O and local surface temperature rises from r ∼ 0.4 to r ∼ 0.6 (i.e. comparable to winter values, cf. Table 3). In the opposite case (convective precipitation above the long-term average) the correlation between δ 18 O and surface temperature decreases to r ∼ 0.3. This suggests that the isotope signal in West Siberian summer precipitation is rather controlled by the temperature variability at the level of condensation than by temperatures on the ground. In principle, a correlation of δ 18 O with vertical temperatures should permit us to identify the altitude range of precipitation formation. However, correlating monthly-mean values we did not arrive at conclusive results. Figure 11 clearly reveals that the summer δ 18 O signal at Kourovka reflects hydrometeorological changes at the regional scale. Regarding precipitation and the seasonal-mean moisture flow, the correlation pattern is asymmetric. Areas in which the correlations are statistically significant are rather small upwind of Kourovka, but more extended downwind. Downwind areas may contribute to the isotopic signal at Kourovka through mixing -i.e. in situations of transient perturbations of the mean atmospheric flow. In fact, according to Fukutomi et al. (2004), Kourovka lies in the Siberian summer storm-track zone with high synoptic-scale eddy activity. This is reflected in our model results by greater monthly variability of atmospheric moisture transport towards Kourovka during summer than during winter. In summer, monthly moisture fluxes vary within ∼ ± 35 % in magnitude and within ∼ ± 25 • in flow direction, while in winter monthly-mean fluxes vary within ∼ ± 10 % in magnitude and ∼ ± 15 • in flow direction, respectively. These findings, as well as the results of our correlation analyses, are probably overly smoothed to fully resolve effects of moisture recycling, atmospheric convection, and transient perturbations, as the effective timescales of these processes can be considerably shorter than a month.
The link between δ 18 O and climate during the West Siberian summer deserves further investigations. The effects of atmospheric convection and transient perturbations could be investigated by analysing the evolution of δ 18 O at the timescale of single meteorological events, which may also involve the use of second order isotopic data (d-excess or 17 O excess; e.g. Landais et al., 2010;Guan et al., 2013). Moisture recycling and the origin of advected moisture could be investigated by moisture tagging, that is by simulating the dispersal of numerical water tracers evaporating from different predefined source regions (e.g. Koster et al., 1986;Numaguti, 1999;Risi et al., 2013). Nonetheless, our findings for the summer are in line with previous studies arguing that variations of δ 18 O in precipitation are rather a regionally integrated signal of several climate variables than a proxy for either local temperature or precipitation changes (e.g. on a global scale: Schmidt et al., 2005; for Western Europe: Langebroek et al., 2011).

Summary and conclusions
Using the few available observations as well as a new simulation from the isotope-enabled atmospheric general circulation model ECHAM5-wiso covering the period 1958-2013, we have investigated the spatiotemporal variations in the isotopic composition of precipitation in Russia during recent decades. In its nudged configuration, the model simulates temperature and precipitation fields over western Siberia within reasonable error margins, providing a realistic framework for investigating the model performance for δ 18 O. The model reproduces the spatial pattern of precipitation δ 18 O when compared with averaged observations from 15 stations of the Global Network of Isotopes in Precipitation between 1970 and 2009. The model has difficulties capturing the amount of δ 18 O depletion in eastern Siberia, while temperature and precipitation are correctly simulated. According to our model results, temperature is the predominant factor, controlling up to 80 % of the variability of annual-mean and winter precipitation δ 18 O in Russia on interannual to decadal timescales. Interannual variations in winter temperature and isotope signals show a strong imprint of the North Atlantic Oscillation. During summer, local temperature has only a minor impact (about 20 %) on the variability of the isotopic composition of West Siberian precipitation. Instead, our analyses indicate that δ 18 O integrates effects of regional hydrometeorological processes on timescales shorter than a month which have not been explicitly considered in this study. The results are in line with moisture recycling through evaporation, involving the delayed reevaporation of isotopically depleted winter precipitation retained in snowmelt and soil water. The isotopic summer signal is significantly influenced by convective precipitation formation, which does not occur in this region in winter. We also find enhanced variability of moisture transports towards western Siberia. The relative importance of these processes should be further investigated with higher temporal resolution, or by using second order isotopic data (e.g. deuterium excess) as well as numerical moisture tagging diagnostics.
Our results indicate that δ 18 O has the potential to reveal hydrometeorological regime shifts in future summers, which are otherwise difficult to identify.
Recent observations reveal significant isotopic variability on the diurnal and daily timescale . The impact of short-term variations on the isotopic signal seen in the monthly GNIP records cannot be analysed, but continuous monitoring of water vapour δ 18 O and daily sampling of precipitation δ 18 O will permit the study of processes on the event scale. Regarding Kourovka Observatory, where such a monitoring programme has recently been established, we find that the simulated variability of temperature and δ 18 O at this location is similar to model results averaged over the entire West Siberian region. Therefore, we conclude that this location is highly suitable to monitor isotopic changes all over western Siberia.