Developing a western Siberia reference site for tropospheric water vapour isotopologue observations obtained by different techniques (in situ and remote sensing)

Water stable isotopologues provide integrated tracers of the atmospheric water cycle, affected by changes in air mass origin, non-convective and convective processes and continental recycling. Novel remote sensing and in situ measuring techniques have recently offered opportunities for monitoring atmospheric water vapour isotopic composition. Recently developed infrared laser spectrometers allow for continuous in situ measurements of surface water vapour δDv andδOv. So far, very few intercomparisons of measurements conducted using different techniques have been achieved at a given location, due to difficulties intrinsic to the comparison of integrated with local measurements. Nudged simulations conducted with high-resolution isotopically enabled general circulation models (GCMs) provide a consistent framework for comparison with the different types of observations. Here, we compare simulations conducted with the ECHAM5-wiso model with two types of water vapour isotopic data obtained during summer 2012 at the forest site of Kourovka, western Siberia: hourly ground-based FTIR total atmospheric columnar δDv amounts, and in situ hourly PicarroδDv measurements. There is an excellent correlation between observed and predicted δDv at surface while the comparison between water column values derived from the model compares well with FTIR estimates.


Introduction
Owing to slight differences in the saturation vapour pressure and diffusivity in air of H 16 2 O, HD 16 O and H 18 2 O molecules, fractionation processes occur during phase changes of the water.As a result, the distribution of the water isotopologues (hereafter δD and δ 18 O expressed in ‰ versus VS-MOW (Craig, 1961)) varies both spatially and temporally in Correspondence to: K. Gribanov (kgribanov@remotesensing.ru) the atmospheric water vapour and in the precipitation.Until recently, our knowledge of their present-day distribution has focused on precipitation, much easier to sample than atmospheric water vapour.This sampling difficulty partly explains why applications dealing with studies of atmospheric processes and atmospheric dynamics have long been limited while they have rapidly developed in such fields as isotope hydrology and isotope paleoclimatology (from ice cores and other archives).
The situation has recently changed thanks to technological advances which now allow for either in situ measurement or remote estimation of δD v and δ 18 O v in atmospheric water vapour.The quantification of water isotopologues in tropospheric water vapour based on space based remote sensing techniques pioneered by Zakharov et al. (2004) is now under rapid development (Worden et al., 2006;Payne et al., 2007;Nassar et al., 2007;Steinwagner et al., 2007;Frankenberg et al., 2009;Schneider and Hase , 2011;Frankenberg et al., 2012;Herbin et al., 2009;Field et al., 2012;Lacour et al., 2012;Boesh et al., 2013) and provides large scale, integrated measurements.Data from ground based high resolution Fourier Transform Infrared (FTIR) spectrometers have been exploited to retrieve information about vertical profiles of water stable isotopologues (mainly δD v ) in water vapour from instruments both from the NDACC (Network for the Detection of Atmospheric Composition Change) sites (Schneider et al., 2006(Schneider et al., , 2010a(Schneider et al., ,b, 2012) ) and columnar δD v values from the TCCON (Total Carbon Column Observing Network) network (Boesh et al., 2013;Risi et al., 2012a).
A third major breakthrough has been accomplished when new infra red (IR) laser spectrometers have reached the same level of precision as mass spectrometers, and have became commercially available (Brand, 2009;Gupta et al., 2009).These devices are sufficiently robust to allow field measurements of the δD v and δ 18 O v composition of water vapour.After the development of calibration protocols, which require the introduction of reference waters and corrections for hu-midity and instrumental drift, such instruments have been deployed from tropical (Tremoy et al., 2012) to polar locations (Steen-Larsen et al., 2013) where they have revealed significant diurnal to seasonal variability in relationship with air mass origins, convection and surface-atmosphere moisture fluxes.Prior to the deployment of a network of stations where the δD v and δ 18 O v of surface water vapour will be continuously monitored, the information brought by water vapour stable isotopologues must be assessed for different climatic conditions.
In parallel, our ability to describe and simulate the distribution of water isotopologues using atmospheric general circulation models in which fractionation processes are embedded (IGCMs) has made considerable progress since the pioneering studies conducted in the eighties (Joussame et al., 1984;Jouzel et al., 1987).High resolution atmospheric models can now be nudged to atmospheric analyses products, allowing for precise comparisons with measurements in a consistent large scale meteorological framework.Sensitivity studies to uncertain atmospheric model parameterizations have shown the potential of water vapour isotopic data to constrain the representation of key processes linked to, e.g.cloud microphysics (Schmidt et al., 2005) or convection (Risi et al., 2012a).
In a comprehensive approach, Risi et al. (2012a,b) have brought together and compared satellite data sets from various instruments (SCIAMACHY, TES, ACE and MIPAS) and ground based remote sensing (FTIR at the NDACC and TCCON sites) and in situ techniques (surface vapour measurements and in situ aircraft data).From this comparison Risi et al. (2012a) extracted the most robust features and then used the LMDZ IGCM (LMDZiso) to understand and quantify the sources of differences between these data sets.They pointed out to significant differences between data sets but their common features appear to be remarkably well reproduced by LMDZiso in the lower and mid troposphere, at large scale.However, in Risi et al. (2012a), the amplitude of seasonal variations, the meridional isotopic gradient and the contrast between dry and convective tropical regions were underestimated by LMDZiso as well as by six other IGCMs involved in the SWING2 (Stable Water INtercomparison Group phase 2) intercomparison project.
Such data model intercomparison is a prerequisite if we want to use the variety of information on isotopic distribution in atmospheric water vapour (satellite data, ground based and in situ measurements) to diagnose biases in the representation of atmospheric processes in GCMs or infer information about, e.g.continental recycling.In their approach Risi et al. (2012b) aimed to use all available isotopic information with the consequence that the various data sets do not cover the same periods and the same locations, a difficulty which however is largely circumvented by applying a rigorous modeldata comparison methodology.
Here, we propose a complementary approach which consists in focusing on one site, the Kourovka Observatory (near Yekaterinburg, close to the western boundary of Western Siberia, 57.038 • N, 59.545 • S, see Fig. 1).This site is characterized by a well marked continental climate, with monthly mean temperatures varying from −16 • C (January) to +17 • C (July) and about 460 mm of annual precipitation, peaking in summer.It is affected by different air mass trajectories and summer continental precipitation recycling (Shalaumova et al., 2010).Its position in a pristine peatland and near the permafrost zone is strategic for the monitoring of the coupling between surface water and carbon budgets.At this site, we have access both to ground based (FTIR) and in situ vapour measurements (PICARRO L2130-i instrument).Comparison between different sets of δD were performed before, thus Lossow et al. (2011) compared between retrievals from different satellite sensors (Envisat/MIPAS, Odin/SMR, SCISAT/ACE-FTS), Schneider andHase (2011) compared IASI andNDACC FTIR retrievals, Boesh et al. (2013) compared GOSAT short-wave infrared and TCCON FTIR retrievals, and Worden et al. (2006) compared TES retrievals with in-situ measurements.Here, we inter-compare two independent data sources (PICARRO and FTIR) using the outputs of the ECHAM5-wiso isotopic AGCM (T63) that has been run in a nudged version using ERA-Interim reanalysis fields (Dee et al., 2011;Berrisford et al., 2009).This intercomparison will focus on a relatively short period between April and September 2012.

In situ isotopic measurements of surface water vapour
A Picarro laser instrument of type L2130-i was received by Ural Federal University in March 2012.Laboratory tests and further periodic calibration in the field were conducted in order to verify the reproducibility of the device using two different reference water samples: (i) DW (distilled water with δD=−96.4‰ ); (ii) YEKA (mixing Antarctic snow with distilled water with δD=−289.0 ‰).A third depleted reference sample (iii) DOMEC ( LSCE water standard with δD=−424.1 ‰) was also used to assess the linearity of the system.Exact isotopic values of the reference water samples were measured at LSCE by IRMS.
The instrument was installed in Kourovka Observatory in mid March 2012, inside the same room as the FTIR spectrometer.Also the Kourovka site is equipped with Gill Instruments MetPak-II meteorological station which provides every second measurements of atmospheric pressure, wind speed and direction, air temperature and relative humidity.Air conditioning was set up to warrant stable temperatures inside the room (around 18 • C).The sampling line consists of O'Brien optical quality stainless steel tubing (3/8 inch diameter) to minimize interactions between the sample and the tube.The length of the sampling line is 6 m, air being sampled about 8 ma.g.l.Self-regulating heating cable is used to keep stable temperature (around 60 • C) along the sampling line.The air input is protected against raindrops by a hard cover and against insects by a net.The station is implemented in the middle of a pine forest.
The measurement protocol consists of continuous ambient air measurements during 6 h, automatically switching to calibration with PICARRO Standard Delivery Module (SDM) performing a sequence of successive vaporization of DW and YEKA reference waters mixed with DRIERITE dried air during 30 min each.Altogether, each calibration sequence lasts about 60 min after accounting for pumping durations.Mean values and standard deviations in humidity and δD v are calculated along the last 20 min of calibrations.Typically, standard deviations of 200 ppm and 1 ‰ respectively are reported for humidity levels around 15 000 ppm.As for the air measurements, after switching from calibration the instrument demonstrated very high variability in the measurement results because of residual traces of water from reference standards in the system.To account for this effect, air measurements were processed only after a time delay of 13 min.This time period was found appropriate for this particular PICARRO device during its installation and calibration.
These frequent calibrations allow to assess the stability of the measurements.Starting from June 2012, instabilities were identified during calibrations, due to leakage in one of the SDM syringes.This may lead to two biases: (i) fractionation in the syringe itself by exchange with ambient air, and (ii) introduction of air bubbles into the SDM and instabilities of injected flux and resulting measurements.The calibration module was subsequently replaced using a new type of glass syringe in September 2012, leading so far to very stable calibrations.
The quality of the post-processed data strongly depends on the stability of the calibrations.Water standard measurements are considered unstable and not taken into account when standard deviations of humidity and δD v are above 600 ppm and 3 ‰, respectively.During the measurement period (from April to September) the total number of successful calibrations was 70 for DW and 130 for YEKA.
Fig. 2 shows the variability in the measured standards fitted with polynomial of 5th degree.Despite the presence of a certain gaps in calibrations due to the problems with power supply, leakage of SDM syringe and instability of SDM software, the overall drift of the system during 6 months is less than 5 ‰.Fitted polynomials were used for transferring the instrument isotopic values to the V-SMOW scale.
As it outlined by Steen-Larsen et al. (2013) isotopic measurements are sensitive to water vapour concentration, so it is required to establish calibration response functions as a function of humidity, based on measurement of reference waters injected at different humidity levels, from 1 000 to 25 000 ppm.These response functions were determined in-situ in April 2012.However for the period of interest we do not observe any significant effect on the measured δD v values.Humidity level changes from 5 000 to 29 000 ppm with the mean value of 14 000 ppm and this leads to variation in humidityisotopologue response less than 3 ‰.
The overall calibration protocol used was realized according to the 6 step protocol, described in Steen-Larsen et al. (2013), except for the humidity correction that was not applied for this case.
Finally, we use both hourly and daily average values to present this data set and compare it with ECHAM5-wiso model results.For the period from 1 April to 30 September 2012 over which we will compare data and model results, 2476 hourly PICARRO measurements were produced which corresponds to 67 % of the total duration (3671 h).As FTIR (Sect.4) data are specifically used to get information about δD v in the water column, we will hereafter focus on this parameter both for the discussion of the dataset and its comparison with model results.
The currently available δD v dataset extends to 21 November.We have, in Fig. 3, displayed both hourly individual measurements and a smooth curve (5 point running mean) limited to periods over which there are at least 6 hourly measurements successively.The amount of water vapour (as measured by the PICARRO instrument) has been reported on the same figure along with surface temperature using either measurements at the site available since 1 April or ERAinterim reanalysis data for the entire period (as used for the simulation).
As expected, the deuterium time series shows a clear seasonal cycle with it lowest values in spring (minimum −232 ‰ on 5 April ) and in fall (minimum −246 ‰ on 18 November) and highest levels during summer (−103 ‰ reached on 14 July and 11 August).While the highest monthly mean values are observed in June-August, the highest single hourly value is recorded to occur in spring(−92 ‰ on 10 May).Indeed, large, and for some of them, rapid δD v variations are superimposed on this seasonal cycle which will be fully described when winter data will be available.These fluctuations are more pronounced in fall with amplitudes reaching about 100 ‰ than during the summer during which no fluctuation exceeds 45 ‰.They are clearly related to large variations in temperature and to associated changes in the amount of water vapour, q v .Although much too simplistic, a Rayleigh type model helps to understand this link between δD v , temperature and q v .This model (Dansgaard, 1964) considers the isotopic fractionation occurring in an isolated air parcel in which the condensed phase is assumed to form in isotopic equilibrium with the surrounding vapour and to be removed immediately from the parcel.The isotopic composition of the water vapour at a given site, δD v , is well approximated by: δD v = (1 + δD 0 )(q v /q 0 ) (αm−1) −1, in which δD 0 and q 0 are the deuterium content and the amount of water vapour at the oceanic origin of the airmass while α m stands for the average value of the fractionation coefficient between the oceanic source and the sampling site.Assuming no change in the conditions prevailing at the oceanic source (which again is too simplistic) this should translate in a linear relationship between ln(1+δD v ) and ln(q v ) while the link with site temperature results from the Clausius-Clapeyron equation.
With this in mind, we have plotted ln(1 + δD v ) versus ln(q v ) hourly (Fig. 4a) and daily (Fig. 4b) means and δD v versus the site temperature using either hourly data for the period over which we have measurements at the site (Fig. 4c) or daily temperature in the Kourovka gridbox (Fig. 4d) as derived from reanalysis data (see Sect. 4.1).In line with the Rayleigh model in which the fraction of water remaining in the cloud is the primary driver of isotopic changes, there is a strong correlation (r 2 = 0.67) between ln(1+δD v ) and ln(q v ) for hourly data which increases (r 2 = 0.71) when considering daily data and thus eliminating the diurnal cycle.The correlation of δD v with temperature is weaker for hourly data either using local meteorological measurements (r 2 = 0.46) or reanalysis data (r 2 = 0.49, not shown).It increases for daily data (r 2 = 0.72) at a similar level as observed for ln(q v ).
Columnar values of δD F T IR v were derived from total column abundances of HDO and H 2 O retrieved from the spectra recorded from July 2012 to June 2013 in Kourovka.For data processing, the standard TCCON software GFIT was used (Wunch et al., 2010(Wunch et al., , 2011)).GFIT retrieves the total number of molecules in the vertical atmospheric column, using the algorithm of profile scaling retrieval with the assumption that the shape of the profile of the retrieved gas is well known.H 2 O, temperature and pressure a priori profiles are based on reanalysis data provided by National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP/NCAR) (Kalnay et al., 1996).The HDO a priori profile is calculated from H 2 O profile as follows (Wunch et al., 2011): x apr HDO = 0.16x apr H2O 8.0 + log 10 x apr

H2O
(1) where x apr HDO is the a priori HDO volume mixing ratio (vmr) profile, and x apr H2O is the a priori H 2 O vmr profile.Examples for H 2 O and corresponding δD F T IR v a priori profiles for each day of July 2012 are shown in Fig. 9. Microwindows containing saturated H 2 O lines were excluded from final results to achieve more robust retrieval.As data base of spectral parameters, the revised water vapour line list was used (Shillings et al., 2011).To check the sensitivity of the retrieval to the a priori vertical profile of HDO, another retrieval run with constant δD v a priori profile equal to zero (corresponding to natural abundance of HDO in the ocean) was performed.Fig. 5 shows a scatter plot for δD v values retrieved using different a priori vertical profiles.This reveals that the retrieval of columnar δD v values does not depend essentially on the initial guess.
4 The ECHAM isotopic model

Model setup
Atmospheric simulations were carried out using ECHAM5wiso (Werner et al., 2011), which is the isotopologuesenhanced version of the atmospheric general circulation model ECHAM5 (Roeckner et al., 2003;Hagemann et al., 2006;Roeckner et al., 2006).Both stable water isotopologues H 18 2 O and HDO have been explicitly implemented into its hydrological cycle (Werner et al., 2011) analogous to the isotope modelling 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 (vapour, cloud liquid, cloud ice) being transported independently in ECHAM5, a corresponding isotopic counterpart is implemented in the model code.The isotopic and the "normal" water are described identically in the GCM as long as no phase transitions are concerned.Therefore, the transport scheme both for active tracers (moisture, cloud liquid water) and for the corresponding passive tracers (moisture, cloud water and cloud ice of the isotopologues) is the fluxform semi-Lagrangian transport scheme for positive definite variables implemented in ECHAM5 (Lin and Rood , 1996).Additional fractionation processes are defined for the water isotopologue variables whenever a phase change of the "normal" water occurs in ECHAM5.Two types of fractionation processes are considered in the model: equilibrium and non-equilibrium processes.An equilibrium fractionation takes place if the corresponding phase change is slow enough to allow full isotopic equilibrium.On the other hand non-equilibrium processes depend even on the velocity of the phase change, and therefore on the molecular diffusivity of the water isotopologues.Processes which involve fractionation processes 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 neglected ( see Hoffmann et al. (1998) for detailed discussion of this issue).
ECHAM5-wiso has been validated with observations of isotopologue concentrations in precipitation and water vapour (Langebroeck et al., 2011;Werner et al., 2011).On a global and European scale, the annual as well as seasonal ECHAM-5-wiso simulation results are in good agreement with available observations from the Global Network of Isotopes in Precipitation, GNIP (IAEA-WMO, 2006).Furthermore, it has been shown that the simulation of water isotopologues in precipitation does clearly improve for an increased horizontal and vertical model resolution (Werner et al., 2011).The simulated near-surface isotopic composition of atmospheric water vapour δD v is also in fairly good agreement with recent observations from five different GNIP stations.Model values and measurements agree well with differences in the range of ±10 ‰.A comparison of the ECHAM5-wiso simulations with total column averaged HDO data determined by the SCIAMACHY instrument on board the environmental satellite ENVISAT (Frankenberg et al., 2009) shows the same latitudinal gradients, but an offset between 20-50 ‰ of unknown origin.Focusing on Europe, the results by Langebroeck et al. (2011) indicate 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.This finding is not just valid for ECHAM5-wiso results, but also supported by other modeling results (e.g.Schmidt et al., 2005) and confirmed by observational data .
Based on our previous findings, we employ in this study the ECHAM5-wiso model with a medium-fine horizontal spectral resolution T63 (about 1.9 • × 1.9 • ).The vertical resolution is 31 hybrid levels.The model is forced with prescribed values of present-day insolation and greenhouse gas concentrations (IPCC, 2000), as well as with sea-surface temperatures and sea-ice concentrations according to ERA-Interim reanalysis data (Dee et al., 2011;Berrisford et al., 2009; data have been obtained from the ECMWF data server).
In order to allow a comparison with observations at the sub-seasonal scale, the ECHAM5-wiso model is nudged to reanalysis data, which ensures that the large scale atmospheric dynamics is correctly represented.Every six hours the dynamic-thermodynamic state of the model atmosphere is constrained to observations by implicit nudging (e.g.Krishamurti et al., 1991;implemented by Rast, 2012), i.e. modeled fields of surface pressure, temperature, divergence and vorticity are relaxed to ERA-Interim reanalysis fields (Dee et al., 2011;Berrisford et al., 2009).If we compare climatological means of measured surface temperatures in Yekater-inburg (Server, operated by the Space Monitoring Information Support laboratory, SMIS SRI RAS) with ERA-40 climatology data, we find a good agreement of the temperature seasonal cycle.The ERA-40 mean monthly surface temperatures show a small warm bias of less than 1 • C for the period May-November, and slightly larger deviations (+1.0 to +2.2 • C) between December and April.
Although the hydrological cycle in our ECHAM5 setup is fully prognostic and not nudged to the ERA-Interim data, in Western Siberia the differences of the simulation results as compared to the hydrometeorological reanalysis fields are small.For instance, modelled daily precipitation agrees within 1 mmday −1 with reanalysis data, and the agreement with observations further improves if monthly averages are considered.The simulated total column water vapour (TCWV) tends to be systematically overestimated by 4-6 mm compared with reanalysis fields.
Our simulation starts on 1 January 2000, with an internal model time step of 12 min.For comparison with the available isotope observational records at Kourovka, we analyze simulation results for the period April to September 2012.We always evaluate model results with a temporal resolution of one hour, if not stated otherwise.For Kourovka, we are using values at the model grid point closest to the station.
For the period April to September 2012, an analysis of ERA-40 and ERA-interim surface temperature data reveals that the region around Kourovka station was anomalous warm, as compared to the long-time average temperatures (reference period 1960-1999).Strongest above-average warming with temperature anomalies of ≈ +4 • C occurred in April and June, while in May and July temperatures were about 1-2 • C warmer than average, only.For August, we find still an above-average warming of 1-2 • C at Kourovka and adjacent regions of Western Siberia, but also cooler than average temperatures of the same order of magnitude in large parts of East Siberia.For September, temperatures in all Siberian regions have been anomalous warm by ≈ 1-3 • C again.

Model results
We briefly describe the simulated near-surface temperature and surface pressure at the location of Kourovka (Fig. 6).A clear diurnal cycle is evidenced with typical day-versusnight temperature changes of ≈ 5-10 • C. Superimposed on this diurnal cycle, the temperature record reveals strong variations within a timescale of a few days.These changes can be as large as 10-15 • C. On the seasonal time scale, the difference between low temperature values in April and September, respectively, and the summer temperature maximum in mid-July to mid-August adds up to ≈ 20 • C.This is slightly higher than the climatological observations from Yekaterinburg.Surface pressure at Kourovka varies between 960 hPa and 1000 hPa.This record also shows some multi-day variations but clearly lacks both a diurnal and seasonal cycle.
The simulated amount of water vapour q v in the lowest atmospheric model layer also shows strong temporal variations at a time scale of a few days.While the water content in the air is rather low (3-6 gkg −1 air) between the beginning of April and early May, it rises thereafter to values of up to 15 gkg −1 air.From mid-July to end of September, the simulated q v values then fall back into the range 5-10 gkg −1 air.
ECHAM5-wiso simulates surface-level water vapour δD v values (hereafter δD v ) mostly in the range −200 to −100 ‰ at the Kourovka site between April and September 2012 (Fig. 6).The model shows isotopic variations of 30-50 ‰ over a few days, over which are superimposed smaller shortterm fluctuations lasting a few hours.The lowest δD v values are found in early April and early May as well as in mid to late September, while summer δD v values are less depleted.A distinct peak event in δD v occurs between 30 August and 3 September.
Both the simulated δD v values of the total water vapour column and δD in precipitation (not shown) are highly correlated with the simulated δD v values near surface (r = 0.90 and r = 0.97, respectively, for hourly values between 1 April and 30 September).Compared to the surface values, the δD v signal of total water column is depleted by ≈ 20-30 ‰.Precipitation occurs at 1216 1-h intervals between April and September (total number of 1-h intervals during this period: 4392) with a mean enrichment of ≈ +70 ‰ as compared to the surrounding vapour.
As seen in Fig. 6, the multi-day variations of δD v , q v and surface pressure are strongly correlated.From our analyses, we find the strongest links between variations of temperature and water amount q v (r = 0.70), while variations of δD v are only weakly linked to local temperature (r = 0.56) and q v (r = 0.60).Our results support previous findings that δD v variations on daily and synoptic time scales are often not strongly correlated with local temperature or water amount changes, but rather represent a more integrated signal of the climatic conditions during the transport of the vapour to a specific site (e.g.Schmidt et al., 2005;Langebroeck et al., 2011).Modeled surface pressure variations at Kourovka are neither strongly correlated to surface temperatures, water vapour, nor to δD v (correlation coefficient |r| < 0.2 in all cases).
In addition to δD, the isotopic signal of δ 18 O of the various water reservoirs and fluxes is also modeled within this ECHAM5-wiso simulation.At the grid point closest to Kourovka, we find a strong linear correlation between hourly values of δD v and δ 18 O v (r = 0.997), with a slope of m = 7.99 and a mean deuterium excess value d (defined as d = δD−8 × δ 18 O) of +10.2 ‰. Between April and September, the modeled hourly excess values range between +5 ‰ and +20 ‰.The potential use of the deuterium excess data to identify different transport regimes of moisture towards Kourovka will be investigated in detail in future studies.
From correlation analyses (not shown) of the simulated daily mean δD v values at Kourovka and the isotopic composition at all other grid cells, we estimate that variations of isotope values in vapour at Kourovka are representative for isotopic changes in a region between 45-75 • E and 48-66 • N, with a correlation coefficient r higher than +0.5.A similar correlation pattern is found for variations of the water vapour amount q, but with slightly higher mean correlation coefficients (r ≥ +0.65).In contrast to these water quantities, the simulated near-surface temperature shows a much stronger and spatially extended correlation between Kourovka and its surroundings (mean r values> 0.9).

Surface δD v : model data comparison
The ECHAM5-wiso results are first compared to the observed hourly water vapour PICARRO data q v (Fig. 3c, red lines).The model correctly captures the patterns and magnitude of variability, with a very large correlation coefficient (r = 0.89).Absolute values of water vapour measured with the PICARRO instrument are up to 20 % higher than the related model values.This might be explained by the facts, that the ECHAM5-wiso values represent the mean of the lowest atmospheric model level (surface to ≈ 60 m) while the PI-CARRO measurements were carried out at a height of 8 m, and that the closest model grid box has much larger size than point of in situ measurements and is located above complicated terrain containing small lakes and rivers with altitude differences about 400 m.
Simulated δD v values are often 30-40 ‰ less depleted than the corresponding PICARRO data.This suggests a lack of depletion either along air mass trajectories or due to boundary layer mixing.Despite the systematic offset, a high correlation (r = 0.77) is obtained between model and observed δD v hourly variations.This result shows that the intra-seasonal δD v variability at Kourovka is dominated by the synoptic variability, which is correctly resolved by the model in the nudged configuration.
The PICARRO data exhibit a stronger correlation between δD v and q v (r = 0.73) than simulated (r = 0.60).We note that this might be partly influenced by the lower number of measured data points (n = 3066) as compared to the total number of hourly modeled values (n = 4392) available for the period between April and September 2012.However, if we limit the analyses of the ECHAM5 values to those points in time, when PICARRO measurements exist, the simulated correlation between δD v and q v strengthens slightly (r = 0.61), only.
The PICARRO observations and ECHAM5-wiso results consistently depict two pronounced δD v negative excursions with minimum and maximum values of −200 ‰ and −100 ‰, respectively, for the first days of April 2012 and May 2012.Another negative excursion occurs on 12 September.Exemplarily, we have chosen the May event with highly depleted δD v values between end of April and early May for a detailed analysis of the atmospheric conditions leading to this fast and strong isotope shift in vapour at Kourovka (Fig. 7).In the model framework, the minimum in δD v lags a local surface pressure minimum by 1 day and precedes a drop in surface air temperature, which reaches its lowest temperatures 4 days later.This sequence of events suggests that such δD v variations at Kourovka are related to passages of dynamic low and high pressure systems and advection of remote air masses.This hypothesis is further investigated by analysis of the isobaric flow at 850 hPa.A few days before this depletion event, the Kourovaka area was receiving southwesterly air masses transporting relatively warm and enriched vapour (Fig. 8, top panel).Around 1 May, the atmospheric circulation changed due to a pronounced low pressure system north of Kourovka.As a result, the main air flow was then transported from central Siberia with depleted δD v levels (Fig. 8, middle panel).During the following days, this northerly airflow caused the cooling at Kourovka.Starting from 7 May, a new high pressure system south of Kourovka was again dominating the atmospheric flow pattern, bringing warm and relatively enriched vapour to this region (Fig. 8, bottom panel).These simulated changes in atmospheric transport to Kourovka between 20 April and 11 May are in good agreement with back trajectory analyses of air masses, available from the AERONET (2012) for the location of Yekaterinburg (not shown).
We conclude that PICARRO measurements and ECHAM5-wiso simulation results of δD v and related quantities (vapour q, surface temperature) between April and September 2012 are in good agreement.Even short-term isotope variations occurring on an hourly time scale are correctly reproduced in this nudged simulation.Thus, one may safely use the ECHAM5-wiso model results for an improved interpretation of observed isotopic variations near Kourovka in future studies.

Comparison of FTIR data with measured at surface and model data
Since the model provides hourly-averaged output data, data retrieved from FTIR measurements taken within 1 h were also averaged.For the comparison between model and FTIR observations we assume that the modeled HDO and H 2 O profiles are true, and we smooth original model values by applying the following equation (Rodgers and Connor, 2003;Risi et al., 2012a): Here, Q is the retrieved total column mass of HDO or H 2 O, respectively, q sim is the specific humidity profile simulated by the model for atmospheric layer i, q apr is the specific humidity in the same layer according to the a priori profile used in the retrieval (converted from wet to dry-mole fractional values according to Wunch et al., 2010), A i is the ith component of the column averaging kernel vector, ∆P i is the thickness of the i-th atmospheric layer, g is the gravity acceleration.Column averaging kernel as a function of pressure for different solar zenith angles of measurements are shown in Fig. 10.TCCON a priori and averaging kernel profiles are tabulated using a different vertical coordinate system than the model profiles (71 pressure levels versus 31 hybrid layers).To ensure numerical consistency, model profiles were interpolated to the same vertical resolution as averaging kernel (71 pressure levels) before the vertical integration was carried out.The δD F T IR v of total column water vapour (δD TCVW ) was then calculated from the normalized ratio of Q HDO and Q H2O .
To check how retrieved columnar δD F T IR v values correlate to collocated ground δD v values we selected FTIR and extended until April 2013 PICARRO measurements taken within the same hours.
Before we enter the comparison between smoothed ECHAM5-wiso results and observations, we consider the effect of the applying column averaging kernels to the original model results.In Kourovka, it shifts the original model results for δD TCVW to more positive values by about 9 ‰ in the average, and also induces a change of the expected slope between smoothed and originally simulated δD TCVW from 1.0 to 0.87.The positive shift of retrieval values is essentially a consequence of the fact that between 1000 and 200 hPa, the isotopic ratio of TCCON a priori profiles is systematically higher than in the ECHAM simulations.The change of the slopes deserves further investigation.11a).We do not find any systematic trend underlying the differences.The observations are systematically shifted to the higher values comparing to the model results and have larger amplitude of variations (by approximately 30 percent).It can be explained by the uncertainties and incorrect temperature dependence in spectroscopic line parameters in the linelist.The measured increase of δD F T IR v during 10 July, 2012 is also found in the model results but with a smaller amplitude (10 ‰).In the model, this fast isotopic enrichment coincides with the temporal evolution of lower tropospheric temperatures, exhibiting for Kourovka a pronounced diurnal cycle during the summer months.Given the limited number of observations that are available so far, a rigid interpretation and assessment FTIR measurements from Kourovka has to be postponed to the future with more detailed investigation of retrieval sensi-tivity and cross-dependance on humidity.
The comparison of hourly averaged columnar FTIR measurements with hourly averaged PICARRO ground measurements shows correlation r 2 = 0.84 (see Fig. 11b).Amplitude of seasonal variations of the retrieved δD F T IR v is approximately 2 times larger than that of δD v values measured by PICARRO at surface.

Conclusions
The present study is part of a project aiming to investigate the water and carbon cycles in permafrost regions and pristine peatlands of Western Siberia and their projected changes under a warming climate.The isotopic approach is a key element of this project and the results that we have presented and discussed in this article should be considered as a first and necessary step to fully exploit the isotopic information contained in water vapour.To this end we have combined two independent methods to acquire data (continuous surface measurements and FTIR) and evaluate them against the results derived from a dedicated simulation of the ECHAM5wiso IGCM focusing on this region.
As expected from a Rayleigh type model, and generally observed in middle and high latitude regions (Rozanski et al., 1992), a significant part of the daily isotopic variations (δD v ) observed in Kourovka water vapour is explained by local changes in the amount of water vapour (r 2 = 0.71) and temperature (r 2 = 0.72).Obviously, a general circulation model which accounts for the origin of the water vapour, for the complexity of weather situations and for the differences of associated fractionations (e.g convective versus nonconvective systems) is a more appropriate tool to examine the link between δD v and climatic parameters.There is indeed an excellent correlation between observed and predicted δD v values including for rapid excursions related to concurrent changes in atmospheric circulation.
This data model comparison fully justifies the use of ECHAM5-wiso to evaluate the method based on the exploitation of FTIR data, allowing remote measurements of δD v in the water column.The method is very satisfying although being limited to a small number of days.
To sum up, the δD v comparison between two observational approaches and a medium-high resolution IGCM, undertaken for the first time at a given site, is quite promising.Data acquisition with the Picarro instrument will be now performed on a continuous basis and a second instrument will be deployed in summer 2013 at Labytnangi located near the Arctic circle (66 • 39 N, 66 • 23 E, see Fig. 1) with the aim to contribute to an Arctic network now under development.Further work will include the exploitation of oxygen-18 and associated deuterium excess from the PICARRO data, a comparison of the algorithm developed to infer column δD v from the GOSAT thermal infrared band and the method currently applied in the short-wave infrared (Frankenberg et al., 2012), as well as the development of an improved algorithm to exploit FTIR data for isotopic purposes.At last the use of a second IGCM (LMDZiso) should help to interpret these data in a larger geographical context.
FTIR measurements were carried out in Kourovka on 32 days during one year since July 10, 2012 until June 30, 2013.Retrieved values of δD F T IR v range from −316 ‰ on February 22, 2013 to −40 ‰ on June 12, 2013.Multiple measurements on July 10, 2012 record an increase of δD F T IR v from morning to noon by about 15 ‰.Observations and retrieved model results are correlated with r 2 = 0.84 and scatter with an absolute standard deviation of 22 ‰ (see Fig.

Fig. 1 .
Fig. 1.Map of the target region (Western Siberia).Kourovka observation site is marked with red star and Yekaterinburg is marked with red circle.White star stands for future site in Labytnangi.

Fig. 5 .Fig. 6 .
Fig. 5. Sensitivity of the retrieval from FTIR observations to different a-priori assumptions.Retrievals started from initial guess defined by the formula 1 are plotted vs retrievals started from constant and equal to zero initial guess δDv vertical profile.Standard deviation of retrieval difference is about 1 ‰.