Satellite-retrieved data of total water vapor (TWV) over the Arctic are patchy, with large areas of missing data because of various limitations of the retrieval algorithms. To overcome these observational difficulties, a new retrieval algorithm has been developed that allows for monitoring the TWV over the Arctic during most of the year. This method retrieves TWV from satellite microwave radiometer data [the Advanced Microwave Sounding Unit B (AMSU-B)]. These new data have been made available for 4 yr (2000–03) and have been used to evaluate high-resolution simulations with the Arctic regional atmospheric climate model HIRHAM at daily, monthly, and seasonal time scales. The strong dynamic TWV variability on the daily time scale, linked with moisture transport by weather systems, is discussed for selected case studies. Both the simulated climatological seasonal mean patterns and the variability on interannual and decadal time scales are in agreement with those of the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) data. Trends in Arctic TWV for 1958–2001, broken down by season, are presented. Although an increase in the TWV is obvious in all seasons, there are also regions where a decreasing trend appears. Significant maximum positive trends are calculated over the western Arctic in summer (up to 0.06 kg m−2 yr−1), and a significant small negative trend is calculated over the East Siberian Sea in winter.
Water vapor plays a key role in the global climate system (Rossow 1996; Chaboureau et al. 1998): Water vapor 1) is the most abundant and the most radiatively important greenhouse gas, 2) can transport large amounts of latent heat and thus strongly influences the dynamics of the atmosphere, and 3) is an important link connecting the various components of the hydrological cycle. There is a need for the continuous collection of global data on the water vapor in the atmosphere, if possible as a function of height (water vapor profiles), or at least the vertically integrated water vapor content of the atmosphere, called column water vapor, precipitable water, or total water vapor (TWV).
There is a global network of meteorological stations that regularly measure vertical profiles of atmospheric humidity (among other variables) by radiosondes. Such measurements are, however, just point measurements, and the station network is very sparse in the Arctic region. Water vapor data are also provided by satellite retrievals, which have the advantage that they are dense in space and time, compared to the direct station measurements. However, most satellite-based retrievals of water vapor have other serious limitations, in particular over the Arctic. They are either hampered by clouds (infrared, visible), need sunlight (visible, near-infrared), or are restricted to open water [Special Sensor Microwave Imager (SSM/I) microwave]. Another source of water vapor data is reanalysis datasets. Here, we note that the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis does not assimilate satellite-derived moisture information (Kistler et al. 2001), while the European Centre for Medium-Range Weather Forecasts (ECMWF) 40-yr reanalysis (ERA-40) does (Uppala et al. 2005). Investigations suggest that the ECMWF moisture processes are more realistically simulated over the Arctic than those in the NCEP–NCAR reanalysis (Bromwich et al. 2000; Groves and Francis 2002a).
To overcome the observational difficulties, a new retrieval algorithm has been developed that allows for monitoring the TWV over the Arctic during most of the year. This method retrieves TWV from satellite microwave radiometer data (Melsheimer and Heygster 2006, 2008) and uses the radiances at five channels: two window channels (no strong absorption lines) at 89 and 150 GHz, and three channels close to the strong water vapor absorption line at 183.31 GHz. It derives the TWV from ratios of the channel differences and has the strength to be independent of daylight and most clouds. However, this method can only retrieve TWV values up to about 7 kg m−2 (15 kg m−2 over ice), which makes it suitable for the polar regions. Total coverage of the Arctic is achieved at least once daily, with a spatial resolution of about 50 km. Therefore, this dataset is particularly qualified for use in verifying high-resolution simulations of the Arctic climate, available from Arctic regional climate model (RCM) simulations.
The aim of the paper is to use the new satellite-derived data for four recent years to evaluate Arctic RCM simulations of TWV for the first time. At the same time, ECMWF analysis data and earlier satellite-derived data are included in this intercomparison. The objectives are, on the one hand, to emphasize the usefulness of the new satellite dataset for model evaluation purposes and, on the other hand, to highlight the reasonable simulation with the Arctic RCM. Due to the restrictions of the different algorithms of satellite retrievals (cloud detection, need of sunlight, restriction to open water), often there are no data available. Here, reanalyses and RCM simulations come into play as they can be understood as a kind of physically based data interpolator and can close data gaps, particularly over the largely ice-covered Arctic Ocean. Within this context, a long-term climatology of simulated TWV is presented and compared with ERA-40 data. Subsequently, recent decadal trends of TWV are discussed based on the RCM simulations because trends in water vapor are of considerable interest due to their major role in climate changes.
The paper continues in section 2 with a detailed description of the new method used to derive TWV from satellite microwave radiometer data as well as a characterization of the other datasets used. Moreover, the applied RCM and the performed simulations are described. In section 3, the simulated Arctic TWV and a comparison with the observational data are presented. Finally, the results are summarized in section 4.
2. Data and simulations
a. Observational data
1) TWV data derived from AMSU-B
Our method of retrieving TWV over polar regions uses data from the Advanced Microwave Sounding Unit B (AMSU-B) microwave radiometer on board the polar-orbiting satellites of the U.S. National Oceanic and Atmospheric Administration (NOAA): NOAA-15, NOAA-16, and NOAA-17. While AMSU-B is designed and operationally used for humidity sounding, this fails over polar regions since there, 1) the TWV content of the atmosphere is so low that the contribution from surface emission is substantial and 2) the surface emission is poorly understood and highly variable because of the variable ice cover of the seas. Our method for retrieving TWV is complementary in that it works exactly where the atmosphere is dry enough for the ground to be “seen” by the sensor, and it is mostly independent of the surface emissivity. The basic idea (Miao et al. 2001) is to use three channels where the surface emissivity is similar but the water vapor absorption is different, such as the three AMSU-B channels centered around the 183.3-GHz water vapor line. Starting from the radiative transfer equation for a not too opaque atmosphere in the approximation of Guissard and Sobieski (1994), an equation for the TWV, denoted by W, can be derived (Miao et al. 2001) that does not depend on surface emissivity:
where θ is the viewing angle, Ti is the brightness temperature measured by AMSU-B at channel i, while the three channels i, j, and k are sorted in such a way that for the corresponding water vapor absorption coefficients, κi, κj, κk, we have κi < κj < κk. The four parameters C0, C1, Fx, and Fy, which we shall call calibration parameters, have to be determined empirically. To do this, radiosonde data (Arctic, 1997–2001, about 27 000 profiles) were taken as input for simulating AMSU-B brightness temperatures with the Atmospheric Radiative Transfer Simulator (ARTS; Buehler et al. 2005) model, using a range of different surface emissivities between 0.6 and 0.96. In addition, the TWV was calculated directly from the humidity data of each radiosonde profile. Several linear regressions then yield the four calibration parameters: C0, C1, Fx, and Fy (for details see Melsheimer and Heygster 2008). Using the three channels near the water vapor line [AMSU-B channels 20, 19, and 18; i.e., (i, j, k) = (20, 19, 18)], the method works up to TWV contents of about 1.5 kg m−2. For higher TWV, channel 18, the one most sensitive to water vapor, becomes saturated and retrieval fails. If we replace channel 18 with window channel 17 at 150 GHz, that is, (i, j, k) = (17, 20, 19), the method works up to TWV contents of about 7 kg m−2. Such water vapor values are typical for the Arctic Ocean, Siberia, and northern Canada in winter and for Greenland almost year-round. So, if we have determined these four calibration parameters, the TWV can be calculated from AMSU-B brightness temperatures without any further input. To extend the retrieved water vapor range to higher values, channel 19 is replaced by window channel 16 at 89 GHz; that is, (i, j, k) = (16, 17, 0). Since the emissivity of sea ice and ocean at 89 GHz is significantly different from the emissivity at 150 and 183 GHz, the equation for W that can be derived now explicitly contains the surface emissivity, unlike Eq. (1). We shall call this equation the extended algorithm:
where r1 = 1 − ɛ89 and r2 = 1 − ɛ150 are the surface reflectivities at 89 and 150 GHz, C depends on the water vapor absorption coefficients but can safely be approximated by 1.1 for TWV above 7 kg m−2, and the other variables have the same meaning as in (1) above. This means that now, some information about the emissivity of sea ice at 89 and 150 GHz is needed. We extracted this information from emissivity measurements over sea ice and open water as part of the Surface Emissivities in Polar Regions-Polar Experiment (SEPOR/POLEX; Selbach 2003) campaign. The airborne radiometer used in that campaign does not have a channel at 150 GHz, but at 157 GHz, and the difference between the surface emissivities at 150 and 157 GHz can be expected to in the range of 0.01. Analysis of these data shows a moderately high correlation of the emissivities ɛ89 and ɛ157 of sea ice at 89 and 157 GHz, respectively. A linear regression yields
where we have imposed the additional constraint that ɛ89 = 1 if ɛ157 = 1. This constraint is necessary to prevent one of the emissivities from exceeding 1.0 (which is physically impossible) when the other one approaches 1.0. Based on these data, the reflectivity ratio r2 / r1 over sea ice can be approximated by a constant value of 1.22 (which assumes that it is constant over time; limitations of this assumption are discussed in section 3). This means that for the extended algorithm, we need, in addition to the four calibration parameters, information on the sea ice cover. The algorithm is then applied over sea ice. Because the emissivity of open water is rather well known, it is in principle possible to adapt the extended algorithm for use over open water. However, we have not done this here since there are other remote sensing methods to retrieve TWV over open water, for example, from other passive microwave sensors like SSM/I. Using this extended algorithm, the upper limit of the TWV that can be retrieved is about 15 kg m−2. Thus, combining all three “subalgorithms” mentioned so far (using channel triples 20, 19, and 18 for low TWV; 17, 20, and 19 for mid-TWV; 16, 17, and 20; and the extended algorithm for high TWV over ice), TWV values from 0 to about 15 kg m−2 can be retrieved from AMSU-B data, using three sets of the four calibration parameters. We start with the most sensitive (low TWV) algorithm and switch to the next choice only when TWV is too high to be retrieved (for details, see Melsheimer and Heygster 2008).
Daily AMSU-B-derived TWV maps have been produced by the combination of all swaths (about 14) for 1 day, interpolated, and gridded on a 0.5° × 0.5° latitude–longitude grid (which is reasonable since the resolution of the data is 15–50 km, depending on the position in the swath). The gridding is done using a near-neighbor algorithm with four sectors, the standard method of the Generic Mapping Tools (GMT, information online at http://gmt.soest.hawaii.edu/). The swaths have more overlaps the nearer they are to the pole. The monthly means have been calculated only for those grid points that have data for at least 25 days in that month. As seen in the later figures, often there are no data available. This happens when the TWV is above the limit that can be retrieved (“channel saturation”), that is, about 15 kg m−2 over sea ice and about 7 kg m−2 otherwise. Note that allowing five or six missing daily TWV values per month (i.e., requiring 25 days with TWV data) can thus introduce a negative bias into the monthly mean.
A comparison of AMSU-B-retrieved TWV data over the Arctic with data from radiosondes and the ERA-40 reanalysis shows good agreement. However, AMSU-B-derived TWV values have a positive bias ranging from 0.3 to 2.7 kg m−2 with respect to the ERA-40 and radiosonde data (Melsheimer and Heygster 2008).
2) TWV data derived from TOVS
Another satellite-derived dataset of TWV over polar regions is based on infrared data of the Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS), on board the TIROS-N satellite and the polar-orbiting series of NOAA satellites (NOAA-6–NOAA-14). This dataset is part of the NOAA–National Aeronautics and Space Administration (NASA) TOVS Polar Pathfinder (so-called Path-P) dataset (Schweiger et al. 2002). This dataset covers the years 1979–2004 and comprises atmospheric temperature and humidity values at several levels, as well as the surface temperature and a number of cloud parameters. The TOVS sensor consists of three parts: the High-Resolution Infrared Radiation Sounder (HIRS), the Microwave Sounding Unit (MSU), and the Stratospheric Sounding Unit (SSU), of which the latter is not used for retrieving the Path-P dataset. The TOVS radiances are processed using a version of the improved initialization inversion (3I) algorithm (Chedin et al. 1985). Starting with an initial guess based on a database of precalculated TOVS radiances, 3I uses sophisticated cloud-clearing methods and Bayesian statistical retrieval for temperature and humidity. The spatial resolution is 100 km; the spatial coverage is poleward of 60° latitude. Daily and monthly data are provided by the National Snow and Ice Data Center (NSIDC; information online at http://nsidc.org/data/nsidc-0027.html). Note that since the humidity retrieval is based on infrared data, TWV in the presence of dense clouds cannot be reliably retrieved; as in that case, the atmosphere below the cloud is invisible to the sensor. The TOVS-retrieved TWV data compare well with radiosonde observations in the Arctic (correlation coefficient up to 0.9), but TOVS-derived TWV values are lower compared to those from radiosondes by about 1 kg m−2 in winter and 3 kg m−2 in summer; the root-mean-square uncertainty ranges from 1.0 to 2.2 kg m−2 (Groves and Francis 2002a).
3) TWV data from ECMWF analysis
For the period 2000–03, the TWV data are taken from the operational ECMWF (ECMWF OD) analysis (T159 spectral resolution, i.e., 1.125° or ∼120 km, 60 vertical levels), while for the time period 1958–2001, the TWV data from the ECMWF reanalysis and the ERA-40 dataset are available. The horizontal resolution of the ERA-40 model is also set to T159 spectral resolution with 60 levels in the vertical. The reanalysis system uses a 3D variational assimilation system. A summary and discussion of the observations available at different times during the 40-yr reanalysis can be found online (http://www.ecmwf.int/research/era/). The data going into the reanalysis have changed markedly over this period. There are three important epochs in ERA-40: 1958–72, which is the presatellite era; 1973–86, which begins with the assimilation of the radiances from the first satellite infrared channels on the Vertical Temperature Profiler Radiometer (VTPR) and, after late 1978, includes infrared and microwave sounder data from the TOVS suite of instruments; and 1987–2001, which includes the addition of information on radiances from the satellite microwave channels of the SSM/I to the atmospheric water vapor assimilation over the ocean (Gerard and Saunders 1999). Note that neither the ECMWF reanalysis ERA-40 nor the ECMWF operational analysis assimilate AMSU-B data over sea ice (N. Bormann, ECMWF, 2008, personal communication); therefore, the AMSU-B-derived TWV data are independent of the TWV data from ECMWF.
b. RCM simulations
The RCM employed in this study is the HIRHAM model (Christensen et al. 1996; Dethloff et al. 1996), which has already been applied for various Arctic climate studies. It has been shown that the model reproduces the Arctic climate and its variability quite well (e.g., Rinke et al. 1999, 2006, and references therein). The physical parameterizations in the model are those from ECHAM4 (Roeckner et al. 1996). The integration domain covers the whole Arctic, that is, the area north of ∼60°N. The model is configured at 0.5° (∼50 km) horizontal resolution and has 19 vertical levels. For this study, HIRHAM has been forced at its boundaries by ECMWF analysis data.
Two HIRHAM simulations have been performed: a 4-yr-long run over the time period January 2000–December 2003 and a 45-yr-long run over the time period September 1957–August 2002. The 4-yr-long run was driven by operational ECMWF analyses. The years 2000–03 were selected in order to compare with the AMSU-B data, which have recently become available starting with the year 2000. Daily, monthly, and seasonal mean TWV data have been investigated. The 45-yr-long run was driven by ERA-40 data and was performed in order to assess the simulated climatology of TWV by comparing it with the ERA-40 data, and to calculate the temporal trends in TWV. Monthly, seasonal, and annual mean TWV data have been investigated for this purpose. Due to the incompleteness of the data during the years 1957 and 2002 in ERA-40, the means over 44 yr from 1958 to 2001 are presented.
For a quantitative comparison of the HIRHAM simulations with the observational data, the latter have been interpolated onto the HIRHAM grid. Standard error measures have been calculated: mean bias, root-mean-square error, and pattern correlation coefficient, defined as (x′ is the deviation of the spatial mean over all model grid points M = 11 000; subscripts a and b indicate the correlated variables).
a. Evaluation of TWV for the period 2000–03
Our primary objective is to evaluate the simulated HIRHAM TWV. The comparison of HIRHAM simulations with observations during these specific 4 yr is focused on the comparison with the AMSU-B-derived data, which recently became available. However, to give a broader presentation and to make a cross-check of other available observational datasets, a qualitative comparison with both the TOVS and ECMWF OD data is included as well.
1) Comparison of monthly mean TWV
For all months during the 4 yr, the HIRHAM-simulated TWV has been quantitatively compared with the AMSU-B-derived TWV data. For this comparison, Fig. 1 shows the mean bias (“HIRHAM minus AMSU-B”), the root-mean-square error, and the pattern correlation coefficient between the monthly HIRHAM and AMSU-B data of TWV. Figure 1 shows that HIRHAM simulates generally lower TWV values than those derived from AMSU-B. During most of the year, the bias is small and is on the order of −0.5 kg m−2. Analogously, the rms error is less than 1 kg m−2 for most of the year. A comparison of the AMSU-B data with those measured by radiosondes during a Polarstern cruise (ARK XIX-1, spring 2003) shows a similar bias in the AMSU-B TWV; that is, it is higher than the radiosonde-derived TWV data (Melsheimer and Heygster 2008). Furthermore, from Fig. 1, a clear seasonal dependence of the mean bias and rms error can be recognized. The largest bias (−1 to −3 kg m−2) and the largest rms error (1.5–3.5 kg m−2) occur during July–October. These larger biases can be attributed to the reduced accuracy of the AMSU-B TWV data during this time of year. As mentioned above, the TWV algorithm works independently of the surface emissivity for TWV values of up to about 7 kg m−2. This condition is fulfilled over most of the Arctic in late autumn, winter, and spring. For TWV values above 7 kg m−2, which is typical for much of the Arctic in summer, the extended algorithm [see section 2a(1)] must be used that includes the ratio of the surface emissivities at 90 and 150 GHz, which is much less accurate than the algorithm for lower TWV values. The needed surface emissivity information was derived from data from the SEPOR/POLEX measurement campaign, which took place over the Norwegian Sea and the Arctic Ocean in winter; there has been no comparable campaign in summer. Deviations in the actual sea ice emissivity ratio from the ratio found during SEPOR/POLEX will cause errors in the TWV retrieval. In late summer and early autumn, after substantial parts of the sea ice have undergone surface melting and refreezing, the behavior of sea ice emissivity can be expected to deviate most from the behavior measured during the SEPOR/POLEX campaign and from that used by the algorithm. In winter, when the emissivity-independent algorithm is used, there is no such effect. Thus, the mean bias and the root-mean-square difference between the TWV computed from HIRHAM and that derived from AMSU-B are much higher in summer and early autumn. Recent advances in satellite retrieval and in the modeling of sea ice emissivities (Mathew et al. 2008) might lead to improvements in the situation in the future. Nevertheless, Fig. 1 further shows that the calculated pattern correlation coefficients indicate a high correlation (r = 0.7–0.95) for almost all months; that is, the HIRHAM-simulated spatial TWV patterns show a substantial relationship with the AMSU-B-derived spatial patterns. Unlike the bias, the mean spatial patterns are in good agreement, also during summer–early autumn. To explore the extent to which the HIRHAM biases contribute to the differences, the rms error (RMSE) in TWV between the HIRHAM simulation and the ECMWF data has been calculated. Figure 1 (middle panel) shows indeed that the simulated TWV values differ from the ECMWF data (RMSE = 0.2–2.5 kg m−2). However, the figure clearly indicates that the above-described bias in AMSU-B-derived TWV data in summer is real. In the other months, the differences between HIRHAM and AMSU-B TWV are attributed to biases in both the HIRHAM and AMSU-B TWV data, relative to the ECMWF TWV.
The comparison of Wei et al. (2002), who evaluated the fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5) for a 1-yr simulation (October 1985–September 1986), showed that the annual cycle of their simulated monthly TWV compares well with the TOVS and NCEP–NCAR results. The simulation overestimates the TOVS-derived TWV data in winter and spring by 0.2–0.5 kg m−2 (see their Fig. 9), particularly over the Arctic Ocean and the North American Arctic Watershed.
In the following (Fig. 2), HIRHAM and AMSU-B monthly mean TWV geographical maps are compared, selecting, as an example, the months in the middle of the two extreme seasons (winter, summer), that is, January and July, for 1 yr (2003). Figures 2a and 2b are intended to demonstrate 1) the general tendency of HIRHAM to simulate systematically lower TWV values than those derived from AMSU-B, 2) the general agreement in the spatial patterns, and 3) the very different data amounts available from the retrieval algorithm in the different seasons. Figures 2a and 2b further show that the HIRHAM and ECMWF analyses agree well in both spatial pattern and magnitude for all seasons. During the critical summer season, both sets of satellite-derived TWV data (from AMSU-B and from TOVS) are subject to errors. AMSU-B-derived data tend to show too large values associated with the assumptions about the sea ice emissivity in the algorithm. TOVS-derived data tend to show too low values resulting from the difficulty of the algorithm in the presence of dense clouds. During the other seasons, both satellite-derived datasets show reasonable patterns. TOVS-derived data do not present TWV values over Greenland because of problems with cloud clearing over cold, white surfaces.
2) Comparison of daily mean TWV
The RMSE and pattern correlation coefficient between HIRHAM and AMSU-B-derived data of the daily TWV have been calculated for all days during the years 2000–03 (Fig. 3). The results consistently show a clear seasonal dependency through all of the years, with a smaller root-mean-square error (1–2 kg m−2) and larger pattern correlation coefficient (0.6–0.8) during the winter season (November–March) and a larger RMSE (4–7 kg m−2) and smaller pattern correlation coefficient (0.1–0.6) during the summer season (June–September). There is also a tendency that for the worst agreement to be found in September. There are two reasons for this: 1) the AMSU-B algorithm, as the sea ice emissivities used in the retrieval algorithm are not suitably appropriate during that time of year, as mentioned above, and 2) the HIRHAM simulations, where the simulated daily TWV significantly differs from that of ECMWF in summer (root-mean-square error of 4–5 kg m−2 in July and August; Fig. 3), associated with differences in the position and strength of cyclones on specific days (see below for specific case studies). Compared to the other seasons, the local surface forcing and therefore the limitations of the physical parameterizations like convection, clouds, and hydrological processes in the Arctic model are more pronounced in summer. Note also that when comparing daily means (Fig. 3), the root-mean-square error is considerably larger and the pattern correlation is considerably smaller than when comparing monthly means (Fig. 1). This emphasizes the strength of HIRHAM in climate measures, like monthly means.
To investigate further the details of the differences, examples of regional patterns at specific days are presented in the following. For this, the TWV patterns of HIRHAM, ECMWF, AMSU-B, and TOVS are given for selected cases in Figs. 4a–c. As the spatial patterns of TWV are linked to the patterns of the atmospheric circulation, illustrated here by the mean sea level pressure (SLP), Figs. 4a–c include the corresponding SLP patterns for those selected cases. The examples have been selected in order to 1) demonstrate the strong dynamic TWV variability on a daily time scale, linked with moisture transport by weather systems, and 2) show examples of both worse (case a) and better (cases b and c) levels of agreement among the different TWV patterns.
(a) Case of 26 September 2003 (Fig. 4a): On this day, the ECMWF TWV map shows the intrusion of two tongues of relatively humid air (10–20 kg m−2) from lower latitudes into the Arctic. One track is from the Labrador Sea into the Davis Strait, and the other is from the Atlantic, northern Europe, and western Russia via the Laptev Sea toward the Arctic. These tracks are attributed to the SLP field, which shows two low pressure systems (see isolines). One is located over the Labrador Peninsula and the other, a very deep cyclone, is located over the Arctic Ocean. They transport on their east flanks relatively warm and humid air northward. HIRHAM simulates these two tongues of humid air. However, differences are seen over the Laptev and Beaufort Seas that are associated with the weaker cyclone pattern, which is split into two branches. The AMSU-B-derived data reproduce the tongue of humid air over the Laptev Sea. The TOVS data are patchy, but also show an indication of this tongue of humid air. Again, the central Arctic is better covered by the AMSU-B-derived data.
(b) Case of 7 January 2002 (Fig. 4b): On this day, the ECMWF TWV map shows the very cold and dry air (0–2 kg m−2) over the Canadian Archipelago–Beaufort Sea and all of northern Eurasia. The latter is associated with the strong wintertime anticyclone over Siberia. Two areas of relatively humid air (8–15 kg m−2) are obvious. One is over the North Atlantic–northern Europe–Spitsbergen region and further extends across the central Arctic toward the East Siberian Sea, and which can be attributed to the transport of humid air on the eastern flank of the very deep and expanded cyclone located northeast of Greenland (see isolines). The other area of relatively high TWV is in the Gulf of Alaska and over northern Canada, associated with the low pressure system over that area that transports relatively warm and humid air from the open water northward. HIRHAM reproduces the SLP distribution and therefore the TWV pattern, although over Siberia the high pressure system, and therefore the area of cold, dry air, extends farther north. The AMSU-B-derived data clearly reproduce the tongue of relatively humid air across the Arctic, as well as the areas of dry air over Siberia and the Canadian Archipelago–Beaufort Sea. The TOVS data are very incomplete but show the dry air over Siberia. The coverage of the AMSU-B data includes that of TOVS and reaches much lower latitudes as compared to the September case in Fig. 4a.
(c) Case of 13 March 2002 (Fig. 4c): The ECMWF TWV map shows very dry air (0–3 kg m−2) over most of the Arctic, associated with the strong anticyclone with its center located over the Beaufort Sea but extending over the entire Arctic Ocean (see isolines). Therefore, the relatively warm and humid air from the open oceans (North Atlantic, Gulf of Alaska) cannot be transported toward the north. Over the Ural mountain area, relatively high TWV values are seen, associated with the small low pressure system located there. HIRHAM clearly reproduces the SLP and the TWV pattern. The AMSU-B-derived data also clearly reproduce the TWV pattern. The TOVS data show the relatively dry air over the Arctic Ocean but overestimate the moisture over the Laptev Sea and Bering Strait. The AMSU-B-derived data in this winter case show very good coverage (containing the TOVS coverage as a subset), similar to the case of 7 January 2002 (Fig. 4b). While the patterns of the very low TWV (<2 kg m−2) over Svalbard, Greenland, and the Bering Sea are positively reproduced, the moist air intrusion over the North Atlantic and the Ural Mountains can only be recognized from the shape of the data missing in the AMSU-B-derived maps.
Generally, both the HIRHAM and AMSU-B data show the regional pattern of dry air over the Greenland ice sheet.
b. Evaluation of TWV for a 44-yr-long climatology
In this section, the time period of 1958–2001 is investigated. The aim is to assess the simulated climatology of TWV by comparing it with the ERA-40 data, and to present temporal trends in Arctic TWV.
1) Climatological mean and interannual variability
Figure 5 shows the climatological mean (1958–2001) annual cycle of TWV, averaged over the Arctic (i.e., averaged over the whole model domain, the area north of ∼60°N), from the HIRHAM simulation and from ERA-40 data. TWV ranges from a maximum in July (15.4 kg m−2) to a minimum in winter (January–February: 2.8 kg m−2). The simulated TWV values tend to be lower than the ERA-40 data. The negative bias is maximal in July with 1 kg m−2, not significant in spring, and in winter it is 0.3 kg m−2. The common picture of the annual cycle is in agreement with earlier estimates, presented for much shorter time periods (Serreze et al. 1995; Groves and Francis 2002a).
To investigate the temporal and spatial TWV variabilities, it is useful to examine the geographical patterns of the simulated seasonal mean TWV (Fig. 6), as well as the corresponding patterns of the standard deviation (i.e., year-to-year variability) of seasonal TWV. As expected, generally, the regional patterns of seasonal mean TWV correspond to those of the temperature (not shown). During winter and the transition seasons, the highest values of TWV (up to 10 kg m−2) are found over the open water areas (North Atlantic, Gulf of Alaska) (Fig. 6). The comparatively abundant TWV largely reflects the ability of a warmer atmosphere to hold more water vapor. Over the Arctic Ocean and Greenland, both observations and simulations show in all seasons the minimum mean TWV value. Over the central Arctic Ocean, the TWV ranges from 1–2 kg m−2 in winter to 10–12 kg m−2 in summer. During summer, the large-scale TWV pattern is characterized by a rather zonally symmetric pattern according to the corresponding temperature distribution. However, on the regional scale, TWV shows pronounced features associated with the topographic forcing (smaller values over high elevation like the mountainous areas in Alaska and Siberia), which become apparent because of the high model resolution. These topographically forced regional TWV patterns are also clearly visible in the transition seasons. The geographical patterns of the year-to-year variability follow closely those of the mean (not shown). Over the central Arctic Ocean, the TWV variability ranges from 0–0.2 kg m−2 in winter to 0.5–1 kg m−2 during summer. The areas of maximum variability are associated with the synoptic activity. In winter and in the transition seasons, the North Atlantic storm track, which extends east through the Barents Sea into the Kara Sea and south of Baffin Island, becomes clearly visible. To evaluate the year-to-year variability in relation to the mean, Fig. 7 shows the ratio of the standard deviation to the mean of the seasonal TWV. Figure 7 demonstrates the weak spatial gradients in summer, when the mean is generally about 10–15 times the variability. In winter and in the transition seasons, the TWV variability is at its maximum, relative to the mean, along the sea ice margin (mean is about 4–7 times the standard deviation) according to its high year-to-year variability. Overall, the HIRHAM-simulated patterns for the mean and standard deviation (Figs. 6 and 7, color) are in agreement with those from the ERA-40 data (Figs. 6 and 7, isolines).
For a quantitative comparison of the simulated TWV with ERA-40 data, the root-mean-square errors, mean biases, and pattern correlation coefficients are given in Table 1, which shows that the RMSE for mean TWV ranges between 0.4 kg m−2 in winter and 1 kg m−2 in summer. The pattern correlation is very high (0.85–0.99) for all seasons. The same good agreement can be seen in the seasonal standard deviation. To summarize, the comparison demonstrates that the simulated mean TWV and its variability agree well with the ERA-40 data, and the model adds smaller-scale features due to the higher model resolution compared to the ERA-40 data (Figs. 6 and 7).
2) Decadal variability
As water vapor plays a major role in the climate, an understanding of its variability and change is important. However, Arctic TWV measurements are difficult and limited due to the region’s remoteness and the extreme nature of its climate. Therefore, the TWV changes have been assessed globally based on the ERA-40 data (Trenberth et al. 2005) and, here, specifically for the Arctic based on HIRHAM simulations at 50-km resolution and on ERA-40 data.
Maps of the seasonal TWV trends based on HIRHAM and on ERA-40 data for the whole 44-yr period of 1958–2001 are shown in Fig. 8. Although an increase in TWV is obvious in all seasons, regions with a negative TWV trend are also found. Moreover, Fig. 8 highlights the seasonally and regionally strongly different TWV trends. To estimate the significance of the trends, the trend–noise ratio, that is, the ratio of the trend to the standard deviation (as a measure of natural variability) has been calculated. Thus, significant trends, that is, grid points with a trend–noise ratio larger than 1.5 (2.0) (Schönwiese et al. 1993) are marked with gray (red) points, based on the ERA-40 data. Generally, the Arctic TWV variability is related to dynamic and thermodynamic influences. The spatial distribution of TWV is strongly coupled with temperature via the thermodynamic constraint of the Clausius–Clapeyron equation (temperature dependence of the saturation water vapor pressure). However, the spatial variability of Arctic TWV is dominated by changes in the large-scale dynamics, accompanied by changes in cyclone paths and/or intensity. Previous studies found a clear dependence of the Arctic moisture budget on the North Atlantic Oscillation–Arctic Oscillation (NAO–AO). Analyzing decadal differences in the moisture budget, Groves and Francis (2002b) found that over the Arctic, the AO index explains a significant amount of variability in the TWV fluxes, especially during winter. The strongest effects are evident over the Atlantic sector, where increased westerlies during the positive AO phase enhance the transport of moist air from the North Atlantic region into the European Arctic. Thus, the presented trends in Arctic TWV are associated with both the dynamics and thermodynamics (e.g., changes in air temperature, sea surface temperature, sea ice coverage). We find that the regional pattern of the TWV trend is closely connected with that of the corresponding 2-m air temperature trend (not shown). To give examples, in winter, the maximum TWV increase over the eastern Arctic is associated with the maximum warming over this region, while the strongest negative TWV trend over the eastern Siberian–Chukchi Seas area and southwest of Greenland is associated with the maximum cooling there. A negative winter TWV trend over those regions persists also for the last 14 yr (1988–2001; not shown), which correlates with the negative moisture convergence trend there, presented by Liu et al. (2007) and based on TOVS Path-P data. Figure 8 further shows that maximum positive TWV trends are calculated during the summer season, with values of 0.015–0.06 kg m−2 yr−1. Here, the trend over the western Arctic, particularly over Alaska and the Canadian archipelago is significant; that is, it exceeds the mean level of natural variability. The modeled trends of TWV show some deviations in the regional trends compared with the ERA-40 data but capture the general spatial patterns. Figure 9 displays the annual mean TWV time series over 1958–2001, first as averaged over the Arctic (i.e., the whole model domain) and, second, as averaged over the northeastern Atlantic. The latter area has been selected as it is an important source region for Arctic atmospheric moisture. It is seen that the simulation reproduces fairly well the decadal-scale TWV variability seen in the ERA-40 data. Linear trends have been calculated for all 44 yr (1958–2001) as well as for the last 24 yr (1978–2001). However, these trends have no statistical significance, except for the HIRHAM trend for the last 24 yr (Arctic, 0.016 kg m−2 yr−1; Atlantic, 0.022 kg m−2 yr−1) with a trend–noise ratio of 2. The upswing in TWV over the Atlantic area in recent years, which is particularly pronounced in the ERA-40 data, can be considered to be connected with the sea ice loss in the Barents and Greenland–Iceland–Norwegian (GIN) Seas (Zuidema and Joyce 2008). In the Barents and surrounding seas, where much of the recent winter sea ice loss has occurred, the thermodynamic coupling leads to an associated TWV increase, despite the relaxation of the NAO index to a more neutral state in recent years.
Long-term and high-resolution measurements of Arctic atmospheric properties are needed to observe and predict the regional climate changes. However, Arctic measurements are often difficult and limited due to the area’s remote and extreme location. This is particularly true for measurements of water vapor under the cold Arctic conditions because of the lack of sensitivity of the conventional water vapor radiometers to low water vapor amounts and the uncertainties of radiosondes. This points out the need to include various datasets in an attempt to cross validate the results. And, satellite-derived TWV data and RCM simulations can contribute toward filling data gaps and to better understanding the process involved.
Satellite-retrieved data of water vapor over the Arctic are often patchy, with large areas of missing data because of various limitations of the retrieval algorithms. This is most obvious for TOVS-derived TWV data on the daily temporal scale. AMSU-derived TWV data are much better in this regard; for example, they are able to present realistic TWV patterns over Greenland. To get the best (i.e., temporally and spatially most dense) observational dataset, one approach could be to combine the different satellite retrievals with all their strengths and weaknesses. However, their mutual biases have to be assessed first. The TWV data derived from AMSU-B that are now available allow for a comparison of model data with such observations on the time scale of days on a much broader base than was possible previously with the TOVS data alone. In particular, regions of extremely low TWV values such as the central Arctic and Greenland are poorly covered by TOVS retrievals (Fig. 4). The coverage of the Greenland ice sheet with TOVS retrievals is so poor that even monthly averages cannot be calculated (Fig. 2). Now, however, the model results of these regions can also be evaluated.
Daily variations in TWV are associated with the atmospheric circulation via the advection of moisture by low pressure systems. To resolve these variations, a certain horizontal resolution of the model and the observational data is required. HIRHAM and AMSU-B data have a resolution of 0.5° × 0.5°, which has been demonstrated to be adequate for this purpose. Hence, to evaluate spatial TWV variations on a daily time scale, HIRHAM simulations and AMSU-B-derived data are suitable.
The HIRHAM-simulated climatological spatial TWV patterns agree well with the ECMWF analysis data and the AMSU-B-derived data. The maximum root-mean-square error between HIRHAM and these data is found in summer and for monthly means of individual years is 1–2.5 kg m−2 and for long-term seasonal means it is 1 kg m−2. During this time of the year, the AMSU-B TWV data have a reduced accuracy. During the rest of the year, the TWV patterns agree well, with root-mean-square errors on the order of 0.5 kg m−2. In addition to the agreement between the simulated climatological seasonal mean TWV patterns and those of the ERA-40 data, the reproduction of the interannual and decadal-scale variabilities lends support to the quality of the HIRHAM simulations. Regional Arctic TWV trends are presented for the 44 yr (1958–2001) of this study. Significant positive trends are calculated in summer over the western Arctic and in spring over Alaska and parts of Russia, while significant negative trends are found in winter over the East Siberian Sea.
This research has been funded by the German Research Foundation (DFG). We thank I. Hebestadt and S. Erxleben for their help with the graphics, and the reviewers for their helpful comments.
Corresponding author address: Annette Rinke, Alfred Wegener Institute, Telegrafenberg A43, 14473 Potsdam, Germany. Email: email@example.com