1. Introduction

Surface snow accumulation is the primary mass input to the Antarctic ice sheets, and it has long been recognized as a potentially important contributor to global sea level change. Precipitation is the dominant term among various components of surface snow accumulation (precipitation, sublimation/vapor deposition, and snow drift) (Bromwich 1988) and its spatial and temporal variability are necessary information to assess the surface mass balance. In the last three decades, three measurement methods have been used to estimate the precipitation and accumulation over the Antarctic ice sheets: direct measurements, remote sensing techniques, and glaciological approaches. No reliable dataset of snow accumulation and precipitation measurements over Antarctica for a long time period is available from these methods for the following reasons. First, the introduction of drift snow creates the problem of distinguishing snow that has been precipitated from what has been picked up by the wind and transported. As a result, snow collection gauges, which are commonly used in middle latitudes and alpine regions, are difficult to employ in the Antarctic, and the direct gauge measurements are not representative of either the local accumulation or the precipitation received. Other direct measurement techniques such as snow stakes, and acoustic depth gauges, are replete with practical difficulties. Second, satellite altimetry measurements (e.g., Wingham et al. 1998) give only the change in surface height. Densification of near-surface snow can decrease the surface height without any snow accumulation occurring, thus complicating the interpretation of surface height changes. For other remote sensing measurements, current retrieval algorithms such as the outgoing longwave radiation–based techniques used by Xie and Arkin (1998) do not work well at high latitudes where ice and snow on the surface affect the detection of new precipitation. Algorithms employing microwave emission characteristics (Winebrenner et al. 2001) and satellite radar scatterometer signatures (Drinkwater et al. 2001) to determine accumulation rates are promising, but are still in the early stages of development. Third, due to costs and logistic difficulties, ice core and snow pit data are collected in few locations over Antarctica, and thus the accumulation estimation from glaciological approaches lacks an adequate and uniform temporal resolution for large areas of the continent. As a result, only long-term-averaged continent-wide maps of snow accumulation (e.g., Vaughan et al. 1999) are presently available, with the interannual variability and trends measured for just a few locations (e.g., Mosley-Thompson et al. 1995).

Fortunately, the surface measurements derived from these methods are not the only source of information, and atmospheric techniques can be used for direct simulation or indirect determination of precipitation and/or accumulation with better temporal and spatial resolution. Three atmospheric methods are available to estimate net accumulation: moisture flux budget calculations, numerical modeling, and dynamic retrieval. The method of moisture flux budgets uses atmospheric moisture transport toward the continent to calculate the area-integrated surface mass balance (Bromwich 1988, 1990). The spatial resolution of this approach is limited because the precipitation minus sublimation (P − E) residual is the small net result of large moisture fluxes into and out of the region of interest. Cullather et al. (1998) used moisture flux calculations to evaluate spatial and temporal variability of Antarctic precipitation from the European Centre for Medium-Range Weather Forecasts (ECMWF) operational analyses for 1985–94.

Atmospheric models can also be used to simulate various physical processes related to the Antarctic atmospheric circulation and to calculate the surface mass balance, which is determined by these processes. This method not only can provide information on the spatial and temporal distribution of the surface mass balance, but also can be used to gain insight into the mechanisms that cause mass balance variations and response to climate change. The advantage of using this approach is that the surface mass balance is obtained at high spatial resolution. Model output can be used to complement sparse measurements. In addition, the results might help to explain the observed surface mass balance distribution. The spatial and temporal variability of net snow accumulation over Antarctica from the ECMWF Re-analysis (ERA-15, 1979–93) have been presented by Turner et al. (1999). Genthon and Krinner (2001) compared net precipitation (precipitation minus sublimation) simulations from seven different general circulation models (GCMs) with the observation-based accumulation compilation from Vaughan et al. (1999), and summarized systematic biases present in those models.

The third atmospheric method, called “dynamic retrieval” uses the dynamic association among meteorological variables to retrieve precipitation values. Chen et al. (1997b) and Bromwich et al. (2001b) developed this method to derive the precipitation over Greenland from a vertical motion calculation, and it has proved to be a relatively economical method to retrieve the snow accumulation over Greenland with good spatial and temporal resolution. In comparison to atmospheric models, the dynamic retrieval method (DRM) has the advantage that no initialization and parameterization of complex physical processes are necessary in the retrieval process, and associated errors in parameterizations and model spinup are reduced. These advantages make the retrieval method attractive.

In this study, the precipitation for Antarctica from 1979 to 1999 has been retrieved by the DRM using ECMWF Tropical Ocean Global Atmosphere (TOGA) analyses (ECT) and ERA-15. Additionally, the Polar MM5, a regional atmospheric model based on the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5; Bromwich et al. 2001a; Cassano et al. 2001; Guo 2002; Guo et al. 2003), has been run over Antarctica for July 1996–June 1999, and the simulated precipitation is compared with the DRM precipitation. Also, to examine the spatial and temporal variability of modern precipitation over Antarctica, DRM-derived precipitation data are contrasted with simulated precipitation from ERA-15, ECT, and the National Centers for Environmental Prediction–Department of Energy (NCEP–DOE) Atmospheric Model Intercomparison Project-2 (AMIP-2) Reanalysis (NCEP2). This study is intended to demonstrate the reliability of the dynamic retrieval method and evaluate its ability in representing the spatial and temporal variability of Antarctic precipitation in comparison with Polar MM5 simulations and global analyses products. A comparison study evaluates the variability of West Antarctic precipitation with respect to the El Niño–Southern Oscillation (ENSO; Guo et al. 2004, hereafter GBH).

2. Methodology

In this section, a brief introduction is given on each of the methods used to evaluate precipitation: the DRM, Polar MM5, and atmospheric analyses and reanalyses data archives.

a. Dynamic retrieval method

A natural method for retrieving precipitation from the observed wind, temperature, geopotential height, and moisture fields is using an atmospheric model and a four-dimensional data assimilation (FDDA) system. Both NCEP and ECMWF analyses and reanalyses projects have produced global analyses of atmospheric fields, which includes the global recovery of precipitation. However, as complex parameterizations of physical processes are involved in atmospheric modeling and FDDA, the spatial resolution of the precipitation derived by the above reanalyses is low, and FDDA systems at high resolution are very computationally demanding. Bromwich et al. (1993) use a statistical diagnostic approach to calculate precipitation over Greenland from the advection of relative geostrophic vorticity at the 500-hPa level. Chen et al. (1997b) and Bromwich et al. (2001b) extend this work by using a generalized vertical motion equation to calculate precipitation, and the method can be used over mountainous regions with steep slopes, such as Greenland and Antarctica. As stated above, this method has the advantage that no initialization and parameterization of complex physical processes are necessary in the retrieval process.

Vertical motion can be calculated from geopotential height, temperature, and wind fields in global analyses and reanalyses data. The topographic effects on precipitation and atmospheric motion are greatly influenced by the computational accuracy of the horizontal pressure gradient force over mountainous regions (Colle et al. 1999), especially near the steep slopes of mountains and ice sheets. In order to improve the estimation of precipitation over steep slopes, the equivalent geopotential and geostreamfunction in σ coordinates are used in the generalized ω equation (Chen et al. 1997a; Bromwich et al. 2001b).

The equivalent geopotential height ϕ_e is derived from the solution of the following equation (Chen and Bromwich 1999):

View Expanded

where ϕ, T, and p∗ are geopotential height, temperature, and surface pressure in σ coordinates, respectively, and R is the gas constant for dry air.

If the model equation is solved in a limited region, the equivalent isobaric geopotential, and geopotential in σ coordinates can be separated into its inner part ϕ_ei, ϕ_i and harmonic part ϕ_eh, ϕ_h (Chen and Kuo 1992) as

ϕ_eϕ_ehϕ_eiϕϕ_hϕ_i(2)

Based on (1), the inner part of the equivalent isobaric geopotential height in σ coordinates ϕ_ei can be derived from the solution of the following Poisson equation:

View Expanded

with zero Dirichlet boundary value.

The vorticity and divergence equations can be transformed into the equations of the inner parts of the streamfunction and velocity potential, respectively, and they are expressed by

View Expanded

where E_i = m²(U² + V²)_i/2; ψ_i and χ_i are inner parts of the streamfunction and velocity potential, respectively; the terms ψ_adv,i and χ_adv,i are the variation rates of the inner parts of the streamfunction and velocity potential caused by advection; and m is the map scale factor.

The inner part of the ageostrophic geopotential ϕ_eia (the difference between ϕ_ei and geostrophic geopotential ϕ_eig = f₀ψ_i) can be written as (Chen et al. 1997a)

View Expanded

where m₀ is the average value of the map scale factor in the integration region: ϕ_e,had,i is the variation rate of the geopotential caused by advection and diabatic heating, and 𝗔 is a matrix comprised of T and σ with units of J kg⁻¹ (Chen et al. 1997a). Here

D_hDD_i(7)

where D_i and D_h are the inner and harmonic parts of the divergence, respectively (Chen and Kuo 1992).

If the tendency of ageostrophic geopotential in (6) is neglected, this approximation is referred to as a balanced ageostrophic approximation (Chen et al. 1996). Thus, Eq. (6) becomes

View Expanded

Equation (8) is a velocity potential form of the generalized ω equation for the balanced ageostrophic approximation in σ coordinates. In this equation, the diabatic and advection terms computed by the ageostrophic wind are the same as those in the generalized ω equation in p coordinates (Pauley and Nieman 1992), but the effect of orography on the vertical motion is much better described than that in p coordinates. If Eq. (8) is transformed into p coordinates and the term ϕ_e,had,ia is computed by the geostrophic wind and expressed by ϕ_e,had,ia,g, Eq. (8) becomes a velocity potential form of the quasigeostrophic ω equation. In order to reduce computational errors in the solution of (8), a harmonic-sine spectral method (Chen and Kuo 1992) is used. It is seen from the above that the equivalent geopotential is used to develop a generalized ω equation in σ coordinates, and this equation can be used to improve ω calculation over mountainous regions.

Using the continuity equation and vertical finite differencing, the pressure vertical velocity ω in σ coordinates can be calculated by

View Expanded

where 𝗜 is a unit matrix, 𝗖 is a lower-triangular matrix comprised of σ and its vertical difference (Chen et al. 1997a), and u, υ are the horizontal wind components; χ_i is derived from Eq. (8).

Only large-scale condensation is considered (Haltiner and Williams 1980, p. 309) in computing precipitation. The condensation rate can be expressed by

View Expanded

where condensation from saturated air is denoted by dq_s/dt, q_s is the saturation specific humidity, L is the latent heat of condensation, C_p is the specific heat of the air at constant pressure, and R_υ is the gas constant for moist air. The Kronecker delta δ is used to indicate the assumption that the condensation begins at some critical relative humidity r_c when upward vertical motion is present; r_c is a tunable parameter in the model. In this paper, the critical relative humidity r_c is assumed to be 94% based upon model tests over the Antarctic continent with different r_c values (the value was found to be relatively insensitive to elevation). It is noteworthy that precipitation over the Southern Ocean is underestimated by using this r_c value although the temporal variability is well captured, as discussed below.

If all of the condensate from saturated expansion is assumed to fall instantly as precipitation, the precipitation per unit area can be calculated from vertical integration of condensation rate dq_s/dt. The precipitation is calculated twice per day based on the ECT at 0000 and 1200 UTC for 1985–99 and four times a day based on ERA-15 at 0000, 0600, 1200, and 1800 UTC for the years 1979–93. The retrieved precipitation is produced from the DRM with ECMWF reanalysis (1979–90) and operational analyses (1991–99). The annual precipitation is derived by adding daily precipitation amounts for the whole year. In this paper, 22 σ levels at σ = 0.995, 0.980, 0.955, 0.920, 0.875, 0.820, 0.750, 0.670, 0.590, 0.505, 0.420, 0.340, 0.270, 0.220, 0.180, 0.140, 0.110, 0.090, 0.070, 0.050, 0.030, and 0.010 are used in the vertical, and a 40-km grid size is used in the horizontal. The square model domain used in this study consists of 181 grid points in each orthogonal direction, centered at the South Pole, with a horizontal polar stereographic projection resolution of 40 km (Fig. 1). The model topography data over the Antarctic continent are also interpolated from a modern 5-km resolution digital elevation model (DEM) of Antarctica (Liu et al. 1999) using bilinear interpolation from the nearest points.

3. The spatial distribution of Antarctic precipitation

The most reliable source for identifying the spatial distribution of Antarctic precipitation comes from the long-term annual accumulation depiction synthesized from glaciological data. Figure 2 shows the most recent accumulation compilation over the Antarctic ice sheets from Vaughan et al. (1999). The surface snow accumulation is the net result of snowfall, sublimation/deposition, drift snow effects, and runoff, and its primary components are precipitation and sublimation. In this section, the annual precipitation minus sublimation simulated by the Polar MM5 is first compared with the accumulation analysis of Vaughan et al. (1999). With the surface wind fields from the model output, the impact of drift snow redistribution on snow accumulation is also assessed. Based on the good performance of Polar MM5 in representing P − E fields, all precipitation fields from DRM, NCEP2, and ECT are then compared with that simulated by Polar MM5.

Figure 3 shows the annual precipitation minus sublimation simulated by the Polar MM5. Sublimation is derived from the simulated latent heat flux. Compared to the accumulation analysis of Vaughan et al. (1999) all major features are reproduced by the Polar MM5. Both maps show large values along the coast of East and West Antarctica, and over the Antarctic Peninsula, and small amounts over the plateau of East Antarctica and around the Ross and Filchner/Ronne Ice Shelves, and Lambert Glacier. The spatial distribution of the model predicted P − E distribution is in good qualitative agreement with the accumulation analysis of Vaughan et al. (1999). Additionally, minimum values around Dronning Maud Land, Victoria Land, and Marie Byrd Land, and maximum values over the Antarctic Peninsula, southern part of the Ross Ice Shelf, and in the coastal region of the southeastern corner of the Bellingshausen Sea match that of Vaughan's analysis (see Fig. 1 for locations). Comparing these two maps, several differences also exist: First, areas enclosed by the 20 and 50 mm yr⁻¹ contour lines in the simulated field are larger than those shown by Vaughan et al. (1999) and the simulated P − E values in the enclosed areas are smaller than the climatologically depicted annual accumulation from Vaughan et al. (1999). Second, the area with small simulated P − E values around the southern part of the Lambert Glacier is much smaller than that from Vaughan et al. (1999). As discussed below, the divergence of drift snow transport due to the strong katabatic winds in this area partially contributes to the difference between the simulated P − E and observed accumulation. Third, although the accumulation peaks along the Antarctic Peninsula in the Polar MM5 output and Vaughan's analysis with a maximum around 2000 mm, the accumulation distributions differ in this area. It is believed that the coarse spatial resolution of the Polar MM5 accounts for this difference. These features also can be observed in Fig. 4 where the Vaughan et al. (1999) accumulation compilation (Fig. 2) is subtracted from the averaged P − E field derived from the Polar MM5 simulations (Fig. 3). Not surprisingly, the largest errors are associated with the coastal regions, and accumulation is underestimated in the central interior while it is generally overestimated in the coastal areas by Polar MM5. This is very similar to Genthon and Krinner's (2001) findings for their seven different GCMs. Cassano et al. (2001) found that Polar MM5 also overestimates the precipitation amounts along the steep slopes of the Greenland Ice Sheet. It is hypothesized that the deficiencies in the interior precipitation simulations reflect the limited ability of Polar MM5 in representing clear sky precipitation. Studies show that clear sky precipitation is continuously formed in the interior of Antarctica without any organized synoptic-scale process, and comprises a large fraction of the total precipitation (Bromwich 1988). The time series of the precipitation events in MM5 over all interior sites appears to be episodic (not shown). This suggests that the simulated deficit accumulation reflects the limited ability of Polar MM5 in representing clear sky precipitation. The annual cycle of accumulation derived from Polar MM5 simulations also supports the above hypothesis. Figure 5 shows the annual cycle of accumulation derived from Polar MM5 simulations and moisture flux budget calculations using ECMWF analysis for the Antarctic continent (Cullather et al. 1998, for 1985–94) above various elevation intervals. A broad peak of accumulation is found during March–October for the three elevation bands examined in moisture flux budget calculations. This peak in the interior of Antarctica is believed to result from the increased clear sky precipitation during the polar night when moisture saturation is enhanced by the strong radiative cooling (Bromwich 1988), and it does not appear in the Polar MM5 simulations for accumulation above 2500 m.

To isolate the effect of sublimation on the comparison of precipitation and accumulation from different datasets, Fig. 6 shows the annual fraction sublimation/precipitation (SU/PR) from the Polar MM5 output. The mean annual sublimation field from Polar MM5 simulations is very close to that estimated by van den Broeke (1997). Similar to van den Broeke (1997), four areas clearly stand out where sublimation on an annual basis removes more than 70% of the precipitation: east Dronning Maud Land, the Lambert Glacier basin, Victoria Land, and the southern part of the Ross Ice Shelf. Averaged over the continent, the model predicts that 13.5% of the annual precipitation over Antarctica is removed by sublimation. Over most areas in the interior of the Antarctic ice sheets, SU/PR is less than 20%. Similar to Genthon and Krinner's (2001) findings, deposition (or inverse sublimation) also contributes to the surface accumulation, and the contribution is not negligible. In addition, areas with a large contribution of sublimation removal are not associated with systematic underestimation of surface accumulation, and areas with small contribution of sublimation removal are not associated with systematic overestimation of surface accumulation. Therefore, the deficiencies found in simulated accumulation are more likely attributed to deficiencies in precipitation rather than sublimation.

Katabatic winds are prominent climatological features of the Antarctic boundary layer. The associated drift snow effects are of importance for the redistribution of the snowfall, and for modification of the local net accumulation by direct removal of snow across the coastline and by enhancing evaporative loss from the surface. The Byrd snow drift project (Budd et al. 1966) executed during 1962–63 found that the observed drift snow transport in the layer from 1 mm to 300 m above the snow surface, Q³⁰⁰_10⁻³ (units: g m⁻¹ s⁻¹), can be represented by the relation

Q³⁰⁰_10⁻³V₁₀(11)

where V₁₀ is the wind velocity at 10 m above the snow surface. The snow transport follows the wind direction. Thus, the drift snow effects can be expressed as the divergence or convergence of the snow transport vectors:

View Expanded

Using the above equation with wind field output from Polar MM5 simulations at 3-h intervals, drift snow effects can be estimated. Figure 7 shows annual mean divergence of drift snow transport estimated from the surface wind fields simulated by Polar MM5 for July 1996–June 1999. The dashed lines are elevation contours. The results show that there are complex divergence and convergence patterns of drift snow transport over Antarctica, especially along the coast. From this figure several features can be observed. First, areas with large drift snow transport convergence and divergence are located around escarpment areas where there is large katabatic wind acceleration. Second, drift snow transport generally diverges over most areas of East and West Antarctica with relatively small amounts. Third, areas with large snow transport divergence are generally accompanied by areas with large snow transport convergence nearby, and this indicates that the drift snow transport is of local importance for redistribution of the snowfall. Finally, the southern part of the Lambert Glacier is one of the areas with the largest divergence of drift snow transport. The divergence of drift snow transport under strong katabatic winds partially contributes to the difference between the simulated P − E and observed accumulation over this area. It is noted that this method neglects the enhancement of sublimation of snow when it is airborne (Gallee 1998; Bintanja and Reijmer 2001) and does not consider variations in the quantity of snow available for transport.

In Fig. 8, the Vaughan et al. (1999) accumulation compilation (Fig. 2) is subtracted from the averaged P − E − D field derived from the Polar MM5 simulations (Fig. 7). When compared to Fig. 4 [Polar MM5 P − E minus Vaughan et al. (1999)], this plot demonstrates that accounting for the drift transport with this method has little impact on the accumulation estimates. Not surprisingly, the largest errors in accumulation are associated with the coastal regions. As before, it is also observed that accumulation is underestimated in the central interior while it is generally overestimated in the coastal areas by Polar MM5.

Figure 9 shows the annual mean precipitation determined from the DRM using ECMWF analysis (DRM_ECT, 1985–99). All major features in the spatial distribution of Antarctic precipitation from the Polar MM5 simulations mentioned above are well captured by the DRM. These features include the maximum values over the Antarctic Peninsula and a large amount of precipitation on the southern coast of the Bellingshausen Sea. Over Marie Byrd Land, the greatest precipitation occurs in the immediate coastal region and rapidly decreases inland. The primary difference is the deficient precipitation over the ocean derived from the DRM compared to that simulated by Polar MM5. Studies show that the dependence of effective precipitation on the relative humidity is spatially complex (Sinclair 1994). The fixed value of r_c is optimized for use over the Antarctic continent and is set too high for use over the ocean (contributing to the underestimation of precipitation here). Also, only large-scale condensation is considered by the DRM in computing precipitation; the omission of mesoscale convection and associated precipitation over the ocean contributes to the difference. The Vaughan et al. (1999) accumulation compilation (Fig. 2) is subtracted from the average P field derived from the DRM (Fig. 9) in Fig. 10. As with Polar MM5, a similar problem in reproducing clear sky precipitation exists in the DRM, and a new scheme is currently being developed to represent clear sky precipitation events.

Figures 11 and 12 show the annual mean forecast precipitation from NCEP2 (1979–99) and ECT (1991–99), respectively. All modeled values from Polar MM5, DRM, ECT, and NCEP2 agree with each other in representing the general spatial features of the modern Antarctic precipitation. Because Polar MM5 and DRM use resolutions of 60 and 40 km, respectively, they resolve more mesoscale precipitation features. In particular, NCEP2 shows a smoothed version of the annual-mean precipitation distribution around the coast of the Bellingshausen Sea and Marie Byrd Land. While all datasets show a reasonable spatial distribution of Antarctic precipitation, it is also noted that they tend to underestimate the precipitation in the interior of Antarctica. Table 1 (following Reijmer et al. 2002) presents the annual mean precipitation from ice core measurements and model outputs at six locations with high elevation (Fig. 1). It shows that most modeled precipitation values are much less than their measured counterparts (around 30%–50% if sublimation is taken into account). This suggests that simulation of the moist physical processes involved in Antarctic precipitation needs further improvement.

Table 2 compares the average annual precipitation and accumulation for all of Antarctica (the Antarctic continent, Antarctic Peninsula, and floating shelves) from modeled values with the results reported by other investigators. The mean annual precipitation for the whole Antarctica, as computed from the DRM with ERA-15 and ECT data, is 179 and 195 mm yr⁻¹, respectively. They are close to the forecast precipitation from ERA-15, ECT, and NCEP2. Considering annual sublimation of 29 mm yr⁻¹ from Polar MM5 simulations, the mean annual mean net accumulation (P − E) is close to previously reported values (Cullather et al. 1998; Vaughan et al. 1999). However, mean annual precipitation and accumulation derived from the DRM and forecast precipitation from ERA-15, ECT, and NCEP2 is much less than that simulated by the high-resolution mesoscale model (Polar MM5) and a general circulation model (Ohmura et al. 1996). The mean annual precipitation for the whole Antarctica estimated by the Polar MM5 is 215 mm yr⁻¹. The sublimation on an annual basis removes about 13.5% of the precipitation, and the drift snow transport removes about 2.3% of the precipitation.

4. The temporal variability of Antarctic precipitation

Here, the temporal variability of Antarctic precipitation and the recent precipitation trends over the Antarctic ice sheets are presented. The temporal changes present in the precipitation time series produced by the DRM are contrasted with those derived from the analysis and reanalysis forecast products, and they are also compared with available ice core accumulation measurements. The ENSO signal in modeled Antarctic precipitation is evaluated in the second part of this work (GBH).

5. Conclusions

Atmospheric numerical simulation and dynamic retrieval method with atmospheric numerical analyses are used to assess the spatial and temporal variability of Antarctic precipitation for the last two decades. First, the Polar MM5, a version of the PSU–NCAR fifth-generation mesoscale model modified particularly for ice sheet applications has been run over Antarctica to study the Antarctic precipitation. With a horizontal resolution of 60 km, the Polar MM5 has been run for the period of July 1996 through June 1999 in a series of short-term forecasts from initial and boundary conditions provided by the ECMWF operational analyses. In comparison with climatological maps, the major features of the spatial distribution of Antarctic precipitation are well captured by the Polar MM5. Drift snow effects on redistribution of surface accumulation over Antarctica are also assessed with surface wind fields from Polar MM5 in this study. There are complex divergence and convergence patterns of drift snow transport over Antarctica, especially along the coast. It is found that areas with large drift snow transport convergence and divergence are located around escarpment areas where there is large katabatic wind acceleration. In addition, areas with large snow transport divergence are generally accompanied by areas with large snow transport convergence nearby, indicating that drift snow transport is of local importance for the redistribution of the snowfall.

The precipitation for Antarctica from 1979 to 1999 has been retrieved by a vertical motion calculation using ECMWF operational analyses (ECT, 1985–99) and reanalysis (ERA-15, 1979–93). The dynamic retrieval method is applied to retrieve the precipitation over Antarctica from 1979 to 1999 in this study. In comparison with glaciological estimates and Polar MM5 simulations, most major features in the spatial distribution of Antarctic accumulation are well captured by the dynamic retrieval results. In comparison with predicted values from ERA-15 (1979–93), ECT (1991–99), and NCEP–DOE AMIP-2 (NCEP2, 1979–99), dynamic retrieval calculations capture more mesoscale features of the precipitation distribution over Antarctica. In comparison with available ice core measurements from Dronning Maud Land the interannual variability of the Antarctic precipitation is reasonably represented by the dynamic retrieval results, having a significant correlation. However, in comparison with snow stake measurements from the South Pole the correlation is not significant. The DRM precipitation is quantitatively more accurate than the NCEP2 and ERA-15–ECT precipitation at both sites, following the measured accumulation more closely. A significant upward trend +1.3 to +1.7 mm yr⁻² for 1979–99 is found from retrieved and forecast Antarctic precipitation that is consistent with results reported by other investigators. This indicates that additional water is being extracted from the global ocean and locked up in the Antarctic ice sheets, about 0.05 mm yr⁻¹. While there is good agreement in this trend between all of the datasets, the interannual variability about the trend is not well captured on the continental scale. However, on the subcontinental scale, the interannual variability about the trend is well resolved for sectors in West Antarctica and the South Atlantic. It is also noted that the precipitation trend is weakly downward over much of the continent.