• Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and other fresh waters in the Arctic circulation. J. Geophys. Res.,94, 14485–14498.

  • Bryazgin, N. N., 1976: Yearly mean precipitation in the Arctic region accounting for measurement errors (in Russian). Proc. Arctic Ant. Res. Inst.,323, 40–74.

  • ——, and T. A. Shver, 1976: Atmospheric precipitation over the Arctic land watershed basin (in Russian). Proc. Arctic Ant. Res. Inst.,323, 75–86.

  • Dumenil, L., K. Isele, H.-J. Liebscher, U. Schroeder, M. Schumacher, and K. Wilke, 1993: Discharge data from 50 selected rivers for GCM validation. Max-Planck-Institut für Meteorologie/Global Runoff Data Centre Rep. 100, 61 pp. [Available from Max-Planck- Institut für Meteorologie, Bundestrasse 55, D-20146 Hamburg, Germany.].

  • Environment Canada, 1994: Annual review: 1994. Climate Perspective,16, 1–10.

  • Gates, W. L., 1992: AMIP: The Atmospheric Model Intercomparison Project. Bull. Amer. Meteor. Soc.,73, 1962–1970.

  • Golubev, V. S., and E. G. Bogdanova, 1996: Point measurements of solid precipitation. Proc. Workshop on the ACSYS Solid Precipitation Climatology Project, Reston, VA, World Climate Research Programme, 18–29.

  • Goodison, B. E., E. Elomaa, V. Golubev, T. Gunther, and B. Sevruk, 1994: WMO solid precipitation measurement intercomparison: Preliminary results. World Meteor. Org. TD-588, 15–20.

  • IPCC, 1996: Climate Change 1995: The Science of Climate Change. J. T. Houghton, L. G. Meira Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds., Cambridge University Press, 572 pp.

  • Ivanov, V. V., 1976: Freshwater balance of the Arctic Ocean. Proc. Arctic Ant. Res. Inst.,323, 138–147.

  • Jaeger, L., 1983: Monthly and areal patterns of mean global precipitation. Variations in the Global Water Budget, A. Street-Perrot, M. Beran, and R. Ratcliffe, Eds., D. Reidel, 129–140.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-year reanalysis project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Karl, T. R., P. Ya. Groisman, R. R. Heim Jr., and R. W. Knight, 1993:Recent variations of snow cover and snowfall in North America and their relation to precipitation and temperature variations. J. Climate,6, 1327–1344.

  • Khrol, V. P., Ed., 1996: Atlas of Water Balance of the Northern Polar Area. Gidrometeoizdat, 81 pp.

  • Korzun, V. I., Ed., 1978: World Water Balance and Water Resources of the Earth. UNESCO Press, 663 pp.

  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gage-corrected global precipitation. Int. J. Climatol.,10, 111–133.

  • Mysak, L., D. K. Manak, and R. F. Marsden, 1990: Sea ice anomalies observed in the Greenland and Labrador seas during 1901–1984 and their relation to an interdecadal climate cycle. Climate Dyn.,5, 111–133.

  • Oechel, W. C., S. J. Hastings, G. Vourlitis, M. Jenkins, G. Riechers, and N. Grulke, 1993: Recent changes in Arctic tundra ecosystems from a net carbon sink to a source. Nature,361, 520–523.

  • Peixoto, J. P., and A. H. Oort, 1983: The atmospheric branch of the hydrological cycle and climate. Variations in the Global Water Budget, A. Street-Perrot, M. Beran, and R. Ratcliffe, Eds., D. Reidel, 5–65.

  • Phillips, T. J., 1994: A summary documentation of the AMIP Models. Program for Climate Model Diagnosis and Intercomparison Rep. 18, Lawrence Livermore National Laboratory, UCRL-ID-116384, 343 pp. [Available from Lawrence Livermore National Lab., P.O. Box 808, L-264, Livermore, CA 94550.].

  • Serreze, M. C., R. G. Barry, and J. E. Walsh, 1995: Atmospheric water vapor characteristics at 70°N. J. Climate,8, 719–731.

  • Vowinckel, E., and S. Orvig, 1970: The climate of the North Polar Basin. World Survey of Climatology, Vol. 14, Climates of the Polar Regions, S. Orvig, Ed., Elsevier, 129–252.

  • WCRP, 1992: Scientific concept of the Arctic Climate System Study (ACSYS). World Climate Research Programme WCRP-72, WMO/TD-486, 89 pp. [Available from World Meteorological Organization, Case Postale No. 2300, CH-1211 Geneva 20, Switzerland.].

  • ——, 1994: Arctic Climate System Study (ACSYS): Initial implementation plan. World Climate Research Programme WCRP-85, WMO/TD-627, 66 pp. [Available from World Meteorological Organization, Case Postale No. 2300, CH-1211 Geneva 20, Switzerland.].

  • Zubenok, L. I., 1976: Evaporation over the land watershed basin of the Arctic Ocean. Proc. Arctic Ant. Res. Inst.,323, 87–100.

  • View in gallery

    Annual mean precipitation rates (mm day−1) in the Arctic from the analyses of Legates and Willmott (1990)—upper panel; Bryazgin (1976)—middle panel; and Jaeger (1983)—lower panel.

  • View in gallery

    Annually averaged zonal mean precipitation rates (mm day−1) from (a) the three observational sources in Fig. 1, and (b) the AMIP models included in this study.

  • View in gallery

    Mean seasonal cycles of area-averaged precipitation (mm day−1) for the ocean area poleward of 70°N: (a) observational estimates, and (b) AMIP model results relative to range of observational estimates (shaded area). Shown in (a) are the observational estimates of Vowinckel and Orvig (1970) for the polar cap including and excluding the Norwegian–Barents Seas.

  • View in gallery

    Annual mean rates of (a) precipitation, P, and (b) precipitation minus evaporation, P − E, for the Arctic Ocean as evaluated from observational estimates (bars at left) and from the AMIP models. Observational sources are B = Bryazgin, L = Legates and Willmott, J = Jaeger, and O = Vowinckel and Orvig (darker bar is for domain excluding Barents–Norwegian Seas).

  • View in gallery

    Regional delineation for results in Figs. 6–10. Asian Arctic watershed consists of regions III, IV, V, and VI; North American Arctic watershed consists of VII, VIII, and IX (including VIII-a and IX-a); European Arctic watershed consists of I and II; Kara Sea watershed is III; East Siberian watershed is V; Canada Straits–Foxe Basin watershed is VIII (including VIII-a). [Figure from Ivanov (1976).]

  • View in gallery

    Mean annual cycle of monthly precipitation (mm day−1) from 23 AMIP models and from observational estimates of Legates–Willmott and Jaeger (dashed lines without symbols). Results are shown in top row for the (a) Asian, (b) American, and (c) European watersheds of the Arctic Ocean; in bottom row for (d) Kara Sea watershed, (e) East-Siberian Sea watershed; and (f) Canada Straits and Foxe Basin watershed (cf. Fig. 5)

  • View in gallery

    Annual mean precipitation rates for the six regions of Fig. 6. Observationally derived estimates are shown at left of each panel.

  • View in gallery

    Mean biases of the ensemble of AMIP models relative to the observational estimates of precipitation (B = Bryazgin, L = Legates–Willmott, J = Jaeger, 01 and 02 = Vowinckel–Orvig with and without the Barents–Norwegian Seas) and precipitation minus evaporation (Bryazgin’s P minus Zubenok’s E).

  • View in gallery

    As in Fig. 6 but for evaporation E.

  • View in gallery

    As in Fig. 7 but for P − E. Observational estimates are areal means of Bryazgin’s P minus Zubenok’s E.

  • View in gallery

    AMIP model-derived P − E averaged over (a) Ob Basin and (b) Mackenzie Basin. Left-most bar of each panel is annual mean river discharge rate (mm3 yr−1) divided by corresponding area (mm2) of river basin.

  • View in gallery

    Annual cycles of model-derived P − E for (a) Kara Sea land watershed, (b) Laptev Sea land watershed and (c) Beaufort Sea land watershed. Also shown are annual cycles of discharge (equivalent mm day−1) of major rivers in each watershed: (a) Ob and Yenisey (total), (b) Lena, and (c) Mackenzie.

  • View in gallery

    Scatterplots of 10-yr (1979–1988) annual mean P vs annual mean E for the AMIP models. Included is the SUNYA “outlier.”

  • View in gallery

    Scatterplots of monthly precipitation vs monthly evaporation for the Mackenzie Basin from the AMIP simulations of 1979–88 by 3 models: CCC (left panels), GSFC (center panels), and ECMWF (right panels). Each panel contains results for one season (3 months × 10 yr = 30 data points).

  • View in gallery

    Monthly mean precipitation, evaporation, and P − E (asterisks) obtained by averaging NCEP reanalyses over all grid points in Mackenzie Basin. Results are for 17 yr, 1979–95.

  • View in gallery

    Time series of monthly P − E from NCEP reanalyses averaged over grid points in (a) Ob River drainage basin and (b) Mackenzie River drainage basin. Dashed spikes are measured monthly river discharges divided by area of river basin.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 86 86 11
PDF Downloads 58 58 12

Arctic Precipitation and Evaporation: Model Results and Observational Estimates

View More View Less
  • 1 Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois
  • | 2 Voeikov Main Geophysical Observatory, St. Petersburg, Russia
  • | 3 Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois
  • | 4 Voeikov Main Geophysical Observatory, St. Petersburg, Russia
© Get Permissions
Full access

Abstract

Observational estimates of precipitation and evaporation over the Arctic Ocean and its terrestrial watersheds are compared with corresponding values from the climate model simulations of the Atmospheric Model Intercomparison Project (AMIP). Estimates of Arctic regional mean precipitation from several observational sources show considerable scatter, and the observational estimates based on gauge-adjusted station data are considerably larger than the other observational estimates. While the AMIP model simulations of precipitation also show scatter, the ensemble mean of the models’ precipitation exceeds even the higher (gauge-adjusted) observational estimates over the Arctic Ocean and its major watersheds. The difference between simulated precipitation and evaporation (P − E), representing the net freshwater gain (runoff) by the surface, also exceeds the observational estimates by 44%–83% over the Arctic Ocean and by generally smaller percentages over the terrestrial watersheds. The ensemble model mean of the annual P − E exceeds the corresponding river discharges of the Ob and Mackenzie Rivers by 62% and 14%, respectively.

The simulated P and E are highly correlated across the AMIP models, and the interannual (as well as the seasonal) variations of P and E are highly correlated in the output of most of the individual models, implying a coupling of the regional P and E in the models. The only formulational feature found to be common to the high-P (and high-E) models is the use of a specified rather than a computed soil moisture. A preliminary examination of the reanalyses of the National Centers for Environmental Prediction shows that the differences between the reanalysis-derived P and E are closer to the observational estimates than are the AMIP estimates. However, the magnitudes of the reanalysis-derived P and E, individually, are higher than the corresponding observational estimates.

Corresponding author address: Dr. John E. Walsh, Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, 105 S. Gregory Avenue, Urbana, IL 61801-3070.

Email: walsh@uiatma.atmos.uiuc.edu

Abstract

Observational estimates of precipitation and evaporation over the Arctic Ocean and its terrestrial watersheds are compared with corresponding values from the climate model simulations of the Atmospheric Model Intercomparison Project (AMIP). Estimates of Arctic regional mean precipitation from several observational sources show considerable scatter, and the observational estimates based on gauge-adjusted station data are considerably larger than the other observational estimates. While the AMIP model simulations of precipitation also show scatter, the ensemble mean of the models’ precipitation exceeds even the higher (gauge-adjusted) observational estimates over the Arctic Ocean and its major watersheds. The difference between simulated precipitation and evaporation (P − E), representing the net freshwater gain (runoff) by the surface, also exceeds the observational estimates by 44%–83% over the Arctic Ocean and by generally smaller percentages over the terrestrial watersheds. The ensemble model mean of the annual P − E exceeds the corresponding river discharges of the Ob and Mackenzie Rivers by 62% and 14%, respectively.

The simulated P and E are highly correlated across the AMIP models, and the interannual (as well as the seasonal) variations of P and E are highly correlated in the output of most of the individual models, implying a coupling of the regional P and E in the models. The only formulational feature found to be common to the high-P (and high-E) models is the use of a specified rather than a computed soil moisture. A preliminary examination of the reanalyses of the National Centers for Environmental Prediction shows that the differences between the reanalysis-derived P and E are closer to the observational estimates than are the AMIP estimates. However, the magnitudes of the reanalysis-derived P and E, individually, are higher than the corresponding observational estimates.

Corresponding author address: Dr. John E. Walsh, Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, 105 S. Gregory Avenue, Urbana, IL 61801-3070.

Email: walsh@uiatma.atmos.uiuc.edu

1. Introduction

The freshwater budget of the Arctic has become an increasingly important consideration in the context of global climate change. Global climate models project substantial increases of temperature and precipitation in northern high latitudes as greenhouse gas concentrations increase (IPCC 1996). Fluxes of CO2 and CH4 from northern land areas to the atmosphere are highly sensitive to the dryness of the surface (Oechel at al. 1993). In addition, the freshwater budget of northern high latitudes may be linked to the intermittency of North Atlantic deep-water formation and the global thermohaline circulation (Aagaard and Carmack 1989; Mysak et al. 1990), which is a major determinant of global climate. The linkage of precipitation over Arctic terrestrial regions to oceanic deep-water formation stems from the fact that, relative to other oceans, the Arctic Ocean receives a disproportionately large ratio of runoff to precipitation. Although the Arctic Ocean contains only about 1.5% of the world ocean water, it receives about 10% of the total global runoff. This runoff contributes to the strong stratification of the upper Arctic Ocean and thus contributes to the stability of the sea ice that covers much of the Arctic Ocean throughout the year. Much of the sea ice and upper-layer freshwater is exported from the Arctic to the North Atlantic through Fram Strait (WCRP 1992, vii).

The Arctic freshwater budget is driven primarily by precipitation. The greenhouse-induced increase of Arctic precipitation projected by global climate models (IPCC 1996, 307–308) thus has broad implications for Arctic and perhaps global climate. The possibilities are even more intriguing in view of recent indications that precipitation has increased in northern high latitudes during the past several decades (Karl et al. 1993; IPCC 1996, Figs. 3.9 and 3.11). The annual mean precipitation averaged over Canada, for example, has exceeded the 1951–80 mean in 20 of the 22 years from 1973 to 1994 (Environment Canada 1994, 2). However, the surface station data used in compiling such statistics are questionable for several reasons. First, the network of surface stations is sparse over many northern land areas and nonexistent over the central Arctic Ocean. Second, the preferential siting of many of the stations in lower elevations and/or coastal regions adds to the difficulty of spatially extrapolating measurements from the stations. Finally, instrumental measurements of solid precipitation contain large uncertainties because of wind effects near the gauge orifice, differences in gauge types, the difficulty of distinguishing falling snow and blowing snow, the treatment of frequent trace amounts, and other difficulties in cold environments (Bryazgin 1976; Legates and Willmott 1990; Goodison et al. 1994; Golubev and Bogdanova 1996).

The difficulties inherent in the use of station data for determining areal mean precipitation add importance to the role of models in quantifying high-latitude precipitation. Models can be run with or without the assimilation of atmospheric data in simulations of climate and climate change. Ongoing reanalysis efforts are examples of the former strategy, while simulations of climate change using general circulation models are examples of the latter. The validity of simulations of climate change, however, will depend on the accuracy of the simulations of the present-day precipitation climatology by these models. An objective of this study is the evaluation of model-derived Arctic precipitation relative to observational climatologies.

Models offer additional advantages with regard to diagnoses of the freshwater budget of regions such as the Arctic. Moisture budget components such as precipitation, evaporation, atmospheric moisture flux convergence, and runoff are readily available on regular grids from model simulations, permitting the closure of the moisture budget over arbitrary temporal and spatial domains. Such closure is exceedingly difficult to achieve observationally in the absence of data on evaporation and, in some regions, runoff. (Note: In order to simplify the terminology, the word “evaporation” will be used for the surface-to-atmosphere moisture flux over the Arctic Ocean and northern land areas. Fluxes due to sublimation and transpiration are included in evaporation as used here.)

The following sections contain an assessment of model simulations of the Arctic hydrologic cycle. The emphasis will be on the simulated precipitation, although we also include evaporation in order to evaluate the net freshwater exchanges between the surface and the atmosphere and between the land surface and the ocean (by runoff). Recent syntheses of observational data make such an assessment more feasible than it would have been in recent decades, especially since at least two of the recent syntheses of precipitation data have included attempts to adjust the gauge measurements for known biases. Specific objectives of the present study are 1) to identify apparent biases in the model-simulated Arctic precipitation, thereby stimulating closer examination of model formulations of Arctic precipitation, and 2) to diagnose the linkage between precipitation and surface evaporation in the models. In order to perform the model assessment and diagnosis, we bring together the output from a suite of global (atmospheric) climate models, precipitation output from recent reanalyses derived from a global model run in a data assimilation mode, Arctic precipitation data from several sources, and river runoff data representing spatially integrated precipitation less evaporation.

2. Model output

A unique source of climate model output is the Atmospheric Model Intercomparison Project (AMIP), which was organized in 1990 by the Working Group on Numerical Experimentation of the World Climate Research Programme in an attempt to promote systematic evaluations and comparisons of atmospheric general circulation models (AGCMs). Approximately 35 groups from the international modeling community, as well as about 20 other groups performing diagnostic subprojects, are participating in the systematic intercomparison of AGCMs run under common initial and boundary conditions (Gates 1992). The output from 24 AMIP model simulations was used in this study. The models are listed in Table 1. (Output from a 25th model was found to differ so sufficiently from the others that it was regarded as an “outlier,” as shown in section 4.) The models used here are those for which the AMIP standard output was available in early 1995. In some cases, the modeling groups have subsequently implemented model versions that are more current than those used to produce the AMIP standard output.

Because the ocean surface temperatures and sea ice boundary conditions are prescribed in the AMIP simulations, the models do not include ocean mixed layers or ocean dynamics. The models determine their own surface temperatures and snow cover over land areas, where each model has its own specification of surface properties such as albedo, roughness, vegetation, and emissivity. The formulations of physical processes (including cloud formation, precipitation, and evaporation) vary from model to model (Table 2), as do the horizontal resolution, number of levels, and the choice of spectral or gridpoint representations of the model dynamics in the horizontal (Table 1). A comprehensive summary of the structure and parameterizations in the various models is provided by Phillips (1994). The AMIP models were run for a 10-yr period, 1979–88, with the ocean boundary conditions varying temporally in accordance with the observed fields from this period. While the prescribed ocean boundary conditions influence the computed evaporation rates to some extent, the model- determined surface winds and air temperatures introduce several degrees of freedom into the evaporation rates computed over ocean areas.

3. Observational data

Several datasets provide the primary observational input to the intercomparison. The first is the gridded database of Legates and Willmott (1990), who adjusted and gridded a global dataset containing monthly precipitation from 22635 land stations and 2233 ocean points. These records were subjected to correction procedures to remove systematic errors attributable to wind, evaporation from the gauge, wetting of the interior walls of the gauge, and the cumulative effect of trace amounts. The adjusted values for each calendar month were then interpolated by Legates and Willmott to a 0.5° × 0.5° latitude–longitude grid.

The second primary source of precipitation data was a Russian set of monthly maps for the Arctic region. Earlier versions of this dataset were described by Bryazgin (1976) and Bryazgin and Shver (1976); a more recent summary is provided by Khrol (1996). The station data used in compiling the Bryazgin dataset were subjected to a variety of corrections, most of which were for the cold season. The annual mean precipitation from this dataset exceeds that from the Legates and Willmott dataset over most of the Arctic, as will be shown in section 4. For the area poleward of 70°N, Bryazgin’s monthly fields have recently been digitized onto a 2.5° × 2.5° latitude–longitude grid at the Main Geophysical Observatory in St. Petersburg. The annual (not monthly) mean fields of precipitation over the Arctic terrestrial watersheds are contained in Bryazgin and Shver (1976).

Additional data on precipitation include the monthly fields of Jaeger (1983) and the area-averaged Arctic Ocean precipitation estimates of Vowinckel and Orvig (1970). The former dataset is based on a consolidation of data from various sources into 2592 latitude–longitude sectors; the Arctic Ocean data came from the six volumes of Marine Climatic Atlas of the World published by the U.S. Naval Weather Service between 1955 and 1965. The latter are based on measurements at drifting ice stations. (It should be noted that all the observational datasets used here include data for years preceding the 1979–88 AMIP period. If secular trends are present in the observational amounts, they may account for small discrepancies between the simulated and observed means.)

Estimates of evaporation, E, over the Arctic Ocean were obtained from the World Water Balance (Korzun 1978, Table 179), hereafter denoted as WWB. Additionally, values of E for the Arctic terrestrial regions were obtained from Zubenok (1976), whose estimates were based on a parameterized soil moisture that varies temporally in response to observationally specified precipitation, radiative fluxes, air temperature, and humidity. River discharge data, used to estimate runoff from several Arctic river basins, were also obtained from the world water balance (WWB) (Korzun 1978) and from Dumenil et al. (1993).

4. Results

Figure 1 shows the annual mean fields of Arctic precipitation from the datasets of (a) Legates and Willmott (1990), (b) Bryazgin (1976), and (c) Jaeger (1983). All three fields contain minima in the central Arctic Ocean. The magnitudes of the Legates–Willmott and Bryazgin minima are slightly larger than the annual mean values of 0.33–0.40 mm day−1 (12–15 cm yr−1) from ice stations in the central Arctic as reported by Vowinckel and Orvig (1970, 192, 226–232), while the minimum of 0.20–0.25 mm day−1 (7.5–9.0 cm yr−1) in the Jaeger field is the smallest. Other differences among the three climatologies in Fig. 1 are Jaeger’s smaller values (relative to the others) between Greenland and Svalbard, and the local minimum near 85°N, 75°E in the Legates–Willmott field. In spite of these differences, the spatial pattern correlations between the fields in Fig. 1 exceed 0.8 in all cases.

The observational estimates are compared with the precipitation from the AMIP models in Fig. 2, which shows the annual zonal mean precipitation amounts. While the models capture the decrease of precipitation with latitude, it is apparent that the models generally simulate more precipitation than is indicated by the observations, especially in the central Arctic. The oversimulation by many of the models is substantial. At 80°N, for example, most of the models simulate 0.60 mm day−1 or more, while the observational estimates are generally less than 0.50 mm day−1. The simulated amounts from several of the models are 50%–100% greater than the observational estimates in the 50°–70°N latitude zone.

Figure 3 shows the annual cycles of mean monthly precipitation for the Arctic Ocean (70°–90°N) as simulated by the AMIP models, together with the range of observational estimates from Legates and Willmott (Fig. 1a), Bryazgin (Fig. 1b), Jaeger (Fig. 1c), and Vowinckel and Orvig (1970). (Two sets of values from Vowinckel and Orvig are shown: one set for the Arctic Basin excluding the Barents and Norwegian Seas and one set for the Arctic Basin including those seas. The former contains generally low values relative to the other observational estimates because the peripheral seas are regions of relatively high precipitation.) In spite of the large range of uncertainty in the observational estimates, the oversimulation of precipitation by most of the models is apparent. The data and most of the models do show generally similar seasonal cycles, with maxima in late summer (July–September) and minima in the spring (April–May). Since this seasonal cycle is consistent with the rawinsonde-derived seasonal cycle of moisture flux convergence across 70°N (Serreze et al. 1995), it is an Arctic precipitation characteristic in which one may place confidence.

The model intercomparison is presented more succinctly in Fig. 4, which shows the decadal (1979–88) annual mean precipitation and P − E from the AMIP models. Also shown in Fig. 4a are the corresponding observational estimates of P of Legates, Bryazgin, Jaeger, and Vowinckel–Orvig. (The latter is denoted by“O” and is again presented as estimates for each of the two areal domains.) Every one of the models simulates excessive precipitation relative to the Jaeger and the lower Orvig estimates. The biases are also generally (but not always) positive relative to Bryazgin (ensemble mean bias of models is +16%), Legates (+26%) and the upper Orvig estimate (+27%). Aside from the “outlier” model of SUNYA, the wettest models are CCC, UIUC, MPI and NCAR, while the least precipitation is simulated by the JMA and UGAMP models. On the basis of the values plotted in Fig. 4b, the models’ ensemble mean biases of P − E are +44% relative to the Legates (P) − WWB (E) difference and +83% relative to the Jaeger − WWB difference.

While the monthly precipitation simulated for the Arctic Ocean differs by a factor of 2 among the various AMIP models (Fig. 3), the precipitation simulated for smaller subregions (Fig. 5) of the Arctic shows even greater variability. Figure 6 shows the seasonal cycles of monthly mean AMIP precipitation amounts averaged over six subregions, which include the Arctic drainage basins of Asia, North America (excluding Greenland), and Europe, as well as the terrestrial watersheds of several peripheral seas (Kara Sea, east Siberian Sea, Canada Strait–Foxe Basin). Also shown in Fig. 6 are the corresponding observational estimates from Legates–Willmott and Jaeger. It is apparent from the figure that the AMIP model amounts are again generally larger than the observational estimates. The models oversimulate the observational amounts by 50% in some regions and in some months. The largest differences generally appear in the summer, when the amounts themselves are largest; the seasonality is greater in many of the models than in the observational datasets. Regionally, the greatest excesses in the model precipitation appear in the Asian Arctic watershed, particularly in the east Siberian Sea watershed (Figs. 6a,e).

The apparent model-data discrepancies in Fig. 6 are, to some extent, reconciled in Fig. 7, which shows the 10-yr mean precipitation amounts for the same regions. Figure 7 includes the gauge-corrected observational estimates from Bryazgin–Shver (leftmost bar of each panel) as well as the observational estimates from Legates–Willmott and Jaeger, together with the simulated amounts from the AMIP models. For regions Fig. 7a–d, the use of the Bryazgin–Shver gauge-corrected data has raised the observational estimates to a level closer to the ensemble means of the models.

As shown in Fig. 8, the biases of the models’ mean P relative to the Bryazgin–Shver observational estimates range from 16% over the Kara region (and the Arctic Ocean) to +56% over the Canada Strait–Foxe Basin. All the biases of P relative to the Bryazgin–Shver estimates are positive but smaller than those relative to the other observational estimates. The next smallest biases of the ensemble means of the model are those relative to the Legates estimates, which, like the Bryazgin–Shver estimates, are based on gauge-corrected station data. For region (Fig. 7c), the relatively small European watershed of the Arctic Ocean, the Bryazgin–Shver gauge-corrected observational estimate is larger than the mean of the model amounts. (However, the small size of the European Arctic watershed makes the areally averaged P highly sensitive to a precise choice of a gridpoint mask and hence to the resolution of the gridded fields.) Figures 7 and 8 underscore the importance of gauge corrections of high-latitude precipitation data, especially if such data are to be used in comparisons with model-derived precipitation amounts that are unaffected by measurement errors.

The annual cycles of evaporation simulated by the AMIP models for the same six regions are shown in Fig. 9. The simulated evaporation shows a strong seasonality, ranging from 0.1–0.5 mm day−1 in the winter to 3–5 mm day−1 in the summer. (The winter amounts represent primarily sublimation.) While the range of E among models is comparable during summer to the range of P, all models show quantitatively similar seasonalities. Observational data on evaporation are inadequate for the inclusion of the corresponding observational curves in Fig. 9.

The difference between precipitation and evaporation, P − E, represents the net input of freshwater to the surface on a local or regional basis. Over land areas, P − E is equal to the runoff if there are no long-term changes in soil moisture storage. Figure 10 shows the annual mean P − E for the AMIP decade as simulated by the suite of models for the various watersheds defined earlier. Observational estimates (the Bryazgin–Shver’s regional mean P minus Zubenok’s regional mean E) are also shown at the left of each bar plot. The annual mean P − E is positive for all models and all regions; averaged over the year, the typical values of 0.5 mm day−1 represent a net freshwater gain of 18.2 cm yr−1. On a regional basis, P − E for the small European watershed is generally greater than for the Asian and North American Arctic watersheds in the AMIP models; the observational estimates are qualitatively consistent with this result, although the geographical differences are greater in the observational estimates. Within the Asian Arctic, the model-derived P − E for the east Siberian watershed exceeds the observational estimate in every case; the models’ P − E are more than twice as large as the observational estimates in several cases, representing a discrepancy of approximately 20 cm yr−1 of freshwater runoff to the east Siberian Sea—an important region of net sea-ice production for the Arctic Ocean (WCRP 1992). As summarized in Fig. 8, the largest positive biases of the models’ P − E are in the east Siberian and Canada Strait–Foxe Basin regions. The biases of approximately 60% in these regions indicate that the AMIP results are characterized by an oversimulation of P and/or undersimulation of E. Figure 8 implies that the former is true.

In order to address more systematically the linkage between P and E, we focus on two terrestrial regions: the Mackenzie watershed in northern Canada and the Ob watershed in Russia. (The Ob watershed used here includes the Irtysh watershed.) These river basins were chosen because they have long-term databases of runoff (discharge), which is an accumulated measure of the difference between the areally averaged P and E when groundwater storage does not change (Korzun 1978; Peixoto and Oort 1983). In this respect, the individual river basins provide opportunities for validation of the areally integrated P − E without the problems in areal integrations of station-measured precipitation, precipitation gauge corrections, or parameterizations of evaporation. [Annual discharge estimates from stream gauges are generally considered to be accurate to within 5%–10%, WCRP (1994, 43).] Mean annual discharges for the Ob and Mackenzie Basins, for example, are 395 km3 and 350 km3, respectively (Korzun 1978). Given the drainage areas of 2.99 × 106 km2 (Ob) and 1.80 × 106 km2 (Mackenzie), the annual runoff amounts are equivalent to area-averaged P − E of 13.2 cm (Ob) and 19.4 cm (Mackenzie). The corresponding annual mean rates of P − E are 0.36 mm day−1 (Ob) and 0.53 mm day−1 (Mackenzie). Figure 11 compares these totals with the annual mean P − E for the same river basins in the AMIP models. In the AMIP results, the annual mean P − E is averaged over all model grid points within the river basins. It is apparent from Fig. 11 that all models overestimate the annual mean P − E in the Ob River drainage basin. The oversimulation is consistent with the general oversimulation of Asian Arctic P and P − E (e.g., Fig. 8). While the overestimate is slight for some models (e.g., GFDL: 0.39 mm day−1, MRI: 0.41 mm day−1, UIUC: 0.42 mm day−1), the ensemble average for the AMIP models—excluding the outlier value of 1.62 mm day−1 for the SUNYA model—is 0.59 mm day−1, which is 63% higher than the observed “discharge” of 0.36 mm day−1. For the Mackenzie River drainage basin, the models’ ensemble average of 0.61 mm day−1 is 15% higher than the discharge-derived 0.53 mm day−1. Several models (COLA, ECMWF, GFDL, NMC, NRL, UGAMP) undersimulate P − E for the Mackenzie Basin.

The comparison of river discharge and P − E over the high-latitude watersheds must be placed into the context of a complex seasonality arising from water storage in the snowpack. Figure 12 shows that the discharge of the major Arctic rivers is highly seasonal and that the maximum discharge occurs in the May–July period. Of the rivers depicted in Fig. 12, the Lena’s discharge has the strongest seasonality and the Mackenzie the weakest. The importance of snowmelt follows from the fact that the river discharge is a maximum when P − E is decreasing rapidly to its minimum. For the Ob–Yenisey and Mackenzie Rivers, most models snow a negative P − E in the month of maximum river discharge. The prevalence of liquid water at the surface immediately after snowmelt evidently contributes to a larger E as well as to the increase of discharge. Of the river basins included in Fig. 12, the Lena is somewhat unique in the sense that most models show little or no decrease of P − E during summer. However, neither is there a systematic increase of the simulated P − E when the Lena’s discharge increases dramatically from May to June.

In an attempt to diagnose the model-to-model differences in P − E for the Arctic drainage basins, precipitation and evaporation were compared across the sample of models. Figure 13 is a plot of the annual mean P against the annual mean E for the Ob and Mackenzie drainage basins. It is apparent that the models with the largest (smallest) P are generally those with the largest (smallest) E. The correlation between the annual mean P and E for the AMIP models (excluding SUNYA) is +0.89 for the Ob and +0.88 for the Mackenzie. (The inclusion of the outlier model, SUNYA, reduces the respective correlations to 0.75 and 0.48.) On a seasonal basis, the correlation is strongest in summer. For example, the correlations between the 10-yr seasonal mean P and E for the Ob are 0.45 for winter (December–February), 0.58 for spring (March–May), 0.89 for summer (June–August), and 0.80 for autumn (September–November).

The positive correlation between P and E is also found in the interannual variations of individual models. Table 3 shows that the monthly P and E are positively correlated in nearly every AMIP model, especially during spring and summer. Many of the correlations in Table 3 exceed 0.8. Figure 14, which contains scatterplots of P and E from the CCC, GSFC, and ECMWF AMIP models, shows that the associations between P and E depend on the season and on the model. However, as suggested by the occurrence of triads in some panels of Figure 14 (see GSFC in spring), some of the common variance between P and E is attributable to the seasonality in both P and E; for example, both variables increase substantially during the spring period, March–May, in most of the models, resulting in the apparent“jumps” when the data are aggregated by calendar month. Table 4 shows the correlations after the seasonal cycle has been removed by subtraction of the 10-yr monthly mean from each value of P and E. While the values are generally lower than in Table 3, many still exceed 0.5 and are statistically significant. In this respect, the P versus E correlations may be considered robust. However, the correlations do not imply causality. On the one hand, P may be larger when E is larger because the evaporation increases the atmospheric moisture available for precipitation. On the other hand, the positive correlation may simply indicate that a greater P leads to a greater E by increasing the soil moisture available for evaporation. Results later in this section loosely support the latter interpretation because P − E is larger when E is larger in the models.

A diagnosis of the model-to-model differences in P, E, and P − E is difficult because the model formulations include different treatments of soil–vegetation processes, atmospheric processes, and surface–atmosphere exchanges. For example, as shown in Table 2, some models (15 of the 23 examined here) include the evaporation of falling rain, the others do not; some (4 of 23) have prognostic cloud liquid water, the others do not; some (3 of 23) allow the conversion of the liquid to the ice phase within clouds; most (but not all) allow the surface roughness to vary spatially; about half include explicit treatments of vegetation, the others treat only the soil explicitly. Soil moisture is handled by the models in one (or more) of three primary ways: the force–restore method, the bucket method, and the diffusion method. Four models prescribe (rather than compute) soil moisture, including its spatial and temporal variations, while two other models prescribe only the deep soil moisture. We composited the AMIP models’ P, E, and P − E for the Ob and Mackenzie basins according to all the distinctions listed above but found a statistically significant dependence on only the final one: the specification (or absence thereof) of soil moisture. Over the Ob Basin, for example, the four models with prescribed soil moisture had an annual mean P − E of 0.70 mm day−1 (vs a mean of 0.59 mm d−1 for the other models, excluding the SUNYA outlier). Over the Mackenzie Basin, the models with prescribed soil moisture had an annual mean P − E of 0.67 mm day−1 (vs a mean of 0.59 mm day−1 for the other models). These values represent increases of 19% and 14% over the means for the Ob and Mackenzie Basins, respectively, in the models with freely evolving soil moisture. The models with prescribed soil moisture have larger E as well as larger P, but the increase of P is greater. The implication is that the convergence of moisture evaporated elsewhere is increased sufficiently by the soil moisture specification that it outweighs the enhanced loss of surface moisture (by E) in the model domain. Thus, the specification of soil wetness increases the bias (relative to the measured river discharge) of the annual mean P − E in the AMIP models. Interestingly, the regions immediately adjacent to the Ob and Mackenzie Basins showed a similar but weaker tendency for P and E to be larger when soil moisture was prescribed. Soil moisture sensitivity experiments with individual models are needed in order to identify the origins of the changes in the moisture budget resulting from the treatment of soil moisture.

It should also be noted that the annual means of P and E evaluated from the NCEP reanalyses (Kalnay et al. 1996) for 1979–95 are 1.61 mm day−1 and 1.16 mm day−1 for the Mackenzie Basin, resulting in P − E = 0.45 mm day−1 (Fig. 15). This value is smaller than the discharge-derived value of 0.53 mm day−1 and slightly less than the value of 0.47 mm day−1 obtained from the NMC AMIP run, which is based on an earlier version of the model used in the NCEP reanalyses. Inspection of the fields of P and E in the NCEP reanalyses reveals a blotchy pattern of lobes that are evidently spurious and attributable to spectral filtering effects in polar latitudes. However, areal averages over regions comparable to the larger Arctic river basins encompass a sufficient number of these lobes that the areal averages may be meaningful measures of the large-scale moisture convergence (M. Serreze 1997, personal communication). Figure 16 shows that the seasonality of P − E depicted by the reanalyses is similar to that in the AMIP models (Fig. 12) and out of phase with respect to the monthly discharge, as in the case of most of the AMIP models. Figure 16 also shows that there is considerable interannual variability of P − E (and discharge) superimposed on the annual cycles.

The computation of NCEP reanalysis-derived annual means for the Arctic Ocean domain poleward of 70°N gives P = 0.76 mm day−1, E = 0.47 mm day−1, and P − E = 0.30 mm day−1. The corresponding estimates of the annual mean P from the observational compilations mentioned earlier are 0.80 mm day−1 (Bryazgin), 0.73 mm day−1 (Legates and Willmott), and 0.59 mm day−1 (Jaeger), while the “best estimates” of annual mean P − E from Bryazgin’s P and the World Water Balance’s E is 0.32 mm day−1 (Korzun 1978). As noted in the discussion of Fig. 8, the AMIP models generally oversimulate P − E relative to the observationally derived estimates for the Arctic Ocean. The biases of P − E as well as P appear to be smaller over the Arctic Ocean in the NCEP reanalyses, suggesting that the assimilation of atmospheric observational data (e.g., winds and moisture) leads to improvements in the simulated P and P − E, at least over the Arctic Ocean.

5. Conclusions

Verification of the AMIP models’ simulations of Arctic hydrologic variables is complicated by the uncertainties in the corresponding observational estimates. As shown here (Fig. 3), the observational estimates of the areal mean precipitation for regions such as the Arctic Ocean vary widely; estimates based on gauge-corrected station data are generally the largest, while other climatological estimates tend to be smaller by as much as 50%. A narrowing of the uncertainty of the observational estimates must be a high priority in the context of model verification, and this priority has indeed emerged in the planning of the international Arctic Climate System Study (WCRP 1994). Uncertainties in regional mean evaporation in the Arctic are even greater, although the difference between regional P and E can be estimated with somewhat higher confidence by drawing upon measurements of river discharge.

In spite of the observational uncertainties, several conclusions emerge from this assessment of the AMIP models.

  • The AMIP models generally oversimulate the annual mean of Arctic precipitation relative to all the observational estimates. While there is considerable scatter among the models, the models’ ensemble mean P is higher than the observed estimates over the Arctic Ocean as well as the northern land areas that drain into the Arctic Ocean. The one exception is the simulation of the European watershed’s P relative to the Bryazgin–Shver observational estimates. However, the European Arctic watershed is far smaller than the Asian and North American Arctic watersheds and estimates of areally averaged P are sensitive to the resolution of the gridded datasets.
  • The areal means of P − E are also generally oversimulated by the AMIP models, although the oversimulations are smaller (as percentages of the observational estimates) than in the case of P over the terrestrial watersheds. While the latter finding suggests that E may also be oversimulated over the terrestrial watersheds, there are large uncertainties in the observational estimates of E.
  • The simulated P and E are highly correlated across models, and the interannual variations (as well as the seasonal variations) of P and E are highly correlated in the output of most of the individual models. These correlations imply that variations of either P or E contribute to variations of the other (or that both are true). Thus local hydrologic processes are strongly coupled in the AMIP models.
  • Of the various model features examined in this study, the only one common to the high-P (and high-E) models was the specification of soil moisture. Controlled experiments with individual models are required in order to identify the role of other elements of the model formulation (e.g., cloud microphysical processes, vegetative processes, etc.).
  • The excess of the models’ ensemble mean discharge (P − E) to the Arctic Ocean from the Ob and Mackenzie drainage basins are 63% and 15%, respectively, corresponding to 21.4 cm yr−1 and 2.7 cm yr−1 of excessive P − E relative to the discharge-derived P − E. Since these rivers provide two of the largest inputs of freshwater to the Arctic Ocean, the implication is that the role of freshwater in the stratification and sea- ice regime of the Arctic Ocean’s coastal areas may be difficult to simulate accurately with coupled models until the biases in the atmospheric models are reduced.
  • A preliminary examination of the NCEP reanalyses shows that the reanalysis-derived P − E is close to the observational estimates for the Arctic Ocean. Although the fields of both P and E appear to contain unrealistic small-scale features over the Arctic Ocean, the reanalyses from NCEP and other centers (e.g., ECMWF) must be regarded as potentially valuable supplements to observational data in depicting the spatial and temporal variations as well as the climatology of Arctic precipitation.

Of the findings obtained in this study, the one of greatest direct relevance to climate modeling is the overestimation of Arctic P and P − E by the AMIP models. The excessive P and P − E call for diagnostic assessments that were not possible with the output of the AMIP models, for which the formulations of model physics differ among models in a large number of ways. However, the close association between the variations of P and E in the AMIP models suggest that the surface parameterization should be an initial focus of the diagnosis. Controlled experiments with individual models will be required in order to identify the parameterizational features responsible for the excessive P and P − E.

Acknowledgments

We wish to thank Dr. Nikolay Bryazgin of the Arctic and Antarctic Research Institute (St. Petersburg) for providing his monthly maps of Arctic precipitation and for helpful discussions. We are also grateful to the following colleagues at the Main Geophysical Observatory: Valentina Gavrilina, Veronika Govorkova, and Tatyana Pavlova for their computational assistance, and Zoya Bryn for the digitization of Dr. Bryazgin’s data.

The Illinois portion of this project was supported by the National Science Foundation’s Climate Dynamics Program through Grant ATM-9319952. We thank Xin Tao and and William Chapman for data processing and assistance with the graphics, and we thank Norene McGhiey for word processing the manuscript.

Finally, we express our appreciation to the staff of PCMDI (Program for Climate Model Diagnosis and Intercomparison), which coordinated AMIP and provided the AMIP model output and documentation. This study was performed as part of AMIP Diagnostic Subproject No. 8.

REFERENCES

  • Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and other fresh waters in the Arctic circulation. J. Geophys. Res.,94, 14485–14498.

  • Bryazgin, N. N., 1976: Yearly mean precipitation in the Arctic region accounting for measurement errors (in Russian). Proc. Arctic Ant. Res. Inst.,323, 40–74.

  • ——, and T. A. Shver, 1976: Atmospheric precipitation over the Arctic land watershed basin (in Russian). Proc. Arctic Ant. Res. Inst.,323, 75–86.

  • Dumenil, L., K. Isele, H.-J. Liebscher, U. Schroeder, M. Schumacher, and K. Wilke, 1993: Discharge data from 50 selected rivers for GCM validation. Max-Planck-Institut für Meteorologie/Global Runoff Data Centre Rep. 100, 61 pp. [Available from Max-Planck- Institut für Meteorologie, Bundestrasse 55, D-20146 Hamburg, Germany.].

  • Environment Canada, 1994: Annual review: 1994. Climate Perspective,16, 1–10.

  • Gates, W. L., 1992: AMIP: The Atmospheric Model Intercomparison Project. Bull. Amer. Meteor. Soc.,73, 1962–1970.

  • Golubev, V. S., and E. G. Bogdanova, 1996: Point measurements of solid precipitation. Proc. Workshop on the ACSYS Solid Precipitation Climatology Project, Reston, VA, World Climate Research Programme, 18–29.

  • Goodison, B. E., E. Elomaa, V. Golubev, T. Gunther, and B. Sevruk, 1994: WMO solid precipitation measurement intercomparison: Preliminary results. World Meteor. Org. TD-588, 15–20.

  • IPCC, 1996: Climate Change 1995: The Science of Climate Change. J. T. Houghton, L. G. Meira Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds., Cambridge University Press, 572 pp.

  • Ivanov, V. V., 1976: Freshwater balance of the Arctic Ocean. Proc. Arctic Ant. Res. Inst.,323, 138–147.

  • Jaeger, L., 1983: Monthly and areal patterns of mean global precipitation. Variations in the Global Water Budget, A. Street-Perrot, M. Beran, and R. Ratcliffe, Eds., D. Reidel, 129–140.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-year reanalysis project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Karl, T. R., P. Ya. Groisman, R. R. Heim Jr., and R. W. Knight, 1993:Recent variations of snow cover and snowfall in North America and their relation to precipitation and temperature variations. J. Climate,6, 1327–1344.

  • Khrol, V. P., Ed., 1996: Atlas of Water Balance of the Northern Polar Area. Gidrometeoizdat, 81 pp.

  • Korzun, V. I., Ed., 1978: World Water Balance and Water Resources of the Earth. UNESCO Press, 663 pp.

  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gage-corrected global precipitation. Int. J. Climatol.,10, 111–133.

  • Mysak, L., D. K. Manak, and R. F. Marsden, 1990: Sea ice anomalies observed in the Greenland and Labrador seas during 1901–1984 and their relation to an interdecadal climate cycle. Climate Dyn.,5, 111–133.

  • Oechel, W. C., S. J. Hastings, G. Vourlitis, M. Jenkins, G. Riechers, and N. Grulke, 1993: Recent changes in Arctic tundra ecosystems from a net carbon sink to a source. Nature,361, 520–523.

  • Peixoto, J. P., and A. H. Oort, 1983: The atmospheric branch of the hydrological cycle and climate. Variations in the Global Water Budget, A. Street-Perrot, M. Beran, and R. Ratcliffe, Eds., D. Reidel, 5–65.

  • Phillips, T. J., 1994: A summary documentation of the AMIP Models. Program for Climate Model Diagnosis and Intercomparison Rep. 18, Lawrence Livermore National Laboratory, UCRL-ID-116384, 343 pp. [Available from Lawrence Livermore National Lab., P.O. Box 808, L-264, Livermore, CA 94550.].

  • Serreze, M. C., R. G. Barry, and J. E. Walsh, 1995: Atmospheric water vapor characteristics at 70°N. J. Climate,8, 719–731.

  • Vowinckel, E., and S. Orvig, 1970: The climate of the North Polar Basin. World Survey of Climatology, Vol. 14, Climates of the Polar Regions, S. Orvig, Ed., Elsevier, 129–252.

  • WCRP, 1992: Scientific concept of the Arctic Climate System Study (ACSYS). World Climate Research Programme WCRP-72, WMO/TD-486, 89 pp. [Available from World Meteorological Organization, Case Postale No. 2300, CH-1211 Geneva 20, Switzerland.].

  • ——, 1994: Arctic Climate System Study (ACSYS): Initial implementation plan. World Climate Research Programme WCRP-85, WMO/TD-627, 66 pp. [Available from World Meteorological Organization, Case Postale No. 2300, CH-1211 Geneva 20, Switzerland.].

  • Zubenok, L. I., 1976: Evaporation over the land watershed basin of the Arctic Ocean. Proc. Arctic Ant. Res. Inst.,323, 87–100.

Fig. 1.
Fig. 1.

Annual mean precipitation rates (mm day−1) in the Arctic from the analyses of Legates and Willmott (1990)—upper panel; Bryazgin (1976)—middle panel; and Jaeger (1983)—lower panel.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 2.
Fig. 2.

Annually averaged zonal mean precipitation rates (mm day−1) from (a) the three observational sources in Fig. 1, and (b) the AMIP models included in this study.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 3.
Fig. 3.

Mean seasonal cycles of area-averaged precipitation (mm day−1) for the ocean area poleward of 70°N: (a) observational estimates, and (b) AMIP model results relative to range of observational estimates (shaded area). Shown in (a) are the observational estimates of Vowinckel and Orvig (1970) for the polar cap including and excluding the Norwegian–Barents Seas.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 4.
Fig. 4.

Annual mean rates of (a) precipitation, P, and (b) precipitation minus evaporation, P − E, for the Arctic Ocean as evaluated from observational estimates (bars at left) and from the AMIP models. Observational sources are B = Bryazgin, L = Legates and Willmott, J = Jaeger, and O = Vowinckel and Orvig (darker bar is for domain excluding Barents–Norwegian Seas).

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 5.
Fig. 5.

Regional delineation for results in Figs. 6–10. Asian Arctic watershed consists of regions III, IV, V, and VI; North American Arctic watershed consists of VII, VIII, and IX (including VIII-a and IX-a); European Arctic watershed consists of I and II; Kara Sea watershed is III; East Siberian watershed is V; Canada Straits–Foxe Basin watershed is VIII (including VIII-a). [Figure from Ivanov (1976).]

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 6.
Fig. 6.

Mean annual cycle of monthly precipitation (mm day−1) from 23 AMIP models and from observational estimates of Legates–Willmott and Jaeger (dashed lines without symbols). Results are shown in top row for the (a) Asian, (b) American, and (c) European watersheds of the Arctic Ocean; in bottom row for (d) Kara Sea watershed, (e) East-Siberian Sea watershed; and (f) Canada Straits and Foxe Basin watershed (cf. Fig. 5)

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 7.
Fig. 7.

Annual mean precipitation rates for the six regions of Fig. 6. Observationally derived estimates are shown at left of each panel.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 8.
Fig. 8.

Mean biases of the ensemble of AMIP models relative to the observational estimates of precipitation (B = Bryazgin, L = Legates–Willmott, J = Jaeger, 01 and 02 = Vowinckel–Orvig with and without the Barents–Norwegian Seas) and precipitation minus evaporation (Bryazgin’s P minus Zubenok’s E).

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 9.
Fig. 9.

As in Fig. 6 but for evaporation E.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 10.
Fig. 10.

As in Fig. 7 but for P − E. Observational estimates are areal means of Bryazgin’s P minus Zubenok’s E.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 11.
Fig. 11.

AMIP model-derived P − E averaged over (a) Ob Basin and (b) Mackenzie Basin. Left-most bar of each panel is annual mean river discharge rate (mm3 yr−1) divided by corresponding area (mm2) of river basin.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 12.
Fig. 12.

Annual cycles of model-derived P − E for (a) Kara Sea land watershed, (b) Laptev Sea land watershed and (c) Beaufort Sea land watershed. Also shown are annual cycles of discharge (equivalent mm day−1) of major rivers in each watershed: (a) Ob and Yenisey (total), (b) Lena, and (c) Mackenzie.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 13.
Fig. 13.

Scatterplots of 10-yr (1979–1988) annual mean P vs annual mean E for the AMIP models. Included is the SUNYA “outlier.”

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 14.
Fig. 14.

Scatterplots of monthly precipitation vs monthly evaporation for the Mackenzie Basin from the AMIP simulations of 1979–88 by 3 models: CCC (left panels), GSFC (center panels), and ECMWF (right panels). Each panel contains results for one season (3 months × 10 yr = 30 data points).

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 15.
Fig. 15.

Monthly mean precipitation, evaporation, and P − E (asterisks) obtained by averaging NCEP reanalyses over all grid points in Mackenzie Basin. Results are for 17 yr, 1979–95.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Fig. 16.
Fig. 16.

Time series of monthly P − E from NCEP reanalyses averaged over grid points in (a) Ob River drainage basin and (b) Mackenzie River drainage basin. Dashed spikes are measured monthly river discharges divided by area of river basin.

Citation: Journal of Climate 11, 1; 10.1175/1520-0442(1998)011<0072:APAEMR>2.0.CO;2

Table 1.

The AMIP simulations used in this study. Information from Phillips (1994, 2 and 6). Models are identified in text by first three letters of acronym in left column.

Table 1.
Table 2.

Treatment of various physical processes affecting precipitation and evaporation in AMIP models (“×” = included). [From Phillips (1994, 15–20).]

Table 2.
Table 3.

Correlations between the AMIP models’ monthly precipitation and evaporation in the Ob and Mackenzie River Basins. Months are pooled into seasons: spring [March–May (MAM)], summer [June–August (JJA)], autumn [September–November (SON)], and winter [December–February (DJF)]. Italicized values are statistically significant at 95% level.

Table 3.
Table 4.

As in Table 3 but for the anomalies (departures from monthly means) of precipitation and evaporation.

Table 4.
Save