• Aagaard, K., and E. C. Carmack, 1989: The role of sea ice and other fresh waters in the Arctic circulation. J. Geophys. Res.,94 (C10), 14 485–14 498.

  • ARCUS, 1991: Arctic System Science Land/Atmosphere/Ice Interactions:A Plan for Action. Arctic Research Consortium of the United States, 48 pp.

  • Armstrong, R. L., and M. J. Brodzik, 1995: An earth-gridded SSM/I data set for cryospheric studies and global change monitoring. Adv. Space Res.,16, 155–163.

  • Arpe, K., H. Behr, and L. Dumenil, 1996: Validation of the ECHAM4 climate model and re-analysis data in the Arctic regions. Proc. Workshop on the Implementation of the Arctic Precipitation Data Archive (APDA) at the Global Precipitation Climatology Center (GPCC), Offenbach, Germany, WMO, 31–40.

  • Basist, A. N., and M. Chelliah, 1997: Comparisons of tropospheric temperatures derived from the NCEP/NCAR reanalysis, NCEP operational analysis, and the microwave sounding unit. Bull. Amer. Meteor. Soc.,78, 1431–1447.

  • Bryazgin, N. N., 1976: Srednegodovoe kolichestvo osadkov v Arktike s uchetom pogreshnosti osadkomerov. Tr. Arkt. Antarkt. Nauch.-Issled. Inst.,323, 40–74.

  • Bryson, R. A., and F. K. Hare, 1974: Climates of North America. Vol. 11, World Survey of Climatology, Elsevier, 420 pp.

  • Colony, R. L., and I. G. Rigor, 1993: International Arctic Ocean buoy program data report for 1 January 1992–31 December 1992. Tech Memo. APL-UWTM 29-93, University of Washington, Seattle, WA, 215 pp. [Available from Applied Physics Laboratory, 1013 NE 40th St., Seattle, WA 98195.].

  • ——, V. Radionov, and F. L. Tanis, 1997: Measurements of precipitation and snow at the Russian North Pole drifting stations. Polar Geogr.,34, 3–14.

  • Cressman, G. P., 1959: An operational objective analysis system, Mon. Wea. Rev.,87, 367–374.

  • Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: The atmospheric hydrologic cycle over the Arctic Basin from reanalyses. Part I: Comparison with observations and previous studies. J. Climate, in press.

  • Dey, C. H., and L. L. Morone, 1985: Evolution of the NMC Global Assimilation System: January 1982–December 1983. Mon. Wea. Rev.,113, 304–318.

  • DiMego, G., 1988: The National Meteorological Center Regional Analysis System. Mon. Wea. Rev.,116, 977–1000.

  • Dorman, C. E., and R. B. Bourke, 1978: A temperature correction for Tucker’s ocean rainfall estimates. Quart. J. Roy. Meteor. Soc.,104, 765–773.

  • Dorman, J. L., and P. J. Sellers, 1989: A global climatology of albedo, roughness length and stomatal resistance for atmospheric general circulation models as represented by the Simple Biosphere Model (SiB). J. Appl. Meteor.,28, 833–855.

  • Fisheries and Environment Canada, 1978: Hydrological Atlas of Canada. Fisheries and Environment Canada, Minister of Supply and Services, 31 pp.

  • Furgeson, H. L., A. D. O’Neill, and H. F. Cork, 1970: Mean evaporation over Canada. Water Resour. Res.,6, 1618–1633.

  • Genthon, C., and A. Braun, 1995: ECMWF analyses and predictions of the surface climate of Greenland and Antarctica. J. Climate,8, 2324–2332.

  • Gibson, J. K., S. Kallberg, A. Hernandez, S. Uppala, A. Nomura, and E. Serrano, 1997: ERA Description. ECMWF Re-Analysis Project Report Series, Vol. 1, European Centre for Medium-Range Weather Forecasts, 72 pp.

  • Gorshkov, S. G., Ed., 1983: Arctic Ocean (in Russian). Vol. 3, World Ocean Atlas, Pergamon Press, 189 pp.

  • Groisman, P. Y., V. V. Koknaeva, T. A. Belokrylova, and T. R. Karl, 1991: Overcoming biases of precipitation: A history of the USSR experience. Bull. Amer. Meteor. Soc.,72, 1725–1733.

  • Hibler, W. D., III, and J. Zhang, 1995: On the effect of sea-ice dynamics on ocean thermohaline circulation. Ann. Glaciol.,21, 361–368.

  • Hortal, M., and J. Simmons, 1991: Use of reduced Gaussian grids in spectral models. Mon. Wea. Rev,119, 1057–1074.

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

  • Kane, D. L., L. D. Hinzman, M. Woo, and K. Everett, 1992: Arctic hydrology and climate change. Arctic Ecosystems in a Changing Climate, F. Chapin III, Ed., Academic Press, 35–57.

  • Khrol, V. P., 1976: Isparenie s poverkhnosti Severnogo Ledovitogo Okeana. Tr. Arkt. Antarkt. Nauchno.-issled. Inst.,323, 148–155.

  • Korzun, 1976: Atlas of World Water Balance (in Russian). Gidrometeoizdat, 65 pp.

  • Ledley, T. S., 1991: Snow on sea ice: Competing effects in shaping climate. J. Geophys. Res.,96 (D9), 17 195–17 208.

  • Legates, D., 1987: A climatology of global precipitation. Publ. Climatol.,4 (1), 85 pp.

  • ——, and C. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected global precipitation. Int. J. Climatol.,10, 110–127.

  • Lydolph, P. E., 1977: Climates of the Soviet Union. Vol. 7, World Survey of Climatology, Elsevier, 443 pp.

  • Mahfauf, J. F., 1991: Analysis of soil moisture from near-surface parameters: A feasibility study. J. Appl. Meteor.,30, 1534–1547.

  • Matthews, E., 1985: Atlas of archived vegetation, land use, and seasonal albedo data sets. NASA Tech. Memo. 86199, NASA/Goddard Institute For Space Studies, New York, 53 pp. [Available from Goddard Space Flight Center, Greenbelt, MD 20771.].

  • Ohmura, A., 1982: Evaporation from the surface of the Arctic Tundra on Axel Heiburg Island. Water Resour. Res.,18, 291–300.

  • ——, 1984: Comparative energy balance study for arctic tundra, sea surface, glaciers and boreal forests. Geo-Journal,8, 221–228.

  • Pan, H.-L., and L. Mahrt, 1987: Interaction between soil hydrology and boundary layer developments. Bound.-Layer Meteor.,38, 185–202.

  • ——, and W.-S. Wu., 1994: Implementing a mass flux convection parameterization package for the NMC medium range forecast model. Preprints, 10th Conf. on Numerical Weather Prediction, Portland, OR, Amer. Meteor. Soc., 96–98.

  • Parish, D. F., and J. C. Derber, 1992: The National Center’s Spectral Statistical-Interpolation Analysis System. Mon. Wea. Rev.,120, 1747–1763.

  • Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate,7, 929–948.

  • Schweiger, A. J., and J. Key, 1994: Arctic Ocean radiative fluxes and cloud forcing based on the ISCCP-C2 dataset, 1983–1990. J. Appl. Meteor.,33, 948–963.

  • Serreze, M. C., 1995: Climatological aspects of cyclone development and decay in the Arctic. Atmos.–Ocean,33, 1–23.

  • ——, and J. A. Maslanik, 1998: Arctic precipitation as represented in the NCEP/NCAR reanalysis. Ann. Glaciol.,25, 429–433.

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

  • ——, J. A. Maslanik, and J. R. Key, 1997: Atmospheric and sea ice characteristics of the Arctic Ocean and the Sheba field region in the Beaufort Sea. Special Rep. 4, National Snow and Ice Data Center, Cooperative Institute for Research in Environmental Sciences, University of Colorado, 219 pp. [Available from National Snow and Ice Data Center, Campus Box 449, University of Colorado, Boulder, CO 80309-0449.].

  • ——, J. R. Key, J. E. Box, J. A. Maslanik, and K. Steffen, 1998: A new monthly climatology of global radiation for the Arctic and comparisons with NCEP–NCAR reanalysis and ISCCP-C2 fields. J. Climate,11, 121–136.

  • Spangler, W. M. L., and R. L. Jenne, 1984: World monthly surface station climatology. Scientific Computing Division, NCAR, Boulder, CO, 14 pp. [Available from National Center for Atmospheric Research, 1850 Table Mesa Drive, Boulder, CO 80307.].

  • Stendel, M., and C. Arpe, 1997: Evaluation of the hydrological cycle in reanalysis and observations, ECMWF reanalysis validation reports. Part 1. ECMWF Reanalysis Project Rep. Ser. 6, 50 pp.

  • Tiedke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev.,117, 1779–1800.

  • ——, 1993: Representation of clouds in large-scale models. Mon. Wea. Rev.,121, 1779–1800.

  • Trenberth, K. E., and C. J. Guillemot, 1998: Evaluation of the atmospheric moisuture and hydrological cycle in the NCEP/NCAR reanalyses. Climate Dyn.,14, 213–231.

  • Viterbo, P., 1995: Initial values of soil water and the quality of summer forecasts. ECMWF Newslett.,69, 2–8.

  • ——, and A. C. M. Beljaars, 1995: A new land parameterization scheme in the ECMWF model and its validation. J. Climate,8, 2716–2748.

  • ——, and P. Courtier, 1995: The importance of soil water for medium range weather forecasting: Implications for data assimilation. Workshop of Slowly Varying Components of Predictable Atmospheric Motions, Beijing, China, WMO, 121–130.

  • Walsh, J. E., and C. M. Johnson, 1979: An analysis of Arctic sea ice fluctuations, 1953–1977. J. Phys. Oceanogr.,9, 580–591.

  • ——, and W. L. Chapman, 1998: Arctic cloud–radiation–temperature associations in observational data and atmospheric reanalyses. J. Climate,11, 3030–3045.

  • ——, X. Zhou, D. Portis, and M. C. Serreze, 1994: Atmospheric contribution to hydrologic variations in the Arctic. Atmos.–Ocean,34 (2), 733–755.

  • ——, V. Kattsov, D. Portis, and V. Meleshko, 1998: Arctic precipitation and evaporation: Model results and observational estimates. J. Climate,11, 72–87.

  • WCRP, 1997: Proceedings of the Workshop on the Implementation of the Arctic Precipitation Data Archive (APDA) at the Global Precipitation Climatology Centre (GPCC), Offenbach, Germany, 10–12 July, 1996, Arctic Climate System Study, World Climate Research Programme. WCRP-98, WMO/TD 804, 127 pp. plus appendices. [Available from World Climate Research Programme, 7 bis, Avenue de la Paix, Case Postale 2300, CH-1211 Geneva 2, Switzerland.].

  • Weller, G., and B. Holmgren, 1974: The microclimates of the Arctic tundra. J. Appl. Meteor.,13, 854–862.

  • Wernstedt, F. L., 1972: World Climatic Data. Climatic Data Press, 522 pp.

  • Willmott, C. J., J. R. Mather, and C. M. Rowe, 1981a: Average monthly and annual surface air temperature and precipitation data for the world. Part 1: The eastern hemisphere. Pub. Climatol.,34 (1), 395 pp.

  • ——, ——, and ——, 1981b: Average monthly and annual surface air temperature and precipitation data for the world. Part 2: The western hemisphere. Pub. Climatol.,34 (1), 378 pp.

  • Woollen, J. S., E. Kalnay, L. Gandin, W. Collins, S. Saha, R. Kistler, M. Kanamitsu, and M. Chelliah, 1994: Quality Control in the Reanalysis System. Amer. Meteor. Soc., 13–14.

  • Zubenok, Z. T., 1976: Isparenie s sushi vodosbornogo basseina Severnogo Ledovitogo Okeana. Tr. Arkt. Antarkt. Nauchno.-issled. Inst.,323, 87–100.

  • View in gallery

    Location of additional land station data for the FSU (+), Canada (○), and North Pole drifting station observations (*) used to improve the Legates and Willmott (1990) climatology. The North Pole data are plotted at the average monthly positions.

  • View in gallery

    Difference in precipitation (mm) between the LWM and LW climatologies (LWM − LW) for (a) Jan and (b) Jul. The contour interval is 20 mm for Jan and 10 mm for Jul.

  • View in gallery

    Annual mean precipitation (mm) from (a) LWM, (b) NCEP, and (c) ERA. The contour interval is 100 mm except for the dotted contour, which encloses areas with precipitation totals less than 150 mm.

  • View in gallery

    Model errors in annual precipitation (mm) expressed as modeled minus observed values for (a) NCEP and (b) ERA. Positive errors are shown by solid lines with negative errors shown by dotted lines. The contour interval is 75 mm.

  • View in gallery

    Mean precipitation (mm) from the LWM climatology for (a) Jan and (b) Jul. The contour interval is 15 mm.

  • View in gallery

    Mean modeled precipitation for Jan (mm) and corresponding model errors (expressed as modeled minus observed totals) for NCEP (a), (b), and ERA (c), (d). Positive errors are shown by solid lines with negative errors shown by dotted lines. The contour interval is 15 mm.

  • View in gallery

    Same as Fig. 6 but for Jul.

  • View in gallery

    Pattern correlations by month for different regions of the Arctic between observed and modeled fields and between the two models.

  • View in gallery

    Seasonal cycles in observed and modeled precipitation (mm) for the same regions used in Fig. 8.

  • View in gallery

    Jan convective precipitation (mm) for (a) NCEP and (b) ERA. The contour interval is 10 mm.

  • View in gallery

    Same as Fig. 10 but for Jul.

  • View in gallery

    Mean Jul evaporation (mm) for (a) NCEP and (b) ERA, and (c) NCEP–ERA differences. All differences are positive. The contour interval is 15 mm.

  • View in gallery

    Mean Jul cloud cover (%) for (a) NCEP and (b) ERA, and (c) NCEP–ERA differences. Positive differences are shown by solid lines with negative differences shown by dotted lines. The contour interval is 10%.

  • View in gallery

    Difference field (NCEP minus ERA) of the downwelling solar radiation flux for Jul (W m−2). All differences are positive. The contour interval is 10 W m−2.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 533 193 11
PDF Downloads 300 140 7

Representation of Mean Arctic Precipitation from NCEP–NCAR and ERA Reanalyses

View More View Less
  • 1 Division of Cryospheric and Polar Processes, Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado
Full access

Abstract

An improved monthly precipitation climatology for the Arctic is developed by blending the Legates and Willmott gridded product with measurements from Russian “North Pole” drifting stations and gauge-corrected station data for Eurasia and Canada. The improved climatology is used to examine the accuracy of mean precipitation forecasts from the National Centers for Environmental Prediction (NCEP) and European Reanalysis Agency (ERA) reanalysis models, based on data for the period 1979–88. Both models capture the major spatial features of annual mean precipitation and general aspects of the seasonal cycle but with some notable errors. Both underestimate precipitation over the Atlantic side of the Arctic. NCEP overestimates annual totals over land areas and to a somewhat lesser extent over the central Arctic Ocean. Except for the North Atlantic–Scandinavia sector, the NCEP model also depicts the seasonal precipitation maximum consistently one month early in July. Overall, the ERA predictions are better. Both models perform best during winter and worst during summer.

The most significant problem with the NCEP model is a severe oversimulation of summer precipitation over land areas, due to excessive convective precipitation. Further investigation for July reveals that both the NCEP analyses and 12-h forecasts are too wet below about 850 mb and have more negative low-level temperature gradients as compared to available rawinsonde profiles. This suggests that low-level observations are not being effectively incorporated in the analyses. Given this finding, the high humidities are consistent with excessive surface evaporation rates. This problem may in turn relate to soil moisture, which NCEP updates by the modeled precipitation. If soil moisture is too high, this would favor excessive evaporation and high low-level humidities, fostering excessive precipitation, in turn keeping soil moisture and evaporation rates high. The NCEP downwelling shortwave fluxes are also much too high, contributing to excessive evaporation and possibly influencing the low-level temperature gradients. By comparison, soil moisture in the ERA model is adjusted using the difference between the model first guess and analysis value (the analysis increment) of low-level humidity, which prevents model drift. The ERA downwelling shortwave fluxes are also closer to observations. These attributes are consistent with the superior ERA precipitation forecasts in summer and suggest avenues for improving the performance of the NCEP model.

Corresponding author address: Dr. Mark C. Serreze, Division of Cryospheric and Polar Processes, CIRES, Campus Box 449, University of Colorado, Boulder, CO 80309-0449.

Email: serreze@kryos.colorado.edu

Abstract

An improved monthly precipitation climatology for the Arctic is developed by blending the Legates and Willmott gridded product with measurements from Russian “North Pole” drifting stations and gauge-corrected station data for Eurasia and Canada. The improved climatology is used to examine the accuracy of mean precipitation forecasts from the National Centers for Environmental Prediction (NCEP) and European Reanalysis Agency (ERA) reanalysis models, based on data for the period 1979–88. Both models capture the major spatial features of annual mean precipitation and general aspects of the seasonal cycle but with some notable errors. Both underestimate precipitation over the Atlantic side of the Arctic. NCEP overestimates annual totals over land areas and to a somewhat lesser extent over the central Arctic Ocean. Except for the North Atlantic–Scandinavia sector, the NCEP model also depicts the seasonal precipitation maximum consistently one month early in July. Overall, the ERA predictions are better. Both models perform best during winter and worst during summer.

The most significant problem with the NCEP model is a severe oversimulation of summer precipitation over land areas, due to excessive convective precipitation. Further investigation for July reveals that both the NCEP analyses and 12-h forecasts are too wet below about 850 mb and have more negative low-level temperature gradients as compared to available rawinsonde profiles. This suggests that low-level observations are not being effectively incorporated in the analyses. Given this finding, the high humidities are consistent with excessive surface evaporation rates. This problem may in turn relate to soil moisture, which NCEP updates by the modeled precipitation. If soil moisture is too high, this would favor excessive evaporation and high low-level humidities, fostering excessive precipitation, in turn keeping soil moisture and evaporation rates high. The NCEP downwelling shortwave fluxes are also much too high, contributing to excessive evaporation and possibly influencing the low-level temperature gradients. By comparison, soil moisture in the ERA model is adjusted using the difference between the model first guess and analysis value (the analysis increment) of low-level humidity, which prevents model drift. The ERA downwelling shortwave fluxes are also closer to observations. These attributes are consistent with the superior ERA precipitation forecasts in summer and suggest avenues for improving the performance of the NCEP model.

Corresponding author address: Dr. Mark C. Serreze, Division of Cryospheric and Polar Processes, CIRES, Campus Box 449, University of Colorado, Boulder, CO 80309-0449.

Email: serreze@kryos.colorado.edu

1. Introduction

Changes in the hydrologic budget of the Arctic may have significant oceanic, climatic, and biological effects. The existence of the Arctic sea ice cover is strongly dependent on the maintenance of a relatively fresh, low-density surface layer. River runoff provides the primary freshwater source to the Arctic Ocean of about 35 cm yr−1 (Aagaard and Carmack 1989). Following freshwater import through the Bering Strait, precipitation less evaporation (PE) over the Arctic Ocean itself provides the next largest contribution of 16–20 cm yr−1 (Serreze et al. 1995; Cullather et al. 2000). Changes in runoff into the Arctic Ocean and PE over the Arctic Ocean itself may influence sea ice extent and thickness, influencing surface albedo and turbulent heat fluxes. Snow cover on sea ice also affects its albedo and thermodynamic growth and decay (Ledley 1991). Freshwater export out of the Arctic, primarily represented by a flux of sea ice and low-salinity water through Fram Strait, may influence the convective regime in the Greenland–Norway–Iceland seas and, potentially, the global thermohaline circulation (Hibler and Zhang 1995). Arctic marine life is conditioned by sea ice, nutrient availability, and water density. Changes in these factors may impact marine ecosystems and biochemical cycling of essential nutrients. Variations in the terrestrial hydrologic cycle may alter soil moisture, impacting plant communities and their grazers. Arctic soils serve as sources and sinks of global carbon dioxide and methane and appear to respond sensitively to altered soil moisture and temperature (ARCUS 1991).

A continuing problem in understanding variability in the Arctic hydrologic budget is uncertainty in precipitation. Gauge limitations result in undercatch of solid precipitation. Although it is possible to correct for the effects of winds, varying gauge types, and wetting losses (Groisman et al. 1991), a more fundamental problem is the sparse observational network. In order to provide accurate monthly mean precipitation estimates for 2.5° latitude–longitude bands, 10–40 stations are required (WCRP 1997). From this definition the station density even for Arctic land areas is inadequate, let alone for the Arctic Ocean.

These shortcomings have raised interest in evaluating precipitation forecasts from numerical weather prediction (NWP) models. While avoiding problems of gauge undercatch and inadequate station density, and providing outputs at regular grid points, models offer the additional advantages of providing fields of other hydrologic budget components such as precipitable water, vapor flux convergence, and PE. As opposed to atmospheric general circulation models (AGCMs), which run in “climate mode,” NWP models are constrained by observed atmospheric data. Such models start with a previous forecast as a “first guess” of present atmospheric conditions. The first guess is then adjusted through data assimilation, with the resulting atmospheric analyses used to generate the next forecast. Assimilation data consist primarily of free atmosphere variables such as upper-air temperature, pressure heights, and humidity obtained from rawinsondes, pibals, dropsondes, and satellite retrievals. Output fields represent analyses (e.g., 700-mb heights) based on the first guess and assimilation as well as forecast surface variables such as precipitation, evaporation, and radiation fluxes.

Modeled surface fields for the Arctic are prone to a number of problems. These stem from simplistic treatment of sea ice (e.g., fixed sea ice surface albedo), poor physical parameterizations of clouds, radiation and boundary processes, difficulties in quality control of assimilation data, scant upper-air observations over the central Arctic Ocean, and difficulties in determining appropriate weights for observations versus first-guess fields. These problems magnify the lack of temporal consistency in archived fields due to changes introduced in model physics and assimilation systems as well as changes in the amount and quality of assimilation data (Basist and Chelliah 1997). Validation of NWP products in the Arctic also presents somewhat of a logical conundrum: while the fairly sparse observational database is the very reason why efforts are needed to examine model output, such efforts require validation against these same sparse data. Limited studies of operational NWP output are nevertheless encouraging. For example, recent investigations suggest that European Centre for Medium-Range Weather Forecasts (ECMWF) precipitation fields may be sufficiently accurate to validate AGCM simulations (Arpe et al. 1996).

Several agencies have completed projects collectively termed “reanalysis.” The most comprehensive of these are the efforts by the National Centers for Environmental Prediction–National Center for Atmospheric Research (hereafter simply termed NCEP), (Kalnay et al. 1996) and the ECMWF European Reanalysis Agency (ERA; Gibson et al. 1997). The intent behind reanalysis is to compile global, quality-controlled datasets of analyzed and forecast fields using “frozen” data assimilation–forecast systems, eliminating pseudo–climate signals introduced by changes in model physics. The NCEP effort provides data from 1958 to the present while ERA fields are available from 1978 to 1993.

Although temporal discontinuities are still inevitable due to changes in the amount and quality of assimilation data, reanalysis products offer the best opportunity to assess the state of modeling of Arctic precipitation. Precipitation patterns over Greenland examined using 6 yr of ERA data appear realistic (Genthon and Braun 1995). In a pilot study conducted in 1996, Serreze and Maslanik (1998) found that the NCEP model captures the major spatial patterns of annual precipitation when compared to Russian climatologies (Bryazgin 1976; Gorshkov 1983) but has problems in depicting precipitation magnitudes and seasonality. However, these Russian climatologies are not provided in digital form, limiting comparisons. As part of a global study, Stendel and Arpe (1997) briefly examined ERA and NCEP precipitation forecasts in the Mackenzie Basin. One conclusion is that NCEP overestimates summer precipitation.

Cullather et al. (2000) have compared balances between precipitation and evaporation (PE) based on the vapor flux convergence (using analyzed wind and moisture fields) and from the model forecasts of P and E. The spatial distribution of mean annual PE from the ERA model forecasts is in qualitative agreement with available estimates (Gorshkov 1983) while the NCEP model performs less well. For both models, forecast and analyzed PE as averaged north of 70°N are not in hydrologic balance, with lower PE in the forecasts. Cullather et al. (2000) also showed reasonable agreement between monthly ERA and NCEP precipitation time series and data from two of the Russian “North Pole” Arctic Ocean drifting stations.

Given the climatic significance of Arctic precipitation and the potential applications of precipitation forecasts, the need exists to expand upon these studies by 1) developing an improved gridded Arctic precipitation climatology to allow for fuller assessment of the ability of the ERA and NCEP models to simulate observed features of Arctic precipitation; and 2) assessing, to the extent possible, the causes for any large, systematic errors in precipitation forecasts and offering strategies that may lead to improved model performance. To this end, we examine mean annual and monthly Arctic precipitation fields as depicted by the two models for the 10-yr period 1979–88. With regard to available assimilation data, this decade spans years before the breakup of the former Soviet Union (FSU), during which the terrestrial rawinsonde network was at its most robust. The FSU North Pole (NP) drifting station program provided upper-air measurements over the Arctic Ocean (Colony et al. 1997). Drifting buoys have also provided an extensive database of Arctic Ocean surface pressures since 1979 (Colony and Rigor 1993). For validation of modeled precipitation, we rely on a version of the Legates and Willmott (1990) monthly climatology improved through incorporation of NP precipitation data and gauge-corrected archives for land areas. Differences in precipitation forecasts are diagnosed using modeled and observed fields of evaporation, downwelling solar radiation, and surface air temperature, as well as through comparisons between vertical profiles of temperature and humidity from rawinsonde data and corresponding model values.

2. Reanalysis systems

a. NCEP

The NCEP reanalysis is an intermittent data assimilation scheme performed with a T62 model with 28 vertical sigma levels and the Operational Statistical Interpolation (SSI) procedure for assimilation. The model is identical to the medium-range forecast global operational model implemented on 10 January 1995 (Basist and Chelliah 1997), except for the horizontal resolution, which is T126 in the operational version. Model outputs are provided on a 6-h basis.

Potential assimilation data are formatted into a common standard World Meteorological Organization (WMO) binary universal format representation (BUFR) and then subjected to quality control during a preprocessing stage and then through an optimal interpolation quality control process (Dey and Morone 1985; Woollen et al. 1994; DiMego 1988; Kalnay et al. 1996). The data sources include rawinsonde profiles, surface marine reports from the Comprehensive Ocean–Atmosphere Data Set (COADS), aircraft observations, surface land synoptic reports, satellite soundings from the Tiros Operational Vertical Sounder (TOVS) and other platforms, surface wind speeds from the Special Sensor Microwave Imager, and satellite cloud drift winds.

Model boundary conditions include 1) National Oceanic and Atmospheric Administration (NOAA) sea surface temperatures (SSTs; Reynolds and Smith 1994) for 1982 onward and U. K. Meteorological Office (UKMO) fields for earlier years; 2) National Environmental Satellite Data and Information System (NESDIS) weekly snow cover analyses; 3) sea ice extent derived from passive microwave data from 1979 onward (both hemispheres) and for earlier years, the Walsh and Johnson (1979) dataset (Arctic) and climatology (Antarctic); 4) soil moisture from the two-layer implicit scheme of Pan and Mahrt (1987); 5) roughness length and vegetation resistance given by the Simple Biosphere Model (Dorman and Sellers 1989); and 6) albedos from the atlas of Matthews (1985).

An ice concentration cutoff of 0.55 is used; above this value, a full (100%) ice cover is assumed. For the UKMO SSTs used prior to 1982, a climatological fit was used to provide concentration-adjusted values of SST. For the later years based on the NOAA dataset, areas with ice concentration greater than or equal to 0.50 are ascribed an SST of −1.8°C. Below this concentration threshold, SSTs are based on interpolation of in situ and satellite data (Reynolds and Smith 1994). This difference in analysis techniques resulted in a discontinuity in SSTs between 1981 and 1982 (R. Reynolds 1998, personal communication).

In SSI, a global dynamical balance is imposed on the analysis, which obviates the need for a separate model initialization procedure (Parish and Derber 1992). As compared to optimal interpolation (OI), SSI was found to provide improved analyses and forecasts, especially in the Tropics, and a reduction in precipitation spinup. A drawback to this approach is the Gibbs phenomenon, which tends to produce “ringing” after a transformation from grid to spectral and back to grid representation.

Two types of precipitation are computed: convective and grid scale (dynamic). Convection is based on a simplified Arakawa–Schubert scheme (Pan and Wu 1994). The scheme was found to result in improved prediction of precipitation over the continental United States and Tropics as compared to the previous Kuo parameterization (Kalnay et al. 1996). Dynamic precipitation is parameterized by starting at the top layer and checking for supersaturation. If supersaturated, latent heat is released to adjust the specific humidity and temperature to saturation, with the excess water falling to the next lower layer. If this next layer is supersaturated, then adjustment to saturation occurs again and the amount of precipitation is added to that from the higher layer. However, if the layer is unsaturated, some or all of the precipitation is evaporated. The process continues downward, with all precipitation that penetrates to the bottom layer allowed to fall to the surface.

b. ERA

The ERA reanalysis is also based on a fixed intermittent data assimilation scheme. The forecast model has a horizontal resolution of T106 with 31 vertical levels and a three-dimensional semi-Lagrangian advection. The assimilation system is a special version of the operational ECMWF scheme. As opposed to the NCEP system, an intermittent statistical OI analysis is used, which requires model initialization. The data archive available for assimilation contains all observations acquired from the WMO Global Telecommunications System since 1979. The database is augmented by TOVS and NESDIS data, cloud track winds, COADS observations over the oceans, observations from various special field programs, PAOBS reports from Australia and supplementary TEMP and AIREP data (Stendel and Arpe 1997; Gibson et al. 1997). As in the NCEP scheme, data are converted into BUFR and subjected to quality control. SSTs and sea ice cover are based on the same basic datasets as for the NCEP reanalysis, and ice concentration is treated in the same way, with points above 0.55 concentration taken as fully ice covered. Albedo is set to 0.55 for sea ice. By contrast, snow cover is not based on satellite data, but is initialized at every analysis time from station snowfall and snow depth data, snow cover climatology, and the model’s snow depth. Orography is given by a spectral T106 fitted version of the gridpoint averages of a 1/6° × 1/6° field, with the mean orography supplemented by additional fields describing height, orientation, anisotropy, and slope of the subgrid orography. This was found to improve precipitation simulations in mountainous areas.

The land surface scheme follows Viterbo and Beljaars (1995), with an additional initialization of the soil water content (Viterbo 1995). The prognostic cloud scheme developed by Tiedtke (1993) is used. Tendencies in specific humidity, cloud water content, and cloud cover rely on advection in gridpoint space rather than spectral space. The convergence of the meridians at the poles and therefore the shrinkage of grids in the east–west direction is circumvented by the introduction of a reduced Gaussian grid. Grid points are removed from the latitude lines provided that the resulting longitudinal grid length does not exceed the grid length at the Gaussian latitude nearest the equator. By reducing the number of grids as a function of increasing latitude, the east–west distance is maintained, reducing computing time without sacrificing accuracy (Hortal and Simmons 1991). Unlike the modified Arakawa–Schubert scheme used in the NCEP reanalysis, the ERA model uses the convective formulation of Tiedtke (1989). Dynamic precipitation is derived using the same general approach as in the NCEP model (Gibson et al. 1997).

3. Validation climatology

a. Legates and Willmott climatology

Validation relies on an improved version of the Legates and Willmott (1990, hereafter LW) climatology. The original long-term monthly climatology, provided on a 0.5 × 0.5° lat–long grid, combines a number of different data sources. The terrestrial database draws from the archives of Wernstedt (1972), Willmott et al. (1981a,b), and Spangler and Jenne (1984). The first two contain long-term monthly averages for about 17 000 and 14 000 stations, respectively, while the third contains data for about 3500 stations. The combined archive is largely representative of the period 1920–89 but is biased to the more recent years. Terrestrial precipitation estimates north of 60°N are based on several hundred stations, with the best coverage for Scandinavia. Oceanic precipitation estimates north of 60°N are provided only for the Atlantic basin south of the sea ice margin (see Figs. 1 and 2 in Legates and Willmott 1990). These ocean estimates are taken from Dorman and Bourke (1978) and cover the period 1950–72. They are based on empirical relationships between observed monthly rainfall totals and “current weather” synoptic codes at coastal and island stations, which are then applied to synoptic weather reports recorded on board ships.

Corrections were applied to the terrestrial network to account for gauge undercatch due to wind, evaporation from gauges, and wetting losses. The largest corrections are for winds [see Legates and Willmott (1990) and Legates (1987) for details]. The crux is to estimate the wind speed at the gauge orifice during precipitation events. Mean monthly climatological wind speeds were used to estimate wind speed during precipitation events at anemometer height from empirical relationships based on the form of precipitation and the number of days in a month when precipitation exceeds 1 mm. Precipitation form (solid vs liquid) was estimated from monthly air temperature, while the number of days exceeding 1-mm precipitation was estimated from the total monthly precipitation, air temperature, and the number of days in the month. Wind speed for precipitation events at the height of the gauge was then estimated using an exposure coefficient, an assumed roughness length, and a logarithmic wind speed profile. Another set of empirical fits, using this orifice wind speed, gauge type, and the form of precipitation, defined the correction factors for rain and snow.

b. Additional precipitation data

In our opinion, the gauge-corrected LW climatology is the best dataset of observed Arctic precipitation, and it has been used in previous studies for this region (e.g., Stendel and Arpe 1997; Walsh et al. 1998). However, it has obvious shortcomings. In particular, the grid estimates for the central Arctic Ocean are based largely on (often excessive) extrapolation from coastal stations. Furthermore, in representing a global climatology, the Arctic was not specifically addressed to assure that all available data were incorporated with corrections specific to this region. Consequently, we have improved the LW climatology by blending it with data for the Arctic Ocean and station values with the best available gauge corrections.

For the Arctic Ocean, we use a recently released dataset of precipitation from the FSU NP series of drifting stations. Thirty ice stations were manned between 1950 and 1991 (NP-2 through NP-31; NP-1, the first ice station, operated during 1937–38). The average duration of each station was 2.4 yr. The records, assembled by Russian scientists at the Arctic and Antarctic Research Institute, are available on a CD-ROM compiled by the University of Washington’s Polar Science Center and the University of Colorado’s National Snow and Ice Data Center (NSIDC). We assembled the daily records into monthly totals by year, taking the monthly position as the average position for the month. Monthly totals based on fewer than 25 days were discarded. Monthly totals that were retained were then adjusted upward to account for missing days. The distribution of the monthly observations is shown in Fig. 1. The precipitation observations were made using the Tretiakov gauge. The instrument is argued as providing for reasonably accurate solid precipitation measurements but may suffer from problems under high wind speeds when snow from the surface is blown into, as well as out of, the gauge. No wind corrections have been applied. There is a correction for wetting losses (Colony et al. 1997).

For terrestrial regions, we utilize two different archives. The first, described by Groisman et al. (1991), represents monthly time series for 622 stations in the FSU. The majority of the records begin in the 1940s but some extend back to 1891. The correction procedures account for changes in gauge type, winds, and wetting losses. We also use data for 34 stations north of 60°N in Canada (P. Louie 1998, personal communication). Adjustments were made for wetting losses, winds, and trace events (J. R. Metcalfe et al. 1994, personal communication). We selected from both archives those stations having at least 10 years of data for every month, and then compiled climatological means. These climatological land station means, along with the monthly totals by year from the NP stations, were then blended with the LW climatology as described in the next section. The selected stations north of 60°N are shown in Fig. 1.

c. Blending procedures

A two-step procedure was used. A Cressman interpolation (Cressman 1959) was first used to translate the additional land station and NP data to the same latitude–longitude grid used in the original LW climatology. The Cressman interpolation has the form
i1520-0442-13-1-182-eq1
where PLW is the precipitation at a desired LW grid point;Ps is the precipitation at a given station, s; Ws is the weight given to that observation; d is the distance between the observation and the LW grid; and N is the search radius at which the weight goes to zero. Typically, an initial search radius is set, with the summations performed using all observations falling within that radius. If no observations fall within the initial radius, the search is extended over larger values of N.

We use a single 500-km search radius, requiring that the interpolated value be based on at least three observations. Otherwise, the grid point was coded as missing. As we use for input long-term monthly totals from land stations and monthly totals by year from the NP stations, this means that for grid points over the Arctic Ocean away from the coast, interpolated values are based on data for at least three different years, while over land, values are based on long-term means from at least three different stations. Coastal grids tend to get weight from both the NP and coastal land stations. Using long-term land station means as opposed to values for individual months and years assures that the interpolations over land are not biased by those stations having longer records. While by our procedure, the individual monthly means (by year) from the NP data are weighted the same as the land values (i.e., they are effectively treated as climatological means) this was felt necessary to provide as much coverage as possible over the ocean.

These gridpoint monthly means were then appended to the gridpoint values representing the LW climatology. The combined dataset was then reinterpolated to a 100- × 100-km version of the NSIDC north polar equal-area scalable earth grid (Armstrong and Brodzik 1995), again using a Cressman interpolation with a 500-km sphere of influence. This procedure gives equal weight to the original LW climatology and the new observations. As just discussed, the individual monthly means by year from the NP data are treated as climatological values. Giving full weight to the NP data over grids for which at least three years of data were available (by masking out grid points in the LW climatology) resulted in spatially noisy fields.

Figure 2 shows the differences in January and July precipitation between our modified climatology (hereafter LWM) and the original LW product. For January over Canada and the central Arctic Ocean, inclusion of the additional data acts to increase precipitation totals, especially in the latter region where, locally, totals are more than doubled. Over Eurasia, there is a mix of positive and negative adjustments. Not surprisingly, adjustments are smaller for July, when most precipitation falls in liquid form. Note, however, that both January and July (as well as most other months) show large positive adjustments over east-central Eurasia, where a local minimum is shown in the original LW field. From Fig. 1, it is clear that there are several stations in this region represented in the Groisman et al. (1991) archive. Apparently, Legates and Willmott (1990) either did not have data from these stations or their gauge corrections were biased.

The mean NCEP and ERA fields were interpolated to the same grid. The Cressman interpolation had the added benefit of smoothing the spurious “blotchy” pattern of lobes in precipitation in the NCEP model. According to Stendel and Arpe (1997) this is not due to the Gibbs phenomenon but rather to the poor approximation of horizontal moisture diffusion in the boundary layer when transforming from sigma to pressure coordinates. NCEP has produced corrected fields, but according to Cullather et al. (1999), these are still prone to problems and are overly dry. NCEP suggests that another fix is to average the data over a sufficiently large area. Our interpolation largely, but not entirely, eliminates this problem.

We are still left with a less-than-ideal validation climatology. The climatology is least reliable over the northern North Atlantic, based on interpolation from coastal sites along Scandinavia (for which coverage in the original LW climatology is good), a few coastal stations along Greenland, Iceland data, and the empirically derived Dorman and Bourke (1978) values. Clearly, we would also like additional gauge-corrected data for the North American side. Our interpolations also smooth real local variations in precipitation. Furthermore, as the validation climatology is pieced together from data representing different time periods, with points for the Arctic Ocean often based on only a few different years, it may not be entirely comparable in a statistical sense to the 10-yr (1979–88) reanalysis averages. Regardless, we consider our climatology to provide an improved representation of the large-scale field of Arctic precipitation allowing for useful assessments of the model output.

4. Total precipitation

a. Mean annual fields

Figure 3 shows the distribution of annual mean precipitation from the LWM climatology and corresponding NCEP and ERA fields. The LWM climatology (Fig. 3a) shows the highest precipitation totals off the southeast coast of Greenland (locally >1400 mm) with amounts decreasing northeast to about 400 mm near Novaya Zemlya. This pattern manifests frequent cyclone activity associated with the Icelandic low and the poleward decay of the mean North Atlantic cyclone track. High totals are also depicted over southern Alaska in association with the Aleutian low. The lowest totals (<150 mm) are found over the Beaufort Sea and northern Canadian Arctic archipelago, where cyclone activity is common only in summer. The LWM climatology is in qualitative agreement with the annual map of Bryazgin (1976) (which does not contain the complete NP dataset) but with the region of minimum precipitation shifted slightly south and east. The original LW climatology shows a larger area with <150 mm.

Comparisons with the LWM climatology confirm earlier findings (Serreze and Maslanik 1998) that the NCEP model captures the general spatial patterns of precipitation (Fig. 3b), with maxima over the Atlantic side of the Arctic and south of Alaska and minimum totals over the central Arctic. The “busy” pattern reflects remaining problems in the original NCEP precipitation fields noted earlier. The mean field from the ERA model (Fig. 3c) is much smoother. This model captures the precipitation maxima along the Atlantic side and south of Alaska, as well as the minimum over the Beaufort Sea and Canadian Arctic archipelago. However, absolute minimum values are shown over northern Greenland. Overall, the ERA field looks decidedly better.

Figure 4 illustrates differences in the model averages with respect to observations (modeled minus observed). Except for the small bull’s-eye just east of southern Greenland, the NCEP model (Fig. 4a) underpredicts precipitation over the Atlantic side of the Arctic. Amounts are also underestimated south of Alaska. Elsewhere, the model shows higher amounts. The oversimulation is worst for land areas where the model locally depicts over twice as much precipitation as observed. The central Arctic Ocean errors of typically 50–100 mm agree with findings by Serreze and Maslanik (1998) based on qualitative comparisons with Bryazgin’s (1976) results. The ERA model (Fig. 4b) also underrepresents precipitation over the Atlantic side and south of Alaska. ERA simulates too little precipitation over Eurasia and most of Canada. Agreement is good over the central Arctic. The apparent precipitation underestimates by both models in the Atlantic sector must be viewed with the caveat that the LWM climatology in this region is based in part on empirical estimates from the synoptic current weather codes. Furthermore, precipitation in this region is notoriously difficult to estimate (D. Bromwich 1998, personal communication) due to strong winds and the orographic effects of Greenland. The results in Fig. 4 again demonstrate the superior performance of the ERA model.

b. Seasonality of precipitation

Figure 5 shows the LWM precipitation fields for January and July. The January field (Fig. 5a) is qualitatively similar to the annual mean pattern. Precipitation ranges from a high of 180 mm southeast of Greenland to typically 20 mm over the central Arctic Ocean, Canadian Arctic archipelago, and east-central Eurasia. High values are also found south of Alaska. The Atlantic and Alaskan maxima largely disappear in July (Fig. 5b). Precipitation is more uniform across the Arctic with markedly higher totals as compared to January over land areas. This reflects changes in synoptic activity (Serreze 1995); although the Icelandic low, North Atlantic storm track, and Aleutian low weaken during summer, cyclogenesis becomes common over land along an Arctic front (north of the polar front), increasing precipitation in these areas. These systems, along with those associated with the weakened but still present North Atlantic storm track, tend to migrate into the central Arctic Ocean where they occlude, dropping precipitation in this region as well. Convective precipitation is also likely over land areas, especially for the southern part of the domain (see section 5).

January precipitation fields from both models and departures from observed totals are provided in Fig. 6. The modeled distributions for January (Figs. 6a,c) again broadly reproduce the observed field. The corresponding January departures (Figs. 6b,d) reflect the annual results. Notably, the NCEP model again yields too little precipitation over the Atlantic side except east of southern Greenland. The ERA model also underpredicts over the Atlantic side, but the performance over land is quite good.

Corresponding results for July (Fig. 7) reveal marked problems with the NCEP model. NCEP displays a very pronounced terrestrial maximum, with amounts over 100 mm in many areas (Fig. 7a). The ERA distribution is more realistic (Fig. 7c). As shown in the difference field (Fig. 7b), the model’s precipitation is more than twice the observed totals over Canada and parts of Eurasia. Errors are much smaller for the ERA model (Fig. 7d), largest (about 30 mm) over northwest Canada.

To summarize the ability of the models to capture the seasonal cycle in precipitation, monthly pattern correlation coefficients were computed between the modeled and LWM fields for different regions (Fig. 8) along with corresponding monthly precipitation totals (Fig. 9). The regions are based on grid points encompassing the entire Arctic domain (north of 60°N) and for the North Atlantic–Scandinavia, (60°–75°N, 45°W–15°E), North America (60°–75°N, 180°–45°W), Eurasia (60°–75°N, 20°E–180°), and the central Arctic (north of 75°N). Figure 8 also provides the monthly pattern correlations between the two models.

For winter months, both models correlate quite highly with the LWM fields as evaluated over the entire domain (>0.80) showing that despite errors in precipitation magnitude, the spatial patterns are reasonably well simulated. The two model fields also correlate highly with each other. Winter correlations between the models and observations are also generally strong for most subregions, illustrating that the high correlations as computed over the entire domain are not simply anchored by the ability of both models to reproduce the precipitation peaks over the Atlantic side and south of Alaska. The ERA winter correlations are higher.

Except for the Eurasian sector, correlations fall for both models during summer. For the North Atlantic–Scandinavia sector, both the NCEP and ERA coefficients drop to below 0.5 in August, while for the central Arctic, the August NCEP correlations are negative, and below 0.5 for ERA. A large part of the central Arctic has a rather flat summer precipitation distribution in the LWM climatology as well as in both models. Hence, even small departures from observed gridpoint values will reduce the correlations. In the sense that both models depict fairly flat fields, it could be argued that despite the low correlations they perform reasonably well. The higher central Arctic correlations in winter stem from the ability of the models to capture precipitation increases toward the Atlantic side associated with the strong North Atlantic cyclone track in this season. The two model fields continue to correlate fairly highly with each other in summer, except for the central Arctic.

It is clear from Fig. 9 that the ERA model captures the seasonal cycles quite well, both in terms of precipitation magnitude and the timing of maxima and minima, but with two exceptions. For the central Arctic, the model underpredicts amounts for all months except May and June. Similarly, the forecast totals are too low over the North Atlantic–Scandinavia sector. The model performance is especially good over North America. The NCEP predictions have generally larger errors. In particular, totals are excessive for land areas during summer (noted earlier for July), and there are also large positive errors over the central Arctic. As discussed shortly, these large terrestrial overestimates reflect convective precipitation. Interestingly, everywhere except the North Atlantic–Scandinavia sector, the timing of the precipitation maximum is in error by one month. The observations show August maxima, while the model depicts July maxima. Serreze and Maslanik (1998) found that for the central Arctic, the NCEP model yielded a July precipitation maximum for every year as examined over the period 1986–93.

5. Convective precipitation

a. Seasonal characteristics

Both models predict dynamic (grid point) and convective precipitation separately. Dynamic precipitation is calculated similarly in the two models. For convective precipitation, NCEP relies on a simplified Arakawa–Schubert scheme while ERA uses the Tiedtke (1989) formulation.

The January and July convective precipitation totals from the two models are shown in Figs. 10 and 11, respectively. For January, both depict convective precipitation only over the ice-free parts of the northern North Atlantic. This is to be expected, for although the downwelling solar radiation flux is small at these latitudes in January and absent at higher latitudes, the comparatively warm ocean surface underlying a cold atmosphere will foster instability. Note, however, the large differences in the convective precipitation totals. Totals locally exceed 60 mm in NCEP (Fig. 10a) compared to maxima of only 30 mm in the ERA reanalysis (Fig. 10b). Recall that both models utilize the same SSTs and sea ice extent and both are constrained by comparable assimilation data.

With the increase in the solar flux during spring and summer, and disappearance of the snow cover, one would expect an increase in the convective component over land areas. In turn, as the temperature contrast between open water and the atmosphere decreases, there should be a decrease in the convective component over the ice-free ocean. This seasonal change is revealed in the July fields. Looking first at the ERA results (Fig. 11b), convective precipitation is essentially absent over ocean areas but over land ranges from less than 10 to about 30 mm, highest over the extreme southern part of the domain over western Eurasia. The convective component typically composes 20%–40% of total (dynamic plus convective) July precipitation. The NCEP convective totals are much higher, ranging from 30 to 140 mm away from the coast (Fig. 11a), in many areas representing nearly all July precipitation. Note the general correspondence between the regions of maximum convective precipitation over land and the largest overestimates in total precipitation for this month (Fig. 7b). The NCEP model shows convective precipitation over land areas beginning in April and extending through September, maximized in June and July. The ERA model shows significant convective precipitation only for June and July.

Although we have no validation data for convective precipitation in the Arctic, information on thunderstorm days is provided in the summaries of Lydolph (1977) and Bryson and Hare (1974). For July, thunderstorm days over Eurasia range from less than two north of 70°N to about seven over the central part of our domain at 60°N. Corresponding thunderstorm days over the North American Arctic are upward of three at 60°N near Great Slave Lake. In terms of personal observations, the first author’s field notes for 1982 remark on the development of cumulonimbus and associated precipitation at 82°N over the Hazen Plateau of Ellesmere Island. While it is difficult to draw conclusions, it is apparent that convective precipitation does occur with reasonable frequency in low Arctic latitudes and occasionally in the high Arctic. Nevertheless, it is obvious that since the convective component alone in the NCEP model for June and July for most land areas exceeds total observed precipitation for these months (e.g., Fig. 5b) and that, for many areas, nearly all precipitation is convective in origin, the model is seriously in error.

b. Diagnosis of differences

Trenberth and Guillemot (1998) undertook a global analysis of the atmospheric and hydrologic cycle in the NCEP reanalysis. They found substantial biases in the moisture fields, associated with both model physics and the quality of the analyses. For example, systematic biases were noted between rawinsonde moisture amounts and those in the analyses. For the southeastern United States in summer, as well as in southeast Asia, the model cannot sustain observed high humidities, giving rise to excessive precipitation. This appears to be related to the properties of the convective parameterization. They also cite problems in soil moisture and errors in the land surface moisture budget as responsible for some errors in evaporation. For the Arctic, the atmospheric and surface moisture budgets are known to be out of balance (Cullather et al. 2000).

As the poor performance of the NCEP model in summer is tied to problems in convective precipitation, it is useful to provide some diagnosis of the differences in the two model predictions. We focus on July, for which problems with the NCEP model are most obvious, and start by comparing mean vertical profiles of humidity and temperature from NCEP forecasts, analyses, and rawinsonde data.

As a special project, NCEP has prepared a set of model forecasts, where at 5-day intervals (e.g., 1, 5, 10, . . . July) the model is initialized at 0000 UTC and outputs forecasts every 12 h over 8-day periods. For example, the 1 July set contains outputs at 12, 24, 36, 48, . . . , 192 h. For each set of forecasts initialized in July, we extracted first 12-h (valid at 1200 UTC) outputs of temperature and specific humidity (convened from relative humidity) over the period 1979–88 for six pressure levels (1000, 925, 850, 700, 500, and 400 mb). This results in six samples for every July and 60 cases over the 10-yr period. We then extracted the 1200 UTC analyses for the corresponding dates and times.

Data were averaged over two regions for Eurasia (60°–70°N, 60°–90°E; 60°–70°N; 120°–150°E) and one for Canada (60°–70°N, 105°–135°W). In terms of local time, the 1200 UTC forecasts in the middle longitude of each region are valid at 1700, 2100, and 0400, such that the forecast periods cover different parts of the diurnal cycle (0500–1700, 0900–2100, and 1600–0400, respectively). There are four to five well-distributed rawinsonde stations in each region, implying a good observational database for assimilation. Mean rawinsonde profiles corresponding to the forecasts and analyses were compiled. The main NCEP archive contains 6-h analyses from which 6-h forecasts are generated. While the 12-h forecasts examined here are hence not entirely comparable, they should still give a good indication of the behavior of the model. Furthermore, it is only at 12-h intervals (0000 and 1200 UTC) that the rawinsonde data are reported.

Results for east-central Eurasia, representative of the other two regions, are shown in Table 1. There is a slight tendency for the forecasts to be drier and cooler than the analyses, especially at low to middle levels, but t tests reveal that with few exceptions these differences are not statistically significant. Consequently, there is no marked model drift over the 12-h forecast period. By contrast, for all three regions the analyses are significantly wetter than the observations at 1000 and 925 mb and up to 850 mb in the east-central Eurasia sector. It is only for the 60°–90°E sector (central Eurasia) at 850 mb that the observations are wetter. Observed temperatures are also significantly different from the analysis values, primarily at low levels, pointing to a reduction in low-level stability as compared to observations. In the Canadian sector, for which the analysis is valid at 0400 LT, the model also fails to capture the observed temperature inversion between 1000 and 925 mb.

Given that multiple rawinsonde stations are distributed in each region, we would expect the analyses to lie between the forecasts and observations. Our results suggest that humidity and temperature data at low levels are being rejected by the assimilation model, such that the analyses are heavily dependent on the model forecast (first guess). In that the amount of convective precipitation will relate to the availability of low-level moisture and stability, our findings are consistent with the results in Fig. 11.

To maintain high convective precipitation rates, the low-level moisture must be replenished through either moisture flux convergence or evaporation. Cullather et al. (2000) note a number of differences between the NCEP and ERA models regarding local patterns of mean annual moisture flux convergence, although for large-scale averages (e.g., over 70°–90°N) the two models are in good agreement. While we cannot discount problems with the moisture flux convergence, we can cite problems in modeled evaporation as likely contributing to the errors in convective precipitation.

Figure 12 provides the mean total July evaporation from the two models. The NCEP totals are small over the Arctic ocean, typically 20 mm (Fig. 12a) This illustrates how the melting sea ice limits the saturation vapor pressure near the surface. Mean evaporation is much higher over land areas, exceeding 100 mm for the southern parts of the domain over both continents. The ERA totals are considerably smaller (Fig. 12b), with the NCEP–ERA differences most pronounced over land, typically exceeding 30 mm and locally much larger (Fig. 12c). Large differences in model evaporation are also shown by Stendel and Arpe (1997) for the Mackenzie River watershed. Pronounced contrasts in evaporation are apparent from May through August, roughly corresponding with the snow-free period; those for October through April are much closer. Assessing the correspondence with snow cover is unfortunately precluded. Because of a processing error, the NCEP model uses 1973 snow cover for the entire period examined here. By comparison, differences in evaporation over ice-free ocean regions are fairly small for all months.

Deciding which model simulates the more correct summer evaporation rates is made difficult by sparse validation data. There are a number of estimates for both land and ocean areas. These are based variously on simple energy budget considerations (e.g., Khrol 1976; Korzun 1976; Zubenok 1976; Lydolph 1977; Fisheries and Environment Canada 1978), the difference between measured runoff and precipitation for individual watersheds (e.g., Kane et al. 1992), and from direct measurements as part of field experiments (e.g., Ohmura 1982, 1984; Weller and Holmgren 1974). Annual means are the most commonly provided values. Since most evaporation over land occurs during summer, it is useful to make some general comparisons with annual values from different sources. These are provided in Table 2, sampled for seven locations along 65°N along with corresponding model values. None of the estimates were available digitally, so all estimates had to be estimated from contour maps.

It is immediately apparent that the “observed” totals themselves range widely for the same region, in some cases by over 100 mm. Although this arises in part from difficulties in estimating values from the contour maps, it more likely reflects different (and generally poorly documented) methods of estimating evaporation. It nevertheless comes out clearly that the NCEP values exceed even the highest estimates by 100 mm or more and that the ERA values are much lower and closer to observations. Lydolph (1977) lists monthly estimates for stations in the FSU. With few exceptions, the NCEP values are much higher for all months, sometimes by a factor of 2, with the ERA values in closer agreement (not shown). We conclude that the NCEP evaporation rates are seriously in error. As high evaporation rates would act to replenish water vapor at low levels lost due to rainout, these findings are consistent with those in Table 1 and Fig. 11. Note in this sense the close correspondence between the patterns of evaporation in the NCEP model and convective precipitation.

Numerous aspects of the land surface treatments, including vegetation distributions, roughness lengths, plant stomatal resistance, and soil moisture could contribute to these high evaporation rates and, in turn, feed back on precipitation. Soil moisture in particular merits consideration. Soil moisture in the NCEP model is treated according to Pan and Mahrt (1987). The salient aspect of this scheme is that the moisture is updated by the modeled precipitation. If the soil moisture (for which there is little information in the Arctic) is too high, this would promote excessive evaporation. This would foster excessive low-level moisture and precipitation, keeping soil moisture high, feeding back to keep evaporation and precipitation rates high. This problem will tend to be avoided in the ERA model, where soil moisture is initialized by a simple nudging method using the difference between the model first guess and analysis value (analysis increments) of the near-surface atmospheric humidity. If the analysis value is greater (less than) the first-guess value, the soil moisture is increased (decreased) (Viterbo and Courtier 1995; Mahfauf 1991). As the soil moisture is constrained to some extent by observations (e.g., humidities assimilated from rawinsonde data), evaporation would tend to be more realistic.

Also meriting scrutiny is the NCEP radiation budget. As part of a recent study of downwelling shortwave radiation fluxes, Serreze et al. (1998) briefly examined NCEP total cloud cover. As averaged for the region north of 70°N, the NCEP cloud cover has an annual average of about 40% and exhibits a weak seasonal cycle, ranging by about 10% from a winter–spring minimum to a summer–autumn maximum. This compares to annual mean values from surface and satellite-derived climatologies of typically 70%–80% with a summer maximum, largely due to the development of persistent, low-level stratus. The problem of too little cloud cover extends to lower Arctic latitudes. Figure 13a shows the average NCEP cloud cover for July, based on data for 1979–88. Terrestrial values range from 40% to 50%, compared with surface and satellite-based measurements, which are 15%–20% higher (Serreze et al. 1997). This results in large oversimulation of downwelling shortwave radiation in the NCEP model for all months except during polar darkness. For June and July, the errors in the shortwave flux are at least 50 W m−2 over land and larger over the Arctic Ocean.

By comparison, the ERA total cloud fractions (Fig. 13b) are typically 0.10 greater over land areas with larger differences over the ocean (Fig. 13c) and hence closer in line with observations. Walsh and Chapman (1998) arrive at a similar conclusion, showing that for the Arctic Ocean, the ERA cloud fractions are closer to observed values from the NP drifting stations. In apparent agreement with the greater cloud amounts, the ERA downwelling shortwave radiation fluxes are much lower. The July NCEP fluxes exceed the ERA values by at least 50 W m−2 over land with larger differences over the central Arctic Ocean (Fig. 14).

It is hence reasonable to suspect that the overly vigorous summer hydrologic cycle of the NCEP model is also related to excessive evaporation associated with these high radiation fluxes. The excessive solar heating may also be adversely impacting low-level temperature gradients, which would further the likelihood of convective precipitation. This problem would not give rise to convective precipitation over the sea ice cover as the surface temperature during summer melt is not free to rise and sea ice extent is initialized to observed distributions. It is not clear, however, why excessive convective precipitation over land would occur in association with limited cloud cover. A reasonable conclusion is that the cloud cover that is present is primarily convective and that stratiform cloud, which would decrease the shortwave flux, is underrepresented. While an archiving oversight by NCEP precludes assessment of the convective cloud component, poor representation of Arctic stratus is known to be a pervasive problem in models. Furthermore, despite the more realistic downwelling shortwave fluxes in the ERA model, the apparent association with the greater cloud amounts needs further clarification in terms of model differences in cloud optical properties, as well as clear-sky scattering treatments. Walsh and Chapman (1998) show that over the central Arctic Ocean, ERA net shortwave radiation fluxes during summer are little different between clear and cloudy conditions (i.e., the cloud forcing is too weak) although a stronger cloud forcing is observed over land (J. E. Walsh 1998, personal communication).

However, further investigation reveals that the excessive solar heating may be in part compensated for. Over land, the model albedos range from 0.25 increasing to 0.35 over Eurasia, compared with values from International Satellite Cloud Climatology Program (ISCCP) data (Schweiger and Key 1994) of 0.10–0.20. While the ISCCP data are certainly open to question, the NCEP albedos may well be in error, which would at least partly correct for the excessive solar flux. Along the line of competing errors, it is also of interest that both models depict the same general patterns of July net radiation (with typical differences of 10 W m−2) that are reasonably in accord with ISCCP values.

In summary, our results suggest that for the Arctic summer over land areas, the NCEP analyses themselves are too wet at low levels with temperature lapse rates that are too negative. Both problems suggest that assimilation data are being rejected such that the analyses rely heavily on the model first guess. While further work is needed to resolve this issue, we can speculate as to how the analysis problems link with other model components to promote excessive summertime convective precipitation. The results are suggestive of a “flywheel effect.” The analyses are too wet at low levels, promoting excessive precipitation, which keeps soil moisture high. High soil moisture is reinforced as the soil moisture is updated by the model precipitation. Evaporation rates consequently remain high, replenishing the low-level moisture, feeding back to keep precipitation rates high. The flywheel may in part owe its existence to excessive solar heating in promoting strong evaporation and influencing low-level temperature gradients, although this may be countered by high albedos.

Other factors that remain to be examined are possible problems with the convective parameterization itself (cf. Trenberth and Guillemot 1998) and errors in moisture flux convergence. In this regard, we stress that the arguments presented above do not apply to ocean sectors. For example, Fig. 9 reveals that both the NCEP and ERA forecasts are too low throughout the year in the North Atlantic–Scandinavia region. While observed precipitation totals in this area are open to question, it may be that both models give insufficient moisture flux convergence. On the other hand, while both models show almost no summer convective precipitation, reflecting small air–sea temperature differences and limited evaporation rates, they differ significantly in their simulation of winter convective totals (Figs. 10–12), implicating the convective schemes. Another problem area is the central Arctic. While, like land areas, the NCEP model simulates too much summer precipitation, this is clearly not related to convection.

6. Summary and conclusions

An improved precipitation climatology for the Arctic is developed by blending the Legates and Willmott (1990) product with observations from the North Pole archive and gauge-corrected station data for the FSU and Canada. The improved archive is used to examine the accuracy of modeled precipitation fields from the NCEP and ERA reanalyses efforts, based on mean fields over the period 1979–88.

Both models capture the major spatial features of annual mean precipitation, including high precipitation totals over the Atlantic side of the Arctic and south of Alaska, and minimum precipitation over the central Arctic. Both undersimulate precipitation over the Atlantic side. The NCEP model generally overpredicts annual amounts elsewhere. Problems are most severe over land, where the model locally depicts over twice as much precipitation as observed. The apparent underestimates by both models over the Atlantic must be viewed with the caveat that our validation climatology is least reliable in this region. Overall, the ERA forecasts are better.

Both models also capture the major aspects of the seasonal cycle. The high precipitation totals over the Atlantic sector and south of Alaska and low amounts over the central Arctic seen in the annual fields are correctly depicted as driven by the winter months. Pattern correlations between the observed and modeled precipitation fields generally exceed 0.80 during winter. The models are also correct in showing a summertime breakdown of these precipitation maxima with attendant increases over land areas and the central Arctic. As in the annual analysis, both models are prone to error in precipitation magnitude, but with larger errors in NCEP, which also has problems in depicting the timing of precipitation maxima. During summer, pattern correlations generally fall in both models.

The most significant problem is the tendency for the NCEP model to yield large overestimates of terrestrial precipitation during summer, due mainly to convective precipitation. This problem does not appear in the ERA fields. The NCEP model has known problems related to both the analyses and model physics. As is evident in the study of Trenberth and Guillemot (1998), it can be very difficult to pin down the specific causes for observed model errors. However, we can identify several different aspects of the model that, because of potential feedbacks, merit closer scrutiny. These include an overly wet boundary layer and spurious low-level temperature gradients in the analyses themselves, the convective parameterization, the treatment of soil moisture, evaporation, and radiative fluxes.

Stendel and Arpe (1997) present similar findings in their summary of model performance over the Mackenzie Basin. Their Fig. 17 shows both models depicting PE in summer as near zero. The tendency for PE to be small or even slightly negative is supported from aerological estimates via analysis of the vapor flux convergence using rawinsonde data (Walsh et al. 1994). The summer minimum in PE occurs despite the summer maximum in precipitation, illustrating that much of the precipitation results from within-basin recycling of water vapor. That both models yield reasonable estimates of summer PE is understood in that the excessive P in the NCEP model essentially balances the excessive E. Put differently, the NCEP model gives reasonably correct values of PE, but for the wrong reasons. This tendency is also evident in the recent analysis of Walsh et al. (1998). In conclusion, we suggest that the NCEP modeling community make efforts to 1) determine if humidity and temperature observations are in fact being rejected; 2) improve the cloud parameterizations, providing more realistic cloud cover distributions to reduce the summertime solar radiation flux; and 3) evaluate the treatment of soil moisture. Efforts are underway to validate output from NCEP/DOE AMIP-II reanalysis (1979–97), which incorporates fixes to a number of problems identified in the NCEP–NCAR effort, along with additional improvements.

Acknowledgments

This study was supported by NSF Grants OPP-9614297, OPP-9524740, and OPP-9732461. J. Walsh and W. Chapman are thanked for assistance in analysis of the ERA data. We also thank P. Louie (AES) for the Canadian precipitation data, P. Groisman (CMDL) for the FSU data, and J. Barsugli, J. Whittaker, and A. Loughe (CDC) for helpful input.

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 (C10), 14 485–14 498.

  • ARCUS, 1991: Arctic System Science Land/Atmosphere/Ice Interactions:A Plan for Action. Arctic Research Consortium of the United States, 48 pp.

  • Armstrong, R. L., and M. J. Brodzik, 1995: An earth-gridded SSM/I data set for cryospheric studies and global change monitoring. Adv. Space Res.,16, 155–163.

  • Arpe, K., H. Behr, and L. Dumenil, 1996: Validation of the ECHAM4 climate model and re-analysis data in the Arctic regions. Proc. Workshop on the Implementation of the Arctic Precipitation Data Archive (APDA) at the Global Precipitation Climatology Center (GPCC), Offenbach, Germany, WMO, 31–40.

  • Basist, A. N., and M. Chelliah, 1997: Comparisons of tropospheric temperatures derived from the NCEP/NCAR reanalysis, NCEP operational analysis, and the microwave sounding unit. Bull. Amer. Meteor. Soc.,78, 1431–1447.

  • Bryazgin, N. N., 1976: Srednegodovoe kolichestvo osadkov v Arktike s uchetom pogreshnosti osadkomerov. Tr. Arkt. Antarkt. Nauch.-Issled. Inst.,323, 40–74.

  • Bryson, R. A., and F. K. Hare, 1974: Climates of North America. Vol. 11, World Survey of Climatology, Elsevier, 420 pp.

  • Colony, R. L., and I. G. Rigor, 1993: International Arctic Ocean buoy program data report for 1 January 1992–31 December 1992. Tech Memo. APL-UWTM 29-93, University of Washington, Seattle, WA, 215 pp. [Available from Applied Physics Laboratory, 1013 NE 40th St., Seattle, WA 98195.].

  • ——, V. Radionov, and F. L. Tanis, 1997: Measurements of precipitation and snow at the Russian North Pole drifting stations. Polar Geogr.,34, 3–14.

  • Cressman, G. P., 1959: An operational objective analysis system, Mon. Wea. Rev.,87, 367–374.

  • Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: The atmospheric hydrologic cycle over the Arctic Basin from reanalyses. Part I: Comparison with observations and previous studies. J. Climate, in press.

  • Dey, C. H., and L. L. Morone, 1985: Evolution of the NMC Global Assimilation System: January 1982–December 1983. Mon. Wea. Rev.,113, 304–318.

  • DiMego, G., 1988: The National Meteorological Center Regional Analysis System. Mon. Wea. Rev.,116, 977–1000.

  • Dorman, C. E., and R. B. Bourke, 1978: A temperature correction for Tucker’s ocean rainfall estimates. Quart. J. Roy. Meteor. Soc.,104, 765–773.

  • Dorman, J. L., and P. J. Sellers, 1989: A global climatology of albedo, roughness length and stomatal resistance for atmospheric general circulation models as represented by the Simple Biosphere Model (SiB). J. Appl. Meteor.,28, 833–855.

  • Fisheries and Environment Canada, 1978: Hydrological Atlas of Canada. Fisheries and Environment Canada, Minister of Supply and Services, 31 pp.

  • Furgeson, H. L., A. D. O’Neill, and H. F. Cork, 1970: Mean evaporation over Canada. Water Resour. Res.,6, 1618–1633.

  • Genthon, C., and A. Braun, 1995: ECMWF analyses and predictions of the surface climate of Greenland and Antarctica. J. Climate,8, 2324–2332.

  • Gibson, J. K., S. Kallberg, A. Hernandez, S. Uppala, A. Nomura, and E. Serrano, 1997: ERA Description. ECMWF Re-Analysis Project Report Series, Vol. 1, European Centre for Medium-Range Weather Forecasts, 72 pp.

  • Gorshkov, S. G., Ed., 1983: Arctic Ocean (in Russian). Vol. 3, World Ocean Atlas, Pergamon Press, 189 pp.

  • Groisman, P. Y., V. V. Koknaeva, T. A. Belokrylova, and T. R. Karl, 1991: Overcoming biases of precipitation: A history of the USSR experience. Bull. Amer. Meteor. Soc.,72, 1725–1733.

  • Hibler, W. D., III, and J. Zhang, 1995: On the effect of sea-ice dynamics on ocean thermohaline circulation. Ann. Glaciol.,21, 361–368.

  • Hortal, M., and J. Simmons, 1991: Use of reduced Gaussian grids in spectral models. Mon. Wea. Rev,119, 1057–1074.

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

  • Kane, D. L., L. D. Hinzman, M. Woo, and K. Everett, 1992: Arctic hydrology and climate change. Arctic Ecosystems in a Changing Climate, F. Chapin III, Ed., Academic Press, 35–57.

  • Khrol, V. P., 1976: Isparenie s poverkhnosti Severnogo Ledovitogo Okeana. Tr. Arkt. Antarkt. Nauchno.-issled. Inst.,323, 148–155.

  • Korzun, 1976: Atlas of World Water Balance (in Russian). Gidrometeoizdat, 65 pp.

  • Ledley, T. S., 1991: Snow on sea ice: Competing effects in shaping climate. J. Geophys. Res.,96 (D9), 17 195–17 208.

  • Legates, D., 1987: A climatology of global precipitation. Publ. Climatol.,4 (1), 85 pp.

  • ——, and C. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected global precipitation. Int. J. Climatol.,10, 110–127.

  • Lydolph, P. E., 1977: Climates of the Soviet Union. Vol. 7, World Survey of Climatology, Elsevier, 443 pp.

  • Mahfauf, J. F., 1991: Analysis of soil moisture from near-surface parameters: A feasibility study. J. Appl. Meteor.,30, 1534–1547.

  • Matthews, E., 1985: Atlas of archived vegetation, land use, and seasonal albedo data sets. NASA Tech. Memo. 86199, NASA/Goddard Institute For Space Studies, New York, 53 pp. [Available from Goddard Space Flight Center, Greenbelt, MD 20771.].

  • Ohmura, A., 1982: Evaporation from the surface of the Arctic Tundra on Axel Heiburg Island. Water Resour. Res.,18, 291–300.

  • ——, 1984: Comparative energy balance study for arctic tundra, sea surface, glaciers and boreal forests. Geo-Journal,8, 221–228.

  • Pan, H.-L., and L. Mahrt, 1987: Interaction between soil hydrology and boundary layer developments. Bound.-Layer Meteor.,38, 185–202.

  • ——, and W.-S. Wu., 1994: Implementing a mass flux convection parameterization package for the NMC medium range forecast model. Preprints, 10th Conf. on Numerical Weather Prediction, Portland, OR, Amer. Meteor. Soc., 96–98.

  • Parish, D. F., and J. C. Derber, 1992: The National Center’s Spectral Statistical-Interpolation Analysis System. Mon. Wea. Rev.,120, 1747–1763.

  • Reynolds, R. W., and T. M. Smith, 1994: Improved global sea surface temperature analyses using optimum interpolation. J. Climate,7, 929–948.

  • Schweiger, A. J., and J. Key, 1994: Arctic Ocean radiative fluxes and cloud forcing based on the ISCCP-C2 dataset, 1983–1990. J. Appl. Meteor.,33, 948–963.

  • Serreze, M. C., 1995: Climatological aspects of cyclone development and decay in the Arctic. Atmos.–Ocean,33, 1–23.

  • ——, and J. A. Maslanik, 1998: Arctic precipitation as represented in the NCEP/NCAR reanalysis. Ann. Glaciol.,25, 429–433.

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

  • ——, J. A. Maslanik, and J. R. Key, 1997: Atmospheric and sea ice characteristics of the Arctic Ocean and the Sheba field region in the Beaufort Sea. Special Rep. 4, National Snow and Ice Data Center, Cooperative Institute for Research in Environmental Sciences, University of Colorado, 219 pp. [Available from National Snow and Ice Data Center, Campus Box 449, University of Colorado, Boulder, CO 80309-0449.].

  • ——, J. R. Key, J. E. Box, J. A. Maslanik, and K. Steffen, 1998: A new monthly climatology of global radiation for the Arctic and comparisons with NCEP–NCAR reanalysis and ISCCP-C2 fields. J. Climate,11, 121–136.

  • Spangler, W. M. L., and R. L. Jenne, 1984: World monthly surface station climatology. Scientific Computing Division, NCAR, Boulder, CO, 14 pp. [Available from National Center for Atmospheric Research, 1850 Table Mesa Drive, Boulder, CO 80307.].

  • Stendel, M., and C. Arpe, 1997: Evaluation of the hydrological cycle in reanalysis and observations, ECMWF reanalysis validation reports. Part 1. ECMWF Reanalysis Project Rep. Ser. 6, 50 pp.

  • Tiedke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev.,117, 1779–1800.

  • ——, 1993: Representation of clouds in large-scale models. Mon. Wea. Rev.,121, 1779–1800.

  • Trenberth, K. E., and C. J. Guillemot, 1998: Evaluation of the atmospheric moisuture and hydrological cycle in the NCEP/NCAR reanalyses. Climate Dyn.,14, 213–231.

  • Viterbo, P., 1995: Initial values of soil water and the quality of summer forecasts. ECMWF Newslett.,69, 2–8.

  • ——, and A. C. M. Beljaars, 1995: A new land parameterization scheme in the ECMWF model and its validation. J. Climate,8, 2716–2748.

  • ——, and P. Courtier, 1995: The importance of soil water for medium range weather forecasting: Implications for data assimilation. Workshop of Slowly Varying Components of Predictable Atmospheric Motions, Beijing, China, WMO, 121–130.

  • Walsh, J. E., and C. M. Johnson, 1979: An analysis of Arctic sea ice fluctuations, 1953–1977. J. Phys. Oceanogr.,9, 580–591.

  • ——, and W. L. Chapman, 1998: Arctic cloud–radiation–temperature associations in observational data and atmospheric reanalyses. J. Climate,11, 3030–3045.

  • ——, X. Zhou, D. Portis, and M. C. Serreze, 1994: Atmospheric contribution to hydrologic variations in the Arctic. Atmos.–Ocean,34 (2), 733–755.

  • ——, V. Kattsov, D. Portis, and V. Meleshko, 1998: Arctic precipitation and evaporation: Model results and observational estimates. J. Climate,11, 72–87.

  • WCRP, 1997: Proceedings of the Workshop on the Implementation of the Arctic Precipitation Data Archive (APDA) at the Global Precipitation Climatology Centre (GPCC), Offenbach, Germany, 10–12 July, 1996, Arctic Climate System Study, World Climate Research Programme. WCRP-98, WMO/TD 804, 127 pp. plus appendices. [Available from World Climate Research Programme, 7 bis, Avenue de la Paix, Case Postale 2300, CH-1211 Geneva 2, Switzerland.].

  • Weller, G., and B. Holmgren, 1974: The microclimates of the Arctic tundra. J. Appl. Meteor.,13, 854–862.

  • Wernstedt, F. L., 1972: World Climatic Data. Climatic Data Press, 522 pp.

  • Willmott, C. J., J. R. Mather, and C. M. Rowe, 1981a: Average monthly and annual surface air temperature and precipitation data for the world. Part 1: The eastern hemisphere. Pub. Climatol.,34 (1), 395 pp.

  • ——, ——, and ——, 1981b: Average monthly and annual surface air temperature and precipitation data for the world. Part 2: The western hemisphere. Pub. Climatol.,34 (1), 378 pp.

  • Woollen, J. S., E. Kalnay, L. Gandin, W. Collins, S. Saha, R. Kistler, M. Kanamitsu, and M. Chelliah, 1994: Quality Control in the Reanalysis System. Amer. Meteor. Soc., 13–14.

  • Zubenok, Z. T., 1976: Isparenie s sushi vodosbornogo basseina Severnogo Ledovitogo Okeana. Tr. Arkt. Antarkt. Nauchno.-issled. Inst.,323, 87–100.

Fig. 1.
Fig. 1.

Location of additional land station data for the FSU (+), Canada (○), and North Pole drifting station observations (*) used to improve the Legates and Willmott (1990) climatology. The North Pole data are plotted at the average monthly positions.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 2.
Fig. 2.

Difference in precipitation (mm) between the LWM and LW climatologies (LWM − LW) for (a) Jan and (b) Jul. The contour interval is 20 mm for Jan and 10 mm for Jul.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 3.
Fig. 3.

Annual mean precipitation (mm) from (a) LWM, (b) NCEP, and (c) ERA. The contour interval is 100 mm except for the dotted contour, which encloses areas with precipitation totals less than 150 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 4.
Fig. 4.

Model errors in annual precipitation (mm) expressed as modeled minus observed values for (a) NCEP and (b) ERA. Positive errors are shown by solid lines with negative errors shown by dotted lines. The contour interval is 75 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 5.
Fig. 5.

Mean precipitation (mm) from the LWM climatology for (a) Jan and (b) Jul. The contour interval is 15 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 6.
Fig. 6.

Mean modeled precipitation for Jan (mm) and corresponding model errors (expressed as modeled minus observed totals) for NCEP (a), (b), and ERA (c), (d). Positive errors are shown by solid lines with negative errors shown by dotted lines. The contour interval is 15 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 7.
Fig. 8.
Fig. 8.

Pattern correlations by month for different regions of the Arctic between observed and modeled fields and between the two models.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 9.
Fig. 9.

Seasonal cycles in observed and modeled precipitation (mm) for the same regions used in Fig. 8.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 10.
Fig. 10.

Jan convective precipitation (mm) for (a) NCEP and (b) ERA. The contour interval is 10 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 11.
Fig. 12.
Fig. 12.

Mean Jul evaporation (mm) for (a) NCEP and (b) ERA, and (c) NCEP–ERA differences. All differences are positive. The contour interval is 15 mm.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 13.
Fig. 13.

Mean Jul cloud cover (%) for (a) NCEP and (b) ERA, and (c) NCEP–ERA differences. Positive differences are shown by solid lines with negative differences shown by dotted lines. The contour interval is 10%.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Fig. 14.
Fig. 14.

Difference field (NCEP minus ERA) of the downwelling solar radiation flux for Jul (W m−2). All differences are positive. The contour interval is 10 W m−2.

Citation: Journal of Climate 13, 1; 10.1175/1520-0442(2000)013<0182:ROMAPF>2.0.CO;2

Table 1.

Comparison of forecast, analyzed, and observed temperature and specific humidity at 1200 UTC for the region banded by latitude 60°–70°N and longitude 120°–150°E (east-central Eurasia). Underlined values indicate statistically significant differences between analysis and forecast means (95% level). Values in bold indicate statistically significant differences between observed and analysis means (95% level).

Table 1.
Table 2.

Annual mean evaporation (mm) along 65°N from reanalyses and observations.

Table 2.
Save