• 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 , 155163.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., J. H. Ball, and P. Viterbo, 2003: Evaluation of the ERA-40 surface water budget and surface temperature for the Mackenzie River basin. J. Hydrometeor., 4 , 11941211.

    • Search Google Scholar
    • Export Citation
  • Bromwich, D. H., S-H. Wang, and A. J. Monaghan, 2002: ERA-40 representation of the Arctic atmospheric moisture budget. Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis, Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 287–297.

  • Cressman, G. P., 1959: An operational objective analysis system. Mon. Wea. Rev., 87 , 367374.

  • Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: The atmospheric hydrologic cycle over the Arctic basin from reanalyses. Part I: Comparisons with observations and previous studies. J. Climate, 13 , 923937.

    • Search Google Scholar
    • Export Citation
  • ECMWF, 2002: Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis. Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 443 pp.

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

  • Gleason, B. E., 2002: Data documentation for data set 9101. Global Daily Climatology Network, V1.0, National Climatic Data Center, 25 pp.

  • Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation measurement intercomparison. Final Rep. WMO TD-872, World Meteorological Organization, Geneva, Switzerland, 212 pp.

  • Groisman, P. Y., 1998: National Climatic Data Center data documentation for TD-9816: Canadian monthly precipitation. National Climatic Data Center, 21 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 , 17251733.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78 , 520.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kistler, R., and Coauthors, 2001: The NCEP-NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climatol., 10 , 111127.

    • Search Google Scholar
    • Export Citation
  • Mekis, E., and W. D. Hogg, 1999: Rehabilitation and analysis of Canadian daily precipitation time series. Atmos.–Ocean, 37 , 5385.

  • Panofsky, H. A., and G. W. Brier, 1963: Some Applications of Statistics to Meteorology. Mineral Industries Continuing Education, College of Mineral Industries, The Pennsylvania State University, 224 pp.

    • Search Google Scholar
    • Export Citation
  • Rogers, A. N., D. H. Bromwich, E. N. Sinclair, and R. I. Cullather, 2001: The atmospheric hydrologic cycle over the Arctic basin from reanalyses. Part II: Interannual variability. J. Climate, 14 , 24142429.

    • Search Google Scholar
    • Export Citation
  • SEARCH SSC, 2001: Study of Environmental Arctic Change, Science Plan. Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, 91 pp.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and C. M. Hurst, 2000: Representation of mean Arctic precipitation from NCEP–NCAR and ERA Reanalyses. J. Climate, 13 , 182201.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and A. J. Etringer, 2002: Representation of Arctic precipitation in ERA-40. Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis, Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 317–331.

  • Serreze, M. C., and A. J. Etringer, 2003: Precipitation characteristics of the Eurasian Arctic drainage system. Int. J. Climatol., 23 , 12671291.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., J. R. Key, J. C. 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 , 121136.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., M. P. Clark, and D. H. Bromwich, 2003: Monitoring precipitation over the Arctic terrestrial drainage system: Data requirements, shortcomings, and applications of atmospheric reanalysis. J. Hydrometeor., 4 , 387407.

    • Search Google Scholar
    • Export Citation
  • Shepard, D., 1984: Computer mapping: The SYSMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., D. Reidel, 133–145.

    • Search Google Scholar
    • Export Citation
  • Vose, R. S., R. L. Schmoyer, P. M. Steurer, T. C. Peterson, R. Heim, T. R. Karl, and J. Eischeid, 1992: The global historical climatology network: Long-term monthly temperature, precipitation, sea level pressure, and station pressure data. ORNL Tech. Rep. CDIAC-53, NDP-041, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, TN, 99 pp. plus appendixes.

  • Wilks, D. S., 1995: Statistical Methods in Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • Willmott, C. J., C. M. Rowe, and W. D. Philpot, 1985: Small-scale climate maps: A sensitivity analysis of some common assumptions associated with grid-point interpolation and contouring. Amer. Cartographer, 12 , 516.

    • Search Google Scholar
    • Export Citation
  • Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model output. Bull. Amer. Meteor. Soc., 78 , 25392558.

    • Search Google Scholar
    • Export Citation
  • Yang, D., 1999: An improved precipitation climatology for the Arctic Ocean. Geophys. Res. Lett., 26 , 16251628.

  • Yang, D., and Coauthors, 2001: Compatability evaluation of national precipitation gauge measurements. J. Geophys. Res., 106 , D2,. 14811491.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The Arctic terrestrial drainage (all shaded areas) and the watersheds of the Ob, Yenisey, Lena, and Mackenzie basins.

  • View in gallery

    Distribution of terrestrial gauge sites for the region north of 40°N in the blended archive.

  • View in gallery

    Time series of the number of terrestrial stations in the blended archive (see Fig. 2) reporting precipitation (1950–2000) for the regions north of 50° and 60°N.

  • View in gallery

    Fields of monthly mean precipitation (mm) from ERA-40 during the period 1979–93.

  • View in gallery

    Distribution of precipitation biases (% deviation from observations) for ERA-40, NCEP-1, GPCP, and ERA-15 for Jan, Apr, Jul, and Oct. Results are based on all grid cells except open-ocean regions. Positive values indicate higher precipitation relative to observations. Biases are computed from monthly means during the period 1979–93.

  • View in gallery

    Same as in Fig. 5, except based on the subset of grid cells for which (as averaged over 1979–93) there were at least four stations within two grid lengths of the cell center.

  • View in gallery

    Spatial patterns of precipitation biases (% deviation from observations) for ERA-40, NCEP-1, and GPCP for Jan, Apr, Jul, and Oct. Positive values indicate higher precipitation relative to observations. Biases are computed from monthly means during the period 1979–93.

  • View in gallery

    Mean annual cycles of precipitation (mm) for the Ob, Yenisey, Lena, and Mackenzie basins from observations: ERA-40, NCEP-1, GPCP, and ERA-15. Means are computed during the period 1979–93.

  • View in gallery

    Fraction of grid cells (y axis) having squared correlations with observed precipitation time series (1979–93) less than the value indicated on the x axis. Results are given for Jan and Jul for ERA-40, NCEP-1, GPCP and ERA-15 based on (left) all grid cells (excepting open-ocean areas) and (right) the subset of grid cells for which (as averaged over 1979–93) there were at least four stations within two grid lengths of the cell center.

  • View in gallery

    Time series (1979–93) of squared field correlations between observed precipitation and precipitation from ERA-40, NCEP-1, GPCP, and ERA-15 for Jan and Jul. Field correlations indicate the ability of each precipitation product to capture the spatial patterns of precipitation across the study domain.

  • View in gallery

    Fields of squared correlations between the time series (1979–93) of observed monthly precipitation and time series from ERA-40, NCEP-1, and GPCP.

  • View in gallery

    (Continued)

  • View in gallery

    (Continued)

  • View in gallery

    Time series of precipitation (mm) for the Ob basin from observations, ERA-40, andGPCP for Jan and Jul. Time series for observed precipitation are only given through 1993.

  • View in gallery

    Same as in Fig. 12, but for the Yenisey basin.

  • View in gallery

    Same as in Fig. 12, but for the Lena basin.

  • View in gallery

    Same as in Fig. 12, but for the Mackenzie basin.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 204 117 14
PDF Downloads 125 61 5

Northern High-Latitude Precipitation as Depicted by Atmospheric Reanalyses and Satellite Retrievals

View More View Less
  • 1 Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado
Full access

Abstract

Monthly precipitation based on forecasts from the new 40-yr ECMWF Re-Analysis (ERA-40) is evaluated for the north polar region (the region north of 45°N), the terrestrial Arctic drainage, and its four major watersheds: the Ob, Yenisey, Lena, and Mackenzie basins. Corresponding evaluations are performed for precipitation from the NCEP–NCAR reanalysis, the earlier 15-yr ERA (ERA-15), and satellite-derived estimates from the Global Precipitation Climatology Project (GPCP). Evaluations rely on an improved gridded dataset of precipitation derived from monthly gauge data during the period 1979–93. The available number of gauges has declined since 1993, making it difficult to perform evaluations for later years. ERA-40 depicts monthly precipitation much better than NCEP–NCAR. This is with respect to both lower mean biases and higher squared correlations between modeled and observed grid-cell time series. Squared correlations between monthly time series of ERA-40 and observed precipitation, averaged over the four major Arctic watersheds, typically range from 0.60 to 0.90. Performance over the central Arctic Ocean is poor in winter and spring, but improves in summer and autumn when precipitation amounts are higher. While the overall performance of ERA-40 is better than NCEP–NCAR, it offers no obvious improvement over ERA-15. In some respects, ERA-15 performs slightly better in summer. This lack of improvement may relate to difficulties in assimilating satellite radiances. All of the reanalyses provide better depictions of monthly precipitation than do the GPCP satellite retrievals. This applies to both land areas and the Arctic Ocean. There is no clear improvement in the GPCP estimates after 1987 when the Television Infrared Observational Satellite (TIROS) Operational Vertical Sounder (TOVS) data began to be used. The GPCP estimates are best in summer.

Corresponding author address: Mark C. Serreze, Cooperative Institute for Research in Environmental Sciences, University of Colorado, Campus Box 449, Boulder, CO 80309-0449. Email: serreze@kryos.colorado.edu

Abstract

Monthly precipitation based on forecasts from the new 40-yr ECMWF Re-Analysis (ERA-40) is evaluated for the north polar region (the region north of 45°N), the terrestrial Arctic drainage, and its four major watersheds: the Ob, Yenisey, Lena, and Mackenzie basins. Corresponding evaluations are performed for precipitation from the NCEP–NCAR reanalysis, the earlier 15-yr ERA (ERA-15), and satellite-derived estimates from the Global Precipitation Climatology Project (GPCP). Evaluations rely on an improved gridded dataset of precipitation derived from monthly gauge data during the period 1979–93. The available number of gauges has declined since 1993, making it difficult to perform evaluations for later years. ERA-40 depicts monthly precipitation much better than NCEP–NCAR. This is with respect to both lower mean biases and higher squared correlations between modeled and observed grid-cell time series. Squared correlations between monthly time series of ERA-40 and observed precipitation, averaged over the four major Arctic watersheds, typically range from 0.60 to 0.90. Performance over the central Arctic Ocean is poor in winter and spring, but improves in summer and autumn when precipitation amounts are higher. While the overall performance of ERA-40 is better than NCEP–NCAR, it offers no obvious improvement over ERA-15. In some respects, ERA-15 performs slightly better in summer. This lack of improvement may relate to difficulties in assimilating satellite radiances. All of the reanalyses provide better depictions of monthly precipitation than do the GPCP satellite retrievals. This applies to both land areas and the Arctic Ocean. There is no clear improvement in the GPCP estimates after 1987 when the Television Infrared Observational Satellite (TIROS) Operational Vertical Sounder (TOVS) data began to be used. The GPCP estimates are best in summer.

Corresponding author address: Mark C. Serreze, Cooperative Institute for Research in Environmental Sciences, University of Colorado, Campus Box 449, Boulder, CO 80309-0449. Email: serreze@kryos.colorado.edu

1. Introduction

Reliable estimates of precipitation with which to assess climate variability and change are difficult to obtain in northern high latitudes. The gauge network is sparse and has declined over the past decade. Accurate measurements are difficult to make. Turbulence around gauge orifices caused by winds reduces catch efficiency, especially for solid precipitation (Goodison et al. 1998; Yang et al. 2001). Snow can also be blown into gauges. Measurement errors may reach 50% and 100% in winter. Different gauge designs and shields are used to reduce wind effects but gauge and shield types differ between countries and even within countries. This leads to artificial discontinuities in precipitation fields. Although gauge data can be adjusted (e.g., Groisman et al. 1991; Yang 1999), the results are by no means perfect. Precipitation output from atmospheric reanalyses and from satellite retrievals represent potential additional data sources for monitoring high-latitude precipitation.

Reanalysis projects provide long time series of analyzed atmospheric fields and modeled surface fields based on frozen forecast and data assimilation systems. Analyzed fields, such as tropospheric pressure heights, represent an optimal blend of a short-term forecast and observations. Modeled (or forecast) fields, such as precipitation and surface radiation fluxes, are not directly influenced by observations of that variable. Using a frozen forecast and data assimilation system removes spurious jumps and trends present in archives from operational numerical weather prediction systems associated with changes in data assimilation techniques and models. Temporal inconsistencies are still present due to changes in observing networks (e.g., rawinsonde and satellite databases). Data from the European Centre for Medium-Range Weather Forecasts (ECMWF) 15-year Re-Analysis (ERA-15) and the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) effort have been widely used in studies of high-latitude and global climate. Here, we evaluate monthly precipitation for northern high latitudes as represented by the new 40-year Re-Analysis (ERA-40). ERA-40 performance is assessed relative to representations of precipitation from the NCEP–NCAR (hereafter NCEP-1) and ERA-15 reanalyses, as well as satellite-derived estimates from the Global Precipitation Climatology Project (GPCP).

Our evaluation has several aims. One is to assess the potential of using ERA-40 precipitation fields as a base for blending with gauge data, satellite-derived estimates, or other reanalysis variables, to yield improved fields of high-latitude precipitation over recent years for which the gauge network has degraded (cf. Serreze et al. 2003); hence, our corresponding interest in the satellite data. The GPCP (Huffman et al. 1997) has produced gridded fields of monthly precipitation since 1979 based on various satellite retrievals calibrated against observations. The Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) also provides gridded fields blending satellite, gauge, and reanalysis data (Xie and Arkin 1997). GPCP estimates of precipitation over open ocean, especially after 1987, are considered to be good (Huffman et al. 1997). However, the quality of these retrievals over northern high latitudes has received little attention. Finally, we aim to contribute to the developing Arctic System Reanalysis (ASR), a flagship activity of the interagency Study of Environmental Arctic Change (SEARCH) program Science Steering Committee (SEARCH SSC 2001). The evolving ASR will likely be based on the new NCEP Weather Research and Forecasting System. One step in ASR development is to establish the current state of the art in reanalysis that must be exceeded, presumably represented by ERA-40.

There have been a number of previous northern high-latitude assessments of precipitation and other hydrologic variables from ERA-15, NCEP-1, as well as the NCEP–Department of Energy Atmospheric Model Intercomparison Project 2 reanalysis (NCEP-2). In general, all reanalysis products capture the major spatial features of observed mean precipitation (Serreze and Hurst 2000; Serreze et al. 2003). Serreze and Hurst (2000) found that ERA-15 performs much better than NCEP-1. A significant problem with NCEP-1 is severe overestimation of precipitation over land areas in summer. This is caused by excessive convective precipitation, which appears to be linked to high evaporation rates, and, as shown by Serreze et al. (1998), excessive downwelling solar radiation that in June exceeds observed values by up to 80 W m−2. NCEP-1 also has a poorly formulated moisture diffusion that leads to anomalous local moisture convergence, causing “blotchy” patterns of precipitation associated with orography. Precipitation fields must be smoothed to make effective use of the data (Serreze et al. 2003). NCEP-1 effectively captures interannual variability of monthly precipitation in many areas of the Arctic but in general performs poorly. NCEP-2 addressed some of the known problems with NCEP-1. However, the Arctic performance of NCEP-2 is no better. Notably, it still greatly overestimates summer precipitation over land areas (Serreze et al. 2003).

Cullather et al. (2000) examined ERA-15 and NCEP-1 representations of the atmospheric and surface moisture budgets (precipitation less evaporation, or PE). The atmospheric (or aerological) budget is obtained by adjusting the vertically integrated moisture flux convergence by the time change in total column water vapor (TCWV, or precipitable water). The surface budget is the difference between the forecasts of precipitation and evaporation. It appears that the aerological budgets are very useful for hydrologic assessments. Subsequent studies have used these data to address variability in PE across the Arctic (e.g., Rogers et al. 2001). However, the aerological and surface budgets for both ERA-15 and NCEP-1 are not in balance, with lower PE in the forecasts (Cullather et al. 2000).

A pilot evaluation of ERA-40 was conducted as part of the ECMWF Workshop on Reanalysis, held 9 November 2001 (ECMWF 2002), see the included papers by Serreze and Etringer (2002) and Bromwich et al. (2002)). This evaluation used 4 yr of data (1989–92) from a preproduction run. In contrast to ERA-15, the surface moisture budget is in rough balance with the atmospheric moisture budget computed from the analyzed fields, a welcome improvement in this respect. However, two areas of concern emerged: 1) precipitation forecasts from ERA-40 were no better than those from ERA-15; 2) relative to both ERA-15 and rawinsonde observations, ERA-40 suffered from a tropospheric cold bias centered over the Arctic Ocean, seen in 500-hPa heights and 1000–500-hPa thicknesses. It was most prominent in summer. Apparently, the cold bias adversely impacted precipitation. It was traced to a problem in the assimilation of satellite radiances over the cloudy Arctic. While a fix was adopted, a tropospheric cold bias still appears to be present in the production run of ERA-40 (D. Bromwich 2004, personal communication).

Our evaluations of ERA-40, NCEP-1, GPCP, and ERA-15 examine the region north of 45°N, the terrestrial Arctic drainage, and the four major watersheds contained within it: Ob, Lena, Yenisey, and Mackenzie (Fig. 1). The Arctic drainage is defined by all areas draining into the Arctic Ocean as well as into Hudson Bay, James Bay, Hudson Strait, Bering Strait, and the northern Bering Sea. Evaluations span the years 1979–93, a common period for all datasets and for which the gauge density is sufficient to compile gridded fields of observed precipitation. As well, it corresponds to a modern assimilation database for the reanalyses that includes satellite observations. Annual cycles, biases, and squared correlations are computed, providing information about errors in the products and how well they depict seasonal and interannual variations in northern high-latitude precipitation.

2. Datasets

a. Estimates from reanalysis

The primary ERA-40, NCEP-1, and ERA-15 archives provide 6-h accumulated precipitation from 6-h forecasts on a 2.5° × 2.5° latitude–longitude grid. High-resolution ERA-40 fields on a reduced Guassian grid with an approximate 125-km spacing are now available at NCAR. We use the lower-resolution data to provide for more direct comparisons with the GPCP estimates (discussed shortly). Data were processed into monthly precipitation sums by year. We use total precipitation, which is the sum of the large-scale and convective components. The reanalysis data can be downloaded from the respective Web sites of NCEP (http://www.cpc.ncep.noaa.gov/products/wesley/reanalysis.html) and ECMWF (http://www.ecmwf.int/). The data are also available at NCAR (http://www.ncar.ucar.edu/ncar/). NCEP-1 and ERA-15 data are also held at the Climate Diagnostics Center (CDC) in Boulder, Colorado (http://www.cdc.noaa.gov/).

The NCEP-1 reanalysis starts in 1948 and is continually updated. ERA-15 covers the period 1979–93. The NCEP-1 system, based on a T62 model with 28 vertical sigma levels, is described by Kalnay et al. (1996), Kistler et al. (2001), and the extensive references therein. The ERA-15 forecast model has a horizontal resolution of T106 with 31 vertical levels. A complete description of ERA-15 is given by Gibson et al. (1997). Serreze and Hurst (2000) also provide brief overviews of both NCEP-1 and ERA-15.

Currently, ERA-40 spans the period 1957–2002. Updates are planned. A broad overview of ERA-40 can be found in the extended abstracts from the 2001 ERA-40 workshop (ECMWF 2002) and the ECMWF Web site. Like NCEP, ERA-40 uses a 3D variational assimilation system. The spectral model has a horizontal resolution of T-159, with 60 levels in the vertical, with a well-defined boundary layer and stratosphere. Conventional data source for assimilation rely strongly on the NCEP–NCAR archives. ERA-40 makes heavy use of multichannel satellite radiances starting with the first Vertical Temperature Profile Radiometer (VTPR) instruments in 1972 and continuing through the present Special Sensor Microwave Imager (SSM/I), the Television Infrared Observational Satellite (TIROS) Operational Vertical Sounder (TOVS), and the Advanced TOVS (ATOVS) instruments. ERA-40 makes more use of satellite data than ERA-15 and NCEP-1. Analysis of ozone is included, as well as improved cloud motion winds. A significant difference with respect to ERA-15 is the assimilation of raw satellite radiance data as opposed to retrieved properties. This avoids trends and variability due to changes introduced in satellite data processing over the years. There are numerous improvements relative to ERA-15 in the land surface scheme. Over half of these improvements were driven by high-latitude concerns (P. Viterbo, personal communication). ERA-40 also features improved sea surface temperature and sea ice boundary fields.

Precipitation spinup in high latitudes is known to be a problem in ERA-40 and has been addressed by Betts et al. (2003) in their study of the Mackenzie basin. To assist later discussions we need to review this issue. Assessing spinup for the entire Arctic is beyond the scope of our paper. Consequently, we assume that the results from Betts et al. (2003) have some general applicability across the Arctic.

Betts et al. (2003) focused on the link between ERA-40 precipitation biases and spinup through the forecast cycle associated with analysis increments in TCWV. They made use of forecasts run out to 36 h from the 0000 and 1200 UTC analyses of each day of the ERA-40 record. From these 36-h forecast, they obtained monthly basin averages of three different daily sums of precipitation that will be progressively less influenced by the analysis update/increments through adding: 1) a pair of 0–12-h forecasts, 2) a pair of 12–24-h forecasts, and 3) a pair of 24–36-h forecasts. Spinup was assessed through comparing the different daily sums (e.g., the change between the 0–12- and 24–36-h forecasts).

Betts et al. (2003) find that in mid- and high latitudes, precipitation in ERA-40 generally increases during the first 36 h of forecasts, with spinup for large-scale precipitation larger than for convective precipitation. Examined as annual averages for the Mackenzie basin, precipitation spinup, bias, and the analysis increments of TCWV vary strongly over the ERA-40 record and in general concert with each other. Bias was calculated from basin averages of pairs of 0–12-h forecasts minus corresponding basin-averaged gauge measurements. Spinup was computed from the 0–12- to 24–36-h forecasts. Biases change from strongly negative in the early part of the ERA-40 record (with the strongest spinup and most negative increments) to positive in the mid-1970s (with the smallest spinup and small positive increments), and finally to the smallest bias in the 1990s (with intermediate spinup and modest negative increments).

In general, and for annual averages, the magnitudes of bias and spinup seem to be largely determined by how much TCWV is removed or added to the analysis. This suggests that the TCWV increments might be used to correct for some of the biases. In the early part of the record, more TCWV is removed from the analysis, which reduces precipitation in the forecast cycle (yielding a strong negative bias). The model then spins up to restore precipitation. While investigations are ongoing, the problems likely relate to changes in conventional data or its use. For example Betts et al. (2003) have identified errors in one of the humidity datasets used for 1958–63, which had significant impacts over North America. Spinup, bias, and the TCWV analysis increments also vary considerably over the annual cycle. For the earlier (1958–70) and later (1987–97) parts of the record, spinup and TCWV analysis increments are largest in summer (when mean precipitation also peaks over the Mackenzie and most Arctic land areas), while a more complex pattern emerges for the middle period (1974–81). There is only general correspondence with the annual cycles of bias. We are not aware of any studies that have focused on high-latitude spinup issues in NCEP-1 or NCEP-2.

b. Satellite retrievals of precipitation

We use the GPCP merged version-2 dataset, which extends from 1979 to the present. It provides monthly precipitation on the same 2.5° × 2.5° latitude–longitude array for which the reanalysis data are available. The merged GPCP version-2 global dataset blends estimates from many different satellite sources covering different periods. All of the retrievals are calibrated against observed precipitation. The form of calibration varies between the different satellite sources. The procedures to merge the different satellite sources are complex. They are based on the known strengths and shortcomings of each satellite estimate for different regions and surface types, and the spatial coverage provided by each estimate (Huffman et al. 1997).

Retrievals from SSM/I are based on microwave emission and scattering techniques. The emission technique infers the amount of column liquid water. More liquid water tends to relate to a higher precipitation rate. The scattering technique infers the quantity of column hydrometer ice. More hydrometer ice typically implies more surface precipitation. TOVS retrievals infer precipitation from various TOVS-based parameters that relate to cloud volume (cloud-top pressure, fractional cloud cover, and relative humidity profiles). The Geostationary Operational Environmental Satellite (GOES) precipitation index dataset (the “GPI product”) makes use of infrared (IR) observations. Colder IR brightness temperatures are directly related to higher cloud tops, which relates to increased precipitation rates. There is also an outgoing longwave radiation (OLR) estimate (the “OPI product”) based on the use of low earth orbiting satellite measurements of OLR.

The merged GPCP dataset is considered to be best over open ocean areas and from about 1987 onward when the SSM/I and TOVS retrievals began to be used. For the polar regions, estimates for this more recent period are primarily from TOVS retrievals; due to the heterogeneous emissivities of land, snow, and sea ice, the SSM/I retrievals are deemed unreliable. For earlier years, polar regions rely on the OPI products. As discussed, little attention has been paid to the high-latitude performance of the merged GPCP product. GPCP also archives a separate product (not examined here) that blends satellite retrievals with gauge data.

c. Gauge precipitation

We employ records of monthly station precipitation based on a blend of many different sources. The aim is to get the most complete coverage possible. The blended archive is an extension of that described by Serreze et al. (2003), which was based on the following: 1) an updated version of the Groisman et al. (1991) dataset of 622 stations in the Former Soviet Union (FSU); 2) National Climatic Data Center (NCDC) dataset (TD-9816) “Canadian Monthly Precipitation” (Groisman 1998), which contains nearly 7000 stations but with the vast majority in the southern part of the nation; 3) an updated version of the Mekis and Hogg (1999) Canadian dataset of 495 stations; 4) the Global Historical Climatological Network (GHCN), described by Vose et al. (1992); 5) records for 105 stations within the major Arctic-draining watersheds of Eurasia (Ob, Yenisey, and Lena) not present in the Groisman FSU or GHCN archives, obtained through collaboration with Russian investigators; and 6) Arctic Ocean records from the Russian “North Pole” (NP) program (Yang 1999). The blended dataset was further improved for the present study through use of 1) the NCDC dataset 9101 “Global Daily Climatology Network, V1.0” (Gleason 2002); and 2) additional records for the coastal and interior FSU assembled by the National Snow and Ice Data Center (NSIDC), Boulder, Colorado, in collaboration with Russian investigators.

All records were obtained as raw monthly totals (or in the case of NCDC 9101 daily values which we then summed) with no bias adjustments. The period of coverage varies between archives and for individual stations. While there is great duplication in coverage between different archives, monthly values for a given station that were available from one archive were often missing in another. Records were merged to obtain the densest possible network of unique stations with the most complete time series possible for each station. Figure 2 shows the location of all land stations for the region north of 40°N. Coverage is quite dense in southern Canada and the northern United States, but is meager over northern Canada and large parts of northeast Eurasia. Given concerns of both data quality and short records, we make no use of Automatic Weather Station (AWS) data that are available since the late 1980s over the Greenland ice sheet. Note that the Russian NP dataset (coverage not shown) provides only one to two observations per month over the Arctic Ocean. The NP program terminated in 1991 with the breakup of the FSU.

The variation through time of the number of land stations in the blended dataset that reported precipitation between 1950 and 2000 is shown in Fig. 3. The decrease in the number of stations after 1990 is striking. The bulk of this loss occurred between 50° and 60°N. This is partly a result of a true (and very disconcerting) decline in the data network. Budget cuts in Canada and Russia have led to the closure of many stations. However, is also reflects a difficulty in obtaining updates for many stations. It appears that there are many updates in the NCDC “Summary of the Day” (SOD) (available online at http://www.ncdc.noaa.gov/oa/climate/climatedata.html) that are not in other archives. There are significant issues of quality control with the SOD data. Efforts are under way to recover as much of these data as possible.

3. Gridding the station data

a. General approach

As previously mentioned, the years 1979–93 represent a common period of overlap between the three reanalyses archives and the GPCP records. This period also has an adequate number of stations to compile gridded fields of observed precipitation although fields for the last several years are based on a lesser network (see Fig. 3). Furthermore, it corresponds to a relatively modern assimilation database for the reanalyses that includes satellite sources and allows the GPCP dataset to be assessed after 1987 when TOVS began to be used.

Comparing output from reanalysis and satellite retrievals directly with station data is unwarranted because of large differences in the spatial scale of the data. Past experience (Serreze et al. 2003) shows that a better comparison is provided by “scaling up” the station data through a gridding procedure. We interpolated the monthly station data to a 100-km version of the NSIDC Equal-Area Scalable Earth (EASE) grid (an equal area grid; Armstrong and Brodzik 1995) for the region north of 45°N using a modified version of the Shepard (1984) scheme. The gridded monthly time series were then adjusted using climatological, monthly varying multipliers that attempt to address gauge biases that are primarily associated with undercatch of solid precipitation. These adjustments are identical to those used in the well-known Legates and Willmott (1990) climatology. The EASE grid has been adopted within an ongoing project to monitor hydrologic variability over the pan-Arctic drainage [Arctic Rapid Integrated Monitoring System (Arctic-RIMS), see online at http://rims.unh.edu/tour.shtml] of which the present study is part of. For compatibility with this adopted scheme, the reanalysis and GPCP data on the 2.5° latitude–longitude grid were reprojected to the 100-km EASE grid using a Cressman (1959) routine.

We have conducted evaluations like those that follow below where instead of translating the reanalysis and GPCP data to the EASE grid, we interpolate the station data to the 2.5° grid. There are no appreciable differences in the results. Having all data on an equal area grid furthermore facilitates more direct assessments of the impacts of the sparse station database (discussed below) on our evaluations.

b. Impacts of the sparse gauge network

While the historically sparse high-latitude gauge network and its recent degradation is the very reason why there is interest in evaluating reanalysis and satellite retrievals, the sparse network complicates these evaluations. This issue was addressed by Serreze et al. (2003) and is useful to briefly review. This study examined the influence of gauge densities on the quality of gridded monthly time series, means, and standard deviations of precipitation for grid cells of different sizes. The analysis was based on Monte Carlo simulations for several regions in the southern Canadian part of the Arctic terrestrial drainage having a dense network.

Dense network (taken as “true”) time series for 1960–89 were obtained by averaging all (N) gauge values within a given grid cell for each month and year. Means and standard deviations were then computed. Time series were then obtained by degrading the network. A total of 500 time series was obtained from a random selection of N-1 gauges. From the 500 realizations, they calculated 1) the absolute mean standard error (mse) of the mean precipitation as a fraction of the dense network mean; 2) the mse of the standard deviation as a fraction of the dense network standard deviation; 3) the mean-squared correlation between the dense network and degraded time series. The process was repeated with a random selection of N-2 gauges, N-3 gauges, etc.

For a grid cell of 175 km, getting a high squared correlation (e.g., exceeding 0.80) with the true monthly time series requires typically three–five gauges in the grid cell. More gauges are needed in topographically complex areas. But when evaluated over the Arctic terrestrial drainage for the period 1960–89, only 38% of 175-km grid cells contain even a single station. For most grid cells, values must be obtained through interpolation of gauge data from well outside grid-cell boundaries. Serreze et al. (2003) used the same basic Shepard scheme employed here. Results for the 100-km EASE grids echo those for the coarser 175-km grid—the interpolated time series of precipitation are typically not very representative of the 100-km scale. However, even with fairly liberal interpolation, the grid-cell means and standard deviations tend to be reasonably well preserved (Serreze et al. 2003). Our evaluations of bias and temporal correlations that follow below include separate assessments of grid cells for which observed precipitation is based on a fairly high density of stations.

c. The Shepard routine

Willmott et al. (1985) provide a useful review of the Shepard (1984) scheme. Interpolation weights are based on three categories of distance:
i1520-0493-133-12-3407-eq1
where d is the distance between the station and the center of the grid cell to be interpolated to, r is the search radius, and S is the interpolation weight. Values are defined for the maximum (MAX) and minimum (MIN) number of data points (station precipitation values) used in the interpolation. An initial search radius around each grid cell center is defined from the area of the spatial domain to be interpolated to and the number of available data points. If the number of data points within the search radius exceeds MAX, the closest data points up to MAX are used. If there are fewer that MIN data points in the search radius, the radius is expanded until at least the MIN points are found. The interpolator uses spherical geometry to calculate distances, adjusts for uneven clustering of stations, and allows for extrapolation beyond the bounds of the station data based on local gradients.

We compiled the 100-km gridded time series as follows. If a grid cell contained at least three stations, precipitation was determined as the simple average of the stations. This provides the least biased grid-cell value (the Shepard interpolation by contrast provides center-weighted values). The Shepard interpolation was used for the remainder of the cells. The Shepard algorithm was modified to use only the four stations closest to each grid-cell center. For most cells, this avoids drawing from distant stations. Values obtained over central Greenland are based on distant coastal sites. Those over the central Arctic Ocean are based on interpolation from coastal sites and Russian NP data through 1991. Results for these regions should be viewed with due caution. We make no attempt to provide values over open-ocean areas far from observations. Open-ocean areas were defined from observed annual maximum sea ice limits, using ice concentration data provided by NSIDC.

4. Mean precipitation

a. ERA-40 monthly mean precipitation and biases

We first examine fields of monthly mean precipitation from ERA-40 averaged for the period 1979–93 (Fig. 4). ERA-40 effectively captures the major precipitation features as they are known. For winter months, the highest precipitation totals, exceeding 150 mm and locally much higher, are found in the Atlantic and Pacific basins, reflecting the northern ends of the primary Atlantic and Pacific storm tracks and the associated Icelandic and Aleutian lows. ERA-40 also captures the areas of strong orographic precipitation along the coasts of Alaska and the Pacific Northwest, and along the Scandinavian coast. Reflecting continentality and generally anticyclonic conditions, the lowest winter totals are found over the northern Canadian Arctic, the Arctic Ocean, and northeastern Eurasia. Weakening of the primary storm tracks and associated sub-Arctic low pressure cells is seen in the decline of the Atlantic and Pacific precipitation maxima in summer. Precipitation increases over most land areas, in part, due to convective precipitation and, especially over northern Eurasia, an increase in cyclone activity. Serreze and Etringer (2003) provide a review of high-latitude precipitation processes over northern Eurasia. Precipitation also increases over the Arctic Ocean in summer. This is due to the migration of lows into the region, many of which are generated over Eurasia.

Biases between monthly mean (1979–93) observed precipitation and precipitation from ERA-40, NCEP-1, GPCP, and ERA-15 were calculated for every grid cell with shared data coverage from each source. This represents all cells north of 45°N excluding the open (ice free) ocean areas for which observations are insufficient. Schematic plots, a variant of box and whisker plots (Wilks 1995) show the distribution of these biases for each product for the four midseason months (Fig. 5). Each box in the figure encloses 50% of the bias distribution, with the median bias represented by the horizontal line in each box. The dark solid vertical lines enclose the 10th and 90th percentiles, while the thin intermittent vertical lines represent biases for the remainder of the grid cells. Biases are expressed as the forecast minus observed precipitation, divided by observed precipitation. Resulting values were then multiplied by 100. As such, the biases represent the percent deviation from observed values.

In general, 50% of biases for all products are between ±50% of observed precipitation. The NCEP-1 biases tend to have the largest spread, most clearly expressed for January, April, and July. Another striking feature in Fig. 5 is that although median biases are typically within about 20%–25% of observed precipitation, they are mostly negative, meaning that precipitation tends to be underestimated. While part of this may reflect precipitation spinup problems [which as discussed have been examined by Betts et al. (2003) for the Mackenzie basin], the negative median biases must also be viewed in light of the climatological bias adjustments applied to the observations. An exception to the general pattern of negative biases is April and July, when median biases in NCEP-1 are positive. Compared to the other data sets, there are also many more positive biases in January, even though the median bias is negative. The July results for NCEP-1 can be related to excessive summertime convective precipitation (Serreze and Hurst 2000). That strong biases exist in other months points to additional problems. However, all products have extreme biases for some grid cells that are mostly positive. These typically correspond to areas where observed precipitation is low, such that a positive bias in millimeters of precipitation (forecast minus observed) is large in terms as a percentage bias, as the divisor (observed precipitation) is small.

Following the study of Serreze et al. (2003), it is useful to assess whether the biases are influenced by the very sparse station database over much of the study domain, especially the Arctic Ocean. To this end, Fig. 6 shows schematic plots based only on those grid cells for which the interpolations were based on a reasonably dense station network. For this example, we use grid cells for which there were more than four stations within two grid lengths (200 km) of the grid cell center, based on the average station network during the period 1979–93. The bias structure (based on 1216 of the 7429 grid cells north of 45°N) is largely similar to that in Fig. 5. It again highlights the preponderance of negative median biases. The exception is that the positive median biases in NCEP-1 for April and July are larger. This follows in that the grid cells used for Fig. 6 are all over land, where NCEP-1 precipitation biases are largest.

Spatial fields of biases for the four midseason months appear in Fig. 7, again expressed as a percent deviation from observed values. White areas represent regions of open ocean where we are unable to provide observed values. Gray areas are where biases are within ±25%. We do not show fields for ERA-15, as they are similar to those for ERA-40. For much of the domain, ERA-40 (hence also ERA-15) biases are within ±25% of observations, with most of the remaining area showing strong negative biases. However, there are also some areas of very large positive biases, especially where mean precipitation is low. The overall pattern for GPCP is similar to that for ERA-40, which is not surprising given that the satellite estimates are calibrated against observations. The presence of very strong positive biases in NCEP-1 for April and July over land areas stands out. All of the products show some tendency for large biases, both positive and negative, along coasts where there are strong gradients in precipitation. Presumably part of this problem arises from the difficulties of interpolating the gauge data in regions of sharp gradients.

b. Annual cycles for major Arctic watersheds

We turn next to the mean annual cycles of precipitation averaged for the Ob, Yenisey, Lena basins (in Eurasia), and the Mackenzie basin (northwest North America; Fig. 8, see Fig. 1 for locations). The focus on these drainages reflects their hydrologic importance—collectively they provide the majority of river runoff from the Arctic land areas to the Arctic Ocean. Annual cycles were compiled by averaging grid-cell values falling within each watershed.

As previously highlighted, NCEP-1 greatly overestimates summer precipitation, as well as for spring. The annual cycles from ERA-40 and ERA-15 are closer to observations, and track each other closely. Overall, both ERA-15 and ERA-40 tend to be too low in the Ob. However, note the generally higher summer month precipitation for ERA-15 relative to ERA-40 in the Ob, as well as for the Yenisey and Lena basins. As with previous results, Fig. 8 must be interpreted with recognition that the observations themselves include bias adjustments. Betts et al. (2003), using the Mekis and Hogg (1999) precipitation dataset for the Mackenzie basin (which is included in our blended precipitation database), find a positive bias in ERA-40 in spring and a negative bias in autumn. While we use different bias adjustments, this pattern also appears in our results although it is not prominent. The GPCP estimates generally fall within the range of the three reanalyses, but this depends on the month and watershed. For example, in the Yenisey basin, GPCP is too low for summer and early autumn.

5. Interannual variability

a. Large-scale assessments

Figure 9 provides a general illustration of how well the precipitation estimates from NCEP-1, ERA-40, GPCP, and ERA-15 capture interannual variability. Squared correlations between time series (1979–93) of observed monthly precipitation and each of the four estimates were calculated by month and grid cell. We summarize these for January and July by showing the cumulative fraction of grids cells (y axis) for which the squared correlation is less than the value indicated on the x axis. We break down the results as in Figs. 5 and 6, first using all cells north of 45°N excluding the open (ice free) ocean areas for which observations are insufficient, and second for the subset of grid cells (all over land) for which there were more than four stations within two grid lengths of the cell center. Note that these correlation analyses are based on rather short 15-yr records.

Looking first at the results using all grid cells, it is clear that ERA-40 does a much better job than NCEP-1 of capturing the interannual variability of monthly precipitation. For example, during January, about 70% of grid cells from NCEP-1 have a squared correlation with observed precipitation less than 0.50 (a squared correlation of 0.50 meaning that half of the variance is shared between the two precipitation estimates). By comparison, about 50% of the grid cells from ERA-40 have a squared correlation of less than 0.50. Put differently, for ERA-40, 50% of the cells have a squared correlation greater than 0.50. ERA-40 performs only slightly better than ERA-15. In July, when convective precipitation is important, the performance of ERA-15 and ERA-40 is nearly identical. Performance of the GPCP product is quite poor. In January, about 95% of the GPCP grid cells have a squared correlation less than 0.50. The satellite product is better in July, but still short of even NCEP-1.

When the analysis is restricted to grid cells based on a comparatively high station density, correlations are expected to improve. This is clearly evident for ERA-40 and ERA-15. For ERA-40 in January, only 30% of the grid cells in this subset have a squared correlation less than 0.50 (70% exceed 0.50). While correlations are still lower in July compared to January, they also improve when the observed time series are more reliable. Again, ERA-40 and ERA-15 have similar performances in July. The implication (which we return to shortly) is that low correlations in data-sparse areas are not necessarily a reflection of poor model performance. By comparison, for NCEP-1, correlations only improve in January, while those for GPCP show little sensitivity to the quality of the observational database.

For each month and year, the field of observed precipitation for the full domain was correlated against respective fields of precipitation from the reanalysis and satellite products. These squared field correlations give an indication of how well the spatial patterns of precipitation across the domain are represented from year to year. Time series of these squared field correlations are given in Fig. 10 for January and July. For January, ERA-40 and ERA-15 track each other closely, with higher field correlations than either NCEP-1 or GPCP. The two Januarys for which ERA-40 and ERA-15 have the lowest squared field correlations, 1985 and 1987, were characterized by strongly negative index values of the Arctic Oscillation (−2.8 and −1.1, respectively). While circulation relationships are left to a follow-up paper, inspection of these years points to problems in the position of precipitation anomalies in the Atlantic sector. For July, ERA-15 consistently beats ERA-40. In some cases, GPCP is actually better than ERA-40. With few exceptions, field correlations for NCEP-1 are comparatively low. The expectation of improved correlations with the GPCP fields after 1987, when TOVS retrievals began to be used, is not readily evident.

We turn next to fields of squared grid-cell correlations with the 15-yr time series of observations (Fig. 11). The better performance of ERA-40 relative to NCEP-1 is prominent. ERA-40 performs remarkably well over western to central Eurasia, where squared correlations generally exceed 0.60. Following from the summaries in Fig. 9, we see lower correlations in summer. This points to problems with both the observations and models. Part of the terrestrial precipitation in summer is of convective origin. Precipitation tends to be more localized than in winter, spring, and autumn, so it is more difficult to obtain good grid-cell estimates with the sparse gauge network. In turn, convective precipitation is very difficult to model. In the later part of the ERA-40 record, precipitation spinup is also largest in summer [at least for the Mackenzie basin; see Betts et al. (2003)], which may also contribute to the lower ERA-40 summer correlations. ERA-40 correlations are low over the Arctic Ocean and northern Canada. However, the gauge network is sparse in both areas. Over the Arctic Ocean, observed time series are based primarily on interpolation from coastal sites and occasional values from the NP program through 1991 (see section 5b for a separate examination of the Arctic Ocean). Given the results in Fig. 9, it is reasonable to conclude that if the station database was more dense, the correlations in these areas would be higher. Also consistent with Fig. 9, both ERA-40 and NCEP-1 are far superior to the GPCP product. The GPCP correlations are somewhat improved in summer as compared to winter. These results may seem at odds with those in Fig. 10 showing that the July field (i.e., pattern) correlations for GPCP are higher than those for NCEP-1. In explanation, the field correlations are influenced by systematic biases (large for NCEP-1 in summer) while the temporal correlations are not.

The maps of temporal correlations between observed and ERA-15 precipitation resemble those for ERA-40 and are therefore not shown. The major differences are in summer. The contrasts between ERA-40 and ERA-15 are evident in maps of squared correlations between the two reanalysis time series. For most of the year, grid-cell-squared correlations exceed 0.80. In summer they drop, locally below 0.50. If changes in the land surface scheme or the convective parameterization were involved, one might expect the lower correlations to be associated with land areas. But there is no clear spatial pattern—the lower correlations are as common over land as they are over the Arctic Ocean.

b. Regional assessments

It is useful to assess the performance of the four products for the major Arctic draining watersheds. Time series were compiled by averaging grid-cell values falling within each watershed. Table 1 lists squared correlations between observed precipitation time series and the four precipitation estimates for the midseason months. On the scale of the large watersheds, squared correlations between observed and ERA-40 precipitation are generally quite high, exceeding 0.90 for some months and watersheds. The worst case is July for the Yenisey basin (0.40), consistent with generally lower correlations in summer for all watersheds. The July results also point to generally better performance of ERA-15 relative to ERA-40 in summer. In this month, ERA-15 explains 30% more variance in the Mackenzie basin than ERA-40, and ERA-40 is surpassed by NCEP-1. The typically poor performance of the GPCP product again emerges.

Basin-averaged times series of observed precipitation and from ERA-40 and GPCP are given in Figs. 12 –15 for January and July. Results are provided for the longer period 1960–2002. Given degradation of the gauge network, observed precipitation is only plotted for 1960–93. In general, ERA-40 tracks the observed time series fairly well. However, there are large departures for individual years and runs of years. For example, in the earlier parts of the records, ERA-40 tracks quite poorly with observed precipitation for July in Yenisey, Lena, and Mackenzie basins. Following from Betts et al. (2003), at least for the Mackenzie basin, the negative bias of ERA-40 in the early part of the record (before the modern satellite era) reflects the large precipitation spinup in this basin for summer. Recall that Betts et al. (2003) identified a source of error in one of the humidity datasets used for 1958–63 that had significant impacts over North America. We stress, however, that there are problems for some basins even in the later years (e.g., the Yenisey basin in July). Time series of GPCP precipitation do not track the observed record as well as ERA-40. Note the poor tracking with observations in Ob even after 1987, when the GPCP product uses TOVS retrievals.

Direct observations of precipitation for the central Arctic Ocean are essentially limited to measurements collected during the Russian NP program. Some idea of reanalysis and GPCP performance for this region is provided in Table 2. Observations from the NP data aggregated for winter, spring, summer, and autumn are compared with the four estimates at the closest grid cell. This yields from 42 to 48 cases for analysis, depending on the season. We recognize the obvious problems of scale differences. For winter, spring, and autumn, means for all of the precipitation estimates are lower than the bias-adjusted observed values. These represent monthly means of precipitation within each season, and not mean seasonal totals (appropriate given the uneven number of cases from each month within a season). For some cases in these three seasons, the estimates are closer to the raw observations (with no bias adjustments, given in parentheses). Each of the four products correctly depicts higher precipitation in summer.

Squared correlations with the NP observations are very low in winter and spring. Part of the explanation is that precipitation amounts are very low. With low precipitation, even small errors in the observations and from the four precipitation estimates will degrade the correlations. The squared correlations are higher in summer and highest in autumn when, as indicated from the means, the precipitation signals are stronger. For both of these seasons, ERA-40 and ERA-15 surpass GPCP.

6. Summary and recommendations

In comparison to NCEP-1, the representation of northern high-latitude precipitation in the ERA-40 reanalysis represents a strong step forward. ERA-40 is superior in terms of smaller biases, its ability to capture large-scale patterns of precipitation and its depiction of interannual variability. ERA-40 precipitation fields are already being used by us within the Arctic-RIMS project to compile time series of high-latitude precipitation through blending with gauge data. The general approach, which builds upon techniques outlined by Serreze et al. (2003), removes biases from ERA-40 using the statistical distributions of observed precipitation in a nonparametric distribution transformation (Panofsky and Brier 1963). The resulting de-biased ERA-40 fields are then adjusted by the observations via optimal interpolation. These will replace products currently based on NCEP-1 data. These blended fields, disaggregated to finer time steps using the 6-hourly ERA-40 forecasts, will also be used by our group in land surface model simulations for the Arctic terrestrial drainage.

Our study results are also troubling in two ways. The first regards the poor performance of the GPCP-blended satellite-derived product over high northern latitudes. The GPCP product has skill in some areas during summer, but as a general statement it cannot surpass NCEP-1, the worst of the three reanalysis products. There is no evidence that the GPCP estimates are greatly improved after 1987, when land and Arctic Ocean retrievals make use of TOVS data. We cannot recommend use of the GPCP data for these areas. Its use should be restricted to open-ocean areas, where performance is considered best. The GPCP also provides a product that is blended with observations, improving depictions of precipitation variability over land, but these fields are inherently limited by the quality of the satellite-derived base. Regarding the CMAP effort, which blends satellite retrievals with gauge data and NCEP reanalysis, we recommend that construction of high-latitude fields explore the use of ERA-40 precipitation estimates.

The second troubling finding is that anticipated improvements in ERA-40 relative to ERA-15 have not been realized. In some respects, ERA-15 is better in summer—the primary advantages of ERA-40 are its longer record and its higher spatial resolution. The degraded performance of ERA-40 in summer shows no clear preference for land areas or the Arctic Ocean. It appears to be a general feature of the reanalysis. It can be argued that the “benchmark” that the developing Arctic System Reanalysis must surpass, at least for precipitation, has not been set by ERA-40, but rather by ERA-15.

It is beyond the scope of the present study to diagnose why ERA-40 offers no improvement. It is likely that part of the problem lies with the satellite retrievals. A fundamental difference between ERA-15 and ERA-40 is that ERA-40 assimilates raw satellite radiances. In the Arctic region this must be done with great care. The Arctic is characterized by strong low-level temperature inversions during winter and is a very cloudy place, especially in summer when low-level stratus dominates. The strong influence of suboptimal assimilation of satellite radiances on tropospheric heights and thicknesses, which should impact precipitation, emerged very clearly in pilot studies using data from the preproduction run of ERA-40 (Bromwich et al. 2002). Although efforts were made to address these problems in the production run, initial indications are that problems still exist. Further investigations are under way. It follows that a key developmental research area for the Arctic System Reanalysis is the optimal use of satellite data.

Acknowledgments

This study was supported by NSF Grants OPP-9910315, OPP-0229769, OPP-0229651; NASA Contracts NAG5-9568, NNG04GH04G, NNG04GJ39G; and NOAA’s support for activities leading to development of the Arctic System Reanalysis.

REFERENCES

  • 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 , 155163.

    • Search Google Scholar
    • Export Citation
  • Betts, A. K., J. H. Ball, and P. Viterbo, 2003: Evaluation of the ERA-40 surface water budget and surface temperature for the Mackenzie River basin. J. Hydrometeor., 4 , 11941211.

    • Search Google Scholar
    • Export Citation
  • Bromwich, D. H., S-H. Wang, and A. J. Monaghan, 2002: ERA-40 representation of the Arctic atmospheric moisture budget. Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis, Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 287–297.

  • Cressman, G. P., 1959: An operational objective analysis system. Mon. Wea. Rev., 87 , 367374.

  • Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: The atmospheric hydrologic cycle over the Arctic basin from reanalyses. Part I: Comparisons with observations and previous studies. J. Climate, 13 , 923937.

    • Search Google Scholar
    • Export Citation
  • ECMWF, 2002: Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis. Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 443 pp.

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

  • Gleason, B. E., 2002: Data documentation for data set 9101. Global Daily Climatology Network, V1.0, National Climatic Data Center, 25 pp.

  • Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation measurement intercomparison. Final Rep. WMO TD-872, World Meteorological Organization, Geneva, Switzerland, 212 pp.

  • Groisman, P. Y., 1998: National Climatic Data Center data documentation for TD-9816: Canadian monthly precipitation. National Climatic Data Center, 21 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 , 17251733.

    • Search Google Scholar
    • Export Citation
  • Huffman, G. J., and Coauthors, 1997: The Global Precipitation Climatology Project (GPCP) combined precipitation data set. Bull. Amer. Meteor. Soc., 78 , 520.

    • Search Google Scholar
    • Export Citation
  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437471.

  • Kistler, R., and Coauthors, 2001: The NCEP-NCAR 50-Year Reanalysis: Monthly means CD-ROM and documentation. Bull. Amer. Meteor. Soc., 82 , 247267.

    • Search Google Scholar
    • Export Citation
  • Legates, D. R., and C. J. Willmott, 1990: Mean seasonal and spatial variability in gauge-corrected, global precipitation. Int. J. Climatol., 10 , 111127.

    • Search Google Scholar
    • Export Citation
  • Mekis, E., and W. D. Hogg, 1999: Rehabilitation and analysis of Canadian daily precipitation time series. Atmos.–Ocean, 37 , 5385.

  • Panofsky, H. A., and G. W. Brier, 1963: Some Applications of Statistics to Meteorology. Mineral Industries Continuing Education, College of Mineral Industries, The Pennsylvania State University, 224 pp.

    • Search Google Scholar
    • Export Citation
  • Rogers, A. N., D. H. Bromwich, E. N. Sinclair, and R. I. Cullather, 2001: The atmospheric hydrologic cycle over the Arctic basin from reanalyses. Part II: Interannual variability. J. Climate, 14 , 24142429.

    • Search Google Scholar
    • Export Citation
  • SEARCH SSC, 2001: Study of Environmental Arctic Change, Science Plan. Polar Science Center, Applied Physics Laboratory, University of Washington, Seattle, 91 pp.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and C. M. Hurst, 2000: Representation of mean Arctic precipitation from NCEP–NCAR and ERA Reanalyses. J. Climate, 13 , 182201.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and A. J. Etringer, 2002: Representation of Arctic precipitation in ERA-40. Proc. ERA-40 Project Report Series 3: Workshop on Re-Analysis, Reading, United Kingdom, European Centre for Medium-Range Weather Forecasts, 317–331.

  • Serreze, M. C., and A. J. Etringer, 2003: Precipitation characteristics of the Eurasian Arctic drainage system. Int. J. Climatol., 23 , 12671291.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., J. R. Key, J. C. 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 , 121136.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., M. P. Clark, and D. H. Bromwich, 2003: Monitoring precipitation over the Arctic terrestrial drainage system: Data requirements, shortcomings, and applications of atmospheric reanalysis. J. Hydrometeor., 4 , 387407.

    • Search Google Scholar
    • Export Citation
  • Shepard, D., 1984: Computer mapping: The SYSMAP interpolation algorithm. Spatial Statistics and Models, G. L. Gaile and C. J. Willmott, Eds., D. Reidel, 133–145.

    • Search Google Scholar
    • Export Citation
  • Vose, R. S., R. L. Schmoyer, P. M. Steurer, T. C. Peterson, R. Heim, T. R. Karl, and J. Eischeid, 1992: The global historical climatology network: Long-term monthly temperature, precipitation, sea level pressure, and station pressure data. ORNL Tech. Rep. CDIAC-53, NDP-041, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, Oak Ridge, TN, 99 pp. plus appendixes.

  • Wilks, D. S., 1995: Statistical Methods in Atmospheric Sciences: An Introduction. Academic Press, 467 pp.

  • Willmott, C. J., C. M. Rowe, and W. D. Philpot, 1985: Small-scale climate maps: A sensitivity analysis of some common assumptions associated with grid-point interpolation and contouring. Amer. Cartographer, 12 , 516.

    • Search Google Scholar
    • Export Citation
  • Xie, P., and P. A. Arkin, 1997: Global precipitation: A 17-year monthly analysis based on gauge observations, satellite estimates, and numerical model output. Bull. Amer. Meteor. Soc., 78 , 25392558.

    • Search Google Scholar
    • Export Citation
  • Yang, D., 1999: An improved precipitation climatology for the Arctic Ocean. Geophys. Res. Lett., 26 , 16251628.

  • Yang, D., and Coauthors, 2001: Compatability evaluation of national precipitation gauge measurements. J. Geophys. Res., 106 , D2,. 14811491.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

The Arctic terrestrial drainage (all shaded areas) and the watersheds of the Ob, Yenisey, Lena, and Mackenzie basins.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 2.
Fig. 2.

Distribution of terrestrial gauge sites for the region north of 40°N in the blended archive.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 3.
Fig. 3.

Time series of the number of terrestrial stations in the blended archive (see Fig. 2) reporting precipitation (1950–2000) for the regions north of 50° and 60°N.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 4.
Fig. 4.

Fields of monthly mean precipitation (mm) from ERA-40 during the period 1979–93.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 5.
Fig. 5.

Distribution of precipitation biases (% deviation from observations) for ERA-40, NCEP-1, GPCP, and ERA-15 for Jan, Apr, Jul, and Oct. Results are based on all grid cells except open-ocean regions. Positive values indicate higher precipitation relative to observations. Biases are computed from monthly means during the period 1979–93.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 6.
Fig. 6.

Same as in Fig. 5, except based on the subset of grid cells for which (as averaged over 1979–93) there were at least four stations within two grid lengths of the cell center.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 7.
Fig. 7.

Spatial patterns of precipitation biases (% deviation from observations) for ERA-40, NCEP-1, and GPCP for Jan, Apr, Jul, and Oct. Positive values indicate higher precipitation relative to observations. Biases are computed from monthly means during the period 1979–93.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 8.
Fig. 8.

Mean annual cycles of precipitation (mm) for the Ob, Yenisey, Lena, and Mackenzie basins from observations: ERA-40, NCEP-1, GPCP, and ERA-15. Means are computed during the period 1979–93.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 9.
Fig. 9.

Fraction of grid cells (y axis) having squared correlations with observed precipitation time series (1979–93) less than the value indicated on the x axis. Results are given for Jan and Jul for ERA-40, NCEP-1, GPCP and ERA-15 based on (left) all grid cells (excepting open-ocean areas) and (right) the subset of grid cells for which (as averaged over 1979–93) there were at least four stations within two grid lengths of the cell center.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 10.
Fig. 10.

Time series (1979–93) of squared field correlations between observed precipitation and precipitation from ERA-40, NCEP-1, GPCP, and ERA-15 for Jan and Jul. Field correlations indicate the ability of each precipitation product to capture the spatial patterns of precipitation across the study domain.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 11.
Fig. 11.

Fields of squared correlations between the time series (1979–93) of observed monthly precipitation and time series from ERA-40, NCEP-1, and GPCP.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 11.
Fig. 11.

(Continued)

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 11.
Fig. 11.

(Continued)

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 12.
Fig. 12.

Time series of precipitation (mm) for the Ob basin from observations, ERA-40, andGPCP for Jan and Jul. Time series for observed precipitation are only given through 1993.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 13.
Fig. 13.

Same as in Fig. 12, but for the Yenisey basin.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 14.
Fig. 14.

Same as in Fig. 12, but for the Lena basin.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Fig. 15.
Fig. 15.

Same as in Fig. 12, but for the Mackenzie basin.

Citation: Monthly Weather Review 133, 12; 10.1175/MWR3047.1

Table 1.

Squared correlations between the time series (1979–93) of observed precipitation and precipitation from ERA-40, ERA-15, NCEP-1, and GPCP for the major Arctic watersheds. Results are provided for Jan, Apr, Jul, and Oct.

Table 1.
Table 2.

Mean monthly precipitation by season for the central Arctic Ocean from the Russian NP records and corresponding values from ERA-40, ERA-15, NCEP-1, and GPCP at the closest grid point (mm). Squared correlations with observed precipitation are also provided. Values in parentheses are for raw observations (no bias adjustments).

Table 2.
Save