• Beljaars, A. C. M., , and P. Viterbo, 1999: The role of the boundary layer in a numerical weather prediction model. Clear and Cloudy Boundary Layers, A. A. M. Holtslag and P. G. Duynkerke, Eds., North Holland Publishers, 287–304.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , and J. Shukla, 1988: Integration of space and in situ observations to study global climate change. Bull. Amer. Meteor. Soc., 69 , 11301143.

    • Search Google Scholar
    • Export Citation
  • Douville, H., , J-F. Mahfouf, , S. Saarinen, , and P. Viterbo, 1998: The ECMWF surface analysis: Diagnostics and prospects. ECMWF Tech. Memo. 258, European Centre for Medium-Range Weather Forecasts, 51 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/tm/001-300/tm258.pdf].

    • Search Google Scholar
    • Export Citation
  • Gibson, J. K., , P. Kallberg, , S. Uppala, , A. Nomura, , A. Hernandez, , and E. Serrano, 1997: ERA description. ECMWF ERA-15 Project Rep. Series 1, European Centre for Medium-Range Weather Forecasts, 72 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id=83253].

    • Search Google Scholar
    • Export Citation
  • Gleckler, P. J., , K. E. Taylor, , and C. Doutriaux, 2008: Performance metrics for climate models. J. Geophys. Res., 113 , D06104. doi:10.1029/2007JD008972.

    • Search Google Scholar
    • Export Citation
  • Haylock, M. R., , N. Hofstra, , A. M. G. Klein Tank, , E. J. Klok, , P. D. Jones, , and M. New, 2008: A European daily high-resolution gridded dataset of surface temperature and precipitation for 1950–2006. J. Geophys. Res., 113 , D20119. doi:10.1029/2008JD010201.

    • Search Google Scholar
    • Export Citation
  • Kållberg, P., 1997: Aspects of the re-analysed climate. ECMWF ERA-15 Project Rep. Series 2, European Centre for Medium-Range Weather Forecasts, 89 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id=83254].

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

  • Kanamitsu, M., , W. Ebisuzaki, , J. Woollen, , S-K. Yang, , J. J. Hnilo, , M. Fiorino, , and G. L. Potter, 2002: NCEP–DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83 , 16311643.

    • Search Google Scholar
    • Export Citation
  • 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
  • Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 , 11381140.

  • Ma, L., , T. Zhang, , Q. Li, , O. W. Frauenfeld, , and D. Qin, 2008: Evaluation of ERA-40, NCEP-1, and NCEP-2 reanalysis air temperatures with ground-based measurements in China. J. Geophys. Res., 113 , D15115. doi:10.1029/2007JD009549.

    • Search Google Scholar
    • Export Citation
  • Onogi, K., and Coauthors, 2007: The JRA-25 Reanalysis. J. Meteor. Soc. Japan, 85 , 369432.

  • Perkins, S. E., , A. J. Pitman, , N. J. Holbrook, , and J. McAneney, 2007: Evaluation of the AR4 climate models’ simulated daily maximum temperature, minimum temperature, and precipitation over Australia using probability density functions. J. Climate, 20 , 43564376.

    • Search Google Scholar
    • Export Citation
  • Pitman, A. J., 2003: The evolution of, and revolution in, land surface schemes designed for climate models. Int. J. Climatol., 23 , 479510.

    • Search Google Scholar
    • Export Citation
  • Schubert, S., and Coauthors, 1995: A multi-year assimilation with the GEOS-1 system: Overview and results. NASA Tech. Memo. 104606, 182 pp. [Available from NASA Center for Aerospace Information, 800 Elkridge Landing Rd., Linthicum Heights, MD 21090].

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., , and J. G. Olson, 1988: An evaluation and intercomparison of global analyses from NMC and ECMWF. Bull. Amer. Meteor. Soc., 69 , 10471057.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012. doi:10.1256/qj.04.176.

  • Vincent, L. A., , X. Zhang, , B. R. Bonsal, , and W. D. Hogg, 2002: Homogenization of daily temperatures over Canada. J. Climate, 15 , 13221334.

    • Search Google Scholar
    • Export Citation
  • Williams Jr., C. N., , M. J. Menne, , R. S. Vose, , and D. R. Easterling, cited. 2006: United States Historical Climatology Network Daily Temperature, Precipitation, and Snow Data. ORNL/CDIAC-118, NDP-070, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy. [Available online at http://cdiac.ornl.gov/epubs/ndp/ushcn/usa.html].

    • Search Google Scholar
    • Export Citation
  • View in gallery

    For the 2-m air temperature in °C (a) mean TMAX difference (NCEP-2 minus ERA-40), (b) difference (NCEP-2 minus ERA-40), (c) skill score for NCEP-2 compared to ERA-40, (d) mean TMAX difference (JRA-25 minus ERA-40), (e) difference (JRA-25 minus ERA-40), (f) skill score for JRA-25 compared to ERA-40, (g) mean TMAX difference (JRA-25 minus NCEP-2), (h) difference (JRA-25 minus NCEP-2), and (i) skill score for JRA-25 compared to NCEP-2.

  • View in gallery

    As in Fig. 1, but for the 1000-hPa temperature.

  • View in gallery

    For the 2-m air temperature (°C) (a) mean TMIN difference (NCEP-2 minus ERA-40), (b) difference (NCEP-2 minus ERA-40), (c) skill score for NCEP-2 compared to ERA-40, (d) mean TMIN difference (JRA-25 minus ERA-40), (e) difference (JRA-25 minus ERA-40), (f) skill score for JRA-25 compared to ERA-40, (g) mean TMIN difference (JRA-25 minus NCEP-2), (h) difference (JRA-25 minus NCEP-2), and (i) skill score for JRA-25 compared to NCEP-2.

  • View in gallery

    As in Fig. 3, but for the 1000-hPa TMIN.

  • View in gallery

    Probability density functions for 2-m TMAX for regions described in Table 1: observed (thick solid line), ERA-40 (thin solid line), NCEP-2 (dotted line), and JRA-25 (dashed line).

  • View in gallery

    Skill score for each reanalysis compared to the observed for 2-m TMAX for each region described in Table 1: ERA-40 (solid bar), NCEP-2 (dotted bar), and JRA-25 (dashed bar).

  • View in gallery

    Difference from observed in for each region described in Table 1: ERA-40 (solid bar), NCEP-2 (dotted bar), and JRA-25 (dashed bar).

  • View in gallery

    As in Fig. 5, but for TMIN.

  • View in gallery

    As in Fig. 6, but for TMIN.

  • View in gallery

    As in Fig. 7, but for TMIN.

  • View in gallery

    Latitudinal-average differences (°C) in the PDF overlap (averaged over the three reanalyses) for (a) TMAX and (b) TMIN in a latitude band from 40° to 70°N and (c) TMAX and (d) TMIN in a latitude band from 15°N to 15°S. The heavy (light) solid line is for the 1000-hPa air temperature (2-m temperature). Note difference in scale between (a),(b) and (c),(d).

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 92 92 3
PDF Downloads 53 53 0

Global and Regional Comparison of Daily 2-m and 1000-hPa Maximum and Minimum Temperatures in Three Global Reanalyses

View More View Less
  • 1 Climate Change Research Centre, University of New South Wales, Sydney, New South Wales, Australia
© Get Permissions
Full access

Abstract

A comparison of three global reanalyses is conducted based on probability density functions of daily maximum and minimum temperature at 2-m and 1000-hPa levels. The three reanalyses compare very favorably in both maximum and minimum temperatures at 1000 hPa, in both the mean and the 99.7th and 0.3rd percentiles of both quantities in most regions. At 2 m, there are large and widespread differences in the mean and 99.7th percentiles in maximum temperature between the three reanalyses over land commonly exceeding ±5°C and regionally exceeding ±10°C. The 2-m minimum temperatures compare unfavorably between the three reanalyses over virtually all continental surfaces with differences exceeding ±10°C over widespread areas. It is concluded that the three reanalyses are generally interchangeable in 1000-hPa temperatures. The three reanalyses of 2-m temperatures are very different owing to the methods used to diagnose these quantities. At this time, the probability distribution functions of the 2-m temperatures from the three reanalyses are sufficiently different that either the 2-m air temperatures should not be used or all three products should be used independently in any application and the differences highlighted.

Corresponding author address: Professor A. J. Pitman, Climate Change Research Centre, Mathews Building, University of New South Wales, Sydney, NSW 2052, Australia. Email: a.pitman@unsw.edu.au

A comment/reply has been published regarding this article and can be found at http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-12-00122.1 and http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-12-00216.1

Abstract

A comparison of three global reanalyses is conducted based on probability density functions of daily maximum and minimum temperature at 2-m and 1000-hPa levels. The three reanalyses compare very favorably in both maximum and minimum temperatures at 1000 hPa, in both the mean and the 99.7th and 0.3rd percentiles of both quantities in most regions. At 2 m, there are large and widespread differences in the mean and 99.7th percentiles in maximum temperature between the three reanalyses over land commonly exceeding ±5°C and regionally exceeding ±10°C. The 2-m minimum temperatures compare unfavorably between the three reanalyses over virtually all continental surfaces with differences exceeding ±10°C over widespread areas. It is concluded that the three reanalyses are generally interchangeable in 1000-hPa temperatures. The three reanalyses of 2-m temperatures are very different owing to the methods used to diagnose these quantities. At this time, the probability distribution functions of the 2-m temperatures from the three reanalyses are sufficiently different that either the 2-m air temperatures should not be used or all three products should be used independently in any application and the differences highlighted.

Corresponding author address: Professor A. J. Pitman, Climate Change Research Centre, Mathews Building, University of New South Wales, Sydney, NSW 2052, Australia. Email: a.pitman@unsw.edu.au

A comment/reply has been published regarding this article and can be found at http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-12-00122.1 and http://journals.ametsoc.org/doi/abs/10.1175/JCLI-D-12-00216.1

1. Introduction

Global reanalyses were proposed by Bengtsson and Shukla (1988) and Trenberth and Olson (1988), resulting in several first-generation products. The 15-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-15) was a 15-yr product starting from 1979 (Gibson et al. 1997). The National Centers for Environmental Prediction (NCEP), in collaboration with the National Center for Atmospheric Research (NCAR), performed a reanalysis from 1948 (Kalnay et al. 1996; Kistler et al. 2001). Finally, a 16-yr reanalysis from March 1980 was produced by the Data Assimilation Office of the National Aeronautics and Space Administration (Schubert et al. 1995).

These first-generation reanalysis projects were successful and the resulting data have become valued in climate and hydrological sciences, for forcing stand-alone components of models, for model evaluation, and for exploring specific trends or phenomenon. However, the first-generation reanalyses were also shown to contain deficiencies (Kalnay et al. 1996; Kållberg 1997) leading to revised products using the latest versions of the numerical models and updated data assimilation systems.

Deficiencies in the original NCEP–NCAR reanalysis were addressed in a revised reanalysis by NCEP/Department of Energy (NCEP-2) by Kanamitsu et al. (2002). ECMWF conducted a new reanalysis (ERA-40) (Uppala et al. 2005) to address deficiencies in ERA-15. Most recently, the Japanese Meteorological Agency (JMA) has conducted a 25-yr global reanalysis aimed at producing higher quality reanalyses over eastern Asia and the tropics (JMA-25) (Onogi et al. 2007). These three reanalyses have been individually assessed by the lead modeling teams and by independent groups. Uppala et al. (2005) note that the ERA-40 product is better than the first-generation ERA-15 and NCEP–NCAR reanalyses, although they note excessive precipitation over tropical oceans. They state that the ERA-40 analysis has considerable potential to contribute to studies of atmospheric, oceanic and surface processes, and to studies of the variability and predictability of climate. Onogi et al. (2007) highlight areas where they suggest the JRA-25 reanalysis is superior to earlier products while noting areas of weakness, such as a dry bias over the Amazon. Overall, based on a reading of the literature, an objective assessment would be that existing global reanalyses are genuinely impressive products.

This paper focuses on a comparison of the daily minimum and maximum temperatures from the three reanalyses derived from the four times per day temperatures reported by each group. Evidence from the literature suggests high confidence in the reanalyses at monthly, annual, and climatological time scales, etc. Kanamitsu et al. (2002), Uppala et al. (2005), and Onogi et al. (2007) each show reanalyses to perform well at the global or continental scale for monthly or annually averaged near-surface air (2 m) temperature. However, monthly averages can hide serious biases and, since subdaily data are provided by each reanalysis group, a comparison of these data may highlight weaknesses. Further, while monthly and annual averages are very useful for model evaluation, reanalyses are increasingly being used to explore finer temporal scales. This includes trend analyses (see Uppala et al. 2005 for discussion) and boundary conditions for modeling extreme events, etc.

This paper explores the similarity between the three reanalyses using two types of modeled data. The first type comprises data that result from fundamental model equations, informed through the assimilation of a variety of observations at various levels in the atmosphere and with varying spatial and temporal density. These data include the 1000-hPa daily maximum and minimum temperatures. The second type is diagnosed temperature, specifically the 2-m air temperature. Gleckler et al. (2008) state that the 2-m air temperature is a diagnostic quantity, computed from complex iterative schemes based on conditions in the lowest model level and near the surface. They comment that, as the 2-m air temperature is not actually needed in any of the equations that affect the model’s simulation, errors in these fields could reflect poor diagnostics rather that fundamental model errors. We note that the diagnosed 2-m air temperature may also reflect errors in soil moisture fields. These are nudged by adding or removing water to minimize errors in quantities in the atmosphere rather than errors in the diagnosed 2-m quantities.

ERA-40 diagnoses a 2-m air temperature indirectly as part of the data assimilation process (Uppala et al. 2005). A separate analysis of measurements of dry-bulb temperature and dewpoint is made (Douville et al. 1998) using a background field derived from the background forecast of the main data assimilation system. This is then interpolated between the surface and the lowest model level using Monin–Obukhov similarity profiles consistent with the assimilating model’s parameterization of the surface-layer part of the planetary boundary layer (Beljaars and Viterbo 1999). JRA-25 also uses a separate surface analysis via a two-dimensional optimum interpolation that Onogi et al. (2007) describe as similar to ECMWF. NCEP-2 diagnoses the 2-m air temperature as a function of surface skin temperature, lowest atmospheric model temperature, vertical stability, and surface roughness (W. Ebisuzaki 2008, personal communication).

The question therefore arises whether the way the three reanalyses diagnose 2-m air temperature from the prognosed lowest model and land surface temperatures introduces more or less variation between the three products. Is agreement in the 2-m temperature similar to the 1000-hPa temperature for example? This matters because, while there are caveats to the use of the diagnosed 2-m air temperature (as highlighted by Gleckler et al. 2008), they are used as if they were observations by some groups.

We are aware that there are differences in how the three reanalyses interpolate the atmospheric temperature from a model level to 1000 hPa. These tend to lead to differences over regions of high orography. We have retained these regions in this paper because users of reanalyses who undertake large-scale analyses do not usually discard these regions. We also note that, as discussed above, there are significant differences in how 2-m air temperatures are diagnosed. While it could be argued that these make the comparison of the three products inappropriate, we note that these differences are not taken into account when the products are actually used. The 2-m air temperatures are available at subdiurnal scales and are commonly used as “observations” without any adjustment for differences in how they were derived.

This paper therefore compares the performance of three second-generation global reanalyses products in terms of their capacity to simulate 2-m and 1000-hPa air temperature, focusing on the daily maximum (TMAX) and daily minimum (TMIN) temperatures. We also compare the three reanalyses capacity to simulate more extreme values—the 99.7th percentile for TMAX and the 0.3rd percentile for TMIN . These percentiles are approximately the annual event. To compare the overall similarity in the simulation of daily 2-m and 1000-hPa TMAX and TMIN, the approach of Perkins et al. (2007) is used to compare the similarity in the overall probability density functions. We compare each reanalysis’ probability density function (PDF) to observed PDFs over regions where daily observed TMAX and TMIN were available, noting that this requires careful interpretation since the reanalyses provide maximum/minimum temperatures four times per day while the observations were near instantaneous.

2. Methodology

a. Reanalysis data

The reanalyses data used in this paper were sourced as follows. The ERA-40 data was sourced from ECMWF (see http://www.ecmwf.int/research/era/do/get/era-40). All fields were selected at a 2.5° × 2.5° resolution. Daily 2-m and 1000-hPa air temperature were downloaded for 1981–2000. The NCEP-2 data was downloaded from http://www.cdc.noaa.gov/cdc/data.ncep.reanalysis2.html for 2-m and 1000-hPa air temperature. The original data were sourced at approximately 1.8° × 1.8° and averaged to 2.5° × 2.5° before use. The JRA-25 2-m and 1000-hPa temperatures were sourced from http://jra.kishou.go.jp/JRA-25/index_en.html and were available at 2.5° × 2.5°. For each of ERA-40, NCEP-2, and JRA-25, daily datasets contained four records per day. The highest value was defined as TMAX and the lowest as TMIN.

We corrected the reanalyses for differences in the orography, which largely results from different model resolutions following the approach of Ma et al. (2008). This simply adjusts the temperatures using a standard lapse rate (6.5°C/1000 m) to a common orography (we used the ERA-40 orography field as the standard). This correction was small over all regions but was largest over the Andes and Himalayas. There are differences in orography of up to 500 m over these regions between the reanalyses, and this can lead to differences in temperature of several degrees. It will be shown that the impact of orography is small compared to the differences between the three reanalyses.

b. Regional observational data

There are no global observational datasets available to derive the daily PDFs for TMAX or TMIN. However, there are daily observations of TMAX and TMIN available for some regions, including many of the Global Energy and Water Cycle Experiment (GEWEX) continental-scale experiment regions (see Table 1). Six of these continental-scale experiment regions are analyzed here.

For the Mississippi basin, daily TMAX and TMIN were obtained from the U.S. historical climatology network (USHCN) (Williams et al. 2006). For the Mackenzie basin, data were obtained from the archive of Vincent et al. (2002) who provide a quality-controlled archive of stations including daily TMAX and TMIN. For the Amazon and Africa, the NOAA daily global summary of day (GOBALSOD) data were used (see http://ingrid.ldeo.columbia.edu/SOURCES/.NOAA/.NCDC/.DAILY/.GLOBALSOD). All stations within the domain defined in Table 1 were used. For the Baltic region, data were obtained from http://eca.knmi.nl/download/ensembles/download.php. These are the data provided through the Ensemble-based Prediction of Climate Change and their Impacts (ENSEMBLES) project (see http://www.ensembles-eu.org; see Haylock et al. 2008). Finally, for the Murray–Darling basin (MDB) daily observed TMAX and TMIN were obtained from the Australian Bureau of Meteorology. Further details on record lengths and number of stations are provided in Table 1.

All observed data within each the region (see Table 1) were concatenated and then used to derive the PDF. Perkins et al. (2007) showed that the resulting PDFs were very robust to sampling of the observed stations, missing data, and large random errors in the observations given the amount of daily data that formed the PDF. However, there are likely limits to the reliability of the observations, most likely where specific quality-controlled products have not been developed (e.g., Amazon, Africa), and we therefore use these data with care.

c. PDF skill score

The PDF score was derived following Perkins et al. (2007). This score measures the overlap between two PDFs such that the common area between two PDFs equals 1.0 for a perfect model (where the two PDFs overlap perfectly) and 0.0 where the two PDFs are independent. In deriving the PDFs, bin sizes were 0.5°C for TMAX and TMIN. The PDF comparison statistic has the useful attribute of measuring the similarity of a distribution in a single number. We explored the sensitivity of our results to smaller bin sizes (up to 0.01°C for temperature) with no substantial impacts.

There is one important caveat to the use of the skill score based on the overlap of the PDFs. Where the variance in a quantity is small (e.g., temperature in the tropics relative to the midlatitudes), a small bias between two PDFs can lead to a very small overlap measure, while that same bias in the midlatitudes would have a small impact. We explored making bin sizes proportional to the variance, or normalizing by the variance, but concluded that in regions of low variance a reanalysis should be more able to capture the PDF than in areas of high variance, and therefore hiding the differences shown over the tropics was potentially misleading.

The overlap of PDFs based on daily data is a very simple measure that quantifies a considerable amount of model behavior. It is sensitive to the mean but also to the shape and range of the distribution. By incorporating the statistics of daily probabilities over the length of data into a single measure, the overlap measure provides more insight than a mean and standard deviation, which are commonly used in model evaluation. While the mean and standard deviation requires an a priori decision on what aspects of a distribution matter, the overlap statistic simply measures the similarity between two distributions, highlighting similarities and differences based on the probability of a quantity being represented.

3. Global-scale results

a. TMAX and

Figure 1a shows the difference between the NCEP-2 and ERA-40 simulation of the 2-m TMAX. Differences exceed 10°C over Antarctica, the Himalayas, Rockies, and Andes. Differences exceed 2°C over most of Africa, parts of northeast North America, and eastern Asia. ERA-40 is systematically warmer than NCEP-2 over land, but NCEP-2 is 2°–4°C warmer over widespread areas of the subtropical oceans. Comparing JRA-25 with ERA-40 (Fig. 1d) provides a similar scale of difference over land, although there are fewer differences over the oceans. Comparing JRA-25 with NCEP-2 (Fig. 1g) highlights a 4°–10°C difference over North Africa, 2°–4°C over the Amazon, and 5°–10°C over parts of Antarctica. JRA-25 is systematically warmer than NCEP-2 over these regions. Not surprisingly, the differences between the reanalyses are typically smaller over well-instrumented continents—including most of North America, Europe, and Australia—and relatively large over Africa and Asia.

An exploration of shows very large differences in some regions between the reanalyses. Figures 1b,e,h show differences exceeding 5°C and locally exceeding 20°C. Differences exceeding 5°C occur in well-instrumented regions, including the U.S. mainland (Fig. 1h), although there are small differences over most continents where observational data are routinely available (e.g., Europe, United States, and Australia, etc).

The skill score proposed by Perkins et al. (2007) was used to measure the similarity between the simulations of the 2-m air temperature via the overlap of the PDFs derived using the daily data. Perkins et al. used this measure to compare climate models; here we compare the reanalyses. Figures 1c,f,i show the amount of overlap in the PDFs between the three reanalyses. There is no objective measure of what is “good,” but Perkins et al. show that some coupled climate models used in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) overlapped the observed PDF over Australia, region-by-region, by over 0.9. Using this threshold to compare the reanalyses, Figs. 1c,f,i show that the three reanalyses overlap by more than 0.9 over Australia, parts of Europe, and the eastern United States. Overlap is below 0.7 in many regions, particularly tropical land and oceans.

Figures 2a,d,g shows the equivalent simulation of TMAX but at 1000 hPa. This shows a higher degree of agreement between the three reanalyses. Differences remain over regions of high orography, but these are small and isolated in comparison to the 2-m TMAX. The largest difference relates to the JRA-25 simulation over the Amazon, which is relatively warm, leading to differences in TMAX of up to 2°–5°C. Over most of the globe, the three reanalyses agree to within ±2°C. This similarity largely extends to (Figs. 2b,e,h) except over the Amazon. Over most regions, the three reanalyses agree to within ±2°C except over high orography.

If the overall overlap of the PDFs is compared (Figs. 2c,f,i), the 1000-hPa TMAX is in agreement to over 0.9 over almost all continental surfaces. Over the tropical oceans, the PDFs agree to below 0.7 and below 0.5 in the western Pacific. In terms of overall agreement therefore, the three reanalyses are very similar over the whole PDF above the continents and agree poorly over the tropical oceans.

b. TMIN and

The NCEP-2 and ERA-40 reanalyses differ by more than 5°C over most of the continental surface (Fig. 3a). Locally, differences exceed 10°C (South Africa, North Africa, Rockies, Andes, etc.). There are no significant regions where NCEP-2 and ERA-40 agree to within ±2°C. JRA-25 and ERA-40 agree to within ±2°C over parts of northern Europe and the Amazon (Fig. 3d). JRA-25 and NCEP-2 agree over most of the continental surfaces to within ±2°C, but disagree by more than 10°C over parts of Antarctica (Fig. 3g).

In the simulation of , agreement between the three reanalyses is negligible over the continental surfaces. NCEP-2 and ERA-40 differ by more than 5°C over widespread regions (Fig. 3b), JRA-25 differs from ERA-40 by more than 5°C over widespread regions (Fig. 3e), and JRA-25 differs from NCEP-2 by more than 10°C over Northern Hemisphere surfaces, by 5°–10°C over North Africa and by 2–4°C over most Southern Hemisphere continental surfaces (Fig. 3h).

In terms of the similarity over the PDFs, Fig. 3c shows agreement between NCEP-2 and ERA-40 is less than 0.5 over tropical and subtropical continental surfaces, but reaches 0.8–0.9 over northern Eurasia and the eastern United States. This is similar to JRA-25 and ERA-40 (Fig. 3f). Agreement between JRA-25 and NCEP-2 is higher but only reaches 0.9 over the continental surfaces of western Europe (Fig. 3i).

Figure 4 shows the equivalent results for the 1000-hPa height. In comparison to the 2-m TMIN, agreement between the three reanalyses is very good. There are differences in the means over high orography and over southern Africa and Australia between JRA-25 and ERA-40 (Fig. 4d), but differences only locally exceed 4°C. A similar result occurs for ; locally there are larger differences between the three analyses, but relative to results shown in Fig. 3 these differences are small. The overall comparison of the PDFs (Figs. 4c,f,i) shows a level agreement above the continental surfaces exceeding 0.9 over large areas and exceeding 0.8 almost everywhere. As with TMAX (Fig. 2), there are large and widespread differences between the reanalyses over the tropics measured by the overlaps of the PDFs.

4. Regional analyses

Reanalysis-to-reanalysis comparison is not as useful as a reanalysis-to-observation comparison. It is not possible to do a global comparison of air temperature reanalyses with independent observations at a daily time scale; the data do not exist, but it is possible to compare the reanalyses with regional observations. However, this has to be performed carefully given the differences between an observed (near-instantaneous measurements) and reanalyses estimates derived from four samples per day.

Six GEWEX continental-scale experiment regions were explored in more detail by focusing on the simulation of the PDFs by each reanalysis of TMAX and TMIN at 2 m. Daily observations were used to derive PDFs for 2-m TMAX and TMIN, but daily observations were not available for the 1000-hPa temperatures.

a. TMAX and

Figure 5 shows the PDFs for TMAX for each region for 2 m. The similarity between each PDF and the observed is shown in Fig. 6 and Table 2. The return values for are shown in Table 3, and their difference from the observed is shown in Fig. 7.

The observed PDF for TMAX for Africa is normally distributed. The three reanalyses all underestimate the probability of values in the 35°–40°C range (Fig. 5a) and overestimate the probability of values below ∼30°C. The overall skill score exceeds 0.7 for all three reanalyses (Fig. 6) with ERA-40 reaching 0.85. The observed (45.0°C) is captured well by all three reanalyses (Fig. 7), although all three reanalyses have a cold bias compared to the observed. Perhaps surprisingly, given the lack of observed data, the African region is captured almost as well as any region by the three reanalyses (Figs. 6 and 7).

The observed PDF for TMAX for the Amazon is also normally distributed (Fig. 5b) but is modeled extremely poorly by all three reanalyses. The overlap of the reanalyses with the observed PDF (Fig. 6, Table 2) shows values between 0.36 and 0.56—very substantially worse than any other region analyzed. The observed (38.2°C) is underestimated by 3°C by ERA-40 and 1.8°C by NCEP-2 and is overestimated by 2.0°C by JRA-25 (Fig. 7). Figure 5b suggests some similarity between the three reanalysis-simulated PDFs; thus it is conceivable that part of the problem relates to observational limits. There is no observational evidence for the shape of the upper tail simulated by JRA-25, and this is likely related to the dry bias—low latent heat flux and therefore high maximum temperatures in this region.

The overall shape of the observed TMAX PDF for the MDB is normal (Fig. 5c). The three reanalyses capture the observed PDF with an overlap of 0.85–0.88 (Fig. 6). There is a systematic underprediction of values in the midrange of the observed PDF, which leads to an overprediction by all three reanalyses of . The observed (39.5°C) is best captured by ERA-40, which is too high by 1.4°C (Fig. 7). NCEP-2 and JRA-25 are 3.5°–3.9°C too warm.

The shape of the observed TMAX PDF for the Baltic is bimodal and all three reanalyses capture this extremely well (Fig. 5d) with an overlap of 0.9–0.93 (Fig. 6). This bimodality is linked to a higher probability of temperatures close to 0°C, probably associated with phase changes. ERA-40 underestimates the observed by 2.0°C and JRA-25 underestimates it by 1.4°C and NCEP-2 by 0.6°C (Fig. 7). However, NCEP-2 simulates a relatively high probability of TMAX at 0.0°C and a low probability of 0.0°–0.5°C, which is not seen in the observed. This would appear to be a systematic bias in the NCEP-2 reanalysis of 2-m air temperatures in cold regions (see Mackenzie basin).

The shape of the observed TMAX PDF for the Mississippi is not normally distributed (Fig. 5e), but the basic shape of the observed is captured by the three reanalyses with overlap statistics of 0.80 or higher (Fig. 6). All reanalyses overestimate the lower tail of the observed distribution, overestimate the probability of values around 30°C, and underestimate by up to 6.6°C (JRA-25, Fig. 7). The “sawtooth” nature of the observed distribution is of concern, and it may be that some of the differences seen over the Mississippi are related to biased observations.

The shape of the observed TMAX PDF for the Mackenzie basin is quite flat (Fig. 5f), but the observed PDF is captured superbly by the three reanalyses with overlap statistics close to 0.9 (Fig. 6). All reanalyses underestimate by more than 1°C and NCEP-2 exceeds 5°C (Fig. 7). Since this is a carefully derived dataset (Vincent et al. 2002) this is unlikely to be related to observational errors. NCEP-2 shows the same high probability of TMAX of 0°C and low probability of temperatures of above 0°C but below 5°C as was noted for the Baltic.

Based on this analysis—and a sample of six regions—there is no clear preferred reanalysis in terms of TMAX or . Figure 6 shows that the skill of each reanalysis is basically similar in the overall PDF, and Fig. 7 shows that the best/worst model is very regionally dependent. We also analyzed the 1000-hPa air temperature PDFs—but without an observed PDF. The shapes of, and differences between, reanalysis PDFs are generally very similar to the 2-m air temperature and are therefore not shown.

b. TMIN and

Figure 8 shows the observed and reanalysis PDFs for each region for TMIN. The distributions for Africa, the Amazon, and the MDB are normally distributed, but there is a bimodality to the distributions for the Baltic and Mackenzie basins associated with high probabilities of temperatures at 0°C. In comparison to TMAX there is a strong sense of weaker similarity to the observed. The African region (Fig. 8a) is well captured by NCEP-2 and JRA-25 with overlap statistics exceeding 0.85 (Fig. 9). ERA-40 is poor in this region for TMIN, as is clear in Fig. 8a, with an overlap statistic of 0.61. ERA-40 is 4.3°C too warm in but less obvious is a 3.3°C bias in NCEP-2 (Fig. 10). JRA-25 captures to within 0.5°C.

The Amazon region displays similar characteristics with ERA-40 too warm, representing the observed PDF very poorly (PDF overlap = 0.54) relative to JRA-25 or NCEP-2 (>0.80, see Fig. 9). In ERA-40 is 6.3°C too warm (Table 3) compared to 3.6°C too warm in JRA-25 and a zero error in NCEP-2. Results are similar for the MDB and Mississippi basin. Figure 9 shows that the overall performance of ERA-40 is substantially worse than JRA-25 or NCEP-2 in capturing TMIN in warm climates (Africa, Amazon, MDB, Mississippi) and marginally weaker than these reanalyses in cold climates. Figure 10 shows that ERA-40 systematically overestimates in all regions relative to the observed and usually relative to the other reanalyses. JRA-25 also always overestimates while NCEP-2 appears to avoid this systematic bias.

In contrast to TMAX, a general feature of the TMIN analysis is a systematic bias in ERA-40 compared to the observed and generally compared to the NCEP-2 and JRA-25 reanalyses. ERA-40 is clearly weakest in capturing the observed PDF of TMIN over all regions (marginally in the case of the Baltic, Fig. 9) and has the largest bias in the simulation of (Fig. 10) in five of the six regions. NCEP-2 captures the best in five of the six regions.

We also analyzed the 1000-hPa air temperature PDFs—but without an observed PDF. The shapes of, and differences between, reanalysis PDFs are generally very similar to the 2-m air temperature and are therefore not shown.

5. Discussion

The overlaps of the PDFs shown in the global-scale maps represent a measure of the similarity of the three analyses. In the simulation of the 1000-hPa temperatures, for TMAX and TMIN, the similarity of the PDFs between the reanalyses exceeds 90% over large fractions of the Northern Hemisphere land and ocean. The agreement is slightly better for TMAX (Fig. 2) than TMIN (Fig. 4) and no reanalysis appears anomalous compared to the other two. This strong agreement between the reanalysis in the 1000-hPa temperatures is very much weaker in the diagnosed 2-m TMAX and TMIN and both TMAX and . ERA-40 appears anomalous compared to JRA-25 and NCEP-2.

While Figs. 1a,d,g show TMAX differences that are relatively localized, Figs. 3a,d,g show TMIN differences over most continental surfaces that exceed 4°C. Figure 11 highlights this across two longitudinally averaged bands (40°–70°N and 15°N–15°S) averaged over the three reanalyses. Figure 11 shows the similarity between the averages of the three reanalyses exceeds 0.85 for TMAX (Fig. 11a) and TMIN (Fig. 11b) and exceeds 0.9 in all regions other than those affected by the Rockies and Himalayas in the latitude band 40°–70°N. The contrast with a tropical latitude region is very clear in Fig. 11c (TMAX) and Fig. 11d (TMIN). Agreement across the reanalyses is weaker (below ∼0.8 at all longitudes) occasionally dropping to ∼0.6 in some regions with the 1000-hPa temperature. However, this is systematically better than the 2-m temperature where the average of the PDF overlaps drops below 0.6 locally for TMAX and over large areas for TMIN.

There are a suite of reasons why this might occur. Surface skin temperatures can be highly variable in time and space and are not routinely assimilated. In contrast, observations of air temperature are more common and their assimilation is more routinely performed into the models at various levels. Second, diagnosing the 2-m temperatures requires a range of procedures that are not well documented between the three reanalyses. The parameterization of the boundary layer, simulation of surface soil moisture, coupling of the land to the atmosphere (Koster et al. 2004), and correct specification of surface characteristics all present challenges in modeling that might limit model skill in diagnosing the 2-m air temperature. While the 1000-hPa temperatures are primarily the result of atmospheric dynamics and physics constrained by observations assimilated into various atmospheric levels, the 2-m air temperature responds to soil temperatures, which in turn respond to soil moisture, land characteristics, the parameterization of turbulent energy fluxes, etc. What is perhaps more surprising is that the reanalyses are more similar in TMAX than TMIN. To capture TMAX requires net radiation and the partitioning of the net radiation to be captured well (Pitman 2003). The TMAX responds to moisture stress in soil evaporation and in transpiration and is a key reason for the incorporation of more sophisticated land surface models. However, TMIN requires “only” nighttime infrared losses to be captured, which would seem easier than the combination of all the factors that contribute to a daytime maximum. Most likely, the differences in TMIN relate to difficulties in simulating nighttime boundary layer structures and cloud impacts on infrared losses. Exploring the reasons for the differences in performance for TMAX and TMIN is beyond the scope of this paper, but we suggest a seasonal analysis may help isolate the mechanisms that lead to these differences.

Our results therefore suggest that the three reanalyses are very similar in their simulation of 1000-hPa TMAX and TMAX away from the tropics, and detailed regional assessments are needed to determine which are correct over the few regions where they differ. More care needs to be taken in using the 1000-hPa temperatures over the tropics, including the Amazon where Fig. 6 suggests all reanalyses are weak in TMAX and Fig. 9 suggests ERA-40 is weak in TMAX. It is not possible to judge if one of the reanalyses is superior to the others in the 1000-hPa temperature field because there are no daily observational data to compare with the modeled PDFs.

The regional analysis for TMAX could not show that one reanalysis was superior to another over most regions (Fig. 6). It is therefore not currently possible to judge if one of the reanalyses is superior to the others in TMAX, and all demonstrate substantial skill. However, Fig. 9 showed ERA-40 as less able to capture the PDF of TMIN than NCEP-2 or JRA-25. Reviewing Fig. 3, it is notable that the JRA-25 comparison with NCEP-2 shows very low differences compared to comparisons that include ERA-40. Comparison of the PDF overlaps shows similarities between NCEP-2 and JRA-25 exceeding 0.8 over almost all continental surfaces in Fig. 3i, while in Figs. 3c,f (where ERA-40 is included) the overlaps of the PDFs fall below 0.5. We therefore note some significant issues with respect to TMIN:

  1. The regional analyses in Fig. 9 for TMIN show ERA-40 to be weak relative to NCEP-2 and JRA-25 compared to observations in tropical and subtropical regions including Africa, the Amazon, the MDB, and the Mississippi basin;
  2. in tropical and subtropical regions, JRA-25 and NCEP-2 show strong similarities to each other (Fig. 3i) and appear able to simulate the regions of Africa, the Amazon, the MDB, and the Mississippi basin well (Fig. 9);
  3. NCEP-2 and JRA-25 appear consistently good across those regions examined (Fig. 9);
  4. all three global analyses (Figs. 3c,f,i) strongly agree with each other north of ∼45°N, and all three reanalyses agree with the observations equally well for TMIN in the two regions examined north of ∼45°N (the Mackenzie and Baltic regions, Fig. 9).

It is therefore reasonable to suggest that for TMIN, JRA-25, NCEP-2, and ERA-40 are indistinguishable north of ∼45°N—they simulate the regions (Fig. 9) equally well and the overlap of the PDFs shown in Fig. 3 is generally 0.8–0.9 over all continental surfaces. However, south of ∼45°N (including Southern Hemisphere land except Antarctica), our results suggest that, for TMIN, JRA-25 and NCEP-2 are indistinguishable (they simulate the regions in Fig. 9 equally well and the overlap of the PDFs shown in Fig. 3 is very similar). However, ERA-40 is systematically weaker in Fig. 9 compared with observations, and ERA-40 appears anomalous compared to both JRA-25 and NCEP-2 over land surfaces (Figs. 3c,f).

In the case of the percentiles, for and , there is considerable agreement between the three reanalyses at 1000 hPa. There are clearly differences on a regional scale between the reanalyses but the level of agreement for shown in Figs. 2b,e,h is not much different from the TMAX. Similarly, the level of agreement for at 1000 hPa shown in Figs. 4b,e,h is not much different from TMIN. This is clearly not true of the 2-m temperatures where agreement between and differ greatly. This is most clear in Figs. 3b,e,h where differences in exceeding 10°C are common.

6. Conclusions

While the mean monthly and annual performance of global reanalyses are routinely assessed and demonstrated to be high quality, the daily performance of the reanalyses is less commonly analyzed. This paper has used probability density functions of daily minimum and maximum temperatures at 1000 hPa and at 2 m to evaluate the similarity between the reanalyses and where possible observations. Key conclusions are the following:

  • The 1000-hPa temperatures, when compared between the reanalyses, are generally very similar between the three products. Regional differences can be identified and are commonly associated with high orography and possibly with desert regions for . However, even at the tails of the distributions over individual regions the 1000-hPa temperatures for the three reanalyses are similar to each other. While this does not mean that they are necessarily correct, it certainly builds further confidence in these products.
  • Estimates of 2-m TMAX and in for the three global reanalyses are more dissimilar to each other than the 1000-hPa temperatures. Large regional differences are apparent but overall no reanalysis appears inferior to the other two (Figs. 6 and 7).
  • Results suggest that for TMIN JRA-25, NCEP-2, and ERA-40 are indistinguishable north of ∼45°N—they simulate the regions equally well and the overlaps of the PDFs shown in Fig. 3 are generally 0.8–0.9 over all continental surfaces. South of ∼45°N our results suggest that JRA-25 and NCEP-2 are indistinguishable, but ERA-40 appears anomalous compared to both JRA-25, NCEP-2, and, where available, observations.
  • At the regional scales there are marked similarities in the PDFs between simulation of TMAX and observations for the three reanalyses, with the exception of the Amazon where all three reanalyses fail to match the observations. NCEP-2 simulates a high probability of 2-m air temperatures at 0.0°C in the Baltic and Mackenzie basins, which does not appear in the observations. Here is substantially better simulated than .

These results imply that users of reanalyses can use the daily data for the 1000-hPa temperatures, including relatively extreme values, and be assured that each of the three reanalyses produce broadly similar results. This paper could not verify whether the three reanalyses were accurate since there are no observed data to compare with them, but no single reanalysis appears inconsistent with the other two. The use of the 2-m air temperatures is substantially less reliable, and the rarer values within a distribution are simulated very differently between the three products (Figs. 7 and 10).

We suggest that the use of the 2-m air temperatures from the reanalyses should either not be used or all three reanalyses should be used independently (we strongly advise against the temptation to average the three products) and results should reflect uncertainty resulting from the variations in the 2-m air temperatures between the reanalyses. The use of the three products needs to be continued until a daily based evaluation of each reanalysis with independent observations can be made.

Finally, we make three suggestions. First, evaluating the three reanalyses using monthly or annual averages hides very significant differences in some modeled products. PDFs can usefully identify some of these weaknesses from daily resolution data by evaluating the differences between the reanalyses via a single metric. Second, we suggest that there is merit in a systematic documentation and comparison of how the 2-m temperatures (and other products) are derived. It is not possible to determine this from the literature in enough detail to explain why the three reanalyses produce such similar 1000-hPa products but so different 2-m products. Finally, it would be useful if all three reanalyses could report comparable near-instantaneous daily TMAX and future TMAX.

Acknowledgments

NCEP-2 reanalysis data were provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their Web site at http://www.cdc.noaa.gov/. The JRA-25 data were provided from the cooperative research project of the JRA-25 long-term reanalysis by the Japan Meteorological Agency (JMA) and the Central Research Institute of Electric Power Industry (CRIEPI). We are also very grateful to ECMWF for providing access to the ERA-40 data. We acknowledge the E-Obs dataset from the EU-FP6 project ENSEMBLES (http://www.ensembles-eu.org) and the data providers in the ECA&D project (http://eca.knmi.nl). Andrew Slater provided very useful advice on a draft of this paper. Two anonymous reviewers provided advice that led to major revisions to an earlier version of this manuscript.

REFERENCES

  • Beljaars, A. C. M., , and P. Viterbo, 1999: The role of the boundary layer in a numerical weather prediction model. Clear and Cloudy Boundary Layers, A. A. M. Holtslag and P. G. Duynkerke, Eds., North Holland Publishers, 287–304.

    • Search Google Scholar
    • Export Citation
  • Bengtsson, L., , and J. Shukla, 1988: Integration of space and in situ observations to study global climate change. Bull. Amer. Meteor. Soc., 69 , 11301143.

    • Search Google Scholar
    • Export Citation
  • Douville, H., , J-F. Mahfouf, , S. Saarinen, , and P. Viterbo, 1998: The ECMWF surface analysis: Diagnostics and prospects. ECMWF Tech. Memo. 258, European Centre for Medium-Range Weather Forecasts, 51 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/tm/001-300/tm258.pdf].

    • Search Google Scholar
    • Export Citation
  • Gibson, J. K., , P. Kallberg, , S. Uppala, , A. Nomura, , A. Hernandez, , and E. Serrano, 1997: ERA description. ECMWF ERA-15 Project Rep. Series 1, European Centre for Medium-Range Weather Forecasts, 72 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id=83253].

    • Search Google Scholar
    • Export Citation
  • Gleckler, P. J., , K. E. Taylor, , and C. Doutriaux, 2008: Performance metrics for climate models. J. Geophys. Res., 113 , D06104. doi:10.1029/2007JD008972.

    • Search Google Scholar
    • Export Citation
  • Haylock, M. R., , N. Hofstra, , A. M. G. Klein Tank, , E. J. Klok, , P. D. Jones, , and M. New, 2008: A European daily high-resolution gridded dataset of surface temperature and precipitation for 1950–2006. J. Geophys. Res., 113 , D20119. doi:10.1029/2008JD010201.

    • Search Google Scholar
    • Export Citation
  • Kållberg, P., 1997: Aspects of the re-analysed climate. ECMWF ERA-15 Project Rep. Series 2, European Centre for Medium-Range Weather Forecasts, 89 pp. [Available online at http://www.ecmwf.int/publications/library/do/references/show?id=83254].

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

  • Kanamitsu, M., , W. Ebisuzaki, , J. Woollen, , S-K. Yang, , J. J. Hnilo, , M. Fiorino, , and G. L. Potter, 2002: NCEP–DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83 , 16311643.

    • Search Google Scholar
    • Export Citation
  • 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
  • Koster, R. D., and Coauthors, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305 , 11381140.

  • Ma, L., , T. Zhang, , Q. Li, , O. W. Frauenfeld, , and D. Qin, 2008: Evaluation of ERA-40, NCEP-1, and NCEP-2 reanalysis air temperatures with ground-based measurements in China. J. Geophys. Res., 113 , D15115. doi:10.1029/2007JD009549.

    • Search Google Scholar
    • Export Citation
  • Onogi, K., and Coauthors, 2007: The JRA-25 Reanalysis. J. Meteor. Soc. Japan, 85 , 369432.

  • Perkins, S. E., , A. J. Pitman, , N. J. Holbrook, , and J. McAneney, 2007: Evaluation of the AR4 climate models’ simulated daily maximum temperature, minimum temperature, and precipitation over Australia using probability density functions. J. Climate, 20 , 43564376.

    • Search Google Scholar
    • Export Citation
  • Pitman, A. J., 2003: The evolution of, and revolution in, land surface schemes designed for climate models. Int. J. Climatol., 23 , 479510.

    • Search Google Scholar
    • Export Citation
  • Schubert, S., and Coauthors, 1995: A multi-year assimilation with the GEOS-1 system: Overview and results. NASA Tech. Memo. 104606, 182 pp. [Available from NASA Center for Aerospace Information, 800 Elkridge Landing Rd., Linthicum Heights, MD 21090].

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., , and J. G. Olson, 1988: An evaluation and intercomparison of global analyses from NMC and ECMWF. Bull. Amer. Meteor. Soc., 69 , 10471057.

    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012. doi:10.1256/qj.04.176.

  • Vincent, L. A., , X. Zhang, , B. R. Bonsal, , and W. D. Hogg, 2002: Homogenization of daily temperatures over Canada. J. Climate, 15 , 13221334.

    • Search Google Scholar
    • Export Citation
  • Williams Jr., C. N., , M. J. Menne, , R. S. Vose, , and D. R. Easterling, cited. 2006: United States Historical Climatology Network Daily Temperature, Precipitation, and Snow Data. ORNL/CDIAC-118, NDP-070, Carbon Dioxide Information Analysis Center, Oak Ridge National Laboratory, U.S. Department of Energy. [Available online at http://cdiac.ornl.gov/epubs/ndp/ushcn/usa.html].

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

For the 2-m air temperature in °C (a) mean TMAX difference (NCEP-2 minus ERA-40), (b) difference (NCEP-2 minus ERA-40), (c) skill score for NCEP-2 compared to ERA-40, (d) mean TMAX difference (JRA-25 minus ERA-40), (e) difference (JRA-25 minus ERA-40), (f) skill score for JRA-25 compared to ERA-40, (g) mean TMAX difference (JRA-25 minus NCEP-2), (h) difference (JRA-25 minus NCEP-2), and (i) skill score for JRA-25 compared to NCEP-2.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 2.
Fig. 2.

As in Fig. 1, but for the 1000-hPa temperature.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 3.
Fig. 3.

For the 2-m air temperature (°C) (a) mean TMIN difference (NCEP-2 minus ERA-40), (b) difference (NCEP-2 minus ERA-40), (c) skill score for NCEP-2 compared to ERA-40, (d) mean TMIN difference (JRA-25 minus ERA-40), (e) difference (JRA-25 minus ERA-40), (f) skill score for JRA-25 compared to ERA-40, (g) mean TMIN difference (JRA-25 minus NCEP-2), (h) difference (JRA-25 minus NCEP-2), and (i) skill score for JRA-25 compared to NCEP-2.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 4.
Fig. 4.

As in Fig. 3, but for the 1000-hPa TMIN.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 5.
Fig. 5.

Probability density functions for 2-m TMAX for regions described in Table 1: observed (thick solid line), ERA-40 (thin solid line), NCEP-2 (dotted line), and JRA-25 (dashed line).

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 6.
Fig. 6.

Skill score for each reanalysis compared to the observed for 2-m TMAX for each region described in Table 1: ERA-40 (solid bar), NCEP-2 (dotted bar), and JRA-25 (dashed bar).

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 7.
Fig. 7.

Difference from observed in for each region described in Table 1: ERA-40 (solid bar), NCEP-2 (dotted bar), and JRA-25 (dashed bar).

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 8.
Fig. 8.

As in Fig. 5, but for TMIN.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 9.
Fig. 9.

As in Fig. 6, but for TMIN.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 10.
Fig. 10.

As in Fig. 7, but for TMIN.

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Fig. 11.
Fig. 11.

Latitudinal-average differences (°C) in the PDF overlap (averaged over the three reanalyses) for (a) TMAX and (b) TMIN in a latitude band from 40° to 70°N and (c) TMAX and (d) TMIN in a latitude band from 15°N to 15°S. The heavy (light) solid line is for the 1000-hPa air temperature (2-m temperature). Note difference in scale between (a),(b) and (c),(d).

Citation: Journal of Climate 22, 17; 10.1175/2009JCLI2799.1

Table 1.

Latitude and longitude of the six regions discussed in the text.

Table 1.
Table 2.

Overlap of the PDFs for each region between the observed and reanalysis for TMAX and TMIN for 2-m air temperature.

Table 2.
Table 3.

Actual return values from the observed and each reanalysis for and for 2 m.

Table 3.
Save