• Arkin, P. A., V. E. Kousky, J. E. Janowiak, and E. A. O’Lenic, 1986:Atlas of the Tropical and Subtropical Circulation Derived from National Meteorological Center Operational Analyses. NOAA Atlas No. 7.

  • Barnett, T. P., and Coauthors, 1999: Detection and attribution of recent climate change: A status report. Bull. Amer. Meteor. Soc.,80, 2631–2659.

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

  • ——, C. F. Ropelewski, and N. C. Grody, 1995: Comparison of tropospheric temperature derived from microwave sounding unit and the National Meteorological Center analysis. J. Climate,8, 668–681.

  • Christy, J. R., R. W. Spencer, and R. T. McNider, 1995: Reducing noise in the MSU daily lower-tropospheric global temperature dataset. J. Climate,8, 888–896.

  • ——, ——, and E. A. Lobl, 1998: Analysis of the merging procedure for the MSU daily temperature time series. J. Climate,11, 2016– 2041.

  • Gibson, J. K., P. Kallberg, S. Uppala, A. Hernandez, A. Nomura, and E. Serrano, 1997: ECMWF Re-Analysis, Project Report Series 1: ERA Description, ECMWF, 72 pp.

  • Hasselmann, K., 1993: Optimal fingerprints for the detection of time-dependent climate change. J. Climate,6, 1957–1971.

  • Hegerl, G. C., H. von Storch, K. Hasselmann, B. D. Santer, U. Cubasch, and P. D. Jones, 1996: Detecting greenhouse-gas-induced climate change with an optimal fingerprint method. J. Climate,9, 2281–2306.

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

  • Hurrell, J. W., and K. E. Trenberth, 1996: Satellite versus surface estimates of air temperature since 1979. J. Climate,9, 2222– 2232.

  • ——, and ——, 1997: Spurious trends in satellite MSU temperatures from merging different satellite records. Nature,386, 164–167.

  • Jones, P. D., 1994: Recent warming in global temperature series. Geophys. Res. Lett.,21, 1149–1152.

  • ——, T. M. L. Wigley, and G. Farmer, 1991: Marine and land temperature data sets: A comparison and look at recent trends. Greenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier Science, 593–602.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Karl, T. R., and Coauthors, 1993: A new perspective on recent global warming: Asymmetric trends of daily maximum and minimum temperature. Bull. Amer. Meteor. Soc.,74, 1007–1023.

  • Parker, D. E., C. K. Folland, and M. Jackson, 1995: Marine surface temperature: Observed variations and data requirements. Climatic Change,31, 429–470.

  • Santer, B. D., T. M. L. Wigley, P. D. Jones, and M. D. Schlesinger, 1991: Multivariate methods for the detection of greenhouse-gas-induced climatic change. Greeenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier Science, 511–536.

  • ——, J. J. Hnilo, and P. D. Jones, 1993: Correlation methods in detection studies. Climate Dyn.,8, 265–276.

  • ——, ——, T. M. L. Wigley, J. S. Boyle, C. Doutriaux, M. Fiorino, D. E. Parker, and K. E. Taylor, 1999: Uncertainties in observationally based estimates of temperature change in the free atmosphere. J. Geophys. Res.,104, 6305–6333.

  • Schubert, S. D., J. Pfaendtner, and R. Rood, 1993: An assimilated data set for earth science applications. Bull. Amer. Meteor. Soc.,74, 2331–2342.

  • Trenberth, K. E., and J. W. Hurrell, 1994: Decadal atmosphere&ndash∘ean variations in the Pacific. Climate Dyn.,9, 303–319.

  • WCRP, 1997: Proc. First WCRP Int. Conf. on Reanalyses. WCRP-104, WMO/TD-876, Silver Spring, MD, WMO, 461 pp.

  • View in gallery

    Time series of the tropospheric mean temperature (1979–97) as estimated by satellite (MSU2) and three reanalysis datasets, NCEP, ERA, and DAO. (See the text for detailed descriptions of these data.) (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

  • View in gallery

    Same as Fig. 1, but for temperature anomalies.

  • View in gallery

    Time series of the tropospheric temperature anomaly differences between the NCEP and MSU2 (solid line) and ERA and MSU2 (dotted line). (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

  • View in gallery

    Time series of the pattern correlations between the MSU2 mean tropospheric temperature estimates and estimates from each of the reanalyses. (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

  • View in gallery

    Same as in Fig. 4, but for pattern correlations between anomalies.

  • View in gallery

    Spatial pattern of the monthly temporal correlations between the MSU2 and (top) NCEP, (middle) ERA, and (bottom) DAO reanalyses for the 1980–93 period. All correlations are positive, and contours are drawn at 0.4, 0.6, 0.7, 0.8, 0.9, and 0.95. The shading is as follows: 0.3–0.6 dark gray, 0.6–0.9 medium gray, >0.9 light gray. Regions with correlations less than 0.3 and areas with no data have no shading (white).

  • View in gallery

    Same as in Fig. 6 except the panels depict anomaly pattern correlations.

  • View in gallery

    Spatial pattern of the linear mean tropospheric temperature trend as estimated by the MSU2 [°C (decade)−1] for the 1980–90 period. Negative (cooling) trends are darkly shaded and positive (warming) trends are lightly shaded. Contour lines are shown at every 0.3, with additional contours at +0.1 and −0.1.

  • View in gallery

    Same as in Fig. 8, but for the reanalyses: (top) NCEP, (middle) ERA, and (bottom) DAO.

  • View in gallery

    Spatial pattern of the differences in linear trend of mean tropospheric temperature, respectively, between MSU2 and the three different reanalyses: (top) NCEP, (middle) ERA, and (bottom) DAO for the 1980–90 period. Negative (cooling) trends are darkly shaded and positive (warming) trends are lightly shaded. Contour lines are shown at every 0.3, with additional contours at +0.1 and −0.1.

  • View in gallery

    Same as in Fig. 10 except the period for the trend computation is 1980–93.

  • View in gallery

    The 2-m temperature anomaly as estimated by the NCEP reanalysis for (top) Jan 1958, (middle) Jan 1979, and (bottom) Jan 1996. Contour interval is 1°C. Negative values are dashed.

  • View in gallery

    The same as in Fig. 12, except the data are surface observations from Jones et al. (1991). Data-void regions are shown shaded gray.

  • View in gallery

    Time series of surface temperature anomaly as estimated by the observed temperature from Jones et al. (1991) and the NCEP reanalysis for the 1958–96 period. (top) The globe, equatorward of 80°. (bottom) The Tropics equatorward of 20°.

  • View in gallery

    (top) Time series of the seasonal pattern correlations between the NCEP reanalysis and the surface observations from Jones et al. (1991). (middle) Time series of the root-mean-square differences (rmsd) between seasonal estimates of surface temperature anomalies estimated by the NCEP reanalysis and surface observations from Jones et al. (1991). (bottom) Time series of the number of 5° lat × 5° long boxes with data by season in the Jones et al. (1991) dataset from 1958 to 1995.

  • View in gallery

    Spatial patterns of temporal correlations between the observed surface datasets of Jones et al. (1991) and the NCEP reanalysis 2-m temperature for (top) Dec–Feb, (second panel) Jun–Aug, (third panel) Mar–May, and (bottom) Sep–Nov.

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

Reanalyses-Based Tropospheric Temperature Estimates: Uncertainties in the Context of Global Climate Change Detection

View More View Less
  • 1 Climate Prediction Center, NCEP/NWS/NOAA, Camp Springs, Maryland
  • 2 International Research Institute for Climate Prediction, Lamont–Doherty Earth Observatory of Columbia University, Palisades, New York
© Get Permissions
Full access

Abstract

Uncertainties in estimates of tropospheric mean temperature were investigated in the context of climate change detection through comparisons of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) 40-yr reanalysis (1958–97), the National Aeronautics and Space Administration Data Assimilation Office (NASA/DAO) 14-yr reanalysis (1980–93), the European Centre for Medium-Range Weather Forecasts Reanalysis Project (ERA) 15-yr reanalysis (1979–94), and the satellite microwave sounding unit channel 2 (MSU Ch2) (1979–97) temperature data. The maximum overlap period for comparison among these datasets is the 14 full years January 1980 to December 1993. This study documents similar shifts in the relative bias between the MSU Ch2 and the ERA and the NCEP–NCAR reanalyses in the 1991–97 period suggesting changes in the satellite analysis. However, the intercomparisons were not able to rule out the changes in the reanalysis systems and/or the input data on which the reanalyses are based as prime factors for the changes in the relative bias between the MSU and ERA and NCEP–NCAR reanalyses.

These temporal changes in the relative bias among the reanalyses suggest their limitations for global change studies. Nonetheless, the analysis also shows that the pattern correlations (r) between the MSU Ch2 monthly mean fields and each of the reanalyses are very high, r > 0.96, and remain relatively high for the anomaly fields, r > 0.8, generally >0.9. This result suggests that reanalysis may be used for comparisons to numerical model–generated forecast fields (from GCM simulation runs) and in conjunction with “fingerprint” techniques to identify climate change.

In comparisons of the simple linear trends present in each dataset for the 1980–90 period, each of the reanalyses had spatial patterns similar to MSU Ch2 except that the NCEP–NCAR reanalysis showed smaller “positive” (warming) trends in comparison with the MSU while the ERA reanalysis showed larger positive trends. The NASA/DAO reanalysis showed a mixed pattern. Many regions of the globe are identified that showed consistent warming/cooling patterns among the major reanalyses and MSU, even though there were disagreements in the exact magnitude among the analyses. The spatial patterns of linear trends changed, however, with the addition of three years of data to extend the trend analysis to the 1980–93 period. This result suggests that such simple linear trend analyses are very sensitive to the temporal span in these relatively short datasets and thus are of limited usefulness in the context of climate change detection except, however, when the signal is large and shows consistency among all datasets.

The long record (1958–96) of seasonal mean 2-m temperature anomalies from NCEP–NCAR reanalysis is well correlated with gridded analyses of station-based observed surface temperature, with correlations between 0.65 and 0.85. It is argued that these correlations might suggest an upper limit to the magnitudes of the pattern correlations that might be obtained by correlating observed surface temperature analyses with those from multiyear GCM simulation runs made in the context of fingerprint climate change detection.

Corresponding author address: Dr. Muthuvel Chelliah, Climate Prediction Center, NCEP/NWS/NOAA, 5200 Auth Road, Camp Springs, MD 20746.

Email: muthuvel.chelliah@ncep.noaa.gov

Abstract

Uncertainties in estimates of tropospheric mean temperature were investigated in the context of climate change detection through comparisons of the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) 40-yr reanalysis (1958–97), the National Aeronautics and Space Administration Data Assimilation Office (NASA/DAO) 14-yr reanalysis (1980–93), the European Centre for Medium-Range Weather Forecasts Reanalysis Project (ERA) 15-yr reanalysis (1979–94), and the satellite microwave sounding unit channel 2 (MSU Ch2) (1979–97) temperature data. The maximum overlap period for comparison among these datasets is the 14 full years January 1980 to December 1993. This study documents similar shifts in the relative bias between the MSU Ch2 and the ERA and the NCEP–NCAR reanalyses in the 1991–97 period suggesting changes in the satellite analysis. However, the intercomparisons were not able to rule out the changes in the reanalysis systems and/or the input data on which the reanalyses are based as prime factors for the changes in the relative bias between the MSU and ERA and NCEP–NCAR reanalyses.

These temporal changes in the relative bias among the reanalyses suggest their limitations for global change studies. Nonetheless, the analysis also shows that the pattern correlations (r) between the MSU Ch2 monthly mean fields and each of the reanalyses are very high, r > 0.96, and remain relatively high for the anomaly fields, r > 0.8, generally >0.9. This result suggests that reanalysis may be used for comparisons to numerical model–generated forecast fields (from GCM simulation runs) and in conjunction with “fingerprint” techniques to identify climate change.

In comparisons of the simple linear trends present in each dataset for the 1980–90 period, each of the reanalyses had spatial patterns similar to MSU Ch2 except that the NCEP–NCAR reanalysis showed smaller “positive” (warming) trends in comparison with the MSU while the ERA reanalysis showed larger positive trends. The NASA/DAO reanalysis showed a mixed pattern. Many regions of the globe are identified that showed consistent warming/cooling patterns among the major reanalyses and MSU, even though there were disagreements in the exact magnitude among the analyses. The spatial patterns of linear trends changed, however, with the addition of three years of data to extend the trend analysis to the 1980–93 period. This result suggests that such simple linear trend analyses are very sensitive to the temporal span in these relatively short datasets and thus are of limited usefulness in the context of climate change detection except, however, when the signal is large and shows consistency among all datasets.

The long record (1958–96) of seasonal mean 2-m temperature anomalies from NCEP–NCAR reanalysis is well correlated with gridded analyses of station-based observed surface temperature, with correlations between 0.65 and 0.85. It is argued that these correlations might suggest an upper limit to the magnitudes of the pattern correlations that might be obtained by correlating observed surface temperature analyses with those from multiyear GCM simulation runs made in the context of fingerprint climate change detection.

Corresponding author address: Dr. Muthuvel Chelliah, Climate Prediction Center, NCEP/NWS/NOAA, 5200 Auth Road, Camp Springs, MD 20746.

Email: muthuvel.chelliah@ncep.noaa.gov

1. Introduction

The detection of “significant” global climate trends and attempts to develop objective methods for attribution of these trends to specific causes has been one of the most difficult, challenging, and contentious areas of climate research of this century (Houghton et al. 1996;Barnett et al. 1999). The conceptual and technical difficulties associated with developing, testing, and applying climate change detection and attribution methods are compounded by a scarcity of appropriate data to study the problem.

In the long run, virtually all of the major observational change/attribution studies have relied on the land/sea surface datasets (e.g., Hegerl et al. 1996; Santer et al. 1991, 1993; Barnett et al. 1999) or some variant of mean temperature (Karl et al. 1993). Although these global and large-scale regional datasets reveal much about global change, they are limited in their use for validation of greenhouse gas (GHG) numerical model simulations. These models generally can provide a full four-dimensional description of the atmosphere. However, it has been extremely difficult to assess the validity of many of the model fields, their vertical structure, and time–space variability because of the lack of comparable observational-based datasets.

Climate analyses based on operational model-assimilated data have been used to monitor and study monthly and seasonal variability since the early 1980s (e.g., Arkin et al. 1986). Although these analyses provided many new insights about the climate system, their utility was hampered by the changes in the analysis systems mandated by operationally driven improvements to numerical weather prediction models. This situation has been addressed by three major projects that produced relatively long records of reanalyzed fields based on “frozen” data assimilation systems.

The National Aeronautics and Space Administration Data Assimilation Office (NASA/DAO) reanalysis (Schubert et al. 1993) produced a 14-yr reanalyzed dataset (1980–93) based on a newly developed semi-Lagrangian gridpoint model. The European Centre for Medium-Range Weather Forecasts (ECMWF; Gibson et al. 1997) produced a 14-yr reanalysis (1979–93) with a T106 spectral model. The National Centers for Environmental Prediction (NCEP) in collaboration with the National Center for Atmospheric Research (Kalnay et al. 1996), produced the most ambitious of these reanalyses (1958 to the present) using a T62 model. This reanalysis will eventually extend back to 1948 and is the only system at present that is being continuously updated.

These reanalyses have already lived up to their great promise with respect to facilitating studies and monitoring of seasonal to interannual climate variability, (see, e.g., WCRP 1997); however, there has been very little work done to document and quantify the uncertainties of these reanalyses in the context of global change studies. One of the indices often used in global change studies is the satellite-derived mean tropospheric temperature (Christy et al. 1995). To obtain quantitative estimates of the uncertainties in the mean tropospheric temperature, the three reanalyses are intercompared and compared with the satellite-derived estimates of the same. Additional intercomparisons are also performed with the observed surface temperature fields (e.g., Jones et al. 1991; Jones 1994), and the NCEP reanalysis to gain an estimate of the uncertainty in these reanalyzed fields. The intercomparisons for NCEP reanalysis and observed surface temperature provide some guidance for the use of reanalyzed fields as validation datasets in GHG-driven climate simulations of surface temperature.

2. Background

This study is an extension of two previous studies (Basist et al. 1995; Basist and Chelliah 1997). Basist et al. (1995) compared tropospheric temperature analyses (1979–90) from the operational global data assimilation system (GDAS) at the National Meteorological Center (now NCEP), known as the climate diagnostics database (CDDB), with temperature measurements from the microwave sounding unit (MSU). Basist and Chelliah (1997) evaluated the reanalyzed temperatures from NCEP against the operational GDAS temperatures with the independent satellite-based MSU. Their study showed that the NCEP reanalysis compared more closely to the MSU than to the CDDB in both both temporal and spatial characteristics. Since then, the NCEP reanalyzed fields have replaced the operational GDAS-based CDDB as the basis for climate monitoring at the Climate Prediction Center.

Basist and Chelliah (1997) also found unexpected (because the model is frozen) changes in the relative bias between NCEP reanalysis estimates of the mean tropospheric temperatures and the MSU. Their study showed that the tropospheric-averaged global mean NCEP temperature anomalies were becoming gradually cooler than MSU by about 0.2°C during the period 1991–95. Their EOF analysis suggested that the relative bias change may have been related to the changes in the National Environmental Satellite, Data, and Information Service (NESDIS’s) satellite temperature retrievals used in NCEP reanalysis.

In the current study we investigate the temporal and spatial covariability of the tropospheric temperatures from three [NCEP, ECMWF reanalysis (ERA), and DAO] major atmospheric reanalyses compared along with the satellite-based MSU temperature. These analyses are done to determine the uncertainties in estimates of tropospheric mean temperature and to examine their utility for climate monitoring and detection. In a more recent study, Santer et al. (1999) also performed a comprehensive comparison between the NCEP and ECMWF reanalyses and radiosonde observations as well as MSU. Their study supplements and is in agreement with the analyses presented here, even though our paper covers a different ground and is in the context of climate change detection.

3. Data

All three reanalyses assimilated satellite temperatures over the global oceans. Both NCEP (Kalnay et al. 1996) and DAO (Schubert et al. 1993) reanalyses employ different versions of the NOAA/NESDIS’s Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) temperature retrievals in the data assimilation. ERA used the NESDIS TOVS cloud-cleared radiances through their own 1D-Var temperature and humidity retrieval system over much of the globe, except above 100 hPa poleward of 30° N/S latitude, where they use NESDIS’s retrievals (Gibson et al. 1997). The differences in the satellite data assimilation systems can influence the temperature field in the three reanalyses.

In this study, temperature data from NCEP’s reanalysis data were available from January 1958 through June 1998. However, ERA’s monthly mean temperature were available only from January 1979 through February 1994, and DAO’s reanalyses from January 1980 through November 1994. Observed Channel 2 MSU (version C) data were available from January 1979 to January 1998. Equivalent MSU Channel 2 temperatures were computed for each reanalysis by weighting the vertical profiles of reanalysis tropospheric temperatures weighted by the MSU weighting function, following Basist and Chelliah (1997). MSU Channel 2 has its largest weights in the troposphere (between 500 and 700 hPa), with relatively little contribution from the surface and stratosphere. For the remainder of the paper, the following convention is used to refer to the various tropospheric temperature estimates: 1) MSU2-Observed Channel 2 Temperature from Christy and 2) NCEP2, ERA2, and DAO2, respectively, for the simulated MSU Channel 2 equivalent temperatures from NCEP, ERA, and DAO reanalyses.

Although we computed the NCEP2, ERA2 simulated Ch 2 temperatures data at all grid points over both land and ocean, the simulated Ch2 temperatures for DAO2 are not computed over much of the land areas and/or in regions where the interpolated 1000-mb (and possibly higher levels) reanalysis temperature values are not available, when the surface pressure is less than 1000 mb. It is a requisite in our calculations of the simulated Ch 2 temperatures from different reanalyses that at each grid point all pressure level (analyzed or interpolated) temperature data be available. This difference in spatial coverage (which can vary from one year to the other) contributes to differences between the DAO2 and estimates of tropospheric temperature anomalies from the other analyses, that is, spatial means of DAO2 are computed using a mask (of missing values) but there is near-complete global spatial coverage in the other three datasets (except MSU).

The observational dataset used for comparison with the NCEP reanalysis surface temperature estimates in the later part of the paper consists of merged near-surface air temperature anomalies over land with in situ, ship opportunity and buoy, sea surface temperature anomalies over marine areas. This monthly dataset has been described in detail by Jones et al. (1991) (and used by the Intergovernmental Panel on Climate Change), and was available to us on a 5° lat × 5° long grid for the period from 1951 to 1996.

4. Analysis

a. Time series of spatially averaged tropospheric temperatures

Spatial averages of MSU2, NCEP2, ERA2, and DAO2 are compared for a near-global domain (top, Fig. 1) and the Tropics (20°N–20°S) (bottom, Fig. 1). Although the nominal period of intercomparison is 1979– 97 there are different data coverage periods for each of the four datasets. Note, in particular that the NCEP2 data used in this analysis start in January 1958 while each of the other datasets start in 1979. The time series of global tropospheric mean temperature averages in each dataset exhibits an annual cycle dominated by the Northern Hemisphere, that is, minima in northern winter, maxima in northern summer. The time series for the Tropics, Fig. 1 (bottom), also show evidence of relatively large interannual variability.

The reanalysis-based simulated channel 2 temperatures are warmer than the MSU2 by about 2°C (NCEP2 and ERA2) to 3°C (DAO2) in the global means and by about 2°C for the tropical mean values. Basist and Chelliah (1997) note a similar mean 2°–3°C bias between the MSU and NCEP2 reanalysis estimated temperature. The relative change in the bias between the NCEP2 and MSU2 in the early 1990s noted in Basist and Chelliah (1997) is matched by similar changes in the relative bias between the ERA2 and MSU2. The analysis presented here also shows that the global means of ERA2 and NCEP2 show less of a relative bias from mid-1991 onward, that is, their estimates of global mean tropospheric temperature are more similar after 1991 than before.

The time series of each global tropospheric temperature anomaly (from a common 1980–83 base period) curve clearly exhibits interannual variability associated with the major El Niño–Southern Oscillation (ENSO) episodes in the period (Fig. 2). Both the ERA2 and NCEP2 anomalies are well within 0.2°–0.3°C of the MSU2 anomalies until late 1993. After 1993, the NCEP2 anomalies become generally more negative than the MSU2 anomalies in both Tropics and global means and remain at these levels until late in 1997.

The global tropospheric temperature anomalies generally have a range of 0.5°C on either side of zero. However, the DAO2 global anomaly time series (Fig. 2, top) shows large, short-lived, anomalous extremes not present in either of the two other reanalyses nor in the MSU2 estimates. The DAO2 does not show this behavior in the tropical anomaly time series (Fig. 2, bottom) following the other time series quite closely. The tropical tropospheric temperature anomaly time series reflects swings in temperature associated with ENSO episodes even more clearly than the global series.

The time series of NCEP2 − MSU2 and ERA2 − MSU2 differences (biases) illustrate (Fig. 3) more directly some of the temporal behavior in relative bias among the two reanalyses with respect to MSU2. In general, the relative biases are quite small, on the order of 0.1°–0.2°C for most of the time series until about 1990. However, the magnitude of the NCEP2 bias (NCEP2 − MSU2), begins to change in the early 1990s and exhibits greater interannual variability. To some extent, the global ERA2 anomaly bias (ERA2 − MSU2) time series follows the same behavior in the early 1990s until the end of the dataset in 1994. The time series of tropical anomaly bias roughly parallels the global time series (Fig. 3, bottom), indicating the global mean is dominated by the Tropics. Basist and Chelliah (1997) explore some of the reasons for this behavior in the early 1990s in the NCEP’s temperature analysis. Incidentally, ERA2 is relatively colder than NCEP2 before 1989/90 and relatively warmer afterward (Santer et al. 1999) in both the global and tropical time series, which gives an overall warmer trend for ERA2 (cf. section 2d).

b. Time series of spatial correlations

At the very least, it is expected that the reanalyses replicate the MSU2-observed annual cycle. The time series of global (80°N–80°S) pattern correlations between the full fields of MSU2 and the corresponding simulated quantities from the various reanalyses (Fig. 4a) shows very large values, in excess of 0.96, in every month. The DAO2 shows even larger pattern correlations (>0.995) but note that these are based on a relatively limited domain of ocean-only points. There appears to be a slight annual cycle in the time series of DAO2, NCEP2, and ERA2 pattern correlations with MSU2. The ERA2 pattern correlation also exhibits a tendency for a semiannual cycle but the magnitudes of this semiannual and the annual correlation variations are extremely small, about 0.02 for NCEP2 and less for ERA2. In the Tropics (Fig. 4b), the pattern correlations between the MSU2 and reanalyzed temperatures are relatively lower as compared with those for the larger global region (80°N–80°S, Fig. 4a), but are still relatively very high, generally in excess of 0.9 and as high as 0.97.

Pattern correlations between the MSU2 and reanalyses for 30°–60°N and 30°–60°S (not shown here) have larger values in the NCEP2 as compared with ERA2 in the Northern Hemisphere and conversely for the Southern Hemisphere. Time series of anomaly (from 1980 to 1993 base period) pattern correlations show (Fig. 5) values that are slightly lower than full field correlations, but these correlations are still quite large, ranging between 0.82 and 0.98 for the globe (wider range for the Tropics), thus indicating a good degree of agreement even among anomaly fields in the different reanalyses and with MSU. The time series of anomaly pattern correlations show a number of coincident periods with lower correlations between both ERA2 and NCEP2 with MSU2. Because ERA and NCEP reanalyses have quite different data assimilation schemes, this suggests the difference could be due to problems with the MSU processing or a lack of satellite data in each reanalysis, for example, in 1985 and 1991. Seasonal mean root-mean-square differences (not shown) between MSU2 and reanalyses’ temperature anomalies do not exceed 0.25°C for both NCEP2 and ERA2.

c. Spatial maps of temporal correlations

Temporal correlations were computed at each grid point between each of the reanalyses’ simulated Channel 2 tropospheric temperatures and the observed MSU2 estimates for the full fields (Fig. 6), as well as the anomaly fields (Fig. 7). Temporal correlations between the MSU2 and the full reanalysis fields (Fig. 6) are in excess of 0.95 poleward of approximately 15°–20° latitudes. In the Tropics, correlations are slightly lower (0.8–0.95) with pockets of even lower values over the western equatorial Pacific Ocean. In general the lowest correlations appear to coincide with regions generally associated with deep tropical convection (i.e., the mean position of the Intertropical Convergence Zone, the Amazon, equatorial Africa, and the western Pacific warm pool). This suggests some inconsistencies between each of the reanalyses and the MSU2 mean tropospheric temperature estimates in areas of deep convection.

Spatial patterns of the different reanalyses’ temperature anomaly correlations with MSU2 (Fig. 7) exhibit generally similar features as the full field correlations (Fig. 6) between them for each of the reanalyses. Note that for the DAO2 much of land areas in either hemisphere and regions south of 60°S are blank in the bottom panel, since DAO2 has no temperature estimates for these regions. The ERA2 correlations generally show no values less than 0.8 for either full or anomaly field while both NCEP2 and DAO2 show some small areas with slightly lower values near the western tropical Pacific.

Spatial maps of root-mean-square (rms) differences between MSU2 and NCEP2 or ERA2 time series (not shown) are under 0.3°C over much of the Tropics with slightly larger values (0.3°–0.7°C) in the higher latitudes with much larger differences exceeding 2°–3°C over Antarctica for both NCEP2 and ERA2. Fields of anomaly RMSD fields (not shown) depict similar features as full RMSD fields.

d. Linear tropospheric temperature trends

The most common and widely cited indexes of global temperature change are area mean surface temperature, mean tropospheric temperature, and their linear trends, most often expressed in degrees per decade. Although such indices provide a useful way to compress a lot of data into one global indicator they do not provide any information about the patterns of temperature trends. These patterns show that linear trends may vary from place to place both in magnitude and in sign. The regional structure (patterns) of temperature trends are of great interest for climate change impact studies. The patterns of global trends have also emerged as important for use in a powerful family of climate change detection techniques often referred to as fingerprint techniques (e.g., Hasselman 1993).

In this section, the spatial patterns of linear trend in the reanalyses estimates of mean tropospheric temperature are compared to those observed in the MSU2. Because of the different temporal spans of the reanalysis datasets and the cautions raised by Hurrell and Trenberth (1997) concerning possible inhomogeniety in the MSU data in the early 1990s, comparisons are initially limited to the 1980–90 time period. Initial computations show unrealistic and spatially incoherent linear trends in the reanalyses at very high latitudes and thus the comparisons are limited to the region 60°S–70°N.

The spatial map of the linear trend in the annual mean MSU2 temperature for 1980–90 (Fig. 8), shows a complex spatial structure, that is, in some ways reminiscent of variations in height patterns associated with extremes in ENSO. This suggests that interannual variability may contribute to the observed spatial patterns of tropospheric temperature trends. Most of the Eastern Hemisphere and global mid- to high latitudes show a general positive trend over the 1980–90 period while the tropical and subtropical Pacific, east of the date line, shows a negative trend. The linear trends for the NCEP2, ERA2, and DAO2 (Fig. 9), show the same general patterns as the MSU2 over this period. However, the magnitude of the North Pacific trend pattern in the ERA2 and DAO2 is somewhat smaller than appears in the MSU2 and NCEP2 trend estimates. Again note, there are no computations over much of the land and in the deep southern oceans for the DAO2.

In summary, there is a general agreement over much of the region both in the spatial details and in the sign and magnitude of the heating/cooling among the three reanalyses as compared with MSU2. There is a general warming, during this period (1980–90), over much of the global region analyzed here, particularly with strong values in excess of 0.5°C/decade over central Europe, along a zonal (40°–55°N) belt between 90°E and 120°W. As noted above, between 30°N and 30°S, there has been a general cooling trend over much of the Pacific Ocean with a pair of east–west oriented cooling centers equidistant from the equator centered near 150°W in MSU2, NCEP2, and ERA2. This pattern closely resembles the upper-level anticyclonic (cyclonic) height pattern associated with tropospheric heating (cooling) usually seen during Northern Hemisphere winter in the mature stages of the ENSO phenomenon. The cooling in the tropical Pacific probably is likely a reflection of the decrease in the tropospheric temperatures associated with the major 1988/89 cold ENSO episode in the later part of the period in comparison with the relatively high temperatures associated with the 1982/83 ENSO episode in the earlier part of the period.

The spatial patterns in the relative differences between the MSU2 linear trends and each of the reanalyses trends (Fig. 10), shows that the NCEP2 is exhibiting relatively more cooling trend than MSU2 over much of the Tropics and subtropics and ERA2 is exhibiting relatively more warming trend than MSU2, while DAO2 (with respect to MSU2) has mixed trend difference pattern.

A note of caution should be exercised in interpretation of these trend patterns computed over any limited period. We computed trends similar to Figs. 8 and 9 shown above, but over a slightly extended (by three more years) period to 1980–93 (patterns not shown). The reasonable agreement that existed (1980–90) among MSU2 and the various reanalyses has now weakened for the extended period, and the suggestion of a link to ENSO-related variability noted earlier in the 1980–90 trend patterns is now missing. We show in Fig. 11 the trend differences of the reanalyses’ temperatures with respect to MSU2 for the extended 14-yr period. Although the major features of the linear trend difference generally remain the same as for the earlier period (Fig. 10), there are local differences: for example, the eastern Pacific in the NCEP2, New Zealand in the ERA2.

These results suggest that when the signal is strong, and especially when different datasets agree, such as the warming/cooling trend patterns from different reanalyses, then these reanalysis datasets may be good tools for monitoring regional patterns of temperature trends around the globe, even though the precise magnitude of such changes in those regions may vary/disagree among the different analyses. The utility of these reanalyses’ trends (at least temperature) may be in question when the signal is weak, or when the analysis method is susceptible to slight changes in the analysis period.

e. Comparison between observed surface temperature and NCEP reanalysis estimates

Intercomparisons of the tropospheric mean temperature are limited to the last two decades because the MSU2 data and ERA2 and DAO2 reanalyses are not available for earlier periods. However, the NCEP reanalysis is currently available from 1958 forward and will be extended back to the late 1940s. In this section, NCEP reanalysis estimates of surface (2-m) temperature are compared with observations of surface temperature data (Jones et al. 1991).

This analysis provides some measure of uncertainty in the NCEP surface temperature data for global change studies. It also allows an indication of the impact of the changing data sources (e.g., the absence of satellite data before the early 1970s) on the NCEP surface temperature reanalysis.

The spatial coverage of the observed surface temperature dataset varies from month to month with considerable gaps in coverage over Africa, South America, and regions north of 60°N and south of 45°S. Because the observational data are gridded to a much coarser 5° lat × 5° long grid, the NCEP reanalysis 2-m temperatures were averaged for this part of the study onto the observational coarse grid from its original Gaussian grid (approximately 2° lat × 2° long grid).

Spatial patterns of NCEP 2-m temperature anomaly and observed surface temperature anomaly, based on 1958–96 base period, for January are shown in Figs. 12 and 13, respectively, in three representative years (1958, 1979, and 1996). Despite some significant areas with no data in the observational dataset, there is close correspondence between the two datasets in the major anomaly features. It has to be noted that the surface temperature observations are not assimilated in the NCEP assimilation system, and it is a model-derived quantity in the NCEP reanalysis (Kalnay et al. 1996).

Time series of global and tropical (Fig. 14, top and bottom) temperature anomalies (from base period 1958– 96) show close correspondence for most major departures in both datasets. The averaged temperature anomalies over the globe and over the Tropics for the NCEP T 2-m time series are computed only based on those grid points for which the observations (Jones et al. 1991) were available. Both datasets clearly show evidence of a “climate shift” in the mid-1970s as was discussed by Trenberth and Hurrell (1994). Considering the large variation in the number and distribution of conventional and satellite data over the course of forty years, the agreement is remarkable on the interannual and low-frequency timescale, even in the earlier years before the “satellite era.” On the other hand, there is a relatively large shift in the bias between the NCEP and observational data beginning after 1991, from which time NCEP became relatively cooler by ∼0.1° to 0.2°C on a global average.

Time series of seasonal anomaly pattern correlations over the global domain (Fig. 15, top) between NCEP 2-m temperature and the observed temperature show that, except for the June–July–August (JJA) season before 1968, the anomaly pattern correlation is in the neighborhood of 0.7 for all seasons and for all years. The December–January–February (DJF) anomaly pattern correlations are generally larger than the other seasons with values greater than 0.75 over almost all years. The JJA seasonal anomaly pattern correlations, with values near about 0.65, are consistently lower than the other seasonal values. Not surprisingly, correlation values for the March–April–May and September–October–November transitional seasons generally fall between the extremum DJF and JJA values. The relatively lower pattern correlations before 1968 could not be ascribed to any particular cause of which we are aware.

The rms differences between NCEP reanalysis and observed (Jones et al. 1991) surface temperature data (Fig. 15, middle) generally range between 0.7° and 1.1°C for all seasons. The largest rms differences occur during the northern winter (DJF) season when the absolute magnitude of the anomalies is the largest. Conversely the lowest rms differences generally occur during the Northern Hemisphere summer (JJA) season when the absolute magnitude of the anomalies is generally the smallest.

The magnitudes of the rms differences and the anomaly correlations do not appear to be very sensitive to the number of surface observations (Fig. 15, bottom) during 1958–96 period. In addition, the introduction of satellite data during the late 1970s does not appear to alter any of the intercomparison statistics in any noticeable way.

Spatial maps of temporal correlations between the observations and the NCEP reanalysis’ 2-m temperature anomalies at all available grid boxes over the global domain for the traditional four calendar seasons are shown in Fig. 16. In general there is no dramatic change in the pattern of the correlations from one season to other. In all seasons, the correlations are generally higher over the midlatitude landmasses of the Northern Hemisphere, with values exceeding 0.9. These regions have a larger signal and a denser observational network than elsewhere and thus provide better agreement between surface temperature in observations (Jones et al. 1991) and NCEP reanalysis. Correlations are also generally lower over mountainous regions, for example, over the Himalaya and the Rockies. Correlations over the world’s oceans range between 0.6 and 0.9 except for lower values over the Indian Ocean, the tropical western Pacific, and the equatorial eastern Pacific. Again, these low-correlation regions can be characterized by poor data coverage and/or weak temperature anomalies.

5. Summary and conclusions

Recently, there has been some concern (Hurrell and Trenberth 1996) about the accuracy of the observed MSU-based datasets, particularly the lower-tropospheric channel 2R, and their utility for providing estimates of tropospheric mean temperature trends. Those authors questioned the stability of the time series and, in particular, the procedure in which the data from various satellite sources were merged to produce a “homogeneous” dataset over the entire period. Christy et al. (1998) performed an intercomparison between the MSU data and available radiosonde observations and could find no evidence of systematic bias in their MSU data. The conflicting evidence from these studies remained unresolved as of the drafting of this paper. The three reanalyses (DAO, ERA, and NCEP) that have formed the basis for the current paper have shown changes in the relative biases with respect to each other and with respect to the MSU estimates of global mean tropospheric temperature over the 1979–97 period. In particular, the NCEP2 has cooled relative to MSU2 since the early 1990s. The ERA2 data are only available for the period 1979–93, but they show changes in bias relative to MSU2 starting in 1990 that are consistent with the changes in bias between NCEP2 and MSU2. ERA2 is colder than both NCEP2 and MSU2 in the earlier part of the period and warmer in the later part of the period. The analyses described in this paper do not dwell on and could not uncover the reasons for these temporal changes in the relative bias of the reanalyses datasets.

The inconsistent temporal changes in the relative bias among the various reanalyses limits their use in global change studies. Nonetheless, the analysis also shows that the pattern correlations (r) between the MSU Ch2 mean fields and each of the reanalyses are very high (r > 0.96) and remain relatively high for the anomaly fields showing r > 0.8, generally r > 0.9. This suggests that the reanalysis may be useful for comparisons with numerical model–generated forecast fields and in conjunction with fingerprint techniques to identify climate change.

In comparisons of the simple linear trends present in each dataset for the 1980–90 period, each of the reanalyses had spatial patterns similar to MSU Ch2 except that the NCEP reanalysis showed smaller positive trends compared to the MSU, while the ERA reanalysis showed larger positive trends. The NASA/DAO reanalysis showed a mixed trend difference pattern. The spatial patterns of linear trends changed however with the addition of 3 yr of data to extend the trend analysis to the 1980–93 period, thus illustrating the inadequacy of short datasets to define patterns and, perhaps, even the sign of long-term trends.

The sensitivity of the estimated temperature trends to the addition of a few more years to the time series suggests that the interannual variability of tropospheric temperatures over relatively short periods profoundly influences all estimates of tropospheric temperature trends. This seasonal “climate noise” on the “trend” signal severely limits the usefulness of trends derived from these short-period data for climate change detection. The unresolved changes in bias discussed above and the unresolved differences in the interpretation of the MSU– radiosonde data biases by Hurrell and Trenberth (1996) and Christy et al. (1998) suggest that none of the reanalysis datasets, nor the MSU data, nor the radiosonde data are reliable measures of tropospheric temperature trend to a high degree of precision in the context of global climate change detection.

On the other hand, the consistency among the analyses in the study presented here also suggests that the tropospheric mean temperatures from the MSU or from any of the three reanalyses are adequate for the investigation of temperature variations on seasonal to interannual temporal scales. This study examined the spatial structure of the temporal correlations between the MSU Ch2 temperature and that of the reanalyses for full fields and anomalies. The full field correlations in the extratropics show high values in each reanalysis, that is, correlations >0.95, but the Tropics show smaller values, that is, correlations in the range from 0.6 to 0.9. The patterns of anomaly correlations show similar magnitudes but with somewhat reduced values in midlatitudes, suggesting a lower signal-to-noise ratio in midlatitudes.

The root-mean-square differences (rms) between the MSU Ch2 temperature and each of the reanalyses’ temperatures were computed to arrive at quantitative estimates of the uncertainty in these analyses. All reanalyses show similar patterns in the rms anomaly differences with magnitudes of 0.2°C or less in the Tropics and between 0.2° and 0.4°C at midlatitudes.

Temporal and spatial variations in the monthly and seasonal mean 2-m above surface temperature anomalies from the NCEP reanalysis exhibit a very coherent relationship with those of the observed gridded analyses of surface temperature (Jones et al. 1991). The time series of the pattern correlations in the seasonal anomaly fields are in the 0.65–0.85 range. The comparison with NCEP 2-m air temperature might be more “like with like” if the Jones (1994) land air temperatures were blended with marine air temperature (MATs) (Parker et al. 1995) rather than sea surface temperatures (SSTs). Although the quality of MATs is generally lower than that of SSTs, the MATs in the last four decades are reliable enough to show air–sea temperature difference anomalies in accord with atmospheric circulation anomalies on monthly timescales. Thus the MATs may have given higher correlations with NCEP 2-m air temperature.

The NCEP reanalyzed surface temperature patterns can be thought of as constrained interpolations of the observed data, with physically based constraints coming from the model’s data assimilation system. Because the station data represent point measurements and the model analysis an average over some area, there will always be some difference between station data observations and a model representation of those observations. Thus the correlations between “boxed average” values of the station data and the NCEP reanalysis will be less than perfect, that is, <1. However, given that the reanalysis (data assimilation) temperature anomaly fields should be more like the observations than any forecast (data simulation) of temperature anomalies, the box averaged versus reanalysis correlations shown here should be larger in magnitude than any box averaged versus forecast, or simulation correlations. Thus in the context of fingerprint climate change detection techniques, the magnitudes of the seasonal anomaly pattern correlations between the NCEP reanalysis and observed surface temperature, generally between 0.65 and 0.85, could suggest an upper limit to the magnitudes of the pattern correlations that might be obtained by correlating observed surface temperature anomalies with the forecast (i.e., simulated) temperature variations obtained from multiyear simulation runs made with current generation circulation climate models.

Acknowledgments

We thank Gene Rasmusson for his many helpful comments and suggestions during the evolution of this work. This work also benefitted through useful and productive interactions with Tim Barnett, Phil Jones, Gabi Hegerl, Ben Santer, and Mike Fiorino. We also wish to thank John Christy for providing the MSU2 data and for providing comments and suggestions during the course of this work and the Climatic Research Unit at the University of East Anglia for providing the surface temperature datasets. Critical comments by the two anonymous reviewers of the Journal of Climate greatly improved the readability and tied several loose ends of this manuscript. Comments by Tom Smith and Vern Kousky on earlier drafts of this paper are greatly appreciated. This work was partially supported by a grant from the Climate Change Data and Detection Program at NOAA’s Office of Global Programs.

REFERENCES

  • Arkin, P. A., V. E. Kousky, J. E. Janowiak, and E. A. O’Lenic, 1986:Atlas of the Tropical and Subtropical Circulation Derived from National Meteorological Center Operational Analyses. NOAA Atlas No. 7.

  • Barnett, T. P., and Coauthors, 1999: Detection and attribution of recent climate change: A status report. Bull. Amer. Meteor. Soc.,80, 2631–2659.

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

  • ——, C. F. Ropelewski, and N. C. Grody, 1995: Comparison of tropospheric temperature derived from microwave sounding unit and the National Meteorological Center analysis. J. Climate,8, 668–681.

  • Christy, J. R., R. W. Spencer, and R. T. McNider, 1995: Reducing noise in the MSU daily lower-tropospheric global temperature dataset. J. Climate,8, 888–896.

  • ——, ——, and E. A. Lobl, 1998: Analysis of the merging procedure for the MSU daily temperature time series. J. Climate,11, 2016– 2041.

  • Gibson, J. K., P. Kallberg, S. Uppala, A. Hernandez, A. Nomura, and E. Serrano, 1997: ECMWF Re-Analysis, Project Report Series 1: ERA Description, ECMWF, 72 pp.

  • Hasselmann, K., 1993: Optimal fingerprints for the detection of time-dependent climate change. J. Climate,6, 1957–1971.

  • Hegerl, G. C., H. von Storch, K. Hasselmann, B. D. Santer, U. Cubasch, and P. D. Jones, 1996: Detecting greenhouse-gas-induced climate change with an optimal fingerprint method. J. Climate,9, 2281–2306.

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

  • Hurrell, J. W., and K. E. Trenberth, 1996: Satellite versus surface estimates of air temperature since 1979. J. Climate,9, 2222– 2232.

  • ——, and ——, 1997: Spurious trends in satellite MSU temperatures from merging different satellite records. Nature,386, 164–167.

  • Jones, P. D., 1994: Recent warming in global temperature series. Geophys. Res. Lett.,21, 1149–1152.

  • ——, T. M. L. Wigley, and G. Farmer, 1991: Marine and land temperature data sets: A comparison and look at recent trends. Greenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier Science, 593–602.

  • Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc.,77, 437–471.

  • Karl, T. R., and Coauthors, 1993: A new perspective on recent global warming: Asymmetric trends of daily maximum and minimum temperature. Bull. Amer. Meteor. Soc.,74, 1007–1023.

  • Parker, D. E., C. K. Folland, and M. Jackson, 1995: Marine surface temperature: Observed variations and data requirements. Climatic Change,31, 429–470.

  • Santer, B. D., T. M. L. Wigley, P. D. Jones, and M. D. Schlesinger, 1991: Multivariate methods for the detection of greenhouse-gas-induced climatic change. Greeenhouse-Gas-Induced Climatic Change: A Critical Appraisal of Simulations and Observations, M. E. Schlesinger, Ed., Elsevier Science, 511–536.

  • ——, J. J. Hnilo, and P. D. Jones, 1993: Correlation methods in detection studies. Climate Dyn.,8, 265–276.

  • ——, ——, T. M. L. Wigley, J. S. Boyle, C. Doutriaux, M. Fiorino, D. E. Parker, and K. E. Taylor, 1999: Uncertainties in observationally based estimates of temperature change in the free atmosphere. J. Geophys. Res.,104, 6305–6333.

  • Schubert, S. D., J. Pfaendtner, and R. Rood, 1993: An assimilated data set for earth science applications. Bull. Amer. Meteor. Soc.,74, 2331–2342.

  • Trenberth, K. E., and J. W. Hurrell, 1994: Decadal atmosphere&ndash∘ean variations in the Pacific. Climate Dyn.,9, 303–319.

  • WCRP, 1997: Proc. First WCRP Int. Conf. on Reanalyses. WCRP-104, WMO/TD-876, Silver Spring, MD, WMO, 461 pp.

Fig. 1.
Fig. 1.

Time series of the tropospheric mean temperature (1979–97) as estimated by satellite (MSU2) and three reanalysis datasets, NCEP, ERA, and DAO. (See the text for detailed descriptions of these data.) (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 2.
Fig. 2.

Same as Fig. 1, but for temperature anomalies.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 3.
Fig. 3.

Time series of the tropospheric temperature anomaly differences between the NCEP and MSU2 (solid line) and ERA and MSU2 (dotted line). (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 4.
Fig. 4.

Time series of the pattern correlations between the MSU2 mean tropospheric temperature estimates and estimates from each of the reanalyses. (top) The globe equatorward of 80° lat. (bottom) The Tropics equatorward of 20° lat.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 5.
Fig. 5.

Same as in Fig. 4, but for pattern correlations between anomalies.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 6.
Fig. 6.

Spatial pattern of the monthly temporal correlations between the MSU2 and (top) NCEP, (middle) ERA, and (bottom) DAO reanalyses for the 1980–93 period. All correlations are positive, and contours are drawn at 0.4, 0.6, 0.7, 0.8, 0.9, and 0.95. The shading is as follows: 0.3–0.6 dark gray, 0.6–0.9 medium gray, >0.9 light gray. Regions with correlations less than 0.3 and areas with no data have no shading (white).

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 7.
Fig. 7.

Same as in Fig. 6 except the panels depict anomaly pattern correlations.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 8.
Fig. 8.

Spatial pattern of the linear mean tropospheric temperature trend as estimated by the MSU2 [°C (decade)−1] for the 1980–90 period. Negative (cooling) trends are darkly shaded and positive (warming) trends are lightly shaded. Contour lines are shown at every 0.3, with additional contours at +0.1 and −0.1.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 9.
Fig. 9.

Same as in Fig. 8, but for the reanalyses: (top) NCEP, (middle) ERA, and (bottom) DAO.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 10.
Fig. 10.

Spatial pattern of the differences in linear trend of mean tropospheric temperature, respectively, between MSU2 and the three different reanalyses: (top) NCEP, (middle) ERA, and (bottom) DAO for the 1980–90 period. Negative (cooling) trends are darkly shaded and positive (warming) trends are lightly shaded. Contour lines are shown at every 0.3, with additional contours at +0.1 and −0.1.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 11.
Fig. 11.

Same as in Fig. 10 except the period for the trend computation is 1980–93.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 12.
Fig. 12.

The 2-m temperature anomaly as estimated by the NCEP reanalysis for (top) Jan 1958, (middle) Jan 1979, and (bottom) Jan 1996. Contour interval is 1°C. Negative values are dashed.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 13.
Fig. 13.

The same as in Fig. 12, except the data are surface observations from Jones et al. (1991). Data-void regions are shown shaded gray.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 14.
Fig. 14.

Time series of surface temperature anomaly as estimated by the observed temperature from Jones et al. (1991) and the NCEP reanalysis for the 1958–96 period. (top) The globe, equatorward of 80°. (bottom) The Tropics equatorward of 20°.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 15.
Fig. 15.

(top) Time series of the seasonal pattern correlations between the NCEP reanalysis and the surface observations from Jones et al. (1991). (middle) Time series of the root-mean-square differences (rmsd) between seasonal estimates of surface temperature anomalies estimated by the NCEP reanalysis and surface observations from Jones et al. (1991). (bottom) Time series of the number of 5° lat × 5° long boxes with data by season in the Jones et al. (1991) dataset from 1958 to 1995.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Fig. 16.
Fig. 16.

Spatial patterns of temporal correlations between the observed surface datasets of Jones et al. (1991) and the NCEP reanalysis 2-m temperature for (top) Dec–Feb, (second panel) Jun–Aug, (third panel) Mar–May, and (bottom) Sep–Nov.

Citation: Journal of Climate 13, 17; 10.1175/1520-0442(2000)013<3187:RBTTEU>2.0.CO;2

Save