• Boyer, T. P., and Coauthors, 2006: World Ocean Database 2005. NOAA Atlas NESDIS 60, 190 pp.

  • Conkright, M. E., and Coauthors, 2002: Introduction. Vol. 1, World Ocean Database 2001, NOAA Atlas NESDIS 42, 160 pp.

  • Gille, S. T., 2002: Warming of the Southern Ocean since the 1950s. Science, 295 , 12751277.

  • Gouretski, V., , and K. P. Koltermann, 2007: How much is the ocean really warming? Geophys. Res. Lett., 34 .L01610, doi:10.1029/2006GL027834.

    • Search Google Scholar
    • Export Citation
  • Hanawa, K., , P. Rual, , R. Bailey, , A. Sy, , and M. Szabados, 1995: A new depth time equation for Sippican or TSK T-7, T-6 and T-4 expendable bathythermographs (XBT). Deep-Sea Res., 42 , 14231451.

    • Search Google Scholar
    • Export Citation
  • Harrison, D. E., , and M. Carson, 2007: Is the world ocean warming? Upper-ocean temperature trends: 1950–2000. J. Phys. Oceanogr., 37 , 174187.

    • Search Google Scholar
    • Export Citation
  • Harrison, D. E., , and M. Carson, 2008: Upper ocean warming: Spatial patterns of trends and interdecadal variability. NOAA Tech. Memo. OAR PMEL-138, 35 pp. [Available online at http://www.pmel.noaa.gov/pubs/PDF/harr3144/harr3144.pdf.].

  • Levitus, S., , J. I. Antonov, , T. P. Boyer, , and C. Stephens, 2000: Warming of the world ocean. Science, 287 , 22252229.

  • Levitus, S., , J. I. Antonov, , and T. P. Boyer, 2005: Warming of the world ocean, 1955–2003. Geophys. Res. Lett., 32 .L02604, doi:10.1029/2004GL021592.

    • Search Google Scholar
    • Export Citation
  • Locarnini, R. A., , A. V. Mishonov, , J. I. Antonov, , T. P. Boyer, , and H. E. Garcia, 2006: Temperature. Vol. 1, World Ocean Atlas 2005, NOAA Atlas NESDIS 61, 182 pp.

    • Search Google Scholar
    • Export Citation
  • Lyman, J. M., , J. K. Willis, , and G. C. Johnson, 2006: Recent cooling of the upper ocean. Geophys. Res. Lett., 33 .L18604, doi:10.1029/2006GL027033.

    • Search Google Scholar
    • Export Citation
  • Press, W. H., , B. P. Flannery, , S. A. Teukolsky, , and W. T. Vetterling, 1992: Numerical Recipes in FORTRAN: The Art of Scientific Computing. Cambridge University Press, 992 pp.

    • Search Google Scholar
    • Export Citation
  • Willis, J. K., , D. Roemmich, , and B. Cornuelle, 2004: Interannual variability in upper ocean heat content, temperature, and thermosteric expansion on global scales. J. Geophys. Res., 109 .C12036, doi:10.1029/2003JC002260.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    (a)–(f) Temperature trends in 3° × 3° regions based on WOD05 OL data anomalies over the period 1955–2003 [°C (49 yr)−1]. Shown are regions with trends passing the 90% CL via the correlation coefficient t test as well as the distribution criterion. The grid size, nominal depth, and percentage of ocean covered by significant trends are listed in Asia in (a)–(f). Note that in this and subsequent figures there are reference lines drawn at 40°N, the equator, 40°S, and 160°W for convenience.

  • View in gallery

    As in Fig. 1, but for all regions passing the distribution criterion (at least three monthly gridded observations per decade in at least four of the five decades of the analysis period).

  • View in gallery

    (a)–(f) Temperature trends based on WOD05 SL data anomalies over the period 1955–2003 [°C (49 yr)−1]. Shown are all regions with data. The SL depths for (a)–(f) are listed in Asia. The data were gridded to a 3° × 3° grid. The color bar is the same as in previous trend figures.

  • View in gallery

    (a)–(e) Temperature trends based on the objectively analyzed annual temperature anomaly dataset (LEV05) examined in Levitus et al. (2005). The data in LEV05 are all yearly temperature anomalies on a 1° × 1° grid. The trends are over the period 1955–2003 [°C (49 yr)−1]. The SL depths for (a)–(e) are listed in Asia. These maps show trends based on the original data and are not smoothed any further here. The color bar is the same as in previous trend figures.

  • View in gallery

    (a)–(e) Temperature trend differences [°C (49 yr)−1] for the LEV05 dataset trends minus the WOD05 SL data trends (regridded to 1° × 1°). The SL depths for (a)–(e) are listed in Asia. The trend differences were all smoothed by a Welch window, which included nearest neighbors in both longitude and latitude to improve readability. The color bar is the same as in previous trend figures.

  • View in gallery

    As in Fig. 5, but temperature trend differences are LEV05 dataset trends minus the WOD05 OL data trends described in the text. The grid size for the WOD05 data is 3° × 3° (regridded to 1° × 1° to match LEV05), and the resulting difference maps were smoothed by a nearest-neighbors Welch window to improve readability.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 4 4 0
PDF Downloads 2 2 0

Is the Upper Ocean Warming? Comparisons of 50-Year Trends from Different Analyses

View More View Less
  • 1 Joint Institute for the Study of the Atmosphere and Ocean, and School of Oceanography, University of Washington, Seattle, Washington
  • 2 Joint Institute for the Study of the Atmosphere and Ocean, and School of Oceanography, University of Washington, and NOAA/Pacific Marine Environmental Laboratory, Seattle, Washington
© Get Permissions
Full access

Abstract

There is great interest in World Ocean temperature trends, yet the historical global ocean database has very uneven coverage in space and time. Previous work on 50-yr upper ocean temperature trends from the NOAA ocean data archive is extended here. Trends at depths from 50 to 1000 m are examined, based on observations gridded over larger regions than in the earlier study. Despite the use of larger grid boxes, most of the ocean does not have significant 50-yr trends at the 90% confidence level (CL). In fact only 30% of the ocean at 50 m has 90% CL trends, and the percentage decreases significantly with increasing depth. As noted in the previous study, there is much spatial structure in 50-yr trends, with areas of strong warming and strong cooling. These trend results are compared with trends calculated from data interpolated to standard levels and from a highly horizontally interpolated version of the dataset that has been used in previous heat content trend studies. The regional trend results can differ substantially, even in the areas with statistically significant trends. Trends based on the more interpolated analyses show more warming. Together with major temporal and spatial sampling limitations, the previously described strong interdecadal and spatial variability of trends makes it very difficult to formally estimate uncertainty in World Ocean averages, but these results suggest that upper ocean heat content integrals and integral trends may be substantially more uncertain than has yet been acknowledged. Further exploration of uncertainties is needed.

* Pacific Marine Environmental Laboratory Publication Number 3073 and National Oceanic and Atmospheric Administration Contribution Number 1402

Corresponding author address: Mark Carson, School of Oceanography, University of Washington, Box 357940, Seattle, WA 98195-7940. Email: mark.carson@noaa.gov

Abstract

There is great interest in World Ocean temperature trends, yet the historical global ocean database has very uneven coverage in space and time. Previous work on 50-yr upper ocean temperature trends from the NOAA ocean data archive is extended here. Trends at depths from 50 to 1000 m are examined, based on observations gridded over larger regions than in the earlier study. Despite the use of larger grid boxes, most of the ocean does not have significant 50-yr trends at the 90% confidence level (CL). In fact only 30% of the ocean at 50 m has 90% CL trends, and the percentage decreases significantly with increasing depth. As noted in the previous study, there is much spatial structure in 50-yr trends, with areas of strong warming and strong cooling. These trend results are compared with trends calculated from data interpolated to standard levels and from a highly horizontally interpolated version of the dataset that has been used in previous heat content trend studies. The regional trend results can differ substantially, even in the areas with statistically significant trends. Trends based on the more interpolated analyses show more warming. Together with major temporal and spatial sampling limitations, the previously described strong interdecadal and spatial variability of trends makes it very difficult to formally estimate uncertainty in World Ocean averages, but these results suggest that upper ocean heat content integrals and integral trends may be substantially more uncertain than has yet been acknowledged. Further exploration of uncertainties is needed.

* Pacific Marine Environmental Laboratory Publication Number 3073 and National Oceanic and Atmospheric Administration Contribution Number 1402

Corresponding author address: Mark Carson, School of Oceanography, University of Washington, Box 357940, Seattle, WA 98195-7940. Email: mark.carson@noaa.gov

1. Introduction

There is great interest in the large-scale, low-frequency variability of oceanic temperature and heat content, regionally and over the World Ocean. Estimates of long-term, large-scale temperature or heat content trends include, for example, Levitus et al. (2000, 2005, Gille (2002), Willis et al. (2004), Lyman et al. (2006), and Gouretski and Koltermann (2007). The analysis approach of Harrison and Carson (2007, hereafter, HC) is extended to test the sensitivity of trends to various data processing methods. The World Ocean Database 2005 (WOD05; Boyer et al. 2006) is used, which has an increased number of profiles—7 900 349 as compared with 7 037 213 in World Ocean Database 2001 (WOD01; Conkright et al. 2002).

Trend results, over the period 1955–2003, are calculated and compared from observed-level (OL) and vertically interpolated standard-level (SL) data using an analysis based on the HC approach. Trends based on these datasets are also compared with those computed from the horizontally objectively analyzed fields used in Levitus et al. (2005). The trend results differ in many regions. The data and methods are discussed further in section 2. Section 3 contains the planview maps of 49-yr (1955–2003) OL trends as well as comparisons with SL data trends and the Levitus et al. (2005) dataset. A summary and discussion of these results are presented in section 4.

2. Data and methods

The methods of HC are used here with some modifications. The primary dataset is the more recent and somewhat larger WOD05 (the same analysis was applied to the previously used WOD01 and results were very similar to those presented here). The HC analysis was extended to larger 3° × 3°, 5° × 5°, and 2° × 10° regions; only 3° × 3° results are presented here [see Figs. 2 and 3 of Harrison and Carson (2008) for results over the other regions]. From each OL observation, the 1° × 1° climatological monthly mean from the World Ocean Atlas 2005 (WOA05; Locarnini et al. 2006) statistical mean dataset was removed. The resulting anomalies were binned and averaged to produce a single monthly time series for each grid box. The depth ranges used for binning were 50 ± 2.5, 100 ± 5, 300 ± 20, 500 ± 20, 700 ± 50, and 1000 ± 50 m.

Fall rate–based depth corrections were applied to the OL XBT data in WOD05 (see, e.g., Hanawa et al. 1995). As in HC it was found that this correction does not affect large-scale trend patterns, but the corrections were made here to enable meaningful comparison of trends from these results with those of Levitus et al. (2005) and the SL data trends, both of which had applied these XBT corrections. There are a considerable number of profiles for which the correction status is uncertain—about 55% of the XBT profiles in WOD05 are of unknown instrument type (Boyer et al. 2006). The correction was applied only to XBTs flagged in WOD05 as needing correction.

The time series in each grid box was fitted with a trend line using linear least squares as in HC. A criterion was used again to select regions that are minimally sampled enough through time so trends are likely to be reasonable long-term estimates. Here we have chosen a criterion of three or more monthly (gridded) observations for at least four of the five “decades” (9.8-yr periods) between January 1955 and December 2003 in each grid box. This reduction from the HC criterion of five observations for four of the five decadal periods could be relaxed to three observations without adding in too much trend noise (high box-to-box trend variability) to the maps. This period was sampled better than the period studied in HC, as reported in Levitus et al. (2005). Planview maps were created of the trend magnitudes over the 49-yr period (January 1955–December 2003) and are presented in the next section for regions passing the data distribution criterion and also for those trends that pass a 90% statistical confidence level (CL).

These OL results are compared with those based on the vertically interpolated standard-level data in WOD05. The same general analysis was performed yielding a 3° × 3° grid of monthly anomalies. No data distribution criterion was applied, which approach was used so as to allow a view of the very noisy trend patterns in the undersampled regions and to allow better comparison with the Levitus et al. (2005) results.

We also compare the results of the OL and SL data trends with those obtained from the dataset of upper ocean yearly anomalies referenced in Levitus et al. (2005). This set was accessed via the National Oceanographic Data Center Internet site (http://www.nodc.noaa.gov/OC5/DATA_ANALYSIS/anomaly_ data.html) and will be referred to hereafter as LEV05. LEV05 is a vertically interpolated and horizontally objectively analyzed, global annual mean anomaly dataset for 1955–2003 on the 16 standard levels of the upper ocean (0–700 m, defined in WOA05). This dataset was built from the data in WOD01 supplemented with 310 000 additional profiles (Levitus et al. 2005) that have subsequently been included in WOD05. Given the general similarity of our trend results using either WOD05 or WOD01, we assume that any large-scale differences with the LEV05 results will arise primarily from analysis technique rather than from dataset differences.

3. Results

a. 1955–2003 trends from WOD05 OL data

Trends were calculated from OL data in WOD05 over the 49-yr interval of 1955–2003. The results for the 90% CL trends are shown in Fig. 1. The trend results for all regions passing the data distribution requirement (i.e., no filtering for statistical significance) are shown in Fig. 2.

The 90% CL trends at 100 m (Fig. 1b) cover a little more of the ocean than the 1950–2000 trends presented in HC (27.5% versus 26.6%). At 300 m (Fig. 1c), the change in coverage for significant trends is 22.5% versus 18.3%. These changes are not large, but some differences in trend patterns are a bit more striking in compared with those of HC (cf. HC Figs. 7a,d and 7b,e). At 100 m (Fig. 1b), the warming along the U.S. West Coast is weaker and fewer regions are significant, but the warming throughout the subtropical North Atlantic Ocean and the central tropical Pacific Ocean (near 160°W) is more consistent and better featured in the later trends. The strong cooling in the western tropical Pacific and Indian Oceans is still well represented by these trends. At 300 m (Fig. 1c), some new coherent features are suggested in the less well-sampled tropical North Pacific, with a warming patch east of 160°W and a cooling patch west of 160°W. There is also a stronger suggestion of warming in the tropical South Pacific east of 160°W (along 20°S) and cooling in more of the tropical/subtropical South Atlantic. Overall though, the HC 2° × 2° results for 1950–2000 trends are very similar to these 3° × 3° trends for 1955–2003.

It is clear from Fig. 1 that the coverage of the ocean with statistically significant trends decreases with depth, especially below 460 m, the nominal terminal depth of the ubiquitous T-4 XBT. The trends at 700 and 1000 m (Figs. 1e,f) are very weak except for strong warming near 40°N in the North Atlantic (which is evident at all depths), at spots along the North Atlantic Current path, and some strong warming near the southern tip of Japan (which is also expressed at most depths but is much weaker at 500 m).

Comparison of Fig. 1 with Fig. 2 indicates that the 90% CL filter generally picks out the larger trends well. Setting aside that filter to look at all regions passing a simple data distribution criterion in Fig. 2 only adds noisy regions (especially in the southern basins) and regions of small amplitude that connect the stronger trend regions. No new information about coherent patterns is revealed. Again, the maps clearly show a strong reduction in coverage with depth; the ocean below 500 m is poorly sampled over this period. Note that the band of sparse data along the equator present in the HC results was due to a flaw in the copy of the WOD01 used and does not appear in the current analyses.

b. Comparison with SL data

Here the OL trend results are compared with trends calculated from SL data in WOD05. Maps of the SL trends are presented in Fig. 3. Note that no data distribution or statistical significance filter has been applied to these results. The vertical interpolation increases the number of data in a grid box relative to OL data so that there is better spatial coverage. Even so, the trend noise in the Antarctic Circumpolar Current region of the Southern Hemisphere is still distinguishable by high small-scale variability, particularly south of 40°S at the upper depths (Figs. 3a–c) and also in more-subtropical regions of the southern basins at lower depths (Figs. 3d–f).

As in the OL data trends, there are large-scale coherent patterns of trends in the Northern Hemisphere and in parts of the tropics in Fig. 3. The same patterns are produced as with the OL data (difference map not shown) with some notable exceptions. The trend differences are noisy in the eastern tropical Pacific in the upper layers, an area particularly undersampled in addition to the Southern Ocean basins (Figs. 3a,b). Also, the cooling in the eastern Indian Ocean and amid Malaysia at 100 m is stronger here (Fig. 3b). Regions of more warming and weaker cooling include the North Pacific basin at 500 m, along with the much weaker cooling in the South Pacific east of Australia (Fig. 3d). Many of these differences would increase the trend of the heat content integral significantly, but elsewhere the patterns match up well with the OL trend patterns, even in magnitude. Interpolating to standard depth levels does not yield quite the same trend patterns that binning OL data in depth does, but the differences are mostly regional and are usually small, with the exceptions noted above.

c. Comparison with LEV05 dataset

Trends were calculated from the dataset analyzed in Levitus et al. (2005) at the depths previously discussed for comparison. The results are displayed in Fig. 4 (1000-m results are not available for the annual anomaly fields, which end at the 700-m standard level). Superficially, the LEV05 trends bear a semblance to the trend patterns evident in the SL and OL data; they show large-scale coherent trends with similar sign to the above results in many regions. However, the warming regions are more strongly warming in the upper depths (Figs. 4a,b), and the region along 40°N in the North Atlantic is very strongly warming at all depths. Also, south of 40°S the spatially noisy trends of the SL analysis are replaced with smaller-amplitude, somewhat larger space-scale trends; the objective analysis procedure dampens noise in the Southern Ocean regions.

It is easier to see the differences between the LEV05 trend results and the OL and SL results by constructing difference maps. Because the LEV05 dataset has been smoothed by its objective analysis procedure, we smooth the OL and SL maps with a Welch window applied to nearest neighbors in longitude and latitude, with a weight set of {0.3, 0.4, 0.3} (see Press et al. 1992). Before smoothing, we also filter out any unrealistic trend with a cutoff of ±4°C · (49 yr)−1 to keep tiny regions with very large trends from contaminating the smoothed maps. The differences between LEV05 and the SL data trends are presented in Fig. 5. Note that noise in the Southern Ocean basins is still present in these maps, and it comes from noisiness in the smoothed SL trend maps. The differences in the upper layers (50 and 100 m) include the stronger warming (darker reds) in LEV05 trends over much of the Indian Ocean and the tropics (Figs. 5a,b). Stronger warming also occurs in much of the subtropical South Atlantic, subtropical North Atlantic, central North Pacific, and between Argentina and the Antarctic Peninsula. The LEV05 trends are less warming along the coast of northern Africa and in the subpolar regions of the North Atlantic. There is also some broad scattered areas of cooler difference in much of the midlatitude North Pacific.

At depths below approximately 100 m, the LEV05 trends are very strongly warming along 40°N in the North Atlantic and overpower any similar warming in the SL trends (Figs. 5c–e). There is also stronger warming in LEV05 over much of the central North Pacific and the tropics, especially near Florida and the Caribbean, and also scattered along 40°S in the South Atlantic. While there are still regions of more cooling or less intense warming trends in the subpolar and subtropical North Atlantic and North Pacific and the western South Pacific, the warming regions are stronger and cover more area on average.

The difference maps comparing the LEV05 trends and the OL trends are shown in Fig. 6. These maps strongly suggest a warming bias in the LEV05 trends. The upper layer trends (Figs. 6a,b) in LEV05 have more positive slope in most places except in the eastern and northern North Atlantic and the western and northwestern North Pacific. There are also pockets of more negative slope in the eastern tropical Pacific, but there are more and stronger pockets of positive slope throughout this region. Certainly, this is one of the more undersampled regions outside of the Southern Ocean basins. At deeper depths, the warming bias is widespread and dominates all basins (Figs. 6c–e); although at 700 m the stronger LEV05 warming is mostly confined to the same North Atlantic region along 40°N.

4. Summary and discussion

The basic results of HC have been shown to be largely unaffected by working with temperature anomalies, by larger grid boxes (3° × 3°), and by imposing a slightly weaker data distribution requirement. Even though more of the World Ocean satisfies the weaker data criterion with larger boxes, few new geographical regions have significant trends. We have not shown our results over 5° × 5° and 2° × 10° grid boxes because they introduce no significant new results. The larger-region analyses are also not shown, as they exhibit strong time and space variability of 20-yr running trends sufficiently similar to that described by HC. All results are in Harrison and Carson (2008), which at the time of writing was available online (http://www.pmel.noaa.gov/pubs/PDF/harr3144/harr3144.pdf).

We have emphasized the comparison of the 1955–2003 trends based on 3° × 3° OL data, on the vertically interpolated SL data, and on the vertically and horizontally analyzed fields reported in Levitus et al. (2005). Some aspects of the consequences of sampling biases (spatial and temporal) on estimates of 50-yr trends based on interpolated fields can be illustrated through these comparisons. These datasets have different trends in many areas because of the different methods of analyzing roughly the same raw data. The most interpolated dataset, LEV05, yields warmer trends over the 1955–2003 interval than do the SL data in many parts of the ocean and at all depths and yields much warmer trends than do OL data. Striking differences occur even in some regions where there are statistically significant trends.

These results, which compare trend results from three versions of substantially the same raw dataset, suggest that interpolation over the very data sparse areas of the World Ocean may have substantial effects on trend results. The use of an objectively analyzed mean (used in producing LEV05) versus the statistical mean climatology dataset in WOA05 might also have an impact on these results. The trend differences, averaged over all regions shown in Figs. 5 and 6 and also for SL trends minus OL trends, are shown for each level in Table 1. They show that the heat content trend estimates presented in Levitus et al. (2005) yield more warming than do trends from the less-interpolated analyses. Unfortunately, we have no way to quantify the World Ocean average uncertainty without doing integrals that we believe are unwise to do.

Vertical integration over different depth ranges (e.g., 0–300 and 0–700 m) and often over oceanic basins has been used to simplify analysis (Levitus et al. 2005; Lyman et al. 2006; and others). Uncertainty due to sampling issues beneath the terminal depths of XBTs (∼700–1000 m) has been considered by Gouretski and Koltermann (2007), who estimated over a 50% sampling error range for the heat content estimates of the 0–3000-m layer. They also demonstrated that there are potentially large instrument biases complicating heat content estimates. Lyman et al. (2006) estimated the 0–750-m sampling error, finding it to be much smaller after 1967, when XBTs became available; this suggests that upper ocean heat content trends using data prior to 1967 are susceptible to large errors.

The substantial decadal temperature variability of the global ocean, coupled with strongly biased temporal and spatial sampling, yields datasets that are incomplete in their description of the evolution of the ocean state. The differences described here may or may not bound the real oceanic trends; from a raw dataset, it is not possible to go further without making additional assumptions about oceanic behavior in the huge poorly sampled regions of the World Ocean. Subsampling output, using historical data distributions, from high-resolution ocean general circulation models that demonstrate qualitative consistency with the historical dataset offers one way to gain additional insight into these uncertainties. It is very important to continue studies to explore uncertainty estimates for World Ocean heat content and temperature trends.

Acknowledgments

This work was supported by the NOAA/Office of Climate Observations and the Pacific Marine Environmental Laboratory (Eddie Bernard, Director). This publication is (partially) funded by the Joint Institute for the Study of the Atmosphere and Ocean (JISAO) under NOAA Cooperative Agreement NA17RJ1232. Much of the work presented here was obtained with the Ferret freeware analysis package (http://www.ferret.noaa.gov).

REFERENCES

  • Boyer, T. P., and Coauthors, 2006: World Ocean Database 2005. NOAA Atlas NESDIS 60, 190 pp.

  • Conkright, M. E., and Coauthors, 2002: Introduction. Vol. 1, World Ocean Database 2001, NOAA Atlas NESDIS 42, 160 pp.

  • Gille, S. T., 2002: Warming of the Southern Ocean since the 1950s. Science, 295 , 12751277.

  • Gouretski, V., , and K. P. Koltermann, 2007: How much is the ocean really warming? Geophys. Res. Lett., 34 .L01610, doi:10.1029/2006GL027834.

    • Search Google Scholar
    • Export Citation
  • Hanawa, K., , P. Rual, , R. Bailey, , A. Sy, , and M. Szabados, 1995: A new depth time equation for Sippican or TSK T-7, T-6 and T-4 expendable bathythermographs (XBT). Deep-Sea Res., 42 , 14231451.

    • Search Google Scholar
    • Export Citation
  • Harrison, D. E., , and M. Carson, 2007: Is the world ocean warming? Upper-ocean temperature trends: 1950–2000. J. Phys. Oceanogr., 37 , 174187.

    • Search Google Scholar
    • Export Citation
  • Harrison, D. E., , and M. Carson, 2008: Upper ocean warming: Spatial patterns of trends and interdecadal variability. NOAA Tech. Memo. OAR PMEL-138, 35 pp. [Available online at http://www.pmel.noaa.gov/pubs/PDF/harr3144/harr3144.pdf.].

  • Levitus, S., , J. I. Antonov, , T. P. Boyer, , and C. Stephens, 2000: Warming of the world ocean. Science, 287 , 22252229.

  • Levitus, S., , J. I. Antonov, , and T. P. Boyer, 2005: Warming of the world ocean, 1955–2003. Geophys. Res. Lett., 32 .L02604, doi:10.1029/2004GL021592.

    • Search Google Scholar
    • Export Citation
  • Locarnini, R. A., , A. V. Mishonov, , J. I. Antonov, , T. P. Boyer, , and H. E. Garcia, 2006: Temperature. Vol. 1, World Ocean Atlas 2005, NOAA Atlas NESDIS 61, 182 pp.

    • Search Google Scholar
    • Export Citation
  • Lyman, J. M., , J. K. Willis, , and G. C. Johnson, 2006: Recent cooling of the upper ocean. Geophys. Res. Lett., 33 .L18604, doi:10.1029/2006GL027033.

    • Search Google Scholar
    • Export Citation
  • Press, W. H., , B. P. Flannery, , S. A. Teukolsky, , and W. T. Vetterling, 1992: Numerical Recipes in FORTRAN: The Art of Scientific Computing. Cambridge University Press, 992 pp.

    • Search Google Scholar
    • Export Citation
  • Willis, J. K., , D. Roemmich, , and B. Cornuelle, 2004: Interannual variability in upper ocean heat content, temperature, and thermosteric expansion on global scales. J. Geophys. Res., 109 .C12036, doi:10.1029/2003JC002260.

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

(a)–(f) Temperature trends in 3° × 3° regions based on WOD05 OL data anomalies over the period 1955–2003 [°C (49 yr)−1]. Shown are regions with trends passing the 90% CL via the correlation coefficient t test as well as the distribution criterion. The grid size, nominal depth, and percentage of ocean covered by significant trends are listed in Asia in (a)–(f). Note that in this and subsequent figures there are reference lines drawn at 40°N, the equator, 40°S, and 160°W for convenience.

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Fig. 2.
Fig. 2.

As in Fig. 1, but for all regions passing the distribution criterion (at least three monthly gridded observations per decade in at least four of the five decades of the analysis period).

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Fig. 3.
Fig. 3.

(a)–(f) Temperature trends based on WOD05 SL data anomalies over the period 1955–2003 [°C (49 yr)−1]. Shown are all regions with data. The SL depths for (a)–(f) are listed in Asia. The data were gridded to a 3° × 3° grid. The color bar is the same as in previous trend figures.

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Fig. 4.
Fig. 4.

(a)–(e) Temperature trends based on the objectively analyzed annual temperature anomaly dataset (LEV05) examined in Levitus et al. (2005). The data in LEV05 are all yearly temperature anomalies on a 1° × 1° grid. The trends are over the period 1955–2003 [°C (49 yr)−1]. The SL depths for (a)–(e) are listed in Asia. These maps show trends based on the original data and are not smoothed any further here. The color bar is the same as in previous trend figures.

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Fig. 5.
Fig. 5.

(a)–(e) Temperature trend differences [°C (49 yr)−1] for the LEV05 dataset trends minus the WOD05 SL data trends (regridded to 1° × 1°). The SL depths for (a)–(e) are listed in Asia. The trend differences were all smoothed by a Welch window, which included nearest neighbors in both longitude and latitude to improve readability. The color bar is the same as in previous trend figures.

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Fig. 6.
Fig. 6.

As in Fig. 5, but temperature trend differences are LEV05 dataset trends minus the WOD05 OL data trends described in the text. The grid size for the WOD05 data is 3° × 3° (regridded to 1° × 1° to match LEV05), and the resulting difference maps were smoothed by a nearest-neighbors Welch window to improve readability.

Citation: Journal of Climate 21, 10; 10.1175/2007JCLI2002.1

Table 1.

Averaged 49-yr temperature change differences on each level. In this table, SL − OL (°C) is the SL data − OL data trends, averaged over all regions (limited by the OL data distribution criterion), LEV05 − SL (°C) is the LEV05 data trends − SL data trends, averaged over all regions (globally), and LEV05 − OL (°C) is the LEV05 data trends − OL data trends, averaged over all regions (limited by the OL data distribution criterion).

Table 1.
Save