| 英文摘要: | The global ocean stores more than 90% of the heat associated with observed greenhouse-gas-attributed global warming1, 2, 3, 4. Using satellite altimetry observations and a large suite of climate models, we conclude that observed estimates of 0–700 dbar global ocean warming since 1970 are likely biased low. This underestimation is attributed to poor sampling of the Southern Hemisphere, and limitations of the analysis methods that conservatively estimate temperature changes in data-sparse regions5, 6, 7. We find that the partitioning of northern and southern hemispheric simulated sea surface height changes are consistent with precise altimeter observations, whereas the hemispheric partitioning of simulated upper-ocean warming is inconsistent with observed in-situ-based ocean heat content estimates. Relying on the close correspondence between hemispheric-scale ocean heat content and steric changes, we adjust the poorly constrained Southern Hemisphere observed warming estimates so that hemispheric ratios are consistent with the broad range of modelled results. These adjustments yield large increases (2.2–7.1 × 1022 J 35 yr−1) to current global upper-ocean heat content change estimates, and have important implications for sea level, the planetary energy budget and climate sensitivity assessments. Numerous studies have examined the long-term (~1950-present) global average and basin-scale evolution of ocean heat content (OHC) change in the upper 0–700 dbar (refs 1, 4, 8, 9, 10, 11, 12) (Supplementary Information) and important advancements have been made to correct for systematic measurement biases6, 13, 14. Evidence exists for a poleward shift of the subtropical gyres and marked warming in the Southern Ocean7, 15, 16, 17, but limitations of methods used to ‘infill’ these data-sparse regions may introduce a conservative bias toward low magnitude (zero) changes5, 6, 7. Recent estimates of OHC change attempt to address sampling deficiencies by relying on coincident sea surface height (SSH) estimates or the modern Argo array8, 9, 12, 18. Additional ocean warming studies apply formal detection and attribution approaches that rely on intrinsic variability estimates from models, and avoid using infilled data by ‘subsampling’ models in space and time, consistent with the sparse historical observations19, 20, 21, 22. Here, we investigate the large-scale spatial structure of OHC changes in five observational estimates (derived independently with differing processing choices) that were evaluated in the IPCC Fifth Assessment Report4. Based on a series of consistency checks with precise altimetry data and a large ensemble of climate models (Coupled Model Intercomparison Project (CMIP) phases 3 and 5), we find that observed Southern Hemisphere (SH) 0–700 dbar OHC changes are significantly underestimated. We analyse the 35-year period (1970–2004) over which both the CMIP5 ‘historical’ data are available and during which observational sampling deficiencies are small enough to yield reliable OHC changes, at least in the Northern Hemisphere (NH; ref. 21). OHC changes from the surface to 700 dbar are first examined in four observational estimates for which infilled gridded data were available10, 11, 18, 23 (Supplementary Fig. 2a and Methods); a fifth data set (Dom08; ref. 8) provides only hemispheric time series, but is included in subsequent analyses below. In Fig. 1a we show one of the observed results (Lev12; ref. 11) alongside the CMIP5 historical multi-model mean (MMM, Fig. 1b). To facilitate comparison of the observed and simulated spatial structure of OHC changes, additional maps show results with the global average removed (Fig. 1c, d and Supplementary Fig. 2a: A2–E2). The regions of inconsistency among the data sets are stippled, indicating where at least one of the four observational estimates disagrees in the sign of the mapped change (Fig. 1a, c and Supplementary Fig. 2a: A1–D1, A2–D2), or where fewer than 75% of models agree with the MMM sign (Fig. 1b, d and Supplementary Fig. 2a: E1, E2 and Fig. 2b: A2–D2, E2–I2). A prominent SH warming feature (30° S–50° S) is evident in MMM trend maps (Fig. 1b, d and Supplementary Fig. 2b), consistent with previous modelling studies24, 25. This strong warming is less distinct in all observational analyses (Supplementary Fig. 2a), which is likely due to SH data sparsity and internal variability that can mask the externally forced warming in this region. As expected, the MMM is smoother than the observed analyses because uncorrelated variability present in individual simulated records is averaged out (Supplementary Fig. 2a: A2–D2).
|
