| 英文摘要: | The rate of global mean sea-level (GMSL) rise has been suggested to be lower for the past decade compared with the preceding decade as a result of natural variability1, with an average rate of rise since 1993 of +3.2 ± 0.4 mm yr−1 (refs 2, 3). However, satellite-based GMSL estimates do not include an allowance for potential instrumental drifts (bias drift4, 5). Here, we report improved bias drift estimates for individual altimeter missions from a refined estimation approach that incorporates new Global Positioning System (GPS) estimates of vertical land movement (VLM). In contrast to previous results (for example, refs 6, 7), we identify significant non-zero systematic drifts that are satellite-specific, most notably affecting the first 6 years of the GMSL record. Applying the bias drift corrections has two implications. First, the GMSL rate (1993 to mid-2014) is systematically reduced to between +2.6 ± 0.4 mm yr−1 and +2.9 ± 0.4 mm yr−1, depending on the choice of VLM applied. These rates are in closer agreement with the rate derived from the sum of the observed contributions2, GMSL estimated from a comprehensive network of tide gauges with GPS-based VLM applied (updated from ref. 8) and reprocessed ERS-2/Envisat altimetry9. Second, in contrast to the previously reported slowing in the rate during the past two decades1, our corrected GMSL data set indicates an acceleration in sea-level rise (independent of the VLM used), which is of opposite sign to previous estimates and comparable to the accelerated loss of ice from Greenland and to recent projections2, 10, and larger than the twentieth-century acceleration2, 8, 10. The satellite-era time series of GMSL is a seminal climate data record2, 3 that describes one of the most robust manifestations of climate change. Accurate estimates and projections of the rate of sea-level rise, and any acceleration or deceleration thereof are of major importance for evaluating model projections and for adaptation planning, particularly for low-lying highly populated, highly productive and environmentally sensitive areas11. The accuracy of these GMSL estimates from data over the past two decades is dependent on the determination of fixed and time-variable systematic errors within and between each of the three successive satellite altimeter missions (TOPEX/Poseidon12 (T/P), Jason-1 (ref. 13) and OSTM/Jason-2 (ref. 14)) used in GMSL studies. Validation of the record (often termed bias drift estimation; that is, estimating drift of the altimeter sea surface height system) requires comparison of the altimeter and tide gauge (TG) sea levels over a network of TG sites (for example, refs 5, 6, 7). This approach has been used previously to successfully diagnose algorithm and instrumental errors4, 15, 16 and, after correction, drift estimates have been small and have not been used to further adjust (or calibrate) the observational records3. However, past implementations of this approach have limitations dominated by uncertainty in their adopted VLM at TGs (refs 5, 7). The validation is also sensitive to a typically poor spatial distribution of suitable TGs, and earthquake deformation at individual TGs (ref. 17). Here we develop an alternative method that addresses these limitations. We expand the network of TGs used (with respect to that used by ref. 7) by a factor of ~2 to 96 TGs (Fig. 1), using high-rate hourly data. Unlike previous work5, 6, for each TG we compute bias drift and residual ocean tide at multiple offshore comparison points (CPs) for each satellite pass, using up to the maximum of four passes surrounding each TG. We correct bias drift estimates for VLM using new data from the expanding network of GPS stations co-located with or near to TGs. These VLM trends are derived from homogeneously reprocessed GPS data (updated from ref. 18), or where they are not available, we use VLM derived from a model of glacial isostatic adjustment19 (GIA) combined with estimates of present-day elastic effects derived from the GRACE mission20. Accurate GPS estimates of VLM are preferable to those from GIA models as GIA is only one component of VLM, and, for many TGs, may not be the dominant signal18. Of our final TGs, 69% have one or more GPS estimates of VLM within 100 km (see Supplementary Methods). We model co-seismic earthquake deformation to exclude TGs with vertical motion above a specified threshold (Fig. 1) and use a data-driven weighting strategy aimed at reducing sensitivity to TG data contaminated by nonlinear VLM or unresolved datum errors. We apply these advances to the most recently updated altimeter data set (1993 to mid-2014) that is processed as homogeneously as possible, with each mission using consistent orbits (see Methods), with respect to the same reference frame as the GPS-derived estimates of VLM (ITRF2008; ref. 21).
|
