Detection of a Large-Scale Mass Redistribution in the Terrestrial System Since 1998

See allHide authors and affiliations

Science  02 Aug 2002:
Vol. 297, Issue 5582, pp. 831-833
DOI: 10.1126/science.1072188


Earth's dynamic oblateness (J 2) had been undergoing a decrease, according to space geodetic observations over the past 25 years, until around 1998, when it switched quite suddenly to an increasing trend that has continued to the present. The secular decrease in J 2 resulted primarily from the postglacial rebound in the mantle. The present increase, whose geophysical cause(s) are uncertain, thus signifies a large change in global mass distribution with a J 2 effect that considerably overshadows that of mantle rebound.

Earth's mean tide-free dynamic oblateness (J 2) ≡ [C– (A +B)/2]/MR 2 = 1.082627 × 10−3, where C > B Aare Earth's mean principal moments of inertia and M andR are the mean mass and radius, respectively. Satellite laser ranging (SLR) has yielded precise determination of the temporal variation in the low-degree spherical harmonic components of Earth's gravity field, beginning with the initial observations ofJ 2 change made by observing Lagoes-1 satellite orbital node accelerations (1, 2). More recent studies have extended the knowledge to higher degree zonals and examined the annual signals in the low-degree geopotential (3–5). The estimated values of theJ 2 rate have ranged from –2.5 × 10−11 year−1 to –3 × 10−11 year−1.

The extension of comprehensive solutions for low-degree geopotential zonal, static, annual, and rate terms and the 9.3- and 18.6-year ocean tide amplitudes to include data since 1997 has resulted in increasingly significant changes in the estimated J 2 rate and 18.6-year tide amplitude (4). These changes implied that the models for these terms were not accommodating the observed signal. Consequently, we estimated a time series of low-degree (maximum degree of 4) static geopotential solutions using SLR observations of 10 satellites over the period from 1979 to 2002. The inclusion of multiple orbital inclinations improves separation of the higher degree zonal components and allows recovery of the gravity coefficients over shorter time periods. All processing used the same algorithms used to develop the EGM96S satellite–only gravity model and to calibrate that model's covariance (5). The 18.6-year and much smaller 9.3-year tide amplitudes were set to the values estimated in the comprehensive solution with data from 1979 through 1997 (4). The applied 18.6-year tide amplitude of 1.41 cm has the equivalent J 2 amplitude of 1.67 × 10−10. The 18.6-year tide–J 2effect is minimized (that is, the geopotential is less oblate) when the lunar node is 0 degrees, which occurred in mid-October 1987.

Shown in Fig. 1 is the estimatedJ 2 as a function of time,J 2(t). Lageos-1 data are present throughout, and Starlette data are present from January 1980 onward. Data completeness issues precluded the use of the earlier Starlette and Lageos-1 data. Other satellites were added when launched: Ajisai from August 1986 and Lageos-2 and several other satellites from 1992 onward. TOPEX/POSEIDON (T/P), which is also tracked by the Détermination d'Orbite et Radiopositionnement Intégrés par Satellite system, was added in January 1993. The formal uncertainties shown reflect the SLR data weights derived from the calibration of the comprehensive 19-year solution and should be realistic (6).

Figure 1

J2 variation observed by SLR as compared with the IB-corrected NCEP atmosphericJ2. The atmospheric series (top) is on the left scale, and the observed series (bottom) is on the right scale. Sampling intervals are 90 days in 1979, 60 days from 1980 through 1991, and 30 days afterward. No detrending has been performed. Units are × 10−10.

Dominant in J 2(t) is a seasonal signal of amplitude 3.2 × 10−10, which is driven by meteorologic mass redistribution in the atmosphere-hydrosphere-cryosphere system (7–10). Also plotted in Fig. 1 is the atmospheric contribution calculated according to the National Center for Environmental Prediction (NCEP) reanalysis data (11), including the inverted-barometer (IB) correction (12). Subtraction of this signal and further empirical removal of the residual seasonal signals (which are attributable to the poorly known seasonal mass redistribution in the oceans and land hydrology) result in a nonatmospheric and nonseasonalJ 2(t) (Fig. 2).

Figure 2

Observed J 2, after subtraction of the IB-corrected atmospheric signal and an empirical annual term before (thin red line) and after (heavy red line) an annual filter has been applied. Error bars are the observedJ 2 uncertainties. The black line is a weighted fit to the (unfiltered) pre-1997 data. The slope is –2.8 × 10−11 year−1. The offset green line is theJ 2 implied by the Greenland plus West Antarctic ice heights derived from ERS-1 and ERS-2 altimetry data. Also shown are the J 2 value implied by the T/P uniform GSL change (blue, offset) and the J 2value when the geographic distribution of the sea height changes is considered (purple, offset). Neither sea height–derived estimate includes steric effects. Units and sampling intervals are as in Fig. 1.

A linear fit to the observed J 2 through 1996 shows a decrease in J 2 of –2.8 × 10−11 year−1 (Fig. 2). For this period, the uncertainty for the J 2 rate in the comprehensive solution (which considers the correlation with the 18.6-year tide) is 0.4 × 10−11 year−1. Despite the lack of data before 1979, the results are in excellent agreement with estimates of the J 2 rate that included those data (2). The secular drift results primarily from postglacial rebound (PGR) (2, 13, 14) in the mantle, plus various secondary contributions of climatic and anthropogenic origin (for example, reservoirs, which are an order of magnitude too small to explain the recent observations) (4,15, 16). At some time during 1997 or 1998, the trend reversed. The post-1996 points have deviated from the pre-1997 slope by about six times the uncertainties, on average, over that period. A linear fit from 1997 onward yields a rate of +2.2 × 10−11 year−1. On the basis of the comprehensive solutions, the uncertainty for this rate is ∼0.7 × 10−11 year−1; however, because of the nonlinearity in J 2(t), the slope can vary by more than the uncertainty value, depending on the period fitted. Another departure may exist around 1980, but excepting a few data points the deviation is only one to two times the uncertainties, making the importance unclear.

An increase in J 2 means a net transport of mass from high to low latitude (the nodal lines of J 2are ± 35.3° latitude). Transport of terrestrial water and/or ice mass to the oceans is one likely cause, because most of the ice mass resides in high-latitude polar caps and glaciers. As an example of the mass flux involved, imagine one fictitious scenario that would increase J 2: a uniform melting of the Greenland ice sheet, resulting in an eustatic global sea level (GSL) rise. For every 10−11 increase ofJ 2, the required loss of Greenland mass as the water volume equivalent is 102 km3, accompanied by a 0.28-mm increase in GSL (15). To overshadow PGR and produce a change comparable to that in the observed J 2rates, nearly five times as much water mass is necessary, amounting to ∼1.4 mm year−1 in additional GSL rise. This is a global change of huge proportions that has not been observed in modern times (17).

Ice height changes over Greenland have been observed with radar and laser altimetry (18, 19). However, the net ice height and implied GSL changes (about ± 0.2 mm year−1) are too small to explain the changes in the observed J 2. We calculated theJ 2(t) for the Greenland plus West Antarctica ice height estimates using ERS-1 and -2 satellite altimetry data (19) up to 81.4°N and S (Fig. 2). Again, the magnitude is too small; and if anything, the ice height variation implies a mass transport that is opposite of the observed anomaly.

Recent studies indicated an acceleration of the mass wastage of mountain glaciers (20, 21). The average loss rate for the subpolar glaciers had been ∼100 km3 of water per year before 1997, with accelerated rates in the past decade (21). For the observed J 2 rate change to be explained by additional glacier mass loss, an additional water mass loss of ∼700 km3 per year would be required. The resultant GSL increase of 2.0 mm/year over the pre-1998 rate has not been observed (22). It is unlikely that transport of terrestrial water mass to the oceans can explain theJ 2 changes; however, more recent glacier and ice height data are needed to definitively rule this out.

Mass redistribution within the oceans could cause a netJ 2 change with little or no GSL signature. The calculated J 2(t) due to the apparent GSL variation within ± 66° latitude according to the T/P GSL data (23), assuming geographically uniform mass addition, is given in Fig. 2. This calculation overestimates theJ 2 effect because much of the observed GSL variation is steric (24, 25), especially at low latitudes. Even so, the change in the GSL-implied J 2before and after the strong 1997–98 El Niño is too small to account for the J 2 observations. Also shown is the J 2 signature calculated from the actual geographic distribution of the sea surface height changes (23) (again assuming no steric contributions) after removal of an empirical annual term. The slope after 1999, when the sea surface temperature had returned to normal after the 1997–98 El Niño, is consistent and comparable with the observed J 2, suggesting that oceanic mass transport could be responsible without having a substantial signature in GSL. Further research is needed and in progress. It is intriguing to consider the possibility of a different climatic mass distribution state (17) triggered by the strong 1997–98 El Niño event. For example, the melting of polar sea ice, which has no direct effect onJ 2, can cause as-yet-unknown changes in the thermohaline circulation and structure, and therefore indirectlyJ 2, as a result of the melting of the insulating freshwater sea ice layer over the Arctic Ocean (17).

So far J 2(t) is the only gravity-time record in which we have found unequivocal evidence for the anomalous mass redistribution. The higher degree zonals, such asJ 3(t) (figs. S1 and S2) andJ 4(t) (figs. S3 and S4), do show interannual fluctuations, but it is unclear whether these data show similar systematic change. A lack of change in J 3would mean that there was no net north-south mass redistribution. Earth rotation records, both length-of-day and polar motion, are potentially useful for delineating global mass transports. However, interpretation of these records is complicated by the interannual signals, which are dominated by dynamic processes within Earth's core.

Judging from the large magnitude and relatively rapid evolution of the observed J 2 changes, one possible cause could be net material flow driven by the geodynamo in the fluid outer core and along the core-mantle boundary. There is evidence of a substantial geomagnetic jerk in 1999. Such jerks have been associated with flow acceleration in the top of the core (26,27), in addition to long-term magnetic dipole changes. Could they be related? To date, no correlation has been demonstrated between the geomagnetic observations and the observedJ 2(t). However, a review of geodynamo simulation results (28) indicates that the core models can possibly explain J 2 changes at the level of 0.1 × 10−11 year−1 to 0.5 × 10−11 year−1, depending on the modeling assumptions.

In principle, climatic general circulation models (GCMs), especially models coupling the atmosphere-ocean-cryosphere with land hydrology, can help to explain the J 2 changes. Indeed, some secondary interannual variability inJ 2 is presumably climatic (Fig. 1). However, traditional GCMs simply do not have sufficient physics or knowledge built into their climatic feedback mechanisms to anticipate sudden changes (17) such as the recent observedJ 2 changes. Further, GCMs are almost invariably too conservative and tend to underestimate the climatic variability when compared with in situ and ground truth data, because of insufficient resolution in the numerical computation (17,29–31).

Supporting Online Material

Figs. S1 to S4

  • * To whom correspondence should be addressed. E-mail: ccox{at}


View Abstract

Navigate This Article