The objective of the TOPEX/POSEIDON (T/P) mission is to measure sea level with an accuracy of a few centimeters for temporal scales of one month or longer and distances of hundreds to thousands of kilometers. It is these larger scales which are the most difficult to observe and yet which bear most significantly on global change. With more than one year of data collected, the T/P goal appears to have been achieved; altimeter errors, including environmental corrections, are estimated to be only 5 cm for single-pass sea level measurements (Fu et al., 1994). This does not include ocean tide model corrections which have uncertainties of comparable magnitude, but the T/P data are enabling these errors to be addressed explicitly, e.g. Wagner et al., (1994a). In this paper, we investigate the extent to which simple averaging in time and space can further reduce the T/P measurement error. By avoiding the usual orbit adjustment process, this approach has the potential of recovering the complete sea level signal on all relevant spatial scales. Based on comparisons with tide gauge data and in situ observations of dynamic height, we show that monthly mean sea level, averaged over spatial scales of a few hundred km, can be determined from T/P data with an accuracy of 2 cm.

Figure 1. Typical T/P sea level time series for the 4-deg longitude x 1-deg latitude cell centered at 8S, sampled 4 times every 10 days. Each plus symbol represents one altimeter pass, where the sea height has been averaged along a 1-deg segment of the satellite track. The solid line is the monthly average. 4x1 monthly means form the basis for all in situ comparisons in this paper.
Statistics generated by the global T/P collinear differences provide an indication of the overall altimeter system precision. In Figure 2a, we show the rms difference (based on the 1-s samples) of each pass from 60N to 60S relative to its collinear partner in cycle 18. The altimeter data are fully corrected as summarized above, but no orbit adjustment has been made. The rms difference for all 12,400 passes together is 14.2 cm, a remarkably small value considering that it includes errors of the altimeter and tide model plus the complete (non-tidal) sea level variability signal of the global oceans over a 16-month period. A number of outliers exist, but virtually all of these are short passes in areas of high natural variability, such as the Agulhas Current, i.e. they represent ocean signal rather than noise.

Figure 2. (a) The rms collinear difference of each pass of T/P data, from 60N to 60S, for cycles 1-51 with respect to cycle 18. (b) Solid line: the same statistics as above, collected in 10-day cycles; selected cycles are labeled. The minimum of 8 cm is the crossover difference value of cycle 18 with respect to itself. Dashed line: same statistics when the additional tide corrections of Wagner et al. (1994a) are not included. The distinct 60-day periodicity is aliasing of the M2 and S2 terms of the Cartwright and Ray (1991) tide model.
It is also apparent from this scatter plot that the variability grows as a function of increasing time between the collinear passes. This is illustrated more clearly by the solid curve in Figure 2b, in which the variability has been computed for each 10-day cycle (still relative to cycle 18). Because collinear differences of cycle 18 cannot be formed with itself, we have substituted the cycle 18 global rms crossover difference value, 8 cm. From this minimum, the variability increases symmetrically, reaching 12-13 cm when the time difference is one month. Thereafter, the increase is more gradual, levelling off at 16 cm for passes 6 months apart. Beyond 6 months, the variability decreases, an indication of the seasonal cycle of sea level change in the ocean (see sections 5 and 6).
The dashed curve in Figure 2b demonstrates the sensitivity of T/P data to the ocean tide model used. Recall that in our basic analysis (the solid curve), the Cartwright and Ray (1991) ocean tide model was used together with corrections to the first four major tidal constituents derived from TOPEX data by Wagner et al. (1994a). The dashed curve shows the result when the Wagner corrections are not included. The distinct 60-day periodicity is an indication of M2 and S2 error in the Cartwright and Ray model (Wagner et al., 1994b). (The 9.916-day repeat period of the satellite track aliases the M2 and S2 tides into 62.1-day and 58.7-day periods, respectively. Errors of the K1 and O1 constituents, with aliases at 173.2 days and 45.7 days, are not apparent due to the relatively long period of the former and the small amplitude of the latter.) Wagner's corrections reduce the overall T/P global variability from 15.4 to 14.2 cm, indicating that the net tidal error removed had an rms amplitude of 6 cm. Such tide model improvement is critical for achievement of the T/P goals.
Maps in Figure 3 summarize comparisons between the T/P data and 16-month records from 50 tide gauges and 36 moorings, all located between 30o N and 30o S.

Figure 3. Based on 16 months of data, comparison of T/P monthly means with sea level from 50 tide gauges (dots) and dynamic height from 36 thermistor moorings (circles). Symbols indicate locations of the 4x1 altimeter cells used, making the few coastal gauges used to appear offshore. Best agreement is within 10 degrees of the equator where rms differences less than 3 cm and correlations greater than 0.8 are common.
The rms difference and correlation distributions tend to be symmetric about the equator with best agreement found in a band from 10S to 10N and gradual degradation toward higher latitudes. We do not interpret this to mean that the altimeter errors grow as a function of latitude, but rather that the decreasing scales of variability away from the equator cause the in situ point measurements to be less representative of the 4x1 altimeter averages. Slowly-propagating ocean signals at certain mid-latitude locations can also introduce temporal lags between the altimeter and tide gauge time series; for example, Mitchum (1994) shows that at Wake Island, better correlations can be obtained if Rossby wave propagation speeds are assumed. It seems clear that for quantitative comparisons at the level of a few centimeters, the altimeter 4x1 averages have no in situ counterpart outside of the tropics .
Figure 4 shows the results only for locations within 10 degrees of the equator where the in situ data can be most confidently used to evaluate the altimeter accuracy. Separate scatter plots indicate that the T/P data agree better with the island tide gauge records than with TAO dynamic heights: the 17 gauges yield an rms difference of 2.2 cm and correlation of 0.88, while the corresponding values for the 36 moorings are 3.1 cm and 0.80 correlation. This result is consistent with the larger errors expected of the TAO dynamic heights.

Figure 4. Scatter plots of all T/P comparisons within the equatorial region, 10N to 10S. (a) 17 island tide gauges yield a tight envelope with rms difference of only 2.2 cm and correlation of 0.88. (b) 36 thermistor moorings show more scatter with rms difference of 3.1 cm and correlation of 0.80.
Figure 5 presents time series at 10 of the island locations where the rms differences range from only 1.1 to 2.0 cm. The existence of this many high-quality comparisons presents a strong case that the T/P 4x1 monthly means are accurate at the level of 2 cm.

Figure 5. Time series from 10 island gauges (thin line) within 10 degrees of the equator at which particularly good agreement was found with T/P data (heavy line). The existence of so many examples such as this argues that T/P is achieving accuracies at the 2 cm level for monthly mean heights in 4x1 cells, without orbit adjustment.

Figure 6. Annual phase and amplitude derived from a harmonic analysis of T/P time series. Phases are in terms of the month of maximum sea level, with the second half of the calendar year shaded. Amplitudes are in cm, greater than 6 cm shaded. Click here to see this figure in color.
Figure 7 compares the patterns of large-scale sea level change seen by T/P with the Levitus (1982) climatology for (a) 20-deg latitude bins, and (b) northern and southern hemispheres. In computing averages, both altimeter and dynamic heights (relative to 1500 dbar) were normalized by the cosine of latitude to account for the fact that neither data set is based on an equal-area grid. In the northern hemisphere, the similarity between the altimetry and climatology is quite remarkable, especially considering that it is a comparison between a 1-year snapshot and an average over many decades. The phases are virtually the same, and the amplitude of the altimeter signal is only about 2 cm larger than the climatology. In the southern hemisphere, however, the agreement is considerably worse. In this case, the dynamic height is larger than the altimeter height by about 1 cm, but the main discrepancy is the phase, which differs by as much as 3 months. Of course, in common with most in situ oceanographic observations in the southern hemisphere, the Levitus data set is quite sparse here; it is therefore likely that T/P is giving the more reliable view. As evidence of this, note that the dynamic height signal is symmetric about the equator, whereas the altimeter signal is significantly larger in the northern hemisphere than in the south, in agreement with the seasonal cycle of sea surface temperature (Shea et al., 1992.)

Figure 7. Variation in sea level from T/P (heavy line) and the Levitus (1982) climatology of dynamic height relative to 1500 dbar (dashed line). Results are shown for (a) global 20-deg latitude bands and (b) the northern and southern hemispheres. Profiles are offset by 10 cm.
Figure 8 shows the globally-averaged (60N to 60S) variation of sea level from T/P data; within these latitude limits, the northern hemisphere ocean makes up 40 percent of the global ocean by area. Excluding the first two months, the curve has a 3-cm peak-to-peak amplitude and phase similar to the northern hemisphere. The smaller undulations (maximum in April, minimum in November) are contributed by the southern hemisphere. Given that satellite altimeters are unique in their ability to collect sea level observations over the entire global ocean, there are no independent data capable of estimating the accuracy of this record. If the 2-cm error for 4x1 monthly means were completely random, the global average would have sub-millimeter accuracy, but a more sophisticated approach to the global error analysis is obviously required. Nevertheless, the T/P data hold great promise for a more reliable determination of the rate of global sea level change.

Figure 8. Variation in global sea level from T/P. Measurements during the first 2 months may not be reliable due to satellite pointing errors which were corrected in December 1992 (Fu et al., 1994). Thereafter a regular seasonal cycle is evident with phase similar to the northern hemisphere (see Figure 7).
It is more difficult to estimate the accuracy of the larger-scale T/P observations because existing in situ data are inadequate, particularly in the southern hemisphere. In the northern hemisphere, the seasonal cycle described by T/P agrees quite well with the best independent data set, the dynamic height climatology of Levitus (1982); for global averages in 20-deg latitude bands, phases are virtually identical and amplitudes differ by 2 cm or less. When expressed as maps of annual harmonics, the T/P data also present a picture that is consistent with seasonal variability of the tropical trade winds together with large-scale heating and cooling at higher latitudes. Altimeter observations of the quality of T/P should ultimately be capable of addressing the issue of global sea level change, although years of data will be required.
In the field of satellite altimetry, it is typical for journal articles to be dominated by discussions of techniques at the expense of science. T/P may change this. The accuracy of these data on both small and large scales is unprecedented. Furthermore, only minimal processing is required. Except for simple differencing, binning, and averaging, our results are based on sea heights taken directly from the geophysical data records (with the exception of the ocean tide model). No special knowledge in areas of geodesy, tropospheric modelling, or orbit mechanics was needed. Availability of such a simple, yet reliable, data set for the global oceans will encourage their use by a broad range of scientists.
Callahan, P.S., TOPEX/POSEIDON project GDR users handbook, JPL D-8944 (internal document), rev A, Jet Propulsion Laboratory, Pasadena, CA, 84 pp., 1993.
Carton, J.A. and E.J. Katz, Estimates of the zonal slope and seasonal transport of the Atlantic north equatorial countercurrent, J. Geophys. Res., 95 (C3), 3091-3100, 1990.
Cartwright, D.E. and R.D. Ray, Energetics of global ocean tides from Geosat altimetry, J. Geophys. Res., 96 (C9), 16897-16912, 1991.
Cheney, R.E., J.G. Marsh, and B.D. Beckley, Global mesoscale variability from collinear tracks of Seasat altimeter data, J. Geophys. Res., 88 (C7), 4343-4354, 1983.
Cheney, R.E., W.J. Emery, B.J. Haines, and F. Wentz, Recent improvements in Geosat altimeter data, EOS, 72 (51), 577-580, 1991.
Douglas, B.C., Global sea level rise, J. Geophys. Res., 96 (C4), 6981-6992, 1991.
Fu, L.-L., On the wavenumber spectrum of the oceanic mesoscale variability observed by the Seasat altimeter, J. Geophys. Res., 88, 4331-4341, 1983.
Fu, L.-L., E. Christensen, M. Lefebvre, and Y. Menard, TOPEX/POSEIDON mission overview, J. Geophys. Res., this issue, 1994.
Miller L., Cheney, R. and J. Lillibridge, Blending ERS-1 altimetry and tide-gauge data, EOS, 74 (16), 185/197, 1993.
Mitchum, G.T., Comparison of TOPEX sea surface heights and tide gauge sea levels, J. Geophys. Res., this issue, 1994.
Levitus, S., Climatological atlas of the world ocean, NOAA Professional Paper 13, 173 pp., 1982.
Lillibridge, J.L. and R.E. Cheney, The 1991-93 Los Ninos from ERS-1 altimetry, EOS, 74 (16), 95, 1993.
Luther, D.S., Observations of long period waves in the tropical oceans and atmosphere, Woods Hole Oceanographic Institution report 80-17, Woods Hole, MA, 1980.
Picot, N. and P. Escudier, Altimeter ionospheric correction from DORIS, J. Geophys. Res., (this issue), 1994.
Rapp, R.H. and Y. Yi, A TOPEX/POSEIDON reference ground track, T/P Research News, 2, 15- 17, Jet Propulsion Lab, Pasadena, CA, 1994.
Shea, D.J., K.E. Trenberth, and R.W. Reynolds, A global monthly sea surface temperature climatology, J. Climate, 6, 987-1001, 1992.
Wagner, C.A. and R.E. Cheney, Global sea level change from satellite altimetry, J. Geophys. Res., 97, (C10), 15607-15615, 1992.
Wagner, C.A. and C.K. Tai, Degradation of ocean signals in satellite altimetry due to orbit error removal processes, J. Geophys. Res., (in press), 1994.
Wagner, C.A., C.K. Tai, and J.M. Kuhn, Improved M2 ocean tide from TOPEX/POSEIDON and Geosat altimetry, J. Geophys. Res., (this issue), 1994a.
Wagner, C.A., C.K. Tai, and J.M. Kuhn, Temporal variability of global TOPEX/POSEIDON altimetry: Tide and annual signals, NOAA Tech. Memo. OES-12, 1994b.
White, W.B, G. Meyers, J.R. Donguy, and S. Pazan, Short-term variability in the thermal structure of the Pacific Ocean during 1979-82. J. Phys. Oceanogr., 15, 917-935, 1985.
Wyrtki, K. and S. Nakahara, Monthly maps of sea level anomalies in the Pacific, 1975-1981, Tech. Rep. HIG-84-3, Univ. Hawaii, Honolulu, 1984.