The results of Pahlow and Riebesell were obtained after systematic corrections for offsets between reoccupied-station data. The offsets were determined at same-depth intervals in deep water, and then applied uniformly over the entire water column (1). Spencer and Mantyla (3) demonstrated that this approach is risky, at least in the North Atlantic, and instead favored evaluating nutrient or oxygen concentration relative to salinity (3). In any event, the systematic corrections applied by Pahlow and Riebesell, and uncertainties in those corrections, actually exceeded the signal in the case of phosphate and were comparable to the signal in the case of oxygen. For example, from figure 1 of (1), we estimate the systematic correction applied to phosphate to be about 0.09 ± 0.04 µM (n = 13). Moreover, although Pahlow and Riebesell corrected offsets using the more reliable modern data (e.g., World Ocean Circulation Experiment) as a reference, even within modern data the offsets and offset errors at reoccupied stations are larger than the inferred phosphate signal (Table 1). We suggest that small temporal signals in phosphate and oxygen can hardly be resolved in view of the large, systematic uncertainties in the data and derived corrections.

Table 1. Summary of absolute offsets for phosphate and oxygen in the deep ocean at crossover stations. South Pacific data were compiled from figure 5.8 of (5); Pacific data from (4). Location Years Phosphate (µmol kg1) Oxygen (µmol kg1) South Pacific 1961-1972 0.1 ± 0.2 3 ± 7 South Pacific 1972-1993 0.05 ± 0.08 1 ± 4 Pacific 1985-1996 0.03 ± 0.02 2 ± 1

A number of studies have reported discrepancies in nutrient and oxygen data collected by different laboratories on different cruises (4-6). Two types of systematic errors must be taken into account in constructing calibration curves. The first is an intercept offset that is related to the background signal and is independent of sample concentration. The second is an error in slope that is related to the accuracy of standards and is proportional to sample concentration. In the absence of certified reference materials for nutrient analyses, it is customary in oceanography for individual laboratories to prepare their own calibration standards. Because of errors involved in standard preparation, the second type of systematic error, slope error, tends to dominate the total systematic error (7) and therefore to account for most of the discrepancies observed in modern data (Table 1).

The constant-offset correction that was determined in deep water and applied over the entire water column by Pahlow and Riebesell (1) would necessarily undercompensate or overcompensate for the actual systematic error in different parts of the water column. In the equatorial Atlantic, for example, applying a correction for slope errors instead of an offset correction would eliminate much of the reported temporal signal (8). In light of these questions regarding systematic-error correction, we conclude that, although Redfield ratios in the oceans may indeed change with time, the changes inferred by Pahlow and Riebesell are probably not valid.

Jia-Zhong Zhang
Cooperative Institute for Marine
and Atmospheric Studies
Rosenstiel School of Marine
and Atmospheric Science
University of Miami
Miami, FL 33149, USA
Calvin W. Mordy
Joint Institute for the Study
of Atmosphere and Ocean
University of Washington
Seattle, WA 98195, USA
Louis I. Gordon
Andy Ross
College of Oceanic
and Atmospheric Sciences
Oregon State University
Corvallis, OR 97331, USA
Hernan E. Garcia
Scripps Institution of Oceanography
University of California at San Diego
La Jolla, CA 92093, USA

REFERENCES AND NOTES

1. M. Pahlow and U. Riebesell, Science 287, 831 (2000) [Abstract/Free Full Text] .
2. Phosphate was back-calculated using an average AOU of 120 mol m² from 500 to 2000 m, with a P:AOU ratio of 1:170 and a (N:P) of 1.9 ± 1.2 per mil per year over 14 years. Similarly, AOU in the North Pacific was 500 mol m² from 500 to 3000 m with a (AOU) of 0.7 ± 0.1 per mil per year over 14 years.
3. D. Spencer and A. Mantyla, in GEOSECS Atlantic Expedition, Hydrographic Data, 1972-1973, A. E. Bainbridge, Ed. (National Science Foundation, Washington, DC, 1981), vol. 1, chap. 1, p. 5.
4. C. W. Mordy et al., EOS. Trans. Am. Geophys. Union 80 (suppl.), OS43 (1999); http://geoweb.princeton.edu/bigscience/GLODAP/details.html#Redfield.
5. V. Gouretski and K. Jancke, World Ocean Circulation Experiment Rep. No. 143/96 (WOCE Hydrographic Program-Special Analysis Centre, Hamburg, Germany, 1996).
6. A. Ross et al., EOS Trans. Am. Geophys. Union 80 (suppl.), OS4 (1999); http://ucs.orst.edu/~rossan/webversion.htm.
7. L. I. Gordon et al., EOS Trans. Am. Geophys. Union 80 (suppl.), OS45 (1999).
8. From the correction range of Pahlow and Riebesell (1) to the nutrient maximum (approximately 1800 m to 800 m), phosphate typically increases by about 0.8 µM in the Equatorial Atlantic. Calibration errors of about 3% [deep offsets in figure 1 of (1) are about 7% of deep phosphate concentrations, and errors in Table 1 are about 1 to 4% of concentrations in the deep Pacific] result in an underestimation of systematic errors in (1) by 0.02 µM phosphate in the nutrient maximum. Over the entire comparison range in (1), calibration errors are roughly half of those in the nutrient maximum, or >90% of the average phosphate trend implied by the results of Pahlow and Riebesell.
9. This work was funded by the NOAA and the NSF, and was carried out under the auspices of the Cooperative Institute for Marine and Atmospheric Studies (a joint institute of the University of Miami and NOAA) and the Joint Institute for the Study of Atmosphere and Ocean (a joint institute of the University of Washington and NOAA) under cooperative agreements NA67RJ0149 and NA67RJ0155.

Response: Zhang et al. use the temporal trends we observed in AOU and oxidative ratios to derive corresponding trends in absolute phosphate and oxygen concentrations, and on that basis raise two main points of criticism of the work reported by Pahlow and Riebesell (1). First, they maintain that the methodological uncertainties exceed the derived trend in phosphate concentration. And, second, they suggest that a correction for slope errors, rather than the offset correction that we applied, would eliminate much of the observed temporal signal. Below, we address each of these points and show that the concerns of Zhang et al. are unfounded.

Zhang et al. estimate a change in phosphate concentration of 0.01 ± 0.01 µM over 14 years, based on an average North Atlantic AOU of 120 mol m². The correct value for the change in phosphate concentration over that time span in our analysis, however, is actually 0.04 ± 0.04 µM (based on a mean phosphate concentration of 1.5 µM, an observed North Atlantic AOU trend of 0.3 ± 0.8 per mil per year, and an observed North Atlantic AOU:P trend of 1.7 ± 1.2 per mil per year). In either case, confidence intervals for changes in absolute nutrient concentrations are indeed similar to or larger than the changes themselves, which precluded the use of absolute concentrations in our analysis. Much of the uncertainty, however, is due to natural variability in AOU and nutrient concentrations between stations. The strong co-variation between nutrients and AOU means that the use of nutrient and oxidative ratios avoids station-to-station variability in absolute concentrations. Hence, our study's approach was to test for temporal changes in nutrient ratios rather than in absolute concentrations. Zhang et al. also suggest that nutrient and oxidative ratios should be evaluated relative to salinity rather than depth; we indeed pursued such an evaluation early in our analysis and achieved identical results.

The statement by Zhang et al. that our systematic corrections were greater than or similar to the reported mean signals does not apply to our data analysis. Our correction procedure was not performed on the basin-wide mean temporal trends, as implied by Zhang et al. Instead, we applied systematic corrections for each set of reoccupied stations individually. These individual corrections were smaller than the respective signals in almost all cases [e.g., figure 1 of (1)].

Zhang et al. point out that our analysis did not correct for systematic slope errors arising from calibration differences between cruises. Based on a difference in phosphate concentration of 0.8 µM between our correction range and the maximum concentration in the example profile that we published [figure 1 of (1)], Zhang et al. suggest that the slope error could explain 90% of the observed North Atlantic phosphate change (which, as noted, they incorrectly estimate at 0.01 ± 0.01 µM) over the 14 years.

A closer examination, however, suggests that correcting for slope error would not materially change the results of our analysis. The average difference between maximum phosphate concentrations and those in the correction range was only 0.3 µM, because most locations used in our analysis did not include data as close to the nutrient maximum as shown in figure 1 of (1), which was chosen to illustrate the correction procedure but not to be representative of all profiles. As we explained [note 16 of (1)], the expected error (E_n) of the mean of n measurements is E_n = E₁/n^1/2, where E₁ is the error of a single measurement. Our analysis for the North Atlantic comprised 20 combinations of different cruises (n = 20). The deviation due to a slope error of 3%, the figure cited in note 8 of the comment by Zhang et al., would therefore be (0.3 µM)(0.03)/ = 0.002 µM. For a change in phosphate concentration of 0.04 µM, therefore, the estimated impact of the slope error would not exceed 0.002/0.04, or 5% of the trend.

Markus Pahlow^*
Ulf Riebesell
Alfred Wegener Institute for
Polar and Marine Research
Post Office Box 120161
D-27515 Bremerhaven, Germany
^* Present address: Bedford Institute of Oceanography
Dartmouth, N.S. B2Y 4A2, Canada

REFERENCES

1. M. Pahlow and U. Riebesell, Science 287, 831 (2000) .