A space-time regression analysis of chlorofluorocarbon (CFC) data from the eastern North Pacific Ocean's subtropical thermocline (25.6–26.6 σθ) is used to extract age and velocity changes from a collection of 1980s and 1990s hydrographic sections. Results indicate substantial increases of CFC ages, from 0.2 yr yr−1 on 25.8 σθ increasing with depth up to 0.4 yr yr−1 on 26.6 σθ, and increases of meridional age gradients, implying a circulation slowdown. Using the regression-derived spatial and temporal derivatives, an isopycnal balance of advection and diffusion, including the effects of mixing on CFC ages, was applied to estimate changes in flow. Southward velocities at 20°N, 145°W generally decreased by ∼ 0.03 ± 0.02 cm s−1 yr−1 on isopycnals 25.6 to 26.4 σθ during the 1980s and 1990s, consistent with overturning changes from historical hydrographic data and with concurrent increases in AOU. The deeper (26.6 σθ) age increase was consistent with mixing effects.
 A slowdown in the North Pacific Ocean's shallow overturning during the 1980s and 1990s has been diagnosed from hydrographic data [McPhaden and Zhang, 2002]. Hindcast general circulation model simulations also indicate a slowdown during the last two decades [e.g., Deutsch et al., 2006]. These findings have been qualitatively supported by repeat occupations of hydrographic sections where Chlorofluorocarbons (CFCs) have been measured. CFCs have been used to diagnose circulation changes from changes in CFC ages, which involve estimating a calendar year when a subsurface water sample was last at the surface by comparing its CFC concentration with the history of CFCs in the atmosphere. This approach relies on the observation that sea surface CFCs are usually close to solubility equilibrium with the atmosphere. Watanabe et al.  measured substantial increases in CFC ages along 47°N, which they attributed to a ventilation slowdown during 1985–2000. In the eastern subtropics a smaller age increase was observed in reoccupations of 1980s and early 1990s CFC sections along 24°N and 152°W in 2000 and 1997, respectively [Mecking et al., 2006]. CFC age changes could also be caused by CFC undersaturation at the sea surface and by mixing in the ocean interior, however [Waugh et al., 2003]. A diagnostic model of the subtropical North Pacific, for example, indicates that much of the 1985–2000 CFC age increase can result from mixing alone [Mecking et al., 2004, 2006], so it is not clear from these repeat CFC sections that a slowdown in the circulation had occurred.
 Here we revisit the eastern subtropical North Pacific thermocline, taking advantage of all of the WOCE sections with CFC observations in this area, and include the 1997 152°W and 2000 24°N sections [Mecking et al., 2006]. These data provide unprecedented spatial and temporal CFC coverage during 1985–2000, making the eastern subtropical North Pacific ideal for employing a regression approach to extract temporal and spatial trends in CFC ages along isopycnals [Robbins and Jenkins, 1998]. The spatial and temporal structure of the CFC age distributions are then applied in an along-isopycnal advective-diffusive (AD) balance that accounts for mixing and the nonlinearity of the atmospheric history of CFCs [Doney et al., 1997]. The AD balance is used to analytically determine circulation speeds and their changes during 1985–2000.
 We compiled all WOCE CFC observations (available at http://whpo.ucsd.edu) from 10°N–30°N, 130°W–160°W (Figure 1), plus two repeat sections along ∼ 152°W in 1997 (STUD97), and along ∼ 24°N in 2000 (GS2000) [Mecking et al., 2006]. The data were distributed across three zonal (10°N in 1989, 24°N in 1985 and 2000, and 30°N in 1994), and two meridional sections (152°W in 1991 and 1997, 135°W in 1993). Temporally, the data were weighted towards 1989–1992. Potential temperature (T), Salinity (S), and CFC concentrations were linearly interpolated onto density surfaces, 25.6–26.6 σθ, in 0.2 σθ increments. From the isopycnal CFC values we computed pCFC ages using the atmospheric CFC history of Walker et al.  and saturation levels of 95% for σθ < = 25.8, 90% for 26.0–26.2 σθ, and 85% for 26.4–26.6 σθ. These saturation levels provided CFC distributions and inventories most consistent with the WOCE data in Mecking et al.'s  diagnostic model of the North Pacific subtropical thermocline.
3. Spatiotemporal Taylor Series Regression
 We quantified isopycnal gradients in CFC ages using a regression approach based on Taylor series expansions [Robbins and Jenkins, 1998]:
Here Age is CFC age, mean is average age within the region, T is sampling time, X and Y are sample longitude and latitude, m are least-squared regression coefficients determined from the isopycnal age data using multiple linear regression, and HOT are higher order terms. Considering first order terms (i.e., mean, mT, mX, and mY) only, the spatial terms capture >84% of the CFC age variability. Inclusion of time as a predictive variable improves the regression by a few %, and the second order terms improve the regression statistics (to 89–96%) significantly. Uncertainties in the time coefficients were small, indicating that the changes in CFC ages can be extracted with confidence from the scattered data set we have assembled. Inclusion of higher order terms did not improve the regression significantly.
 The first order coefficients were determinable to ∼±10%, and the second order terms to only ∼ ±20–40%. No age changes were detectable on 25.6 σθ while age changes increased with density (and depth) from 0.2 years per year on 25.8 σθ up to 0.4 years per year on 26.6 σθ (Figures 2–3). The age distributions were dominated by meridional, as opposed to zonal, gradients (Figures 2 and 4) . On all isopycnals the meridional gradient in age, , increased during 1985–2000 ( > 0). Both the increases in ages and the steepening age gradients imply a slowdown in circulation in this region. Both features are also consistent with the effects of mixing, which would cause older CFC ages to increase more than younger ones, because mixing biases in CFC ages due to curvature of the CFC history were larger in the earlier portion of the atmospheric CFC history.
4. Isopycnal Balance of Advection, Diffusion, and Mixing
 To quantify impacts of mixing on CFC ages, the space-time coefficients from the regressions were applied in a two dimensional balance of advection and diffusion [Doney et al., 1997],
where K and are along-isopycnal turbulent diffusivity and advection, respectively, and f′ and f″ are first and second temporal derivatives of the atmospheric CFC history evaluated at the apparent outcrop date (sampling year − CFC age) of the water sample. Following Robbins and Jenkins , the Taylor Series regressions yield all of the necessary spatial and temporal age information to solve for the component of velocity that is normal to the CFC age isopleths (Figure 4).
To estimate the effects of along-isopycnal mixing we assigned an isopycnal diffusivity (K) of 1000 ± 500 m2 s−1 [Ledwell et al., 1998]. The resulting velocities, oriented primarily southwards (Figure 4), decrease with increasing density from ∼ 0.6 ± 0.07 cm s−1 on 25.6 σθ to ∼0.3 ± 0.05 cm s−1 on 26.6 σθ (Figure 5). These velocities are ∼ half those calculated from climatological T and S data [Mecking et al., 2004], but agree well with absolute velocities determined at 35°N, 155°W using a variation of the Beta-spiral technique [Coats, 1981]. Using a higher diffusivity (2000 ± 500 m2 s−1) increases the upper water velocities (to 0.72 ± 0.1 cm s−1 on 25.6 σθ) and the velocity shear, but causes a discrepancy between CFC-11 and CFC-12 speeds on 26.4 and 26.6 σθ (Figure 5).
 We solved for temporal changes in · by differentiating equation (3) with respect to time. All of the necessary spatial and temporal derivatives to compute ∂ · /∂t are available from the Taylor Series regression with one exception: The third order term, (∇ · (K∇CFCage)), was not distinguishable from zero in a higher order regression. Assigning mixing of 1000 ± 500 m2 s−1, as above, on shallow (σθ = 25.6 to 26.4) isopycnals, the steepening CFC-12 age gradients indicate an average slowdown of –0.03 ± 0.02 cm s−1 yr−1 (Figure 5). The circulation changes are indistinguishable from zero on 25.8 and 26.6 σθ, even though significant age increases occurred on those surfaces. The steepening age gradients (Figure 3e) implied a slowdown on 25.6 σθ, where no increase in average age was observed (Figure 3b). A larger mixing coefficient (2000 ± 500 m2 s−1) implies about the same slowdown within uncertainties (Figure 5). Assuming instead that the sea surface was saturated with respect to the atmosphere did not significantly affect the velocities and velocity changes because these are determined primarily by the age gradients. A more vigorous overturning circulation in the 1960s and 1970s than in the 1980s and 1990s [McPhaden and Zhang, 2002], and the rapid growth rates of CFCs in earlier decades [Walker et al., 2000] could have caused decadal increases in sea surface CFC saturation, however. Assigning 15% lower saturation levels before the 1980s implied a larger, yet within the error bounds reported, slowdown on 25.6–25.8 σθ, while the deeper isopycnals were unaffected.
 The slowdown in southward transport we infer from CFCs is comparable to that diagnosed from hydrographic data. Based on geostrophic calculations, McPhaden and Zhang  computed a shallow (σθ < = 26.0) slowdown on the order of 30% comparing 1970–1977 with 1980–1989, and an additional 50% slowdown comparing the 1980s with the 1990s. Our analysis indicates that the slowdown extended to 26.4 σθ.
 While we find that recent increases in CFC ages can be substantial, they may be consistent with the impacts of moderate amounts of mixing in the ocean interior. Furthermore, increases in age gradients are to be expected in recent decades even under steady ocean circulation conditions because older ages age faster than younger ages due to mixing (Figure 3). The combined spatio-temporal regression approach [Robbins and Jenkins, 1998] and isopycnal AD balance for CFC age [Doney et al., 1997] can be used to separate the effects of mixing on CFC ages from true changes in ocean circulation. It will be useful to apply this method to other subsets of the WOCE data, and to the data that are emerging as part of the CLIVAR/CO2 Repeat Hydrography (RH) program. While the RH cruises will not provide the same coverage as WOCE, they will provide key repeat CFC sections that can be combined with existing data using the regression approach.
 Repeating our calculations using WOCE data only, i.e. not including any section repeats, yielded the same results with only slightly larger error bars, indicating that repeat sections are not critical to the approach. In our study region, CFC age isopleths are oriented zonally, so the meridional terms dominate the isopycnal balance (Figures 2 and 4). To test the utility of the approach on RH cruises, we repeated our calculations using only three sections: the 1985 24°N zonal section and the 1991 and 1997 152°W meridional sections. The resulting meridional velocities and their changes were comparable to those derived from all available data (Figure 5). While the spatiotemporal approach can be applied to RH section reoccupations, use of a pair of repeat sections alone would only be appropriate in regions where CFC age isopleths are aligned orthogonal to the axis of the cruise sections.
 CFC age changes track not only circulation changes and mixing, but also changes in outcropping. Farther to the North (40°–45°N), Mecking et al.  concluded that the 1991–1997 increases in CFC ages (and in AOU) on σθ = 26.6 were mainly due to a reduction or cessation of outcropping of this isopycnal. To evaluate this idea, we performed an analogous regression analysis of the AOU fields from 10°–30°N, 130°–160°W, and found a significant increase rate on 25.8–26.6 during 1985–2000 (Figure 6). The AOU changes, about 3% yr−1 on 25.8, 26.2 and 26.4 σθ, and 5% yr−1 on 26.0 σθ, were comparable to the CFC-diagnosed slowdown (5–10% yr−1), so neither a decrease in outcrop efficiency nor an increase in oxygen utilization rates is required to explain the AOU changes on 25.8–26.4 σθ. On 26.6 σθ, the AOU increase (1.6% yr−1) is not accompanied by a decrease in velocity and thus may support a decrease in outcropping frequency, duration, or intensity on this isopycnal [Mecking et al., 2006].
 The regression approach adopted here captures only gradual changes in space and in time. In other words, the regression terms likely do not reflect variations (e.g., ENSO) occurring on temporal and spatial scales shorter than or smaller than the scale of the data sets used, in this case about a decade and 10°–30°. The spatiotemporal regression approach requires an a-priori assignment of the intensity of isopycnal mixing. In our analysis, the temporal evolution of CFC-11 and CFC-12 were not different enough to provide independent constraints on both the flow and mixing rates. However, in the future use of emerging Sulfur Hexafluoride datasets [Bullister et al., 2006] with the CFCs show promise in simultaneously determining circulation changes and ocean mixing rates based on transient tracer data.
 Changes in CFC ages in the eastern subtropical North Pacific during 1985–2000 suggest that there has been a decrease in southward velocities for σθ < = 26.4. While dramatic changes of CFC ages in the subpolar gyre and at the subpolar-subtropical gyre boundary have been reported [Watanabe et al., 2001; Mecking et al., 2006], our study indicates that CFC age changes also provide support for a reduction in the shallow overturning circulation connecting the subtropics and the equatorial Pacific during the 1980s and 1990s [McPhaden and Zhang 2002]. Recent efforts document a rebound (since the late 1990s) in shallow Pacific meridional overturning [McPhaden and Zhang, 2004], and space-time regression analysis of CFC data from ongoing Repeat Hydrography cruises in the North and South Pacific may confirm these findings.
 Thus study was supported by the NOAA Global Carbon Cycle Program. This is PMEL contribution 2932, JISAO contribution 1386.