Warren's  “Why is no deep water formed in the North Pacific” is revisited. His box model of the northern North Pacific is used with updated estimates of oceanic volume transports and boundary freshwater fluxes derived from the most recent data sets, using diverse methods. Estimates of the reliability of the result and its sensitivity to error in the data are given, which show that the uncertainty is dominated by the large observational error in the freshwater fluxes, especially the precipitation rate. Consistent with Warren's conclusions, it is found that the subpolar Atlantic-Pacific salinity contrast is primarily explained by the small circulation exchange between the subpolar and subtropical gyres, and by the local excess of precipitation over evaporation in the northern North Pacific. However, unlike Warren, we attribute the latter excess to atmospheric water vapor transports, in particular the northern moisture flux associated with the Asian Monsoon. Thus the absence of such a large transport over the subpolar North Atlantic may partly explain why it is so salty, and why deep water can form there and not in the North Pacific.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 “A notable and well-known asymmetry in the deep circulation is that sinking of surface water to great depth in high northern latitudes occurs only in the North Atlantic, and not at all in the North Pacific” [Warren, 1983] (hereinafter referred to as BW). This thermohaline circulation pattern leads, in the Northern Hemisphere, to large differences between the Atlantic and the Pacific oceans, especially concerning the mode of meridional heat transport. Talley  estimates that intermediate and deep water formation are associated with at least 75% of the meridional heat transport in the North Atlantic, whereas in the North Pacific it is the shallow gyre overturning that accounts for 75% of this transport. The oceanic meridional heat transport is a significant component of the climate system, and it has been hypothesized in a large number of studies to play a key role in climate change on decadal to millenial timescales, from the onset of the Little Ice Age [Weyl, 1968] to glacial-interglacial transitions [Broecker and Denton, 1989]. Whether these theories are correct or not, improved understanding of the climate system requires a better knowledge of how the meridional heat transport is determined.
 In attempts to explain this asymmetry, BW posed the central question: “Why is no deep water formed in the North Pacific?” For this problem, the critical variable is the potential density contrast between surface and deep waters. BW related this to the low salinity of the northern North Pacific surface water (32.8‰ on annual mean), “so much lower than the underlying deep water salinities (34.6‰) that even when the surface temperature is reduced to the freezing point,… the surface water cannot be made dense enough to sink very deep (wintertime convection seems to be limited to the upper 150 m, as noted by Reid ).” In contrast, the mean surface salinities of the northern North Atlantic (34.7‰ between 45°N and 60°N) are much closer to the deep water values (34.9‰), allowing for sinking at a sufficiently low surface temperature.
 Three answers have been offered to BW's question. First, the contrast between the two oceans may be related to the fact that the Atlantic is open to the Arctic whereas the Pacific is largely closed by the Bering Strait and the Aleutians at about 62°N. However, as BW points out, deep water is formed in the Labrador Sea, close to 60°N. In addition, the connection to the Arctic, if anything, would tend to make the North Atlantic fresher. Second, in the North Atlantic, the high salinity of the near-surface layer may be accounted for by the weak freshwater input, due to a much greater evaporation rate than in the North Pacific. This is attributed to the wind-stress curl pattern, which induces northward flow of warm subtropical water in the North Atlantic, whereas in the North Pacific these waters flow largely zonally within the Kuroshio current. Thereby isolated from subtropical water, the northern North Pacific upper layer is eventually freshened by the combined effects of the upwelling, which, while bringing up salty water, also cools the SST and reduces the evaporation, with the latter effect dominating in the salinity budget. This was BW's explanation. Third, the high salinity of the northern North Atlantic upper layer may be more directly due to the high salinity of the Tropical Atlantic, through direct salt advection by the water masses. This salinity is related to atmospheric freshwater fluxes across the Isthmus of Panama [Weyl, 1968; Zaucker et al., 1994]. In parallel, the export of water out of the Atlantic basin may be freshening the Tropical Pacific, and consequently the northern North Pacific, through a similar advective mechanism.
 A motivation for this study is the growing evidence from paleoceanographic studies that the distribution of several water column properties has fluctuated in the North Pacific on time scales of thousands to hundreds of thousands of years. Such changes have been inferred from sediment cores using proxy records of concentrations of either nutrients [Boyle, 1992, and references therein; van Geen et al., 1996; Keigwin, 1998] or oxygen [Dean et al., 1989; Behl and Kennett, 1996; Zheng et al., 2000] in overlying bottom waters. Various explanations have been proposed to account for these observations, including the possibility of deeper and/or more intense convection at higher latitudes of the North Pacific. Variations in surface water productivity changes, and their repercussions at depth, provide an alternative mechanism that cannot be ruled out [Wyrtki, 1962; Mix et al., 1999]. The notion of climatically sensitive and fairly rapid convection at intermediate depths of the North Pacific has been supported by observations of the penetration of man-made and natural tracers [Van Scoy and Druffel, 1993; Warner et al., 1996; Reid, 1997] and a time series of hydrographic sections across the basin spanning several decades [Wong et al., 1999]. The objective of this paper is to show with a simple box model that ventilation of the North Pacific is controlled by the transport of water vapor into the region.
 The salinity distribution is the result of the balance of oceanic volume transports and boundary freshwater fluxes, i.e., precipitation, evaporation, and continental river runoff. As shown by BW, the simplest way to represent this equilibrium is to consider a steady-state box model in which the freshwater fluxes are balanced by horizontal and vertical advective changes in the salinity of the water passing through the near-surface layer. Here we follow his formulation of the problem, and repeat his calculations with updated estimates of the boundary fluxes involved in the freshwater and salt budgets (section 2). Results and sensitivity analysis are given in section 3 that highlight the importance of moisture convergence over the northern North Pacific. Discussion follows in section 4, and conclusions are given in section 5.
2. A Box Model for the Surface Waters of the Northern North Pacific
 Our study will follow BW and use a box model of the northern North Pacific near-surface layer (hereinafter nNP) to investigate the processes that control its salinity. The box-model formulation is believed to be sufficiently robust for this portion of the ocean since the spatial variability of the salinity field does not exceed 0.5‰. This can be seen in Figure 1. The weak annual variability (less than 0.1‰, using the World Ocean Atlas of Conkright et al. ) allows us to consider the system in a steady state for the current equilibrium. On decadal timescales, however, it has been shown to vary significantly: Using hydrological sections along 47°N, Wong et al.  inferred a maximum freshening of 0.1‰ in the source region of NPIW over approximately 22 years. As we shall see, this is well below the accuracy permitted by such a crude representation, and will hence be neglected.
 The nNP surface layer is taken to be 200 m deep, so that it includes the well-mixed layer in winter. The northern boundary is set at about 60°N by the bathymetry, and the lateral boundaries are the coastlines, between longitudes 143°E and 125°W. The case of the southern boundary is more delicate. To simplify the circulation scheme, BW made this boundary correspond with the zero wind-stress curl line (roughly the boundary between the subtropical and the subpolar gyre), so that the meridional component of the barotropic Sverdrup transport equals zero across this boundary. The wind-stress curl derived from the Atlas of Surface Marine Data is presented in Figure 2. Clearly, the zero wind-stress curl line is not zonal, but if the southern boundary is set at 43°N, the positive and negative areas across the latitude circle may cancel each other. As we shall see later, this is of little importance in our case, for we will estimate the transport across 43°N directly.
 Within these boundaries, the average salinity of the nNP box is then Ss ≃ 33.07‰, using work of Levitus and Boyer , compared with 34.7‰ at 3000 m. In the same latitude domain of the North Atlantic (the northern North Atlantic, hereinafter referred to as nNA), the surface layer salinity is 34.94‰, and about 35‰ at depth.
2.2. Freshwater and Salt Budget for the Surface Waters of the Northern North Pacific
 Here we will briefly review BW's formulation, using the same notation. Consider the previously described nNP box of uniform salinity Ss. The boundary fluxes are defined as follows, with the convention that any flux leaving the box is negative:
 • The outflow through the Bering Strait VBS carries water of salinity of Ss.
 • The Ekman transport occurs across the southern boundary VE. Since the zonal component of the wind stress is eastward along this line, this flow is southward, hence also transporting water with salinity Ss. However, Sverdrup theory [Sverdrup, 1947] requires the following balance in the interior:
where H is the depth of the level of no motion, typically 2000 m. Here is the wind-stress vector, and is the local vertical unit vector. This transport vanishes along the zero wind-stress curl line, leading to
where VE is given by and VG = ∫−H0vGdz, vG being the geostrophic meridional velocity. However, while all the Ekman transport occurs in the upper layer, the geostrophic flow is distributed over the whole water column and only a fraction α actually occurs in the upper 200 m (from 0 to −h),
BW assumed it to be small, namely α ∼ 1/3, and we will further show that this estimate lies within an acceptable range of values. Note that the assumption of zero Sverdrup transport (equation (2)) is no longer required here, since we will compute the geostrophic transport independently from the Ekman transport, but the notation will be kept for easier comparison with BW's work. The Sverdrup balance (equation (1)), together with our sign convention, thus yields a net outflow in the upper 200 m at the southern boundary of ∫xwxe (1 − α)VE dx. However, the relevant quantity for the salt budget is the advective, cross-gyre salt transport MS. Thus we must not restrict ourselves to a zonally averaged salt transport, for the sign of the integrand at each longitude determines whether subtropical water with salinity SWB (see below) is flowing in, or whether nNP water is being carried out. In contrast with BW, we thus compute the local VE + as a function of longitude, with even grid spacing. We then group the positive terms in V+ and the negative terms in V−, such that the total mass transport is V = V+ − V−. The meridional, interior salt transport now takes the form
which is a linear function of Ss.
 • The integrated upwelling Vu is produced by the large cyclonic atmospheric circulation, and carrying a salinity Su.
 • The atmospheric freshwater flux F is the difference between the area-integrated precipitation P, and evaporation E, over the nNP box.
 • The river discharge R flows into the northern North Pacific basin. The total freshwater input is then R + F.
 • The “western-boundary current” transport VWB actually groups together the effect of ageostrophic interior flow and nonlinear boundary-layer flow. Its sign is uncertain: Observations of surface currents, especially the Oyashio, suggest a southward, negative VWB, that would carry waters out of the box with salinity Ss. However, as BW argues, there could be an inflow from the subtropical gyre, in this case an extension of the Kuroshio, which “strikes roughly east northeastward after separating from the western boundary farther south and some small part of it may enter the South West portion of the box (VWB northward, positive).” The inflow would then carry water with subtropical salinity SWB. In the model, VWB is calculated as a residual from the mass conservation equation, since it is the most ill-constrained flux.
 • In addition, we now have a better, though largely imperfect, knowledge of the eddy salt fluxes. Using the classical turbulent closure, they can be expressed as the gradients of large-scale quantities. Let us note and the meridional and vertical salt flux and the southern and lower boundaries of the box, respectively. Then,
We make the assumption that these do not transport freshwater (in the sense of Wijffels et al. ), but only contribute to the salt budget.
 The model consists of two conservation equations: (1) mass conservation, where the steady-state assumption together with the continuity equation leads to the following mass balance:
and (2) salt conservation, where the steady-state assumption, together with our parameterization of eddy salt fluxes, reduces the usual advection-diffusion equation to
if VWB is northward, and
if VWB is southward.
2.3. Updated Estimates of the Boundary Fluxes
 The fluxes were computed from some of the most recent sources, using a 1° × 1° regridding for VE, F, and Vu. All the data presented here are annual means.
VBS, the outflow through the Bering Strait, is now estimated by Roach et al.  to be 0.83 Sv ± 30% (1 Sv = 106/m3 s−1), from moored current meters and direct property measurements over the water column. This result is virtually identical to that of Coachman and Aagaard  used by BW (0.8 Sv). Here we have ignored the temporal variations of this transport and simply taken the annual mean, which is northward.
R, the continental runoff, was computed with the Global Runoff Data Center (GRDC) web-based database (available at http://www.grdc.sr.unh.edu) [Fekece et al., 2000]. This raised several problems: (1) Not all the rivers belonging to the Pacific catchment basin are listed in this database, but the contribution of the remaining (smaller) rivers is believed to be negligible. (2) The gauging stations are not always at the mouth of the river, so we took for each one the river discharge time series of the closest station to the ocean; (3) Although the sampling time was not the same for all stations, we performed time averages over the complete time series for each station; this may bias the total river runoff, taken to be the sum of the individual mean river discharges. (4) We made the assumption that all the rivers listed as part of the Pacific catchment basin between 43°N and 60°N were entirely contributing to R, i.e., directly flowing into the nNP box. This might not be the case for the Amur, for example, because the sea of Okhostk is not included in the box we are considering. We find a value of 0.03 Sv, which is likely to be an upper limit, but it might compensate for the problem that not all rivers are accounted for. This error should be small, however, for the five largest rivers (Amur, Yukon, Columbia, Fraser, and Stikine) already account for 87% of this flux. Thus we expect the 11 river flows used in this calculation to capture at least 90% of the real runoff; that is, the error bar should be about 10%. As a comparison, BW used the value of 0.056 Sv, derived from Baumgartner and Reichel , which computed the runoff as the residual P-E over land, probably a more unreliable approach.
 The value of the net atmospheric freshwater flux F is not well-known, due to the large uncertainty in the precipitation fields. The error is reported to be about 30% in this region for the NCEP CPC Merged Analysis of Precipitation (CMAP) [Xie and Arkin, 1996], considered as the best product, according to Trenberth and Guillemot . The evaporation rate calculated with the bulk aerodynamic formula is believed to be more accurate, though still subject to considerable uncertainty. In Table 1 we present results extracted from various data sets.
Table 1. Atmospheric Freshwater Fluxes Integrated Over the Northern North Pacific (nNP) and Northern North Atlantic (nNA)(Sv)
 Integrating the CMAP precipitation field over the box yielded P = 0.45 ± 0.15 Sv. This is the value we will retain for our sensitivity tests. It is noteworthy that the NCEP/NCAR Reanalysis and the da Silva et al.  Atlas of Surface Marine Data (ASMD) yield similar, although largely independent, results for the precipitation field. For the integrated evaporation rate we will take E = 0.18 Sv (compare Table 1), with an error bar of 20%, as done in the ASMD data set [da Silva et al., 1994]. Thus F = P − E = 0.29 ± 0.15 Sv over our box, almost 3 times higher than the Baumgartner and Reichel  estimate used by BW (0.095 Sv), but very close to the freshwater divergence calculated by Roemmich and McCallister  between 47°N and the Bering strait (0.27 Sv). This introduces a tendency to freshen the nNP relative to BW's calculations.
 The Ekman transport through the southern boundary, VE, was computed using various data sets for the wind stress. Results are presented in Table 2.
Table 2. Annual Mean Wind-Stress and Associated Ekman Transports at 43°N
 The wind stress is also quite uncertain, but no explicit error estimate could be found in the references used here. It is generally admitted to be no larger than 25%. The field belongs to category B in the NCEP/NCAR terminology [Kalnay et al., 1996], which means that its determination involves some model outputs, and should therefore be considered with care. The value given by the TRENBERTH data set is anomalously high, and probably unreliable. Thus we will only consider the first three data sets, and take their arithmetic mean VE ∼ 5.1 Sv. BW derived a value of VE = 4.5 Sv from Fofonoff and Dobson , which is very similar and only slightly smaller, perhaps because this zonal-integrated quantity was computed over a narrower box.
 The interior geostrophic flow VG can be estimated by a thermal-wind calculation above a reference level (2000 m, as suggested by Roemmich and McCallister ), using the density data of Levitus and Boyer . We must note, however, that surprisingly swift currents have been found below that depth in the northwest corner of the basin [Owens and Warren, 2001], and that the very existence of a level of no motion should be considered with caution. As shown by equation (3), the relevant quantity for the salt transport estimate is the sum of the Ekman and geostrophic flow at each longitude, which we computed over boxes of Δλ × Δz = 2° × 200 m. By consistency, we can verify that the total geostrophic transport between 150.5°E and 125.5°W leads, for this grid spacing, to VG = 4.94 Sv (using a linear interpolation between 42.5°N and 43.5°N). This result is virtually identical to our estimate of VE (5.1 Sv) at 43°N; thus the assumption of Sverdrup balance across this latitude is not unrealistic, which Figure 2 did not suggest.
 The ratio of shallow to total geostrophic transport (0–200 compared to 0–2000), α, is about 0.53, larger than BW's preferred value of 1/3. In Figure 3 we present the Ekman and geostrophic transports as well as the estimated net transport. Integrating along the latitude circle, the total northward transport is V+ = 2.02 Sv, while the southward transport is V− = 3.45 Sv. The net transport is −1.43 Sv. We assume an error bar of 30%, to account for errors is the density field, wind-stress and computational procedure.
 For the integrated upwelling Vu, different strategies are available. BW used the upward velocity at the base of the Ekman layer computed by Fofonoff and Dobson  (4.5 Sv) and supposed that the upwelling would be two thirds as large at z = 200 m, i.e, 3 Sv. This was broadly consistent with his assumption that α ∼ 1/3.
 In our case, estimating yields a total upwelling of 4.74 Sv at the base of the Ekman layer. Assuming that α is the same over the whole box (0.53) implies 2.23 Sv. This might not be true. Instead, if we integrate the Sverdrup balance from −h (200 m) to 0,
Summed over the nNP box, the right-hand side equals 1 Sv, thus Vu ≃ 3.74 Sv. By consistency with the meridional transport, we consider an error bar of 30% (1.12 Sv).
2.4. Eddy Salt Fluxes
 The meridional eddy flux, , has been estimated from satellite altimetry by Stammer . His Figure 8b suggests that it is about 10 × 106 kg s−1 at 43°N for the upper 1000 m. As we seek its value over the top 200 m, the transport per unit depth it must be multiplied by a factor of 2 or 3 in order to account for surface intensification of the geostrophic flow (D. Stammer, personal communication, 2002); hence ∼ [4–6] × 106 kg s−1. These numbers are subject to considerable uncertainty.
 The vertical eddy flux, , may be estimated the same way (equation (4b)), albeit with even greater uncertainty concerning the vertical (approximately diapycnal) eddy diffusivity κz (between 10−5 and 10−4 m2 s−1). Probably, κz ∼ [1; 5] ×10−5 m2 s−1 is more realistic. Integrating across the area of the basin (deeper than 200 m), this yields ∼ [0.46; 2.3] × 106 kg s−1.
 In Table 3 we summarize the comparison between the boundary fluxes used for the salt budget by BW and this study, together with the uncertainties. Note that BW's estimates all lie within the range of variation of ours, allowed by these error bars. In the row corresponding to BW, V+ refers to his αVE, which is equivalent in his formulation. The salinity values of water flowing into the box remained almost the same, i.e., Su = 33.7‰ (the salinity at 200 m) and SWB = 34‰. See BW for a discussion.
Table 3. Boundary Water Fluxes for the nNP Box (Sv)
 In this section we present the results for the nNP box model using the values for fluxes described in the previous section. We first compute the northern North Pacific salinity for the values presented in Table 3 (the so-called basic state), then we evaluate the sensitivity of the model.
 As the functional form of Ss (equations (6) and (7)) depends on the sign of VWB, we first need to estimate the latter. It is determined as a residual from equation (5) using the values for the other quantities given in Table 3. We find that VWB is negative for this range of parameters (about −1.8 Sv), which corresponds qualitatively to the existence of the Oyashio current. Rearranging equation (7), we obtain
which is the ratio of a salt flux Fs to the throughflow rate V0 (more correctly, the “outflow volume transport”). It is noteworthy that the salinity does not depend directly on VBS and V− in this formulation, but indirectly through mass balance. For the parameters given above, we find Ss = 32.93‰, with V0 = 4.8 × 109 kg s−1 and Fs ≃ 160 × 106 kg s−1, which we will call the reference state. How sensitive is this result to observational and computational errors?
 To first order, variations in Ss around the reference state are related to those of the parameters ζi, i ∈ [1; 8] by
The analytical expression of these derivatives, and their value for the error bars (δζi) given above are presented in Table 4. Note that for all terms, the second derivatives are of order 1/V02 for terms with a small δζ or 1/V04 for those with a large δζ, and are null for the variables of the last four columns; hence the first order captures the essential part of the variations.
Table 4. Sensitivity Terms in the nNP Salt Budget
 It is obvious that the large error in the freshwater fluxes prevents any determination of Ss with an accuracy better than 0.8‰. We find that this result is rather independent of the reference state used for the linearization. Thus our estimate of 32.93‰ is as close to the observed Ss (33.07) as one can hope. The form of the derivatives suggests that the model is more sensitive to changes in the parameters, particularly the freshwater fluxes, at low throughflow rates V0 (i.e., for a longer residence time in the box). We note that the second biggest source of uncertainty is the eddy salt transport (about 0.3‰), confirming that mesoscale mixing can affect large-scale quantities to first order, but this does not alter our main result. Nevertheless, it does call for improved estimates of mixing in the ocean.
 With the same model, BW came to the conclusion that “the large freshwater flux and the small through-flow rate seem equally important in reducing the surface salinity by so great an amount that no convection can occur. The relatively low salinity of midlatitude surface water entering the region (i.e., the general freshness of the North Pacific in comparison with other oceans) influences somewhat the freshness of the surface water farther north, but that contribution alone is far from sufficient to prevent deep-water formation.” Our findings are entirely consistent with these views. We shall now compare this situation to the northern North Atlantic.
3.2. Comparison With the Atlantic
 The North Atlantic is roughly 2‰ saltier at the same latitude of 43°N (i.e., SWB ∼ 36‰), and P-E is much smaller there (Table 1). How would the model respond to such a situation? With SWB = 36‰, the model produces Ss = 33.45 for the nNP, which is only a modest increase over our basic state. This is a direct consequence of the weak cross-gyre salt transport: For a geostrophic flow and Ekman transport akin to the nNA (scaled by the width of the Pacific), and consistent upwelling, an Atlantic-like SWB = 36‰, the model gives Ss = 34.33. For an Atlantic-like value of F = 0.05 (compare Table 1), the salinity is raised to 34.34‰. With allowance for temporal and spatial variations of 0.3‰ BW, this value is close enough to the deep salinity (34.7‰) to permit wintertime deep convection in some areas of the nNP. We remark that this value of F is well below the lowest estimates of its value over the nNP (Table 1). A less extreme difference is present with other composite difference, for example the 0.17 Sv of the CMAP-ASMD, in which case Ss is only about 33.6‰. Yet the northern North Atlantic has other particularities: The eddy salt transport is about 5 times larger there at 43°N [Stammer, 1998], and the cross-gyre flow is much greater. Although the existing density structure may contribute by feeding turbulent eddy mixing through baroclinic instability, and by determining the meridional overturning circulation bringing salty water north, the main part of this cross-gyre flow is wind-driven. The atmospheric flow, through the wind-stress curl pattern, may therefore largely contribute to the ocean salt transport into the nNA, whereas the more zonal wind-stress pattern seems to prevent it in the nNP.
 From several independent data sets and reanalyses (Table 1), it is clear that the freshwater flux difference is almost entirely due to the precipitation field (∼0.40 Sv in the nNP, ∼0.30 Sv in the nNA). In contrast, the integrated evaporation is strikingly similar in the two basins (about 0.20 Sv). In order to explore the cause of these differences, we use the NCEP Reanalyses [Kalnay et al., 1996], for it also contains information about the moisture and wind field. We remember, however, that their freshwater balance is not entirely consistent with the other data sets, but we make the assumption that the difference between the Atlantic and the Pacific is qualitatively correct.
 In Figure 4 we plot the total P-E, E and P fields over the nNP (thick solid line), the nNA (dash-dotted line) and the difference (thin solid line). The difference noted earlier (i.e., that the evaporation is almost identical in the two regions, while the precipitation is about 0.2 Sv greater in the Pacific) seems to hold in all seasons in the Reanalyses. This results in a P-E difference (Pacific -Atlantic) between 0.12 and 0.19 Sv, peaking in May. However, this comparison is biased by the much larger area of the nNP (11.6 versus 6.7 × 1012 m2). In fact, the values per unit area look quite different: In the nNP, P = 3.47 × 10−5 kg m−2 s−1, E = 1.70 × 10−5 kg m−2 s−1. In the nNA, the precipitation per unit area is almost as high (P = 3.28 × 10−5 kg m−2 s−1), and the evaporation is 1.6 times greater (E = 2.75 × 10−5 kg m−2 s−1).
 Hence BW's conclusion remains apparently unchanged: As the nNP is rather isolated from southerly oceanic influences, it is the local excess of precipitation over evaporation that makes it so fresh. However, should we be surprised by the fact that the precipitation is relatively high over an ocean that evaporates a lot, or rather, by the high precipitation of the nNP compared to its low evaporation? The real issue is how the evaporative moisture source is locally balanced: In the absence of atmospheric moisture transports, we would expect P-E to vanish over each region. In fact, it nearly does over the nNA (at least in the Reanalyses), but there is a large excess over the nNP. Thus the question “Why is no deep water formed in the North Pacific” becomes “Why is the North Pacific so rainy despite its low evaporation?”
 The reanalysis of the box model developed by BW shows that the local excess of precipitation over evaporation largely sets the stability of the water column, and therefore the absence of deep convection, in the subarctic gyre of the North Pacific. However, one must acknowledge the shortcomings of this model. Intermediate waters (NPIW) are thought to be formed in winter in the Sea of Okhotsk and in the Gulf of Alaska [You et al., 2000], and to be occasionally ventilated in the Alaskan gyre [Van Scoy and Druffel, 1993]. NPIW consists of density classes heavier than σθ = 26.7, which is the mean potential density at 200 m in the nNP. Thus there is some diapycnal mixing with waters below 200 m (the lower boundary of our box), possibly not accounted for by our estimate of vertical eddy mixing. However, the error in ignoring this small flux ought to be negligible compared to other errors in the budget.
 The usefulness of this box model lies in its simplicity, and it is only appropriate to investigate the mean equilibrium salinity of the nNP box under a variety of conditions, but we cannot address the timing of the adjustment to varying fluxes, for the steady state assumption led us to ignore seasonal and decadal [Wong et al., 1999] changes in salinity and density. We have also left out any advective feedback, as described by Rahmstorf , which makes the model less appropriate for the North Atlantic where this feedback occurs. The model does not take account of deep convection, which is expected to occur in the North Pacific for Ss larger than 34.4 or so. Essentially, the model suggests that under current climatic conditions, it is the low throughflow rate and the large freshwater input over the North Pacific that are responsible for the large-scale relative freshness of the nNP compared to its Atlantic counterpart. Hence we now turn to the source of moisture responsible for the large excess of precipitation over the North Pacific (consistent in the three data sets presented in Table 1). For the western-boundary region, Roemmich and McCallister  hypothesize that the moisture excess is due to the monsoon circulation.
 In Figure 5 we present the moisture transport, horizontal convergence, and freshwater fluxes for the June–July–August period (JJA) derived from the NCEP Reanalyses. We clearly see the marked difference in the P-E term (bottom panel) of the nNA and nNP. The transport is convergent over the whole nNP, and the moisture is carried from the western subtropical Pacific by the large-scale low-level winds associated with the Asian Monsoon, which corroborates the hypothesis of Roemmich and McCallister . No such convergent circulation is found in the Atlantic at the same latitude [Trenberth et al., 2000], which may account for the difference.
 The MAM and SON periods are similar, albeit with smaller P-E differences (not shown). However, for the winter period (DJF, Figure 6), the interbasin P-E contrast is generally strong, albeit weakened along the western boundaries. In this case the outflow of exceedingly dry air from the Asian continent is allowing transfer of moisture from the western boundary current of the subtropical gyre into the subpolar Pacific. Downstream, the moisture is converged rather strongly in the Alaskan gyre, where the P-E contrast is greatest. This means of atmospheric freshwater communication between the subtropical and subpolar gyres appears to be stronger in the North Pacific than in the North Atlantic, perhaps because the Atlantic is too narrow for this to happen. This might explain why the total P-E is consistently larger in all season in the nNP (Figure 4): In summer the monsoon circulation transfers important amounts of moisture, presumably originated from the Indonesian seas and the Warm Pool, and converging over the western nNP; in winter, the circulation picks moisture from the western subtropical gyre and converges it into the northeast North Pacific.
 We performed an updated analysis of the processes that determine the salinity of the northern North Pacific using the box model described by BW, together with new, more abundant, and presumably more reliable data. We also added estimates of eddy salt transports, unavailable in BW's time. In the light of these firmer, though still imperfect, constraints, we propose that the salinity contrast between the Pacific and the Atlantic subarctic gyres is due to two major factors: (1) The limited oceanic cross-gyre flow and eddy mixing, preventing warm, salty subtropical water from reaching far northern latitudes in the Pacific, and maintaining low evaporation over the nNP. (2) The strength of the Asian monsoon and associated moisture transports. Both factors critically depend on the current land/sea configuration. As emphasized by BW, it is the location of the Himalayas and the Rockies that places the zero-wind-stress-curl line so far south over the North Pacific, thereby isolating the northern North Pacific. The Monsoon itself is the atmospheric response to the seasonal variations in land/sea thermal contrasts.
 As a number of conclusions rely on the (questionable) results of the NCEP Reanalyses, the study indicates the need for more complete analysis of global atmospheric freshwater fluxes and their variability, in particular those associated with the Asian Monsoon, and suggest that they might substantially influence the thermohaline circulation asymmetries.
 We thank two anonymous reviewers for what the eldest of us describes as “the most intelligent, perceptive reviews I have ever seen (in a sample size of more than 200).” Julien Emile-Geay is grateful to Jennifer Miller, Gerd Krahmann, Phil Mele, Marc Stieglitz, and Lawrence Rosen for their invaluable help. The authors thank Martin Visbeck, Takashi Kagimoto, Detlef Stammer, Arnold Gordon, and Robbie Toggweiler for useful discussions. Much of the data collecting and processing was done on the web-based IRI/LDEO Climate Library developed and maintained by Benno Blumenthal. J. E. G. was supported by the Ecole Normale Supérieure. Additional support comes from UCSIO PO 10196097 and ATM 9986515.