Response to Referees’ Comments
2004GL020022
"Interdecadal changes in mesoscale eddy variance…”
A. J. Miller et al. (2004)
Overall:
We thank the two referees for their detailed and perceptive comments. Indeed, most of the issues that are raised by the referees are points that we have already discussed and discounted before submitting the paper, as discussed below. We hope that the referees will acknowledge the novelty of the results and the uniqueness of the GRL format in quickly communicating these potentially influential results. We do have plans to continue this research along the lines suggested by the referees. The present letter is, we believe, good science that has impressed many biological oceanographers and intrigued many physical oceanographers that have heard these results in presentations at meetings and in informal discussions. A year or two of additional research is needed in high-resolution PE modeling of the dynamics of the seasonal cycle of the Gulf of Alaska circulation, the transient response to fluctuating winds, the spin-up of the boundary currents, and the stability properties and mixing characteristics of the mesoscale eddy field, combined with a careful assessment of the limited observational database. Long papers will indeed follow this tantalizing note.
Response to Comments of Referee 1:
Wind stress curl: The dominant pattern of wind-stress curl change after the shift is indisputably the pattern shown in Figure 1a. This pattern occurs if one computes 5, 6, 10, 15, or 20 -year averages before and after the shift, and then differences them. This was shown by Lagerloef (1995) for COADS winds, and by Capotondi et al. (2004) for NCEP winds. We have re-iterated the importance of this pattern in Figure 1a. The secondary pattern (Figure 1b) has no strong decadal component in its PC associated with the shift. We note that Lagerloef found Fig1b (1a) was the first (second) EOF in his analysis of COADS, although this has no bearing on the decadal signature of the forcing.
The referee suggests that the interannual variability of the wind-stress curl may be important. This may be true, and this interannual variability is indeed included in the forcing of the PE model. But Cummins and Lagerloef have shown that the time-dependedent response to fluctuating winds in the open ocean of the Gulf of Alaska (GoA) is essentially a static response, whereby the thermocline heaves up and down but does not generate Rossby waves. On the decadal timescale, this broad-scale thermocline heave sets up a broad-scale thermocline gradient that drives a strengthening of the Alaskan Stream; see the discussion in Capotondi et al. (2004) which uses a coarse resolution model.
Sverdrup response?: The referee is absolutely correct in suggesting that if Sverdrup dynamics apply, then the wind-stress curl pattern of Figure would lead to a weakening of the Alaskan Stream. We do not have adequate space to show the plot of this theoretical streamfunction response. But Rossby waves are required to equilibrate this theoretical response and these waves do not appear in nature (Cummins and Lagerloef), in coarse models (Capotondi et al), or in the eddy-permitting model here. Instead, a broad, static thermocline heave response to the curl pattern in Fig. 1a drives northwestward geostrophic flow that impinges on the western boundary and drives the increase (not decrease) in the Alaskan Stream. We have not worked out the details of how this fascinating response occurs. But it clearly does occur in the models and we will be exploring it in future simulations.
Recirculation east of the Alaskan Stream: The referee is correct in noting that the strong countercurrent recirculation seaward of the Alaskan Stream (to the east and north of Kodiak) is not likely to be an observed feature. This type of inertial recirculation feature does occur in other western boundary currents, like the Gulf Stream where it has been studied and measured intensively for decades (and many issues remain unsolved in that domain). It is likely that the model frictional parameters are somewhat weak, resulting in an overly energetic eddy field, which rectifies this mean return flow. An extensive model tuning procedure is warranted to select optimal vertical, lateral and bottom drag frictional parameters for model in this area. We believe, however, that the qualitative characteristics of the eddy field will remain unchanged, viz., enhanced variance where the flow is strengthened and weakened variance downstream. This eddy-driven countercurrent is not vital to the theme of the paper.
The Reed, Royer and Emery, and Bograd et al. references were meant to help substantiate the case for a seaward countercurrent Alaskan Stream along its entire path. A Pacific-wide, but coarser resolution, ROMS runs reveals a mean countercurrent that follows the topographic contours of the southeastward side of the trench that is present. This suggests the countercurrent is influenced strongly by the topography. Previous numerical models (of which we are aware) have not included this deep trench in their topographic representation, so this may be preliminary theoretical evidence for the existence of the countercurrent. Note that TOPEX altimetry cannot provide mean currents due to the unknown geoid; time differencing of sea level observations can provides information only on changes in the 1990s and beyond. Typical hydrographic surveys stop at the shelf break or slope, and do not cross the trench. XBT surveys exclude the salinity part of the density signal which may be important. An extensive analysis of the data is warranted by these theoretical results.
Velocity variance changes: The referee suggests that the velocity variance changes may be due to mean flow changes and not eddy changes. This impression is a consequence of our being unclear in describing what is plotted in Fig. 4. The figure contains variance of monthly-mean velocity anomalies variance after removal of the seasonal cycle monthly mean velocities based on the epoch (67-76 for Fig 4a; 79-88 for Fig 4b). Hence all the variance is due to either eddies (that is, due to unstable currents) or wind-driven anomalous currents. To distinguish whether the variance is due to changes in wind variance, we had also run a pair of simulations with no anomalous winds. One was forced for 10 years with monthly-mean seasonal cycle winds based on the 1970-76 (6-yr) average. The other was forced for 10 years by the 1977-82 (6-yr) average. The averaged mean currents and velocity anomaly variance plots of these two runs were so similar, respectively, to Figs. 3a,4a and 4b,4b (except for a basin-wide increase in variance in the anomalous-wind forced runs) that it was clear that the distribution of variance is controlled by the eddies, not the anomalous winds. We thought it was better to discuss only the 1952-99 hindcast, but we now mention these runs for support of our thesis that the unstable eddies are responsible for the changes.
Ecosystem implications: We agree that it is a long, arduous process to connect our mesoscale eddy variance mechanism to a conclusive effect on Steller sea lions. Indeed, it is likely that one could never ascribe a physical change to an ecosystem response with certainty. Ecosystems are far too complex. However, these experiments were motivated by the Steller sea lion decline (funded by NOAA CIFAR “climate regime shift hypothesis of the SSL decline” program) and provide an additional physical/climate mechanism to the suite of processes being considered as candidates for the way that physics influences biology (e.g., thermal stratification changes; E-P stratification changes, streamflow changes that influence the Alaska Coastal Current; changes in wind-driven coastal mixing, etc.). Moreover, the mechanism proposes the first explanation for the east-west asymmetry in the SSL response. Without the SSL connection, the results are pretty bland. With the SSL connection, they take on profound scientific and economic significance.
Response to Comments of Referee 2:
1. Observational comparison: We are preparing a proposal that includes extensive analysis of the observations. As a quick, but more complete, comparison with Lagerloef (1995), we offer the following. Lagerloef’s time sequence of sea level (estimated from XBT data) reveals a preponderance of low sea level (shoaled thermocline) in the northeast GoA during the 1970’s and a pre preponderance of high sea level (deepened thermocline) in the northeast GoA during the 1980’s (see his Figure 5). This is consistent with the static response to Ekman pumping of EOF1. It also agrees with the model results of Capotondi et al. (2004) who showed thermocline deepening to the northeast. Capotondi et al also did a comparison of their model output with GAK1 and PAPA, showing significant correlation between the pycnocline depth changes. The possible problem with the Lagerloef analysis only lies in the way that objective-analysis (OA) contours close around these broad large-scale patterns. The OA closes the contours around the aforementioned patterns to generate closed cells, suggesting a weakening of the Stream after the shift. The mass-conserving dynamically consistent ocean model sends the flows southwestward along the western boundary, increasing the Stream strength.
It is not our intent to call out Gary in the literature here. I talked with Gary about this and he agreed that the analysis procedure should be re-investigated in the context of the boundary current response of these model experiments. This will be a long process, however, that the Ph.D. student co-author (Ms. Hey-Jin Kim) plans to undertake in the coming years.
2. Model resolution and the internal deformation radii: The term “eddy-permitting” has come into fashion among modelers as a more accurate replacement for the presumptuous term “eddy-resolving”. Modelers have long recognized that increasing resolution in an eddy-resolving model seems to inevitably result in a richer spectrum of phenomena rather than convergence to a unique solution. Even if one “resolves” the deformation scale with, say, 6-to-10 grid points, one typically finds ageostrophic frontal flows, submesoscale coherent vortices, etc., that themselves are improperly “resolved”. It is a confounding pursuit to decide what has been resolved and what is partly resolved and what is completely missed in a given model run.
In the present situation, the grid resolution is roughly equivalent to the first internal deformation radius in the GoA. Emery et al. (JPO, 1984) and the Chelton et al. atlas (OSU web site) show that the deformation radius ranges from 10-18 km depending on local stratification in the GoA. The model grid ranges from 18.5 km in the south to 14.5km in the northern basin. Given our experience in other modeling domains and frameworks, it a reasonable supposition that increasing the resolution to marginally better resolve the Rossby radius would yield a solution with qualitative the same behavior. The mean currents would strengthen and the eddy variance would be enhanced (reduced) around the strengthened (weakened) mean flows. Smaller scales would be introduced to the mesoscale, of course, but there would little chance of validating this change mesoscale response. Since we already have uncovered an interesting result with our current simulation, we feel confident that the conclusions (east-west asymmetry, strengthened flows to the northwest GoA, altered eddy variance distribution as described) will be robust.
Cross-shelf transport: We are in the early stages of including a 3D NPZD-type ecosystem model in these simulations that will directly address both the changes in cross-shelf transport of nutrients and the changes in productivity associated with them. Changes in cross-shelf mixing are inevitable once the mesoscale eddy distribution changes. The details of this process will be evaluated in the coming years in the proper context of an ecosystem model solution.
1976-77 Climate Shift: The shift has been discussed extensively in the literature as both a Pacific-wide and regional GoA phenomenon. The winter of December 1976 through Febraury 1977 experience a monster anomaly of the Aleutian Low and commenced a long-term period of stronger-than-normal winters in the North Pacific (Miller et al., 1994). One can see this peak at the start of 1977 in PC1 (dark line). The previous year (1976) was in the weak AL regime. Since the total wind stress fields from 1952-1999 forced the run, it is only the interpretation of the output that requires specifying epoch-differencing intervals. People in the literature have used a variety of epoch periods and the results are not sensitive as long as 1976-77 is the transition point. Note that the extra runs (now discussed in the text, and discussed above) used 6-yr periods pre- and post-shift for computing the climatological winds. Others have used 20-year periods. Some people allow a few-year lag after 1976-77 to account for some spin-up. None of this is crucial to this short contribution.
3. Lead-Lag relation of EOF1 and EOF1: Propagation requires a 90 deg phase lag relationship in space and in time. The spatial patterns indeed exhibit the 90 deg lag as the referee notes. Over the duration of the 1952-1999 interval, howver, there is no consistent lead-lag relationship in time. Right before the shift, the two time series both have roughly 8-year wiggles that slightly lagged with respect to each; hence one might be able to interpret a transition from 1976 to 1977 as a switch from EOF1 to EOF2 (propagating the red maximum of EOF2 to the southwest in EOF1.) But a simple difference plot using any reasonable choice of epochs yields a wind stress curl pattern very similar to EOF1. And we are discussing inter-decadal changes in the paper rather than interannual. Lastly, these EOF patterns are distinct and not mixed or degenerate; the COADS analysis of Lagerloef (1995) recovers these same two dominant patterns. Although the results are not new, it useful to verify that NCEP RA2 winds produce correct patterns.
Linking physics and biology: See the Ecosystem Implications comments in our response to Reviewer 1. References are now noted that link cross-shelf exchange to biological productivity. The articles by Benson and Trites (2002) and Trites and Donnelly (2003) include various linkages between climate and Steller sea lions, all the way up the through the food chain. The present results are a novel explanation of the fundamental changes physical that may have precipitated this response.
5. The model has 20 generalized sigma-coordinates levels.
6. 1951 is correct.
7. The bathymetry shows the degree of smoothing that was needed for stability of the model and for minimizing pressure gradient error. Since no previous models that we are aware of have included the full depth of the water column and a realistic Aleutian trench, we feel it is useful to show the topography for comparison with previous model simulations. We feel plotting the shelf-slope boundary on Figure 3 or 4 would obscure the contours, since that is where all the action is.