Geochemistry, Geophysics, Geosystems Supporting Information for Ventilation and dissolved oxygen cycle in Lake Superior: Insights from a numerical model Katsumi Matsumoto1, Kathy S. Tokos1, and Chad Gregory2 1Dept. of Earth Sciences, University of Minnesota, 310 Pillsbury Dr. SE, Minneapolis, MN 55455 2University of Michigan Contents of this file Text S1 to S3 Figures S1 to S5 Additional Supporting Information (Files uploaded separately) Captions for Movies S1 to S3 Introduction This Supporting Information provides information on four topics: (a) modifications to our previous model of Lake Superior; (b) model-data comparison; (c) idealized model tracer Dye; and (d) movies of Dye simulations. Text S1: Modifications to the Lake Superior Model The model previously used at the University of Minnesota [White et al., 2012] has been updated to ROMS version 3.6. The dynamic ice model, as described by White et al. [2012] and based on Hunke and Dukowicz [1997] and Hunke [2001], was then incorporated into the new version of ROMS. A modification of the ice code was included that results in a more robust agreement with observations of both surface temperature and wintertime ice coverage of Lake Superior. The net heat flux from the atmosphere to lake ice, which in turn determines the surface temperature of the ice and its subsequent melting or formation, is calculated by combining the net fluxes of sensible heat, latent heat, short- and longwave radiation. In the original 1 formulation, under boundary conditions described by White et al. [2012], the net longwave radiative flux (LWnet) from the atmosphere to the ice was calculated as: LWnet = LWin + 4 Ta3 (Ts - 2Ta) , (Eq 1) where LWin is incoming longwave radiation from the model forcing, and σ, the StefanBoltzmann constant, is set equal to 5.67 x 10-8 W m-2 K-4. The emissivity of ice is ϵ = 0.97, and Ta and Ts are air and surface temperature. The second term in Eq. 1 appears to be an approximation of downwelling longwave radiation, and it is unclear why it is added to the incoming radiation that forces the model. In the new version, we use instead the simple StefanBoltzmann relationship to calculate the outgoing longwave radiation: LWout = ϵTs4 , (Eq 2) and calculate LWnet as: LWnet = LWin - ϵTs4 . (Eq 3) Figure S1 illustrates the resulting difference in terms of ice concentration, which we define as the percent of the lake that is covered by ice, for the year 1997. Although the maximum amount of ice formed (blue line) is slightly less than with the previous formulation (green line), the rapid melting of lake ice is captured much better as seen by comparing with the ice atlas observations (red line). Also shown in the figure (black line) is ice concentration from 1997 of White et al. [2012], with lake ice remaining until well into the summer. Another source of the discrepancy between White et al. [2012] and the latest version of the model is the forcing. We have chosen to force the model with the NARR reanalysis, which is more objective and readily available from 1979 to the present. In addition, rather than an ad hoc application of cloud cover effects to incoming shortwave radiation, we account for cloud albedo by using the net incoming shortwave radiation that is included in the NARR output. Text S2: Model-Data Comparison With the described modifications to the model, we find good agreement of daily mean lake surface temperature (LST) and percent ice cover between model output and the observation-based data. NOAA’s Great Lakes Ice Atlas, obtained from the Environmental Research Laboratory (GLERL, glerl.noaa.gov) provides daily measurements of the total fraction of Lake Superior covered by ice from December 1 to May 30 or 31. Daily, lake-wide mean LST is calculated from Great Lakes Surface Environmental Analysis (GLSEA) maps, also from GLERL. In Figure S2a is shown daily, lake-wide average LST for the years 1995-2003 from the model (blue line), observations (red line), and the model of White et al. [2012] (green line). With a root mean square error (RMSE) value of 1.9°C over the 9 year period, modeled LST maxima and minima compare well with the seasonal thermal cycle of the lake. The model reproduces observed LST best in cold winters, at the cost of overestimating LST during following stratified period. In warmer winters, the model overestimates LST, but with less deviation from 2 observations the following summer. We note that Bennington et al., [2010] also reported a warm bias in their MITgcm-based model of Lake Superior and attributed the bias to the NARR air temperatures. Our warm bias may also be related to the NARR forcing, but we believe there is another more important cause, because LST of White et al. [2012] is also warm. Daily RMSE between White et al. [2012] and the observations is 2.0°C, similar to that in our current model. However, the forcing for White et al. [2012] was observationally based and not NARR, pointing a possibly different cause; for example, the coefficients of vertical mixing may need adjusting. Figure S2b compares daily ice coverage of model output with observations over the same time period. The model clearly shows the interannual fluctuations of ice coverage seen in the data. The daily RMSE between the model and observations is 10.2%, while that of White et al. [2012] and observations is 13.9%. However, this method of comparison could have a very large RMSE if the ice coverage were identical, but the dates of onset and melting were simply off by days or weeks. Another way to assess the ice production in the model is to compare the average amount of ice cover over all days with ice greater than 1% over a calendar year (Figure S2c). For example, the annual mean ice coverage for 1996 was 40%, although the maximum coverage for that year is 100%. This allows a better comparison of the thermodynamic properties of ice formation. The RMSE between the annual averages of the model and observations is just 5.5%, while that of White et al. [2012] was 10.0%. Text S3: Tracer Dye As described by England [1995], tracer Dye is a passive tracer like tracer Age, and the two tracers are entirely consistent. In the main text, only the description and results of Age are presented. Here we describe and present tracer Dye. The time evolution of Dye is governed by: 1, z 0 Dye(x, y, z, t) Dye(x, y, z, t - 1) {Dye(x, y, z, t)}, z 0 Tracer Dye is conserved in the lake interior and therefore does not age. The Dye concentration is continuously set to 1 at the surface (Figure S3a). As surface waters penetrate the lake interior, the interior Dye concentration will increase with time (Figure S3b). As illustrated, an interior grid box, which is ventilated readily, will quickly reach the maximum concentration of 1, while those boxes that are poorly ventilated will only slowly increase in Dye concentration. Eventually the entire lake will reach a concentration of 1 (Figure S3c). From that time, the Dye concentration will no longer reflect any changes in ventilation and thus becomes useless. We therefore release Dye on the first of the years of interest. Before Dye concentration reaches one, its transient penetration into the lake interior provides information not contained in tracer Age. Specifically, the first derivative of the time series of Dye concentration at any point in space gives the age spectrum, or the frequency distribution, of ventilation age (Figure S3d, e). The frequency-weighted mean of the age spectrum obtained from the Dye tracer would correspond to the value of the Age tracer. In other words, the spectrum gives information about the make up of an interior water parcel in terms of multiple surface water source histories. 3 As with the Age tracer, the Dye tracer maps on the 100 m level also show that surface waters penetrate into the interior more readily over undulations in bottom topography in the eastern basin in both 1997 and 1998 (Figure S4a, b). For example, the cross sections at 47.1c (Figure S4c, d) show that the Dye concentration is higher in the topographically rougher eastern arm (e.g., 84-88°W) than in the smoother and shallower western arm (e.g., west of 88°W). The Dye concentration and its first derivative reveal more detail of the mixing history of a water parcel. Figure S5 shows two time series for 1998. In the first Dye concentration is shown at the same water depth of 175 m at three different locations, all where the lake floor is very deep (220-300m). One is near the north shore in the western arm where there are frequent wind-driven upwelling events (green line, Figure S5a; see filled circle, main text Figure 1 for location), one is in the deepest basin (blue line; filled triangle in Figure 1), and one is in the eastern arm in a region of very rough bottom topography (red line; filled square, Figure 1). Figure S5a shows that the water on the north shore (green line) remains somewhat isolated after the initial mixing in January. Then in June, springtime overturning mixes down surface waters, until almost 80% of the water at that depth has been ventilated. Mixing is nearly complete by the end of November as another, possibly wind event brings more surface water to depth. Dye concentration does not reach 1 in 1997 at this location, indicating that overturning is not complete until the following year. In the very deep basin, the initial mixing is evident (blue line), as is the spring overturning in June, when most of the dye reaches 175m. In September, upwelling or horizontal advection brings in water with lower Dye concentration, and the water here also refrains from mixing completely with the surface until the end of the year. The deepest basins generally tend to remain free of the Dye the longest, which is consistent with Age being the oldest in the same basins. But, despite being in a deep part of the lake, surface waters quickly penetrate to175 m depth in the region of rough topography (red line), so that the Dye concentration there reaches 1 almost immediately. The first derivatives of the Dye time series reveal the composition of the ventilation age distribution (Figure S5b). Where Dye reaches 1 almost immediately, such as in shallow waters or over rough topography, the first derivative has a single high peak at the very beginning, indicating a large fractional contribution of surface water at the beginning. Elsewhere there may be multiple peaks that indicate multiple contributions of surface waters to give the value of Age. For example, the first derivatives of Dye concentration on both the North Shore and in the deepest basin have two large peaks (Figure S5b): one early in January when the model starts in nearly isothermal conditions, and a second in June when late spring overturning finally reaches the 175 m. This indicates that the water in each of these regions is composed primarily of two sources of water: one is surface water from the fall overturning; and a second source comes from the surface during spring overturning. Since the area under the first peak is larger than the one in June, the fractional contribution of the fall overturning is larger than of the spring overturning at this depth for this year. The smaller peaks, especially for the North Shore water (green line), indicate many smaller mixing events, each contributing a small volume of surface water to 175m depth. This is in contrast to the region over rough bottom topography. There is only one peak of surface water injection (red line), when the water column was thoroughly mixed, and no more information of surface water history is possible. Similarly, Figure S5b depicts the initial major mixing event to all depths in January, and the lesser event in June. The deeper waters are able to receive a greater contribution from the surface during June than the shallower ones, since the shallower ones are already nearly completely mixed. 4 Figure S1. Simulations and observations of ice concentration, defined as the percent of the lake that is covered by ice, for 1997. Red line is from NOAA’s Great Lakes Ice Atlas; blue line is ROMS v3.6 with the current ice modification, black line is ROMS v3.6 without ice modification, and the green line is from White et al. [2012]. 5 Figure S2. Daily, lake-wide average LST for the years 1995-2003 from the model (blue line), observations from GLSEA (red line), and the model of White et al. [2012], (green line): a) daily, lake-wide mean surface temperature, b) total fraction of lake covered by ice, and c) annual mean of fraction of lake >1% covered by ice. 6 Figure S3. How tracer Dye works. The effect of vertical mixing in the interior is to increase Dye concentration (a-c). Given a time series of Dye concentration (d) of some water parcel in the model domain, the first derivative of the time series (e) is the age spectrum or the frequency distribution of ventilation age. In this example, this water parcel is largely composed of waters Figure S2 that left the surface at times centered on t1 and t2. The frequency weighted mean age would correspond to the value of tracer Age. d) Dye =1 Dye = 0 b) Dye concentration a) Dye = 1 c) Dye = 1 e) Frequency 0 < Dye < 1 Dye = 1 t1 Time t2 7 Figure S4. Model-simulated Dye distribution in 1997 (a, c) and 1998 (b, d) at 100 m water depth (maps) and along 47.1° N (cross sections). The lines in the maps indicate location of the sections. 8 Figure S5. Time series of Dye (a) and its first time derivative (b) at three locations at 175 m depth. See Figure 1 in the main text for the three locations. 9 Movie S1. Model output of percent of Lake Superior covered by ice in 1997 (upper panel) and 1998 (lower panel). Movie M2. Model output of tracer Age, in days, at 100 m water depth in 1997 (upper panel) and 1998 (lower panel). Movie M3. Model output of tracer Age, in days, at 47.1 N in 1997 (upper panel) and 1998 (lower panel). References Bennington, V., G. McKinley, N. Urban, and C. P. McDonald (2012), Can spatial heterogeneity explain the perceived imbalance in Lake Superior’s carbon budget? A model study, J. Geophys. Res., 117(G03020), doi:10.1029/2011JG001895. England, M. H. (1995), The age of water and ventilation timescales in a global ocean model, J. Phys. Oceanogr., 25(11), 2756–2777. Hunke, E. C. (2001), Viscuous-plastic sea ice dynamics with the EVP model: Linearization issues, J. Comput. Phys., 170(1), 18–38. Hunke, E. C., and J. K. Dukowicz (1997), An elastic–viscous–plastic model for sea ice dynamics, J. Phys. Oceanogr., 27(9), 1849–1867, doi:10.1175/1520-0485. White, B., J. Austin, and K. Matsumoto (2012), A three dimensional model of Lake Superior with ice and biogeochemistry, J. Gt. Lakes Res., 38, 61–71. 10