1 2 3 4 5 A dynamical analysis of the shedding of a Loop Current eddy in the Gulf of Mexico F.-H. Xu,1 Yu-Lin Chang,1 L.-Y. Oey1,* and P. Hamilton2 1: Princeton University; 2: SAIC *Corresponding Author: lyo@princeton.edu 6 Abstract 7 The shedding of Loop Current eddies in the Gulf of Mexico has long been considered to 8 be driven by internal chaotic dynamics. Recent studies suggest, however, that eddy-shedding is 9 both internally-driven and wind-forced. An example of such a case is analyzed here based on the 10 shedding event in July of 2011 using the outputs of a time-dependent, primitive-equation 11 numerical model (the Princeton Ocean Model). The model results are first validated using 12 satellite and in situ observations. It is then shown that model initialized from a data-assimilated 13 analysis on Jul/01, 2011 predicted eddy shedding near the end of July in agreement with 14 altimetry data. From May through June, the Yucatan transport increased in response to the 15 strengthening trade wind in the Caribbean Sea. The Loop Current grew and ultimately reached a 16 sufficiently large size that β effects became important. Trade wind and Yucatan transport 17 weakened in early July, and the Rossby wave speed (-βRo2) exceeded the Loop Current growth, 18 providing a favorable condition for eddy shedding. Instability analysis shows that at incipient 19 shedding, baroclinic instability growth rate increases north of the Campeche Bank, thereby 20 accelerating the eddy shedding. 21 22 1. Introduction 23 The Loop Current (LC) is the dominant feature of the circulation in the eastern Gulf of 24 Mexico (GOM) and the formation region of the Florida Current-Gulf Stream system. It is a 25 component of the meridional overturning circulation, playing a key role in the global climate. Page | 1 26 The LC episodically sheds large warm-core eddies or rings (200-300km wide, 500-1000m deep) 27 that generally propagate westward at 2~5 km day-1, with maximum swirl speed ≈1.8-2 ms-1 (see 28 Oey et al. 2005 for a review). The LC and eddies affect every aspect of oceanography in the 29 GOM, either directly or indirectly through the action of their smaller-scale subsidiaries. 30 The LC eddy shedding was observed (e.g. Vukovich 1988; Sturges 1994; Sturges and 31 Leben 2000) and simulated in several numerical models (Oey, 1996; Oey et al. 2003; Chang and 32 Oey 2011; Chang and Oey 2012a; Le Hénaff et al. 2012). Chang and Oey (2011) describe the 33 ideal case of the Loop Current Cycle: LC expansion, shedding, and retraction cycle. However, 34 LC eddy often detaches and re-attaches to the LC several times before final separation (Oey et al. 35 2003; Hénaff et al. 2012), suggesting a highly nonlinear and complex process. 36 The shedding of Loop Current eddies can be understood as a competing imbalance 37 between the mass influx through the Yucatan Channel, which grows the Loop, and westward 38 Rossby wave which tends to ‘peel’ eddy from the Loop; this will be referred to as the Pichevin- 39 Nof mechanism (Pichevin and Nof, 1997; Nof, 2005). The Loop Current grows larger and deeper 40 with mass influx from the Yucatan Channel. When its Rossby radius Ro (based on the Loop’s 41 upper-layer depth) reaches a certain size, the variation of the Coriolis parameter (f) becomes 42 significant ( effects), and the westward Rossby wave velocity (-βRo2) exceeds the LC growth 43 rate due to the mass influx. At this point, the Loop Current eddy begins to detach. The idea may 44 be extended to the case when the mass influx (i.e. Yucatan transport) varies slowly in time 45 (longer than 1~2 months), so that eddy-shedding may also depend on this variation. Oey et al. 46 (2003) show that models forced by time-dependent wind produce strong Yucatan transport 47 fluctuations, of order of a few Sv’s and larger, which in turn also influence the shedding periods. 48 Chang and Oey (2012a) identify biannual preferences in summer and winter of the LC eddy Page | 2 49 shedding by analyzing long-term observational data and numerical model results. They found a 50 strong dependence of eddy-shedding on Yucatan transport. The biannual variation in the trade 51 wind forces a corresponding biannual transport through the Yucatan Channel; as a consequence, 52 the Loop Current has a tendency to shed eddies as the wind weakens from summer to fall, and 53 also from winter to spring. 54 The Pichevin-Nof mechanism is fundamental for Loop Current eddy-shedding. Other 55 factors such as baroclinic instability and upper-lower layer coupling do not determine eddy 56 shedding, but they can modify it, by accelerating the timing when eddies are shed (see the 57 companion submitted manuscript: “Chang and Oey, 2012b: Loop Current EOF modes”). Oey 58 (2008) showed that north of Campeche Bank is a fertile ground for baroclinic instability which 59 can generate deep cyclones below 1000m. 60 accelerates the upper-layer LC eddy shedding. The upper and lower layer (shallower and deeper 61 than 1000m) mass coupling in the eastern Gulf (east of 900W) can also accelerate eddy shedding 62 (see Fig.5 of Chang and Oey, 2011, hereafter CO2011). They examined mass exchanges between 63 the upper and lower layers, and also between the eastern and western Gulf of Mexico. At 64 incipient shedding, the westward-extended Loop Current forces a deep return flow into the 65 eastern Gulf where the upper layer then becomes divergent while the lower layer becomes 66 convergent. The strong upwelling in the eastern Gulf again accelerates eddy-shedding. The cyclone in turn produces upwelling that 67 In late July, 2011, an eddy was observed to separate from the Loop Current, coincident 68 with the weakening of the summer trade wind in that year. The scenario appears to match the 69 idea discussed in Chang and Oey (2012a), and outlined above. In this paper, we attempt to 70 support the idea based on a realistic case-study. We describe a free-running (i.e. non data- 71 assimilated) model initialized on Jul/01 and forced by realistic wind that predicts eddy shedding Page | 3 72 near the observed shedding time. We analyze this case in detail in order to understand the 73 underlying eddy-shedding dynamics. Section 2 describes the model and experiment setup. 74 Section 3 presents the results, and here we also validate the model against observations. Section 75 4 is discussion and section 5 concludes the paper. 76 77 2. The Numerical Model 78 The Princeton Regional Ocean Forecast System (PROFS) for the Gulf of Mexico is used 79 for this study; this model is based on the Princeton Ocean Model (POM) and has been 80 extensively tested for (i) process studies to understand the Loop Current and eddy-shedding 81 dynamics (see, e.g. above quoted works by Oey et al. and Chang and Oey); (ii) comparison 82 against observations (e.g. Wang et al. 2003; Fan et al. 2004; Oey et al. 2004; Lin et al. 2007); 83 and (iii) hindcasts and forecasts (e.g. Oey et al. 2005; Yin and Oey, 2007). (For a more complete 84 list of references please visit http://www.aos.princeton.edu/WWWPUBLIC/PROFS/publications). 85 The present version of the model has MPI directives implemented into POM by Dr. Toni Jordi 86 (http://www.imedea.uib-csic.es/users/toni/sbpom/) and achieves good (i.e. linear) scalability. 87 Most of the original PROFS (and POM) features are retained. There are 25 terrain-following 88 sigma levels with (logarithmically) finer resolutions near the surface and bottom, but a fourth- 89 order pressure gradient scheme (Berntsen and Oey, 2010) is used to guarantee small pressure- 90 gradient errors. An orthogonal curvilinear grid is used in the horizontal with x and y 2~5 91 km in the Gulf of Mexico. 92 modified by Craig and Banner (1994) to include input of turbulence energy due to breaking 93 waves near the surface is used. A Grant and Madsen (1979) type bottom-drag scheme is used in 94 the bottom boundary layer to empirically account for increased bottom roughness due to surface The Mellor and Yamada’s (1982) turbulence closure scheme Page | 4 95 gravity waves; this is particularly effective over shallow shelves and under strong wind 96 conditions. The Smagorinsky’s (1963) shear and grid-dependent horizontal viscosity is used with 97 a nondimensional coefficient = 0.1; the corresponding horizontal diffusivity is made five times 98 smaller. 99 (http://gcmd.nasa.gov/records/GCMD_NAVOCEANO_MCSST.html) with an e-folding time 100 constant of 1 day-1. However, tests (and previous experience – e.g. Yin and Oey, 2007) indicate 101 that the effects of SST boundary conditions on Loop Current dynamics are minor. The sea surface temperature (SST) is relaxed to AVHRR MCSST The model domain includes the north-west Atlantic Ocean west of 550W and from 50N 102 World Ocean Atlas data (“climatology”) from NODC 103 ~550N. 104 (http://www.nodc.noaa.gov/OC5/WOA05/pr_woa05.html) is used for boundary condition along 105 the eastern open boundary at 550W. The topography was set up according to Etopo2 and edited 106 on shelves using NOS digitized maps. An analysis field was obtained for 19 years from 1993- 107 2011 108 (http://www.aviso.oceanobs.com/) is assimilated into the model using a statistical surface- 109 subsurface projection method (see Yin and Oey, 2007 for details and references). The model is 110 forced by six-hourly winds (1993-2009: 0.25o×0.25o CCMP (cross-calibrated Multi-platform), 111 and 2009-2011: NCEP 0.5o×0.5o GFS (Global Forecast System)), by M2, S2, K1 and O1 tides 112 specified along the open boundary at 55oW, and by daily river discharges obtained from the U.S. 113 Geological Survey at 51 locations along the U.S. coastline (34 rivers in the Gulf and 17 rivers in 114 the eastern coast). The during which satellite sea-surface height anomaly (SSHA) from AVISO 115 Results obtained from the above hindcast analysis run are used as initial conditions for 116 two free-running experiments to study July/2011 Loop Current eddy-shedding dynamics: an 117 experiment initialized from July/01/2011 hindcast field (Exp.Jul01) and another experiment Page | 5 118 initialized from May/15/2011 hindcast (Exp.May15). The rationale for these experiments will 119 become clear shortly. For these free-running experiments, neither the AVISO SSHA, nor the 120 MCSST data were used; other forcing are the same as in the hindcast analysis run. 121 convenience, these experiments will be called “forecasts” even though strictly-speaking they 122 really are not since wind (and other forcing) are used. Additional, sensitivity experiments were 123 also conducted, as will be mentioned below. For each experiment, daily-averaged fields are used 124 for analyses. For 125 126 3. Results 127 The Exp.Jul01 predicts a LC eddy shedding during the last week of July in agreement 128 with that observed in AVISO satellite sea surface height (SSH) data (Fig.1). The hindcast and 129 AVISO both show on Jul/01 (Fig.1a) a northwestward-extended Loop Current, although AVISO 130 zero-SSH contour is some 50km more extended. Ten days later (Jul/11), the forecast LC 131 intrudes northwestward, and which compares well with AVISO. On Jul/21, both AVISO and 132 model Loop developed a thin neck typical of the situation at incipient eddy shedding (Oey, 1996; 133 Schmitz et al. 2005; Oey, 2008). The forecast shows that a large eddy was shed shortly thereafter 134 on Jul/25 in agreement with AVISO. 135 136 Model Skill Assessment Against Satellite Observations: 137 To evaluate forecast skill, time series of the spatial correlation coefficient and root mean 138 square errors between the model and AVISO SSH anomalies are calculated and compared 139 against persistence for the open-ocean region of the GOM: north of 23oN and west of 84oW in 140 water depths deeper than 500m (Fig.1b). The correlation 0.8 on Jul/01 – the initial hindcast Page | 6 141 value; it remains above 0.65 during the first 5 weeks of the forecast in agreement with Yin and 142 Oey (2007) who concluded based on bred-vector analyses that the limit of Loop Current and 143 eddy forecast horizon in the Gulf of Mexico is 4~6 weeks (see also Oey et al. 2005). At week#6, 144 the correlation drops to approximately 0.5 but it remains above this value from weeks6~7 before 145 degrading further to about 0.45 at the end of week#8. The RMS errors increased from 0.18 to 146 0.23 over the first 8 weeks. In both measures, the model forecast beats persistence. 147 148 Model Validation Against in situ Measurements: 149 In addition to evaluating the model skill against AVISO, we have also conducted an 150 extensive comparison of both model hindcasts and forecasts against direct current-meter (ADCP) 151 measurements in the Loop Current collected for Bureau of Ocean Energy Management (BOEM) 152 by Dr. Hamilton and co-investigators. 153 separately. The study confirms the above model-AVISO comparison that during the 2011 July 154 eddy-shedding event the Exp.Jul01shows considerable skills. We now examine more closely the 155 model results with the objective of understanding the relevant dynamics and forcing that can 156 explain the eddy-shedding event in July 2011. The hypothesis is that, according to Chang and 157 Oey (2012), the trade wind and hence Yucatan transport are important to eddy-shedding over the 158 seasonal time scale. To simulate the corresponding variation over several months prior to the 159 July shedding, we analyze in details another free-running run (Exp.May15) initialized from the 160 May/15 hindcast field. The free-running experiment then eveolves in accordance to wind and 161 edd-forced model dynamics unhampered by spurious force balances that may result from a data- 162 assimilated run. Details of this comparison study will be reported 163 Page | 7 164 Analysis of the Exp.May15: 165 The Exp.May15 produces an eddy-shedding event on Jul/8, 2~3 weeks earlier than the 166 observed (and Exp.Jul01) shedding date. The eddy-shedding date is some 7~8 weeks into the 167 forecast horizon which is near the limit of modeled Loop Current predictability, and some 168 discrepancies from observation are therefore to be expected (Yin and Oey, 2007). Overall, 169 however, the Exp.May15 shows good skill (plots not shown), and the fact that the model predicts 170 an eddy-shedding, albeit a little earlier, is of great interest, and suggests an underlying dynamics 171 to be learnt. 172 173 Eddy-Shedding and Its Conneciton to Caribbean Trade Wind & Yucatan Transport Variations: 174 Figure 2 shows the SSH on June 10th from Exp.May15. This shows that both model and 175 observed LC are well-extended at this time. As the LC is fed by inflow through the Yucatan 176 Channel, it grows to a sufficiently large size (diameter 200~300km), and is sufficiently deep 177 (upper-layer thickness h(t) 500m) that -effects become important (Nof, 2005). The westward 178 Rossby wave velocity = Ci = - βRo2, where Ro = (g’h(t))1/2/f, then exceeds the inflow growth rate, 179 and a portion of the Loop’s mass – i.e. a warm eddy – is “peeled” westward. The Ci is calculated 180 at the star location (-89.40, 260) of Fig. 2, and its temporal variation is shown in Fig. 3 (bottom). 181 The Ci increased from March to June, became maximum at the end of June, and then dropped 182 rapidly in early July when eddy shedding occurred. 183 The growth of the Loop Current is caused by continual influx of warm water into the 184 Gulf of Mexico from the Caribbean Sea. 185 increases with the trade wind strengthens from spring to summer, and it weakens from summer 186 to fall concomitant with a corresponding weakening of the trade wind. During 2011, the trade Chang and Oey (2012) showed that this influx Page | 8 187 wind stress (negative zonal component) in the Caribbean Sea strengthened from near-zero in 188 mid-May to -10-4 m2 s-2 in late June, and it weakened to -0.3×10-4 m2 s-2 in mid-August (Fig. 3, 189 top). The modeled Yucatan transport increased from mid-May to late June when it peaked, and it 190 then weakened through August (Fig. 3, middle). This relation between the trade wind and the 191 Yucatan transport agrees well with Chang and Oey’s (2012) conclusions. 192 correlation coefficient between the daily zonal wind stress and Yucatan transport is -0.6 (above 193 the 95% significance; -0.75 if 10day-running mean time series is used). The authors analyzed a 194 long-term (20 years) model run forced by reanalysis wind stress dataset derived from satellite 195 and buoy observations and model, and found that the Yucatan transport peaks (weakens) at 196 approximately the same time (with a short lag ~ 1 month) as when the trade wind in the 197 Caribbean Sea peaks (weakens). They concluded that the weakening of wind and Yucatan 198 transport together with the matured development of the Loop Current (i.e. with large Ci) provide 199 a favorable condition for the separation of the LC eddy. In the present case, the weakening of 200 the trade wind and Yucatan transport occurred from the end of June through mid-August 201 coincident with the near-maximum Ci (Fig.3), and an eddy is shed shortly thereafter (on Jul/08); 202 the scenario agrees well with Chang and Oey’s (2012) analysis. The zero-lag 203 204 Upper & -Lower-Layer Coupling 205 Now we examine the upper and lower layer coupling. As CO2011, we define a semi- 206 enclosed volume in the eastern Gulf, bounded by 900W (west), Yucatan Channel (South), and the 207 Straits of Florida (east). The transports through west, south, and east boundaries in the upper and 208 lower (shallower and deep than 800m) layers were calculated (not shown). Note that the east 209 boundary of the lower layer is closed due to the shallow sill of the Straits of Florida. The Page | 9 210 divergence/convergence of the upper and lower layer is calculated as transport out of the layer 211 (Trout) subtracts transport into the layer (Trin). Therefore, the positive value indicates divergent, 212 and the negative is convergent. The time series of 10-day running average of Trout-Trin of upper 213 (solid) and lower (dash) layers are shown in Fig. 4. From the end of March to mid June (before 214 Jun/18), the upper layer was mainly convergent and the lower layer was divergent, indicating the 215 growth of the LC. This corresponds to the Loop reforming phase described in Fig. 5a of CO2011. 216 The mean SSH in the eastern Gulf (grey line in Fig. 4) also shows the higher sea level before 217 Jun/18, consistent with the convergent. After Jun/18, the sea level dropped, the lower layer is 218 convergent, and upper layer is divergent, corresponding to the incipient shedding phase in Fig. 219 5b of CO2011. An upward motion induced by the two layer coupling contributed to the shedding. 220 Considering the eddy size = 250km and the Rossby wave vecolity C i=0.05 m s-1, the time 221 required for the eddy to move over its own size is about 2 months. If Yucatan inflow and Rossby 222 wave dynamics alone, the eddy shedding would occur in mid August instead of Jul/08. There 223 must have other processes accelerating the eddy shedding. 224 Baroclinic instability was calculated over two 30-day periods, May/16 - Jun/15 and Jun/01 - 225 Jun/30 at depth 750m (Fig. 5). During the first period, a dominant instability (positive; shown in 226 red; right panel) BC was seen north of the Campeche Bank around 24.80N, 88.30W where the LC 227 flows over the deeper regions of the Gulf. The dominant instability was getting larger and 228 spreading northwestward in the second period. These results are consistent with the 10-yr 229 ensemble average of baroclinic instability in Oey (2008). The corresponding growth rate is about 230 1/40 ~ 1/10 per day. Therefore, the instability accelerated the eddy shedding for about 10-40 231 days after Jun/18. This is consistent with the model estimated shedding on Jul/08. Page | 10 232 The cyclonic vorticity ζ on the western (~50km) portion of the Yucatan Channel divided by f 233 gives a good predictor of the LC northern boundary (Chang and Oey 2012). A high correlation, 234 R2 = 0.83, was obtained (see Fig.2c of Chang and Oey 2012). This relation is in good agreement 235 with the Reid’s formula (Reid 1972). The larger the ζ/f is, the more northward the LC extends. 236 The Loop retracts as ζ weakens. The 10-day average time series of the ζ/f and northern boundary 237 of the LC represented by latitude was shown in Fig.6 for the free model run May/15. From the 238 end of the June to mid July, the ζ/f (red) decreases from 0.475 to 0.41, and meantime the 239 northern boundary of the LC retracts from 26.80 N to 25.50 N, coincide with the eddy shedding. 240 1. 241 This paper studies an eddy shedding event of the Loop Current in Jul, 2011 using a mpi-version 242 of the Princeton Ocean Model. A real-time forecast was conducted starting from Jul/01, 2011. 243 The model predicted the eddy shedding on Jul/22/2011, consistent with the shedding time during 244 Jul 22-29 from the satellites. To study the underlying dynamics of the eddy shedding, we 245 conducted a free-model run starting from May/15 to eliminate the influence of data assimilation 246 on the ocean state in late May and June for the Jul/01 case. The May/15 run created an eddy 247 shedding on Jul/08. We found that the underlying dynamics of the eddy shedding are: 248 1) 249 fed by the Yucatan influx, and ultimately reached a large size where the Rossby wave velocity (- 250 βRo2) was about -0.05 m s-1. Meantime, the easterly wind stress in the Caribbean Sea decreased 251 and consequently the Yucatan inflow dropped. The Rossby wave speed exceeds the LC growth 252 fed by Yucatan inflow. This provided a favorable condition for the eddy shedding; 253 2) Summary The primary dynamics is the β effect as discussed by Nof (2005). The growing LC was The baroclinic instability north of the Campeche Bank accelerated the eddy shedding; Page | 11 254 3) 255 to the eddy shedding too. 256 Reference: 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 Anderson, D. L. T., and R. A. Corry, 1985: Seasonal transport variationsin the Florida Straits: A model study. J. Phys. Oceanogr.,15, 773–786. Berntsen, J. and L.-Y. Oey, 2010: Estimation of the internal pressure gradient in σ-coordinate ocean models: comparison of second-, fourth-, and sixth-order schemes. Ocean Dyn. 60, 317-330. Chang, Y.-L., and L.-Y. Oey, 2012:Why does the Loop Current tend to shed more eddies in summer and winter? Geophys. Res. Lett., doi:10.1029/2011GL050773, in press. Chang & Oey, 2011: Loop Current Cycle: Coupled Response of the Loop Current with Deep Flows. J. Phys. Oceanogr., 41, 458-471. Cooper, C., G. Z. Forristall, and T. M. Joyce, 1990: Velocity and hydrographic structure of two Gulf of Mexico warm-core rings, J. Geophys.Res., 95, 1663– 1679. Craig, P.D. and M.L. Banner. 1994: Modeling wave-enhanced turbulence in the ocean surface layer. J. Phys. Oceanogr., 24, 2546-2559. DiNezio, P.N., Gramer, L.J., Johns, W.E., Meinen, C.S., and Baringer, M.O., 2009: Observed interannual variability of the Florida Current: wind forcing and the North Atlantic Oscillation, J. Phys. Oceanogr. 39 (3), 721–736 Elliott, B.A., 1982: Anticyclonic rings in the Gulf of Mexico, Journal of Physical Oceanography, 12,1292-1309. Forristall, G. Z., K. J. Schaudt, and C. K. Cooper, 1992: Evolution and kinematics of a loop current eddy in the Gulf of Mexico during 1985, J. Geophys. Res., 97, 2173– 2184. Le Hénaff, M., V. H. Kourafalou, Y. Morel, and A. Srinivasan (2012), Simulating the dynamics and intensification of cyclonic Loop Current Frontal Eddies in the Gulf of Mexico, J. Geophys. Res., 117, C02034, doi:10.1029/2011JC007279. Nof, D., 2005: The momentum imbalance paradox revisited. J. Phys. Oceanogr., 35, 1928-1939. Oey, L.-Y., Ezer, T., Forristall, G., Cooper, C., DiMarco, S., Fan, S., 2005: An exercise in forecasting loop current and eddy frontal positions in the Gulf of Mexico. Geophysical Research Letters, 32, L12611. Pichevin, T., and D. Nof, 1997: The momentumimbalance paradox. Tellus, 49, 298–319. Reid, R. O. (1972), A simple dynamic model of the Loop Current, in Contributions on the Physical Oceanography of the Gulf of Mexico, Tex. A&M Univ. Oceanogr. Ser., vol. 2, edited by L. R. A. Capurro and J. L. Reid, pp. 157–159, Gulf, Houston, Tex. Yin and Oey, 2007: Bred-Ensemble Ocean Forecast of Loop Current and Eddies. Ocean Modelling, 17, 300-326. Oey, 2008: Loop Current and Deep Eddies. J. Phys. Oceanogr. 38, 1426-1449. Oey, L.-Y., T. Ezer, G. Forristall, C. Cooper, S. DiMarco and S. Fan, 2005. An exersise in forecasting loop current and eddy frontal positions in the Gulf of Mexico. Geophys. Res. Let., 32, L12611, 10.1029/2005GL023253. Sturges, W. (1994), The frequency of ring separations of Loop Current, J. Phys. Oceanogr., 24, 1647–1651, doi:10.1175/1520-0485(1994) 024<1647:TFORSF>2.0.CO;2. Sturges, W., and R. Leben (2000), Frequency of ring separations from the Loop Current in the Gulf of Mexico: A revised estimate, J. Phys. Oceanogr., 30, 1814–1819, doi:10.1175/15200485(2000)030<1814: FORSFT>2.0.CO;2. The upwelling induced by the upper-lower layer coupling in the eastern Gulf contributed Page | 12 298 299 300 Vukovich, F. M. (1988), Loop Current boundary variations, J. Geophys. Res., 93(C12), 15,585– 15,591, doi:10.1029/JC093iC12p15585. Page | 13