Limnol. Oceanogr., 54(6), 2009, 2041–2057 2009, by the American Society of Limnology and Oceanography, Inc. E Flow paths and spatial heterogeneity of stream inflows in a small multibasin lake Francisco J. Rueda,a,* and Sally MacIntyreb a Instituto del Agua and Departamento de Ingenierı́a Civil, Universidad de Granada, Granada, Spain of Ecology, Evolution, and Marine Biology, and Marine Science Institute, University of California, Santa Barbara, California b Department Abstract We describe the flow paths of negatively buoyant river inflows in a small lake with multiple basins separated by sills (Toolik Lake, Alaska) using field data and three-dimensional simulations. Comparisons of field observations, analytical computations, and simulations show that in small basins in which the timescale for filling (tf) is less than the timescale of an event (t0), overflow is the dominant mechanism of interbasin exchange. Exchange is mediated by internal wave displacements if upwelling reaches the height of the sills, zs. This criterion is met if the Lake number, the inverse of which is proportional to the degree of metalimnetic tilting, is less than zi /(zi 2 zs), where zi is the vertical displacement of the intrusion. Entrainment followed by horizontal dispersion is the dominant mechanism of interbasin transport and can account for ,65% of the exchange between large subbasins. Entrainment is enhanced when submerged sills force river water to flow close to the surface layer. Depths of intrusions depend upon discharge. They occurred in the lower metalimnion for the coldest events analyzed but were near the top of the metalimnion during low-discharge events, such as during brief cold fronts. Persistence of intrusions depends upon the time interval between cold fronts; they ranged from a few days to three weeks. Timescales of horizontal mixing vary with meteorological forcing; they ranged from on the order of 1 d for wind events with speeds of up to 6–8 m s21 to on the order of 10 d for light winds. Storm-inflow events are one of the major disturbances affecting the seasonal and long-term behavior of lacustrine ecosystems (Robarts 1987; Elber and Schanz 1990; Barbiero et al. 1999). These events provide a large fraction of the annual nutrient loading. For example, Inamdar et al. (2006) found that the largest storms contributed 15% of the annual discharge but one-third of the annual inputs of ammonium, dissolved organic nitrogen, and dissolved organic carbon in glaciated lakes in New York. MacIntyre et al. (2006) estimated that loading of inorganic nitrogen from one storm event in an oligotrophic arctic lake was ,34% of the loading that occurs in more typical years. The response of aquatic ecosystems to such storm-runoff events largely depends on the fate of the incoming storm water and the associated dissolved and particulate constituents. It is important to know whether storm water forms intrusions and, if so, the depth and persistence of the intrusions. If the intrusions are enriched with nutrients or labile organic matter and are introduced into the surface layer, they will immediately be available for phytoplankton or bacterial growth. Alternatively, if they are introduced into the metalimnion, they may fuel growth within a chlorophyll maximum or be below the euphotic zone. Then, they may be available to phytoplankton if mixed vertically at another time, or they may be sequestered via adsorption to particulates or consumption by bacteria. Further, if nutrients, dissolved organic carbon, particulates, or organisms are introduced in discrete intrusions that persist, layered communities may result. Hence, an understanding of the flow paths of stream inflows is essential assess their importance to lake ecosystem function. * Corresponding author: fjrueda@ugr.es The pathways of storm river water depend on a lake’s geometry, stream density relative to lake water, stream hydraulics, the hydrodynamics and mixing away from river inlets, and the ambient stratification (Fischer and Smith 1983; Akiyama and Stephan 1987; Johnson et al. 1987), which can vary both as a consequence of meteorological forcing and the stream flow itself. In long and narrow lakes and reservoirs with simple geometries in which lateral motions are not restricted, the pathways of distribution of stream water and their relevant timescales and spatial scales have been well described in the literature. Our understanding of the fate of river plumes in these cases is based on an extensive body of literature that includes laboratory experiments (Ellison and Turner 1959; Britter and Linden 1980; Hallworth et al. 1996), analytical work (Savage and Brimberg 1975; Jain 1981; Hauenstein and Dracos 1984), numerical simulations (Chung and Gu 1998; Bournet et al. 1999; Kassem et al. 2003), and field data (Hebbert et al. 1979; Fischer and Smith 1983; Dallimore et al. 2001). For negatively buoyant inflows (i.e., river water that is denser than lake water), the stream water initially pushes the stagnant lake water ahead of itself until it is arrested by buoyancy forces due to the difference in density between lake and stream water. At this point, the colder stream water will plunge beneath the surface. Once submerged, it will flow downward along the bottom as a gravity-driven density current, gradually entraining water until it reaches the level of neutral buoyancy, where the densities of the flowing current and the ambient fluid are equal (Stevens et al. 1995; Ahlfeld et al. 2003), or the bottom of the basin (Hebbert et al. 1979; Finger et al. 2006). In the first case, the density current separates, forming intrusions that spread horizontally. Once the intrusions have formed, their fate is tightly linked to vertical mixing in the water column. 2041 2042 Rueda and MacIntyre Fig. 1. Interbasin exchange mechanisms along a longitudinal section of Toolik Lake: (a) overflow, (b) internal displacement, and (c) entrainment due to wind mixing and heat loss. That is, if vertical mixing is weak or infrequent, intrusions may remain at depth for extended periods of time. If winds or cooling intensify mixing and deepening of the upper mixed layer, the intrusions could be entrained into the surface mixed layer. Lakes with multiple basins occur frequently in the landscape. The sills found at depth within them will modify the flow path of incoming water. In order to develop generalities about the movement of incoming water, the resulting intrusions, and the persistence of spatial heterogeneity in the vertical and horizontal directions, the dominant mechanisms of cross-basin exchange within multibasin lakes must be determined and put into the context of plume dynamics, within-lake stratification, and meteorological forcing. With negatively buoyant plumes, exchange between basins could occur as the incoming water fills basins sequentially along a flow path (Fig. 1a) due to mechanisms similar to seiche-pumping (Van Senden and Imboden 1989), in which the up and down motions of internal waves facilitate exchanges (Fig. 1b), or via entrainment of intruding water and its subsequent lateral transport (Fig. 1c). The presence of topographic features may further moderate exchange because these features will force deep intrusions closer to the surface, where they may be entrained into the surface layer. The importance of these processes to cross-basin exchange in small multibasin lakes is currently unknown and likely depends upon discharge, the intensity of mixing as the incoming water first enters a lake, the size of subbasins relative to the volume of incoming water, and the ambient physical forcing. Due to the differences in composition between lake and storm water, the occurrence of storm-inflow events will likely induce spatially variable concentration fields in the lake. The persistence of heterogeneous fields in the lake will depend upon the incidence of vertical and horizontal mixing. Horizontal heterogeneity at depth will develop when the lateral spreading of intrusions is restricted or due to changes in inflow characteristics during events (Fernández and Imberger 2008). Horizontal heterogeneity in the surface layer, in turn, can be created as a result of wind forcing inducing preferential flow paths of incoming water. It is also created after entrainment of deep intrusions that have not spread horizontally due to restricted interbasin transport. The efficacy of vertical mixing entraining the intrusions into the surface layer will depend upon the magnitude of cooling coupled with storm-related factors that determine the depth of the intrusions. Persistence of horizontal patchiness in the surface layer depends on turbulent diffusion in the horizontal direction caused by fluctuations in the velocity field smaller than the cloud or patch size of entrained solutes; horizontal velocity gradients at scales larger than patch size, which, together with small-scale horizontal fluctuations, cause horizontal shear dispersion; and vertical shear dispersion from vertical gradients in the flow field associated with small-scale vertical velocity fluctuations (Fischer et al. 1979; Stocker and Imberger 2003). Stocker and Imberger (2003) argued that dispersion from horizontal shear should be the dominant mechanism of dispersion in small- to mediumsize lakes. The presence of submerged features could make aspects of the larger flow field more important. In the following, we use simulations conducted with a three-dimensional (3-D) hydrodynamic model supported by experimental data collected in situ to describe and understand the behavior of storm river inflows and the spatial variability they induce and to develop generalities. Our goals include describing mechanisms that cause interbasin transport, assessing the conditions under which each process is dominant, and characterizing the depth and persistence of the resulting intrusions as a function of stream hydraulics, thermal stratification, and changing meteorological forcing. Secondly, we quantify the timescales of the physical processes involved in the transport and mixing of storm river and lake water to enable predictions for multibasin lakes of different sizes. Our effort is based on the analysis of field data and modeling results from a multibasin lake, Toolik Lake, Alaska (Fig. 2). Study site Toolik Lake, Alaska (68u389N, 149u389W), lies in the northern foothills of the Brooks Range. It is an oligotrophic kettle lake with a surface area of 1.5 km2, mean depth of 7.1 m, and maximum depth of 25 m (Fig. 2). Shifts between warm and cold air masses occur frequently; cold air masses typically come from the north (Miller et al. 1986; MacIntyre et al. in press) and may lead to increased rainfall and high discharge events (MacIntyre et al. 2006). The diffuse attenuation coefficient ranges from 0.5 to 0.9 m21 Pathways of stream inflows in lakes 2043 periods of time (which we will refer to as the study period): one from day 190 to day 210 (1999), and the second from day 188 to day 225 (2004). The model was forced using hydrological and meteorological information collected in the field, and its results were validated by comparing temperature simulations against the field data. The model velocity fields were used to drive the simulations of tracer releases during four inflow events that occurred during the study period. One of the events occurred in 1999, and the other three occurred in 2004 (events 2004-1, 2004-2, and 2004-3). All simulations started at least 4 d before the inflow events. Hence, the effects of the artificial initial conditions on model results were deemed negligible during the storm events. Experimental and numerical results from 1999 were used to analyze pathways of riverborne substances across moraines and to the surface layers and to quantify the importance of different mechanisms of transport. The 2004 data and simulations extend our understanding of the effects of varying forcing conditions on the pathways of storm river water in Toolik Lake. Fig. 2. Bathymetric map of Toolik Lake, with isobaths shown every 2 m. Gray areas mark shallow zones, or moraines, which separate the Toolik basin into subbasins: inlet, main, and west basins. The symbols mark the approximate locations where microstructure profiles (MacIntyre et al. 2006) and/or thermistor chains were deployed and sites where model results were checked (IB 5 inlet bay; TM 5 Toolik main; CN 5 Central; WB 5 west basins; SWB 5 southwest basin). In 1999, thermistor arrays were located at TM and CN. In 2004, arrays were located at each symbol. Also shown in the figure is the M-axis, a line that originates at the inlet basin and is perpendicular to the moraine that separates the main from the west basins. (Miller et al. 1986). The basic limnology of the lake is reviewed in O’Brien et al. (1997). Its geometry is characterized by the presence of several subbasins that are separated by shallow, ,3-m-deep moraines. The smallest subbasin to the southeast of the lake is the Inlet Bay, and it is also the one that receives the largest inflow. The next basin, in the direction of the flow from inlet to outlet, is the main basin. Its long axis runs from northeast to southwest. Its deepest point, ,25 m, is toward the south; its eastern portion is nearly flat and shallow (,4–6 m) with some small depressions. The west basin is adjacent to the main basin and is separated from it by shallow moraines arranged from SW to NE. The southwest basin is further separated from the west basin by a small promontory. The lake’s outlet is to the northeast. Methods Approach—We studied the space–time distribution of storm river water as it flows into Toolik Lake through the analysis of field observations and a series of simulated tracer experiments conducted with a 3-D hydrodynamic model. Data and simulations correspond to two different Meteorological and temperature methods—Meteorological instrumentation for both years and temperature sensors used in 1999 are described in MacIntyre et al. (2006). In 2004, Brancker TR-1050 self-contained temperature loggers with accuracy of 0.002uC, resolution of 0.0001uC, and time constant of 3 s were deployed at three stations: inlet bay (IB), Toolik main (TM), and Central (CN) (see Fig. 2). Onset Stowaways, with accuracy of 0.2uC and time constant of 1 min were deployed in the west basin (WB), southwest basin (SWB), and an additional station between CN and the outlet. Analysis of the meteorological and water temperature data followed MacIntyre et al. (2002, in press). Model description—Numerical simulations of circulation and transport in Toolik Lake were conducted with a threedimensional (3-D) free-surface hydrodynamic model (Smith 2006), which has been extensively validated with both analytical solutions and field data (Rueda and Cowen 2005). The model is based on the continuity equation for incompressible fluids, the Reynolds-averaged form of the Navier-Stokes equations for momentum, the transport equation for scalars, and an equation of state relating the concentration of active scalars to fluid density. The governing equations are solved in layer-averaged form using a semi-implicit, three-level, iterative leapfrog-trapezoidal finite-difference scheme on a staggered Cartesian grid. Turbulent mixing is represented in the 3-D model using diffusion-like terms. A Laplacian operator with constant mixing coefficients (horizontal eddy viscosity Ah or diffusivity Kh) is used in the model to represent horizontal mixing of momentum and scalars (see below). Vertical eddy coefficients of mixing Kz are calculated using a two-equation turbulence model (Kantha and Clayson 1994). Nonactive (i.e., tracers) scalar transport equations are solved using a two-level semi-implicit scheme, in which only vertical diffusion is discretized implicitly. The advection terms in the transport equation for nonactive scalars are discretized with flux-limiter methods (Durran 1999). 2044 Rueda and MacIntyre Fig. 3. (a) Lake numbers with times when wind was from the northwest identified with gray histograms; dashed line indicates LN 5 3. (b) Short-wave radiation, (c) air and surface-water temperatures (thick black line), (d) relative humidity, (e) wind speed, (f) wind direction, (g) heat fluxes (LW 5 long-wave radiation; L 5 latent heat flux; S 5 sensible heat flux), (h) stream discharge Q and rainfall (note the rain is inverted), (i) stream temperature and specific conductance SC (conductivity normalized to 25uC), and (j) water temperatures. The thick white line in (j) indicates the theoretical depth of neutral buoyancy as if it were only a function of stream temperature for Toolik Lake during the study periods (1999 and 2004). The dashed line marks the level of the moraines separating the different basins. Model setup and validation—The computational model was set up to simulate the processes of transport and mixing in Toolik Lake during the two study periods. The lake was discretized using 20 3 20 3 0.5 m grid cells, and, for stability purposes, the time step was set to 20 s. The model was forced using observed inflow rates and temperature in Toolik inlet, and surface heat and momentum fluxes estimated as in MacIntyre et al. (2002) from local atmospheric variables (Fig. 3). Heat-flux estimates were corrected for atmospheric stability. Mass-balance equations were used to estimate outflow rates from observed water-level fluctuations in the lake, rainfall, and inflow rates. The time series of estimated outflow rates was used as boundary condition at the outflow section. With the grid resolution adopted in the model, the advective terms should be able to resolve transport features on the order of (O) 102 m, which, according to Knauer et al. (2000), is the scale of the eddies that will contribute most to the horizontal mixing in lakes with horizontal length scales of O(103) m. The horizontal eddy viscosity, Ah, and diffusivity, Kh, in the model were set to 2 3 1022 m2 s21. This value corresponds to the lowest limit of the range of values encountered by Peeters et al. (1996) during tracer studies in the metalimnion of lakes with dimensions similar Pathways of stream inflows in lakes 2045 Table 1. Characterization of the four inflow events. Those events were identified as times when the inflow rate exceeded 2 m3 s21. The symbol Qave is the average inflow rate during the event, Qmax is the maximum inflow rate, F0 is the inflow densimetric Froude number, xp is distance to plunge point, and C is dilution rate. Columns 10 and 11 show the length of time that river water took to reach the main and west basins. These values were estimated from model simulations. The final column represents the time to horizontal mixing, th, which was estimated from tracer simulations in which storm river water was traced. The variance of tracer concentration S2(t) in the layer just below the surface was initially zero before the storm; it increased to a maximum value S 2max during the inflow event, and decreased thereafter. Horizontal mixing is presumed to occur by the time S2(t) first drops to 0.05 of S 2max . (1) Start date (2) (3) Duration Qave (h) (m3 s21) 1999 2004-1 198.69 191.88 63 80 7.5 5.6 2004-2 2004-3 200.92 213.46 69 116 3.1 3.6 (4) (5) Qmax (m3 s21) Date of Qmax 16.2 7.7 11.3 5.0 6.2 5.0 199.00 192.25 (193.10) 201.23 213.90 (216.75) (6) Volume discharged at Qmax (m3) (7) (8) (9) (10) (11) (12) F0 xp (m) C 216,950 123,077 5.59 6.07 183 195 2.14 2.21 2 6 11 40 3.19 7.09 66,960 117,850 2.80 3.67 111 133 1.66 1.83 5 5 33 24 4.74 5.59 DtreachMB* DtreachWB* (h) (h) th (d) * The arrival time was defined as the time when the tracer loading to any subbasin reached an arbitrary threshold set to 105 mmol d21. to Toolik’s, but it is the result of applying the empirical law proposed by Lawrence et al. (1995) for a length scale (,) of 40 m (52 3 Dx), which represents the horizontal scale of the smallest flow features that can be resolved by the numerical model. The diffusion terms with Ah and Kh set to 2 3 1022 m2 s21 should therefore represent the effects on mixing due to nonresolved subgrid-scale transport processes. In the model, we presumed that temperature is the only active scalar. We ignore the influence of sediments in suspension and salinity in the density of the water, given the low conductivity values (ranging from 40 to 80 mS cm21; MacIntyre et al. 2006) and total suspended sediment (TSS) concentrations (ranging from 0.4 to 35 mg L21; Kriet et al. 1992) reported for stream water in the nearby but larger Kuparuk River. For a change in TSS of 35 mg L21, the relative change in density Dr/r, estimated as in Finger et al. (2006), is O(1025), which is one order of magnitude smaller than the change in density (Dr/r) that results from changing the river-water temperature by 1uC from a reference value of 10uC (see values of thermal expansion coefficient in Fischer et al. 1979). Our model of Toolik Lake was validated by comparing simulations and observations collected during 1999 and 2004 at station TM (Fig. 2). The mean root mean squared error (RMSE), a measure of the discrepancy between observations and simulations, calculated from daily temperature profiles was 0.25uC in 1999 and 0.15uC in 2004. This level of accuracy is consistent with other 3-D modeling studies in lakes (Hodges et al. 2000; Jin et al. 2000; Rueda and Cowen 2005). Later herein, we illustrate the comparison of the measurements and simulations from four of the temperature arrays deployed in 2004. Comparisons of modeled and computed vertical eddy diffusivities (Kz) indicated that they were similar in the mixed layer and at the depths of intrusions. Further, a comparison of measured and simulated profiles of specific conductance indicates that both gave similar estimates of the vertical dimensions of the intrusions at TM, CN, and WB (Rueda and MacIntyre, in press). Tracer experiments—The fate of river water during the four inflow events that occurred during the study period was analyzed by simulating a series of tracer-release experiments. In this analysis, Toolik Lake was subdivided into three subbasins (referred to in Fig. 2 and throughout this work as inlet bay, the main basin, and the western basins [including both the west and southwest basins]). The amount of simulated tracer existing within each subbasin was monitored during the whole period simulated. The tracer was released in the simulations continuously at all times whenever the river inflow Q was larger than 2 m3 s21 (i.e., Qevent 5 2 m3 s21; see Table 1). The river water was presumed to have a constant tracer concentration C, which was set to 1 mmol m23. These values for inflow threshold (Qevent) and tracer constant concentrations (C), although arbitrary, allowed us to define quantitatively when an inflow event started and finished and to directly quantify the space–time changes in the fraction of river water within the lake. Simulated tracer-release experiments conducted during the inflow event that occurred in 1999, together with field data collected during and after that event and presented in MacIntyre et al. (2006), were first used to study the dominant mechanisms of interbasin exchange and the timescales relevant for river intrusion transport and dispersion. We then analyzed and compared the tracerrelease simulations conducted during the inflow events in 1999 and 2004 to establish the dominant processes involved in dispersing intrusions and to relate the variability in dispersion timescales with differences in hydraulic or freesurface forcing. Throughout, we assessed the consequences of morphometry on the physical pathways of distribution of the river intrusions. Dimensionless indices and mixing rates—Here we introduce definitions and define procedures used throughout the text to estimate the scales of transport and mixing processes affecting storm river water in lakes. Lake–river mixing near the inflow can be parameterized in terms of the densimetric Froude number F0 (Stevens et al. 1995) calculated as F0 5 2046 Rueda and MacIntyre U0/(g9h0)1/2, where U0 is the inflow velocity, h0 is the depth of the river channel, and g9 is the reduced gravitational acceleration. Inflow velocity was obtained by dividing the inflow rate Q by the area of the inflow channel, which was assumed to be 10 m wide and 1.5 m deep. Reduced gravity is calculated as g|r0 2 r|/r0, where g is gravity, and r and r0 are densities of the river and the lake surface water, respectively. The dilution rate C at the plunge point is the ratio between the inflow discharge at the plunge point (after mixing) and the discharge at the inflow section, and it quantifies the initial mixing. An alternative means of defining the rate of river–lake mixing is through the mixing ratio c, which is defined as the fraction of river water in a mixture. Note that the dilution rate C is related to the fraction of river water c in the intrusion as C 5 c21. The initial dilution rate C was estimated from average inflow densimetric Froude numbers and using empirical equations taken from the literature. We first followed Hauenstein and Dracos (1984) to estimate the distance to the plunge point xp in terms of F0 and river channel geometry, as " xp ~ S{1 0 0:774F0 h0 S0 h0 =b0 Rates of horizontal dispersion at the center of Toolik Lake were determined from a series of tracer-release experiments. In these experiments, a Gaussian tracer cloud with standard deviations (sx, sy, sz) set equal to (100, 100, 1) m was released at intervals of 6 h, where the center of mass was located ,1 m below the free surface and near the geometric center of the lake. The original size of the cloud was such that the most energetic eddies, which should be of O(100) m for lakes of the size of Toolik Lake (Knauer et al. 2000), would be effective in modifying the shape of the cloud. The evolution of the tracer concentration field was followed for 6 h after the releases. We defined the time varying horizontal extension of the vertically integrated concentration distribution, s2(t), as twice the product of sma and smi, where s 2ma and s 2mi are the variances of the vertically integrated tracer concentration in the direction of the major and minor principal axis, respectively. The apparent dispersion coefficient was then estimated, from the initial and final size of the cloud after 6 h of simulations, as (Peeters et al. 1996) # 1=4 z 1:16 Kapp ~ ð1Þ ð3Þ Results where S0 is the bottom slope, and h0 and b0 represent the depth and width of the river upstream from the lake, respectively. This equation applies for large divergence angles, as found in many lakes, where inflows initially behave as jets and entrain water along their margins until they plunge, and for large densimetric inflow Froude numbers (F0 . 1). We then used Eq. 4 in Johnson et al. (1989) to estimate C in terms of the ratio xp /b0, as C < 0.5(xp /b0)1/2. The fraction of river water in the intrusions (c) was estimated from changes in the observed conductivities following McNamara et al. (1997). Entrainment rates into the surface-mixed layer due to heat losses at the free surface were computed using the onedimensional mixed-layer model of Fischer et al. (1979). In that model, the rate at which the depth of the mixed-layer h increases with time t is determined as h i dh CTf w2 z gaDTh ~ q3 CKf dt CTf ~ 0:50,CKf ~ 0:13 1 Ds2 4 Dt ð2Þ where w is a velocity scale of thermals generated by * cooling, DT is the temperature change at the bottom of the mixed layer, and q is a velocity scale defined in terms of w * and the wind friction velocity u , as q 5 (w 3 + 1.23u 3)1/3. * * * The thermal coefficient of expansion, a, was taken as 1.51 3 1024 uC21, which corresponds to water at 10uC (Fischer et al. 1979). Shear-induced mixing was not included, so our calculations of entrainment are conservative. The Lake number was calculated as in MacIntyre et al. (1999). This dimensionless index is an integral form of the Wedderburn number (Imberger and Patterson 1990), which provides an estimate of the degree of thermocline tilting in response to the wind (Stevens and Imberger 1996; Stevens and Lawrence 1997; Horn et al. 2001). The variations in discharge in the four storm events investigated are illustrated in Fig. 3, and additional details are given in Table 1. The volume of water discharged was similar in the event in 1999 and in the first and third events in 2004, whereas the volume discharged in the second event of 2004 was ,50% of the other three. The durations of the events varied considerably: the shortest lasted 63 h, and the longest lasted 116 h. Consequently, average discharge varied by a factor of two. In addition, the rate of increase in discharge at the onset of the storms varied, and volumetric flow rates increased most rapidly in 1999 (Fig. 3). Together, these factors affected maximum discharge, which varied from 5 to 16 m3 s21. The four discharge events all occurred during cold fronts, accompanied by concomitant increased cloudiness and higher relative humidity (Fig. 3b,d). Importantly, the degree of cooling varied between events. Heat losses during 1999-1, 2004-1, and 2004-3 events exceeded 250 W m22, whereas the heat loss in 2004-2 was ,100 W m 22. Related to these differences, stream temperatures determining inflow density dropped below 10uC in the colder events but not in the warmer 2004-2 event. The average densimetric Froude number (F0) during the inflow events, which parameterizes the mixing associated with the inflow (Johnson et al. 1989; Stevens et al. 1995), was F0 .1, which suggests that inflow mixing was vigorous in all events. It was ,3 during 2004-2 and 2004-3, and 6 during 2004-1 and 1999 (Table 1). Estimated dilution rates (C) before the plunge point were ,2 (Table 1). C was larger for events with the largest inflow rates. Wind speeds were higher in the three colder events, and prolonged periods with winds over 6 m s21 were observed before and sometimes during the event. The upper mixed layer at Toolik main was 4–5 m deep at the onset of all events, and the metalimnion was strongly stratified. Despite the stratification, wind forcing was sufficient to Pathways of stream inflows in lakes 2047 Fig. 4. Observed and simulated thermal structure in (a, e) IB, (b, f) TM, (c, g) CN, and (d, h) WB from day 190 to day 223 (2004). The right frames represent simulated results, while the frames to the left are drawn from observations. Isotherms are shown every 1uC. The 6uC is marked on frame (b). drive the Lake number to values O(1) for several hours in all events but 2004-2 (Fig. 3), which is indicative of upand downwelling of the thermocline and intensified mixing in the metalimnion (Boegman et al. 2005; MacIntyre et al. 2006, in press). As a consequence of heat loss and increased wind forcing at the start of the events, the mixed layer deepened by at least 1 m. The temperature time series at IB, TM, CN, and WB provide insight into the flow paths of the storm water in 2004, as well as model validation (Fig. 4). During the two colder events with higher discharge, overall temperature of the inlet basin became much colder. In 2004-1, the isotherms spread vertically above and below 6-m depth, suggesting that the river inflow intruded initially at that depth. The upward deflections of the isotherms in 2004-3 occurred above 8 m depth, which suggests that incoming water flowed deeper at that time. The abrupt vertical deflections of isotherms in both events indicate that incoming water quickly filled the inlet basin. The temper- ature gradients at the depths of the intrusion indicate partial mixing of incoming water with ambient water. In 2004-2, isotherm spreading occurred near 4 m and was restricted to about 1.5 m. The thermocline thickened in 2004-1 at TM and CN such that the upper mixed layer was only 2 m deep at those stations but was 4 m deep at WB. From these differences, we infer that much more storm water intruded into the main basin than into the west basin. The magnitude of the changes in isotherm depths was so small in 2004-2 that we cannot assess the flow path of the intrusion from temperature data alone. The 2004-3 inflow event was accompanied by intense cooling at its onset, which would have entrained storm water into the mixed layer. However, the subsequent spreading of the isotherms between 3- and 8-m depth suggests that storm water intruded at those depths. These observations suggest that storm river water may form large intrusions in the main basin, but the dispersion of storm water into the west basin is limited by the presence of shallow sills. 2048 Rueda and MacIntyre Comparison of model and observations at multiple sites— A comparison of the temperature records collected in summer 2004 at multiple locations with the results of our simulations shows that the temperatures, temperature gradients in the metalimnion, and, in general, the longterm evolution of stratification in the water column are all well-captured by the model (Fig. 4). The model captured the temporal patterns of heating and cooling with respect to the storms, the cooling in IB due to the large stream inflows, and the separation of the isotherms in the main basin that occurred in response to storm inflows. Discrepancies included lower temperatures in the epilimnion and more pronounced diurnal heating and cooling. The model predicted water temperatures about a degree colder than values measured in IB near the depth of the sill. The overall successful comparison of field observations and modeled temperature structure indicates that the model can be used to assess the mechanisms enabling interbasin exchange and to quantify how the exchange flow varies in response to different forcing. Fig. 5. (a) Conductivity profiles collected with a temperature-gradient microstructure profiler at TM at different times in 1999. (b) Conductivity profiles collected in different locations along the M-axis in Toolik Lake on day 200 (2004). The locations of the profiles are shown in Fig. 1. (c–d) Fraction of storm river water derived from the conductivity profiles shown in frames (a) and (b), respectively. The fractions were calculated using a mixing model (see Methods) in which it was presumed that the conductivity of lake (CL) and storm river water (CR) were constant. Different fractions are obtained depending on the values assumed; the range of values obtained taking into account the variability in CL and CR is indicated by the shadowed areas. CL was varied from 62 to 66 mS cm21, which corresponds to the minimum and maximum values of specific conductance measured in TM on day 190 prior to the storm event. CR was, in turn, varied between 38 and 45 mS cm21, the minimum and the maximum values of stream conductivity during the inflow event. Profiles of specific conductance (SC) taken in 1999 illustrate the flow paths during the early part of the storm in that year (Fig. 5a,b). From those profiles and using a mixing model (McNamara et al. 1997), we computed the fraction of river water at each depth. Specific conductance in the inlet basin 13 h after peak discharge was depressed below 3-m depth and was 80% river water (Fig. 5b,d). At TM and near the far moraine in the main basin, SC was depressed between 2 and 6 m, and the intrusion was 40% to 60% storm water (Fig. 5a,c). In contrast, on the other side of the moraine at FJ, SC was low, between 3 and 5 m (Fig. 5b), and the intrusion was less than 20% storm water (Fig. 5d). Interestingly, a SC profile at station FI on the far side of the west basin (Fig. 2) showed an intrusion at the same depth as at FJ, as well as one centered at 2 m. The higher concentrations of storm water at FI provide evidence for an additional pathway of storm-water inflow. Hydraulically forced exchange: Overflow mechanism— The length of time until water spills into adjacent basins by overflow is determined by the particular bathymetry of the subbasin, the inflow rate, the depth of the intrusion, and the rate that inflow water mixes with ambient water. For the average inflow rate during the 1999 event, the inlet basin would fill in ,3 h at depths between the top of the sill and the bottom of the intrusion (see Tables 1, 2). After that time, river water would start overflowing over the moraines to adjacent basins. Similar computations for the 3–8-m depth range in the main basin, and assuming no vertical mixing, indicate that the layer would fill in 2–3 d (Tables 1, 2). Our simulations, however, indicate that spillage over the moraine to the west basins occurred only after 6 h. The shorter time may be due to dilution and enlargement of intrusions as inflow and lake water mixed (Fig. 5). The fraction of river water (c) in the intrusions formed in TM, estimated from the observed conductivities, was approximately 0.44. This estimate agrees, first, with the simulations (Rueda and MacIntyre, in press) and, second, with our estimates of lake–river mixing rates obtained from the average values of inflow densimetric Froude numbers F0 (Table 1). The filling timescale, including the effects of the initial lake–river mixing, can be estimated as tf ~ V c Q ð4Þ where Q is the average inflow rate, and V is the volume of the basin between the theoretical intrusion depth and the top of the sills. Assuming that c is constant, tf is ,20 h. These results indicate that initial mixing may significantly reduce the transit time between basins. The estimates of transit time, however, are larger than the observed and simulated values (Table 3) and indicate that overflow may not be important for large basins during the first stages of a storm, and other mechanisms of horizontal exchange, as described in Fig. 1, could be operative. Pathways of stream inflows in lakes Table 2. Estimates of the time to fill the volume between two depths (receiving layer) with and without mixing with ambient water in the main and inlet basins (Fig. 2) during the 1999 event. The estimates were constructed by first dividing the volume of the receiving layer by the average inflow rates during the 24-h period from 12:00 h on day 198 to 12:00 h on day 199 and multiplying the result by a mixing ratio c. The mixing ratio defines the fraction of river water in the intrusion mixture (c 5 1 if no mixing occurs; c 5 0.44 when mixing is taken into account). After this filling time, river water would start to flow over the moraines separating the main and the west basins. Depth ranges for receiving layer (m) Vlake (m3) tf (h) no mixing tf (h) mixing IB 3–8 MB 3–8 MB 3–7 MB 3–6 MB 3–5 MB 3–4 71,320 2,396,200 2,046,400 1,659,800 1,205,000 625,600 2.7 56 48 39 28 15 25 21 17 12 6 Internal displacements driving inter-basin exchange— Winds during the inflow events analyzed tended to be from the NW and perpendicular to the moraines separating the main and west basins (Fig. 3). The tilting of the metalimnetic intrusion in the model (Fig. 6a,b,c) and the increases in simulated concentrations of tracer at 8-m depth in the west basin (Fig. 6d,e,f) after such an event provide evidence for this mechanism. With a two-layered model of the lake and the assumption that rotational effects and variability in wind stress are negligible, the amplitude of internal waves required for overflow into the next basin can be computed following Monismith (1987) and Horn et al. (2001). The magnitude of the thermocline displacement zmax is zmax 5 2L/Ri*, where L is the fetch length of the lake, and Ri* is the bulk Richardson number, given by Ri ~ agDTh u2 ð5Þ Here h is the depth of the water column above the intrusion, DT is the temperature difference between the 2049 intrusion and the surface layer, and u is the water friction * velocity due to wind stress. For h < 5 m, DT < 5uC, as observed on day 199 (1999) in Toolik main, and letting L < 1000 m, the minimum wind speed to induce the 2-m isotherm displacements large enough to bring river intrusions over the 3-m-deep moraines and into the west basin is 6–7 m s21. From 00:00 h to 12:00 h on day 199 (1999) (Fig. 3), winds exceeded this threshold. During that time, the tracer loads into the west basins reached a maximum (Fig. 7f), and intrusions appeared in WB (Fig. 6c). A threshold for exchange by internal displacement can be similarly defined in terms of the Lake number LN. If we assume that the metaliminion is centered at 5 m, the threshold value for a displacement to reach the depth of the sill and then overflow is LN < 3. Lake numbers dropped to values of 2 for several hours prior to peak discharge in 1999 (Fig. 3), confirming that this dimensionless index can be used to infer when interbasin transport occurred via the displacement mechanism. As a further test of the importance of wind-induced upwelling for interbasin exchange, we ran simulations in which the surface shear stress was reduced to 1% of its measured value. With this reduction, the tracer was primarily confined to the metalimnion in the main basin, and concentrations in the west basin were only 14% of those computed with the actual wind forcing (Fig. 6g,h,i). Thus, internal wave displacements, as opposed to overflow, mediated transport across submerged features between larger basins for at least the initial stages of storm events. Entrainment of shallow intrusions—When river water intrudes just below the mixed layer, as occurs when it fills the inlet basin and starts flowing over the moraines into the main basin (Fig. 4), it can be entrained into the surface mixed layer. The likelihood of entrainment can be analyzed using the mixed-layer model of Fischer et al. (1979). During the first part of day 199 (1999) and shortly after the peak discharge, the lake was subject to ,6 m s21 winds and lost energy at a rate of < 200 W m22. Subject to that forcing, the model in Eq. 2 predicts that a 2-m-deep surface mixed layer (Fig. 3) overlying inflow water 5uC colder would deepen and entrain river water at a rate dh/dt of ,7.5 3 Table 3. Persistence (tintr) of intrusions; averaged depth for the top of the intrusion (zintr-min), averaged depth of their bottom (zintr-max), and averaged location where maximum concentration is achieved (zintr-avg). Note, because of the averaging, these data do not show the full depth range of the intrusions. Intrusions were identified from inspection of one-dimensional vertical projection of the tracer concentrations fields C1D(z). An intrusion was deemed to exist if the following conditions were met: C1D(z) has a maximum at a given depth below the surface, zintr, and the value of maximum aerial averaged concentration C1Dmax is at least 10% larger than the concentration calculated by averaging C1D(z) from zintr to the surface C1Davg. The intrusion layers were identified as the layers where the tracer concentration was at least 0.80 times the maximum value of C1D at each time. Basin IB WB MB MB MB MB Scale 1999 2004-1 2004-2 2004-3 tintr (d) tintr (d) tintr (d) zintr-max (m) zintr-min (m) zintr-avg (m) 11.2{ 22.2* 13.2* 1.4 3.4 6.89 4.39 6.03 10.6 20.5 8.6 4.2 7.23 1.0 3.7 5.32 2.88 4.66 11.5{ 0.5 4.7 7.94 4.70 6.67 * Washed out during third event in 2004. { Limited by the extent of simulation. 2050 Rueda and MacIntyre Fig. 6. Tracer concentration field predicted by the model at 03:00 h on day 199 along three longitudinal sections parallel to the M-axis (see Fig. 1): (a) a section 200 m north of the M-axis, (b) a section along the M-axis, (c) a section 200 m south of the M-axis. (d–f) Same as (a–c) but at 13:00 h on day 199 (1999). (g–i) Tracer field predicted by the model at 13:00 h on day 199 (1999) for a synthetic wind field, in which wind speeds were reduced to 10% of their measured values on days 198 and 199 (1999). Note that the shear stress, being proportional to the square of the velocity, is only 1% the value of the ‘‘measured shear stress’’ when the velocities are multiplied by the factor 0.1. In all simulations, the heat fluxes were set to the measured values. 1026 m s 21. During the 12–14-h period when these conditions prevailed, ,37% of the river water just below the mixed layer could have been entrained and rapidly mixed into the surface layers. If the shallow incoming layers had a tracer concentration of 1 mmol m23, the concentration of the tracer in the 2-m-thick surface layers of IB after mixing vertically would be ,0.2 mmol m23. This value is similar to our modeled results, 0.25 mmol m23, and implies that 20% of the water in the surface layer in IB would be river water. These values are consistent with the observations, which suggest that a non-negligible fraction of lowconductance river water might have reached the surface layers in the inlet (Fig. 5d). As a consequence of entrainment and subsequent surface overflow, nearly 30% of the total mass of tracer in the main basin at 12:00 h on day 199 (1999) was within the upper 3 m of the water column. The remaining 70% was between 3 and 8 m (see also Fig. 5a,b). After 12:00 h on day 199 (1999), the wind ceased, and the cooling rate decreased. As a result of the reduced entrainment, the river water flowed mainly as an intrusion between 3 and 8 m, and the percentage of tracer mass between 3 and 8 m by 00:00 h on day 200 increased to reach ,80%. These results indicate that the percentage of inflowing water in the surface layer versus metalimnion strongly depends on the extent of heat loss and wind mixing throughout the storm period. Entrainment of deep intrusions—Once river water has reached its depth of neutral buoyancy zintr, the resulting intrusions will persist (see Table 3) until the upper mixed layer deepens to reach zintr. The intrusion will then be rapidly incorporated in the surface layers, where it can be freely transported to reach other basins. The depth of the surface mixing layer was defined, from modeled values of eddy diffusivity Kz, as layers in which Kz . 1025 m2 s21. In the field data, it was defined as the depth to which temperatures were within 0.02uC of the surface temperature. These two methods gave comparable estimates of mixed-layer depth. In 1999, the average diffusivity Kz in the surface mixing layer was O(1022) m2 s21. With an estimated mixing time of h2/Kz, an 8-m-thick layer with that diffusivity will mix in ,2 h. Hence, once the bottom of the mixed-layer reaches the intrusion, it will be dispersed rapidly. During the event in 1999, the mixedlayer depth was shallower than the depth of the intrusion in the main basin until day 201 (Fig. 7b), and, consequently, tracer remained confined in a layer between 3 m to 8 m. On day 201, maximum heat fluxes into the lake were at most 100 W m22, fluxes out of the lake exceeded 200 W m22, and NW winds averaged ,6 m s21 (Fig. 3). Toward the end of day 201, the surface mixing layer reached 6.5 m. As a result, tracer concentrations decreased in the upper 7 m of inlet bay and the main basin and increased rapidly in the west basins (Fig. 7). Sixtyfour percent of the total mass of tracer in the WB at the end of the simulations entered during the 2-d period from day 201 to 203. The two deep-water exchange mechanisms that occurred on days 199–201 only transported 32% of the total mass that reached WB (Fig. 7f). Thus, deep vertical mixing combined with lateral wind-driven transport and horizontal mixing effectively transfers inflow water across moraines and can be responsible for most of Pathways of stream inflows in lakes 2051 the exchange to basins away from those that receive the initial inflows. Fig. 7. (a–c) Area-averaged tracer concentration in (a) inlet bay, (b) Toolik main basin, and (c) west basins for the tracerrelease experiments in 1999. The time is represented on the x-axis, the depth is on the y-axis, and the color represents the areaaveraged tracer concentration (note, scale of color bar changes for each panel). Frame (b) also shows the evolution of a mixing-layer depth (white line). Black lines in (a–c) show the evolution of the top, bottom (discontinuous), and center (continuous) of the intrusion structure. The vertical extent of the intrusion layer was identified from an analysis of a one-dimensional vertical projection of the tracer concentration fields C1D(z). The intrusion layers were identified as the layers where the tracer concentration was at least 0.80 times the maximum value of C1D at each time. (d–f) Time series of tracer loads into (d) inlet bay, (e) Toolik main basin, and (f) west basin. Tracer mass loadings were estimated on a 6-h basis by subtracting the mass existing at any given time and Influence of discharge and surface forcing on the fate of storm river water—The variations in meteorological and hydrological forcing among the four events allow us to assess the effects of different forcing on the interbasin transport rate (Table 1) and depth and persistence of intrusions (Table 3). The sequence of events in 2004, furthermore, allows us to analyze the influence of subsequent storms on the fate and dispersion of previously formed intrusions. The depth of the intrusion is largely set by the stream temperature, the volume of introduced storm water, and the mixing of the introduced water with that in the inlet bay. As a result, intrusion depths in the main basin centered between 6 and 7 m for the three larger storms and at 4.7 m for the smallest one (2004-2). Note that attempts to predict mean intrusion depth based on stream temperatures alone (Fig. 3) would be at least 1 m too deep. This discrepancy attests to the importance of mixing dynamics in the inlet bay and modification of the temperature, and thus density, of the storm water as water flows over the sills between basins. Travel times in 1999 were two to four times faster than during the events in 2004 (Table 1). These faster times were driven by the greater discharge. However, for discharges between 5 and 11 m3 s21, travel times from TM to WB were independent of discharge and varied between 24 and 40 h. Due to lower discharge, wind forcing, and heat losses in the first two events in 2004, the overflow, internal displacement, and entrainment mechanisms were all slower than in 1999 and 2004-3. Consequently, travel times were longer (Table 1), and tracer persisted longer within intrusions in TM (Table 3). Wind forcing and rate of cooling in 2004 were greatest during the third event (event 2004-3). Winds were from the northwest and Lake numbers were near 1 for the first 24 h of the event. The net heat flux during the first hours of day 214 was 2400 W m22, and it remained negative almost until 08:00 h on day 215. Hence, not only internal displacement, but also entrainment led to faster rates of transport from TM to WB relative to the first two events in 2004 (Table 1). The persistence of intrusions of stream water in the lake varied with the proximity of each basin to the incoming stream, with the physical forcing during and after the time stream water flowed into the lake, and with the proximity in time of storm events that could flush stream water out of a basin (Table 3). For instance, intrusions tended to persist longer in the inlet basin than at the main basin. The intrusion in the inlet basin from 2004-1 persisted for 22 d. r at any given subbasin from the mass existing in that same subbasin 6 h later. (g) Evolution of the apparent or effective dispersion coefficient calculated from pulsed tracer-release experiments. Finally, frame (h) represents the evolution of areaaveraged vorticity (dashed line) and the area-averaged absolute vorticity (solid line), estimated from the simulated velocity field for the 1999 runs calculated in the uppermost layer. 2052 Rueda and MacIntyre Vertical mixing was insufficient to entrain it into the surface layer; also, the next event was warmer, and much of the inflow water flowed in on top of it. Intrusions from both 2004-1 and 2004-2 in the inlet basin were flushed during the third event. In the main basin, the persistence of intrusions varied from 3.4 to 20.5 d. The intrusion in 1999 was dispersed after 3.4 d, in part by internal displacements, but mainly by entrainment into the mixed layer (Fig. 7b). The intrusion in 2004-1, in turn, lasted 20.5 d because the mixed layer shoaled due to warming shortly after inflow, resulting in a reduction in the potential for entrainment. Mean Kz values were O(1026) m2 s21 in the center of the intrusion and O(1025) m2 s21 near its upper boundary. The timescales for mixing across the 4-m-thick intrusion, estimated as d2/ Kz (with d 5 4 m), were on the order of one to several months. Exchanges over the 0.5-m upper boundary, though, would have occurred daily. The 2004-2 intrusion came in on top of the first and did not contribute toward its dispersion. The intrusion from the 2004-1 event was finally entrained into the surface mixed layer during the 2004-3 event, mainly due to strong cooling and wind mixing. In the west basin, the intrusions were short-lived and occurred discontinuously (e.g., Fig. 7c). They generally formed during strong NW wind forcing, which suggests that they were fed by upwelling from intrusions in the main basin. The largest exchanges between the west and the main basins occurred after the intrusions in the main basin had been entrained and were mixed horizontally. Horizontal heterogeneity—The tracer concentration fields simulated by the model were highly variable, both horizontally and vertically within events and between events (Figs. 8, 9). The fact that the model is based on physical laws, and that the simulations agree with the available field observations (see Model setup and validation in the Methods section; Fig. 5; Rueda and MacIntyre, in press), allows us to place confidence in the simulated horizontal fields. The model shows detailed patterns of river inflows at time and space intervals that cannot be duplicated by field sampling. For example, the patterns 36 h after the start of the event varied between storms (Fig. 8). During event 1999, concentrations in surface waters extended throughout much of the main basin, whereas, in the three events with lower discharge, there was a tendency for incoming water either to flow along the lake margins, where they bathed the benthic algal communities, or to form gyrelike structures. Concentrations remained heterogeneous 72 h after the start of the event. The moraine separating the main and the west basins contributed to the heterogeneity, even at depths above the moraine (Fig. 9), as did prevailing winds (Fig. 9, surface layer TM). Considerable dispersion results from frequent shifts in wind direction (2004-3). Overall, the persistence of horizontal heterogeneity depended upon the time from the onset of the discharge event until the next event with strong winds. For instance, the surface layer in the main basin became homogeneous shortly after the wind events at the end of day 201 (1999) (Fig. 10). It took several more days until tracer concentrations were uniform in the main basin and the west basin. Discussion In multibasin lakes, the introduction of negatively buoyant water during storm inflow events induces significant spatial–temporal variability in the composition of water. These changes occur in the vertical direction as a consequence of the formation of intrusions. They occur in the horizontal direction because of the presence of shallow sills that restrict the lateral spreading of intrusions and beacuse of variations in surface currents caused in part by these same features. We identified three mechanisms that induce cross-basin exchange across sills, and we showed a variety of factors that influence the depth of intrusions and their persistence. In the following, we compare the timescales and dimensionless indices which determine the importance of these different mechanisms, so that our results can be generalized to other storm events in other locations and in the context of changing weather patterns. Role of different mechanisms for interbasin exchange— The inflow rate, Q, the size of the basins, V, and the rate at which the incoming water mixes with ambient water prior to the formation of the intrusion will determine the rate at which sequential basins are filled and, hence, the rate of lateral transport by overflow. The ratio of the filling timescale (tf), as given by Eq. 4, to the length of the inflow event (t0) provides an indication of the importance of overflow as an exchange mechanism between basins during any given inflow event. If an event ceases before a basin has filled or before an intrusion has thickened enough to allow overflow (i.e., tf /t0 & 1), this mechanism will not be important. The inlet basin was, for example, characterized by ratios tf /t0 % 1, whereas the main basin was characterized by tf /t0 & 1 during the four events analyzed. Consequently, overflow was the dominant form of exchange out of the inlet basin (Fig. 7d) and into the main basin (Fig. 7e), but it did not contribute significantly to the exchange between the main and west basins (Fig. 7f). Large values of F0 (&1) and consequently large initial mixing rates (C 5 c21 &1) are also expected for large inflow rates (Q), as reported by Johnson et al. (1989), which should further decrease the ratio tf /t0 and further contribute to the importance of overflow. Internal wave–mediated exchanges have been observed previously in a number of lakes (Lawrence et al. 1997; Umlauf and Lemmin 2005; Laval et al. 2008), and the process has been called seiche pumping (Van Senden and Imboden 1989). In most of these cases, the ratio of the depth of the thermocline to the depth of the sill was ,1 (Laval et al. 2008). The internal waves drove bidirectional exchange of hypolimnetic water through sills, and the magnitude of exchange was dependent on the amplitude and duration of the internal waves (Laval et al. 2008). In contrast, in Toolik Lake, and as expected in many kettle lakes, the sills are shallow, and the intrusions tend to occur within the metalimnion. Thus, the ratio of the depth of the thermocline to that of the sill is .1, and intruded water can only be transported between basins when winds are large enough that isotherm excursions are larger than the difference in depth between the top of the intrusion (zi) Pathways of stream inflows in lakes 2053 Fig. 8. Tracer concentrations 36 h after the beginning of each event as a function of depth for the four events simulated: (a) 05:00 h on day 200 (1999); (b) 09:00 h on day 193 (2004); (c) 09:00 h on day 202 (2004); and (d) 23:00 h on day 214 (2004). The color scale is the same for all plots. and the sill (zs) and in a direction that induces upwelling perpendicular to the sill. Our calculations showed that the Lake number can be used to predict when interbasin exchange will occur by isotherm displacement. Given that the magnitude of the displacement depends on LN, the criterion for exchange is LN , zi /|zi 2 zs|. In arctic lakes, such events occur frequently due to the relatively weak stratification and frequent windy periods (MacIntyre and Melack 2009; MacIntyre et al. in press). In lakes with stronger stratification or greater sheltering from wind, this mechanism would likely occur less frequently. 2054 Rueda and MacIntyre Fig. 9. As in Fig. 8, but 36 h later (i.e., 72 h after the beginning of the event): (a) 17:00 h on day 201 (1999); (b) 21:00 h on day 194 (2004); (c) 23:00 h on day 203 (2004); and (d) 11:00 h on day 216 (2004). Entrainment occurs as wind and cooling deepen the surface mixed layer to the depth of intrusions. Entrainment distributes storm water throughout the surface mixed layer in a timescale te 5 h2/Kz, which is typically of O(1) h (Imboden and Wüest 1995; MacIntyre et al. 2002), i.e., much shorter than the wind or inflow events. Once in the surface layer, the entrained water will rapidly be dispersed to other basins on timescales that depend on horizontal dispersion rates. Our simulations suggest that entrainment probably accounts for most of the transport between large basins downstream of inlet basins. For instance, almost 65% of the water exchanged between TM and WB in 1999 occurred during a 2-d period when the intrusion in TM was entrained into the surface mixed layer. In addition, a Pathways of stream inflows in lakes Fig. 10. Evolution of the tracer concentration near the surface and at four different sites in Toolik Lake for the simulated tracer-release experiment in 1999. significant fraction of the storm river water is entrained into the surface layers before reaching the theoretical depth of neutral buoyancy. These fluxes mainly occur as water flows through the small inlet basin and through and over the sills. We estimated that almost 30% of riverborne substances entering during the initial stages of the storm in 1999 could have been entrained into the surface layers while flowing near the surface in the inlet bay and adjacent shallow regions. In consequence, small inlet basins are not mere conveyors of storm river water from the inlet to the deep intrusions in the largest basins. Instead, the smallest inlet basins behave as mixers, facilitating the incorporation of river water into the surface layers from where they can be freely transported to other subbasins over the sills. In basins with complex morphometry, this mechanism might significantly increase fluxes to the surface layer relative to lakes in which inflows occur directly into a large basin and sink quickly to their depth of neutral buoyancy. Thus, entrainment and interbasin transport are facilitated in basins with shallow shelving areas or sills that force incoming water to depths where vertical mixing is induced by surface forcing. Timescales of horizontal heterogeneity—Horizontal dispersion depends upon wind, depth of the mixed layer and, thus, vertical shear, and bottom topography and basin morphometry, which affect the magnitude of velocity gradients. Our simulations with tracer clouds gave values of Kapp of O(1021) m2 s21 during calm periods and ,1.4 m2 s21 during the strong wind events on days 199 and 202 (Fig. 7g). These values are comparable with those in Lawrence et al. (1995) for length scales , 5 3s of O(102) m and , 5 O(103) m, the latter being the horizontal scale of Toolik Lake, and they are consistent with those in Kootenay Lake (,3 km wide) (Stevens et al. 1995). Stocker and Imberger (2003) and Stevens et al. (1995) argued that 2055 the irreversible spreading of solutes and particulates in the surface layer in small- to medium-size lakes is largely the result of dispersion driven by horizontal shear. Our simulations suggest that large horizontal velocity gradients develop in response to strong and persistent wind events (see Figs. 2, 7g), and our time series figures of tracer distribution suggest that these may be accentuated by basin morphometry (Figs. 8, 9). Area-averaged vorticity, f, which indicates the magnitude of the velocity gradients, was two to three times higher during storm events and cooling periods (<2.5 3 1025 s21) than during calm periods (1 3 1025 s21). If we assume that the lateral dispersion rate is proportional to the square of the magnitude of the velocity gradients, as dimensional analysis suggests (Smagorinsky 1963; see also Fischer et al. 1979), then the rate of dispersion during strong winds should be nearly one order of magnitude larger than under calm conditions. The order-of-magnitude difference during calm and windy periods in both f and Kapp supports this hypothesis and indicates that large velocity gradients induced by strong wind events contribute substantially to rapid horizontal mixing. The timescale for a tracer to become uniformly distributed over the whole lake, th, is computed from the rate of spreading of a tracer cloud, where the initial and final variances are set to s 20 5 A0 /p and s2 5 A/p, respectively. For example, A0 can be taken as the surface area of the main basin, and A is the area of the whole lake. The time th for a cloud to grow from s 20 < 3.2 3 105 m2 to s2 < 4.7 3 105 m2 when Kapp 5 O(1) m2 s21, as on windy days and using Eq. 3, is O(1) d. During calm periods, with Kapp , O(1021) m2 s21, mixing time is O(10) d. Thus, spatial heterogeneity can rapidly be reduced when winds are moderate to high, such as during the passage of fronts, but this reduction persists with the low to moderate winds that are prevalent at other times (Fig. 3). The implications of the temporal heterogeneity depend upon the rates of the biological and chemical reactions induced when the intruding water mixes with ambient water (Knauer et al. 2000). For instance, whether benthic algae or pelagic phytoplankton preferentially acquire nutrients will depend upon the amount of time the stream inflows hug the nearshore regions (Fig. 8), the timing of nutrient loading relative to changes in the storm hydrograph, and uptake rates. Hydraulic and meteorological forcing—The ecological response of lakes as weather patterns vary under different climate regimes will depend upon the pathways of inflow events as they vary with stratification and surface forcing. All the discharge events described here were initiated during cold fronts, but other conditions varied. The depth of penetration of the incoming water was linked to the degree of cooling associated with the event, since decreased air temperatures and increased evaporation rates affected both the stream temperatures and the average temperature of the inlet basin. The discharge rate and related volume of incoming water determined the amount of the incoming water that mixed with water in the inlet basin. Given these combined factors, intrusion depths were deeper during cold fronts, which were accompanied by the largest heat losses 2056 Rueda and MacIntyre and largest incoming volumes of water. The volume of water at peak discharge contributed strongly to the rate that water moved between basins. The frequency of shifts between warm and cold air masses and the associated increases in wind speed affected the frequency of events having Lake numbers low enough to cause upwelling and overflow into adjacent basins, persistence of intrusions, and the timescale for horizontal uniformity in the surface layer. Thus, the effect of incoming waters and the associated nutrients, microbes, and phytoplankton, whether they will become well mixed on timescales of days or weeks, is highly dependent upon climate-related factors that influence the frequency and intensity of storm events (Serreze and Barrett 2008; MacIntyre et al. in press). The combined modeling and field studies presented here illustrate the space–time scales of pathways of stream inflows in small lakes with complex bathymetry. Results indicate that synoptic sampling is required even in small lakes to quantify the biological consequences of increased stream discharge. Variations in dominant patterns and in mixing rates depend on atmospheric forcing in predictable ways. With an understanding of these patterns, we are poised to predict the implications for benthic and pelagic communities as climate changes. Of major importance, modeling allows confirmation and extension of the intuition gained from field studies. The synoptic view allows questions to be raised and strategies to be developed to more fully address the implications of physical processes on ecosystem function. Acknowledgments We thank James King, Neil Bettez, Chris Wallace, Chris Crockett, Mary Anne Evans, and Jim Laundre for help with field measurements. We thank Brice Loose, Chad Helmle, Lorenz Moosmann, and Chris Wallace for assistance with processing and analysis of physical data. The Arctic Long-term Ecological Research (LTER) provided meteorological and stream discharge data. We particularly thank George Kling for his assistance with this data. Logistic support was provided by the University of Alaska Toolik Lake Field Station. Financial support was provided by National Science Foundation (NSF) Division of Environmental Biology grants DEB-0508570, DEB-0423385, and DEB-9810222, and Office of Polar Programs (OPP) grants OPP9911278 to the Arctic LTER, and DEB-9726932, -0108572, -0640953, OCE-9906924, and ARC-0714085 to Sally MacIntyre. References AHLFELD, D., A. JOAQUIN, J. TOBIASON, AND D. MAS. 2003. Case study: Impact of reservoir stratification on interflow travel time. J. Hydraul. Eng. 129: 966–975. AKIYAMA, J., AND H. G. STEPHAN. 1987. Onset of underflow in slightly diverging channels. J. Hydraul. Eng. 113: 825–843. BARBIERO, R. P., W. F. JAMES, AND W. BARKO. 1999. The effects of disturbance events on phytoplankton community structure in a small temperate reservoir. Freshwater Biol. 42: 503– 512. BOEGMAN, L., G. N. IVEY, AND J. IMBERGER. 2005. The energetics of large-scale internal wave degeneration in lakes. J. Fluid Mech. 531: 159–180. BOURNET, P. E., D. DARTUS, B. TASSIN, AND B. VINCON-LEITE. 1999. Numerical investigation of plunging density current. J. Hydraul. Eng. 125: 584–594. BRITTER, R. E., AND P. F. LINDEN. 1980. The motion of the front of a gravity current travelling down an incline. J. Fluid Mech. 99: 531–543. CHUNG, S., AND R. GU. 1998. Two-dimensional simulations of contaminant currents in stratified reservoir. J. Hydraulic Eng. 7: 704–711. DALLIMORE, C. J., J. I MBERGER , AND T. ISHIKAWA . 2001. Entrainment and turbulence in a saline underflow in Lake Ogawara. J. Hydraul. Eng. 127: 937–948. DURRAN, D. R. 1999. Numerical methods for wave equations in geophysical fluid dynamics. Springer. ELBER, F., AND F. SCHANZ. 1990. The influence of a flood event on phytoplankton succession. Aquat. Sci. 52: 330–344. ELLISON, T. H., AND J. S. TURNER. 1959. Turbulent entrainment in stratified flows. J. Fluid Mech. 6: 423–448. FERNANDEZ, R., AND J. IMBERGER. 2008. Time-varying underflow into a continuous stratification with bottom slope. J. Hydraulic Eng. 134: 1191–1198. FINGER, D., M. SCHMID, AND A. WÜEST. 2006. Effects of upstream hydropower operation on riverine particle transport and turbidity in downstream lakes. Water Resour. Res. 42: W08429. doi: 10.1029/2005WR004751. FISCHER, H. B., E. J. LIST, R. C. KOH, J. IMBERGER, AND N. H. BROOKS. 1979. Mixing in inland and coastal waters. Academic Press. ———, AND R. D. SMITH. 1983. Observations of transport to surface waters from plunging inflow to Lake Mead. Limnol. Oceanogr. 28: 258–272. HALLWORTH, M. A., E. H. HERBERT, J. C. PHILLIPS, AND R. S. SPARKS. 1996. Entrainment into two-dimensional and axisymmetric turbulent gravity currents. J. Fluid Mech. 308: 289–311. HAUENSTEIN, W., AND T. H. DRACOS. 1984. Investigation of plunging density currents generated by inflows in lakes. J. Hydraul. Res. 22: 157–179. HEBBERT, B., J. IMBERGER, I. LOH, AND J. PATTERSON. 1979. Collie River underflow into the Wellington reservoir. J. Hydraul. Div. ASCE 105: 533–545. HODGES, B. R., J. IMBERGER, A. SAGGIO, AND K. B. WINTERS. 2000. Modeling basin-scale motions in a stratified lake. Limnol. Oceanogr. 45: 1603–1620. HORN, D. A., J. IMBERGER, AND G. N. IVEY. 2001. The degeneration of large-scale interfacial gravity waves in lakes. J. Fluid Mech. 434: 181–207. IMBERGER, J., AND J. C. PATTERSON. 1990. Physical limnology. Adv. Appl. Mech. 27: 303–475. IMBODEN, D. M., AND A. WÜEST. 1995. Mixing mechanism in lakes, p. 83–138. In A. Lerman, D. Imboden, and J. Gat [eds.], Physics and chemistry of lakes. Springer-Verlag. INAMDAR, S. P., N. O’LEARY, M. J. MITCHELL, AND J. T. RILEY. 2006. The impact of storm events on solute exports from a glaciated forested watershed in western New York, USA. Hydrol. Process. 20: 3423–3439. JAIN, S. C. 1981. Plunging phenomena in reservoirs, p. 1249–1257. In H. G. Stephan [ed.], Proceedings of the Symposium on Surface Water Impoundments. ASCE. JIN, K.-R., J. H. HAMRICK, AND T. TISDALE. 2000. Application of three-dimensional hydrodynamic model for Lake Okeechobee. J. Hydraulic Eng. 126: 758–771. JOHNSON, T. R., C. R. ELLIS, AND H. G. STEPHAN. 1989. Negatively buoyant flow in a diverging channel. IV: Entrainment and dilution. J. Hydraulic Eng. 115: 437–456. ———, G. J. FARRELL, C. R. ELLIS, AND H. G. STEPHAN. 1987. Negatively buoyant flow in a diverging channel. I: Flow regimes. J. Hydraulic Eng. 113: 717–730. Pathways of stream inflows in lakes KANTHA, L. H., AND C. A. CLAYSON. 1994. An improved mixed layer model for geophysical applications. J. Geophys. Res. 99: 25235–25266. KASSEM, A., J. IMRAN, AND J. A. KHAN. 2003. Three-dimensional modeling of negatively buoyant flow in diverging channels. J. Hydraulic Eng. 129: 936–947. KNAUER, K., H. M. NEPF, AND H. F. HEMOND. 2000. The production of chemical heterogeneity in Upper Mystic Lake. Limnol. Oceanogr. 45: 1647–1654. KRIET, K., B. J. PETERSON, AND T. L. CORLISS. 1992. Water and sediment export of the upper Kuparuk River drainage of the North Slope of Alaska. Hydrobiologia 240: 71–81. LAVAL, B. E., J. MORRISON, D. J. POTTS, E. C. CARMACK, S. VAGLE, C. JAMES, F. A. MCLAUGHLIN, AND M. FOREMAN. 2008. Winddriven summertime upwelling in a fjord-type lake and its impact on downstream river conditions: Quesnel Lake and River, British Columbia, Canada. J. Great Lakes Res. 34: 189–203. LAWRENCE, G. A., J. M. BURKE, T. P. MURPHY, AND E. E. PREPAS. 1997. Exchange of water and oxygen between the two basins of Amisk Lake. Can. J. Fish. Aquat. Sci. 54: 2121–2132. ———, N. YONEMITSU, AND J. R. ELLIS. 1995. Natural dispersion in a small lake. Limnol. Oceanogr. 40: 1519–1526. MACINTYRE, S., K. M. FLYNN, R. JELLISON, AND J. R. ROMERO. 1999. Boundary mixing and nutrient flux in Mono Lake, CA. Limnol. Oceanogr. 44: 512–529. ———, J. P. FRAM, P. J. KUSHNER, N. D. BETTEZ, W. J. O’BRIEN, J. E. HOBBIE, AND G. W. KLING. In press. Climate-related variations in mixing dynamics in an Alaskan arctic lake. Limnol. Oceanogr. ———, AND J. M. MELACK. 2009. Mixing dynamics in lakes across latitudes, p. 603–612. In G. Likens [ed.], Encyclopedia of inland waters. Springer. ———, J. R. ROMERO, AND G. W. KLING. 2002. Spatial–temporal variability in surface layer deepening and lateral advection in an embayment of Lake Victoria, East Africa. Limnol. Oceanogr. 47: 656–671. ———, J. O. SICKMAN, S. A. GOLDTHWAIT, AND G. W. KLING. 2006. Physical pathways of nutrient supply in a small, ultraoligotrophic arctic lake during summer stratification. Limnol. Oceanogr. 51: 1107–1124. MCNAMARA, J. P., D. L. KANE, AND L. D. HINZMAN. 1997. Hydrograph separations in an Arctic watershed using mixing model and graphical techniques. Water Resour. Res. 33: 1707–1719. MILLER, M. C., P. SPATT, P. WESTLAKE, D. YEAKEL, AND G. R. HATER. 1986. Primary production and its control in Toolik Lake, Alaska. Arch. Hydrobiol. Suppl. 74: 97–131. MONISMITH, S. G. 1987. Modal response of reservoirs to wind stress. J. Hydraulic Eng. 113: 1290–1306. 2057 O’BRIEN, W. J., AND oTHERS. 1997. The limnology of Toolik Lake, p. 61–106. In A. M. Milner and M. W. Oswood [eds.], Freshwaters of Alaska. Ecological studies. Vol. 119. Springer-Verlag. PEETERS, F., A. WÜEST, G. PIEPKE, AND D. M. IMBODEN. 1996. Horizontal mixing in lakes. J. Geophys. Res. Oceans 101: 18361–18375. ROBARTS, R. D. 1987. Effect of rain storms on heterotrophic bacterial activity in a hypertrophic African lake. Hydrobiologia 148: 281–286. RUEDA, F. J., AND E. A. COWEN. 2005. The residence time of a freshwater embayment connected to a large lake. Limnol. Oceanogr. 50: 1638–1653. ———, AND S. MACINTYRE. In press. Modelling the fate and transport of negatively buoyant storm-river water in small multi-basin lakes. Environ. Model. Software. SAVAGE, S. B., AND J. BRIMBERG. 1975. Analysis of plunging phenomena in water reservoirs. J. Hydraul. Res. 13: 187–205. SERREZE, M. C., AND A. P. BARRETT. 2008. The summer cyclone maximum over the central Arctic Ocean. J. Clim. 21: 1048–1065. SMAGORINSKY, J. 1963. General circulation experiments with the primitive equations: I. The basic experiment. Mon. Weather Rev. 91: 99–164. SMITH, P. E. 2006. A semi-implicit, three-dimensional model of estuarine circulation. Open-File Report 2006-1004. USGS., Sacramento, California 2006. Available from http://pubs.usgs. gov/of/2006/1004/pdf/ofr2006-1004.pdf. STEVENS, C., AND J. IMBERGER. 1996. The initial response of a stratified lake to a surface shear stress. J. Fluid Mech. 312: 39–66. ———, P. F. HAMBLIN, G. A. LAWRENCE, AND F. M. BOYCE. 1995. River-induced transport in Kootenay Lake. J. Environ. Eng. 121: 830–837. ———, AND G. A. LAWRENCE. 1997. Estimation of wind-forced internal seiche amplitudes in lakes and reservoirs, with data from British Columbia, Canada. Aquat. Sci. 59: 115–134. STOCKER, R., AND J. IMBERGER. 2003. Horizontal transport and dispersion in the surface layer of a medium-sized lake. Limnol. Oceanogr. 48: 971–982. UMLAUF, L., AND U. LEMMIN. 2005. Interbasin exchange and mixing in the hypolimnion of a large lake: The role of internal waves. Limnol. Oceanogr. 50: 1601–1611. VAN SENDEN, D. C., AND D. M. IMBODEN. 1989. Internal seiche pumping between sills-separated basins. Geophys. Astrophys. Fluid Dynam. 48: 135–150. Associate editor: Chris Rehmann Received: 28 August 2008 Accepted: 11 May 2009 Amended: 24 June 2009