Model Results


During model calibration, a subset of model parameters was adjusted within reasonable bounds to optimize a comparison of measured and modeled results for this river reach. Manual model calibration consisted of iterative model runs where the values of key model parameters were adjusted to better fit the results during unique time periods where one or more processes could be somewhat isolated from the many processes affecting water quality throughout the year. The magnitude and seasonal and spatial patterns between measured data and model output should correspond; however, an exact match is not expected because measurements are made at a point and model output represents conditions in model cells that encompass about 1,000 ft of river length. Concentration values for suspended particulate material, including algae, particulate carbon and nitrogen, and total nitrogen and phosphorus, had more spatial variability than values for dissolved constituents (Sullivan and others, 2008, 2009); the model is laterally averaged and does not simulate any bank-to-bank differences. Models are simplifications of nature, and not every process is simulated, but a good fit between the spatial and temporal patterns in the modeled and measured data provides confidence that major processes are included and that the model can be used to provide insight into system behavior under various management options.


Model calibration began with balancing the water budget followed by assessing water velocity, ice cover, and water temperature. Finally, the water-quality aspects of the model were calibrated, beginning with some of the constituents that exhibited conservative (nonreactive) properties, such as TDS and ISS, then moving on to nutrients (nitrate, ammonia, dissolved phosphorus) and organic matter, and finally addressing the more challenging constituents such as algae and dissolved oxygen. pH was initially included in the model, but is not considered to be calibrated. This system has high concentrations of dissolved organic matter which can affect pH and buffering capacity, as indicated theoretically and also from the results of alkalinity titrations in this reach (U.S. Geological Survey, Miller Island, unpub. data, 2007). The CE-QUAL-W2 model does not yet include algorithms that include the acid–base properties of organic matter in the simulation of pH.


The parameter optimization software PEST (Doherty, 2010) was linked to CE-QUAL-W2 during the model calibration process and thus was used to refine the model calibration and to examine parameter sensitivity and parameter correlation. Much knowledge was obtained about the nature and characteristics of water-quality processes acting in the Klamath River during the model calibration process. 


Water Velocity


Simulation of water velocity allows the calculation of travel time through the Link–Keno reach. Slower velocities allow more time for settling, decay, and nutrient transformation to occur within the reach; faster velocities move more material downstream. Generally, the variation in velocity from bank to bank was low and less than the vertical variability in velocity for most of the Lake Ewauna to Keno Dam reach according to the 2007 cross-sectional measurements. Flow direction usually was aligned with the upstream–downstream axis of the channel. The lack of substantial lateral variability and the alignment of velocities with the longitudinal axis of the river confirmed that the two-dimensional model CE-QUAL-W2 was a suitable choice for the Link–Keno reach. Although average measured cross-sectional velocity was not directly comparable to simulated velocity, because model velocity is a representation of average conditions that occur over about 1,000 ft of river length, measured and modeled velocities were in the same range (fig. 7). Modeled and cross-sectional average measured velocities ranged from near 0 to 0.5 ft/s on the measurement dates in 2007. September velocities were lower than May velocities, which was expected because streamflow in this reach tends to be lowest in August and September. Maximum modeled velocity (volume averaged over segment depth) in 2006–09 for the whole modeled Link–Keno reach was 3.0 ft/s near the Keno sampling site in mid-April 2006 during high flow; minimum modeled velocity was near 0 ft/s at the upper end of Lake Ewauna at several times over the modeled period. Generally, modeled velocities were highest near the Lost River Diversion Channel inflow and near the Keno sampling site, and lowest in Lake Ewauna and midreach near the Miller Island and KRS12a sites. Simulated travel times from model tracer tests for the Link River to Keno Dam reach ranged from about 4 days at 2,000 ft³/s flow to 12 days at 700 ft³/s flow.


The 30-minute ADCP measurements at the Keno site also showed that velocity direction was aligned with the longitudinal axis of the channel (fig. 8A). Although measured and modeled velocity are not directly comparable, results in figure 8B show that the two are similar in magnitude and that the temporal patterns in the measured data are well represented by the model. Daily average modeled and measured velocities from June to early December 2008 ranged from near 0 to almost 0.6 ft/s. Velocity variations at the Keno site over periods of hours to days were the result of changes in operations at Keno Dam; abrupt changes in flow at the dam led to abrupt changes in upstream water velocity.


The hydrodynamics of the Lake Ewauna reach were complex. Flow often was aligned with the upstream–downstream axis of the river channel, but the direction of flow at EWA1 and EWA2 periodically would reverse (fig. 9A). This flow pattern was affected by both the inflow from Link River, just upstream at the northwest corner of figure 9A, and by the wind direction. For June and July 2008, with relatively higher flows from Link River, when the wind direction was towards the southeast, in the same direction as the flow from Link River, flow at both EWA1 and EWA2 usually was in the downstream direction (fig. 9B). If the wind was towards the northwest for the same period, flow at the deeper western side (EWA1) was still towards the southeast, but flow on the shallow eastern side (EWA2) was reversed towards the northwest, thus suggesting a counterclockwise circulation pattern. During mid-July through October, with lower flow from Link River, periods still occurred when flow at both EWA1 and EWA2 were in the downstream direction, but wind had a greater influence on circulation. For example, sometimes when wind direction was towards the northwest, a strong counterclockwise circulation was suggested, with flow at EWA1 to the southeast and flow at EWA2 to the northwest. At other times, when wind direction was towards the southeast, flow at EWA1 was to the northwest and flow at EWA2 to the southeast, thus suggesting clockwise circulation. Flow at EWA2 tended to be in the same direction as the wind, whereas flow at EWA1 was downstream during high flow, but could be reversed at low flow when winds were toward the southeast.


For a shallow lake with a trench to the west, clockwise circulation with wind towards the southeast, and counterclockwise circulation with wind towards the northwest is expected theoretically (Ji and Jin, 2006). When most wind is toward the southeast, a similar clockwise pattern dominates the circulation in Upper Klamath Lake (Wood and others, 2008), a shallow lake that also has a deeper trench on its western edge. The circulation pattern in Lake Ewauna, like that in Upper Klamath Lake, was strongly affected by the wind, but was complicated by the strong advective flow from Link River, especially in the spring. The two ADCPs have provided valuable insight into the complex circulation in Lake Ewauna. Additional ADCP measurements could be valuable to gain a more detailed understanding of these circulation patterns.


The two-dimensional CE-QUAL-W2 model cannot capture the details of these flow patterns in Lake Ewauna because lateral variations are not simulated. Instead of simulating higher flows varying between upstream and downstream direction, the model simulates an average flow from bank-to-bank, most often in the downstream direction. The complex flow patterns in Lake Ewauna result in a wide distribution of residence times there, with some water parcels that traverse quickly downstream and bypass much of the large volume of the reach and other water that becomes entrained in a recirculating flow and spends a longer time there. An increased or decreased travel time has an important effect on water quality because more or less time is available for settling or decay or other processes. The model, in contrast, will simulate a narrower distribution of travel times and, therefore, a narrower distribution of water-quality effects, although the average effect on water quality may not be biased. Although the circulation pattern in Lake Ewauna undoubtedly has some interesting water-quality effects, it was decided that because the reach is relatively short, the increased effort and complexity associated with using a different type of model for the Lake Ewauna reach was not appropriate for this study. A more detailed examination of water-quality processes in Lake Ewauna in the future could benefit from the development of a three-dimensional model. 


Ice Cover


Ice forms at times on the Link–Keno reach of the Klamath River, because winter air temperatures are commonly below freezing. The CE-QUAL-W2 model can simulate the onset, buildup, and break up of ice. These processes in the model are affected by the locations and temperatures of inflows and outflows, the ice-to-water heat-exchange coefficient, and sublimation effects of wind over the ice surface. Proper simulation of ice cover is important to the accurate simulation of water temperature and winter reaeration.


The model simulated ice to have occurred on some segments and dates in December 2006, January–February and December 2007, January–February and December 2008, and January and December 2009. This simulation matches well with field observations of the presence of ice by the field crews. At times, the crew noted partial ice cover along the banks. Because CE-QUAL-W2 is a laterally averaged model, simulation of lateral differences from bank-to-bank were not represented. The model would instead simulate partial lateral ice cover as a thin layer of continuous ice from bank to bank. The model does simulate differences in ice-cover from upstream to downstream and those spatial and temporal patterns of ice cover simulated by the model match well with the available field observations.


Water Temperature


Water temperature in rivers is affected by the temperature and flow of inflows and outflows, heat exchange at the air–water interface, light extinction, and mixing by wind. Water temperature is usually one of the first constituents to be calibrated in CE-QUAL-W2 model applications because many other simulated water-quality processes, such as chemical reactions, algal growth, and SOD, are dependent on temperature. The physics of heat-exchange processes for waterbodies is well known and has been accurately translated into mathematical formulas and algorithms that are readily coded into numerical models. The CE-QUAL-W2 model includes the most important algorithms describing heat-transport and exchange processes.


The seasonal and daily temperature cycles in the Link–Keno reach were consistent in the 4 years simulated for this study (figs. 10and 11), with cold winter water temperatures near freezing and warm summer water temperatures near 30 °C at times. Seasonal water temperature patterns were largely a result of the inflow from Upper Klamath Lake, the large shallow lake just upstream; seasonal temperature patterns were similar at the upstream and downstream ends of the reach.


Daily temperature cycles were present in summer, and heating of the river surface is enhanced in the Klamath River because of high light extinction coefficients that cause most of the short-wave solar radiation to be scattered, absorbed, and converted to heat near the top of the water column. For most of the year and at most locations in the Link–Keno reach, temperature differences between the river top and bottom were present only during daytime; at night, cooling of the river surface often decreased vertical temperature differences sufficiently to allow vertical mixing. At some locations, especially in the downstream reaches, temperature differences of up to several degrees Celsius between the river surface and bottom persisted, up to about 1–2 weeks, thus inhibiting vertical mixing. Intermittent thermal stratification downstream of the Klamath Straits Drain inflow may be enhanced by the sinking of that inflow due to its higher TDS concentration (discussed in the next section, “Total Dissolved Solids and Specific Conductance”).


Goodness-of-fit statistics were calculated to compare hourly measured temperatures and model output at the same location and time (table 6). The mean error (ME) for 2006–09 averaged over all sites for a year was between -0.17 and 0.17°C depending on the year. The ME is an indication of the overall bias in model predictions. An ME <0.2 °C indicates that overall bias of the model is on the same order of magnitude as measurement error, which typically is ±0.2 °C. The computed mean absolute error (MAE) for model temperature predictions was between 0.54 and 0.64 °C. The MAE is an indication of the typical error associated with any one model-data comparison. Sources of error in modeling water temperature include the estimation of Lost River Diversion Channel temperatures in 2006, estimation of distributed tributary temperatures, and any errors in the model representation of channel width or calculation of extinction coefficients. The model consistently produces a slight underprediction of water temperature in the fall, which may be due to the lack of an algorithm in CE-QUAL-W2 to transfer stored summer heat from river sediments to the water column. Overall, the model accurately simulated water temperature in all 4 years with an MAE of 0.64 °C or less; an MAE of less than 1.0 °C is the usual benchmark for a well-calibrated CE-QUAL-W2 model.


Total Dissolved Solids and Specific Conductance


Total dissolved solids (TDS) is the sum of all dissolved substances in water, such as dissolved nutrients, dissolved organic matter, and all dissolved ions. Total dissolved solids contribute to density gradients in the model, which can affect how constituents are mixed through the water column. The CE-QUAL-W2 model treats TDS as conservative (unreactive), in that its concentration is only affected by inflows, outflows and hydrodynamic mixing. This is a good approximation, but it is important to realize that TDS in rivers also is affected by a number of chemical and biological processes, and those processes do not affect TDS in the model. Total dissolved solids, as discussed in section, “Methods,” are related to specific conductance; specific conductance was converted to TDS for use in the model and model output was converted back to specific conductance to compare with field calibration data.


The inflow specific conductance at Link River generally was lower than specific conductance from point sources and tributaries. Model options were set so that tributary inflows were placed in the river at a depth based on the density of the inflow and that of the river, where the densities were determined from TDS and water temperature. The effect of this density placement of inflows was most obvious at the KRS12a site downstream of the inflow of Klamath Straits Drain. The higher density Klamath Straits Drain water plunges to the bottom of the Klamath River at certain times of year; the Klamath Straits Drain has especially high specific conductance in spring. That density stratification was especially evident in the hourly data and specific conductance profiles for early 2007–09 (figs. 12 and 13). The density stratification decreases by the time the water reaches the Keno site, indicating that additional vertical mixing has occurred. Specific conductance from the NPDES point sources was as high as or higher than that from Klamath Straits Drain, but the point source flows were low enough that the density effect was not evident in the profiles at Railroad Bridge. The higher specific conductance from the NPDES point sources, however, does affect the depth at which those flows entered the river.


The model does well at simulating specific conductance at most sites (figs. 12 and 13), which indicates that inflows, transport, and mixing were the major controlling processes. Some of the small-scale daily variation in specific conductance that occurs at the Railroad Bridge site, especially Railroad Bridge-bottom, is not simulated by the model. This can indicate that the processes that produce those variations originate from a process not represented in the model.


Inorganic Suspended Sediment


Inorganic suspended sediment (ISS), such as suspended clay or silt, is a natural component of lakes and rivers, though excessive concentrations can impair certain uses of a river. For example, high ISS loads can reduce the efficiency of water treatment plants, and, over time, high concentrations of sediment can settle and impair fish habitat or reduce available water storage in a reservoir. High concentrations of ISS also affect light penetration, which affects the vertical distribution of heat and the depths at which photosynthesis can occur. Some of the tributaries in the Link–Keno reach, such as Klamath Straits Drain, had relatively high ISS concentrations over periods of several days to weeks in winter. These high concentrations usually were associated with storms and could come from stormwater picking up sediment as it runs over soil or increased flow and turbulence causing resuspension of bed sediment. In-river modeled ISS concentrations downstream of Klamath Straits Drain were elevated (above 30 mg/L) for periods in late February to early April in 2006, 2007, 2008, and 2009. ISS also was elevated for most of the entire Link–Keno reach in early December 2007 and late March to early April 2008. Summer concentrations of suspended sediment typically were low in this reach at about 3–5 mg/L.


Phytoplankton


The presence, magnitude, and status of algal blooms are influenced by available light and nutrients, water temperature, travel time (flow), transport from upstream reaches, settling, mortality, and zooplankton grazing and other factors such as viruses. Algae can be beneficial because they are an important part of the food chain, and algal photosynthesis produces dissolved oxygen in the water column. Conversely, algal respiration and decay consumes dissolved oxygen. Some species of algae (for example, Anabaena, Microcystis) can produce toxins that may be dangerous to aquatic life or humans.


By simulating the algal community in the Link–Keno reach with three distinct types of algae (blue greens, diatoms, and other), the model was able to reproduce most of the major spatial and temporal trends in algae observed in 2006–09 (fig. 14). Diatoms typically bloomed in the spring, and large blooms of blue-green algae entered the reach from Upper Klamath Lake through Link River in summer (note that the maximum scale on the blue-green algae graphs in figure 14 is about 20 times higher than that for diatoms or other algae). The model simulates the characteristic decrease in blue-green algae concentration that occurs with increasing distance downstream of Link River (Sullivan and others 2008, 2009). Populations of diatoms and other algae do not decrease in the downstream direction, and at certain times, increase in the downstream direction. It is unknown why the blue-green algae populations from Upper Klamath Lake are not able to sustain themselves to the same levels in this reach. Several explanations have been proposed, including physical cell damage as a result of transport past Link Dam, differences in the characteristics and vertical thermal structure of the hydrodynamic system of the lake versus the river, or algal mortality due to low dissolved-oxygen concentrations. The latter mechanism was hypothesized and invoked by a previous model of this reach; that model separated algae into a “healthy” or, upon exposure to low dissolved oxygen, an “unhealthy” algae group. The two groups were assigned different growth, respiration, excretion, and mortality rates (Tetra Tech, 2009; Rounds and Sullivan, 2009, 2010). However, at present, insufficient evidence is available to support a specific mechanism for the decline in algal populations through the Link–Keno reach. The model described herein does not simulate the details of this unknown process, but it does simulate the end result, which is an increase in settling and mortality losses for blue-green algae in this reach. If future research reveals the causal mechanism for blue-green algae losses in this reach, that process could be encoded into the model.


The CE-QUAL-W2 model keeps track of whether light, nitrogen, or phosphorus is limiting algal growth in each model cell for each time step. That information showed that light was the major limiting factor to algal growth, due to large light extinction coefficients that result from the dark color and prevalence of dissolved organic matter and suspended matter. The model did indicate that nutrients were the major limiting factor for algal growth for some periods, locations, and depths; this included occasional phosphorus limitation for the blue-green algae group and nitrogen and phosphorus limitations for the other two algal groups when sufficient light was available.


Models used to simulate algal communities, even when multiple algal groups are separated and simulated separately as in this study, can only begin to capture the complexity of those communities and their interactions and responses to light, flow, vertical mixing, nutrients, and dissolved oxygen. Unlike heat-exchange processes, the model algorithms used to represent algal communities are gross simplifications of actual processes, and large uncertainties in the predictions are inherent. Sources of error for algae in this model include uncertainties and variations in almost all of the growth rates, reaction rates, settling rates, and stoichiometric factors. In addition, errors are inherent to the estimation of Link River algal inflow concentrations in 2006 and 2009 as well as short-term variation in the Link River algal concentration that was not captured by weekly grab sampling in 2007 and 2008. Fewer algal data existed for tributaries such as the Lost River Diversion Channel and Klamath Straits Drain. Furthermore, CE-QUAL-W2 does not include an algorithm to account for blue-green algae changing their buoyancy as a function of their physiological state. Despite these uncertainties and limitations, the major spatial and temporal patterns for algae were well-simulated by the model, which suggests that most of the processes affecting algae were simulated with sufficient accuracy to be predictive and useful.


Organic Matter


Organic matter is of paramount importance to the water quality of the Link–Keno reach, and its cycling is closely tied to the concentrations and dynamics of regulated constituents such as nutrients and dissolved oxygen. Particulate organic matter is composed of a diverse collection of material such as dead algae, zooplankton, leaf litter, bacteria, and other organic materials in various stages of decay. Dissolved organic material is made up of organic molecules of varied origin and composition, typically with complex chemical structures that defy exact characterization.


The composition, and therefore the decay rates, of organic matter varies widely. Quickly decaying (labile) organic matter can result in the consumption of large amounts of oxygen over short time periods on the order of hours to days, with a concomitant release of dissolved nutrients. In contrast, slowly decaying (refractory) organic matter causes less oxygen to be consumed over short periods, but can continue to cause oxygen loss and release of nutrients over longer periods (and over longer downstream distances) when more labile materials might have been exhausted.


Long-term (30-day) BOD tests of water in the Link–Keno reach have shown the presence of both labile and refractory forms of organic matter; dead algal material was especially labile, and dissolved organic matter was particularly refractory (Sullivan and others, 2010). These results for the Link–Keno reach are similar to observations reported for other systems in the scientific literature. For example, in another eutrophic system, algal material was determined to be the most labile substrate and dissolved organic matter the most refractory; treated wastewater treatment plant effluent and pulp and paper waste were of intermediate lability in that system (Hendrickson and others, 2007).


Although many data were available to characterize the lability of organic matter in this reach in summer, less is known about the character of organic matter in winter. Degradation of organic matter is fastest soon after death of the source material (Canuel and Martens, 1996), so winter organic matter is likely made up of older, more recalcitrant organic matter. Organic matter was assumed to be less labile in winter than in summer when abundant algae were present. Model sensitivity testing confirmed that to correctly simulate winter dissolved-oxygen concentrations, organic matter must be more refractory and less labile in winter than in summer.


Organic matter can decay either in the water column or on the river bottom. The amount and decay characteristics (labile or refractory) of organic matter that falls to the river bottom is tracked by the model (fig. 14, plots of “Bottom sediment”) and is used to produce an appropriate and seasonally responsive level of SOD. The organic material in the first-order sediments compartment in the models built up and was mostly consumed within a year. This representation has implications for management of water quality in this reach, and suggests that if the materials that most contribute to first-order SOD (algae and LPOM, especially from Upper Klamath Lake) are removed from the system or decreased in load, oxygen levels could increase significantly.


The model simulated the seasonal and spatial patterns of particulate carbon and dissolved organic carbon well, as evidenced by comparisons to weekly grab samples (fig. 14, table 6). Measured particulate carbon includes algae and POM, as does model output for particulate carbon. Given the typical concentrations of POC and DOC (0.5–18 mg/L and 5–14 mg/L, respectively) in the Link–Keno reach, these error statistics (MAE of about 1.4 and 0.7 mg/L for POC and DOC, respectively) indicate that model errors typically were small relative to the amount of dissolved and particulate carbon moving through the system.


Nitrogen and Phosphorus


Nitrogen and phosphorus are essential nutrients contributing to river primary productivity. Dissolved nitrate and ammonia are the most commonly measured nitrogen species, along with total nitrogen, which includes nitrogen in organic matter, both dissolved and particulate. Nitrite can occur in natural waters, but measurements showed its concentration to be low in this reach (Sullivan and others, 2008, 2009); therefore, nitrite results are combined with nitrate results in this discussion. Sources of nitrate in the model include ammonia nitrification, and sinks include denitrification and algal uptake. Model sources of ammonia include sediment release, algal respiration, and organic matter decay; sinks include nitrification and algal uptake. Tributary inflows and outflows also affect concentrations. 


Nitrate concentrations were highest in winter and lowest in summer; this temporal pattern was common and was observed at the Link River inflow and elsewhere in the Link–Keno reach at most sites and years (fig. 14). The model simulated the temporal and spatial patterns of nitrate well. Like nitrate, ammonia concentrations were relatively high in winter. Unlike nitrate, ammonia concentrations increased in midsummer from relatively low concentrations in early summer. Ammonia concentrations showed a distinct spatial pattern in summer. Summer concentrations were less than 0.1 mg/L in the Link River inflow, but usually increased to greater than 1 mg/L downstream at Keno. Model fluxes showed that the source of most ammonia was from the decay of organic matter in the sediments and water column with algal processes as an additional source. Concentrations of un-ionized ammonia (NH₃ as opposed to the ionized form NH₄⁺), which has high toxicity to aquatic life, also increased through the reach in summer (fig. 15).


Phosphorus exists in the dissolved phase as orthophosphorus, and also as a part of dissolved organic matter. Measurements of total phosphorus include orthophosphorus adsorbed to inorganic particles (not simulated in this model) and amounts contained in particulate organic matter and algae. Sources of orthophosphorus in CE-QUAL-W2 include various inflows, algal respiration, anoxic sediment release, and organic matter decay; sinks of orthophosphorus include algal uptake. Spatial and temporal patterns of total phosphorus and orthophosphorus were simulated well by the model (fig. 14). Orthophosphorus patterns were matched well by the model in spring, but the model sometimes overpredicted the concentration slightly in the fall. This could indicate that there may be some seasonal differences in the stoichiometric ratios for phosphorus, whereas the model allows only one ratio to be set for the whole year. Given the typical concentrations of these constituents in the river, the overall nitrogen and phosphorus error statistics (table 6) indicate that model errors were relatively small compared to the amount moving through the system.


Modeled loads of total nitrogen and total phosphorus at the reach inflow at Link River and at the reach outflow at Keno Dam (fig. 16) showed differences among years, with the highest loads in the high flow year 2006. This difference is reasonable, since loads are calculated by multiplying flow by concentration. Some consistent seasonal patterns were present. Higher nutrient loads were exported downstream compared to incoming loads from Upper Klamath Lake for the period February through April, a result that likely is due to loads entering from tributaries. However, nitrogen and phosphorus loads exported downstream were lower than the respective loads imported from Upper Klamath Lake for the summer period of July and August. These lower loads are due, in part, to less flow entering the river in summer and greater withdrawals, compared to winter in addition to particulate-associated settling losses in summer.


Dissolved Oxygen


Fish and most other aquatic organisms need suitable levels of dissolved oxygen to survive and function. In the model, sources of dissolved oxygen included the inflows, exchange with the atmosphere across the river surface, and phytoplankton photosynthesis. Sinks of dissolved oxygen included atmospheric exchange (when concentrations were supersaturated), algal respiration, ammonia and nitrite oxidation, and organic matter decay both in the water column and sediments. The dissolved-oxygen budget is complex and dynamic, and can be particularly difficult for a model to accurately simulate when algal communities and oxygen demands are large. 


A plot of all 4 years of measured dissolved-oxygen data against water temperature data from the Miller Island (top) site reveals a repeating annual cycle that reflects the effects of temperature (oxygen solubility), photosynthesis, and large oxygen demands becoming predominant at different times of the year (fig. 17). Figure 17 includes contour lines of constant percent oxygen saturation (as a function of temperature), with those lines indicating saturation and supersaturation (solid contour lines) or subsaturation (dashed contour lines). Data collected during different months are plotted in different colors to illustrate the annual cycle. In January and February, dissolved-oxygen concentrations are near saturation with only minor daily variations. In March through June, the system is warming and algae are actively producing oxygen through photosynthesis because supersaturated conditions in excess of 150 percent occur. In July or August, the algal community cannot produce enough oxygen to counteract the oxygen demands from respiration, BOD, and SOD, and the oxygen levels decrease to values near zero at times. For the rest of the summer and into early fall, photosynthesis continues to decrease and oxygen demands remain high; oxygen levels slowly recover toward saturation as the labile oxygen demands are exhausted and the water cools in late fall and into winter.


The model was able to simulate most of these patterns of dissolved oxygen in the river, both near the surface and at depth (fig. 18). A notable difference was that the measured data showed large daily variations in near-surface dissolved-oxygen concentrations along most of the river, but the model simulated large daily variations only in the upstream river reaches, not at the KRS12a or Keno near-surface sites. The model responds to field observations and data indicating that algal biovolume usually decreased longitudinally from Link River to Keno. It is not known with certainty why the near-surface measured data continued to show large daily oxygen cycles, even in the presence of fewer algae downstream. Observational evidence indicated that rooted aquatic plants (macrophytes) may be more common in the downstream portions of this reach; they may exist in large enough quantities that their photosynthesis and respiration patterns were contributing substantially to the near-surface daily oxygen cycles there. Macrophytes are not yet included in this Link–Keno model, although CE-QUAL-W2 does include algorithms to simulate macrophyte populations (Cole and Wells, 2008).


An analysis of the modeled oxygen fluxes for a representative year (fig. 19) indicated that besides algal photosynthesis, reaeration from the atmosphere was an important source of oxygen in fall, when hypoxic or anoxic conditions in the river caused a large oxygen deficit relative to its solubility and, therefore, a large gradient for absorbing atmospheric oxygen. The reaeration flux also showed that there were short periods, during supersaturation due to algal photosynthesis, when the river outgassed oxygen to the atmosphere. Sinks of oxygen, as expected, were dominated by decay of algae and organic matter primarily on the channel bottom and in the water column. Algal respiration and ammonia nitrification were smaller sinks of dissolved oxygen in this reach.


Simulated total SOD, the sum of first-order and zero-order demand, varied both spatially and temporally. For example, from January through mid-June and late December 2008, modeled SOD was less than 1.5 (g O₂/m²)/d (fig. 20). Modeled SOD was elevated in summer and fall 2008 with values up to about 15 (g O₂/m²)/d. Elevated SOD was due to warmer temperatures and to the settling and decay of labile particulate organic matter and blue-green algae. An in-situ SOD study for the Link–Keno reach took place in early June 2003, before the summer influx of algae and organic matter (Doyle and Lynch, 2005). Those early June SOD measurements ranged from 0.3 to 2.9 (g O₂/m²)/d, with a median of 1.8 (g O₂/m²)/d, values in line with early June SOD from the model. In-situ SOD measurements require oxygen to be present in the water column and so cannot be made during the anoxia that was present through much of the reach in summer. However, summer in-situ SOD greater than 10.2 (g O₂/m²)/d has been measured in Upper Klamath Lake (Wood, 2001). A 1-in. thick layer of “fluffy” material topped the sediments during that measurement and it was theorized that the source of this material was settled algal mats.


The actual exerted SOD was less than the full potential SOD for times or locations where the river was anoxic. For example, the highest simulated SOD in July and August 2008 was near Link River (fig. 20), because enough dissolved oxygen was present in the water column to support the full expression of SOD. Substantial deposition of organic particulate materials occurred both there and downstream, but SOD could be expressed only when oxygen was present; anoxia prevents the expression of additional SOD. The organic sediment continued to build up, however, and this led to periodic spikes in the simulated SOD rate in September, when dissolved-oxygen concentrations increased and allowed SOD to be expressed, as long as decaying organic matter was still present in the sediment.


The model does not currently simulate mechanical reaeration in Keno Dam releases, which results in an underprediction of dissolved-oxygen concentrations downstream of the dam during summer. If prediction of dissolved-oxygen concentrations downstream of Keno Dam becomes important, equations with empirical coefficients will need to be determined to relate dissolved-oxygen reaeration and spill rates (Cole and Wells, 2008); these equations will need to be developed specifically for Keno Dam. Reaeration equations also could be developed for use outside of the CE-QUAL-W2 model as a post-processing routine.


Goodness-of-fit statistics were calculated to compare hourly measured dissolved-oxygen data and model output at the same location and time (table 6). For 2006–09, the ME averaged over all sites for a year was between -0.66 and 0.51 mg/L. The MAE was between 1.01 and 1.30 mg/L. These results show relatively little bias and demonstrate that overall seasonal patterns were captured by the model. Some improvements might be possible through the simulation of macrophytes and a deeper understanding of some of the more-dominant factors and processes affecting dissolved oxygen.


First posted July 14, 2011