Topography and Hydrology

The Beaver Creek basin (fig. 1) is in the flat, northwestern lowlands of the Kenai Peninsula. The peninsula is defined by Turnagain Arm to the north, Cook Inlet to the west, Prince William Sound to the east (not shown), and the Gulf of Alaska to the south and southeast. The Kenai Peninsula has a sharp topographic divide between the Kenai Mountains rising to the east and the flat Kenai Lowlands (from Karlstrom, 1964) stretching west. The Kenai Lowlands are 20–50 miles wide, 106 miles long, and have local topographic relief ranging from a few feet to more than 200 feet. The topography in this lowland landscape reflects a series of glacial advances and retreats that have left behind a poorly drained, wetland-rich, flat surface disrupted by moraines, kettle depressions, and relict glacial drainageways (Karlstrom, 1964).

The Beaver Creek Basin is about 13 miles wide north to south and east to west. The major hydrologic features near Beaver Creek are highlighted in figure 2. The creek generally runs north to south through a valley that was carved by glacial outwash and is now oversized relative to the modern Beaver Creek. The western edge of the basin is about 2 miles from a steep bluff down to Cook Inlet, an embayment of the Pacific Ocean. Areas west of the basin drain into Cook Inlet. Bishop Creek forms a surface-water divide to the northwest and the Swanson River forms a boundary to the northeast. East and southeast of the study basin is a large, poorly drained hummocky area with numerous lakes at the boundary of the Moose River, Swanson River, and Soldotna Creek basins. The headwaters of the west fork of the Moose River begin to form a mapped channel about 2.7 miles east of the edge of the Beaver Creek Basin. Southeast of the basin is Soldotna Creek, and the Kenai River forms the southern boundary. The Kenai River is a key hydrologic feature in this region and is tidally influenced from Cook Inlet for several miles upstream from its outlet; this zone includes the outlet of Beaver Creek (Morton and others, 2015). There are four named lakes that drain into the headwaters of Beaver Creek including Beaver Lake, Tern Lake, Lake Ootka, and Ermine Lake. Several smaller lakes are present across the basin, particularly along the eastern and northern edges. Wetlands (fig. 2) cover 23 percent of the basin (Morton and others, 2015), with large wetland complexes on the western side of the basin and throughout the wide, relict glacial drainageway that the Beaver Creek channel now occupies (U.S. Fish and Wildlife Service, 2020).

Climate

The Kenai Peninsula has a subarctic climate that is in the transition zone from the dry, cold continental climate of interior Alaska to the wet, mild coastal climate along the Gulf of Alaska (Karlstrom, 1964). Summers are cool and short, with frequent cloud cover. Winters are long, with snowpack often lasting from October to May. Winters bring a mix of temperatures with periods well below freezing and times where temperatures remain above freezing (Hayward and others, 2017). The Beaver Creek Basin straddles the divide between the Köppen climate classification of “Dsc” to the west and “Dfc” to the east. “Dsc” and “Dfc” are cold, subarctic continental climates that differ only by the amount of summer precipitation relative to winter precipitation; “s” climates have dry summers, whereas “f” climates are fully humid with no significant precipitation differences between seasons (Kottek and others, 2006; Geodiode, 2022). Storms across the Kenai Peninsula typically move in a counterclockwise direction from the Gulf of Alaska, into Prince William Sound, and then west. This predominant weather pattern causes the Kenai Mountains to form a partial rain shadow in the northwestern part of the Kenai Peninsula where Beaver Creek Basin is located (Hayward and others, 2017).

Weather data have been collected since 1899 (with a large gap from 1908 to 1943) at the Kenai airport weather station (National Oceanic and Atmospheric Administration [NOAA], 2023a) located just southwest of the basin edge (fig. 2). Continuous daily weather records from this station were used to assess climate trends in temperature and precipitation. The Kenai weather data were compared to nearby weather stations in Sterling (NOAA, 2023b) and Soldotna (NOAA, 2023c) for more recent periods when all three stations were collecting data. Precipitation totals were similar at these stations until 2015 when the Kenai station record started to deviate with lower total precipitation despite showing a complete record for this period. It is unknown why the Kenai station recorded less precipitation, but based on streamflows at the Beaver Creek near Kenai, Alaska, streamgage (USGS site 15266500; hereafter referred to as the “Beaver Creek streamgage”), it appears the Kenai station precipitation data do not match conditions in the basin for recent years and may be erroneous. Therefore, Kenai station precipitation data were used through 2014 and Soldotna station data were used from 2015 to 2023; Kenai station temperature data were reasonable and used for the full period. The annual precipitation for 1944–2023 is shown in figure 3. Average annual precipitation is 19.4 inches, with the study period of 2019–23 having above average precipitation.

The average monthly temperatures (fig. 4) show a maximum average monthly temperature of 16.8 °C in July and a minimum average monthly temperature of −14.8 °C in January. A plot of average growing season air temperatures (May–Sept; fig. 5) shows a warming trend, with the warmest average growing season temperature of 12.3 °C occurring in 2019. The summer of 2019 was notably warm in Alaska with stream temperatures in the Yukon River basin exceeding the thermal thresholds above which adult salmon show decreased spawning success and early mortality (von Biela and others, 2020) while Alaska’s most expensive wildfire to date burned close to 170,000 acres just east of Beaver Creek Basin (University of Alaska Fairbanks, 2022).

Looking forward, a 2017 report (Hayward and others, 2017) on climate change vulnerability in the Kenai Peninsula summarized Scenarios Network for Alaska and Arctic Planning (SNAP) climate projections for the region. The SNAP climate projections show a 2–3-°C increase in July air temperatures and a 3–3.5-°C increase in January air temperatures across the Kenai Peninsula during the next 50 years. Spring thaw is expected to come 3–10 days earlier by the 2060s, with an increase in the warm season length from 200 to 230 days for Kenai (Hayward and others, 2017). By the end of the 21st century, precipitation in Kenai is generally expected to increase (SNAP, 2022). The fraction of precipitation days that include snow (snow-day fraction) in 2030–59 is expected to decline by 22.7 percent in the Kenai Lowlands relative to the 1970–99 average (Hayward and others, 2017). However, the SNAP projections for precipitation are more uncertain than projections for temperature and have a known wet bias in the Kenai Mountain rain shadow where Beaver Creek is located (Bieniek and others, 2016). Additional discussion of the SNAP climate projections and their use in the model is provided in appendix 7.

Geology

The geology of the study area consists of thick Quaternary deposits over Tertiary sedimentary bedrock of the Kenai Group. The bedrock forms a structural trough (Karlstrom, 1964) with reported depths to bedrock of 500–600 feet (ft) near Kenai and as much as 750 ft near Nikiski (Freethey and Scully, 1980; Bailey and Hogan, 1995; Glass, 1996). The bedrock surface becomes shallower towards the southern Kenai Peninsula, where it is exposed in road cuts and beach cliffs, and to the east, where it eventually outcrops as the Kenai Mountains. Thick glacial deposits have since covered the bedrock surface during a complex glacial history. Five major glacial advances occurred in the Pleistocene, from oldest to youngest: the Mount Susitna, Caribou Hills, Eklutna, Knik, and Naptowne glaciations of Karlstrom (1964) (hereafter referred to as “Mount Susitna,” “Caribou Hills,” “Eklutna,” “Knik,” and “Naptowne”). During these periods, ice moved from the surrounding mountain ranges, many of which still have glacial ice today, toward Cook Inlet—east and south from the Alaska and Aleutian Ranges, south from the Talkeetna Mountains, and west from the Chugach Mountains and Kenai Mountains. An overview of the relevant glacial geologic interpretations of the area and the deposits they left across the study basin is provided in appendix 1.

Wilson and Hults (2012) compiled available geologic maps into a geologic map of the region, which became the basis of the surficial geologic units used for this study. Boundaries from Wilson and Hults (2012) were improved using information in a USGS (2010) digital elevation model to better align unit edges with topography. These surficial units were further updated with information from Reger and others (2007) on the more recent Killey stade of Karlstrom (1964) moraine and outwash braidplain deposits. The major surficial geologic units across the study area, as interpreted for this study, are shown in figure 6.

Hydrostratigraphic Units and Aquifer Properties

The interpretations of the glacial geologic history of the study area vary somewhat in the available literature (for example, Karlstrom, 1964; Reger and others, 2007), but there is general consensus among prior hydrologic studies in the region on major hydrostratigraphic units (for example, Anderson, 1971; Anderson and Jones, 1972; Freethey and Scully, 1980; Nelson, 1981; Bailey and Hogan, 1995; Homer Soil & Water Conservation District, 2006; R&M Consultants, Inc., 2007; DOWL, 2015). These hydrostratigraphic units are shown in table 2 and include the following:

The model hydrostratigraphic units were modified from the surficial geology in Wilson and Hults (2012) and Reger and others (2007), with depths and thicknesses of the subsurface units informed by available borehole data (AKDNR, 2022). The borehole data were plotted in several cross-sections across the study area to assess the spatial distribution of the confining unit as well as spatial patterns in the lithologies of the surficial units mapped in figure 6. A generalized cross-section (fig. 7) was developed using the pattern of lithologies observed in the borehole data. Additional details on the development of the geospatial geology datasets for the model is discussed in appendix 1.

Water Use

Water use data for permitted withdrawals were obtained from the Alaska Water Use Data System (AKDNR, 2023). The reported data are for larger withdrawals that exceed typical household water use from a private well. The only reported water uses within the basin in recent years (2020 and after) were the City of Kenai municipal wells. Well construction information including screen depths and lengths were obtained from the Alaska Well Log Tracking System database (AKDNR, 2022). Pumping rates were available in Alaska Water Use Data System for 2011–22. City of Kenai total pumping rates declined slightly from 2011 to 2018 and then were more consistent through 2022 (fig. 8).

Conceptual Model

Water enters the basin primarily as precipitation (rainfall or snowmelt), which is partitioned into evapotranspiration, surface runoff, and net infiltration through the soil zone to the groundwater system (recharge). Groundwater originating from net infiltration or surface-water leakage ultimately discharges to streams, lakes, and wetlands, but may also be taken up directly by vegetation. A small amount of groundwater (approximately 0.03 cubic meter per second [m³/s] or about 4 percent of average base flow in Beaver Creek) is withdrawn by the City of Kenai municipal wells, and some groundwater originating in the westernmost part of the topographic basin likely discharges to Cook Inlet, a competing sink that is closer and lower in elevation. Similarly, some groundwater originating in the hummocky (and streamless) area between the Beaver Creek, Soldotna Creek, and Moose River basins probably flows into the Beaver Creek Basin along its eastern margin. Within the basin, groundwater may cycle through streams, lakes, and wetlands, depending on their elevations relative to the water table. Nearby age dating of wells in the upper aquifer showed relatively young groundwater ages of about 20 years (Glass, 2002).

Historical records of stream discharge where Beaver Creek crosses the Kenai Spur Highway (USGS site 15266500) indicate a base flow-dominated system, where base flow between discharge peaks constitutes approximately 70 percent of the total discharge during the months of May through September (fig. 9). In many natural systems, groundwater discharge makes up the bulk of this base flow. In Beaver Creek, flat topography and extensive wetland areas may slow the flow of surface runoff compared to steeper basins and provide surface storage that results in a substantial surface-water component of base flow. This slow runoff has more exposure to solar radiation and can heat up when at or near the land surface (Sjöberg and others, 2021), providing a steady but warmer source of base flow to Beaver Creek. Deeper and cooler groundwater discharge to Beaver Creek may originate primarily from the upper unconsolidated aquifer, with the upper aquitard limiting interactions between the lower semi-confined aquifer and the surface-water system. Near the City of Kenai wells, pumping from the lower semi-confined aquifer may pull water downward through the leaky aquitard, diverting some water that would otherwise discharge to the creek.

Water temperature in Beaver Creek is affected by air temperature; the temperatures of inflowing groundwater; surface runoff; surface water from features such as Beaver Lake; and the interplay of solar radiation, shading, channel width, and flow velocity, among other factors (for example, Bartholow, 1989). Although vegetative shading can mitigate the warming effects of solar radiation (for example, Garner and others, 2017; Fehling and Hart, 2021), shading potential along much of Beaver Creek is limited by the wide valley and predominance of short wetland vegetation. Narrow tributaries that extend into wooded areas or are overhung by alder and other shrub or scrub vegetation may have the best shading potential. Conversely, upstream areas with low flow velocities through wide wet areas of short emergent vegetation may be especially vulnerable to summer temperature increases.

In general, groundwater temperature in cold regions such as the Kenai Lowlands reflects the mean annual air temperature plus some offset representing the insulating effects of the snowpack (Zhang, 2005). Temperature fluctuations at the land surface are dampened exponentially with depth and shifted linearly in time as heat is carried downward from the surface by infiltration, conduction, and vertical groundwater flow (Suzuki, 1960; Stallman, 1965). The temperature of groundwater discharge entering the stream is therefore a function of groundwater flowpath depth (Kurylyk and others, 2015). Shallow groundwater discharge may fluctuate in temperature seasonally but lag air temperatures (for example, with cooler temperatures in early summer and higher temperatures in late summer or early fall). Groundwater originating from deeper than a few meters will be mostly attenuated in temperature amplitude, reflecting the mean annual air temperature plus the offset from snowpack insulation. Groundwater temperatures at greater depths will increasingly lag multidecadal changes in air temperature at the surface, reflecting, in part, past temperatures (Kurylyk and others, 2015).

As noted earlier in the “Introduction” section, Mauger (2013) determined Beaver Creek to have a low sensitivity to air temperature, presumably owing to thermal buffering from cold groundwater. However, similar seasonal amplitudes and phases for air temperatures and stream temperatures collected for this study (fig. 10) indicate stronger “air coupling” (for example, Hare and others, 2023). “Tributary 6” and the lower part of the west fork (fig. 11; sites TL–8 and TL–14 in fig. 10) may have greater proportions of groundwater discharge, as evidenced by cooler summer temperatures in 2022 and winter temperatures above 0 °C. These and other similar locations may serve as thermal “refugia” (for example, Sedell and others, 1990) during warm periods. Other locations in the stream network, including TL–14 during June and July of 2023, show stream temperatures greater than the air temperature (fig. 10). Stream temperatures in excess of air temperature could be an artifact of those locations being farther inland from Cook Inlet than the Kenai airport (and therefore warmer), or indicative of the slow release of water from surface-water storage that is subject to radiative warming. Warm surface runoff during mid- to late summer rain events that are common in this area could dramatically affect temperatures throughout the stream network for days or longer (Feinstein and others, 2022).

A key question for the future sustainability of salmon habitat in Beaver Creek is the extent to which base flow is dominated by groundwater, and the thermal sensitivity of that groundwater to long-term warming of air temperatures. Thermal sensitivity can be conceptualized across different timescales by an “effective depth” parameter that represents the average depth of groundwater flowpaths discharging to a stream (Kurylyk and others, 2015). Deep groundwater can potentially lag surface temperatures in warming for decades and therefore provide a buffer against warming air temperatures, especially during dry periods (Kurylyk and others, 2015; Briggs and others, 2018).

Monitoring Well Network

Groundwater levels and water temperatures were measured by the Kenai Watershed Forum (https://www.kenaiwatershed.org/) in five shallow monitoring wells installed for this study (fig. 11). Well depths varied from 1.3 to 4.3 meters below land surface (table 3). Wells 3 and 4 were colocated to observe vertical hydraulic gradients. The wells were instrumented with pressure transducers for continuous recording of hourly groundwater level and temperature data from July 2022 to September 2023. The transducers record absolute pressure, and a barometric logger was deployed to compensate the data for air pressure changes. Manual measurements of depth to water from the top of the well casing were made during each visit to correct for instrument drift and convert the transducer data from water levels to depths below land surface. Raw well data and data with the applied corrections are available in a USGS data release (Koch and others, 2024).

Temperature Sensors

Fourteen temperature loggers were installed in Beaver Creek from July 2022 to September 2023 to record continuous water temperature at a network of locations across the basin (fig. 11). Temperature loggers were operated by the Kenai Watershed Forum following methods from Mauger and others, 2015 for stream temperature monitoring in south-central Alaska, including verifying that sites are well-mixed, pre- and post-deployment quality control checks against a National Institute of Standards and Technology thermometer, and biannual site visits. Site TL–12 is a “long-term” monitoring site where Kenai Watershed Forum has collected stream temperature data since 2010.

Streamflow

Continuous stream discharge data for Beaver Creek at the Kenai Spur Highway are available from June 2022 to September 2024, as well as in the 1960s and 1970s (USGS, 2023; Beaver Creek streamgage; fig. 11). Streamflow at 12 upstream sites along Beaver Creek were measured during synoptic flow surveys on June 8 and July 4–7, 2022 (Koch and others, 2024; locations and values shown on fig. 12).

Shading

Stream channel shading data were collected in July 2022 near the synoptic streamflow locations (fig. 12) using a densiometer and following the methods of Fitzpatrick and others (1998). Shading data were grouped by wetland type (fig. 13) to estimate ranges of shade for the riparian vegetation communities (emergent, scrub-shrub, mixed emergent and scrub-shrub; and non-wetland for riparian areas not mapped as wetlands; U.S. Fish and Wildlife Service, 2020). Representative values and ranges were then applied across the basin for the SNTEMP model using the mapped wetland communities (appendix 5). A summary of the measured shading information for the wetland types is provided in table 4; shading measurements by point location are available in a USGS data release (Koch and others, 2024).

Based on the wetland descriptions in USGS (1996), palustrine emergent wetlands in Alaska are generally marshes and wet meadows dominated by herbaceous plants and sedges that are typically not tall enough to provide much shade to the stream channel. These wetland corridors had the lowest estimated shade in the Beaver Creek Basin (10 percent) and are the most common wetland plant community along the main stem and east fork valleys. Palustrine scrub-shrub wetlands are common near stream channels in Alaska, are dominated by woody plants and perennial herbs, and should provide more shade to a nearby stream channel than emergent wetlands. These wetland corridors had an average estimated shade of 37 percent and are common in the southern tributaries and near the confluence of the main stem and east forks (fig. 13). Mixed palustrine and scrub-shrub wetlands should have a range in shading values between emergent and scrub-shrub but only one measurement site was in a mixed setting because many of these reaches were in inaccessible parts of the basin (along the north and west forks). Therefore, the range and average in table 4 for the mixed category are from all the emergent, scrub-shrub, and mixed measurement points. Most of the creek valley is covered by some type of mapped wetland. The non-wetland category covers areas that are mostly in the smaller tributary valleys, which had narrower valleys with more shrub cover. Two representative shading measurements were collected in the west fork in non-wetland stream corridor locations and had an average of 62 percent shade.

Discretization

The MODFLOW 6 model is discretized on a uniform grid of 50-meter cells with four layers that represent the hydrostratigraphic units described in table 2. Temporally, the historical MODFLOW 6 model simulates an initial steady-state period representing average conditions between January 1, 2015, and January 1, 2019, followed by transient 7-day periods from January 1, 2019, to October 1, 2023. Periods of 1 or 2 days are included at the end of each calendar year so that a new 7-day period starts on January 1, and the periods within each year span consistent days of the year (a requirement of SNTEMP).

Boundary Conditions

Boundary conditions include sources, sinks, and surface-water levels that control the groundwater flow system. The following boundary conditions were included in the MODFLOW 6 model (fig. 15):

Aquifer Properties

Aquifer properties define how easily water can move through the modeled aquifer (hydraulic conductivity) or be released from aquifer storage (storativity). The limited information from previous studies on the aquifer properties for the hydrostratigraphic units beneath the Beaver Creek Basin is summarized in table 5. Most of the estimates do not specify if they are horizontal or vertical hydraulic conductivity. In addition, the applicability of these measurements to the groundwater model is limited by small measurement volumes (especially for drill cuttings and lab permeability tests) and the potential for bias in methods that disturb the original pore structure. In general, larger hydraulic conductivities would be expected at groundwater modeling scales compared to lab measurements (Rovey and Cherkauer, 1995). The lower confined aquifer is thought to be more conductive than the upper unconfined aquifer (Anderson and Jones, 1972).

An aquifer test in a production well near North Kenai indicated a storativity of 2.8x10⁻⁴ for the confined aquifer (Anderson and Jones, 1972) but it is unclear whether the test represented the regional confined aquifer or the local upper confined aquifer that is present near Nikiski. An aquifer test in the City of Kenai production wells in the confined aquifer beneath the Beaver Creek valley produced storativities between 2.8x10⁻⁴ and 3.3x10⁻³ (Anderson and Jones, 1972; AKDNR, 2022). Assuming a 60-meter thick aquifer, this would correspond to specific storage values of between 4.7x10⁻⁶ and 5.5x10⁻⁵ (per meter of aquifer thickness).

History Matching Results and Discussion

History matching results for streamflow at Kenai Spur Highway (Beaver Creek streamgage) and stream temperature at the Kenai Watershed Forum long-term monitoring site TL–12 (located just downstream from the streamgage; fig. 11) are shown in figures 17 and 18. Detailed history matching results for all observations are available in appendix 6 and the companion data release (Leaf and others, 2024). These figures include the base member of the history matching ensemble that was selected as the “best” model and results for each ensemble member, with the spread in these lines indicating model uncertainty (in the input parameter values that were adjusted during history matching). Places where the measured values fall outside of the ensemble results indicate structural error in the model (stemming from oversimplification, missing processes, or other factors in the model design) that was not interrogated in the history matching process. For example, the offset between the observed and simulated spring freshet peaks in May 2023 may stem from snowmelt being simulated too quickly in the SWB model, which was not included in formal parameter estimation; unsaturated zone processes that are not considered in the groundwater flow model (Hunt and others, 2008); or surface runoff delays or storage that are not considered in the SWB model. Oversimulation of low flows during the summer of 2022 and winter of 2022 and 2023, and undersimulation of streamflow in the fall of 2022 and other times may indicate a tradeoff in optimizing the fit of the full time series.

In general, stream temperatures are well-fit by the model across the longest time-series at site TL–12 (fig. 18). Across all sites, a root mean squared error in temperature of 1.35 °C was achieved, with some bias in the ensemble varying by site and through time (appendix 6).

MODFLOW Model Results and Discussion for Study Period and Future Scenarios

The MODFLOW model provided an estimate of groundwater flow throughout the Beaver Creek Basin including exchanges between the groundwater and surface-water system and the water table surface during the study period (fig. 19). The average growing season basin conditions show neutral or loss of stream water to the groundwater system in the headwaters of most tributaries, with the strongest groundwater inflows to the stream along the lower two-thirds of the main stem of Beaver Creek. These conditions are consistent with many of the basin headwaters being in topographically flatter, wetland-rich areas and the main stem occupying the relict glacial drainageway that provides more topographic relief to support steeper groundwater gradients.

The simulated water balance during the warm growing season of May through September was evaluated for the historical and potential future climate scenarios. Average total flows simulated May through September at the mouth of Beaver Creek are shown in figure 20. Mean growing season streamflow in Beaver Creek at the outlet with the Kenai River is between one-third and one-half runoff and the rest is base flow, making this system slightly base flow dominated. During future summer periods, streamflow may remain similar to the study period or increase somewhat with more runoff. The slight jump in streamflow between the study period, which uses observed climate data, and the subsequent 2024–32 future period, which uses climate projection data, indicates that with the rain shadow adjustment, running the model with projection data can produce reasonable streamflow estimates, but they might be biased slightly high. The lack of obvious trends from 2024 to 2032 and 2042 to 2050 indicates that total streamflow volumes and the base-flow portion in Beaver Creek Basin is not expected to change dramatically by the mid-21st century.

Stream Network Temperature Model Results and Discussion for Study Period and Future Scenarios

Simulated stream temperatures were assessed using the Alaska water quality standards for stream temperature. Based on the anadromous fish life cycles discussed in the “Introduction” section, the egg and fry incubation and spawning standard of 13 °C is critical during the late summer and fall for coho salmon, early to mid-summer for Chinook salmon, mid- to late summer for sockeye salmon, and late fall to early winter for Dolly Varden. Between these four fish species, the 13-°C standard could be applicable for spawning and egg incubation throughout the full May–September simulation period. If species-specific timing is of interest to a future user, monthly exceedances could be readily calculated using the SNTEMP results in the data release (Leaf and others, 2024). The migration and rearing standard of 15 °C is applicable across the full growing season and for all four anadromous fish species, which each spend at least one full summer season in the basin. The 18 AAC 70 Alaska water quality standards also state that stream temperature should not exceed 20 °C at any time (Alaska Department of Environmental Conservation, 2024).

Maximum 7-Day Mean Stream Temperatures

Maximum weekly stream temperatures were selected from the SNTEMP results as the maximum 7-day model period mean stream temperature in each year at a given location. Observed equivalents were developed from the field data by averaging temperatures during each 7-day simulation period and selecting the annual maximums. This method is equivalent to the method in Kyle and Brabets (2001), who computed means for 7-day periods occurring on the same days of each year, but differs from the maximum weekly average and maximum weekly maximum temperatures presented by Mauger and others (2017), which were based on 7-day rolling averages of daily mean and daily maximum temperatures, respectively. The 7-day temporal resolution of the SNTEMP model in this study precludes the use of rolling means. Discrete 7-day averaging periods may underrepresent extremes compared to rolling means if the extreme temperature events are split between two 7-day periods.

The maximum 7-day mean stream temperatures simulated in each year, across the seven future scenarios, are compared in figure 21. To correct for bias between observed and simulated temperatures while retaining the relative differences between the “shallow” and “deep” (referred to here as “steady”) groundwater variants, the mean offset from the observed 2019 maximum 7-day temperature (for a model stress period) was computed across all variants and subtracted from the simulated temperatures for that scenario. For the three climate scenarios, subtracting the mean offset resulted in 2019 simulated results where the “shallow” groundwater variant was greater than the 2019 observed maximum, and the “deep” (“steady”) groundwater variant was less than the 2019 observed maximum (fig. 21). The results indicate summer maximum temperatures like 2019 becoming more frequent or even common by mid-century. Considering a typical difference of about 1.5 °C between maximum weekly average temperatures (similar to the 7-day model period maximums shown in fig. 21) and maximum weekly maximum temperatures (Mauger and others, 2017), the results indicate that maximum weekly maximum temperatures in excess of 20 °C may be common by the 2040s. Nearly identical results for the “ccsm4” and “ccsm4-pop-growth” scenarios, which differ only in the simulation of the municipal pumping increase, indicate that a 0.44-percent annual increase in municipal pumping from 2023 to 2050 would have little to no effect on annual maximum 7-day temperatures in Beaver Creek. This finding was consistent across the other temperature metrics presented in this section (Leaf and others, 2024).

Duration of Temperatures Above Water Quality Standards

Simulated shifts in the distribution of 7-day mean stream temperatures at site TL–12 during the annual simulation period of May through September are shown in figure 22. The 5-year periods of 2019–23 and 2046–50 were combined to produce the temperature duration curves. Overall, the results indicate roughly 34 to 63 additional days (on average each year) with average temperatures above the 13-°C threshold. For the 15-°C threshold, 14 to 81 additional days with average temperatures above the threshold are anticipated. In 2019–23, weekly average temperatures above 15 °C only occurred for 3 weeks in 2019 and 2 weeks in 2023 (Leaf and others, 2024).

Average annual exceedances of the 15-°C migration and rearing standard (fig. 23), the 13-°C egg and fry incubation and spawning standard (fig. 24), and the 20-°C maximum water temperature standard (fig. 25) were calculated as the average number of 7-day periods exceeding the thresholds during the time periods indicated, expressed as a number of days per year. Only the “deep” or “steady” groundwater scenario variants are shown in this report; results for the “shallow” groundwater scenarios are available from Leaf and others (2024). Exceedances generally show an increase from the study period to the late 2040s. Exceedances are greatest for the gfdl3 scenario, followed by the cssm4 and snap-mean scenarios. These exceedances are nonuniform across the basin, with the northern headwaters at tl3 (main stem headwater) showing the highest number of exceedances (between 60 to almost 100 days a year with temperatures greater than 15 °C, with 22–42 additional days per year from the study period to the mid-21st century). Site TL–8 (tributary 6) had the least number of exceedances (less than 20 days per year and only 2–13 additional days per year from the study period to the mid-21st century). Similar to other tributaries in the basin, tributary 6 is dominated by low-gradient wetlands upstream with little to no groundwater input but has cooling groundwater inflows in its lower reaches. This tributary also has higher shade than the upper basin or west fork. The headwaters of the main stem have relatively small groundwater inflows and little shade. Lower reaches have more shade and larger groundwater inflows that moderate temperatures and reduce the number of exceedances moving downstream.

Exceedances of the 20-°C standard are rare in the study period but may occur regularly by the mid-21st century, with the main stem headwaters having 1–20 days per year greater than 20 °C and the lower main stem 0–2 days per year. The tributaries vary from few to no exceedances in the west fork or tributary 6 to as much as 15 additional days per year greater than 20 °C at the east fork downstream from Beaver Lake.

Spatial Distribution of Temperatures During Heatwaves and Effects of Surface Runoff

The spatial distribution of simulated stream temperatures in relation to the stream hydrology during a hypothetical mid-century heatwave is shown in figure 26. Hot and dry conditions simulated for the week of July 30, 2050, in the gfdl3 “deep” groundwater scenario are shown in figure 26A–C. Most reaches of Beaver Creek and its tributaries are receiving groundwater inflow, which accounts for 100 percent of flow into Beaver Creek. Stream temperatures are generally coolest in the downstream reaches of tributaries 5, 6, and 9 (fig. 12) and in the main stem. Cool temperatures simulated in the headwaters of the West Fork may not be accurate. There are no stream temperature data in this area, and groundwater levels in well 2, which is within about 2 kilometers of the creek, were oversimulated by several meters in the model (fig. 6.4), which likely means that groundwater discharge to the West Fork headwaters is also being oversimulated, resulting in cooler than actual temperatures.

Hot and wet conditions simulated for the following week of August 6, 2050, in the gfdl3 “deep” groundwater scenario are shown in figure 26D–F. Air temperature is similar to the previous week, but surface runoff constitutes 85 percent of flow into the creek. Most of the lower main stem and lower reaches of tributaries 5, 6, and 9 are still receiving groundwater, but the overall volume of surface runoff is sufficient to raise stream temperatures to approximately 20 °C throughout the system. The lower reaches of tributaries 5, 6, and 9 maintain the coolest temperatures while the mainstem is relatively warm compared to the tributaries. These results illustrate the importance of surface runoff as a variable affecting stream temperature, especially in the Kenai region, where late summer rains are typical.

Limitations of Soil Water Balance

The SWB simulation partitions precipitation into evapotranspiration, surface runoff, and net infiltration past the root zone. The pattern of precipitation across the basin was assumed to be adequately represented by the weather data collected at the Kenai and Soldotna airports based on the small basin size, flat topography, and lack of appreciable spatial trends in gridded precipitation products. The overall partitioning of precipitation into evapotranspiration and runoff to Beaver Creek was constrained by manual history matching of simulated and observed annual mean total flows at the Beaver Creek streamgage. Partitioning of the runoff component into groundwater recharge and surface runoff was partially constrained during history matching of the groundwater flow and stream temperature model.

Global multiplier estimates of 1.7 for surface runoff and 0.6 for groundwater recharge indicate undersimulation of runoff by the SWB model and oversimulation of recharge, which may be a result of shallow water table depths throughout much of the study area. In the modeling analysis, net infiltration is calculated by SWB without accounting for the water table position; all net infiltration is assumed to enter the groundwater system as recharge. In places where the water table is shallow, this assumption can result in unrealistic recharge and simulated heads that are above the last surface. In reality, groundwater discharge and surface runoff occur when the water table reaches the land surface. Although the global recharge and runoff multipliers compensate for this defect at a coarse scale, future work could potentially improve the spatial and temporal simulation of the recharge and surface runoff components by formally including the SWB model in the formal history matching, and by representing the effects of a shallow water table with the MODFLOW Unsaturated Zone Flow Package (for example, Feinstein and others, 2020).

In this analysis, surface runoff generated by SWB is assumed to reach the stream channel within each 7-day time period; no surface-water storage is considered. This assumption could potentially be problematic if appreciable surface-water storage is actually creating longer recession times for warm rain events. Inspection of streamflow peaks in figure 17 and simulated and observed temperatures (fig. 18 and appendix 6) indicates that the assumption of little to no surface-water storage is reasonable.

Early simulation of the observed spring 2023 freshet may indicate an issue with the simulation of snowmelt in the SWB model. This possible issue could be evaluated further by extending the historical model to simulate the 2024 spring freshet and could be potentially corrected by including the SWB model in the formal parameter estimation. In the context of evaluating warm stream temperatures, early simulation of snowmelt is likely conservative, in that temperatures simulated later in the spring would be warmer than the actual stream temperatures.

Limitations of Groundwater Model

The groundwater model in this study was primarily designed to simulate streamflows and the stream water balance for the stream temperature model. Relatively little is known about the subsurface beneath the Beaver Creek Basin other than that it consists of heterogenous glacial deposits of greater than or equal to 500 ft in thickness, which reflect a complex glacial history (appendix 1). Large ranges in simulated heads across the historical model ensemble reflect uncertainty in the aquifer property inputs. The places or times where the ensemble results are poorly fit to observed heads reflect additional structural error in the model stemming from simplifications including relatively coarse vertical resolution, coarse spacing of pilot point input parameters, coarse grid spacing that may skew simulation of groundwater levels at the study wells close to Beaver Creek, and the previously mentioned unrealistic treatment of places where the water table is shallow. Simulated heads are most constrained near surface-water boundary conditions that are relatively well characterized and exert strong influence on nearby groundwater levels.

Inspection of the streamflow history matching results (fig. 17 and appendix 6) indicates a model tendency to oversimulate low stream flows and undersimulate some high flows. These errors could be the result of limiting assumptions in the surface-water balance as well as the neglection of unsaturated zone processes including delays in groundwater recharge from infiltration time, evapotranspiration of shallow groundwater, and groundwater seepage to the land surface. Good agreement between simulated and observed stream temperatures and limited variability in the ensemble of simulated stream temperatures (fig. 18 and appendix 6) indicate that the effect of these deficiencies on the simulation of stream temperatures is acceptable.

Water use by the City of Kenai was assumed to be constant across years for each month of the year (for example, the same water use was assumed in July of each year). Given the small magnitude of water use, which represents approximately 4 percent of average base flow in Beaver Creek, the effects of this assumption are assumed to be negligible.

Limitations of Stream Temperature Model

The 7-day temporal resolution of the stream temperature model carries several limitations. A key limitation is the inability to simulate individual daily maximum temperatures. Extreme temperatures occurring at subweekly scales cannot be represented. The results here are similar or equivalent to the weekly average temperatures reported by past studies (for example, Kyle and Brabets, 2001; Mauger and others, 2017), but fixed 7-day periods can underrepresent extremes compared to rolling averages if a warm period is split between 2 weeks.

Although the results here cannot be compared directly to the maximum weekly maximum temperatures presented by Mauger and others (2017) and KPFHP (2022b), which average daily maximum temperatures, supplementary information from Mauger and others (2017) and data collected for this study indicate a typical difference between maximum weekly average temperatures and maximum weekly maximum temperatures of about 1.5 °C. Adding 1.5 °C to the maximum 7-day mean temperatures shown in figure 21 can provide an approximation of the maximum weekly maximum temperature metric.

The transient effects of warm runoff from individual summer rainstorms may be overrepresented (that is, the temperature of the rain may in reality be cooler than the average temperature for that week and the residence time of the runoff in the basin may be substantially less than the 7-day model period length). Conversely, the assumption that all surface runoff generated within a 7-day period reaches the stream within that period may underrepresent the effects of surface-water storage across the flat topography of the basin, which may slowly release water to the stream that has been warmed by solar radiation. This phenomenon may explain stream temperatures observed in June and July 2023 that were consistently greater than air temperature (fig. 10).

As explained in appendix 6, the representation of groundwater temperatures is greatly simplified. Similar history matching results with thermally sensitive and insensitive conceptual models indicate a lack of discriminating information in the field data about groundwater thermal sensitivity. Inclusion of the thermally sensitive and thermally insensitive conceptual models in the future scenarios indicates a potential effect of groundwater thermal lag of about 1 °C, which is well within the range of outcomes stemming from uncertainty in the future climate. Colder winter temperatures in the shallow subsurface resulting from a reduced snowpack may counteract this effect, resulting in less warming of groundwater in the near term.

The representation of shading in the model is simplified to four vegetation zones based on a small number of field measurements, which could be highly variable at the scale of an individual site (for example, depending on position relative to individual alder shrubs, and so forth). The model did not account for effects of large undercut banks, which are common throughout the main stem and may provide persistent cool habitat. The model also did not account for heterogeneity in shading, which may explain some of the mismatch between simulated and observed temperatures.

Despite all of these limitations, a mean error of 0.15 °C in stream temperatures simulated by the base member of the ensemble indicates a general lack of bias in the model results. An approximate 4-°C spread in maximum weekly mean temperatures shown in figure 21 indicates greater uncertainty in the future climate, compared to the model uncertainty in simulating historical temperatures (root mean square error of 1.35 °C for the base model). Uncertainty in the future climate is reflected in the wide ranges of projected future temperature increases shown in figures 21–25.

Limitations of the Future Climate Scenarios

The limitations of any future projection based on general circulation models are well understood (for example, Pörtner and others, 2023), with projected temperatures and precipitation varying by model and by scenario. The range of results across the ensemble of climate scenarios presented in this report reflects uncertainty in the nonlinear dynamics of the atmosphere and in future human behavior. Although a wide range of potential outcomes is presented in this report, the actual uncertainty may be wider because only a limited number of downscaled climate projections were available for this area.

Numerous issues were encountered in the available GFDL–CM3 and NCAR–CCSM4 projections. Precipitation from these projections was reduced by 41 and 47 percent, respectively, to match observed precipitation for the historical period. Both projections had smaller, more frequent rain events; a compressed daily temperature amplitude; and some general bias in sun fraction, wind speed, and humidity (appendix 7). The implications of these issues are unclear, but it is perhaps notable that GFDL–CM3 and NCAR–CCSM4 projections (with precipitation reduced as described) and the snap-mean scenario (which repeats historical climate with only the temperatures increased) indicate little change in the overall water balance between the near future and midcentury (fig. 20). A step change in the amount of simulated runoff between the historical period and near future (fig. 20) gives some indication of the effects of the climate projection limitations on the water balance. Differences between historical temperatures simulated by GFDL–CM3 and NCAR–CCSM4 and observed historical temperatures were addressed by applying bias corrections to the results presented in figures 21 and 26. Evaluation of simulated changes instead of absolute quantities (for example, fig. 22) is more robust.