Purpose and Scope

This report presents the results of a study conducted in cooperation with the Alaska Department of Transportation and Public Facilities to identify and improve understanding of special conditions for peak flows through water year 2022 in support of hydrologic analyses for Alaska. By convention, a water year is defined as the period from October 1 to September 30 of the following year and named for the calendar year in which it ends. In this report, water years are denoted with the abbreviation WY. The study focuses on peak flows for USGS streamgages in Alaska and considers basin characteristics for conterminous basins in Canada to facilitate understanding of transboundary basins. For this report, floods were defined by their streamflow magnitude, not the areal extent of flooding, and were limited to annual peak flows because the time series of annual peak flows is readily available and commonly used for flood-frequency analysis (England and others, 2018). Although many special conditions affecting peak flows exist (England and others, 2018; U.S. Geological Survey, 2024b), this report is limited to selected streamflow, basin, and extreme flood conditions that were obtainable from existing datasets and required minimal site-specific analysis to maximize application to a large number of peak flows. The streamflow special conditions identified for this report were regulation and diversion, GLOFs, other outburst floods, snowmelt floods, and floods associated with ARs. The association of peak flows with ARs was limited to considering the presence of an AR on the day of a peak flow to provide a first-order inventory of AR-related peak flows in Alaska while minimizing data analysis beyond the scope of this project. The basin special conditions identified were urbanization, indeterminate drainage area, and drainage areas less than the minimum used for regional analysis. Extreme floods are defined in this report as peak flows exceeding a threshold based on flood-frequency statistics or an empirical measure of relative magnitude, depending on data availability, and were limited to floods observed in modern gaged records (that is, floods since the early 1900s), which excluded paleofloods such as megafloods associated with much older glacial lakes.

Previous Studies

Previous USGS reports discussing special conditions for USGS streamflow data for Alaska include flood-frequency reports (most recently, Curran and others, 2016) and other regional statistical analyses, such as Wiley and Curran (2003). These reports describe selected special conditions, including regulation and GLOFs, and detail their treatment (inclusion or exclusion) for the specific streamgages and analyses in the respective reports.

Historical reports documenting flood-generating mechanisms for regionally extreme and societally impactful floods in Alaska have described the nature and extent of flooding, primary flood-generating mechanisms, secondary processes, and contributing factors. Detailed, peer-reviewed reports described historical floods in calendar years 1967 (Boning, 1972; Childers and others, 1972); 1971 (Lamke, 1972); 1986 (Jones and Zenone, 1988; Lamke and Bigelow, 1988); 1994 (Meyer, 1995); and 2002 (Eash and Rickman, 2004). Other reports in newspapers or newsletters described flooding in 1971 (McGee, 1974), 2006 (Joling, 2006), 2012 (Pearson, 2012), and 2022 (Plumb and Ostman, 2023). Nearly all these large regional flooding events were associated with rainfall. However, regional floods in 1971 included unusually large high-elevation snowmelt floods, and outburst floods occurred in association with processes related to the 1971 and 1986 rainstorms. GLOFs pose hazards to downstream communities and infrastructure and have been described in numerous publications—for example, floods on the Knik River (Bradley and others, 1972); Taku River (Neal, 2007); Snow River (Beebee, 2023); and Mendenhall River (Kienholz and others, 2020).

Primary flood-generating mechanisms were previously inferred for seasonal concentrations of peak flows within nine subclasses of seasonal flow regimes in Curran and Biles (2021). Two subpopulations of peak flows, spring snowmelt and summer-to-winter rainfall, occurred in most seasonal flow regimes dominated by snowmelt or rainfall. In seasonal flow regimes dominated by summer high-elevation-melt processes, peak-flow subpopulations were less distinct and consisted of two groups (spring snowmelt and summer-to-fall rain) or three groups (spring snowmelt, summer high-elevation melt, and fall rain).

Seasonal Flow Regimes

Seasonal flow regimes describe similarities in flow patterns across seasons for a group of streams. The cold climate of the Alaska region and variations in timing of delivery of precipitation facilitate the seasonal dominance of various rainfall-based and melt-based streamflow-generating mechanisms, creating a range of strongly seasonal streamflow patterns. Curran and Biles (2021) grouped hydrographs for 253 streamgages in Alaska into three classes of seasonal flow regimes that each had a different primary daily mean flow driver: (class 1) rainfall, primarily in fall; (class 2) snowmelt, primarily in spring; and (class 3) high-elevation melt, including glacier melt, primarily in summer. These classes were subdivided into nine subclasses that were named using the class plus a letter. The letters were assigned in rough order by increasing dominance of spring snowmelt for classes 1 and 2 and increasing dominance of summer high-elevation melt for class 3. For example, 1A is the most rainfall-dominated regime, 2D is the most strongly snowmelt-dominated regime, and 3C is the most glacier-melt-dominated regime (refer to table 1 of Curran and Biles, 2021, for more details). Hereinafter, this report refers to the subclasses as seasonal flow regimes. Within each seasonal flow regime, Curran and Biles (2021) identified two to three seasonal peak-flow subpopulations corresponding to seasonal spikes in the peak-flow day-of-year density for all streams in the regime and inferred the spring, summer, and fall-to-winter subpopulations to be associated with snowmelt, high-elevation melt, or rainfall, respectively. The dominant peak-flow subpopulation in a regime sometimes varied from the dominant daily mean flow driver, as for the case of regime 2A, a weakly snowmelt-dominated seasonal flow regime where the largest daily mean flows occurred in spring but fall rainfall generated the most peak flows.

Basin Characteristics

Cluster Analysis of Drainage Area-Precipitation Groups

Independently, the drainage area and mean annual precipitation of a basin have a strong influence on peak-flow magnitudes at a streamgage, and the values of each of these basin characteristics vary widely throughout the Alaska region. Understanding the distribution of combinations of drainage area and mean annual precipitation can provide a first-order expectation of peak-flow magnitude characteristics. For example, the largest drainage basins are unlikely to only include the wettest areas, limiting the likelihood of the largest basins having the highest specific peak flow, or peak flow per unit drainage area. To find groups of streamgages with similar combinations of drainage area and mean annual precipitation for analysis of group peak-flow magnitude characteristics, a cluster analysis was performed.

Of streamgages where mean annual basin precipitation estimates were available, 455 had unregulated flows and determinate, non-urbanized basins with drainage areas greater than 0.4 mi² and were selected for a drainage area-precipitation cluster analysis. The selected streamgages were grouped into three drainage area-precipitation categories using a k-means cluster analysis (kmeans function in the stats R package [R Core Team, 2024]) with the logarithms of drainage area and mean annual precipitation as variables. The number of groups was selected using a within-cluster sums of squares plot (fviz_nbclust function in the factoextra R package (Kassambara and Mundt [2020]), aided by visual observation that most streamgages grouped into three quadrants on a logarithmic plot of precipitation against drainage area.

The streamgages clustered relatively cleanly into small and wet (cluster 1), large and dry (cluster 2), and small and dry (cluster 3) drainage area-precipitation groups (fig. 2; Curran, 2025). Cluster-group boundaries were manually simplified to provide a classification tool for future use with other streamgages or ungaged sites (table 1). The simplified boundaries misclassified three streamgages (USGS stations 15264000, 15267900, 15896700) used in the development of the groups; these were not reclassified for the analyses in this study.

Table 1.    

Drainage area and mean annual precipitation values determined as boundaries for estimating drainage area-precipitation groups for streamgages and ungaged sites in Alaska.

[Cluster group boundaries shown in figure 2. <, less than; ≥, greater than or equal to]

Table 1.    Drainage area and mean annual precipitation values determined as boundaries for estimating drainage area-precipitation groups for streamgages and ungaged sites in Alaska.

Cluster group identifier	Drainage area precipitation category	Drainage area (square miles)	Mean annual basin precipitation (inches)
1	Small and wet	<150	≥50
1	Small and wet	≥150 and <300	≥70
2	Large and dry	≥150 and < 300	<70
2	Large and dry	≥300	<120
3	Small and dry	<150	<50

Identification of Streamflow Special Conditions from NWIS Codes and Basin Characteristics

The USGS peak-flow qualification code, peak_cd (table 2), denotes special conditions including selected data-collection conditions, peak-flow-generating mechanisms, basin conditions, and other conditions. The codes are published with the peak-flow data and defined in the NWIS help system (U.S. Geological Survey, 2024b). The selected codes inventoried included those commonly used as site or period-of-record selection criteria in hydrologic analysis.

Peak-flow codes 3 and 9, related to less common flood-generating mechanisms, were inventoried first to provide the primary basis for each code and to provide more detail about the type of flood-generating mechanism (Curran, 2023). A summary of selected special basin and streamflow conditions for all streamgages, including generalized information from the detailed code 3 and 9 inventory, was then prepared (Curran, 2024c). In this summary, the conditions are identified by water year or noted as applicable to all years. Water years for daily mean flows affected by regulation or outburst floods were also inventoried because daily mean flow data are often used to support analysis of peak flows. The summarized basin and streamflow special conditions were used in this report to guide the censoring of entire streamgages or selected water years of peak flows. Details of which conditions were censored varied by analysis and are specified in each respective section of the report.

Inventory of Extreme Floods

Definitions of extreme floods vary with the intended purpose of the results. A goal of this study was to identify a sample of floods as a dataset for assessing the flood-generating mechanisms of very large floods in the region. Additional goals were to provide considerations of the range of possible floods to inform flood-frequency analysis of short-record streamgages, to inform flood forecasting, and to inform assessments of how the prevalence of flood-generating mechanisms could vary in the future with climate change. Although more than one flood can occur in a given water year, the gaging record of peak flows provides a readily available sample of large floods and is unlikely to omit many gaged floods that could be considered extreme.

Extreme floods, given the study goals and dataset, were defined as peak flows that (1) were very large relative to the rest of peak flows in the record or (2) were very large relative to other peaks in Alaska given the basin drainage area (and basin mean annual precipitation, if available). Peak flows could meet either of these criteria for magnitude to be considered extreme, and streamgages could have more than one extreme flood. Flood-frequency statistics (Curran and others, 2016) were used as an extreme flood threshold for the first group. Flood-frequency statistics were available for 49 percent of the streamgages used in this study, which collectively included 76 percent of the peak flows used. The second group consisted of streamgages where flood-frequency estimates had not yet been computed, or where criteria required for flood-frequency analysis such as record length or randomness could not be met. For this group, the 95th percentile of an empirical metric that scaled as a non-linear function of drainage area (Creager and others, 1945) was used to provide a measure of the departure of the flood magnitude from more common floods.

Prior to identification of extreme floods, streamgages and individual peaks were censored to remove peak flows affected by regulation or basin special conditions shown in table 3 using summaries of basin and streamflow special conditions in Curran (2024c). Peak flows having a peak-flow qualification code of 4 for “Discharge less than indicated value which is Minimum Recordable Discharge at this site” (U.S. Geological Survey, 2024b; table 2) were also removed. Although outburst and non-outburst floods were considered separately for some analyses in this report, they were subject to the same criteria for identification of extreme floods. Because flood-frequency statistics in Curran and others (2016) were developed from input data that excluded outbursts for streamgages having enough non-outburst data, this meant that outbursts were evaluated relative to non-outburst flood-frequency statistics, where available.

Floods Exceeding Empirical Relative Magnitude Metrics

For streamgages where flood-frequency estimates were not available, thresholds incorporating specific peak flow (the peak-flow value divided by the drainage area of the basin) provided an alternative measure of the departure of the flood magnitude from more common floods. Specific peak flow for the largest peak flow per streamgage varied by about two orders of magnitude across Alaska as a function of drainage area and precipitation (fig. 3), such that specific peak flow alone was not a sufficient metric for identifying extreme peaks. A method that quantifies the variation of specific peak flow with drainage area using empirical curves (Creager and others, 1945) was adapted to adjust for the wide range of precipitation in the Alaska region and applied to peaks as a group and within drainage area-precipitation clusters to determine extreme flood thresholds.

Table 4.    

Threshold values of Creager’s coefficient C that were used to identify extreme peak flows in Alaska through water year 2022.

[Cluster group membership for each streamgage is provided in Curran (2025). Creager’s coefficient C calculated using methods described in Creager and others (1945). Minimum Creager’s C value for extreme peak flows does not apply to streamgages in table 5. Source of Creager’s C threshold excludes streamgages in table 5]

Table 4.    Threshold values of Creager’s coefficient C that were used to identify extreme peak flows in Alaska through water year 2022.

Group of streamgages	Minimum Creager’s C value for extreme peak flows	Source of Creager’s C threshold
Cluster 1 (small, wet basins)	18.5	95th percentile of maximum gaged peak flows per streamgage within cluster
Cluster 2 (large, dry basins)	19.2	95th percentile of maximum gaged peak flows per streamgage within cluster
Cluster 3 (small, dry basins)	5.22	95th percentile of maximum gaged peak flows per streamgage within cluster
Streamgages not clustered	17.0	95th percentile of maximum gaged peak flows per streamgage

Creager and others (1945) developed a non-linear equation to fit curves of roughly equal relative magnitude to empirical observations of specific peak flow plotted against drainage area on a logarithmic scale. A coefficient, C, scaled the curves by relative magnitude. This resulted in a series of roughly concentric curves that showed the relation of specific peak flow to drainage area for large floods. In the original application, using a global sample of streams that were mostly from the United States, notably large floods were considered those between the Creager curves with values of Creager’s coefficient C equal to 30 and 100. The curves have been used for describing notably large events in Canada (Neill, 1986; refer to fig. 1 of Neill [1986], which shows a reproduction of the original Creager figure and equation) and adjusted to support estimating future design floods in Alberta (Maria and others, 2024).

Creager curves were computed using the Creager equation, with various values for C:

$q=46C{A}^{0.894A^{-0.048}-1}$

,

(1)

where

q

is the specific peak flow, in cubic feet per second per square mile;

C

is Creager’s coefficient, in variable units; and

A

is the drainage area, in square miles.

Equation 1 was rearranged to solve for Creager’s C:

$C=\frac{q}{46A^{0.894A^{-0.048}-1}}$

,

(2)

and applied to the peak-flow values to find Creager’s C for each peak flow. To identify peak flows as extreme floods using the computed Creager’s C values, threshold Creager’s C values for each cluster group and a separate overall threshold were developed in several steps. First, draft thresholds were developed from a dataset of the maximum peak flow per streamgage, which leveraged use of any large floods at short-record streamgages and avoided overweighting long-record streamgages that might have multiple large floods. The 95th percentile of the Creager’s C values for the dataset of maximum peak flow per streamgage was selected as a threshold because it focused on very large floods and produced a reasonable sample size of extreme floods. Using these draft thresholds, the Creager’s C values for all peak flows were then evaluated to find streamgages having anomalous fits to the Creager curves. Following removal of anomalous streamgages from the dataset of maximum peak flow per streamgage, the Creager’s C thresholds were re-computed.

The Creager curves provided a reasonable approximation of relative magnitude for Alaska floods with several exceptions, such as streamgages on the Yukon River, where most or all peak flows exceeded a Creager’s C value of 30. A total of 17 streamgages (table 5) had more than 2 peak flows in the top 5 percent of Creager’s C values overall (Creager’s C greater than 21.1) or in their cluster group (Creager’s C greater than 19.2, 36.7, or 5.76 for cluster groups 1, 2, and 3, respectively) and were considered anomalous Creager’s C fits. Most of these streamgages had very large drainage areas, suggesting that the curves might not fit well for drainage areas larger than 10,000 mi² in Alaska, results supported by analysis for large drainage areas in Alberta (Alberta Transportation, 2007). Anomalous fits for other basin sizes could relate to variations in the coefficient for different settings other than the drainage area-precipitation clusters used here (Maria and others, 2024).

Table 5.    

Streamgages in Alaska having anomalously high Creager’s coefficient C values, determined as more than two peak flows per streamgage having Creager’s coefficient C above thresholds for uncensored data through water year 2022.

[Station numbers and names from U.S. Geological Survey (2024a). USGS, U.S. Geological Survey; R, River; nr, near; AK, Alaska; C, Creek]

Table 5.    Streamgages in Alaska having anomalously high Creager’s coefficient C values, determined as more than two peak flows per streamgage having Creager’s coefficient C above thresholds for uncensored data through water year 2022.

USGS station number	USGS station name	Drainage area (square miles)
15024800	Stikine R nr Wrangell, AK	19,630
15081497	Staney C nr Klawock, AK	48.9
15129120	Alsek R at Dry Bay nr Yakutat, AK	11,000
15212000	Copper R nr Chitina, Alaska	20,770
15214000	Copper R at Million Dollar Bridge nr Cordova, AK	24,030
15215990	Nicolet C nr Cordova, AK	0.7
15356000	Yukon R at Eagle, AK	111,600
15389000	Porcupine R nr Fort Yukon, AK	29,410
15453500	Yukon R nr Stevens Village, AK	194,000
15468000	Yukon R at Rampart, AK	196,900
15478040	Phelan C nr Paxson, AK	12.1
15478093	Suzy Q C nr Pump Station 10, AK	1.2
15564800	Yukon R at Ruby, AK	256,600
15564900	Koyukuk R at Hughes, AK	17,990
15565200	Yukon R nr Kaltag, AK	292,500
15565447	Yukon R at Pilot Station, AK	318,300
15875000	Colville R at Umiat, AK	13,860

The streamgages with an anomalous fit to Creager’s C curves (table 5) were removed from consideration for developing Creager’s C thresholds, although they were still evaluated for extreme floods using flood-frequency statistics, if available. After this censoring, the 95th percentile of Creager’s C values was re-computed overall and within each cluster group from the rest of the dataset of maximum peak flow per streamgage. These revised 95th-percentile values are shown in table 4 and were used as an extreme flood threshold for all peak flows at streamgages with no flood-frequency statistics, except for the anomalous streamgages shown in table 5.

Inventories of Floods from Selected Flood-Generating Mechanisms

Groups of peak flows from selected flood-generating mechanisms were identified from qualification codes or inferred from associations with ARs and peak-flow subpopulations for seasonal flow regimes. The resulting information, along with date-specific data sources including historical flood reports and the National Oceanic and Atmospheric Administration Storm Events Database (National Oceanic and Atmospheric Administration, 2024), were used to help assess flood-generating mechanisms of extreme peaks.

Association of Floods with Atmospheric Rivers

To determine AR conditions on or near the dates of peak flows, a dataset of AR presence or absence on an hourly timescale for calendar years 1980–2019 was prepared using an algorithm applied to climate reanalysis data following the methods and data sources of Nash and others (2024). The Tracking Atmospheric Rivers Globally as Elongated Targets (tARget) algorithm (version 3) identifies ARs from their characteristic long, narrow shape, intense flow of water vapor, and poleward direction of movement (Guan and Waliser, 2019). The tARget algorithm was applied to the global, 6‐hour, 1.5-degree horizontal resolution European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA) climate reanalysis dataset “ERA-Interim” (Copernicus Climate Change Service, 2023) to identify ARs, and a bounding box and additional grid points defining land areas were used to determine if the ARs reached land (table 6). If an AR was identified at any land grid point for any of the 6-hour timesteps for the day (coordinated universal time [UTC] 00, 06, 12, and 18), an AR was defined as present for the 6-hour period starting with the timestep. The hourly dataset was converted to Alaska standard time (UTC–9) to match the definition of a day for Alaska peak flows, then reduced to a daily timestep by defining an AR as present for the day if an AR was present for any hourly timestep in the day. Two land areas were considered (fig. 4): the Alaska region truncated at the end of the Alaska Peninsula and a subregion consisting of the Southeast Alaska area used in Nash and others (2024). The bounding box for Southeast Alaska from Nash and others (2024) includes a small area north of Yakutat and conterminous areas in Canada (fig. 4). This bounding box differs from the definition of Southeast Alaska used in this report but is used synonymously in the AR analysis of this study because the differences are small relative to the size of the Alaska region, and the area in the bounding box of Nash and others (2024) does not drain to any streamgages outside Southeast Alaska as used in this report. The resulting binary datasets identified days when at least 1 AR was present (AR days) and days when no ARs were present (non-AR days) within the respective areas for the 1980–2019 period and are available in the Curran (2025) data release.

Table 6.    

Latitude and longitude of bounding box and additional grid points used to define land areas for detection of landfalling atmospheric rivers in the Alaska region and Southeast Alaska through water year 2022.

[Southeast Alaska bounding box and additional grid points are defined in Nash and others (2024). Latitude and longitude are referenced to North American Datum of 1983]

Table 6.    Latitude and longitude of bounding box and additional grid points used to define land areas for detection of landfalling atmospheric rivers in the Alaska region and Southeast Alaska through water year 2022.

Latitude north (decimal degrees)	Longitude west (decimal degrees)
Alaska region bounding box
50	168
75	168
50	129
75	129
Alaska region additional grid points
58.5	139.5
58.5	138
57	136.5
57	135
55.5	135
55.5	133.5
54	166.5
54	165
55.5	163.5
55.5	162
55.5	160.5
55.5	159
57	159
57	154.5
57	153
58.5	153
Southeast Alaska bounding box
54	141.5
61.5	141.5
54	130.0
61.5	130.0
Southeast Alaska additional grid points
58.5	139.5
58.5	138
57	136.5
57	135
55.5	135
55.5	133.5

Because ARs can cover large areas, the ARs detected within Southeast Alaska are not necessarily exclusive to the subregion. However, ARs detected in the Alaska region but not in Southeast Alaska can be considered exclusive to areas outside Southeast Alaska. Thus, the nested analysis areas allowed for comparisons of ARs occurring in the Alaska region outside Southeast Alaska to ARs occurring in Southeast Alaska alone or Southeast Alaska plus areas in the Alaska region outside Southeast Alaska.

The definition of an AR day as a day when at least one AR was present in the defined land area assumes that any precipitation falling in that area on that day is related to the AR and associated processes. Nash and others (2024) noted that exceptions to this assumption for Southeast Alaska could include convective post-frontal precipitation. This exception holds for the Alaska region as well, but additional exceptions must be assumed for the larger area of the Alaska region, where other processes unassociated with the AR, such as extratropical cyclones and associated frontal precipitation, could affect streamflow in areas distant from the AR footprint. The identification method also assumes flood-generating mechanisms other than intense precipitation, including smaller amounts of precipitation and warming-related processes such as snowmelt or precipitation falling as rain instead of snow, that occur on an AR day are related to the AR and associated processes. Although AR-related processes can extend beyond the footprint of the AR, the association of these additional flood-generating mechanisms with degraded AR remnants or with areas adjacent to ARs is less well defined and can generate additional errors.

The AR presence or absence data for the Alaska region were joined with the dates of peak flows to identify peak flows occurring on AR days, or AR peak flows. AR days on which at least one peak flow occurred (peak-flow days) were identified as AR-peak-flow days and used for most analyses to minimize overweighting densely gaged areas. The simple association of peak flows with AR days misses peak flows that were associated with an AR that ended on the previous day or previous several days, which could result in an undercount. Delayed response to ARs could occur for complex reasons, but one simple explanation is delayed runoff to the stream for large basins. For example, Sharma and Déry (2020), after constraining AR data to ARs delivering intense precipitation within the basin, considered the AR day and previous 6 days as potentially associated with a peak flow for Southeast Alaska and British Columbia. For the present study, the risk of an overcount related to the collectively high frequency of ARs in the large land areas of the study area contributed to the decision to limit the association of peak flows and ARs to the AR days. AR peaks consisted of unregulated peaks in basins that have none of the special conditions described in table 3 that occurred during the 1980–2019 period (using the peak-flow date, not water years). Stream regionality was assigned as Southeast Alaska for streamgage locations along the southeastern coast of Alaska from Yakutat south, and outside Southeast Alaska for locations north of Yakutat.

Regulated Floods and Selected Basin Conditions

Streamflow affected by regulation or diversion is one of the special conditions most frequently excluded from regional analysis. Peak-flow qualification codes identified regulation or diversion affecting 607 peak flows, or about 5 percent of the peak flows, at a total of 39 streamgages (Curran, 2024c). Basin special conditions, including urbanization, indeterminate drainage areas, and drainage areas less than the minimum used for regional analysis, affected another 22 streamgages.

Glacial Lake Outburst Floods

The USGS peak-flow record for Alaska contained 150 GLOFs at 15 streamgages (table 7) distributed from the south side of the Alaska Range to Southeast Alaska (fig. 5). Although all the ice-dammed lakes associated with the GLOFs repeatedly released outburst floods, four of the streamgage records included only one or two identified GLOFs (table 7) and were omitted from analysis of patterns in the rest of this section. The gaged records for the 11 streamgages with more than 2 GLOFs are shown in fig. 6.

GLOF magnitude is dependent on lake volume and how quickly water can drain through pathways formed in the impounding glacier during the flood (O’Connor and others, 2013). Contributing factors are time-varying and site-specific, including glacier thickness, lake water levels, and rates of thermal erosion of subglacial drainage conduits. In addition, lakes can generate multiple GLOFs in a year, generate a GLOF every few years, or stop generating GLOFs completely when the impounding glacier retreats or thins below the minimum lake level. This makes GLOF magnitude and timing difficult to predict or generalize, although some GLOF peak flows have shown patterns in magnitude and timing that have varied over time.

GLOF peak flows varied in magnitude, with some being similar to floods from other mechanisms, and others being extreme floods. As a statistical group, however, the subpopulation of GLOF peak flows was consistently larger than the subpopulation of non-GLOF peak flows at the streamgage. At the eight study streamgages in table 7 with more than two peaks in both GLOF and non-GLOF peak-flow subpopulations, the median magnitude of GLOF peaks exceeded the median magnitude of non-GLOF peaks (fig. 7). Statistical tests comparing the subpopulations (Welch’s two-sample t-tests using t.test in the stats R package [R Core Team, 2024]) showed a strongly statistically significant (p ≤ 0.005) difference in the mean of the logarithms of the subpopulations for most streamgages in table 7 but no statistically significant difference in the mean of the logarithms of the subpopulations using a level of significance (α) of 0.05 for the streamgages on the Taku and Copper Rivers.

Differences in magnitude between subpopulations for the streamgages shown in fig. 7 were notable for streamgages subject to extreme GLOF peaks (table 7). Extreme GLOFs occurred repeatedly at the Salmon River streamgage, especially during the first period of discontinuous data collection, WYs 1964–73, and the Knik River streamgage, where GLOFs last occurred in WY 1966. A downstream decay in the strength of the difference between GLOF and non-GLOF subpopulations can be seen by comparing the response to Snow Lake GLOFs at the Snow River streamgage (USGS station 15243900; about 3 miles upstream from 22-mile-long Kenai Lake on the Kenai Peninsula) to the response at the Kenai River streamgage at the downstream end of Kenai Lake (USGS station 15258000), where the GLOF peaks diminished in proportion to non-GLOF peaks but remained larger than non-GLOF peaks, on average (figs. 6, 7). For the Taku River streamgage, Neal (2007) extracted the largest annual non-GLOF peak from the gaging record through WY 2004 and showed a considerable interannual range in the relative magnitudes of GLOF and non-GLOF peaks (fig. 10 of Neal [2007]).

Trends in GLOF magnitude are difficult to characterize and not consistent among streamgages. Several apparent patterns exist in the gaged records for the streamgages shown in fig. 6, including (1) a clear shift at the Salmon River streamgage from larger GLOF peaks during the initial period of gaging (ending in WY 1973) to smaller GLOF peaks during the second period (starting in WY 2010); (2) a weak positive trend at the Snow River streamgage (USGS station 15243900) found by Beebee (2023) using additional estimated magnitudes and analysis; and (3) a positive trend at the Knik River streamgage from the start of gaging in WY 1948 until the early 1960s, followed by a declining trend until the cessation of GLOFs after the WY 1966 GLOF. In many cases, however, systematic gaging began after the onset of GLOFs, such that patterns in gaged GLOFs might not be an accurate characterization of long-term annual maximum GLOF patterns. For example, the annual GLOFs from Lake George that affected the Knik River streamgage extend back to 1914 (Bradley and others, 1972), and Tulsequah Lake GLOFs, which affect flows at the Taku River streamgage, have occurred since the early 1900s (Neal, 2007).

The timing of the day of year of occurrence for recurring GLOFs ranged from June to December across the 11 streamgages and spanned several months during the period of record for most streamgages. At some streamgages, GLOF seasonality visibly trended toward occurring earlier in the year during the period of GLOF occurrence (fig. 8). Specifically, Salmon River GLOF peaks occurred later in the year during the initial period of gaging and earlier in the second period; Taku River GLOF peaks showed a trend toward earlier occurrence throughout the period of gaging, and Knik River GLOF peaks showed a strong trend toward earlier occurrence from the start of gaging through the last GLOF. Although this report only considers the largest GLOF in a year, and only GLOFs that are also the annual peak, Rick and others (2023) performed a more rigorous analysis on all GLOFs, regardless of size, from selected lakes and noted a trend toward earlier seasonality for Tulsequah Lake GLOFs, which affect streamflow at the Taku River streamgage. Because the gaging record for many GLOF streamgages starts well after the onset of GLOFs, the data shown in this report provide a snapshot of a variable period of GLOF occurrence.

Other Outburst Floods

Outburst floods other than GLOFs in the peak-flow record included 10 peaks from a wide range of flood-generating mechanisms (table 8), all occurring in Southeast Alaska or south-central Alaska (fig. 9). The outburst floods other than GLOFs included beaver dam breakups at two different streamgages (USGS stations 15238986 and 15261000) in different years and breakup of temporary landslide debris dams at three streamgages (USGS stations 15238000, 15238010, and 15238600) that occurred during an October 1986 rainstorm in Seward (Jones and Zenone, 1988). In addition, a large regional rainstorm in south-central Alaska in August 1971 triggered outburst peaks through three separate mechanisms: (1) for the outburst flood at USGS station 15294500, rapid change of the outlet of Chakachamna Lake, the source lake for Chakachatna River, which was dammed by glacier ice and sediment at a glacier terminus (Lamke, 1972); (2) for the outburst flood at USGS station 15284000, failure of a sediment dam (reported as a moraine by McGee, 1974, but could be interpreted from elevation data as landslide debris) that had impounded a lake in the basin of Granite Creek, a tributary of the Matanuska River (Lamke, 1972); and (3) natural diversion of a stream from a flow path along Sheep Creek to a flow path along Goose Creek, creating an outburst flood at USGS station 15292900 that is considered an indeterminate drainage area case for the analyses in this report. In 2002, a mass failure involving a moraine and saturated sediments from a tributary valley entered a pro-glacial lake and generated an outburst flood at USGS station 15056210 in Southeast Alaska (Capps, 2004). Finally, detonation-induced ground subsidence generated a flood at USGS station 15297690 coded as an outburst and which is considered a regulated peak for the analyses in this report.

The relative magnitude of the outburst floods other than GLOFs ranged considerably, from less than the maximum peak on record for the streamgage to peaks considered extreme for this study. Of the five streamgages in table 8 with weighted flood-frequency estimates, three streamgages had floods that exceeded the 0.2-percent AEP (500-year return interval) flood value. One of these was the 1971 flood recorded at the Matanuska River streamgage (USGS station 15284000), where flow had already crested at a peak of record from the 1971 rainstorm (Lamke, 1972) when the outburst floodwaters from tributary Granite Creek arrived and further swelled the river. The relative magnitude of the gaged peak flow, although notable for the large Matanuska River Basin, pales in comparison to the relative magnitude for the much smaller tributary basin (Lamke, 1972), providing insight into the potential magnitudes of the full subpopulation of outburst floods other than GLOFs in Alaska.

The conditions associated with these outburst floods other than GLOFs (table 8) were isolated events and have been considered unlikely to reoccur multiple times, or at all, at these streamgages for the purposes of flood-frequency analysis (Curran and others, 2016). However, beaver dams can be rebuilt, and landslides, although infrequent, can reoccur, highlighting the need for user interpretation for this type of special condition. Although the Taiya and Chakachatna River outburst floods involved lakes at or near the terminus of glaciers, the flood-generating mechanisms were associated with mass failure (Taiya River) or erosion of sediment (Chakachatna River), conditions that would be difficult to reconstruct.

Snowmelt Floods

The peak-flow qualification code for snowmelt, intended to differentiate snowmelt peak flows from rainfall peak flows, created a dataset of peaks with spring snowmelt as their dominant flood-generating mechanism. From WY 2006, when snowmelt codes began to be more systematically applied to many streamgages, to WY 2022, a total of 530 peaks that were not affected by regulation or diversion and were not in basins with a special condition affecting flow (table 3) received the code. This dataset is opportunistic in that the peak-flow codes were not applied to every streamgage where snowmelt peak flows existed, but for the streamgages where codes were applied, this dataset provides an opportunity to examine subpopulation characteristics.

The 141 streamgages with at least one snowmelt-coded peak flow were distributed across a wide range of the State (fig. 10), but 96 percent were located in areas outside Southeast Alaska. The dominant months for snowmelt-coded peak flows were May (70 percent of the peak flows) and June (23 percent). Six percent of the snowmelt-coded peaks occurred in April, and the 1 percent that occurred in winter (December–February) were likely rain-on-snow peak flows.

For streamgages where seasonal day-of-year ranges for flood-generating mechanisms were available for seasonal flow regimes from Curran and Biles (2021), the primary flood-generating mechanism inferred from the peak day of year for the snowmelt-coded peak flows was consistently snowmelt, providing a one-directional check on the accuracy of the inferred primary flood-generating mechanism from the day-of-year data. The snowmelt-coded peak flows occurred at streamgages in every seasonal flow regime except regime 1A, the most rainfall-dominated regime, supporting observations of the existence of snowmelt-peak subpopulations in most flow regimes in Alaska.

Snowmelt peaks were assumed to be smaller, as a group, than peaks from other flood-generating mechanisms for all but the most snowmelt-dominated seasonal flow regimes (Curran and Biles, 2021), such as regime 2D, which includes USGS station 15896000, (Kuparuk River near Deadhorse, Alaska), where 16 of 17 peaks during WYs 2006–22 are snowmelt-coded. To quantify the difference in magnitude between snowmelt-coded peaks and other peaks, streamgages with (1) at least 3 snowmelt-coded peaks and 3 non-snowmelt-coded peaks, (2) a total of at least 10 years of unregulated, non-outburst peak flows, and (3) none of the basin special conditions affecting flow listed in table 3 were selected. The median magnitude of snowmelt-coded peaks was smaller than the median magnitude of non-snowmelt-coded peaks for 70 percent of the 50 selected streamgages.

Floods Associated with Atmospheric Rivers

ARs occur frequently in the Alaska region, with AR days amounting to 67 percent of the days in the 1980–2019 period. The frequency of ARs for the Alaska region and their ability to persist for multiple days created such frequent AR conditions that 92 percent of the days were AR days or within 3 days of an AR day. The count of days on which an AR occurred anywhere in the region is inflated relative to the count for any one location but can be used as a metric for assessing ARs that can cover multiple parts of the State. Although true subregion comparisons were not possible with only one subregion, ARs were frequent inside and outside Southeast Alaska (fig. 11A). The percentage of days in the period when an AR occurred in the Alaska region can be subdivided into the percentages when ARs were present or absent in Southeast Alaska. ARs were present in Southeast Alaska on 35 percent of the days in the period, occurring 8–15 days a month (first presented in Nash and others, 2024; identical results from the present study are shown in fig. 11A). ARs were present in the Alaska region but absent from Southeast Alaska on 32 percent of the days in the period (fig. 11A). This shows the relatively high frequency of ARs in Southeast Alaska (for its land area) and shows a substantial presence of ARs in the Alaska region outside Southeast Alaska.

ARs occurred in every month of the year, most frequently during July–September for the Alaska region (fig. 11A). ARs had a slightly summer-dominated seasonality (July–August) in areas outside Southeast Alaska and a primary fall (August–October) and secondary winter (December–January) seasonality in Southeast Alaska. This pattern is consistent with the general southward-progressing timing of AR landfalls for the West Coast of North America (Gershunov and others, 2017). Relative to ARs, peak flows are more strongly seasonal, occurring primarily from spring to fall (May–October) for the Alaska region, and summer to winter (July–January) for Southeast Alaska (fig. 11B). Seasonal differences in amounts of moisture transport in ARs (Gershunov and others, 2017) and precipitation falling as snow contribute to the differences in seasonality of the relation of AR processes and peak flows.

Despite the high frequency of AR days, only 21 percent of the AR days in the full period and 35 percent of AR days from May to October coincided with peak-flow days. However, 78 percent of the peak-flow days in the full period and 78 percent from May to October occurred on AR days (fig. 11B), an association with ARs that is statistically greater than the percentage of days that are AR days (using a 1-sample proportions test with the “greater” alternative and p < 0.5 in the prop.test function in the stats R package [R Core Team, 2024]). Although streamgages with peak flows associated with ARs occurred across most of the State (fig. 12), peak-flow days for Southeast Alaska streamgages were more strongly associated with ARs, on average during the year, than those for streamgages outside Southeast Alaska. For streamgages in Southeast Alaska, 86 percent of peak-flow days were on AR days, whereas for streamgages outside Southeast Alaska, 76 percent of peak-flow days were on AR days. The observation that a small percentage of ARs are responsible for extremes in streamflow echoes findings of other studies relating ARs to extreme precipitation (Nash and others, 2024) or streamflow (Sharma and Déry, 2020) in Southeast Alaska. Using the same Southeast Alaska AR presence or absence dataset as this study, Nash and others (2024) found that 6 AR days per year accounted for as much as 91 percent of the precipitation extremes in 6 Southeast Alaska communities, despite the presence of ARs on 122 days per year on average during the period. They identified the strength of the water vapor transport and the direction of the water vapor transport with respect to complex local coastal topography as important factors for generating extreme precipitation events. For the present study, 458 streamgages, or 69 percent of the 663 streamgages used for analysis, had at least 1 peak associated with ARs (fig. 12). AR peak flows occurred in all 9 seasonal flow regimes and were most common in regime 1A, the most rainfall-dominated regime, where 96 percent of the peak flows occurred on AR days, and least common in regime 2D, the most snowmelt-dominated regime, where only 61 percent of the peak flows occurred on AR days.

The association of peak-flow days with ARs varied seasonally and to some degree with geographic area (fig. 11B; table 9). For the May–October period, the association with ARs for Alaska was greatest during the late summer and early fall, reaching a high of 89 and 88 percent of peak-flow days occurring on AR days in August and September, respectively (using the two areas shown in table 9 combined). The association was slightly stronger (91 percent) and peaked later (October) in Southeast Alaska. The association was notably lower for Alaska in spring, when snowmelt peaks are common. For WYs 2006–19, when snowmelt peak-qualification codes were consistently applied and AR data were available, ARs occurred from May to June on 67 percent of the days, and 63 percent of the 244 snowmelt-coded peak flow days coincided with an AR day. The snowmelt-coded peak association with ARs was not statistically significantly less than the general association with ARs for that time of year using a one-sample proportions test with the “less” alternative and p < 0.5 in the prop.test function in the stats R package (R Core Team, 2024). The overlap of peak-flow days and AR days in December and January, the winter months when peak flows were common enough to assess, was high for areas inside and outside Southeast Alaska, ranging from 86 to 93 percent (table 9).

Comparing the magnitude of AR peak flows to non-AR peak flows within streamgages shows some modest tendencies for AR peaks to be larger, especially for larger flow percentiles. To quantify the difference in magnitude, streamgages with (1) at least 3 AR peaks and 3 non-AR peaks, (2) a total of at least 10 years of unregulated, non-outburst peak flows, and (3) none of the basin special conditions affecting flow listed in table 3 were selected. The median magnitude of AR peaks was greater than the median magnitude of non-AR peaks for 60 percent of the 128 selected streamgages. The 75th percentile of AR peaks (corresponding to the upper quartile of a boxplot) exceeded the 75th percentile of non-AR peaks for 70 percent of the streamgages.

Some of the associations of peak flows with ARs are likely the result of ARs with rain-on-snow events, where processes associated with warmer ARs produce snow water equivalent loss that increases streamflow-to-precipitation ratios. For example, in winter months in the Sierra Nevada, Guan and others (2016) noted that AR conditions, more common in winter in that region, were associated with 50 percent of the rain-on-snow events and that ARs with rain-on-snow were warmer than ARs without rain-on-snow. The presence of ARs in the Alaska region during many snowmelt-coded peaks at many of the northernmost streamgages (streamgages near or north of the Brooks Range), where ARs footprints might not overlap the drainage basin of the streamgage, could be a non-causal association or could indicate that regional warming in association with an AR plays a contributing role for spring snowmelt peak flows.

Extreme Floods

The gaging record contained 149 peak flows identified as extreme floods, which amounted to 1.5 percent of the unregulated peaks in basins with no special conditions. Most of the extreme floods were identified using the 1-percent AEP flood magnitude as a minimum, although 19 percent did not have flood-frequency statistics and were identified using the Creager’s C values from table 4 as minimum thresholds. Figure 13 shows non-outburst extreme flood magnitudes and identification methods and figure 14 shows extreme outburst flood magnitudes by streamgage. All but one of the extreme outburst floods identified using flood-frequency statistics also exceeded the Creager’s C threshold for the streamgage. Extreme floods occurred in locations that were widely distributed around the State, although extreme outburst floods (fig. 14) occurred only in Southeast Alaska and south-central Alaska (fig. 15). A total of 12 streamgages recorded more than 1 non-outburst extreme flood, and 3 streamgages recorded multiple extreme outburst floods. Details of flood-frequency statistics exceeded, Creager’s C, flood-generating mechanisms, and other information are provided for the extreme peaks in a data release (Curran, 2025).

Considerations for Future Changes to Floods

Future flood-generating mechanisms are likely to include existing mechanisms shifted in their timing, intensity, or geographic extent. For example, modeling of changes associated with loss of sea ice along the western and northern coasts of Alaska and other warming-related effects predicts that convective storms will triple in number and increase in intensity in the northern parts of Alaska, including the North Slope and western Alaska (Poujol and others, 2020). A study of ice-dammed lakes in Alaska (Rick and others, 2023) found no change in GLOF frequency but also found decreases in lake size that could signal reduced GLOF magnitudes. The relatively recent (2011) onset of Suicide Basin GLOFs on the Mendenhall River demonstrates that new GLOFs in basins with no history of outbursts can be expected as glaciers thin and retreat. However, GLOF magnitudes continue to have the potential for a high degree of variability. A sudden and large increase in Suicide Basin GLOF magnitudes on the Mendenhall River occurred just after the period of record used for this study, impacting hundreds of homes near Juneau, Alaska. Mendenhall River GLOFs in WY 2023 and 2024 (U.S. Geological Survey, 2024a) were more than twice the magnitude of any GLOFs on the Mendenhall River through WY 2022 (fig. 16) and would have met the definition of extreme for this study. Accelerating glacier loss (Hugonnet and others, 2021) is expected to initially produce additional high-elevation melt in some basins and later less melt as glacier size diminishes (Huss and Hock, 2018). In other basins where substantial glacier loss has already occurred, less melt can be expected, and in either case, changes in melt that differ by month can be expected (Huss and Hock, 2018). As scaling metrics for AR intensity, duration, and effect are developed for Alaska (for example, Nash and others, 2024), a better understanding of future AR effects will emerge.

The translation of predictions in changes in flood-generating mechanisms to changes in flooding is facilitated by understanding the historical range and dominance of flood-generating mechanisms. However, the complex relations between rain-based and melt-based runoff, and the host of secondary factors associated with extreme flooding, create complicated feedbacks that can be explored by modeling. For example, a change in the rain-to-snow fraction resulting in precipitation falling as rain that would have fallen as snow can decrease or increase flooding, depending on the season. Reduced snowpack in winter can lead to a smaller snowmelt peak flow in spring, whereas late fall flooding associated with a warm AR can be enhanced by additional rain at elevations where precipitation would normally fall as snow that late in the season.

Identifying Special Conditions and Other Information

Identification of special conditions for peak flows or daily mean flows that form common site and period-of-record selection criteria—including regulation, selected basin conditions, and outburst flooding—can be obtained from NWIS (U.S. Geological Survey, 2024a) and selected data releases associated with this report (Curran, 2023; Curran, 2024c). First, qualification codes available from NWIS for peak flows (peak_cd) or gage height (gage_ht_cd and ag_gage_ht_cd) flag data requiring review and can help a user understand whether data censoring or special interpretation is necessary. Code definitions are provided in NWIS (U.S. Geological Survey, 2024b; refer to table 2 for peak-flow qualification code definitions in use at the time of this study [2024]). Second, for peak flows assigned a peak-qualification code of 3 or 9, code disambiguation and additional details in the Curran (2023) data release can be used to identify snowmelt, GLOF, and other outburst peak flows. Finally, the following can be obtained by streamgage from the Curran (2024c) data release: (1) types of flow data available, (2) data-collection special conditions, (3) basin special conditions, and (4) a water-year summary of peak flows and daily mean flows interpreted as affected by regulation or diversion or by outburst floods.

Peak flows from subpopulations associated with selected flood-generating mechanisms can be identified from peak-flow qualifications codes in NWIS, data releases, and other documents, such as the flood reports cited in table 10. Peak-flow qualification codes coupled with the detailed peak-flow special conditions data release (Curran, 2023) identify a selected group of snowmelt peak flows and all GLOF and other outburst floods in the gaged record. Peak flows that occurred on the day an AR was present or absent in the Alaska region or in Southeast Alaska can be identified using peak-flow dates together with the binary files of AR presence or absence for 1980–2019 for the Alaska region and Southeast Alaska that are provided in the Curran (2025) data release. For more generalized identification of subpopulations of peak flows that occur during periods inferred to be dominated by snowmelt, high-elevation melt, or rainfall, the day-of-year boundaries provided in Curran and Biles (2021) can be applied to peak-flow dates.

Extreme floods identified in this report are listed along with data used to determine their relative magnitude and flood-generating mechanisms in the Curran (2025) data release. For exploration of relative magnitude for other large floods, the Creager’s C, which provides one measure of relative magnitude, can be computed using equation 2. To compare the Creager’s C of the flood to the extreme flood thresholds for various drainage-area precipitation clusters shown in table 4, the drainage area-precipitation cluster for the streamgage can be found in Curran (2025) for the streamgages used in the analysis or estimated using the bounds in table 1 for other streamgages or ungaged sites.

Other resources can contribute to understanding site or peak-flow conditions. Streamgage drainage basin boundaries (including those for selected USGS streamgages in Curran [2024a, b]) can be used with other geospatial data to obtain additional climate, land cover, or other basin characteristics. To further understand GLOF conditions at a streamgage, and to identify GLOFs other than those associated with peak flows, inventories of glacier-dammed lakes or GLOFs can provide additional information. Compilations with recent data for Alaska include the National Weather Service glacier-dammed lake web page (National Oceanic and Atmospheric Administration, 2023), an inventory of ice-dammed lake drainage events in Alaska (refer to data available with Rick and others, 2023) and the global Glacier Lake Outburst Flood Database (Lützow and Veh, 2024).

Treatment of Flows with Special Conditions

Understanding and treating streamflow special conditions can be critical to obtaining accurate hydrologic analyses. This discussion focuses on peak flows and regional flood-frequency analysis using Curran and others (2016) as a general example but is generally applicable to flood-frequency analysis for a single streamgage and to other hydrologic analyses. For flood-frequency analysis, an important assumption is that the peaks are from a homogeneous population (England and others, 2018). When special conditions are present, a determination must be made whether the conditions represent a distinct subpopulation unsuitable for analysis together with other peak flows. Treatment of special conditions can include censoring selected peak flows or entire streamgages or applying advanced analysis techniques, depending on the special condition and the intent of the analysis. Although independent and analysis-specific evaluation of the data and site-selection criteria should always be conducted, several cases are common for regional studies in Alaska.

Basin special conditions typically considered ineligible for regional analysis of peak flows include indeterminate drainage areas, urbanized basins, and drainage areas less than the minimum drainage area for analysis. The entire record is omitted unless the condition did not apply for some part of the gaged period, in which case the unaffected period of record could be used. Periods of regulated streamflow are generally omitted from the record for analysis. In isolated cases (for example, when flow is generated below a complete diversion) regulated flow can be associated with a contributing drainage area and considered for use.

GLOFs and other outburst floods can be considered distinct subpopulations of peak flows that generally do not satisfy the assumptions for regional analysis and require site-specific review. The outburst floods other than GLOFs that were identified in this report are from events considered unlikely to reoccur and are generally omitted from the record for flood-frequency analysis. GLOFs can be considered likely to recur only as long as the basin conditions conducive to outbursts persist and are generally omitted from the record for analysis for streamgages where those conditions are presently absent. Recurring GLOFs represent future flood risk that could depart substantially from typical regional flood-generating mechanisms. Without site-specific assessment, GLOFs are assumed to (1) not be random and independent because of the potentially progressive basin conditions that are generating the GLOFs, (2) as a group, generally exceed the magnitude of non-GLOF peaks, and (3) not be responsive to meteorologic conditions in a manner similar to floods from primary regional flood-generating mechanisms. In addition, during the period when GLOFs occur, peak and daily mean flows can be potentially affected by changes in the basin response to meteorologic conditions because GLOF-related processes control storage and release of precipitation and meltwater runoff, whether or not GLOFs occur every year. Regional studies have, therefore, generally omitted from analysis entire periods of record when GLOFs are occurring.

Site-specific studies of future flood risk from recurring GLOFs can determine if assumptions for conventional flood-frequency analysis can be applied to produce an at-site estimate, even if the result is not applicable for regional analysis. Alternatively, more advanced analysis could improve at-site estimates. Advanced treatments can factor in special scenarios, such as for the Taku River, where Neal (2007) considered the possibility of simultaneous release of two lakes. Studies could consider mixed-population analysis, where the likelihoods of different subpopulations can be considered and combined. In some cases, studies might consider a worst-case scenario analysis that estimates compound peaks from synchronous sources, such as releases from multiple lakes or a GLOF occurring at the peak of large rainstorm- or snowmelt-generated flood or during tidally influenced high water levels. Finally, for analysis of recurring GLOFs in consideration of infrastructure design or community safety, long-term GLOF patterns might be of less interest than recent, or projected future, patterns. Time-variable lake volumes and impoundment conditions could make analysis of the basin conditions more productive than analysis of past floods.

Peak flows from rainfall and snowmelt flood-generating mechanisms have been considered as part of the same population for many regional flood-frequency analyses in Alaska (most recently, Curran and others, 2016). However, recent improvements in the understanding of the extent and distribution of multiple seasonal peak-flow subpopulations related to different flood-generating mechanisms (Curran and Biles, 2021, and this report) and the role of various types of rainstorms in producing extreme floods demonstrated in this report indicate that some subpopulations of peak flows from selected flood-generating mechanisms could be addressed using mixed-population analysis (England and others, 2018). Bulletin 17C (England and others, 2018) notes that mixed populations can result in abnormally large skew coefficients and abnormal slope changes in flood-frequency curves and that computing separate curves for each subpopulation and then combining the results can improve the estimates. Barth and others (2019) applied mixed-population analysis to 43 long-record streamgages across the western United States with AR and non-AR subpopulations and found notably different estimates for the upper tail of the distribution (the smaller AEPs, or larger floods) in 20 percent of the streamgages when compared to the single population method.

For Alaska peak flows, this report shows that streamgages for which mixed-population analysis might be considered include those with sizable snowmelt and rainfall subpopulations or AR and non-AR subpopulations. As shown by the Barth and others (2019) study, mixed-population analysis is not universally appropriate or helpful; there must be enough peak flows in each subpopulation to produce defensible results, and the flood-frequency curves must show differences between the subpopulations. Additionally, event-based identification beyond the inferred identifications for groups of peaks in this report might be required to isolate peak-flow subpopulations for a single streamgage. Treatment of GLOF and non-GLOF peak flows using mixed-population analysis could only be undertaken in cases where a flood-frequency curve could be determined for GLOF peak flows, which do not necessarily meet the requirements for randomness and might not be a representative sample of the population of future floods.

The treatment of peak flows from subpopulations not suitable for analysis with other peak flows in the record might require adjustments in PeakFQ (Siefken and others, 2024) beyond the built-in treatments for peak flows with selected qualification codes, which are summarized in Siefken and others (2024). In particular, peak flows assigned a code 9 will not be automatically excluded, using the version of PeakFQ that is current at the time of this report (8.0.2). Because most GLOFs and other outburst floods in Alaska were coded 9 as of the time of this report, the user should carefully review all code 9 peaks and supporting information to understand the peak-flow special condition, then interpret whether the peak represents future flood risks and is appropriate to analyze together with peaks from other flood-generating mechanisms. To omit peak flows assigned a code 9 from analysis, the user can adjust settings for each peak flow within PeakFQ following the guidelines in England and others (2018). Other techniques, such as the multiple Grubbs-Beck test for potentially influential low floods (PILFs) described in Bulletin 17C (England and others, 2018), help focus analysis on the largest floods. For streamgages dominated by ice-jam flooding not related to peak flows, other data and analysis types, such as stage frequency analysis using the maximum annual stage, might provide more useful information than flood-frequency analysis.

Regional and At-Site Exceptionality for Extreme Floods

When an independent estimate of flood-frequency statistics, such as a regional regression estimate, is available, weighting the station and regression estimate can reduce the uncertainty of the estimate (England and others, 2018). When floods are extreme and the differences between the station and regression estimates are large, it can be difficult to determine if the regression estimate is a reliable estimate. Considerations could include the intended use of the estimate, the record length, and the representation of the flood-generating mechanisms for the streamgage in the regional regression dataset. For example, convective storms that generated extreme floods, such as those in 2022 (table 10; Best, 2023), are not common and, therefore, not well-represented in the regional regression equations. Although the weighted estimate might more accurately convey the regional exceptionality of the flood, the station estimate might be a more representative estimate for at-site purposes such as the design of infrastructure. Appendix 9 of Bulletin 17C (England and others, 2018) provides guidance for the use of weighted estimates, including consideration of record length and whether the station estimate was used to develop the regional regression model.