Volume 57, Issue 4 e2020WR028713
Research Article
Free Access

Changing River Network Synchrony Modulates Projected Increases in High Flows

David E. Rupp

Corresponding Author

David E. Rupp

Oregon Climate Change Research Institute, College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, OR, USA

Correspondence to:

D. E. Rupp,

[email protected]

Search for more papers by this author
Oriana S. Chegwidden

Oriana S. Chegwidden

CarbonPlan, San Francisco, CA, USA

Search for more papers by this author
Bart Nijssen

Bart Nijssen

Department of Civil and Environmental Engineering, University of Washington, Seattle, WA, USA

Search for more papers by this author
Martyn P. Clark

Martyn P. Clark

Coldwater Laboratory, University of Saskatchewan, Canmore, Alberta, Canada

Search for more papers by this author
First published: 19 March 2021
Citations: 7

Abstract

Projections of change in high-flow extremes with global warming vary widely among, and within, large midlatitude river basins. The spatial variability of these changes is attributable to multiple causes. One possible and little-studied cause of changes in high-flow extremes is a change in the synchrony of mainstem and tributary streamflow during high-flow extremes at the mainstem-tributary confluence. We examined reconstructed and simulated naturalized daily streamflow at confluences on the Columbia River in western North America, quantifying changes in synchrony in future streamflow projections and estimating the impact of these changes on high-flow extremes. In the Columbia River basin, projected flow regimes across colder tributaries initially diverge with warming as they respond to climate change at different rates, leading to a general decrease in synchrony, and lower high-flow extremes, relative to a scenario with no changes in synchrony. Where future warming is sufficiently large to cause most subbasins upstream from a confluence to transition toward a rain-dominated, warm regime, the decreasing trend in synchrony reverses itself. At one confluence with a major tributary (the Willamette River), where the mainstem and tributary flow regimes are initially very different, warming increases synchrony and, therefore, high-flow magnitudes. These results may be generalizable to the class of large rivers with large contributions to flood risk from the snow (i.e., cold) regime, but that also receive considerable discharge from tributaries that drain warmer basins.

Key Points

  • Projected changes in extreme high flows at river confluences are due partly to changes in mainstem-tributary hydrograph synchrony

  • Decreasing synchrony between snowmelt-dominated basins dampens anthropogenically forced increases in high flows

  • Under large forcing, increased synchrony between a historically snowmelt-dominated and rainfall-dominated basin increases high flows

1 Introduction

Projections of change in high-flow extremes with global warming vary widely among midlatitude rivers of North America (Bürger et al., 2011; Chegwidden et al., 2020; Das et al., 2013; Demaria et al., 2016; Huang et al., 2018; Maurer et al., 2018; Naz et al., 2018), Europe (Alfieri et al., 2015; Huang et al., 2018; Köplin et al., 2014; review in Madsen et al., 2014; Thober et al., 2018), and Asia (Gu et al., 2018; Huang et al., 2018; Shkolnik et al., 2018). Interbasin variability in projected changes has been mainly attributed to differences in precipitation changes and differences in the rain:snow partitioning across basins (Alfieri et al., 2015; Chegwidden et al., 2020; Dankers & Feyen, 2008; Thober et al., 2018).

Curiously, some large rivers show little projected change in high-flow extremes despite large relative increases in high-flow extremes in their headwaters (e.g., Maurer et al., 2018; Shrestha et al., 2017) that result from increased rainfall (Chegwidden et al., 2020). Sharma et al. (2018) proposed two explanations for why fractional increases in high-flow extremes with respect to fractional increases in precipitation (i.e., “sensitivity”) may be smaller in large basins than in small basins. One explanation is that antecedent available basin storage has a larger impact on the conversion of precipitation to streamflow in larger river basins. For high-flow extremes in small basins, compared to large basins, the amount of antecedent available storage relative to the amount of precipitation on a per unit area is likely to be smaller because the small basin is more likely to be covered by the more intense part of a storm. The consequence for smaller basins is, therefore, a larger relative increase in flow for the same relative increase in precipitation during extreme events. One can also consider snowmelt as an input, or the impact of evaporation on increasing the antecedent storage, and arrive at the same conclusion that small basins are more sensitive.

A second explanation for the lower sensitivity in larger basins proposed by Sharma et al. (2018) is that the spatial extent of storms may decrease in a warmer climate, reducing the fraction of the basin area receiving heavy rainfall (Chang et al., 2016; Wasko et al., 2016). However, Prein et al. (2017) show simulated mesoscale convective systems delivering high rain rates over substantially larger areas in response to warming. Moreover, floods in many parts of the world are frequently driven by large atmospheric rivers (e.g., Dettinger et al., 2011; Kingston et al., 2016; Konrad & Dettinger, 2017; Lavers & Villarini, 20132013; Neiman et al., 2011; Paltan et al., 2017) and there is currently no strong evidence to support that there will be meaningful reductions in the areal extent of precipitation sourced from atmospheric rivers. Atmospheric rivers may even deliver precipitation over larger areas as they penetrate farther inland in a warmer climate (e.g., Mahoney et al., 2018), though the conversion of the farther-penetrating water vapor to precipitation will depend on the air-mass saturation and factors that drive lift, such orography (Payne et al., 2020).

A third reason why larger basins may exhibit lower sensitivity lies in the temporal patterns of high and low flows (i.e., the flow regimes) among the tributaries that contribute to high-flow extremes on large rivers. Specifically, we refer here to the synchrony of high flows coming from different tributaries within a river network. The topic of synchrony in high-flow extremes has recently received increased attention, both in the context of regional (i.e., multicatchment) flood risk assessment (Berghuijs et al., 2019; Brunner et al., 2020; Keef et al., 2009; Kemter et al., 2020; Quinn et al., 2019) and in the context of superposition of flood waves at river confluences (Geertsema et al., 2018; Guse et al., 2020; Neal et al., 2013; Vorogushyn & Merz, 2013). In either context, both higher spatial correlation of the meteorological forcing event (Berghuijs et al., 2019) and higher spatial similarity of the land-surface processes driving the extreme flow (Brunner et al., 2020; Keef et al., 2009) tend to increase synchrony. Changes in the built environment that dampen or accelerate the propagation of flood waves on a river can also alter synchrony (e.g., Vorogushyn & Merz, 2013).

Changes in synchrony may be prominent in large basins where flow regimes respond to climate change at varying rates across tributaries. Even where tributaries are subjected to an equivalent increase in temperature, the rates that flow regimes change will depend on the initial climates of the basins. In warmer basins that have a substantial snowpack, the snow regimes, and consequently the flow regimes, are more sensitive to a change in temperature (e.g., Hamlet et al., 2013). Differential rates of change in flow regimes can lead to changes in the synchrony of high flows of a mainstem and its tributary at their confluence. All other things being equal, high-flow extremes at a confluence should increase (or decrease) where high flows of the mainstem and tributary become more (or less) synchronous.

Both increases and decreases in the spatial scale of high-flow synchrony have been detected across Europe over the years 1960–2010 (Berghuijs et al., 2019; Kemter et al., 2020). Kemter et al. (2020) attributed these opposing trends to disparate changes in the relevance of different flood generating processes across Europe during this period. For example, they associate the decrease in the spatial extent of synchronous high flows in Eastern Europe with a concurrent decrease in the contribution of snowmelt to high flows. This is a particularly noteworthy result given that shrinking of the snow season—in all but the coldest regions of the world—is a robust hydrological response to a warming planet (e.g., Räisänen, 2008).

Despite the importance of synchrony to flooding, its discussion in the literature in the context of climate change has been limited mainly to conceptualizing how synchrony changes may manifest due to changes in flood generating mechanisms (e.g., Brunner et al., 2020). Our work examines synchrony during high-flow extremes in a large, climatically diverse basin—the 7 × 105 km2 Columbia Basin in North America (Figure 1)—using hydrologic modeling under commonly assumed scenarios of future anthropogenic greenhouse gas forcing. Previous studies of the basin indicate little projected change, and even small decreases, in the magnitudes of high-flow extremes in the lower reaches of the Columbia River, even as headwater reaches in the Rocky Mountains, Cascade Range, and other mountainous regions show relatively large increases (Maurer et al., 2018; Queen et al., 2021; Tohver et al., 2014). Queen et al. (2021) speculated that the relatively large projected increases in headwater basins do not persist to the downstream reaches of the Columbia because downstream reaches integrate a larger diversity of hydroclimates and individual weather events. However, Queen et al. (2021) did not describe specifically how greater hydroclimatic diversity leads to a decreased sensitivity in high-flow response to anthropogenic warming, leaving the matter unresolved.

Details are in the caption following the image

Köppen-Geiger climatic regions over the hydrological model domain and the locations of the 18 mainstem + tributary confluences (black dots) in the Columbia Basin (outlined in black) analyzed in this study.

We investigate to what extent synchrony can help explain the relatively small changes in projected high-flow extremes in the downstream reaches of the Columbia River and its major tributary, the Snake River. Our work extends the current literature in two ways. First, we focus specifically on the impacts of changes in the snow regime, adding to the literature on the synchrony of flood peaks in rainfall-dominated regimes. Second, we focus on anthropogenically forced changes, illustrating how the spatial variability in shrinking snow seasons can affect century-long changes in flow synchrony over a large river basin. Our analysis uses a simple index to quantify mainstem-tributary synchrony in the historical record and in simulations driven by projections of climate change. We use this index to approximate the effect that century-long synchrony changes have in dampening, or amplifying, the magnitude of high-flow extremes at major confluences.

2 Data and Methods

2.1 Data

We analyzed both reconstructed and simulated naturalized daily streamflow. The reconstructed streamflow approximates natural streamflow by removing the effects of regulation, irrigation, and reservoir evaporation from the observed streamflow. The reconstructed naturalized streamflows account for variations in both reservoir storage time and wave celerity as functions of river discharge (Bonneville Power Administration, 2011). This “no-regulation no-irrigation” (NRNI) data spans the years 1928 and 2008 and exists for 197 locations in the Columbia Basin and adjacent smaller basins (RMJOC, 2017).

The model simulations of naturalized streamflow, along with other hydrological variables such as the snow water equivalent (SWE) of the snowpack, come from a multimodel ensemble of hydrological simulations (Chegwidden et al., 2017). Forty “historical” simulations spanning 1950–2005 and 80 “future” simulations spanning 2006–2099 are the product of four hydrological models driven with downscaled daily temperature and precipitation output from 10 global climate models (GCMs). The GCMs were forced by observed natural and anthropogenic greenhouse gas forcings during the historical period, and forced by two Representative Concentration Pathways (RCPs) during the future period (Chegwidden et al., 2019; RMJOC, 2018). The two RCPs represent midrange (RCP4.5) and high (RCP8.5) greenhouse gas concentration scenarios used for modeling the effects of future anthropogenic activity on climate (Meinshausen et al., 2011). Chegwidden et al. (2017) selected 10 GCMs (Table S1) from the Coupled Model Intercomparison Project Phase 5 (Taylor et al., 2011) based on the availability of GCM data downscaled to the desired spatial resolution and on GCM performance as measured by comparison with both regional and larger-scale properties of the observed 20th century climate (RMJOC, 2018; Rupp et al., 2013). Daily GCM temperature and precipitation were downscaled to the resolution of the hydrologic models (1/16° or ∼6 km) using the Multivariate Adaptive Constructed Analogs (MACA) method (Abatzoglou & Brown, 2012). The four hydrologic models include three implementations of the Variable Infiltration Capacity model (VIC-P1, VIC-P2, and VIC-P3) and one implementation of the Precipitation Runoff Modeling System (PRMS-P1). The three VIC models differ by their calibration methods which result in three unique parameter sets, denoted by P1, P2, and P3. In addition to the simulations forced by downscaled GCM output, the four hydrological models were also forced with observation-based gridded temperature and precipitation data over the period 1950–2011 (Livneh et al., 2013). See Chegwidden et al. (2019) for additional details on the multimodel ensemble including calibration information on the hydrological models.

For this study, we used streamflow near 18 river confluences along the Columbia, Snake, and Willamette Rivers (Figure 1 and Table 1). The Snake River is by far the largest tributary to the Columbia; at roughly 2.6 × 105 km2 in area, the Snake basin is roughly equal to the size of the contributing area of Columbia Basin at its confluence with the Snake. The Willamette basin, while only one-tenth the area of the Snake basin, is the most populated subbasin in the Columbia Basin and provides a strong climatic contrast to other major Columbia watersheds. Being west of the Cascades Divide, the Willamette basin is both warmer and much wetter in winter. In the Columbia Basin where the snowpack largely disappears each summer, the ratio of the maximum water year (WY: 1 October to 30 September) daily SWE (SWEmax) to the accumulated precipitation since the first day of the WY to the date of maximum SWE (SWEmax:P) provides a measure of the dynamic range of relative snow storage and therefore of the importance of snow in a basin. In the Willamette basin, snowmelt has a smaller role in runoff generation, with SWEmax:P being only 0.12, compared to higher values (reaching 0.80) in large tributaries east of the Cascades Divide (Table 1). As a result, individual precipitation events of one to a few days typically generate the highest river flows in the Willamette River, whereas snowmelt and rain-on-snow events are the primary generators elsewhere in the Columbia Basin (e.g., Berghuijs et al., 2016).

Table 1. Catchment Areas and Snow Regime Characteristics of the Mainstem and Tributary Rivers at Each of 18 River Confluences
Contributing area (103 km2)a Median date of max SWEb, d Median max SWE/cumulative Pc, d
Mainstem and tributary Mainstem Tributary Mainstem Tributary Mainstem Tributary
1 Columbia + Willamette 609.4 26.0 March 18 March 4 0.51 0.12
2 Columbia + Snake 275.7 261.0 March 18 March 17 0.62 0.46
3 Columbia + Yakima 248.7 14.9 March 19 March 8 0.64 0.49
4 Columbia + Spokane 179.4 15.1 March 23 March 8 0.68 0.41
5 Columbia + Pend Oreille 86.9 68.2 March 23 March 21 0.75 0.59
6 Columbia + Kootenay 36.6 49.6 March 26 March 23 0.81 0.70
7 Snake + Clearwater 221.3 23.4 March 17 March 21 0.46 0.54
8 Snake + Grande Ronde 210.7 8.5 March 17 March 10 0.47 0.39
9 Snake + Salmon 170.6 35.0 March 12 March 27 0.41 0.65
10 Snake + Payette 143.5 7.5 March 10 March 18 0.39 0.62
11 Snake + Malheur 133.6 9.9 March 11 February 13 0.39 0.29
12 Snake + Boise 123.5 10.1 March 16 March 15 0.37 0.57
13 Snake + Owyhee 93.8 29.7 March 17 January 27 0.43 0.16
14 Snake + Bruneau 72.9 7.0 March 21 February 22 0.49 0.24
15 Snake + Henrys Fork 15.2 7.6 March 31 March 27 0.73 0.65
16 Willamette + Clackamas 26.0 2.1 March 4 March 4 0.12 0.30
17 Willamette + Santiam 12.4 2.5 March 9 March 3 0.17 0.28
18 Willamette + McKenzie 3.0 2.8 March 10 March 10 0.29 0.32
  • a Contributing areas are based on gauging station locations (see Figure 1 and Table S2) so will differ from contributing areas exactly at each confluence.
  • b 1951–2000 median of the date of the water year maximum of the catchment-averaged daily SWE (SWEmax).
  • c 1951–2000 median of max SWE divided by the cumulative catchment-averaged precipitation (P) from the start of the water year to the date of max SWE (SWEmax:P).
  • d Four hydrological model ensemble average forced with observed temperature and precipitation.
  • Abbreviation: SWE, snow water equivalent.

The selection of confluences was based primarily on the availability of both NRNI and simulated streamflow data on the mainstem and tributary near a confluence. The smallest tributaries were excluded on the basis that they would contribute very little to extreme streamflow on the mainstem. On each of the three rivers, we went upstream to a point where the contributing areas of the mainstem and its tributaries are comparable in size, for example, to the confluence of the Kootenay River at the Columbia River, where the Kootenay Basin is actually larger than the contributing Columbia Basin (Table 1). Table S2 in the Supporting Information lists the streamflow sites used to estimate the flow at each confluence.

Chegwidden et al. (2019) discussed the projected changes in hydrology, including changes to snow, across the Columbia Basin. Under RCP8.5, ensemble median SWEmax decreased nearly everywhere over the 21st century. Absolute decreases in SWEmax were greatest in the middle elevations (1,000–1,600 m) where the peak snowpack lost ∼150 mm SWE century−1, or about a 60% loss over 100 years. Along with the decreases in SWEmax, the shift in the ensemble median date of SWEmax occurred at rates ranging from −40 to 0 days century−1. For reference, Figure S1 of the Supporting Information shows ensemble median projections of SWEmax:P and date of SWEmax for the 18 mainstem and tributary catchments.

2.2 Methodology

2.2.1 Quantifying Mainstem and Tributary Flow Synchrony

We developed a synchrony index (α) to quantify how coincident in time extreme flows in a mainstem and tributary are at their confluence. The index is based on the sum of the cumulative distribution functions of the flows on the mainstem and tributary
urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0001(1)
where Fmain(Qmain(t)) is the cumulative distribution function of the daily flows in the mainstem (Qmain), evaluated at time t, and Ftrib(Qtrib(t)) is the same but for the tributary. Importantly, Fmain and Ftrib are determined anew for each water year (WY). The factor n/(n − 1), where n is the number of days in the WY (365 or 366), is used to standardize α so that −1 urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0002 α urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0003 1, as will be apparent from examples given below. We calculated the cumulative distribution functions as Fmain = (mmain − 0.5)/n and Ftrib = (mtrib − 0.5)/n, where mmain and mtrib are the ranked orders, from low to high flow, of mainstem and tributary daily flows, respectively, within each WY. Substituting the above definitions of Fmain and Ftrib into Equation 1, the synchrony index can be expressed in terms of the ranks of mainstem and tributary daily flows at time t
urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0004(2)
We provide hypothetical examples of how extreme flows in the mainstem and tributary lead to different values of α. First, the case of highest synchrony occurs when the flows in both the mainstem and tributary reach their WY peaks simultaneously. In that case, the ranked orders (m) are mmain = mtrib = n = 365 and α = 1. Second, a case of complete asynchrony occurs when the mainstem flow is at its WY peak when the tributary flow is at its WY minimum (i.e., mmain = n and mtrib = 1), leading to α = 0. Though not the topic of this study, the synchrony index α is less than zero for synchrony in low flow occurrence.

We calculated α on the days of the annual (by WY) maximum daily flows (AMF) at the 18 selected river confluences. As will be shown, the frequency distribution of α for AMFs is strongly negatively skewed with α exceeding 0.9 at many confluences. We found it useful to use a transformation of the synchrony index urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0005 when examining high-flow extremes. The transformed index α* provides more resolution at high values of α and has a more symmetrical distribution.

2.2.2 Change in Annual Maximum Flows and the Effect of Synchrony Changes

We extracted daily flow Q for a historical (hist: WY 1951–2000) and future (fut; WY 2050–2099) period from each simulation and identified the confluence AMF in each WY in each period. For each period, AMFs were ordered from high to low. Changes in AMF magnitude from the historical to the future periods were calculated separately by their order i in AMF magnitude; for example, the largest AMF in the historical period was paired with the largest AMF in the future period, both indexed by i = 1. Changes were expressed as a ratio R of future to historical AMF for each i
urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0006(3)
We summarized change as the multimodel ensemble average of the changes over the AMFs with urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-00070.2 probability of exceedance per year, that is, a 5-year return interval or greater, which results in the 10 largest AMFs per 50-year simulation period. These rarer, and larger, events are of greater concern for flood hazards. The ensemble average (Rens) was calculated as
urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0008(4)
where j indexes the 10 GCM, and k indexes the four hydrological models. The ensemble averaging filters the effect of internal (i.e., not anthropogenically forced) climate variability (e.g., Hawkins & Sutton, 2009) and assumes the best estimate of the forced change is an average of the GCMs’ and hydrological models’ response. Recent studies have examined sources of variability in projections of the climate (Ahmadalipour et al., 2018; Rupp et al., 2017) and hydrology (Chegwidden et al, 20192020; Queen et al., 2021) for the Columbia Basin, so we did not make variability across the ensemble a focus here. However, we do briefly discuss conspicuous differences in simulations among hydrological models where they occur.

We estimated 95% confident intervals on the ensemble mean of the changes in AMFs (and the synchrony index) using standard bootstrapping (Efron & Tibshirani, 1986). The units of resampling were entire ensemble members to maintain any nonstationarity or temporal dependencies within each simulation.

A change in a high extreme in total confluence flow can result from a combination of changes in magnitude in the mainstem or tributary flow and changes in synchrony. We isolated the effect of synchrony changes through construction of AMFs in an alternate (alt) future in which we resampled the simulated historical mainstem and tributary flow contributions to the confluence AMF, as described below. The historical mainstem and tributary flows that comprise an alternate AMFi have the same cumulative probabilities as the mainstem and tributary flows that comprised the AMFi in the future simulation, therefore differences between the historical and alternate AMFi will only result from differences in synchrony since the mainstem and tributary magnitudes of the alternate AMFi come from the historical simulations. Figure 2 illustrates the process in detail.

Details are in the caption following the image

Schematic for constructing the alternate (alt) annual maximum flow (AMF) at the confluence of the mainstem (main) and tributary (trib) from the ith-largest future (fut) AMF. F denotes the cumulative distribution function and F1 denotes the inverse distribution function. The steps are as follows: (1) From the simulations of future streamflow, the ith-largest future AMF at the confluence is identified, as are its contributions from the mainstem (Qmain,fut,i) and tributary (Qtrib,fut,i); (2) The within-year cumulative probabilities of Qmain,fut,i are Qtrib,fut,i are computed; (3) From the inverse distribution function of the year containing the ith-largest simulated historical AMF, flow magnitudes at the mainstem and tributary are computed that have cumulative probabilities F(Qmain,fut,i) and F(Qtrib,fut,i); and (4) the resulting mainstem (Qmain,alt,i) and tributary (Qmain, trib,i) flow are added to construct the alternate future AMF.

The change in AMF due to the change in probabilities was estimated as Ralt,i = AMFalt,i/AMFhist,i. The ensemble mean of Ralt (Ralt,ens) was calculated using Equation 4 but with Ralt,i,j,k replacing Ri,j,k.

Lastly, we approximated the fraction of change in AMFs (fut − hist) attributed to the change in synchrony. We refer to the attributed fraction as FAS and calculated it as
urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0009(5)
Negative FAS values imply that the change in AMF magnitude due to synchrony alone is in the opposite direction to the total change. For example, if Rens > 1 and FAS is negative, then the synchrony change is dampening the increase in the AMF magnitude.

In general, we expected that a decrease (or increase) in the synchrony index would result in a decrease (or increase) in the AMF of the alternate future relative to the historical period. However, this need not always be true if the changes in the mainstem and tributary cumulative probabilities are in opposing directions. For example, an increase in α could occur if the mainstem flow probability decreases while the tributary flow probability increases, so long as the net probability change is positive. These opposing changes in probability could lead to a decrease in the AMF if the decrease in the magnitude of the mainstem flow outweighs the increase in the tributary flow.

We also expected that the magnitude of Ralt would depend not only on the magnitude of the synchrony change but also on the relative contribution of the tributary to the total flow. We therefore tested if weighting the ensemble mean synchrony change, urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0010, by the mean historical fractional tributary contribution to the AMFs could explain the variability in Ralt,ens and the fractional change in Rens attributable to a synchrony change (FAS) across confluences.

3 Results and Discussion

3.1 The WY 1929–2008 Reconstructed Naturalized Annual Maximum Flows

We begin with a discussion of AMFs in the reconstructed NRNI flow records to provide context for the projected changes in AMFs discussed later. At each of the 18 confluences, we focus our analysis on what we refer to for brevity as “rare” AMFs: Those with 0.2 or smaller probability of exceedance per year. For the 80 water years in NRNI record, these are the largest 16 AMFs.

High-flow extremes on the Columbia River and two major tributaries, the Snake River and Willamette River, can be separated into two regimes (e.g., Hamlet & Lettenmaier, 2007). The first regime consists of AMFs that occur in May and June (e.g., Figure 3, top row). The primary source of water for these AMFs is a relatively quick melting of a snowpack that accumulates throughout the winter and early spring (i.e., the spring freshet). We call this the “cold” regime because the source headwaters remain below 0°C for much of the winter, allowing snow to accumulate. Excluding the confluences along the Willamette and the Columbia-Willamette confluence, which are all west of the Cascades divide, all but one AMF examined here belonged to the cold regime. (Note: The one AMF east of the Cascades divide that did not occur in May or June is the extraordinary March 1993 event at the Snake-Owyhee (Vandal, 2007; see also Supporting Information).) The second regime consists of AMFs that occur from late fall through winter and result from heavy rain or heavy rain on snow. We call this the “warm” regime because the source headwaters rise above 0°C frequently during the winter so that rainfall is common and snowmelt events occur throughout the winter. All rare AMFs on the Willamette River were in the warm regime (see Figure S2).

Details are in the caption following the image

(Row 1) Timing of naturalized (NRNI) annual daily maximum flows (AMFs) with at least a 5-year return interval against the synchrony index α of the mainstem and tributary contributions to the AMF. (Rows 2 and 3) Same as Row 1 but for simulated AMFs in water years 1951–2000 and 2050–2099 (RCP8.5). Symbols for the simulated AMFs are color-coded by hydrological model from the 40-member global climate model-hydrological model ensemble. (All rows) The x-axis is scaled by 1−(1−α)1/2 to increase resolution at values of α near 1. The time axis spans the water year from October (O) to September (S). For each confluence, symbols are sized relative to the smallest NRNI AMF. The gray shading shows the estimated probability density of α and the vertical line marks the mean of 1−(1−α)1/2. NRNI, no-regulation no-irrigation; RCP, Representative Concentration Pathway.

At the confluence of the Columbia and Willamette River, the cold and warm regimes merge. Here, most of the rare AMFs were in the cold regime (Figure 3a, top row) and were dominated by the Columbia; the Willamette contributed <5% of the total streamflow during these events (Figure S3). However, the third and sixth largest AMFs were two warm-regime events occurring in winter (December 1964 and February 1996). Both of these AMFs resulted from widespread and exceptionally heavy rains (Halpert & Bell, 1997; Rantz & Moore, 1965) though snowmelt in the Cascades Range may also have made up a large proportion of the total runoff during the 1996 event (Marks et al., 1998). The Willamette contribution during these two AMFs exceeded 40%, even though the Willamette Basin is only about 4% the size of the contributing area of the Columbia Basin at the confluence.

Most confluences (two-thirds) had high synchrony during rare AMFs (Table 1), using a mean synchrony index urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0011 = 0.98 as an arbitrary threshold for “high.” Three confluences stand out from the rest as having much lower synchrony: the Columbia-Willamette (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0012 = 0.60), the Snake-Malheur (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0013 = 0.62), and the Snake-Owyhee (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0014 = 0.78). Interconfluence variability in the synchrony index of rare AMFs can be qualitatively associated with the similarity of climates of the mainstem and tributary headwaters, using the Köppen-Geiger climate classification as a reference. In the low-synchrony case of the Columbia-Willamette, for example, the headwaters of the Willamette are on the eastern side of the Oregon Coast Range and western side of Oregon Cascades, both relatively warm, while the primary source headwaters of the Columbia are in the much colder and snowier Rocky Mountains (Figure 1). The Snake-Malheur and Snake-Owyhee also both have warmer, lower elevation tributary headwaters (and therefore lower basin SWEmax:P; Table 1) relative to the major Snake headwaters located in the Rocky Mountains.

We highlight four confluences where the tributary has provided a very large fraction of the total flow during AMFs because in such cases, we expect a change in synchrony to have a larger impact on the magnitude of AMFs. Historically at the Columbia-Snake, Columbia-Pend Oreille, and Snake-Salmon, the tributaries provided, on average 29%, 30%, and 44%, respectively, of the total flow magnitude during rare AMFs. By area, the Snake and Pend Oreille are the first and second largest tributaries to the Columbia, respectively, and the Salmon is the largest tributary to the Snake (Table 1). The Willamette River historically has contributed only 8% to the rare AMFs at the Columbia-Willamette confluence, but, as stated above, the Willamette accounted for over 40% of the total flow during two rare AMFs.

The Columbia-Snake confluence holds interest because the synchrony index is high during the rare AMFs (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0015 = 0.990; Figure 3b, top row) although the two rivers have headwaters that are over 1,000 km apart. The nearly identical timing of median SWEmax points to similar snow regimes in the two basins, although the Columbia Basin accumulates more snow relative to total precipitation (Table 1). The high synchrony and the May-June timing of the AMFs suggests that large snow accumulation occurred during the winter in both basins prior to the AMFs and region-wide warm weather drove synchronous snowmelt. In comparison, the Columbia-Pend Oreille confluence has a similar and slightly lower synchrony index (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0016 = 0.986; Figure 3c, top row) although the headwaters are much closer together (a few hundred kilometers apart; Figure 1).

The Snake-Salmon confluence has the highest mean synchrony index (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0017 = 0.994; Figure 3d, top row) of any of the nine confluences on the Snake. Although the Salmon is sourced from the western side of the Idaho Rockies and the upper Snake headwaters are in the Rockies of western Wyoming, the two sources are climatically similar (Figure 1).

As noted above, the Columbia-Willamette on average has the lowest synchrony of all 18 confluences, owing primarily to the dissimilar flow regimes of the Columbia and Willamette (cold and warm, respectively). A secondary cause of the lower synchrony at this confluence may be the different times that it takes a flood wave to travel from the farthest upstream reaches of these two basins to their confluence: roughly 72–96 h on the Columbia River and 24–36 h on the Willamette River. While the average synchrony index is low, the two warm-regime AMFs discussed above have a synchrony index (∼0.95) that is much higher than the others at this confluence.

3.2 Projected Changes in Annual Maximum Flows

Before discussing projected changes in synchrony and the impact of these changes on AMFs, we briefly summarize the projected changes in AMFs themselves. The ensemble average change in rare AMFs, given as the ratio of the future to historical ensemble mean of AMFs with a 5-year return interval or higher (Rens), is greater than one under RCP8.5 at all of the 18 confluences and there is a wide range in increases, from as low as Rens = 1.02 at the Columbia-Snake, to as high as Rens = 1.52 at the Willamette-McKenzie (Table 2, Column B). These results are consistent with Queen et al. (2021) who estimated changes in the 20- and 100-year return interval AMFs using the same data set but based their results on fitted generalized extreme value distributions. Interestingly, we find that increases in AMFs under RCP8.5 are not greater than under RCP4.5 everywhere; along the Columbia River upstream from the confluence with the Willamette, increases are smaller with the higher anthropogenic forcing. Although Queen et al. (2021) do not show detailed results from RCP4.5, they state that that “to first order the changes in flood magnitude in RCP4.5 were ∼2/3 those in RCP8.5.” The Columbia River, along much of its mainstem, is an important exception to their generalization and the cause for the nonmonotonic response to increased anthropogenic forcing may lie in changes in mainstem-tributary synchrony, which we discuss in Section 10.

Table 2. Mean of Synchrony Index α of Annual Maximum Daily Flows (AMFs) Immediately Below Confluence of Mainstem and Tributary and With at Least a 5-Year Return Interval
Mainstem and tributary NRNIa All HMb VIC-1c VIC-2c VIC-3c PRMS-1c
1 Columbia + Willamette 0.599 0.581 0.548 0.571 0.628 0.578
2 Columbia + Snake 0.990 0.978 0.983 0.976 0.976 0.978
3 Columbia + Yakima 0.986 0.918 0.938 0.879 0.927 0.929
4 Columbia + Spokane 0.921 0.910 0.889 0.909 0.914 0.927
5 Columbia + Pend Oreille 0.986 0.990 0.991 0.989 0.991 0.989
6 Columbia + Kootenay 0.994 0.993 0.993 0.991 0.996 0.993
7 Snake + Clearwater 0.994 0.991 0.993 0.989 0.990 0.992
8 Snake + Grande Ronde 0.974 0.960 0.946 0.975 0.967 0.951
9 Snake + Salmon 0.994 0.994 0.995 0.991 0.994 0.995
10 Snake + Payette 0.985 0.980 0.990 0.972 0.972 0.984
11 Snake + Malheur 0.622 0.911 0.904 0.894 0.934 0.913
12 Snake + Boise 0.984 0.979 0.978 0.971 0.985 0.980
13 Snake + Owyhee 0.776 0.911 0.941 0.832 0.977 0.896
14 Snake + Bruneau 0.967 0.945 0.945 0.930 0.958 0.947
15 Snake + Henrys Fork 0.984 0.995 0.994 0.995 0.997 0.995
16 Willamette + Clackamas 0.991 0.977 0.937 0.986 0.994 0.992
17 Willamette + Santiam 0.987 0.997 0.994 0.998 0.999 0.999
18 Willamette + McKenzie 0.999 0.998 0.995 0.999 1.000 1.000
  • a No-regulation, no irrigation (1929–2008).
  • b All hydrological models (VIC-1, VIC-2, VIC-3, PRMS-1) and all global climate models.
  • c Hydrological models forced with surface meteorology downscaled from 10 global climate model simulations (1951–2000).
  • Abbreviations: NRNI, no-regulation no-irrigation; PRMS, Precipitation Runoff Modeling System; VIC, Variable Infiltration Capacity model.

Queen et al. (2021) did not include projections for the Columbia at, or below, the confluence with the Willamette, so a comparison of their results to ours cannot be made for the Columbia-Willamette. We find that increases in AMFs at the Columbia-Willamette are much larger than those at the Columbia-Snake upstream (3.5 times greater for RCP8.5). Moreover, nearly all of this increase can be attributed to increased discharge from the Willamette (Figure S4), which is impressive given the Willamette contributes on average only 4% to the Columbia-Willamette AMFs in the historical simulations.

3.3 Projected Synchrony Changes

The ensemble mean of the change in the synchrony index, urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0018, varies widely across the 18 confluences (Table 3, column C). Along the Columbia River, the mean synchrony index is mainly projected to decrease in the future. These decreases become larger (i.e., the changes are more negative) from RCP4.5 to RCP8.5. AMFs with relatively low synchrony also occur more frequently in the future scenarios, extending the lower tail of the distribution of α (e.g., Figures 3b and 3c). One explanation for the decreased synchrony is the region-wide shift toward an earlier freshet occurring at different rates among tributaries (Queen et al., 2021). Because the snowpack in the colder headwaters, such as the upper Columbia headwaters in the Canadian Rockies, is less impacted by the projected warming than the snowpack in warmer headwaters that tend to feed the downstream tributaries (e.g., Chegwidden et al., 2019), the relative timing of the freshet can be expected to diverge across many tributaries to the Columbia. A second, and related, explanation is that the rate that snowmelt becomes less of a primary driver of peak flows varies across the Columbia Basin. The headwaters of basins on the eastern side of the Washington State Cascades, such as the Yakima Basin, show a faster rate of transition from snowmelt- to rainfall-driven AMFs than do headwater basins in the Rocky Mountains (Chegwidden et al., 2020).

Table 3. Multimodel Ensemble Mean Properties of the Annual Maximum Daily Flows (AMFs) With at Least a 5-Years Return Interval at 18 River Confluences (A) Fractional Tributary Contribution to Historical Confluence AMF, (B) Ratio of Future to Historical Confluence AMF (Rens; RCP4.5/RCP8.5), (C) Change in Synchrony Index of Confluence AMFs (urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0019; RCP4.5/RCP8.5), and (D) Fraction of Change in Confluence AMF Attributed to Change in Synchrony of Tributary and Mainstem AMF (RCP4.5/RCP8.5)
Mainstem and tributary A Ba Cb D
1 Columbia + Willamette 0.04 1.04/1.07 0.005/0.212 −0.10/0.35
2 Columbia + Snake 0.31 1.03/1.02 −0.012/−0.018 −1.21/−2.3
3 Columbia + Yakima 0.03 1.06/1.03 −0.111/−0.221 −0.11/−0.25
4 Columbia + Spokane 0.04 1.07/1.06 −0.084/−0.190 −0.12/−0.25
5 Columbia + Pend Oreille 0.33 1.09/1.09 −0.006/−0.010 −0.32/−0.40
6 Columbia + Kootenay 0.52 1.09/1.08 −0.001/−0.001 −0.20/−0.23
7 Snake + Clearwater 0.25 1.09/1.19 −0.008/−0.002 −0.15/−0.09
8 Snake + Grande Ronde 0.05 1.10/1.23 −0.022/0.002 −0.06/−0.02
9 Snake + Salmon 0.40 1.10/1.24 −0.008/−0.007 −0.23/−0.11
10 Snake + Payette 0.15 1.16/1.37 −0.013/−0.031 −0.07/−0.04
11 Snake + Malheur 0.03 1.19/1.41 −0.022/−0.008 −0.02/0.00
12 Snake + Boise 0.16 1.18/1.40 −0.007/−0.016 −0.10/−0.04
13 Snake + Owyhee 0.12 1.20/1.42 −0.028/0.027 −0.17/−0.03
14 Snake + Bruneau 0.04 1.19/1.36 −0.049/−0.054 −0.05/−0.03
15 Snake + Henrys Fork 0.27 1.16/1.28 0.000/0.001 −0.05/−0.02
16 Willamette + Clackamas 0.07 1.28/1.35 0.017/0.017 −0.03/−0.01
17 Willamette + Santiam 0.22 1.32/1.43 0.001/0.001 −0.03/−0.01
18 Willamette + McKenzie 0.47 1.39/1.52 0.000/0.000 −0.04/−0.03
  • a At all 18 confluences, the 95% confidence interval does not include (i.e., all changes are statistically significant at α = 0.025).
  • b Values in bold where 95% confidence interval does not include 0 (i.e., bold values are statistically significant at α = 0.025).

The Columbia-Willamette provides a noteworthy exception to this pattern of decreased synchrony along the Columbia River. At this confluence, an earlier freshet on the mainstem and more frequent melt episodes during the winter lead to higher mainstem flows on the Columbia when the Willamette is in peak-flow season. The result is an increase in synchrony and many more AMFs occurring in winter, particularly under RCP8.5 (Figure 3a, middle and lower row).

As it does along the Columbia River, synchrony decreases along the Snake River under RCP4.5. This decrease does not persist everywhere with larger anthropogenic forcing, however, as some confluences under RCP8.5 show no change in mean α, with the Snake-Grande Ronde, Snake-Malheur, and Snake-Owyhee being the most notable examples. These three confluences have historically lower synchrony indices relative to the others on the Snake. It may be that while the flow regimes on these tributaries diverge from the mainstem flow regime as they lose snow under RCP4.5, the tributaries and mainstem become more alike under RCP8.5 as the entire Snake basin becomes more rain-dominated.

The synchrony index increases or stays essentially unchanged at the confluences along the Willamette River. The Willamette-Clackamas confluence has by far the largest increase in the synchrony index on the Willamette. At this confluence in the future scenarios, low synchrony AMFs disappear (not shown) and the Willamette and Clackamas become much more similar to each other with respect to their snow regimes (Figure S1).

To this point, we have made largely qualitative associations of synchrony with spatial variation in climate, principally with respect to snow regimes. We have characterized the snow regime mainly in terms of median SWEmax:P, which is related to the date of median SWEmax:P across the Columbia Basin: Catchments with higher SWEmax:P tend to have later dates of median SWEmax:P (Figure S1). To be more quantitative in this association, we can relate the difference in snow regimes between mainstem and tributary—quantified as the absolute difference between ensemble medians of SWEmax:P for the mainstem and tributary—to the ensemble mean transformed synchrony index at each of the 18 confluences. Pearson's correlation coefficient r for this relationship equals −0.73 for the historical period, implying a decrease in synchrony with increasing difference in mainstem and tributary snow regimes.

3.4 Effect of Synchrony Changes on Annual Maximum Flows

AMFs decrease in the alternate future relative to the historical period at nearly all confluences, implying that projected changes in timing generally have a dampening effect on AMFs. As a consequence, the fraction of the change in AMFs attributed to synchrony change (FAS) is negative nearly everywhere (Table 3, column D). Synchrony changes have the largest relative effect along the Columbia upstream from the confluence with the Willamette, with FAS ranging from −0.23 to −2.30 under RCP8.5. The Columbia-Snake is the most prominent example, where the decreases in AMFs due to synchrony changes alone are of greater magnitude than the projected increases in AMFs (Figure 4b). The implication is that increases in AMFs at the Columbia-Snake would be more than double and more than triple those projected under RCP4.5 and RCP8.5, respectively, were there no decreases in synchrony in the future (compare heights of navy blue and slate blue bars in Figure 4b). The effect of decreasing synchrony is also large at the Columbia-Pend Oreille, where AMF increases would be about 30% and 40% higher under RCP4.5 and RCP8.5, respectively, if synchrony remained unchanged. As noted in Section 8, increases in AMFs are smaller under RCP8.5 than RCP4.5 at the Columbia River confluences (6%–38% smaller), with the exception of the Columbia-Willamette. Our results indicate that the effect of decreased synchrony at many Columbia River confluences is sufficient to counter the larger increases in headwater discharge under RCP8.5 relative to RCP4.5.

Details are in the caption following the image

RCP4.5 and RCP8.5 ensemble mean projected change in 5-year return interval and higher annual maximum daily flows (AMFs) and the isolated change due to synchrony and magnitude to change only, at four major river confluences. (Navy blue bar) Change in AMF as the ratio Rens of future (fut) to historical (hist) AMF. (Orange bar) Change in AMF accounting for the projected change in synchrony of mainstem and tributary flow only, given as the ratio Ralt,ens of alternate (alt) to historical (hist) AMF. (Slate blue bar) Change in AMF accounting for the projected change in magnitude only of mainstem and tributary flows, calculated as RensRalt,ens + 1. Vertical line segment shows the bootstrapped confidence limits (2.5th to 97.5th percentile) on the estimate of Rens. See Figure S5 for results at all 18 confluences. RCP, Representative Concentration Pathway.

The effects of synchrony changes are small (−0.1 < FAS < 0.1) along most of the Snake and all the Willamette confluences. The Snake-Salmon (Figure 4d) and Snake-Clearwater are two exceptions, where AMFs would be about 23% and 15% greater under RCP4.5 without the synchrony changes. While the Snake-Owyhee has a similar FAS to the Snake-Salmon, and Snake-Clearwater under RCP4.5, the hydrological models greatly overestimate urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0020 at the Snake-Owyhee (Table 2), diminishing our confidence in results for the Snake-Owyhee.

In distinct contrast to the other confluences, synchrony changes at the Columbia-Willamette contribute a sizable positive fraction (35%) to the changes in AMFs, though only under RCP8.5 (Figure 4a). The relatively large warming under RCP8.5 causes a large enough shift toward earlier peak flows on the Columbia Basin such that mainstem flows are higher during the winter peak discharge season in the warm-regime Willamette.

As a simple way of predicting where changes in synchrony will have the largest effect on the ensemble mean change in AMFs, the change in ensemble mean of the transformed synchrony index urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0021 weighted by the historical contributing tributary fraction, does moderately well (Figure 5). The Pearson correlation coefficient is 0.77 when predicting variability in FAS across all 18 confluences and both RCPs. The transformation provides more resolution in α at high values of α and delivers a tighter relationship between synchrony and FAS. Values of the exponent in transformation index ranging from 0.25 to 0.75 were tried, but any improvements were marginal.

Details are in the caption following the image

Fraction of the ensemble mean change in annual maximum daily flows (AMFs) attributed to a change in synchrony (FAS) versus the ensemble mean of the change in the transformed synchrony index α* = 1−(1−α)1/2 weighted by the historical mean fractional contribution of the tributary to the AMFs. Each point represents a different river confluence and different RCP. Only the AMFs with at least a 5-year return interval are included in the ensemble means. Because all changes on AMF magnitude are increases, a negative FAS means the changes in synchrony dampen the increase in the AMF magnitude. RCP, Representative Concentration Pathway.

The sign alone (negative or positive) of urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0022, is a good, but imperfect, indicator of whether synchrony changes contributed negatively or positively to AMF changes. Our general expectation was that increases in urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0023 would be associated with FAS > 0 (or Ralt,ens > 1), and in the 36 cases we examined (18 confluences ×2 RCPs), we found only three cases counter to expectation (i.e., three had statistically significant increases in urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0024 associated with FAS < 0, or Ralt,ens < 1). FAS was relatively close to zero (−0.03 to −0.01) in these three cases and could be considered negligible given the uncertainties in projected AMF changes. As discussed in Section 3, the relationship between urn:x-wiley:00431397:media:wrcr25226:wrcr25226-math-0025 and FAS (or Ralt,ens) is not completely straightforward and depends on the particular cumulative distribution functions Fmain(Qmain(t)) and Ftrib(Qtrib(t)), the magnitudes of the changes in Fmain(Qmain(t)) and Ftrib(Qtrib(t)) between the historical and future periods, and the relative contributions of the mainstem and tributary to the AMFi at each confluence.

3.5 Study Limitations and Uncertainty

A large portion of the uncertainty in the response of peak flows to anthropogenic forcing stems from incomplete and imperfect representation of key hydrological processes (e.g., Queen et al., 2021; Thober et al., 2018). Evaluating the simulations in this study with respect to fidelity in reproducing the properties of interest (e.g., mainstem-tributary synchrony and contributing tributary fraction during AMFs) is hindered by the small sample size of AMFs in the reconstructed streamflow data. Still, large discrepancies between observations and simulations can suggest where results should be viewed with more caution. For example, all the hydrological models greatly overestimate the synchrony index of AMFs at the Snake-Malheur and Snake-Owyhee confluences (Table 2), indicating there may be important model deficiencies in peak-flow generating mechanisms at these tributaries that could result in substantial errors in the response to anthropogenic forcing. Also, where the simulated historical fractional contribution of the tributary to the total flow is much too low or too high (see Table S4 in Supporting Information), the impact of the change in synchrony on the AMFs may be, respectively, under- or over-estimated. Given that the hydrological models underestimate the historical tributary fraction of the Willamette at the Columbia-Willamette confluence, for example, we can speculate that the consequence of the synchrony changes to the AMFs is underestimated at this confluence.

We can also direct a critical eye to potential errors in the simulations based on our knowledge of how the hydrological models were built. Although all the hydrological models were run at the same spatial resolution, the three VIC implementations track subtile snowpack on elevation bands whereas PRMS tracks tile-average snowpack only. As a result, snow loss is more spatiotemporally diffuse, and some snow remains longer, in the VIC models as the climate warms. Chegwidden et al. (20192020) found that PRMS-P1, lacking the subtile snow, simulated the greatest increases in winter streamflow and largest transition from snowmelt to rainfall-driven high flows. We also see PRMS-P1 simulating more winter AMFs than do the VIC models (Figure 3, bottom row). The lower effective spatial resolution of PRMS-P1 would lead it to simulate changes that are too sharp and too large given the applied anthropogenic forcing, assuming the model were otherwise perfect.

Whereas spatial averaging, not just with PRMS-P1 but with all the hydrological models, leads to overly “step-like” sensitivity of snow-related hydrological processes to warming, known errors in meteorological inputs may have the opposite effect. Three of hydrological models in the study were calibrated using gridded meteorological data with a known cold bias at high elevations (Alder & Hostetler, 2019). A possible consequence is that the higher elevations retain snow too long under the applied anthropogenic forcing. The implication to synchrony is not straightforward, however, because a change in synchrony depends on the rate of change in timing of high flows on the mainstem relative to the rate of change on the tributary. The cold bias could initially lead to overestimation of synchrony decreases if the coldest headwaters remain relatively unimpacted for longer into the future while the warmer headwaters experience the transition to more rainfall, and fewer snowpack accumulation, events.

Lastly, it is important to mention that we do not report results of bootstrapped confidence intervals on FAS (or Ralt,ens). The confidence intervals generated by resampling Ralt,i,j.k and Ri,j.k implied statistical significance at all confluences (all inner 95% confidence intervals on FAS excluded 0), which certainly overestimates our actual confidence in the FAS estimates. We think the narrow confidence intervals result from the way we created the alternate future scenario from the historical period simulations. We recommend further research into how to adequately represent statistical uncertainty in estimates of FAS.

4 Conclusions

Results from a large ensemble of simulations of naturalized river flow indicate that the synchrony of mainstem and tributary discharge during high-flow extremes along much of the Columbia and Snake River will decrease under a moderate anthropogenic forcing scenario (RCP4.5). These projections are consistent with observed decreases in a flood synchrony length scale across regions of Europe where there has been a concurrent decrease in snowpack (Kemter et al., 2020) and the hypothesis that the decreased snowmelt will lead to decrease in high-flow synchrony in rivers sourced from the Rocky Mountains of North America (Brunner et al., 2020). On the Columbia, synchrony will decrease further with higher anthropogenic forcing (RCP8.5). This monotonic decrease in synchrony is consistent with a divergence in flow regimes across subbasins of the Columbia as they respond to climate change at different rates depending on the rate at which spring snowmelt becomes a smaller contributing factor to high flows.

Over most of the Snake, however, the change in synchrony between RCP4.5 and RCP8.5 reverses direction: six of nine confluences show smaller decreases or no decreases at all in synchrony under the higher forcing. This reversal in response suggests an initial divergence of water regimes until a certain amount of warming is reached, beyond which water regimes begin to converge as snows plays an ever-diminished role even in the coldest subbasins of the Snake.

The relative effects of the decreased synchrony on AMF magnitudes are greatest on the larger Columbia mainstem upstream from the Willamette, helping to make the projected AMF increases much lower than elsewhere in the basin. Moreover, the decreases in synchrony under RCP8.5 are of sufficient size to lead to smaller increases in AMFs under RCP8.5 than under RCP4.5.

More than one factor likely contributes to the small increases in high-flow extremes projected along the lower reaches of the Columbia and Snake relative to increases at headwater basins. Though not examined here, antecedent available storage and spatially varying changes in heavy precipitation will affect how high-flow extremes respond to climate change differently in large and small basins. Further research is warranted on this topic. Overall, decreasing synchrony serves as one factor that dampens the forced increases in AMFs along lower reaches of Columbia in the multimodel ensembles simulations of Chegwidden et al. (2017) and, we speculate, also contribute to the very small to negative changes in high-flow extremes reported in Tohver et al. (2014) and Maurer et al. (2018).

The Columbia-Willamette confluence offers a special case in the Columbia River network because the mainstem and tributary begin with historically dissimilar flow regimes (cold and warm, respectively) and the transition to more rain-driven high flows in the Columbia mainstem leads to large increases in synchrony under high forcing. The larger role of the Willamette River discharge to extremely high flows on Columbia mainstem will complicate flood risk management at the confluence, the location of a large urban center (Portland, Oregon), where management also must consider the tidal influence (Helaire et al., 2019; Wherry et al., 2019). The season of potential flooding will extend from November to June and reservoir managers on the Willamette will need to consider the discharge on the Columbia to a greater degree than they have before.

While this study examined large rivers in the Columbia Basin, results may be generalizable to the class of large rivers in snowmelt-dominated regimes around the world. In particular, our results may be generalized to rivers where snowmelt provides a large contribution to high-flow extremes, and therefore flood risk, but that also receive considerable discharge from tributaries that drain warmer basins. In these large rivers, flow regimes across colder subbasins may initially diverge with warming as each subbasin responds to climate change at different rates, leading to a general decrease in synchrony. Eventually, if the warming is sufficiently large, most subbasins may transition to the rain-dominated, warm regime, effectively reversing the decreasing trend in synchrony. Where large river flow regimes are initially different from their tributaries, increases in synchrony with warming may lead to larger high-flow extremes, and greater flood risk.

Acknowledgments

D. E. Rupp was supported by the Innovations at the Nexus of Food, Energy and Water Systems (INFEWS) Program of the National Science Foundation, Grant #1740082. We acknowledge the World Climate Research Programme's Working Group on Coupled Modeling, which is responsible for CMIP, and we thank the climate modeling groups for producing and making available their model output.

    Data Availability Statement

    The multimodel ensemble hydrological data are publicly available through a data release (Chegwidden et al., 2019) and accessible via a DOI link: https://doi.org/10.5281/zenodo.854763 and also the from the University of Washington at http://www.hydro.washington.edu/CRCC/. The NRNI streamflow data are publicly accessible from the Bonneville Power Administration at https://www.bpa.gov/p/Power-Products/Historical-Streamflow-Data/Pages/No-Regulation-No-Irrigation-Data.aspx.