1 INTRODUCTION

Interactions between surface water and groundwater underlie the riparian zones of river corridors (Naiman et al., 2005; Wohl, 2014). Riparian plant species are distributed in relation to past and present river flows, particularly where flows replenish alluvial soil water and groundwater (Diehl et al., 2018; Lytle & Poff, 2004; Sargeant & Singer, 2021; Stromberg & Merritt, 2016). Across the globe, human management of rivers has changed streamflow patterns (Nilsson et al., 2005). The resulting novel hydrologic regimes can significantly alter species distribution and competition (Merritt & Cooper, 2000; Volke et al., 2019), disrupting ecological processes and initiating ecosystem state transitions that forfeit benefits of the natural system (Martínez-Vilalta & Lloret, 2016; Nilsson & Berggren, 2000; Poff et al., 2010).

Flow regulation can initiate hydrological drought (i.e., drought that reduces water supply for biota in freshwater ecosystems). In the most extreme cases all surface water is removed, eliminating the freshwater habitat required by aquatic species (Perkin et al., 2015) and aquatic-terrestrial food webs (Power et al., 2008). Other diversions remove all flow up to a threshold, above which water remains in the channel. Riparian ecosystems may persist under these modified conditions. Combining a management-induced hydrological drought with an additional stressor such as atmospheric drought (i.e., a climate-induced drought from low precipitation and/or high temperature) may trigger a threshold response in riparian ecosystems. The resilience of species and the resistance of the community reflect an ecosystem's capacity to persist through such disturbances and may create time lags that obscure the identification of cause and effect relationships (Chin et al., 2014; Frank et al., 2015; Poeppl et al., 2017), such as those between river regulation and riparian decline. Therefore, being able to predict how streamflow reduction affects riparian ecosystems remains a key challenge for land and water managers.

Cottonwoods dominate the overstorey of many riparian forests across the northern hemisphere (Friedman et al., 2005; Karrenberg et al., 2002; Stettler, 1996). Although they occupy both humid and arid regions, cottonwoods are highly drought-sensitive because their xylem cells cavitate under relatively small water tension caused by low soil water availability (Pockman & Sperry, 2000; Sparks & Black, 1999; Tyree et al., 1994). Cottonwoods cannot survive on local precipitation alone in semi-arid regions, and they require soil water or shallow water tables recharged by streamflow (Tai et al., 2018). The rooting depth of phreatophytes, such as cottonwoods, is related to the prevailing groundwater depth during plant maturation (Butler et al., 2007; Rood et al., 2011), and changes in water table depth can threaten tree survival (Cooper et al., 2003). Reductions in available water can initiate cottonwood branch sacrifice that reduces transpiration and facilitates whole-tree survival (Horton et al., 2001; Rood et al., 2003; Schook et al., 2020; Schook, Carlson, et al., 2016). However, branch loss also limits photosynthesis and can eventually lead to tree mortality and forest loss via carbon starvation (Grossiord et al., 2020; McDowell et al., 2008).

Riparian dendrochronology is a tool for revealing historical relationships between streamflow and connected ecosystems (Philipsen et al., 2018; Stella et al., 2013), especially in semi-arid regions where streamflow replenishes alluvial groundwater. Cottonwood ring width is often more highly correlated to flow than to local precipitation along free-flowing rivers (Johnson et al., 1976; Meko et al., 2015), and tree-ring width can be used as a proxy for discharge (Meko et al., 2020; Schook, Friedman, & Rathburn, 2016). Regulating streamflow can force trees to become more dependent on local precipitation (Reily & Johnson, 1982) or lead to tree death if changes are too extreme and local precipitation is insufficient (Cooper et al., 2003).

Drought may not decrease annual ring growth in all trees due to complex competitive interactions (Keyser & Brown, 2016). Mortality of some trees can release survivors from competition, negating or even reversing the expected tree-ring growth decreases of surviving trees during stress events (Aakala et al., 2011; Berg et al., 2006; Scott et al., 1999). Therefore, ring width from surviving trees alone may not accurately reveal past droughts. A more complete drought reconstruction can be developed by incorporating stand-level information on changes in forest area and tree density (Martínez-Vilalta & Lloret, 2016; Vlam et al., 2014). Given the value of cottonwoods as foundation species in riparian corridors, an improved understanding of their ecohydrological limitations is essential for understanding how to manage the habitats they create.

We investigated riparian cottonwood response to a water diversion that began in 1961 along a mountain stream in the Great Basin of the western United States. The study stream included a mostly dewatered reach, an upstream reference reach above the diversion, and a downstream reference reach below where water is returned to the channel. We set out to answer three questions: Does the removal of low-to-moderate flows affect riparian forest productivity even when peak flow remains mostly unchanged? Do tree rings more clearly indicate ecosystem conditions if scaled up from the tree- to the forest-scale? How soon after flow diversion begins do changes occur? We combined cottonwood growth measurements with stand-level demographic data of establishment and mortality to calculate changes in stemwood production over time. The study approach included tree-ring measurements from living trees, death dates of standing and fallen dead trees, aerial photo analysis of changes in forest area over time, and groundwater investigations to describe a half-century of riparian forest changes.

2 METHODS

2.1 Study site

2.1.1 Regional and local setting

Snake Creek flows from the southern Snake Range in Great Basin National Park, Nevada, USA, and naturally terminates in the closed basin Snake Valley in western Utah (Figure 1). The region has normal faulting from crustal expansion that causes streams to cross diverse rock units over short distances. For 2 km, Snake Creek flows over Pole Canyon limestone, a porous carbonate-rich bedrock where surface water naturally recharges groundwater. In 1961, downstream water users seeking to limit surface water loss built a 5-km water diversion pipeline to move water across this reach and deliver it downstream. The pipeline diverts up to 0.28 m³ s⁻¹, and higher flows exceed the pipeline capacity and remain in the channel.

We studied three reaches at Snake Creek: the pipeline-dewatered reach (DW) and reference reaches immediately upstream (RU) and downstream (RD) of DW (Figure 1 and Table 1). The DW reach is 5 km long from the pipeline inlet to outlet. The pipeline leaks and groundwater discharges into DW, producing intermittent or ephemeral flow in a few small areas. RU extends 780-m upstream from DW and has perennial baseflow of <0.03 m³ s⁻¹. RD extends 340-m downstream from DW and has perennial baseflow of <0.03 m³ s⁻¹ from emerging groundwater and pipeline return flow. It is unknown how much the pipeline augments flow in RD above natural conditions, but any change in water table depth and seasonality is likely small.

TABLE 1. Characteristics of the three study reaches, including their top five riparian woody plants by relative cover

	Reference upstream (RU)	Dewatered (DW)	Reference downstream (RD)
Elevation (m.a.s.l.)	2358–2296	2296–2067	2067–2047
Valley slope	5.7%	5.2%	5.3%
Reach length (m)	780	4900	340
Number of live cottonwoods	276	1132	230
Number of dead cottonwoods	34	449	32
Riparian trees and shrubs
#1 (most common)	Abies concolor (46%)	Populus spp. (32%)	Populus spp. (39%)
#2	Populus spp. (30%)	Juniperus osteosperma (18%)	Betula occidentalis (28%)
#3	Populus tremuloides (13%)	Pinus monophylla (17%)	Juniperus osteosperma (16%)
#4	Betula occidentalis (9%)	Abies concolor (8%)	Pinus monophylla (12%)
#5	Juniperus osteosperma (3%)	Betula occidentalis (6%)	Artemisia spp. (3%)

The Snake Creek riparian corridor is composed of the deciduous species narrowleaf cottonwood (Populus angustifolia) and hybrid cottonwoods (P. angustifolia × P. trichocarpa), as well as aspen (P. tremuloides), water birch (Betula occidentalis), sandbar willow (Salix exigua), Wood's rose (Rosa woodsii) and chokecherry (Prunus virginiana). Conifers include single-leaf pinyon pine (Pinus monophylla), Utah juniper (Juniperus osteosperma) and white fir (Abies concolor). Plant nomenclature follows the USDA PLANTS Database. Our analyses here focus on the cottonwoods that dominated much of the riparian corridor (Figure 2).

2.1.2 Streamflow and weather

Snake Creek has a snowmelt-dominated flow regime that typically peaks in May or June (Figure 3a). The Snake Creek flow record at NPS gauging station SNK3, located in the RU reach, began in 2011 and was insufficient for long-term analysis. Therefore, we used the 1960–2019 flow record from Cleve Creek (USGS #10343700), located 45 km northwest of Snake Creek and in a similar landscape position (Prudic et al., 2015; Schook et al., 2020). Their overlapping 2011–2017 period was used to correlate daily and monthly flows (r = 0.89). We estimated natural monthly discharge in Snake Creek using a linear best-fit equation constructed from shared months. Because 22% of the Cleve Creek record was missing, we imputed missing months using regression equations from 40 years of overlap with the highly correlated and adjacent Steptoe Creek (USGS #10244950; monthly r = 0.87 to 0.97).

Snake Creek mean annual flow for 1960–2019 was calculated to be 0.12 m³ s⁻¹. We estimated flow bypassing the pipeline inlet and entering the DW reach by subtracting its intake capacity of 0.28 m³ s⁻¹, although its capacity occasionally decreased when debris-clogged during high run-off. Subtracting pipeline intake from the calculated mean monthly flow indicated no flow occurred in the DW reach in 88% of months after the pipeline was installed (Figure 3b).

The 1 April snow water equivalent at the Baker Creek #3 Snow Course (NRCS) during the study years with groundwater data was 60% of the 1980–2010 average in 2018, 163% in 2019 and 77% in 2020. The DW reach had continuous surface flow in 2019 but not 2018 or 2020.

Annual precipitation varied spatially from as little as 16 cm year⁻¹ near Snake Creek's terminus in the Snake Valley to 81 cm year⁻¹ at high elevations where mountain peaks reach 3982 m.a.s.l. (Prudic et al., 2015). The Great Basin National Park weather station (GHCND USC00263340) at 2080 m.a.s.l., the same elevation as the DW reach, averages 35 cm year⁻¹ of precipitation, with slightly higher amounts in winter than in summer. Mean daily temperature ranges from −7°C in December and January to 29°C in July.

In addition to the hydrologic drought imposed by diversion, atmospheric drought was characterized by vapour pressure deficit (VPD) for 1937–2017 using PRISM (Daly et al., 2008) for a grid cell in the centre of the study area. We averaged daily minimum VPD values for the June–August growing season. Minima were used because leaf water potential measurements revealed stomatal closure by 9:00 am (unpublished data), indicating that trees largely guarded against daytime VPD highs and conducted more gas exchange during lower VPD periods. In addition to a significant increasing VPD trend across the study period (linear regression p < 0.01), the years 2000–2003 all had higher VPD than any prior year (Figure 4), which we characterize as extreme atmospheric drought.

3 RESULTS

3.1 Hydrology

Snake Creek flows sustained alluvial groundwater within its riparian corridor. No surface flow or groundwater was detected at any DW reach transect when the monitoring wells were installed in July 2018, indicating the water table was deeper than 4.8 m at all seven wells. This was below the maximum rooting depth previously reported for Populus species of 0.8 to 3.7 m below ground (Holloway et al., 2017; Rood et al., 2011; Singer et al., 2013; Sprackling & Read, 1979; Stromberg, 2013; Williams & Cooper, 2005). At the time of installation, the water table was 1.9 m below ground in the single well in the RD reach.

The high volume 2019 snowmelt run-off surpassed the pipeline's intake capacity and caused surface water to flow through the entire DW reach for 102 days. Snake Creek became strongly hydrologically losing at two of the DW groundwater transects and neutral at the other two (Figure 5; expanded dataset in Figure S1).

For example, the water table was within 5 m of the ground surface at the hydrologically losing Transect 2 only when surface water was present in the creek. Groundwater rose in Transect 2's Well 3, nearer the stream, 1 h after flow began and rose into Well 2, father from the stream, 3 h later. The alluvial groundwater had a negative gradient, sloping at 17% to 36% downward away from the stream (hydrologically losing; Figure 5a). Surface flow ceased on 22 August, 2 months after Well 2 went dry, and the water table in Well 3 plummeted 1 h later to confirm the groundwater's dependence on surface water.

At an example location where DW was hydrologically neutral (Transect 3), groundwater was 3.2 m below ground, or 1.8 m below the streambed, then rose 1.6 m within 12 h after surface flow started. The abrupt rise occurred in the same hour in the stream and wells. During the April–July 2019 snowmelt run-off period, groundwater elevation here was usually within 0.20 m below surface water, although it rose to 0.13 m higher for 7 days in June (gradients <2%, Figure 5b). The groundwater recession rate increased immediately after surface flow ended at the end of the summer, confirming groundwater dependence on surface water. Transects 1 and 4 responded similarly to Transects 2 and 3 (Figure S1).

In contrast, Transect 5 in the RD reach always had surface water and alluvial groundwater within 0.6–2.0 m below ground in 2018–2020. The surface water–groundwater gradient shifted from slightly losing in the baseflow season to slightly gaining during the high run-off period in 2019 (Figure 5c), and it remained losing in the drier 2020, with the gradient always less than ±6%.

Electrical resistivity imaging also indicated that Snake Creek surface water supplied alluvial groundwater. Resistivity values at the reference cross-section in RU indicated hydrologically losing conditions where infiltrating streamflow supplied shallow groundwater directly beneath the channel (Figure 6a). Surveys were conducted when the DW reach likely had no flow for 10 consecutive months and indicated that groundwater was much deeper. For example, the inferred water table across most of the Transect 2 floodplain was at least 6 m below ground, beneath the bottom of the two monitoring wells and 3 m beneath the streambed (Figure 6b). The water table appeared to be shallowest near the channel, possibly because groundwater sourced from infiltrated streamflow.

3.2 Cottonwood forest dynamics through time

3.2.1 Riparian area from historical aerial photos

The RU and RD riparian forests remained dense and continuous through time, but tree and canopy dieback in DW resulted in the riparian area fragmenting and its area being reduced (Figure 7). By 2017, the riparian area in RU and RD had increased to 103% and 110% of baseline, from 8.6 to 9.1 hectares (Figure 8). At DW, the riparian area decreased to 55% of baseline, from 10.2 to 5.6 ha. All three reaches had a temporary increase in riparian area during the wet period of 1981–1984, likely due to crown expansion. There was a strong divergence between DW area and the reference reaches after 1999, when extreme atmospheric drought began.

3.2.2 Cottonwood age structure and mortality

Censusing all cottonwoods and cross-dating of a subset of them provided estimates of the numbers of living trees by year per reach. Nearly all trees established prior to the 1946–2016 SBAI assessment period: 93% at DW, 100% at RU and 96% at RD (Figure 9). During the 2016–2017 field survey, 44% of the DW reach cottonwoods were dead, compared to 21% at RU and 18% at RD (Figure 9). This decrease in DW compared to RU and RD mirrored its greater decline in aerial photos (Figure 8). Cross-dated dead trees in DW had died between 1958 and 2017, with similar numbers standing (n = 449) and fallen (n = 454). Trees transitioned from standing to fallen approximately 15 years after death: 89% of fallen dead died before 2002 and 84% of standing dead died after 2002 (Figure 10). The largest pulse of cottonwood death occurred after 2005, soon after the extreme atmospheric drought, although our results may have captured more of the recent mortality events because older cottonwoods decompose and become unrecognizable or undatable. A smaller pulse of DW reach tree death also occurred from 1967 to 1975, 6–14 years after the pipeline was installed, although the event's magnitude is likely underrepresented because of the mortality bias described above.

3.2.3 Tree-ring growth (BAI)

Periods of relatively high and low BAI preceded a mid-2000s decline that was most severe at DW (Figure 11). The reach's four lowest growth individual years in the 117-year assessment were the last 4 years (2013–2016). Over the study period, the DW reach BAI ranged from a low of 2.41 cm² in 2015 to 11.86 cm² in 1983, with a mean of 7.42 cm² year⁻¹. The highest run-off years in the 1960–2019 flow record were 1983 and 1984, followed by 2005 and 2011; 1983 and 1984 resulted in high BAI (>11.72 cm² year⁻¹). However, modest growth in 2005 (8.52 cm² year⁻¹) and 2011 (6.25 cm² year⁻¹) indicated that the declining cottonwoods were less able to respond to favourable water conditions. The two reference reaches also had periods of high and low growth (Figure 11a); however, their post-2005 growth did not have the decline as occurred at DW (Figure 11b).

3.2.4 Stand-level cottonwood growth (SBAI)

The aerial photo approach revealed that cottonwood SBAI peaked in the DW reach at 0.52 m² km⁻¹ year⁻¹ in 1954 and again at 0.47 m² km⁻¹ year⁻¹ in the wet 1983–1984. Peaks in 1983–1984 also occurred in RU at 0.48 m² km⁻¹ year⁻¹ and RD at 0.26 m² km⁻¹ year⁻¹. We used the SBAI results normalized to prediversion baseline values to highlight changes following diversion (Figure 12a). The lowest SBAI occurred at the end of the study period in DW, with a 2013–2016 mean of 0.07 m² km⁻¹ year⁻¹ (20% of baseline growth). In contrast, RU had a less extreme decline down to 0.17 m² km⁻¹ year⁻¹ (52% of baseline), and RD growth increased to 0.21 m² km⁻¹ year⁻¹ (124% of baseline). After the 2000–2003 atmospheric drought, SBAI declined from 2003 to 2016 at 0.014 m² km⁻¹ year⁻¹ in a linear regression test for DW trees (p < 0.001), while it did not decline in RU or RD trees, trending at 0.001 and 0.000 m² km⁻¹ year⁻¹ (p = 0.79 and 1.00).

The field-based SBAI approach using a cottonwood census and cross-dating revealed similar reach-level trends as the aerial photo approach (Figure 12b). The SBAI decline at DW would have appeared even more severe relative to reference reaches if it was not partially offset by younger trees entering the population in this reach (Figure 9). SBAI in DW diverged from RD soon after the 1961 diversion but not from RU until the 2000 atmospheric drought. From 2003 to 2016, DW reach SBAI declined the same as in the aerial photo approach, at 0.014 m² km⁻¹ year⁻¹ (p < 0.001), while RU and RD still had no significant trend, declining by 0.003 and 0.002 m² km⁻¹ year⁻¹ (p = 0.43 and 0.54).

The field-based approach produced absolute annual SBAI production values for DW, RU and RD averaging 0.29, 0.48 and 0.38 m² km⁻¹ year⁻¹, while the aerial photo approach produced SBAI averaging 0.26, 0.26 and 0.20 m² km⁻¹ year⁻¹. Differences between the approaches reflect the differing assumptions and the reference reach lengths, again favouring the evaluation of trends over absolute values.

4 DISCUSSION

Tree-ring data are often used to reveal environmental patterns at the stand, forest or regional scales, but tree-based analyses provide an incomplete picture if stand-level dynamics are not considered. We used two approaches to analyse stand-level forest production to illustrate that the DW reach cottonwood forest decline was more severe than indicated by the growth rings of the surviving trees. We analysed the products of tree-level BAI along with stand-level assessments of either forest areas or field-based tree counts to upscale the tree-ring data. By doing so, we found that the riparian forest in the DW reach diverged from the reference reaches at two different times and to a greater extent than could be inferred from the tree rings of survivors alone. The first decline was soon after flow diversion began, and the second was delayed until the effects of flow diversion appear to have been amplified by a severe atmospheric drought. Delayed and heterogeneous Populus declines from drought have also been documented elsewhere (Stella et al., 2013), illustrating the negative effects of compounding stressors on terrestrial ecosystems globally (Frank et al., 2015).

Each of the two approaches to stand-level cottonwood characterization had advantages and disadvantages. The aerial photo approach had the advantage of incorporating riparian forest expansion and contraction, and it was unaffected by the wood decomposition of dead trees. However, delineation of the riparian forest area was affected by species composition changes due to conifer invasion of the valley bottom and variable photo quality through time. The field-based approach had the advantage of quantifying live trees each year by using field counts and cross-dated living and dead trees. However, this analysis was limited by the smaller study reaches necessitated for field counts and by decomposition of dead trees that resulted in undercounting of long-dead trees. Together, the similar trends derived from the two approaches provide strong evidence of forest changes through time.

Riparian forest production in the DW reach diverged from the two reference reaches at different times. SBAI for the DW reach diverged from that in RD approximately when flow diversion began in 1961. However, both DW and RU saw similar declines after 1961 and did not diverge from each other until 40 years later following extreme atmospheric drought in the early 2000s. The initial DW decline beginning in the 1960s is interpreted as being due to drought stress, while at RU, competition for light associated with a successional transition to white fir-dominated forest is a likely cause (Table 1). Higher drought stress at DW compared to RU has been demonstrated by higher stable carbon isotope ratios beginning abruptly in 1961, stronger correlation between BAI and stable carbon isotope ratios, stronger correlations relating tree-ring metrics to precipitation and flow, higher branch mortality and lower canopy vigour (Schook et al., 2020). Here, we add that the DW reach also had higher tree mortality (Figure 10) and fragmentation of the forest canopy (Figure 8). Historical forest management practices appear to have caused the tree-ring narrowing at RU. Nearly all of the RU cottonwoods established from 1876 to 1899 (Schook et al., 2020) after timber harvests opened the forest canopy. Conifers have since regrown and now shade out cottonwoods, providing an alternate mechanism for tree-ring narrowing (Cailleret et al., 2017). This demonstrates that although SBAI provides a better indication of stand-level productivity than BAI alone, it too reflects factors other than drought.

Drought-induced tree mortality has affected forests across the world in recent decades (Eamus et al., 2013; Martínez-Vilalta & Lloret, 2016; Williams et al., 2013), including the western United States where increasing temperatures have increased VPD (Martin et al., 2020). Although we document cottonwood dieback in the DW reach in the first two decades following flow diversion, extensive dieback and separation from both reference reaches did not occur until after the beginning of the 2000–2003 atmospheric drought that occurred 40 years after diversion began. The historically unprecedented VPD apparently marked the onset of a drought too severe for hundreds of DW reach cottonwoods to survive, whereas trees in the perennial reference reaches were less affected. Because high VPD increases the hydrologic gradient across a leaf surface and increases transpiration demands (Grossiord et al., 2020), trees become more susceptible to death from xylem cavitation and carbon deficit (Allen et al., 2015; McDowell et al., 2008). The DW reach cottonwood BAI most clearly decreased in 2007, however, a few years after the VPD extremes. Higher flow in 2005 and 2006 may have masked or delayed the cottonwood decline. Prior research documented narrow Populus tree rings in the year of and year following low flow (Rood et al., 2013), a tightly coupled relationship that suggests summer VPD may not be the climatic variable that best characterizes drought. Like ours, however, previous studies have also found a delayed shift to lower tree-ring growth (Stella et al., 2013), highlighting the complexity of different growing environments.

Two distinct types of drought occurred in the Snake Creek watershed. Hydrological drought appears to be the factor that differentiated cottonwood forest production in Snake Creek's DW reach relative to the RD reach. Hydrological drought can lead to gradual tree and forest decline (Cailleret et al., 2017; Schook et al., 2020) or cause immediate and widespread forest mortality (Cooper et al., 2003; Scott et al., 2000), depending upon the severity of the drought. A compounding atmospheric drought four decades after diversion likely triggered a second decline of DW riparian forest production that differentiated it from the RU reach. The 2000s and 2010s were unusually hot and dry across the western United States (Udall & Overpeck, 2017), producing conditions known to surpass the survival threshold of trees globally (Batllori et al., 2020). Although all cottonwoods along Snake Creek experienced similar early 2000s atmospheric drought, the hydrological drought imposed by flow diversion in DW led to widespread tree decline, whereas perennial flow in the two reference reaches allowed most trees to survive and maintain moderate growth. Further challenges will come as hydrological and atmospheric droughts are forecasted to continue along rivers region wide (Miller et al., 2021; Milly & Dunne, 2020). Our findings from Snake Creek demonstrate that a stand-level view can reveal how compounding stressors affect ecosystem stability, and how local management actions can mitigate or exacerbate extreme weather events and global climate change.