River channel response to invasive plant treatment across the American Southwest

Invasive riparian plants were introduced to the American Southwest in the early 19th century and contributed to regional trends of decreasing river channel width and migration rate in the 20th century. More recently, efforts to remove invasive riparian vegetation (IRV) have been widespread, especially since 1990. To what extent has IRV treatment reversed the earlier trend of channel narrowing and reduced dynamism? In this study, paired treated and untreated reaches at 15 sites along 13 rivers were compared before and after IRV treatment using repeat aerial imagery to assess long‐term (~10‐year) channel change due to treatment on a regional scale across the Southwest USA. We found that IRV treatment significantly increased channel width and floodplain destruction. Treated reaches had higher floodplain destruction than untreated reaches at 14 of 15 sites, and IRV treatment increased the rate of floodplain destruction by a median factor of 1.9. The effect of treatment increased with the stream power of the largest flow over the study period. Resolving observations of channel change into separate measures of floodplain destruction and formation provided more information on underlying processes than simple measurements of channel width and centerline migration rate. Restoration practitioners who perform IRV treatment projects often focus on wildlife or vegetation response; however, geomorphic processes should be considered in restoration planning because they drive aquatic habitat and vegetation dynamics, and because of the potential for damage to downstream infrastructure. Depending on the restoration goal, management practices can be used to enhance or minimize the increase in channel dynamism caused by IRV treatment.

IRV has been targeted for treatment in the SW as a part of environmental restoration efforts since the 1960s (González et al., 2015;Shafroth et al., 2005), with treatment efforts increasing exponentially since 1990 (Bernhardt et al., 2005). Goals of river restoration practices that remove IRV generally focus on improving wildlife habitat or promoting native riparian plant species (Shafroth et al., 2005(Shafroth et al., , 2013. IRV treatment may also be carried out to increase river flow, but this approach may not be feasible (Hart et al., 2005). Treatment of the dominant invasive vegetation can decrease hydraulic roughness and bank stability, at least temporarily, and can lead to channel widening and increased rate of channel migration (Jaeger & Wohl, 2011;Keller et al., 2014;Vincent et al., 2009). It is unknown, however, whether this is a frequent and widespread response. IRV exerts fundamental controls on river morphodynamics through mechanical and hydraulic processes acting at the streambank (Bertoldi et al., 2014;Martínez-Fernández et al., 2018). Treatment of woody plants promotes erosion by reducing root reinforcement of banks and reducing roughness, which increases flow velocity and shear stress (Griffin et al., 2005;Pollen-Bankhead et al., 2009). Where vegetation treatment is followed by flooding, large increases in channel area and migration rate may result (Vincent et al., 2009). A more mobile channel may be seen as a return to the condition prior to arrival of IRV and may benefit native disturbance-dependent species Sher et al., 2002). On the other hand, increased sediment fluxes from IRV treatment may fill reservoirs, and increased channel migration rates may put inhabited or cultural areas at risk (Jaeger & Wohl, 2011) or remove valued patches of existing vegetation.
While IRV treatment within river corridors has continued to increase along with overall channel restoration efforts, the post-IRV treatment monitoring has not kept pace (Rubin et al., 2017). Restoration goals and post-eradication of IRV monitoring efforts overwhelmingly focus on vegetation (González et al., 2017a(González et al., , 2017bHarms & Hiebert, 2006) or wildlife (Sogge et al., 2008;Valente et al., 2019) response; however, hydrological and geomorphic processes are key drivers of aquatic habitat and vegetation dynamics in riparian ecosystems . At the Escalante River, Scott et al. (2018) found that the shade-tolerant invasive shrub Russian olive grew in native Fremont cottonwood (Populus fremontii) understories, and modified channel morphology by trapping sediment and forming levees. Russian olive control efforts along the Escalante River have occurred for approximately 10 years, but channel morphological response to treatments is still not well understood. After Russian olive treatment along Chinle Creek in Canyon de Chelly National Monument, Arizona, Pollen-Bankhead et al. (2009) and Jaeger and Wohl (2011) found that channel widening occurred. IRV treatment may introduce significant sediment contributions to Chinle Creek, but large floods would be needed to shift channel morphology from the present meandering condition back to pre-invasion braided form. Along the Rio Puerco and San Rafael River, channel movement, channel area, and sediment supply all increased where treatment of tamarisk was followed by flooding (Keller et al., 2014;Vincent et al., 2009). These isolated studies show that treatment can lead to increased channel width and mobility in some cases, but there have been no regional studies investigating the likelihood of these channel morphological responses or the factors promoting or resisting it.
This study focuses on three dominant introduced floodplain plant species found in the SW: tamarisk, Russian olive, and giant cane.
Diverse methods to treat these species exist, including biological, mechanical, chemical, grazing, burning, flooding, and integrated control methods (Shafroth et al., 2010). Tamarisk and Russian olive are commonly targeted using chemical, mechanical, or biological methods (Zavaleta, 2000). The tamarisk beetle (Diorhabda spp.) was released in 2001 to control tamarisk and has increased plant mortality by defoliation (Pattison et al., 2011). Giant cane spreads downstream by rhizome dispersal (Gilbert & Wilcox, 2021), and plants resprout easily after mechanical cutting (Briggs et al., 2021). Therefore, commonly used treatment methods for giant cane include application of herbicide alone, as well as integrated methods of mechanical cutting or prescribed fire with follow-up herbicide application (Briggs et al., 2021).
We focus our analysis on mechanical and chemical methods of IRV treatment. Mechanical treatment includes the use of heavy machinery to remove the entire plant (whole-plant, WP), or hand/ chain saws to cut the plant to its stump (CS). The CS method is commonly used to treat IRV because it can be more cost efficient than WP (Shafroth et al., 2010). Helicopter herbicide application (HH) was included in this analysis as a chemical method to treat IRV, considering its use over large spatial extents. Biocontrol sites were not examined because the dispersal of treatment organisms makes the identification of paired treatment and untreated control sites difficult.
We tested the hypothesis that IRV treatment increases streambank erosion and channel mobility. Because response varies strongly across rivers, we investigated a number of different rivers (15 study sites on 13 rivers across six states). We paired each treated site with a nearby untreated site along the same river to distinguish the effect of IRV treatment from those of other local factors. We quantified channel change at a broad spatial scale using aerial imagery collected before and after IRV treatment to meet the following objectives: (i) quantify the response of river channels to IRV treatment; and (ii) determine the influence of flood magnitude on the response to IRV treatment.

| Study region and reaches
Our 15 study sites are located along 13 rivers across six states in the SW region: California, Arizona, Colorado, Texas, New Mexico, and Utah ( Figure 1). All of the rivers we selected are alluvial, with predominantly sand-sized sediment comprising channel beds. Each site included a treated reach and a nearby untreated reach (Table 1). IRV treatment data were compiled from published literature (Barz et al., 2009;González et al., 2017a;Harms & Hiebert, 2006;Hart et al., 2005), databases of restoration organizations (RiversEdge West, the Escalante River Watershed Partnership), and through personal communication. Study reaches were located based on the following criteria: IRV abundance was strongly reduced in a clearly documented reach length large enough to affect channel dimensions (>1 km); a nearby comparable untreated reach was present; and high-resolution, repeat aerial imagery was available. To avoid pseudoreplication we chose not to inflate sample size by sampling multiple nearby sites on the same river. Untreated reaches were selected to be similar in length to treated reaches, and, where possible, were located upstream of treated reaches ( Figure 2) to avoid transport of sediment eroded from IRV treatment into the untreated reach. Treatment was not always complete throughout the treated reach; for example, at two sites (Rio Grande NM, Colorado River CO), treatment occurred on only one side of the channel. At one site (Escalante River), the untreated reach contained a small amount of treatment.

| Aerial imagery collection
Treated and untreated reaches were compared before and after treatment using repeat aerial imagery to assess long-term ($11-year average, Table 3) channel change due to treatment on a regional scale across the SW. Imagery was obtained through the USGS Earth Explorer and consisted of high-resolution orthoimagery (HRO), from the National Agriculture Imagery Program (NAIP), and multispectral aerial imagery (MAI). Imagery collected nearest to the year prior to treatment efforts was used to delineate a pre-treated channel, and the most recent imagery that was available for a location was used to delineate the post-treated channel (Table 3). Imagery resolution ranged from 0.2 to 1 m (Table 3). Although all imagery obtained was georeferenced, georeferencing errors due to image warping (Fryer & Brown, 1986), issues with georeferenced control points (Hughes et al., 2006;Mount et al., 2003), or image quality may cause uncertainty in channel measurements (Donovan et al., 2019).

| Change metrics
We imported aerial imagery into a geographic information system (ArcGIS Pro, version 2.6.0) to digitize stream channel boundaries at the active-channel shelf level (Osterkamp & Hedman, 1982), a surface indicated by the lowest occurrence of established perennial vegetation. The active channel shelf level, which is generally lower than the bankfull level (Osterkamp & Hedman, 1982), has proved to be useful  (Table 4).
Channel centerlines were generated to examine migration using the polygon to centerline tool in ArcGIS Pro, with occasional manual adjustments to correct for in-channel features such as mid-channel bars and vegetated islands. If multiple channels were present, the was divided by the length of the centerline for A migration (w migration = A migration /L migration ;

| Statistical and hydrologic analyses
In a paired design (n = 15 treated/untreated pairs), we analyzed the effect of IRV treatment on floodplain destruction, floodplain formation, width difference, and channel centerline migration using the Wilcoxon signed rank test (Wilcoxon). This test was selected instead of the traditional paired t-test because the effects data were not normally distributed (Rey & Neuhauser, 2014). The significance threshold for Wilcoxon tests was set at p = 0.05.
To explore river channel change with respect to flood disturbance (objective ii), we estimated the highest instantaneous unit stream power for each site between the dates of the pre-and post-treated imagery. We used the peak flood under the assumption that the most geomorphic change is associated with this large flood (Tooth, 2013).
We acquired instantaneous annual peak streamflow data from the USGS for the stream gaging station that best represented each treatment site (Table 2). We found the highest flow peak in the years between the pre-and post-treated imagery (Tables 2 and 3). Four simple linear regressions were run on each change metric using log of unit stream power as the predictor variable (Table 6). Regression models used all treated and untreated data combined.
Unit stream power was calculated to evaluate the energy available for geomorphic change across sites (Merritt & Wohl, 2003): Conceptual diagram of change metrics measured and calculated using ArcGIS pro. Yellow and green polygons are generated using the symmetrical-difference tool in ArcGIS pro by overlaying preand post-treated polygons to quantify floodplain destruction and floodplain formation in treated and untreated reaches. Blue polygons are generated using the feature-to-polygon tool in ArcGIS pro and summing the area between the pre-and post-treatment channel centerlines to quantify centerline migration. Channel width is measured by dividing channel area by channel length. [Color figure can be viewed at wileyonlinelibrary.com] T A B L E 2 Peak stream discharge information obtained from USGS stream gaging stations near study sites. Unit stream power was calculated using the peak discharge, channel slope (Table 1), and pre-treated average channel width where ω is the unit stream power (W m À2 ), γ is the unit weight of water (9800 N m À3 ), Q is peak discharge (m 3 s À1 ), s is the channel slope (m m À1 ) approximated using USGS topographic maps within ArcGIS Pro, and w pre is the pre-treatment channel width (m).  Figures 4c and 5). In contrast, the median ratio of floodplain formation for treated to untreated reaches was 1.08, indicating no overall effect of IRV treatment on this metric. The median ratio of migration was 1.69, and ranged from 0.12 (Rio Grande NM site) to 21.81 (Rio Puerco site) ( Figure 5). We did not detect consistent differences among treatment methods, in part because of the T A B L E 3 Aerial imagery information used in the analysis of change metrics, including imagery source, spatial resolution, and the year the imagery was collected  width difference (p = 0.013, Table 6).

| DISCUSSION
This study quantified river channel response to IRV treatment at the decadal scale using change metrics of floodplain destruction, floodplain formation, channel width difference, and channel migration. We Floodplain formation did not differ significantly between treated and untreated reaches (p = 0.934, Table 5), and the median ratio for treated to untreated reaches was close to 1.0 (Figures 4 and 5). The relationship between stream power and formation was not statistically significant ( Table 6) There was an overall weak positive effect of stream power on width difference for treated and untreated reaches (Table 6). Higher stream powers are generally correlated with increases in channel width (Schumm & Lichty, 1963), but the presence of vegetation in the untreated reaches limits channel widening by increasing bank stability.
Moreover, pre-treated minus post-treated channel width differences were mostly negative in the untreated reaches, suggesting that processes such as vegetation establishment on in-channel deposits, and the hydrological and mechanical effects of plants, are contributing to increased bank stability and channel narrowing.
More centerline migration appeared to occur in the treated reaches compared to untreated (median ratio of 1.69, Figure 4), but this effect was not statistically significant (Table 5). Increasing channel migration following IRV treatment was expected because, when vegetation is treated, the additional strength provided by roots decreases, leaving banks unstable and prone to erosion (Pollen-Bankhead et al., 2009;Vincent et al., 2009). On the other hand, narrowing of many channels in the absence of high flows resulted in centerline migration without much floodplain destruction, especially in untreated channels.
T A B L E 6 Linear model results for change metrics using stream power as the predictor variable. Stream power and change metrics were log transformed to meet model assumptions The centerline for the untreated reach at the Santa Clara site was not mappable due to a multithreaded channel system, so the Santa Clara site was not considered in this statistical test. *Significant differences at the 0.05 level.
The power of the largest flow in the study period was correlated with both floodplain destruction and channel width (Table 6) F I G U R E 4 Floodplain destruction (a), floodplain formation (b), width difference (c), and migration (d) for treated (black) and untreated (green) reaches by study site.
[Color figure can be viewed at wileyonlinelibrary.com] The threshold for river channel change could differ for sites where giant cane is treated. Compared to tamarisk and Russian olive, giant cane has a lower root tensile strength, but a denser root system at shallower depths, influencing the type of bank failure that may occur (Stover et al., 2018). More specifically, during moderate to high flows, banks occupied by giant cane can fail by undercutting and slumping (Stover et al., 2018). Therefore, more change through lateral instability may be expected where channels are lined with giant cane, depending on the height of the streambank, bank sediment composition, and other local factors. We did not observe this effect in our dataset, most likely due to small sample size.
IRV treatment projects that use techniques such as the WP method target both the aboveground and belowground plant biomass, and cause a high level of disturbance (Shafroth et al., 2010). We suspect that, in comparing treatment methods, increased channel change will result from more disruptive treatment methods (Jaeger & Wohl, 2011).
However, the data in this study were not sufficient to test the relationships between IRV type, treatment method, and river channel response.
Additional information on site-specific conditions at study sites and study reach replicates are needed to further evaluate these relationships. Furthermore, this study correlated change metrics with peak unit stream power, which overlooks other components of the flow regime, antecedent channel conditions, and geological floodplain constraint, all of which may have a strong influence on channel change.

| MANAGEMENT IMPLICATIONS
This study shows that IRV treatment causes a large and consistent increase in erosion across a broad range of rivers. In river restoration, increased erosion following IRV treatment is a benefit in some cases (Pollen-Bankhead et al., 2009) and a cost in others (Barz et al., 2009;Gilbert & Wilcox, 2021;Jaeger & Wohl, 2011;Vincent et al., 2009); therefore, IRV treatment projects would benefit from a geomorphic restoration goal that answers the question, "How much of an increase in channel activity is desired?" Setting this goal is based on consideration of desired process changes, downstream threats, and susceptibility of the system to change (Figure 6). Desired process changes may be explicit; for example, land managers may desire a return to a more dynamic channel that characterized the system prior to arrival of the IRV. In addition, desired process changes are often implicit in the list of species the practitioner is interested in promoting or impeding.
For example, many native plant species (e.g., willow) are dependent upon physical disturbance and channel change. On the other hand, some invasive species (e.g., tamarisk) are also favored by channel change, while others are not (e.g., Russian olive; Katz & Shafroth, 2003). Refining our understanding of the geomorphic effects of IRV treatment requires more data on decadal-scale geomorphic response.
Long-term monitoring of channel geometry as illustrated in this study is inexpensive and would help to provide the needed information. In addition, practitioners should consider designating an untreated reach to evaluate treatment geomorphic effectiveness, as long as this does not pose an unacceptable reinvasion risk. Lastly, we recommend increased availability of treatment data to promote analysis of channel and floodplain response to treatment.

| CONCLUSIONS
We investigated channel morphological change associated with IRV treatment at 15 sites along 13 rivers across the American Southwest using repeat aerial imagery over $10 years. We paired treated and untreated channel reaches and found that 14 of the 15 sites analyzed showed higher floodplain destruction at the treated reaches compared to untreated reaches. IRV treatment increased the rate of floodplain destruction by almost a factor of 2. Floodplain destruction also increased with the stream power of the largest flow during the study period. Resolving observation of channel change into separate measures of floodplain destruction and formation provided more information on underlying process than measurements of channel width and centerline migration rate. We suggest a management framework to assist restoration practitioners in determining how much of an increase in channel activity is desired when IRV is removed based on consideration of desired process changes, downstream threats, and susceptibility of the system to change.

ACKNOWLEDGEMENTS
We thank RiversEdge West for providing spatial data for treated reaches at the Dolores River site, and Jeff Renfrow for help with finding suitable treated and untreated reaches at the Rio Grande TX site.
Brian Cade and Ann Hess provided useful suggestions on the statistics. Financial support was provided by the US Geological Survey, Geological Society of America, and Colorado Riparian Association.
Eduardo Gonzalez and two anonymous referees provided thorough, constructive reviews. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the US Government. F I G U R E 6 Determining a geomorphic restoration goal: How much of an increase in channel activity is desired when invasive riparian vegetation (IRV) is removed? This determination is based on consideration of desired process changes, downstream threats, and susceptibility of the system to change. Susceptibility to change is a function of the hydrologic regime and channel characteristics. For example, if threats and susceptibility to geomorphic change are both high and there is not an interest in promoting the occurrence of disturbance-dependent native vegetation, then the practitioner may choose to minimize any increase in channel activity resulting from IRV treatment using techniques described in the text.