Riparian vegetation shade restoration and loss effects on recent and future stream temperatures

Abstract

River temperatures are expected to increase this century harming species requiring cold-water habitat unless restoration activities protect or improve habitat availability. Local shading by riparian vegetation can cool water temperatures, but uncertainty exists over the scaling of this local effect to larger spatial extents. We evaluate this issue using a regional spatial stream network temperature model with covariates representing shade effects to predict mean August stream temperatures across 78,195 km of tributaries flowing into the Columbia River in the northwestern US. We evaluate nine scenarios predicting stream temperatures for three riparian shade conditions (current, restored, and no riparian vegetation) within three different climate periods (2000s, 2040s, and 2080s). Results suggest riparian shade restoration (2000s climate) could decrease mean August stream temperatures by 0.62°C across the study network. Under the same restored shade conditions, temperature predictions for tributaries at their confluence with the Columbia River range from 0.02–2.08°C cooler than under current shade conditions. The climate warming effect predicted for the 2040s and 2080s, however, is greater than the cooling effect from restoring riparian shade. Streams less than 10m bankfull width cooled more frequently with riparian shade restoration. In Oregon, the proportion of fish habitat for salmon and trout rearing and migration that meet temperature numeric water quality criteria could be increased by 20% under restored shade conditions although net habitat declines may still occur in the future. We conclude riparian vegetation restoration could partially mitigate future warming and help maintain cold-water habitats that function as thermal refuges if implemented strategically.

Introduction

Ecological forecasting is an important part of managing environmental resources (Clark et al. 2001). Predicting future conditions and the uncertainty around those conditions provides a baseline for measuring the success and effectiveness of restoration activities (Palmer et al. 2005). Therefore, scenario planning is a useful exercise to first identify potential benefits that alternative actions may have on a target ecosystem prior to initiating restoration (Peterson et al. 2003). Temperature is an environmental variable that plays an important mechanistic role in many ecosystem and organismal processes (Caissie 2006) and is also a climate variable predicted to change in the future (IPCC 2013). Consequently, environmental temperature is an important ecological forecasting metric for evaluating the potential risk to species and their habitat.

Ectothermic organisms, whose distributions and behaviors are directly tied to ambient temperatures, can be disproportionately affected by changes in thermal regimes (Buckley et al. 2012). Aquatic communities are often dominated by ectotherms and therefore are highly sensitive to any changes in water temperature. Pacific salmon and char are freshwater ectotherms and marquee cold-water habitat specialists that complete part of their life cycle in freshwater river networks (Isaak et al. 2018). A key stage of their life cycle in river networks includes upstream movements, often during warm summer seasons, to headwater streams where spawning and juvenile rearing occurs. During these migrations, continuous or patchy distributions of cold-water habitat are necessary for these species to survive and reach their spawning habitat (Strange 2010; Snyder et al. 2020). The thermal regimes for networks within Pacific salmonids’ ranges have changed due to human landscape modifications, damming of rivers, and altered climate patterns (Fullerton et al. 2015; Isaak et al. 2018). Consequently, the availability of cold-water habitat may already be reduced and may be more apt to shrink in the future. Because of these impacts, it is important to understand where cold-water habitats exist now, in the future, and what potential restoration actions may have for their maintenance.

Management of stream temperatures is important in the Columbia River basin in the northwestern United States due to its historically large and now reduced populations of migratory Pacific salmon that require cold temperatures during warm summer months (EPA 2021a; EPA 2021b). The states of Oregon and Washington have Clean Water Act water quality standards (WQS) for temperature to protect these and other cold-water fishes. In Oregon, for example, the State’s temperature WQS includes a provision (Oregon Rule 340–041-0028(4)(d)) requiring water bodies to “have cold water refugia that are sufficiently distributed so as to allow salmon and steelhead migration without significant adverse effects from higher water temperatures elsewhere in the water body”. The Oregon “refugia” provision is an example of the need to protect and restore the spatial thermal heterogeneity and understand the dynamics and drivers of cold-water habitats across the landscape. Cold-water habitats can manifest at tributary junctions as localized cold-water plumes within the warmer mainstem Columbia River (EPA 2021b) or as cold reaches within a tributary network driven by a combination of biological and physical landscape characteristics (EPA 2021a). Consequently, a primary goal for cold-water species conservation managers within the Columbia River basin is to understand the spatial distribution of cold water across a riverscape and to strategically implement restoration efforts at sites with the greatest potential to reduce stream temperatures.

Solar radiation has long been recognized as a dominant component of stream thermal budgets (Brown & Krygier 1970; Johnson 2004). Therefore, reducing the solar heat flux reaching a stream should cause cooling, an outcome that has been successfully demonstrated at the reach-scale by restoring riparian vegetation and increasing shade (Nusslé et al. 2015). However, broader implementation of riparian restoration across large spatial extents (regions or entire drainage basins that often encompass thousands of stream kilometers) has not been attempted, is potentially expensive, and has unknown potential to mitigate warming patterns associated with climate change.

Since experimental manipulation of riparian shade at very large spatial extents (10⁵ km²) is generally unrealistic, predictive modeling exercises are the best tools available to assess the magnitude of effect from large restoration efforts (Seixas et al. 2018). Previous modeling efforts investigating the relative impact of shade on stream temperatures have taken different approaches. One approach measures shade restoration potential by comparing past riparian shade conditions with current disturbed or reduced riparian shade conditions (Seixas et al. 2018). Other studies focused on identifying the relative influence different types of riparian vegetation have on shade and the resulting temperature implications (Dugdale et al. 2018). Johnson & Wilby (2015) investigated the combined topographic and riparian vegetation shade impact on stream temperature across river reaches. Still others have focused on the spacing and location of shaded reaches within a network to identify what constellation of shaded reaches and riparian buffer widths may be necessary to reduce network-wide stream temperatures due to downstream heat propagation (Rutherford et al. 2004). Research efforts that combine several or all of these goals are less common and rare at large spatial extents (landscape area ~10⁵ km² or stream length ~10⁴ km).

We designed a geostatistical modeling approach that uses several different reach-shade components to assess how manipulating reach shade conditions may affect stream temperature under current and future climate periods at large spatial extents (~10⁵ km²). We applied this modeling approach to the tributary networks of the Columbia River downstream of its confluence with the Snake River in the Pacific northwest United States. Five primary questions were the focus of our study.

1. What is the potential of riparian shade restoration to decrease stream temperatures under recent and future climate periods?

2. How do stream temperatures vary across the landscape under the maximum and minimum riparian vegetation shade conditions when compared with the current riparian shade condition under recent and future climates?

3. What size stream reaches are related to the largest thermal response from riparian restoration?

4. How much does extensive riparian shade restoration or loss affect tributary outflow temperatures into the mainstem Columbia River migration corridor?

5. What proportion of fish habitat in OR and WA (for the fish uses and temperature criteria evaluated) meet State temperature numeric water quality criteria under different riparian shade conditions and climate periods?

Methods

Study system

This study was conducted in the Pacific Northwest region of the United States (Figure 1) and focused on tributary networks of the Columbia River downstream of its confluence with the Snake River (Figure 1A). These networks drain approximately 123,038 km² and were split into two processing units (Oregon Coast – OC: 45,788 km²; Mid-Columbia – MC: 77,250 km²) sensu Isaak et al. (2017) that have distinct physiographic characteristics (Figure 1A and Supplement S1). The OC is adjacent to the Pacific Ocean on the west side of the Cascade Mountains with a slightly cooler and wetter climate than the MC region (OC mean August air temperatures 14.4–17.7°C and mean annual precipitation 523–4149 mm). The MC is east of the Cascade crest and has a more continental climate (mean August air temperature of 14.8–18.7°C and mean annual precipitation of 207–3251 mm). Topographically, the OC sits lower in elevation (0–1589 m) than the MC (23.5–2369 m). Due to these differences and those in land cover (Tables S1 and S2), we fit temperature models separately for each processing unit.

Figure 1.

Study region (A) encompassing the Columbia River tributaries downstream of the confluence with the Snake River. Highlighted, for each processing unit region (OC: Oregon Coast; MC: Mid-Columbia), are the temperature sensor locations (points) for observed temperature data and (B) the locations of the confluences of the tributaries (stars) to the mainstem Columbia River.

After excluding the mainstem of the Columbia River, the tributary networks in the study area extend for approximately 78,195 km (OC: 30,946 km; MC: 47,248 km; length estimates derived from the National Hydrography Dataset Plus Version 1 [NHDPlus V1 – McKay et al. 2010]). Temperatures in these networks were previously modeled using spatial stream network (SSN) models in the NorWeST project and predictions were limited to perennial reaches that were accessible to fish species as described in Isaak et al. (2017). The same networks were used in our SSN models.

Along the mainstem of the Columbia River, 198 tributary confluences (OC: 116; MC: 82) had August discharge based on National Hydrography Dataset Plus Version 2 (NHDPlus V2) Enhanced Runoff Method attributes (McKay et al. 2012) (Figure 1B; Supplement S2). Most (~83%) of these tributaries have minimal mean August discharges (< 0.5 m³/s) and only ~11% of tributaries have a mean August discharge > 1 m³/s (Table S3). Larger tributaries to the mainstem Columbia River can create cold-water plumes at their confluence when they are cooler than the mainstem Columbia River. These cooler habitats have been observed as important stepping-stone habitats for migrating salmon (Keefer et al. 2009). The potential for cold-water habitat generation along the mainstem Columbia River migration corridor is a reason we focused on tributary outflow temperatures at their confluence with the mainstem Columbia River (study question 4).

Stream temperature observations

Our study focused on mean August stream temperatures for two reasons. First, August is the summer period with the longest duration of warm-water temperatures in our study system and, second, critical life history stages for species with cold-water habitat requirements, such as Pacific salmonids migrating to their headwater spawning habitat, also occur in August. Conveniently, August is also the summer month with the most temperature data available for developing models. Temperature data were obtained from the NorWeST project website database (https://www.fs.fed.us/rm/boise/AWAE/projects/NorWeST.html) and included 10,129 observations of mean August stream temperature (OC: 3,140; MC: 6,989) across the study area from 3,336 independent sites (OC: 1,206; MC: 2,130) between 1993 and 2011 (Figure 1A).

SSN model structure

We used spatial stream network (SSN) models (Peterson & Ver Hoef 2010) to predict mean August stream temperatures across our study tributary networks (Figure 1A). SSN models differ from traditional generalized linear mixed models because they include autocovariance functions (as random effects) that explicitly quantify the spatial autocorrelation associated with dendritic river network branching structure and the heavy influence of downstream flow. These accommodations make them ideal models for examining data collected within stream networks and especially for variables that have strong dependencies on flow (e.g., stream temperature) (Isaak et al. 2014).

Rather than develop entirely new SSN models for our study region, we modified previously published SSN temperature models from the NorWeST project (Isaak et al. 2017). The original NorWeST models underwent a rigorous model selection process that involved selecting the best set of covariates that also covered the entire western conterminous United States. Consequently, these models are well suited to predict stream temperature across large spatial extents as in our study. The original NorWeST covariates are categorized into two main types (spatial or temporal) to provide static (physical landscape characteristics) as well as dynamic (climate patterns) inputs to the SSN model to account for interannual variability in temperature. The NorWeST models had 10 spatial covariates including site channel elevation, reach slope, proportion of upstream watershed covered by lakes (or glaciers), mean annual precipitation, latitude, baseflow index, upstream drainage area, hypolimnetic dam tailwater presence, and percent riparian canopy coverage (Table 1). There were two temporal covariates to describe the interannual variability in the observation data that included mean August air temperature and mean August discharge.

Table 1.

Model covariates for predicting mean August stream temperature. All parameters except “Reach shade” are original to NorWeST SSN models (Isaak et al. 2017).

Parameter	Abbreviation	Units	Covariate data source
Mean August air temperature	air	°C	Dynamically downscaled NCEP RegCM3 reanalysis (Hostetler et al. 2011; http://regclim.coas.oregonstate.edu/index.html) (15-km or 50-km grid)
Mean August stream discharge	discharge	m3/s	Averaged across reference USGS flow gages with long-term records and minimal water abstraction or storage reservoirs (http://watersdata.usgs.gov/nwis/rt)
Elevation	elevation	m	National Elevation Dataset with NHDPlus V2 (30 m grid)
Latitude	latitude	m	Derived by snapping agency coordinates to NorWeST stream network
Reach shade	shade	%	DOI 2014 and NHDPlus V2 databases. See SSN model structure and covariates and Supplement S4. Shade covariate development for complete details.
Cumulative drainage area	drainage	km2	NHDPlus V2 (McKay et al. 2012)
Reach slope	slope	%	NHDPlus V2 (McKay et al. 2012)
Mean annual precipitation	precipitation	mm	NHDPlus V2 (McKay et al. 2012, based on PRISM 1971–2000)
Base flow index	bfi	unitless	https://water.usgs.gov/GIS/dsdl/bfi48grd.zip
Glacier proportion	glacier	%	Fountain et al. 2006; http://glaciers.research.pdx.edu/Downloads (1:100,000)
Lake proportion	lake	%	NLCD (MRLC-NLCD 2011) in NHDPlus V2 (McKay et al. 2012)
Tailwater	tailwater	unitless	Binary variable indicates whether a stream temperature site was in a reach downstream of a deep reservoir that is anomalously cold due to releases of hypolimnetic waters

Our study used the same NorWeST SSN model structure (same fixed effects covariates listed above and autocovariance function types [exponential tail-up, tail-down, and Euclidian distance]), but used an alternative riparian shade covariate. We substituted the NorWeST “Riparian Canopy” covariate (percent canopy cover based on the National Landcover Dataset (NLCD) 2011 “Canopy Cover” data product that was averaged across 1-km reaches) with another riparian shade estimate from the “Shade.xls” model (available at: ecology.wa.gov/Research-Data/Data-resources/Models-spreadsheets/Modeling-the-environment/Models-tools-for-TMDLs). Just as in the NorWeST SSN models, we split our study area into two processing units to accommodate regional variation across our large study extent and fit SSN models separately for the MC and OC regions. We included 69,961 prediction sites (OC: 28,008; MC: 41,953) from the NorWeST database that are spaced at ~1-kilometer intervals. A detailed comparison of the original NorWeST and our SSN model structures is available in Supplement S3. All data analysis was conducted in R v 4.0 statistical software (R Development Core Team, 2020) and model fitting used the “SSN” package (Ver Hoef et al., 2014).

Riparian vegetation shade

We used riparian vegetation reach shade estimates from the Shade.xls model instead of the NorWeST Canopy Cover covariate in our models. This was necessary to have a more realistic shade covariate that could be manipulated for assessing shade condition scenarios. Moreover, the NorWeST Canopy Cover covariate does not account for topographic shading (shade due to channel morphology, channel aspect, and/or hillslope/terrain shade angles), and is not a direct measure of reach shade, but an estimate of how much canopy is available to shade a reach. In contrast, the Shade.xls covariate more directly estimates the role riparian vegetation plays in shading the channel and incorporates components of topographic shading for us to leverage (Supplement S4).

The Shade.xls model has been used since the early 2000s in Total Maximum Daily Load (TMDL) development by Idaho, Oregon, and Washington State agencies. Algorithms in the Shade.xls model have been modified from Chen et al. (1998) and inputs were derived from opensource datasets (Supplement S4). The shade covariate represents the percent of the channel width that is shaded where 0% indicates no shade and 100% indicates the entire channel cross section is entirely shaded over the course of a day. We estimated mean August stream shade at each observation and prediction site for August 1^st. We used the 2014 LANDFIRE spatial data products named “Existing Vegetation Canopy Cover” and “Existing Vegetation Height” (DOI 2014) for inputs in the development of our current shade covariate from the Shade.xls model. See Supplement 4 for additional details on stream shade calculation methods, software, and data requirements.

Three different shade conditions were developed by modifying our current reach shade covariate. These three shade conditions are: (A) topographic shade, (B) current vegetation shade plus topographic shade (hereafter referred to as “current shade”), and (C) restored vegetation shade plus topographic shade (hereafter “restored shade”). Because topographic shade is integrated in all three shade scenarios, the difference between each shade condition is driven by riparian vegetation. Topographic shade is most relevant in mountainous areas and represents a condition with all riparian vegetation removed from the zone of influence and therefore is the minimum shade possible for a reach or a “worst-case” shade scenario for stream cooling. This is representative of reaches following intense forest fires that may remove all riparian vegetation along streams. The current riparian shade condition was described in the “SSN model covariates” section (also see Supplement S4) and is derived from existing vegetation height and canopy cover LANDFIRE data sets (DOI 2014). The restored riparian shade condition is partially derived from the “Environmental Site Potential” vegetation classification LANDFIRE product and represents late-successional or climax vegetation communities that would become established across the landscape if the system developed in the absence of anthropogenic disturbance (DOI 2014). Published temperature TMDL documents for rivers within the study area were also reviewed for any expected riparian vegetation height or community statistics for restored riparian vegetation conditions. Restored riparian condition, therefore, represents the maximum vegetative shade condition at each site in our study region. We compared these various shade conditions by looking at difference maps to identify regions with many reaches that have low current riparian vegetation that could see large vegetation shade increased from restoration, as well as locations that have lots of current shade that could be lost due to riparian vegetation degradation.

SSN model fit diagnostics

The SSN model fits for each processing unit were compared using Leave One Out Cross Validation (LOOCV) R² and Root Mean Square Prediction Error (RMSPE). The LOOCV R² statistic describes how much of the observed data variance the model explains while the RMSPE statistic is the average absolute model prediction error in degrees Celsius. Additionally, the percent variation explained by fixed effect covariates and autocovariance functions (random effects) were evaluated for overall model fit behavior and to determine how spatially structured the observed temperature data were in each model. It is common for fixed effects to explain small percentages of the variance when SSN models are fit to dense temperature datasets (Detenbeck et al. 2016; Isaak et al. 2017) because of spatial redundancy among observations (Sály and Erős 2016; Marsha et al. 2018), which results in the model’s autocovariance function explaining most of the variation. This does not suggest the fixed effects are not important in the model but instead provides an indication of the importance of using a model that accounts for spatial autocorrelation so that fixed effects are estimated accurately. Finally, the relative importance of individual covariates was assessed by comparing the standardized coefficients (magnitude of their absolute value) of each fixed effect within a processing unit. Larger absolute value standardized coefficients indicate a greater influence of that covariate in explaining observed temperature variability and also a greater influence on temperatures when using the model for prediction scenarios.

SSN model prediction scenarios

Once stream temperature datasets were fit, we used the SSN models to predict mean August stream temperature across the entire study network. We made predictions for nine different scenarios generated from the pair-wise combinations of three riparian shade conditions with three climate periods. Different climate periods for the prediction scenarios were represented in the same way the NorWeST SSN temperature models predicted future stream temperatures (Isaak et al. 2017). Changes in mean August air temperatures and stream flows based on the A1B trajectory were multiplied by their respective parameters in the SSN models and the resulting water temperature deltas were added to the temperatures in recent baseline conditions. The air temperature and flow predictions for two future climate periods were average values from a suite of ten global climate change models encompassing the decade of interest in each climate period (the 2040s and 2080s; Hamlet et al. 2013). For the 2040s climate period, the average climate values characterized the period from 2030 to 2059 and for the 2080s period from 2070 to 2099. The third climate period is derived from an historical average of mean August air temperature and stream discharge from 1993 to 2011 (as in the NorWeST model – Isaak et al. 2017). We refer to this historical climate period as the “recent” or “2000s” decade climate period, and it represents our understanding of near-present conditions. Summary analysis of the temperature predictions for each of the nine scenarios focused on two spatial extents: 1) network-wide for the entire study area and 2) tributary outflow points at their confluence with the mainstem Columbia River.

Network spatial extent prediction summary

The first spatial extent characterized the network-wide response for stream temperatures among scenarios and helps address our first three study questions. Network-wide summary statistics included the mean stream network temperature and the mean difference in network-wide stream temperatures between prediction scenarios. Difference maps among scenarios provide spatial patterns of reach temperature change as a result of riparian vegetation restoration or loss as well as change between climate periods.

Confluence spatial extent prediction summary

The second summary spatial extent focused on the temperature predictions for the tributary reaches flowing directly into the mainstem Columbia River (Figure 1B stars) and is referred to as the “confluence” or “outflow” summary spatial extent. This evaluates the potential for creating cold-water habitat (either as cold-tributaries fish swim upstream to, or as plumes within the mainstem channel) for salmonids migrating along the Columbia River mainstem and helps address our fourth study question. We evaluate these tributary junctions by comparing their outflow temperatures among scenarios to determine how upstream riparian shade loss or restoration affects the outflow temperatures among tributaries along mainstem Columbia River.

Reach size relationship to maximum cooling

We identified reaches that cooled by 1°C or more under restored vegetation conditions (when compared to current vegetation in 2000s climate) to identify network locations with the greatest restoration potential. As a proxy for river size, we used channel bankfull width (BFW) to characterize these maximum cooling reaches. This provided us with an estimate for which stream sizes could be cooled using riparian shade restoration activities. We also compared the frequency distribution of all stream BFWs in the study to those that cooled by 1°C or more to see if the maximum cooling potential reaches come from a select portion of stream sizes when compared to those present across the entire study region. Finally, we compared the percent change in reach shade between the current and restored shade conditions with the temperature change for reaches under these two shade conditions (for the 2000s climate) to extract what percent change would be required for a 1°C cooling effect.

Temperature criteria to protect fish uses

To address our fifth study question, we evaluated the thermal suitability of stream reaches for different fish species and uses in the study area. Numeric temperature water quality criteria (WQC) in streams of Oregon and Washington States were used as a proxy for thermal suitability. The fish uses evaluated in this study (Table 2) are a subset of designated fish and aquatic life uses in Oregon and Washington rules (for additional information regarding Oregon and Washington water quality standards regulations, see https://www.epa.gov/wqs-tech/water-quality-standards-regulations-oregon and https://www.epa.gov/wqs-tech/water-quality-standards-regulations-washington, respectively). We also only evaluated numeric temperature WQC that are applicable year-round (including the summer months) and, for example, did not include seasonal uses and their applicable numeric criteria. Therefore, our reference to fish uses in each state and their associated temperature numeric WQC are not exhaustive and only reflect the fish uses we selected for analysis due to their relevance to our study area and the mean August temperatures we are predicting. Finally, we do not address any WQSs specific to the mainstem Columbia River as we did not model water temperature within the mainstem reaches. We refer the reader to the TMDL for Temperature in the Columbia and Lower Snake Rivers (EPA 2021b) for a thorough review of the WQSs in the mainstem Columbia River. The interpretation of these numeric temperature WQC made in this study is for research purposes only.

Table 2.

State fish uses evaluated in this study and their associated seven-day average daily maximum (7DADM) numeric criteria values. Temperature numeric water quality criteria from Oregon Administrative Rules Database 340-041-001 and Washington Administrative Code 173-201A-200

State	Fish Use	7DADM numeric criterion
Oregon	Char spawning and rearing	12°C
	Core cold-water habitat	16°C
	Salmon and trout rearing and migration	18°C
	Salmon and steelhead migration corridor	20°C
Washington	Char spawning and rearing	12°C
	Core summer habitat	16°C

For each state, we evaluated how much stream length (km) for a particular fish use (Table 2) would meet the state numeric WQC for temperature under our nine prediction scenarios. Each state has public geospatial databases (Oregon Spatial Data Library - https://spatialdata.oregonexplorer.info/geoportal/; Washington Department of Ecology GIS Data - https://ecology.wa.gov/Research-Data/Data-resources/Geographic-Information-Systems-GIS/Data) containing hydrography that can be filtered to specific fish uses. We used each state’s hydrography data to locate reaches within our study system associated with the fish uses (see Supplement S5 for fish use maps related to each numeric temperature WQC we evalulated).

The Oregon and Washington temperature numeric WQC for the fish uses we evaluated are based on the seven-day moving average of the daily maximum (7DADM) temperature metric. Our SSN models predict mean August stream temperature, so we developed a binomial logistic regression model relationship between mean August stream temperature and the probability of exceeding a 7DADM numeric criteria value. This relationship links empirical mean August stream temperature (explanatory variable) to whether or not a site exceeded the 7DADM criterion (the binomial response variable) during the month of August (Supplement S6). From the logistic regression fits, we selected a 90% probability of exceedance to back out what mean August temperature for each fish use numeric criterion would likely result in an exceedance. This 90% probability does not represent a policy recommendation, refer to any specific WQS guidance, or represent a threshold used by Oregon or Washington to determine impairment. Impairments are determined by the states through application of their own state listing and assessment methodologies. The EPA then reviews and takes action on individual state impairment lists pursuant to the Clean Water Act Section 303(d). Instead, we set this 90% probability as a conservative estimate for when reaches with mean August temperature above a threshold likely exceed the specified state numeric criterion for each binomial logistic relationship.

Using the mean August temperature threshold for each 7DADM numeric criterion, we then used our reach-scale scenario predictions of mean August temperature to identify which reaches remained below the threshold and would likely meet the 7DADM numeric criterion. We fit logistic binomial regressions (Supplement S6) for each evaluated fish use numeric criteria value (Table 2). It is worth noting that mean August and 7DADM stream temperature statistics are highly correlated (r = 0.97; Isaak et al. 2017), but mean August temperatures have been more accurately predicted than 7DADM temperatures (Detenbeck et al. 2016). In addition to the better prediction accuracy of mean monthly temperature statistics, we also required mean monthly statistics to apply future climate period deltas to the mean August air temperature and discharge values in our models. This binomial logistic regression analysis extension provided us with predictions for how riparian vegetation shade restoration or loss affects available thermally suitable habitat for each fish use and how it might change in future climates.

Results

Riparian vegetation shade

Average reach shade percentages varied widely among the three shade conditions (Figure 2A–C). Topographic shading had an average reach shade of ~9% (Figure 2A), while restored vegetation shade averaged ~85% (Figure 2C). Current vegetation shade averaged across all stream reaches was ~50% (Figure 2B). The difference between current and topographic shade conditions emphasizes the high current percent shade in the heavily forested Cascade Mountains that separate the Oregon Coast and Mid-Columbia processing units (Figure 2D). Shade differences between the current and restored vegetation conditions indicate where riparian restoration could have the greatest effect on stream temperatures, most notably at lower elevations in the Willamette River valley and east of the Cascade Mountains (Figure 2E).

Figure 2.

Percent of reach shaded for (A) topographic – no vegetation –, (B) current vegetation, and (C) restored vegetation shade conditions as well as the differences between topographic (D) and restored (E) shade conditions from the current shade condition. Percent values in the lower left corner indicate landscape mean percent shade or percent shade difference values for each map.

SSN model fit diagnostics

Both OC and MC processing unit models had LOOCV R² values near 0.9 and RMSPE values at approximately 1°C (Table 3). Overall, the fixed effect covariates in the models explained a small proportion of the variance in the observed temperature (OC: 19%; MC: 10%). The spatial autocovariance functions explained a large percentage of the variance in the data (Table 3) indicating the temperature data set is highly spatially structured. The percent variance explained by fixed effects was higher for the OC model fit than the MC model fit suggesting the fixed effects were more useful in explaining variation in the observed temperature data than in the MC processing unit.

Table 3.

Model fit diagnostic statistics in Mid-Columbia (MC) and Oregon Coast (OC) processing units. Statistics for comparison include Akaike’s Information Criterion (AIC; not comparable between models from different processing units), leave one out cross validation (LOOCV) R², root mean square prediction error (RMSPE), percent of variance explained by the fixed effect covariates, the three exponential autocovariance functions (tail-up, tail-down, and Euclidean), the random effects of site and year, and the nugget.

	MC	OC
Diagnostic statistics
Obs. Site n	6989	3140
AIC	20789	9528
LOOCV R2	0.93	0.913
LOOCV RMSPE (°C)	0.94	1.005
Variance components (%)
Fixed Effect Covariates	9.8	18.5
Exponential tail-up	35.5	41.3
Exponential tail-down	12.7	4.8
Exponential Euclidean	24.4	14.6
site	5.2	6.7
year	1.6	2.0
nugget	10.7	12.1

Parameter estimates for most fixed effect covariates in the OC model were statistically significant except for Upstream Lake Area and Mean August Discharge (Reach Slope was marginally significant at the α = 0.05 level; Table 4). Similarly, almost all parameter estimates were significant in the Mid-Columbia model fit except for Drainage Area and Mean August Discharge (Reach Slope was marginally significant; Table 5). Non-significant covariate parameters were retained in these models to maintain the same model structure as the original NorWeST SSN models (Supplement S3), and because the large sample sizes of temperature data minimized the effect of their retention on other parameter estimates.

Table 4.

Oregon Coast SSN model covariates raw (Raw) and standardized (Std.) coefficient estimates (Est), standard errors (SE), t-statistics (t), and p-values (p). Parameter abbreviations as in Table 1. Covariates ordered from largest to smallest absolute value of the standardized coefficient estimate (“Std. Est”).

Covariate	Raw Est	Raw SE	Std. Est	Std. SE	T	p
tailwater	−3.42	0.465	−3.421	0.465	−7.352	<<0.001
elevation	−0.00437	0.000492	−2.277	0.256	−8.88	<<0.001
shade	−0.0237	0.00314	−1.325	0.175	−7.553	<<0.001
precipitation	−0.000933	0.000232	−0.933	0.233	−4.012	<<0.001
latitude	−0.0000047	0.00000149	−0.877	0.278	−3.157	0.002
bfi	−0.0444	0.015	−0.853	0.289	−2.952	0.003
air	0.436	0.0955	0.524	0.115	4.563	<<0.001
drainage	0.00000591	0.00000207	0.497	0.174	2.855	0.004
glacier	−47	13.8	−0.354	0.104	−3.414	0.001
slope	3.26	1.72	0.264	0.139	1.899	0.058
lake	0.109	0.0873	−0.254	0.203	−1.249	0.212
discharge	−0.00303	0.0902	0.005	0.16	0.034	0.973

Table 5.

Mid-Columbia SSN model covariate raw (Raw) and standardized (Std.) coefficient estimates (Est), standard errors (SE), t-statistics (t), and p-values (p). Parameter abbreviations as in Table 1. Covariates ordered from largest to smallest absolute value of the standardized coefficient estimate (“Std. Est”).

Covariate	Raw Est	Raw SE	Std. Est	Std. SE	T	p
elevation	−0.0049	0.000357	−4.381	0.319	−13.741	<<0.001
tailwater	−4.07	0.909	−4.071	0.909	−4.479	<<0.001
lake	0.716	0.073	1.764	0.180	9.806	<<0.001
bfi	−0.0931	0.0188	−1.693	0.342	−4.947	<<0.001
precipitation	−0.00119	0.000277	−1.323	0.309	−4.28	<<0.001
shade	−0.024	0.00319	−1.072	0.142	−7.543	<<0.001
latitude	−0.00000676	0.00000289	−1.070	0.457	−2.342	0.019
air	0.413	0.0702	0.726	0.123	5.875	<<0.001
drainage	0.000064	0.0000478	0.195	0.145	1.339	0.181
slope	−3.71	2.19	−0.189	0.112	−1.692	0.091
discharge	−0.0845	0.0851	−0.140	0.141	−0.992	0.321

Network spatial extent prediction summary

Mean August stream temperature predictions for the current shade condition in the 2000s climate period averaged 14.25°C (range 1.9–23.1°C) across the river network, which was ~1°C cooler than the 15.22°C average (range 2.9–24.1°C) stream temperature predicted for the 2040s with current shade conditions (Figure 3). The average stream temperature with current shade condition in the 2080s of 16.11°C (range 3.7–25.0°C) was nearly 2°C warmer than the 2000s current shade condition (Figure 3).

Figure 3.

Density distributions for all nine scenarios that represent the combinations of three climate periods (2000s, 2040s, 2080s) and three shade conditions (topographic – no vegetation shade, current vegetation shade, restored vegetation shade). Distribution labels represent the distribution mean water temperatures (left labels) and the mean temperature difference (right labels) for each scenario from the 2000s climate period and current vegetation shade condition scenario (top distribution).

Stream temperature predictions for the scenario using restored shade and the 2000s climate had a mean stream temperature of 13.63°C (range 1.3–22.9°C), which was ~0.6°C cooler than the baseline of current shade and the 2000s climate (Figure 3). The average stream temperature with restored shade in the 2040s was 14.6°C (range 2.3–23.9°C) and 15.49°C (range 3.1–24.8°C) in the 2080s (Figure 3). The average warming between the baseline current vegetation and 2000s climate and the predictions for restored vegetation shading in the 2040s and 2080s was 0.35°C and 1.24°C, respectively (Figure 3).

When comparing the temperature predictions between the current vegetation and topographic shade scenarios, we see how much warmer stream temperatures would get if riparian vegetation were removed from the system. For topographic shading, average temperature increases across the study network were 1.06°C for the 2000s climate, 2.03°C in the 2040s, and 2.92°C in the 2080s (Figure 3). Average temperatures for the topographic shade scenarios were 15.32°C (range 3.4–23.1°C), 16.29°C (range 4.3–24.1°C), and 17.18°C (range 5.2–25.1°C) for the 2000s, 2040s, and 2080s climates respectively. See Supplement S7 for temperature prediction and difference maps among all scenarios.

Confluence spatial extent prediction summary

Individual tributary outflow temperatures ranged from ~10°C to ~25°C across all nine scenarios (MC: 10–25°C, OC: 11–24°C; Figure 4). As reference, the mainstem Columbia River into which these tributaries flow has a mean August temperature of 21–22°C (EPA 2021a; EPA 2021b). The mean temperature of all tributary outflows (n=198) for each prediction scenario ranged from 15.5°C to 19.2°C (Figure 4). The mean outflow temperature difference between current and restored shade for the 2000s climate was ~0.9°C cooler under restored shade conditions. Additionally, the predicted warming between the 2000s and 2040s climates for these outflows with current shade indicates about a 1°C increase in temperature. The mean outflow temperature in the 2040s for the restored vegetation shade scenario is only 0.1°C warmer than the 2000s climate and current vegetation shade scenario (Figure 4).

Figure 4.

Scenario density distributions for the 198 tributary outflows that reach the mainstem Columbia River. Mean tributary outflow temperatures are labelled on the left and the mean temperature differences with the recent 2000s climate and current shade condition scenario (top distribution) are labelled on the right.

Individual tributaries of the mainstem Columbia River are predicted to have different response magnitudes to riparian restoration at their outflows (Figure 5). The temperature of some large tributaries (e.g., Deschutes River and John Day River, which are both > 100 m wide) at their confluence with the Columbia River were mostly unaffected by either restoring or removing riparian vegetation. For example, in the 2000s climate the predicted outflow temperatures for all three shading scenarios of the Deschutes River are within 0.1°C. In contrast, some tributary outflows cooled considerably (e.g., >1.4°C cooling in Little White Salmon and Rock Creek) under restored riparian shade conditions (Figure 5). Finally, there are also tributaries that appear more susceptible to stream warming from riparian vegetation loss (e.g., Eagle Creek [0.9°C warmer] and Tanner Creek [0.8°C warmer]; Figure 5). For a flow-weighted tributary outflow temperature analysis see Supplement S8.

Figure 5.

Tributary outflow temperatures for all three shade conditions under the 2000s climate period. Only tributaries with mean August discharge greater than 1 m³/s shown. For similar statistics/details for all 198 tributaries in this study see Supplement 2.

Reach size relationship to maximum cooling

Spatially, streams that cooled by 1°C or more under restored riparian conditions for 2000s climate were more prevalent across the OC (especially within the Willamette River floodplain) than the MC (Figure 6A). The Cascade Mountain region did not have many stream reaches that cooled in response to riparian vegetation restoration (Figure 6A). Most reaches that cooled more than 1°C had BFWs < 10 m and cooling prevalence was greatest in streams with BFW of 2.5–7.5 m (Figure 6B). The density distribution for reaches that cooled at least 1°C is similarly shaped to the density distribution of all stream reaches, suggesting that reaches most responsive to restored riparian vegetation are not from a particular subset of study reaches.

Figure 6.

Stream segments with at least 1°C temperature decline when (A) comparing current and restored shade condition temperatures for the 2000s climate. Black lines in part A indicate the Columbia River mainstem and some of the major tributaries for reference to Figures 1 and 2. Density distribution plots (B) show the range in size of streams (as bankfull width) across the entire study system (All reaches) and the reaches that cool by 1°C or more in part A. A scatter plot (C) compares the difference in shade percent between the restored and current shade conditions as bankfull width (note log scale) changes across all study reaches. A linear relationship (D) between the percent shade difference and temperature difference for the reaches cooling by 1°C or more.

A comparison between the percent change in shade (between the current and restored conditions in 2000s) versus the bankfull width of all reaches in the study indicates most streams with large percent changes in shade are < 10m BFW (Figure 6C). For reaches from 10 to 100m BFW, there is an exponential decline in the maximum potential increase in percent change in reach shade. Reaches with a BFW > 65m have a maximum percent shade change of less than 7% (Figure 6C). The relationship between the percent shade change and temperature change between the restored and current shade conditions (2000s climate) is linear and for both processing unit regions the percent shade change necessary for a 1°C cooling effect is a little more than 40% (Figure 6D).

Temperature criteria to protect fish uses

Within a climate period, four of the six fish uses we evaluated (Char spawning and rearing [OR and WA], Core cold-water habitat [OR], and Core summer habitat [WA]; Table 2), had the proportion of fish habitat respond more to the loss of riparian vegetation (topographic versus current shade conditions) than to the restoration of riparian vegetation (restored versus current shade conditions) (Figure 7; see Supplement S5 for fish use maps). For these four fish uses, the proportion of reach length lost was 3–5 times larger than the proportion of reach length gained for a given climate period. For example, along the char spawning and rearing reaches, removing the current riparian shade resulted in a nearly 50% loss of habitat meeting the numeric temperature WQC (12°C 7DADM), while restoring vegetation only increased the proportion of fish habitat meeting the numeric temperature WQC by ~10%.

Figure 7.

Proportion of designated fish use reaches that meet the Oregon and Washington water quality standard for temperature. Both states use a 7DADM numeric water quality criterion metric and for similar designated fish uses, each state has the same numeric criterion value.

The proportion of fish habitat meeting Oregon’s salmon and trout rearing and migration fish use numeric temperature WQC responded somewhat equally (15% loss or gain) to the loss or restoration of riparian shade among its fish use reaches (within a given climate period). The same response was observed in the proportion of fish habitat for salmon and steelhead migration corridor fish use with one exception. Under the 2000s climate, riparian restoration resulted in nearly three times the length of habitat meeting the numeric temperature WQC than under the current shade condition. This large improvement was not observed in the two future climate periods for the salmon and steelhead migration corridor fish use.

Across climate periods (from 2000s to 2080s) in both states, all fish uses evaluated had an ~50% decline of fish use reaches meeting the numeric temperature WQC for a given shade condition (Figure 7). The largest decline (>70% loss) across climate periods in the proportion of reach length meeting the numeric temperature WQC was observed in the salmon and steelhead migration corridor fish use under restored shade conditions (Figure 7).

Discussion

Our results suggest riparian vegetation shade restoration across large spatial extents (i.e., entire networks) could meaningfully-reduce stream temperatures (0.62°C cooler on average) from their current state. This reduction could offset significant amounts of climate warming predicted by mid-century but would be exceeded by late century if projected trends are realized. Consequently, restoration activities in addition to riparian shade restoration will be necessary to fully mitigate the predicted effects of future climate change. Streams most responsive to riparian restoration had BFWs less than 10 m and the ability to increase riparian shade by 40%. Tributaries’ confluences with the mainstem Columbia River cooled on average 0.87°C under restored shade conditions (0.02–2.08°C range), but some individual tributary outflows were not responsive to either shade restoration or removal. Depending on where cooling is desired from restoration activities, the location of the restoration efforts should be carefully assessed because, for example, improving headwater stream temperatures is unlikely to translate to significant downstream cooling in some streams (e.g., Deschutes River). Designated fish use habitats appear only moderately sensitive (small increase in habitat meeting numeric temperature WQC) to riparian shade restoration with the notable exception being increases in salmon and trout rearing and migration designated fish use reaches meeting numeric temperature WQC in OR. In contrast, designated fish use reaches were very sensitive to the loss of riparian vegetation shade under recent and future climates (large habitat loss meeting numeric temperature WQC) suggesting that protecting current riparian vegetation shade may be as important as restoring shade. Viewed collectively, the spatial variability of stream temperature responsiveness suggests restoration actions need to be undertaken strategically to maximize desired biological beneficial.

Implications of riparian management

Studies have empirically investigated the direct influence of riparian shade restoration at the reach-scale through manipulation of the system (Hannah et al. 2008; Johnson 2004; Leach et al. 2012). However, these studies are unable to fully explore how their reach-scale responses scale across networks. At the network-scale, the orientation of the solar path relative to the stream channel can be equally, if not a more important predictor of stream temperature than riparian vegetation shade (Garner et al. 2017). Therefore, combining the reach- and network-scale drivers of stream temperature is important when building predictive models exploring the influence of solar radiation on stream temperature across landscapes. This is especially important when considering thermal requirements of species with large home ranges or spatially extensive migration routes. Channel orientation and solar paths were included and combined with our riparian vegetation shade estimates for this study to account for both of these important temperature drivers (Chen et al. 1998).

SSN models provide an approach that addresses downstream propagation of stream temperature and also functions at river network spatial extents (Peterson & Ver Hoef 2010). The SSN models we fit connect spatially disparate empirical data points across river networks and appropriately interpolate between them by accounting for the downstream propagation of stream temperature (network-based spatial autocorrelation) and the Euclidean spatial autocorrelation inherent in landscapes. The large percent variation explained by spatial autocorrelation in the observed temperature data highlights why it is important to account for spatial autocorrelation based on both network and Euclidean distances. While our scenarios took an “all-or-nothing” approach to restoration, future research could evaluate restoring riparian vegetation in portions of a river network to see how strategically placed restoration activities affect stream temperatures locally and regionally. This could specifically look at restoring only low stream order tributaries to see whether downstream mainstem temperatures respond with continuous cold-water habitat or patchy cold-water reaches within the mainstem that cold-water species could use for behavioral thermoregulation (Corey et al. 2020, Snyder et al. 2020)

The simulation of our topographic shade condition allows us to estimate the importance of the natural capital existing in the current riparian vegetation for shade. This natural capital can be valued for its cooling benefits and leveraged when making ecosystem service quantifications (Vermaat et al. 2021). Furthermore, this natural capital could help identify equivalent zones for restoration required by tradable permits and mitigation banking instruments for rivers (Pirard 2012). Natural capital loss can also be quantified for natural disasters such as forest fires which remove riparian vegetation (Pettit & Naiman 2007). Our topographic shade scenarios could be used as baseline expectations for areas that recently burned or have high fire risk in the future.

Reach and confluence cooling

Our modeling scenarios indicated that stream riparian shade restoration is most effective at cooling stream reaches with BFWs < 10 m. This was not unexpected as vegetation height limits how much of a channel cross-section is shaded (Davies-Colley & Rutherford 2005). Additionally, our models indicated that reaches with the capacity to increase reach shade by 40% would cool by 1°C. Given our methods for estimating reach shade, this was only observed for streams with BFWs of up to 50m. Identifying stream-size and shade increase bounds for effectiveness can direct managers toward systems that may benefit more from restoration. Additionally, when channel modification is also considered in restoration activities (e.g., Merrriam & Petty 2019), narrowing artificially widened channels (e.g., Belsky et al. 1999) could put systems within BFWs where stream cooling is more likely. A more thorough evaluation of these cooling reaches and their characteristics would be a fruitful analysis for managers looking to identify additional features associated with maximum potential cooling.

Some tributaries reach the mainstem Columbia River at nearly the same temperature under all three riparian shade conditions (within a given climate period). This suggests the mainstems of these tributaries contain physical characteristics or conditions that do not respond to riparian vegetation shade restoration (e.g., wide channel widths impacting the ability of the riparian vegetation to cast a shadow over the wetted channel) and/or the cooling predicted in upstream reaches does not persist downstream in larger reaches (EPA 2021b). The Deschutes River has an outflow temperature that varies little between different shade conditions. Under the 2000s climate, the outflow temperature at the Columbia River only varies by ± 0.1°C among the three riparian shade conditions. However, the Deschutes River headwaters reaches (Strahler order 1–3) do have temperature decreases of than 1°C or more under restored shade conditions. If the motivation for riparian shade restoration activity is primarily to improve downstream mainstem river temperatures, restoration planning should first conduct modeling exercises to determine if the restoration has the potential to be effective at the target location.

Within the study area, we can overlay our estimated shade data across the landscape with the stream reaches that cool by more than 1°C. A region of intersection is clear in the Willamette River valley. This area is heavily influenced by both agriculture and urbanization (National Land Cover Database 2016 – Yang et al. 2018) which are associated with our lower estimated current riparian vegetation shade. One reason this area has a large number of the reaches that cool is because there are many streams that could potentially have much greater reach shade after riparian vegetation is restored to its targeted late-successional/climax vegetation community (high proportion of upland and wetland forests [DOI 2014]). Alternatively, low elevation areas in western half of the MC also have lower current average reach shade, but this is likely not primarily due to human modification of the land cover, but more likely due to different expected vegetation communities (the predicted late-successional/climax vegetation is a high proportion of upland shrubs [DOI 2014]), and thus this area had relatively few reaches that cool by 1°C or more.

Developing predictive models, as in this study, provides a baseline to scenario plan for different restoration priorities as they change through time. Just as we modified NorWeST SSN temperature models (Isaak et al. 2017) to investigate riparian shade’s influence on stream temperatures, the models fit for this study can be used as initial starting points for additional restoration scenarios. The output from our prediction scenarios could benefit other modeling efforts by feeding the spatially continuous stream temperature information into bioenergetics models for salmonids moving throughout networks during migrations (Snyder et al. 2019). Additionally, a more thorough understanding of the large spatial extent effects from habitat restoration, may also provide a means to better evaluate the effects of food-web-based restoration for threatened species (Naiman et al. 2012).

State WQC and fish use habitat

Our evaluation of the role riparian shade plays in controlling stream temperature demonstrates how there is significant natural capital currently in the landscape as the loss of current riparian shade resulted in large proportions of fish habitat no longer meeting the Oregon and Washington state numeric temperature WQC. WQC that had colder temperatures were most vulnerable to habitat loss following riparian vegetation loss. Warmer numeric temperature WQC fish uses, such as the 20°C salmon and steelhead migration corridor fish use, already had small proportions of the fish habitat likely meeting the numeric temperature WQC under recent climate conditions. This makes these reaches especially important for targeted protection and restoration as our predictions indicate little habitat remains suitable under our recent climate scenario and even less in the future scenarios.

According to the 2000s climate and current riparian shade condition scenario, the salmon and steelhead migration corridor reaches are at a critical threshold for being able to improve a large proportion of available habitat meeting numeric temperature WQC using riparian shade restoration. Under the recent 2000s climate, riparian restoration can triple the length (200% increase) of river meeting the 20°C 7DADM criterion. In contrast, under the 2040s climate, the improvement in available habitat meeting the numeric temperature WQC is only a 50% increase in available habitat. This suggests that a large portion of the salmon and steelhead migration corridor reaches are very close to meeting the 20°C 7DADM numeric temperature WQC. The warming expected for the 2040s, however, pushes many of these reaches past the cooling effects of riparian vegetation shade restoration.

The location of the fish habitat is also important for determining the benefit from riparian shade restoration or loss. The Oregon salmon and trout rearing and migration fish habitat overlaps with the Willamette River valley region which included a higher prevalence of reaches cooling by 1°C or more in response to riparian shade restoration. This is likely why we see a large increase in proportion of this designated fish use habitat meeting numeric temperature WQC under restored shade conditions. Similarly, the fish uses that demonstrated a large loss of fish habitat when riparian shade was removed were generally located at high elevations in the headwaters of catchments where riparian shade is already high.

Despite each of these designated fish use habitats being defined and regulated separately (and with different numeric criteria), for many of the species using them, maintaining all designated fish use habitats in concert is often necessary to provide adequate conditions for all life cycle stages. This may not require continuous suitable habitat for each designated fish use, as patches of habitat are known to be used successfully by migrating individuals in river networks (Keefer et al. 2009). However, reducing continuously suitable habitat to patches within the network can come at a cost to individual survival (Snyder et al. 2020).

Study limitations

We modeled mean August stream temperatures as an estimate of general thermal habitat suitability, but maximum daily or monthly temperatures can also be thermally limiting factors for coldwater species (Isaak et al. 2010). Daily maximum stream temperatures have been more responsive to stream shade manipulation than mean stream temperature metrics (Johnson 2004). However, predicting maximum temperatures has been less accurate than predicting means (Detenbeck et al. 2016; Isaak et al. 2010). Therefore, choosing a metric (e.g., mean or maximum statistic) and its temporal aggregation (e.g., daily, weekly, or monthly) should incorporate weighing the relative importance of prediction accuracy versus any ecological impact that may be associated with those predictions (Turschwell et al. 2016).

The effort required to restore riparian vegetative shade across the spatial extent of this study is likely unachievable. The purpose of generating a scenario with an unattainable level of restoration effort is to determine what the theoretical bounds are for the restoration method in the study region. With an upper bound on what maximum riparian shading can produce as a restoration tool, other restoration and management activities can be added to the management plan to investigate their influence on stream temperature in concert with riparian shade. Additional restoration efforts could address management of surface water extractions and returns to retain water in the channel, narrowing of stream channel widths leading to increased shading potential and lower channel width to depth ratios, and groundwater/hyporheic restoration (Poole & Berman 2001).

In simulating future climate periods, we applied a standard temperature and discharge change across each of our modeled processing units (MC and OC regions) as in Isaak et al. (2017). Given the variability in mean August air temperature and discharge observed across study basins much smaller than our study extent (e.g., Steel et al. 2019) this may not be the most accurate assumption for future climate periods and will undoubtedly miss the local heterogeneity in temperature change across a landscape for future climate periods. However, given the large spatial extent of our study, identifying more localized responses to mean August air temperature and discharge were outside the bounds of our research scope, but would provide an extremely valuable modification to these models if pursued in the future.

The linear relationship established in our SSN models between reach shade and stream temperature indicates an increase of ~40% reach shade should cool a stream reach by ~1°C. This relationship is constant across the study networks and thus suggests that both small headwater streams and larger rivers will cool by 1°C if either has an increase in riparian shade of 40% or more. Despite this relationship being the same for both small and large rivers, is unlikely for larger rivers (BFWs > 40m) to attain a 40% increase in reach shade. For example, a river that is 100m wide requires an additional 40m of the channel cross section to be completely shaded over the course of a day. From our estimated shade conditions, reaches with BFWs >43m never reached this 40% shade increase threshold. Furthermore, reaches with a BFW >65m never had restored shade increases of more than 7%.

Our study evaluated just one potential response (stream temperature) to the modification of riparian vegetation, but there are several other stream ecosystem processes that are also related to riparian vegetation condition. Ecosystem processes such as benthic primary production (Kiffney et al. 2003), nutrient and sediment water quality from nonpoint source pollution (Dosskey et al. 2010), food-web subsidies crossing aquatic-terrestrial boundaries (Nakano & Murakami 2001), geomorphic stability (Krzeminska et al. 2019), and stream community composition (Benstead et al. 2003) all have associated relationships with riparian vegetation. Consequently, the riparian vegetation conditions evaluated in this study would also lead to changes in these other ecosystem processes. Future modeling efforts could attempt to better understand the effects and potential benefits of riparian vegetation restoration on these additional ecosystem components.

Implications for Practice.

• Riparian vegetative shade restoration can improve cold-water habitat availability for aquatic species (both residents and migrants) dependent on cold temperatures under recent climate conditions.

• Streams with bankfull widths less than 10 m and the capacity to increase reach shade by 40% are most likely to cool from riparian vegetation restoration.

• Future climate warming is predicted to outpace the beneficial cooling from riparian vegetative shade restoration by the 2040s.

• The proportional loss of fish habitat meeting numeric temperature water quality criteria from riparian vegetation shade removal is greater than the proportional gain from riparian vegetation shade restoration.