Large wood as a confounding factor in interpreting the width of spring-fed streams

. Spring-fed streams throughout volcanic regions of the western United States exhibit larger widths than runoff-fed streams with similar discharge. Due to the distinctive damped hydrograph of spring-fed streams (as compared to large peaks visible in the hydrographs of runoff-fed streams), large wood is less mobile in spring-fed than runoff-fed stream channels, so wood is more likely to remain in place than form logjams as in runoff-fed streams. The consequent long residence time of wood in spring-fed streams allows wood to potentially have long-term impacts on channel morphology. We used high-resolution satellite imagery in combination with discharge and climate data from published reports and publicly available databases to investigate the relationship between discharge, wood length, and channel width in 38 spring-fed and 20 runoff-fed streams, additionally responding to a call for increased use of remote sensing to study wood dynamics and daylighting previously unpublished data. We identiﬁed an order of magnitude more logjams than single logs per unit length present in runoff-fed streams as compared to spring-fed streams. Histograms of log orientation in spring-fed streams additionally conﬁrmed that single logs are immobile in the channel so that the impact of single logs on channel morphology could be pronounced in spring-fed streams. Based on these observed differences, we hypothesized that there should be a difference in channel morphology. We found that spring-fed streams in our study are about 2 times wider than runoff-fed streams with similar mean discharge. A model for stream width in spring-fed streams based solely on length of wood is a better model than one derived from discharge or including both discharge and wood length. This study provides insights into controls on stream width in spring-fed streams

Figure 1 caption --"are have" isn't correct (presumably you mean just "have"?) --it might help readers to specify which of these is spring fed and which is runoff fed Figure 2 caption --Google should be capitalized --remove "from" before 50-90°?Line 222: just "imagery were clear"?Lines 228-235: I was surprised to find this text here, because you had discussed these two systems earlier, at lines 193-200.Is there merit in considering moving the text at 228-235 up so that it brings together the discussion of these two systems?That would also mean switching the order of Figures 3 and 4, which also makes some sense to me.Line 246: "When" should not be capitalized Figure 5: this is blurry in the version of the manuscript that I have; please just double check when you receive the proofs that this (and all of the other figures) appear clearly Line 286: extra space before 36% Line 287: Do you mean that models 1 and 2 resemble model 4? Otherwise this logic is not completely clear to me, with reference to the prior sentence.
Line 305: what does "time of LW" mean?Line 332: either make point cloud one word (as at line 328) or two words, as here Line 337: no comma after El Tatio streams Thank you for providing the data via GitHub!Response from the authors: Thank you for your feedback and for accepting the manuscript for publication.We are excited that this work will appear in esurf.All of the suggestions made are greatly appreciated, and we have made corrections for all of them.We reviewed the manuscript for tense consistency and adjusted the tense where appropriate.Any comments that could have additional explanation are below.Thank you so much.
Line 154-155: The length of these segments varied based on how much wood was easily visible.For streams with a great deal of wood and clear imagery (like Cultus River), a reach of about 1.5 km was sufficient to collect about 100 data points on wood orientation.For other streams, especially runoff-fed streams, this was more difficult.For some runoff-fed streams, the entire length of the stream was scoured to get as many data points as possible.McCloud River is the most extreme case, with a reach of about 30 km explored to find single logs in the channel.
1 Introduction ?first proposed a set of power laws to describe channel morphology based on discharge.Subsequent studies confirmed the existence of a relationship between discharge and width (e.g., ???), but the scatter in the relationship is large.There is a wealth of empirical correlations to describe width based on environmental conditions; however, the best relationships exhibit limited capacity to describe real channels (?).

1
In certain cases, though, it may be possible to predict channel width more precisely.One example is that of spring-dominated or spring-fed streams.Spring-fed streams receive the bulk of their discharge from groundwater sources and thus exhibit relatively stable hydrographs (e.g., ??).Compared to runoff-fed streams, spring-fed streams transport a proportionally larger amount of sediment in everyday flows than high-flow events, leading to different channel responses to disturbance, such as flow obstacles (?).Spring-fed streams are a promising test group for understanding some of the controls on stream width since their stable hydrographs reduce the number of variables impacting the channel.
Previous studies have identified differences between runoff-and spring-fed channels (e.g., ??).? studid streams in the western US, primarily in the Oregon Cascades, and found that the spring-fed streams in their study (0.005-8 m 3 /s) are signficantly wider than their runoff-fed counterparts.Conversely, a study comparing spring-fed to runoff-fed streams in Arizona (10 −3 m 3 /s) found that spring-fed streams exhibited :::::: exhibit lower width-to-depth ratios than runoff-fed streams (?).The streams studied by ? and ?are comparable in every aspect save discharge and the presence of large wood (LW).The streams studied by ?had high discharge and significant amounts of LW, while the streams studied by ?had very low discharge and essentially no LW.
? found that the presence of wood increases mean water depth, implying lower mean velocities but local velocity increases.?
demonstrated that single logs can increase bank erosion via those local velocity increases, providing a mechanism for channel widening with the presence of LW.However, with multiple single logs in a stream, the effect is enhanced when single logs are very close together but dampened when they are moderately closely spaced (?).In contrast, removal of LW has been observed to cause rapid changes to channel form, including rapid channel widening (???).The mechanism for LW constriction of channel width is streambank stabilization by LW (?).
Despite evidence that LW impacts channel dimensions, LW was absent from early discussions of channel geometry (?).We hypothesize that LW widens spring-fed streams.In general, the stability of LW in channels is related to flow characteristics of the stream and the size of LW (????).Notably, ?show :::::: showed : that peak annual discharge has a large impact on LW mobility, and generally, hydrology is a good predicter of wood mobility (?).Thus, due to differing hydrograph behavior, peak events in runoff-fed streams may be able to mobilize wood, whereas the more stable hydrographs of spring-fed streams generally lie below the threshold for wood mobility, making LW more likely to be immobile in spring-fed but not runoff-fed streams.In order to assess this hypothesis, ?measured orientations and diameters of wood in Oregon streams to determine whether wood was oriented with respect to the thalweg.They found that wood in runoff-fed channels was generally more oriented with flow, demonstrating mobility, and wood in spring-fed channels was generally aligned randomly or more perpendicular with flow, implying immobility.
We hypothesize that mobility promotes the development of logjams in runoff-fed streams (e.g., ?) and explains the paucity of logjams in spring-fed streams, where single logs may dominate the population of LW.In addition to the impacts on channel widening, the presence of logjams may impact morphology by forcing a multi-threaded rather than a single-thread channel (??).With a low abundance of logjams in spring-fed streams, we thus expect that the wood interaction mechanism explored by ?for single logs in single-thread streams (i.e. an increase in bank erosion) may dominate, leading to channel widening associated with the presence of LW.With sufficient logs immobile in a channel, the consequent bank erosion would increase the reach-averaged width-to-depth ratio.In contrast, logjams may produce more variable effects on channel morphology or locally stabilize banks, cause channel constriction.
The purpose of this study is to examine the empirical relationship between LW and the morphology of spring-fed streams in order to identify statistically significant relationships.We also respond to a recent call by ? to employ remote sensing to study wood dynamics and to daylight unpublished data on wood dynamics.Specifically, we investigated (1) wood orientation and frequency of logjams, (2) discharge and width of stream channels, and (3) length of LW and width of stream channels.

Field Area
In The streams located in eastern Idaho and southwestern Montana are located in the easternmost part of the Columbia Plateau (Snake River Plain) and neighboring Middle Rocky Mountains physiographic provinces (?).The annual precipitation is 300-600 mm with about 150 mm snowfall (?).Mean Annual temperatures range from 1-9 • C (?).The area is underlain by Quaternary rhyolite and basalt (?).The streams in this region primarily run through oak/pine woodland.
The spring-dominated streams in southwest Oregon and northern California are located along the border of the Cascade-Sierra Mountains and Basin and Range physiographic provinces (?).This area lies in the rain shadow of the Cascades to the west.Mean annual precipitation, dominated by snow, decreases from over 1 m to the west to about 0.5 m in the southern part of the study area (?), and mean annual temperatures range from 8-12 • C (?).The area is underlain by Quaternary basalt and basaltic andesite.Typical land uses ::::: cover for the studied streams in this region are oak or pine woodland, grassland, shrubland, wetland, and some small farms.
The streams studied in northern Arizona are located along the Mogollon Rim (?).The high relief of the Mogollon rim at 2100 m induces a strong orographic effect (?), yielding some of the highest precipitation in the state, an annual average of more than 800 mm (?), and the mean annual temperature is 17 The streams in El Tatio Geyser Basin, Chile are located on the San Pedro formation (?).Located in the Atacama desert, precipitation is very low at 0.025 m/yr, but the high elevation means that the mean annual temperature is 3.6 • C (?).This area is underlain by andesites, dacites, and rhyolites (?), with the streambed material consisting of glacial outwash.The streams in this area run through desert landscapes above treeline.These streams are included for comparison between spring-fed streams with and without wood since these streams are above treeline and have no recent history of LW.Other spring-fed streams with no visible LW in this study may have had LW in recent history since the watersheds they run through contain forests.
Spring-fed streams occur in specifically defined geological settings in which a highly permeable material overlays an impermeable layer, such as in the volcanic regions explored in this study (?).The geologic setting is important for producing the conditions for spring-fed streams to exist and sustain.Due to these particular geological constraints, it is difficult to find a large, comparable set of runoff-fed streams.We selected ::::: select a set of streams that are located as closely as possible to the spring-fed streams in this study to control for geology as much as possible.We can verify that the labeled runoff-fed and springfed streams display different hydrograph behavior by examining the mean and standard deviation of flow, when available.All spring-fed streams with available data exhibit standard deviations :: in :::::::: discharge smaller than their mean ::::::: discharge, whereas the runoff-fed streams show standard deviations larger than their mean.When unavailable, we rely on the cited authors to correctly identify the flow source for the stream.
For 25 spring-fed and 19 runoff-fed streams containing wood, we measured the length of 10 or more pieces of LW found in or near the channel in this same reach (Table 1).Additional measurements were taken for streams exhibiting a high degree of variability in wood length.This measurement is meant to characterize the wood source to the streams, so wood found near the streams should be representative of the wood that enters the channel.If wood were only measured in the channel, then the results may be biased since we only measured wood for which we could confidently identify both ends.In the channel, this criteria rules ::::::: criterion :::: ruled : out many pieces of wood, often excluding smaller pieces or pieces where one end is obscured by trees.Wood outside the channel is ::: was sometimes more clearly identifiable in aerial imagery.To verify the validity of this technique, we compare :::::::: compared field measurements of wood length at one site to results from remotely sensed measurements.
While fully submerged logs likely have an impact on stream morphology as well, they are :::: were : largely not included in this study due to unreliable identification via satellite imagery.For the remainder of the paper, the term "wood length" refers to the average wood length.
To test the precision of our technique of measuring length in Google Earth Pro, we measured the length of a single log 10 times in a row to yield a length of 17.6±0.2m with 90% confidence.The small size of the confidence interval (1.2%) suggests :::::::: suggested relatively high precision for the technique.All LW observed via satellite imagery and in the field at this location was long, so no estimates on accuracy of the method for measuring small pieces of LW were possible.
For streams marked by a † in Table ??, we also took histograms of log orientation for single logs in each stream.Histograms were taken using Google Earth Pro imagery.Ideally, we could measure wood orientation on a scale from 0 • (directly in line with flow) to 180 • (directly opposite to flow).This is possible in the field, but due to limitations in imagery resolution, we were unable to reliably distinguish the bottom and top of LW in this study.As a result, we noted orientation of LW on a scale from 0 • (parallel to flow) to 90 • (perpendicular to flow), unable to note orientation (± 90 • ).
We additionally observed, for streams with multiple dates of clear imagery, whether there was any detectable change in wood placement for 20+ observed logs between dates.Dates were typically from about 2005 to about 2018 with variation in the specific years and time periods when imagery were available.Regional precipitation records do ::: did : not indicate persistent drought through the entire time period at any site (?), although local conditions may deviate from regional averages.We primarily observed single pieces of LW with few or no logjams in the studied spring-fed streams.We quantified this observation by measuring the density of single logs and the density of logjams over a reach about 500 m in length for streams with adequately clear imagery.These data also allow :::::: allowed for a sense of how close LW is to one another.This is important since the effect of LW on bank erosion is increased when single logs are close together (?).We found all best fit parameters using the Marquardt-Levenberg algorithm.
Discharge data are :::: were : obtained from a range of sources.When available, mean and standard deviation are :::: were : reported.
For spring-fed streams, mean is ::: was similar to bankfull discharge (e.g., ???), so when bankfull discharge is :::: was not available, mean discharge is ::: was : used for analyses.For runoff-fed streams, if bankfull discharge is ::: was unavailable, 1.25-year return period is ::: was : used as an estimate for bankfull discharge.Statistics are :::: were repeated with and without estimated bankfull discharge.
Data are :::: were : modeled to determine which physical factors are most statistically related to stream width.We begin where w is stream width, l wood length, Q discharge, and c and : a, : b, :::: and : c : are constants.Models 3 and 4 appear nearly the same, but we fit them separately since model 4 requires fewer fit parameters.These formulae align with the body of research that confirms a power law relationship between stream width and discharge, while taking into account a power law or linear relationship between wood length and stream width for spring-fed streams.We assess ::::::: assessed : the value of candidate models using adjusted R 2 (?), which accounts for the number of predictive variables included in the model, and Akaike's Information Criterion (AIC), which measures the amount of information lost when data are approximated by a given model as compared to other candidate models also accounting for the number of predictive variables (?).An adjustment for small sample sizes (AICc) is ::: was : presented by ?, which we use :::: used in this study.If the set of AICc values is {AICc i }, then the probability that model i is the best of a set of candidate models is given by e (min({AICci})−AICci)/2 .

Wood Dynamics
We begin with a description of the observed wood dynamics within the studied streams.In order for single logs to drive changes in morphology, we assume that logs must be immobile in the channel.In order to confirm that this is the case in spring-fed, but not runoff-fed, streams, we examined ::::::: examine : histograms of wood orientation.
Following ?, we note that from the histogram of aggregated data for spring-fed streams in Figure ?? (a), it appears that wood is preferentially oriented around 50-90 • (see supplement for individual stream histograms).If wood were mobile in streams, we would expect to see preferential orientation at 0-20 • (?).We compare the histogram for spring-fed streams to that for runoff-fed streams in this study, where wood is preferentially oriented around 0-20 • .While the aggregate histograms exhibit clear results, many individual histograms demonstrate differences from these trends (see supplement).We considered whether basin size impacted the results since larger basins tend to transport more wood (?), but that observation does not explain the data aberrations ::::::::: differences.For instance, Chick Creek, ID (a spring-fed stream), contains wood mostly oriented around 0 • or 50 • , while Moose Creek, Deer Creek, and Buck Springs Canyon (runoff-fed streams) show random orientation, and Boulder Creek (runoff-fed) is preferentially oriented around 30-50 • .In Chick Creek, LW is significantly longer than the width of the stream, so the flow regime in the channel may have little impact on the orientation of wood.In the runoff-fed streams, the deviations from the trend are likely due to other aspects of wood dynamics noted during data collection.First, most wood observed in runoff-fed streams was found in logjams, and identifying single logs to measure the orientation was difficult.In runoff-fed streams in this study, there were on average 37 pieces of single wood per km as opposed to the 130 pieces of single wood per km found in spring-fed streams, as shown in Figure ??.The high density of single logs means that LW is closely spaced in the streams in this study.This disparity also prevented us from collecting as much data in certain streams due to a dearth of single logs.We noticed about 5 logjams per km in runoff-fed streams compared to about 1 per km in spring-fed streams.This indicates that there may be a bias toward new wood when measuring single pieces in some runoff-fed channels since older wood may be moved to logjams already.This also led to more difficulty in measuring orientation of single logs in some runoff-fed channels when multi-threaded channels made determining orientation with flow more difficult.
We verify conclusions about residence of LW by examining imagery from multiple dates on the streams marked by an a in Table 1.Imagery data were clear for a period of 3-10 years, depending on the site, and we examined at least 20 pieces of LW at each site.In each spring-fed stream, we were unable to detect any changes in wood placement at any site.In all of the runoff-fed streams except for Buck Springs Canyon, AZ, we observed a change in orientation or location for at least one  observed piece of LW.We suggest that no large run-off events occurred during the 3-year period for which clear imagery are available at Buck Springs Canyon.We thus confirm that there is little mobility of wood in the spring-fed streams in this study, distinct from the motion observed in runoff-fed streams.
Boxplot representing the number of (a) single logs and (b) logjams identified per km via satellite imagery on spring-fed and runoff-fed streams.

Discharge and Width
A common relationship used to describe stream width is the Leopold power law relating width w and discharge Q by constants a and b (?): w = aQ b .Typically, the value of b is close to 0.5, but b can vary depending on the streams being analyzed (?). ?
found b = 0.57 for the spring-fed streams in their study.The finding of ?suggests that discharge impacts the width of streams in their study to a similar degree as for most channels.We verify the result of ?for the streams in their study by finding b = 0.55 ± 0.1.
For the full set of spring-fed streams in this study containing wood, we find that a = 9.9 ± 1.2 and b = 0.42 ± 0.09 with a Pearson correlation coefficient of 0.52.Spring-fed streams without wood are fit by a statistically different trendline given by a = 14.4 ± 1.4 and b = 0.67 ± 0.08 with a Pearson correlation coefficient of 0.87.Runoff-fed streams are significantly different from spring-fed streams containing wood only in the coefficient a, with a = 5.1 ± 1.1 and b = 0.36 ± 0.03 with a Pearson correlation coefficient of 0.89 (When :::: when : repeated without estimated bankfull discharges, the results are statistically indistinguishable except an increase in R 2 to 0.99).The value of a is significantly smaller for the runoff-fed streams than the spring-fed streams included in this study.This corresponds to much narrower widths for the runoff-fed streams, confirming the results of ?.It is also noteworthy that the correlation coefficient for spring-fed streams with wood is much lower than for the other two groups, indicating that there is another very important factor needed to describe width adequately.indicating that runoff-fed streams are narrower than spring-fed streams at the same bankfull discharge.All runoff-fed streams contain wood, and no runoff-fed streams without wood were available for comparison.

LW and Width
We compare the stream widths we measured to those measured by ?for the subset of streams included in both studies.For all of the streams contained in both studies, the widths measured by ?fall within the confidence interval for the widths measured in this study via remote sensing.
For the 25 spring-fed streams containing wood, we find that there is a power law relationship between LW length and stream width, as shown in Figure ?? (b), with a Pearson correlation coefficient of 0.66.For streams lying below the dashed  AICc.The results from Adjusted R 2 match very well with the AICc results in ranking.For both runoff-fed and spring-fed streams, we note that models 3, 4, and 5 are essentially identical when fit for all streams since parameters a and c in model 5 are indistinguishable from 1.

Using LW and Discharge to Describe Stream Width
There are comparably large Pearson correlation coefficients for the relationships between wood and width as well as discharge and width for spring-fed streams, implying both are important descriptive factors for stream width.There is, however, a ln-ln correlation between discharge and wood length with Pearson correlation coefficient of 0.44, indicating that the two parameters do not contain totally unique information but do contain a significant amount of unique information.Since discharge and wood length are both significant descriptors for stream width and contain unique information, we examine a model for stream width incorporating both parameters.Full results for all tested models are shown in Table ??.For all cases, model ranking is very similar for AICc and Adjusted R 2 .
For all spring-fed streams, model fittings of parameter a in model three and a and c in model 5 are indistinguishable from 1, making models 3, 4, and 5 nearly identical, so we discuss only models 1, 2, and 4. Model 4 performs significantly better than models 1 and 2, as demonstrated by a high adjusted R 2 and a low AICc value in Table ??, although there : .::::: There : is still a significant probability that model 1 or 2 could be the most effective model (36% and 24 :: 23% respectively).This is unsurprising given that , ::::::: though, ::::: since models 1 and 2 resemble model 5 very closely.For spring-fed streams with an average width less than 30 m (the group of streams which are close to or narrower than available LW), models 3 and 4 are indistinguishable and models 2 and 5 are indistinguishable, so we discuss only models 1, 2, and 3. Model 1 (based only on discharge), drops in significance from an adjusted R 2 of 0.25 to 0.16 while all other models rise in significance, most notably model 2 which rises Figure ?? shows a distinction between spring-fed streams with and without wood in the relationship between discharge and width.There is, however, only a small set of data points available to identify the relationship for spring-fed streams without wood, and 5 of the 12 streams in this group are unusually narrow for the study group.The remaining points are not visually distinct from the pointcloud :::: point ::::: cloud for spring-fed streams with wood.For the streams in the Ozarks and Eastern Idaho, we speculate that these streams may once have had significant amounts of wood due to their size, location in wooded areas, and a history of "management" that may have included wood removal (????).If this is the case, then the presence of wood may have had a lasting impact on the channel morphology that is still measurable despite the present lack of wood, explaining why those streams lie in the point cloud for streams containing wood.While many, if not all, streams in the study may have been subject to wood removal at some point, we take the current wood load as representative of the type of wood dynamics that would have existed prior to wood removal.Additional management is not expected to have had much impact on results since geomorphic restoration efforts are typically not attempted over large reaches such as those used in this study (?).
In contrast to the U.S. streams : , the El Tatio streams , are above the treeline so would not have had wood in the past.It is possible that the channels were shaped by a different hydrological regime, but the streams run through glacial outwash, so the shape of the channel is dynamic and is probably controlled by the contemporary, spring-fed fluvial regime.Including all spring-fed streams in calculating the relationship between stream width and discharge does not significantly change the relationship parameters.This finding indicates that we are unable to reliably distinguish between spring-fed streams with wood and those without, an analysis which may be confounded by the minimal availability of spring-fed streams without wood for data collection.
There is, however, a robust distinction between spring-fed and runoff-fed streams in terms of the relationship between discharge and stream width, demonstrated in the fitted parameter a.This parameter indicates that for streams larger than those measured by ?, it is generally the case that spring-fed streams are wider than runoff-fed streams.

LW and width
We expect wood to be most important for describing the width of streams when it is comparable in size to the streams.When wood is much longer than the width of the stream, then additional increases in wood length do not change the way wood interacts with the channel since the majority of the wood piece is outside of the channel since nearly all wood observed in spring-fed streams is oriented closer to perpendicular to the bank than parallel, causing wood to either span the channel or interact with the channel only for part of the LW length.?note that LW longer than 2.5 times the channel width are : is generally immobile.While LW is immobile, though, the full length of the LW is relatively unimportant for its impact on stream width beyond the fact that it is longer than the channel is wide.Conversely, when the stream is much wider than the wood, LW can only be close to the bank on at most one side of the stream.?found that when LW at a given orientation is closer to the bank, the impact on shear stress is greater.Taking distance from the bank as the most important predictor of how important a single log is in altering channel properties, then decreasing the size of LW after a certain point does not change the ability of the wood to be close only to one bank.Thus, we expect LW to be less important in two cases: 1) where streams are very narrow in comparison to LW length and 2) where streams are very wide in comparison to LW length.In other words, when discharge is outside a certain range, we expect the impact of LW on stream width to decrease since channels are either very wide in comparison to wood length or very narrow.We see visually in Figure ?? (a) that when streams are wider than about 25 m, the points deviate significantly from the otherwise apparently linear trend.For streams in this study wider than 30m, streams are much wider than the LW found in or near them.In fact, we find that there is a linear relationship with Pearson correlation coefficient of 0.75 for streams smaller than 30 m wide, more significant than the ln-ln relationship for all data.This stronger correlation aligns well with our hypotheses about when wood should have an impact on stream morphology, i.e. when LW is comparable in length to stream width.While we are unable to say with confidence whether or not there is a difference between spring-fed streams with or without wood, we find that deviation from the relationship occurs where expected if wood were driving the relationship.
In the case of runoff-fed streams, although the best fit matches closely with that for spring-fed streams, we find it likely that this relationship does not hold in general for runoff-fed streams.Since there is a strong bias in our set of runoff-fed streams toward high-discharge streams, with over 70% of the runoff-fed streams exhibiting a discharge higher than 5 m 3 /s and most over 50 m 3 /s, it may be coincidence that the runoff-fed streams included in this study are about as wide as the wood found in them.The difficulty in identifying runoff-fed streams in geologic settings in which spring-fed streams occur prevents us from assessing more fully the relationship between wood length and stream width in runoff-fed streams in a comparable geologic setting.

Figure 1 .
Figure 1.Orientation of wood was measured from adjacent bank for approximately 100 pieces of wood using Google Earth Pro (green).?measured orientation of wood in the field (transparent black).Data are shown together for (a) Cultus River and (b) Cultus Creek.The distributions align very well for Cultus River and are have the same qualitative shape for Cultus Creek, although the center peak is displaced between the two sets of measurements.

Figure 2 .
Figure 2. Using google ::::: Google : Earth Pro, orientation of wood was measured from adjacent bank for approximately 100 pieces of wood in each stream which had clear enough imagery to reliably identify LW (marked by a † in Table ??).Histogram data are aggregated for (a) spring-fed and (b) runoff-fed streams.Wood in spring-fed streams is preferentially oriented from 50-90 • , whereas wood in runoff-red streams is more randomly oriented with a significant portion of wood oriented 0-20 • .

Figure 3 .
Figure 3. Google Earth Pro high-resolution imagery showing (a) Cultus River (q95 = 2.8 m 3 /s) and (b) Cultus Creek ((q95 = 2.3 m 3 /s).Stream channels are outlined in white, and flow direction is down from the top of the image in both panels.These images are representative of the general wood dynamics in the two streams, where most of the wood in (a) is single logs, and most of the wood in (b) is in logjams, so little of the wood in panel (b) would contribute to the histogram shown in Figure ?? (b).

Figure 5 .
Figure 5. Relationship between bankfull discharge (or 1.25 year flow as an approximation to bankfull discharge for some runoff-fed streams marked in Table ??) and stream width plotted on a ln-ln plot for spring-fed streams with wood (dark green), spring-fed streams without wood (orange), and runoff-fed streams (light blue).The line of best fit for streams containing wood is shown (w = aQ b , b = 0.42 ± 0.09, a = 9.9 ± 1.2); 95% confidence interval for the fit is shaded.Stream types are denoted by color, as shown in the top left, and locations are denoted by shape, as shown in the bottom right.Runoff-fed streams are fit by a statistically significant different value of a = 5.1 ± 1.1,

Figure 6 .
Figure 6.Wood length and stream width were measured using Google Earth Pro satellite imagery.The relationship between wood length and stream width for spring-fed streams is shown (a) on a plot with width = length shown as a dashed line and error bars showing the standard deviation and runoff-fed streams marked with black dots (error bars left off for clarity of viewing) and (b) on a ln-ln plot with the line of best fit (w = al b with b = 2.4 ± 0.4 and a = 0.04 ± 0.03), error bars and runoff-fed streams left out for clarity.The 95% confidence intervals for the line of best fit is shaded.In both panels, the data symbols represent the geographic locations of the streams.There is no apparent significant clustering by location.In panel (a), streams that fall above the dotted line are wider than the wood load entering the streams, whereas the streams falling below the line are narrower than the wood load.

Table 2 .
Fit statistics for candidate models for spring-fed and runoff-fed streams.Adjusted R 2 and Akaike's Information Criterion (AICc) account for the number of predictive variables.A larger R 2 value indicates better fit, while a smaller AICc value indicates that less information is lost.The AICc Probability is the likelihood that a given model is the best model based on the criterion of lost information as measured by