New remote method to systematically extract bedrock channel width of small catchments across large spatial scales using high‐resolution digital elevation models

Bedrock river width is an essential geometric parameter relevant to understanding flood hazards and gauging station rating curves, and is critical to stream power incision models and many other landscape evolution models. Obtaining bedrock river width measurements, however, typically requires extensive field campaigns that take place in rugged and steep topography where river access is often physically challenging. Although prior work has turned to measuring channel width from satellite imagery, these data present a snapshot in time, are typically limited to rivers ≥ 10–30 m wide due to the image resolution, and are physically restricted to areas devoid of vegetation. For these reasons, we are generally data limited, and the factors impacting bedrock channel width remain poorly understood. Due to these limitations, researchers often turn to assumptions of width‐scaling relationships with drainage area or discharge to estimate bedrock channel width. Here we present a new method of obtaining bedrock channel width at a desired river discharge through the incorporation of a high‐resolution bare‐earth digital elevation model (DEM) using MATLAB Topotoolbox and the HEC‐RAS river analysis system. We validate this method by comparing modeled results to US Geological Survey (USGS) field measurements at existing gauging stations, as well as field channel measurements. We show that this method can capture general characteristics of discharge rating curves and predict field‐measured channel widths within uncertainty. As high‐resolution DEMs become more available across the United States through the USGS three‐dimensional elevation program (3DEP), the future utility of this method is notable. Through developing and validating a streamlined, open‐source, and freely available workflow of channel width extraction, we hope this method can be applied to future research to improve the quantity of channel width measurements and improve our understanding of bedrock channels.

place in rugged and steep topography where river access is often physically challenging. Although prior work has turned to measuring channel width from satellite imagery, these data present a snapshot in time, are typically limited to rivers ≥ 10-30 m wide due to the image resolution, and are physically restricted to areas devoid of vegetation. For these reasons, we are generally data limited, and the factors impacting bedrock channel width remain poorly understood. Due to these limitations, researchers often turn to assumptions of width-scaling relationships with drainage area or discharge to estimate bedrock channel width. Here we present a new method of obtaining bedrock channel width at a desired river discharge through the incorporation of a high-resolution bare-earth digital elevation model (DEM) using MATLAB Topotoolbox and the HEC-RAS river analysis system. We validate this method by comparing modeled results to US Geological Survey (USGS) field measurements at existing gauging stations, as well as field channel measurements. We show that this method can capture general characteristics of discharge rating curves and predict field-measured channel widths within uncertainty. As high-resolution DEMs become more available across the United States through the USGS three-dimensional elevation program (3DEP), the future utility of this method is notable. Through developing and validating a streamlined, open-source, and freely available workflow of channel width extraction, we hope this method can be applied to future research to improve the quantity of channel width measurements and improve our understanding of bedrock channels.

K E Y W O R D S
bedrock rivers, channel geometry, channel width, LiDAR, river incision 1 | INTRODUCTION River channel geometry measurements (e.g., slope and width) are essential in scientific and societal applications. Among other applications, channel geometry measurements are applied in hydrologic models (Neal et al., 2015;Yanites, 2018), stream restoration studies (Kondolf et al., 2002;Nelson et al., 2015), and sediment transport models of alluvial rivers (Merritt et al., 2003;Prosser et al., 2001).
They are further used to explain bedrock channel sinuosity (Turowski, 2018), and implemented in bedrock-channel incision models to provide insight into the topographic, morphologic, tectonic and climatic history of an area (Allen et al., 2013;Finnegan et al., 2005;Fisher et al., 2012Fisher et al., , 2013. Performing these calculations, especially at a large spatial scale, requires knowledge of how bedrock channel width changes across the landscape. Analyses frequently assume spatial changes in channel width via scaling relationships between channel width and other more easily or typically measured proxies, such as drainage area or discharge (DeLong et al., 2007;Knighton, 1998;Whipple & Tucker, 1999;Willett, 1999). Although empirical data support such scaling relationships, most measurements are from alluvial river systems, and the data used to derive the scaling relationships shows significant scatter (Whipple et al., 2022).
Because of these limitations, researchers have recently taken advantage of the increasing accessibility to satellite imagery and remote sensing data to derive channel width measurements of large river systems (typically > 30 m channel width) and some smaller river systems with minimal vegetation cover across broad spatial scales (e.g., Allen et al., 2013;Fisher et al., 2012Fisher et al., , 2013Lin et al., 2020;Yamazaki et al., 2019). Although valuable, these data sets extract measurements from one snapshot in time (not necessarily during comparable flow periods), typically use imagery that is limited in resolution (≥ 10-30 m), and are restricted to performing analysis where vegetation cover does not impede the view of the river. Therefore, for all other scenariossmall river systems (e.g., areas with channel widths < 30 m) and climates that allow for moderate to heavy vegetation growthresearchers largely rely on width measurements from field campaigns that are time-consuming, labor-intensive, often spatially limited in extent (tens of km 2 ), and logistically challenging due to limited river access. As a result, field width measurements of bedrock channels, often found in steep, remote, and hard-to-access landscapes, are sparse, forcing researchers to generally embrace the use of scaling relationships derived from alluvial river systems to estimate channel width of bedrock rivers (Whipple et al., 2022). The use of such scaling relationships, however, might be inappropriate for bedrock channels. Bedrock channel width is a function of a multitude of factors beyond discharge (or, by proxy, drainage area), including rock uplift rate, bedrock lithology and rock strength, sediment supply, hydraulic roughness, vegetation, slope, and climate (Duvall et al., 2004;Finnegan et al., 2005Finnegan et al., , 2007Spotila et al., 2015;Walsh et al., 2012;Whittaker et al., 2007). Such studies highlight a current incomplete understanding of controls on bedrock channel width and the need to develop new remote sensing methods capable of measuring bedrock channel width across large spatial scales.
Existing studies highlight the need for more field data on channel width in bedrock channels, which would enable a better understanding of the factors that control channel width (Finnegan et al., 2005;Whipple et al., 2022). However, even when field channel width measurements exist, these data can be misleading for a number of reasons. As a dynamic stream variable, channel width measurements can differ based on both the channel geometry and changes to the flow magnitude (i.e., discharge). As a result, field measurements of channel width (a.k.a. wetted width, which we refer to as channel width or bedrock channel width herein) must be acquired or referenced to the same flow magnitude. These flow conditions, however, can change in both space and time across a field campaign, limiting the ability to accurately correlate between the various measurements. Most bedrock channel studies do not know the reference flow conditions at the time of their field campaigns, and therefore turn to measuring "bank-full" channel widthdefined by Wohl and Merritt (2005) as the wetted channel width during "discharge that recurs on average every 1-2 years". Because field data is typically not collected during such discharge conditions or streams are not gauged, and this reference discharge is unknown, the bedrock channel width is often measured using assumed "bank-full" indicators such as changes in bank geometry, a lack of vegetation, scour marks, or stains observed on clasts or bedrock (Lague, 2014;Turowski, 2018;Wohl & Merritt, 2005;Zondervan et al., 2020). However, where these "bankfull" measurements are positioned in bedrock channels can often be variable based on the subjective opinion of a particular observer or field personnel and differences in climate and speed of vegetation regrowth, likely leading to inconsistent results across different landscapes. Due to the ambiguities associated with collecting bedrock channel width field measurements, it becomes difficult to fully standardize, compare, and understand field measurements of channel width based on data derived from numerous field campaigns, researchers, and climates. These underlying limitations to the field measurements highlight the need to derive new and unbiased methods for determining comparable values of bedrock channel width for moderate to small river systems across large spatial scales.
In this research, we overcome these limitations by developing and presenting a new method to determine bedrock channel width that leverages the recent growth and availability of high-resolution bareearth digital elevation products (e.g., LiDAR [light detection and ranging]) to fill this data gap. We introduce a method for systematic (objective and under similar flow conditions), high-resolution bedrock channel width extraction. Recently, Bernard et al. (2022) implemented a similar method, employing a two-dimensional (2D) morphohydrodynamic model to systematically extract channel width measurements from an approximately 17 km 2 drainage basin using LiDAR.
Although demonstrating promising results, the model output is effective flow width (i.e., width of channel section with most flow for a given runoff intensity). While generally comparable to wetted width, which is commonly reported in other studies, at low runoff intensity in steep confined channels, effective width is not consistent with wetted width for channels wider than $5 m and for high runoff intensities and low gradient reaches. Furthermore, this approach was not benchmarked for accuracy relative to field measurements.
In our methodology, we employ TopoToolbox in MATLAB in concert with a freely available and widely-used hydrology modeling software (HEC-RAS) and a high-resolution digital elevation model (DEM) to perform this analysis in a way that does not require extensive coding experience from the end-user. To facilitate the application of this method, we present a general workflow of the analysis and a more indepth user guide in the Supporting Information (Methods S1). We validate this method by comparing model results with US geological Survey (USGS) gauging data and field measurements. Next, we automate this process in a case study of a watershed in Puerto Rico (Figure 1b,c). We chose to perform analysis in Puerto Rico due to the availability of high-resolution ($1 m) LiDAR across the island (OCM Partners, 2022), because it is located in a tropical landscape that is not impacted by snowfall, the relatively high density of existing USGS gauging stations, and because its mountain rivers are generally sediment supply-limited bedrock rivers. Our analysis in Puerto Rico enables us to efficiently glean hundreds of channel width measurements at desired locations along a stream network, across different bedrock types, and at various elevations. By validating this method and presenting its effective application, we hope to open avenues for researchers to extract bedrock channel width data more safely and efficiently from the watershed-to-landscape scale, and provide a process that can significantly increase the quantity of bedrock channel width measurements in small (< 30 m wide) bedrock rivers.
Although we acknowledge that deriving channel geometry from highresolution DEMs has its limitations (e.g., DEM data should be acquired during a low-flow regime), this method provides a path forward to improving data richness and our understanding of bedrock channel width by taking advantage of the ever-increasing accessibility of global LiDAR data. In the United States, for example, the USGS threedimensional elevation program (3DEP) provides a freely-available, high-resolution LiDAR data set across the continental United States. year-round, with temperatures typically ranging between 70 F and 90 F (The Southeast Regional Climate Center, 2019). Northeast trade winds that cross the island generally determine precipitation and climate and are modified by tropical depressions, storms, and hurricanes (Calvesbert, 1970;Ehlmann, 1968). Precipitation patterns are orographically controlled, with the high-elevation north-eastern portion of the island receiving the most precipitation (mean annual precipitation of over 4000 mm), whereas the flatter south-western portion of the island tends to be the driest (with an average annual precipitation of approximately 732 mm) (Figure 1b; NOAA, 2011).
Topography varies dramatically across the island and can be classified into three physiographic regions: the mountainous interior, the coastal lowlands, and the karst area (US Geological Survey, 1998). The mountainous interior is comprised of the Cordillera Central, which is characterized by highly deformed volcanic rocks and less deformed or undeformed plutonic rocks (mainly volcaniclastic and granodiorite, respectively) formed in the Cretaceous to Paleocene and early Eocene (US Geological Survey, 1960) (Figure 1c). Towards the north and south of the island, the older volcanic rocks are overlain by younger (Tertiary in age), gently-dipping carbonates and sedimentary rocks (Schellekens, 1998). Reef Carbonates, deposited in the Oligocene to early Pliocene, which are generally located in the northern portion of the island, form the rugged and distinct karst topography (Moussa et al., 1987). In contrast, the coastal lowlands in the southern portion of the island are composed of Miocene to Quaternary sedimentary rocks that form a gentle topography (Volckmann, 1984).
As noted earlier, we chose to test our new method in Puerto Rico for a variety of reasons. First, the island contains a high density of USGS monitoring stations (Figure 1b,c), with 73 gauging stations that monitor a cumulative drainage area of > 7000 km 2 (equivalent to approximately 80% of the island area). Many of these stations further offer temporally continuous data, with some stations providing > 60 year records (Supporting Information Table S1). Second, Puerto Rico's mild, subtropical climate offers an area devoid of snow. This was an important consideration, as snow can alter how precipitation is temporally and spatially received and processed by the landscape.
Although Puerto Rico's climate is characterized by a "dry" and "wet" season, the relatively consistent temperature across the year ensures minimal impact of a more extreme seasonality seen across different areas of the world. Finally, the crystalline rock types (the focus of this study) that comprise most of the mountainous topography on the island are defined by predominantly bedrock channels where sediment transport capacity exceeds sediment supply.

| HEC-RAS
To efficiently route flows of desired discharges across channel networks derived from a LiDAR DEM of Puerto Rico, we turned towards the freely available HEC-RAS River Analysis System developed by the Hydrologic Engineering Center for the US Army Corps of Engineers, and widely used across academic and professional river modeling fields. Due to minor bugs found in the more current version of HEC-RAS, we used version 5.0.7 (released in March 2019) for this analysis.
Within HEC-RAS, we use a one-dimensional (1D) Steady Flow hydraulic model rather than more computationally-expensive 2D models because (1) the streams we analyzed had a dominant flow direction with a known general flow path, (2) the streams are mainly characterized as steep bedrock channels with minimal overbank areas, and (3) our analysis lacked detailed channel bathymetry information since we were using DEMs acquired through near-infrared LiDAR (1064 nm), which cannot penetrate through water. This reasoning and choice of employing a 1D hydraulic model is supported by the HEC-RAS 2D Modeling User Manual (Brunner, 2021). In addition, it is important to note that a 1D hydraulic model is substantially more computationally efficient and typically has fewer unconstrained parameters than the more complicated 2D models, allowing us to perform modeling simulations more efficiently across many locations in a larger landscape. Furthermore, as shown later (Section 3), the simpler 1D model performs well in predicting observed data, suggesting that the additional computational expense of running a 2D model might not add much value.
The 1D Steady Flow model in HEC-RAS generally uses the Energy equation to compute the energy grade line and water surface elevations across desired locations of interest under a given discharge scenario. Data requirements to run the model include geometric data and flow data. Geometric data is acquired by HEC-RAS through the input of a DEM (e.g., terrain model), river channel shapefiles (providing the river path, flow direction, reach length, river system schematic), and cross-sectional shapefiles (identifying the location and extent of desired cross-sections, drawn perpendicular to the flow lines).
Because of this, we strongly recommend that care is taken to ensure that the DEM used with this method was acquired during low flow conditions (i.e., not in the days and weeks following a storm or during seasonal high flow regimes) to minimize the impact of water from significantly obscuring the channel geometry.
Flow data is entered into the program to establish the flow regime and boundary conditions. Because we are mainly modeling subcritical flow conditions, we define a water surface at the downstream end of the river network, which allows the hydraulic model to proceed with calculations in the upstream direction. When the downstream water surface is unknownwhich is the case for our calculationsit is recommended to use an estimated water surface elevation above the channel bed. To avoid the impact of the downstream water surface boundary condition, it is recommended to extend the river network so that the study reach is well upstream (Figure 2; USACE, 2016). This allows model results to converge to a consistent answer once the computations reach the upstream study area (USACE, 2016). An additional model flow input includes the Manning's n channel roughness coefficient, for which we used a n = 0.05, representative of "normal" values for mountain streams devoid of inchannel vegetation, characterized by steep banks, and with channel bottoms comprised of mainly cobbles and large boulders (USACE, 2016). Based on field observations, we found this roughness value to best characterize bedrock channels in Puerto Rico, but also want to highlight the ability to easily modify this value based on variable types of river channels in different environments or locations (see Methods S1). As a final flow input, we entered a desired water discharge for each river reach, which the 1D Steady Flow model assumes remains constant across the length of the reach.
A breadth of output parameters can be extracted from each model run, including average flow velocity, hydraulic radius, mean flow depth, shear stress, and total stream power. For this analysis, we F I G U R E 2 A sketch based on figure 3-1 in the HEC-RAS Hydraulic Reference Manual, showing the placement of the study reach location (e.g., data values used) upstream of the starting boundary conditions, thereby limiting the effect that initial boundary conditions (such as an incorrect water surface elevation estimate) can have on the used data values (USACE, 2016) [Color figure can be viewed at wileyonlinelibrary. com] mainly focused on the wetted channel width across a specific location of the reach, and thus primarily extracted this result across all crosssections and river reaches.

| TopoToolbox in MATLAB
We used TopoToolbox version 2an open-source MATLAB-based softwareto extract field-specific information and manipulate data inputs in preparation for HEC-RAS modeling. More specifically, Top-oToolbox version 2 (Schwanghart & Scherler, 2014), enables us to easily combine elevation data, geologic maps, and precipitation data into one set of layers, which in turn lets us systematically choose the specified modeling locations based on desired characteristics. Key TopoToolbox components used in MATLAB included deriving the stream network from a LiDAR DEM by filling sinks and calculating the drainage network upstream of a given point. Through spatially extracting this information using TopoToolbox, our analysis is more flexible, as we can combine and overlay this data with existing USGS gauging discharge data, to better calculate parameters, such as discharge, that we then implement into the HEC-RAS model. Lastly, Top-oToolbox and MATLAB were instrumental in automating our method, providing a base for us to develop a code that automatically created river network and cross-section shapefiles that could then be directly uploaded into HEC-RAS with limited user interference.

| PROOF OF CONCEPT AND APPLICATION IN PUERTO RICO
In this article, we explore a new method of obtaining channel width measurements remotely across a range of drainage areas. We use a LiDAR-derived DEM with 1 m horizontal resolution acquired in July 2018, during a relatively dry month (National Weather Service, 2018; OCM Partners, 2022). To demonstrate the method's utility and accuracy, we first validate the method results by comparing these data with USGS channel measurements acquired at existing USGS gauging stations across different discharge scenarios. To bolster this analysis, we then conduct an island-wide comparison (providing for a large range in drainage areas) of channel width measurements from USGS gauging stations with modeled results under mean annual discharge scenarios.
A third way that we validate our method is by comparing model results to our field measurements. Because these field measurements are not acquired near gauging locations, we explore the utility of a simple approach to estimate discharge of a given magnitude based on correlations with upstream mean annual precipitation. Having validated the method, we lastly examine the efficiency of the model by conducting analysis at a catchment scale, and perform first-order exploratory analysis of the results to demonstrate how these channel width measurements fit into the context of existing research of bedrock channels.

| Method validation using USGS gauging stations
To evaluate the accuracy and utility of the derived method, we compared modeled width measurements to USGS field measurements collected at 49 of the 73 gauging stations across Puerto Rico (Figure 1b,c). We eliminated stations draining carbonate bedrock to avoid areas potentially impacted by karst hydrology, and stations directly downstream of dams where flow is modulated, and the mean annual discharge (MAQ) does not reflect natural changes in precipitation. We chose to use the USGS field measurements of channel width, as they are field data that have undergone a quality control process and are deemed of sufficient quality to be incorporated into the rating curve of the respective gauging station. In addition, these measurements included relevant datanamely channel width measurements across different discharges.
Initially, we used the mean daily discharge values across the time of record for each station to determine the MAQ at each gauged location that meet our criteria. Next, we analyzed the USGS field width measurements, which provide insight into changes to the channel width at the same location across different discharge scenarios. Performing such verification across a range of discharge scenarios is an important aspect of validating the utility of the model, as width intrinsically varies with discharge; as discharge increases, so too does channel width. This relationship is expressed in the following power-law equation by Leopold and Maddock (1953).
where w denotes channel width, Q is water discharge, and a and b are empirical constants. As a result, discharge has a first-order impact on the resulting modeled channel width.
Because the USGS field measurements were collected both close to the gauge location and from distances of up to > 700 ft from the gauging station, and our field observations at numerous gauging sta-

| Model validation with field measurements
To further validate our method, we sought to compare our model results to field measurements collected across the island, largely at ungauged locations. During two field campaigns in March 2020 and January 2022, we obtained 165 channel width measurements at 27 field locations, with an average of six measurements at each location (Figure 1a,b). In the field, we used a laser rangefinder to measure the approximate "bank-full" channel width, which we assumed to be marked by a lack of vegetation along the channel banks. Considering the fast vegetation regrowth in Puerto Rico, we presume that our channel width field measurements reflect vegetative stripping and erosion that occurs during a discharge with a 1 to 2-year reoccurrence interval (RI), and less frequently than the MAQ. Because discharge has a first-order impact on channel width, it is important to consider channel width across a similar reference flow.
Deriving this reference flow, however, can be challenging at ungauged locations. Previous studies have used satellite-based measurements of mean annual precipitation (MAP) as a proxy for calculating discharge at ungauged locations (Desormeaux et al., 2022;Rossi et al., 2016).
Comparing MAPcalculated by averaging the mean annual precipitation between 1963 and 1995 obtained from the PRISM data set (PRISM Climate Group, 2022) across the respective location's drainage basinto the MAQ at the known USGS gauging locations reveals that MAP is a good proxy (r 2 = 0.82) for predicting MAQ in Puerto Rico when assuming a power-law relationship ( Figure 5, Supporting Information Figure S1). However, this strong positive relationship between the two parameters remains predictive but weakens when MAP is compared to the 1-year and 2-year return interval discharge magnitudes at these locations ( Figure S2). As completed by Rossi et al., 2016, precipitation probability can be used with discharge frequency to improve this relationship, yet even in these scenarios, the strength of the relationship appears to degrade with increasing RI (Rossi et al. [2016] found an r 2 = 0.48 for 2-year RI events). With this in mind, we may expect our 1 to 2-year RI discharge estimates derived from MAP to underestimate the true 1 to 2-year RI discharges at the field locations and the modeled channel widths using these estimated discharge data to produce lower channel width values.
For comparison between modeled channel width and our field measurements, we chose to use the average channel width at the point of measurement because (1)  field width measurements shows that MAQ-based channel width modeling results are consistently lower than field measurements ( Figure 6a,d). This supports our presumption that the field measurements reflect 1 to 2-year RI flow conditions rather than MAQ conditions. Although the modeled width measurements only fall within the range of field measurements at 18 of the 27 locations, there appears to be reasonable alignment between the modeled and measured width values when measurement error is considered (Figure 6b,c,e,f).
Comparing modeled channel width with "bank-full" field measurements highlights the utility of this method in multiple ways. Primarily, it emphasizes the subjectivity of "bank-full" field measurements, as these measurements can be delineated differently based on the subjective opinions of what "bank-full" looks like to various field personnel, disparities in the rates of vegetation regrowth, among others. In addition, the recurrence interval associated with these width measurements is largely unknown but is assumed to fall within a 1 to 2-year time frame. The discharge associated with such recurrence intervals is more difficult to predict at ungauged locations.
However, analysis of the relationship between precipitation and discharge reveals that MAP can be used as a good predictor for MAQ (r 2 = 0.82), but it does not predict the discharge of 1 and 2-year RI as well (r 2 = 0.44 and 0.3, respectively). These results highlight the utility of using MAQ as a better predictor of the spatial pattern of discharge at ungauged stations. However, the comparison with our field measurements shows that the 1-and 2-year RI provide a better match to these data and suggests that our field measurements are consistent with typical "bank-full" definitions ( Figure 6b,c,e,f).

| Automating the method to enhance the quantity and spatial distribution of bedrock width channel measurements
In the previous sections, we validated the accuracy of our method's results by comparing modeled channel width measurements with both USGS gauging station data and our field measurements. These results show that our method, within uncertainty, reproduces field-based width measurements at a given reference flow condition for a reasonable range of discharge magnitudes (Figures 3a,b, 4, and 6). As a next step, we aim to use this method to gather channel width measurements remotely and efficiently with the intent of increasing the number of bedrock channel width measurements obtained across the world. The general steps of modeling channel width are outlined in We eliminated reaches that were close to the Caonillas Reservoir or those that were generated and less than 500 m from one another. For this analysis, we use MAQ as our reference discharge based on the relationship given in Figure 5 and a map of MAP, but we also ran similar simulations for the 1-and 2-year RI discharge magnitudes.
It is important to note that our modeled width resolution is sensitive to the DEM resolution ($1 m 2 ). Therefore, we consider width measurements that span twice the DEM resolution ($2 m) as reliable. For our MAQ reference flow condition, this width resolution corresponds to a minimum drainage area of roughly 1 km 2 , so we only analyzed F I G U R E 7 A general overview of the steps required to model channel width measurements. Bullet points outline relevant data products obtained from different inputs, or outline the general steps within a larger process. The hourglass symbols indicate the steps that take significantly more time ( river reaches ≥ 1 km 2 after running our automated analysis. However, we note that higher magnitude reference discharges yield resolvable width results at lower upstream drainage areas. After filtering our results based on these considerations, 116 stream reaches remained for analysis in the watershed. Preliminary data analysis reveals interesting channel width trends and other metrics across the watershed. Analyzed channel measurements cover a large range of drainage areas, spanning two orders of magnitude (1 km 2 to 104 km 2 ). Results show a strong positive relationship between drainage area and channel width, with an increase in drainage area associated with a wider channel ( Figure 9)a trend that has consistently been recognized in the literature (e.g., DiBiase & Whipple, 2011;Duvall et al., 2004;Whipple et al., 2022;Wohl et al., 2004). Interestingly, the exponent of a power law regression through the data, $0.47, is generally similar for the MAQ, 1-year, and 2-year recurrence interval reference discharges Channel width values correlate to the drainage area (Wohl et al., 2004), so we further compare channel width by using the normalized wideness index (k wn ), which is calculated by using Equation (2): where W is the channel width, A is the drainage area, and b ref is the reference wideness exponent (set to 0.47 based on our data) (Allen et al., 2013). Accounting for drainage area through the normalized wideness index shows that channels in this drainage basin that are underlain by granodiorite had, on average, narrower channels than those underlain by volcaniclastics and sedimentary rocks.
Comparing width to basin average elevation reaffirms that volcaniclastics are generally limited to the higher elevations of the watershed, whereas granodiorites and sedimentary rocks can be found in both low-and high-elevation areas (Figure 11a,b). Figure 11 (b) reveals that the variability of channel width is larger for rivers underlain by granodiorites and sedimentary rocks, whereas it is mostly confined to channel widths < 10 m for areas underlain by volcaniclastics.
Lastly , S1 in Wohl and David [2008]), most notably with Montgomery and Gran (2001) and Tomkin et al. (2003). Comparing our preliminary analysis of the Caonillas watershed to alluvial fits further suggests that, with an increase in drainage area, bedrock channels in Puerto Rico increase in width at a faster rate than both mixed bedrock-alluvial rivers and gravel-bedded rivers ( Figure 12).

| Implications and outlook
The coupled use of TopoToolbox with HEC-RAS provides a straightforward and relatively simple approach to derive channel width mea-  (Stoker, 2022).
Where high-resolution topographic data are available, an additional advantage is the affordability and simplicity of data acquisition requiring remotely sensed data, rather than expensive, risky, and time-consuming field campaigns. Gathering these data, therefore, becomes much more accessible for researchers with a limited budget.

| Method limitations
In light of the promising data results, we recognize that our presented method has a number of intrinsic limitations. Most importantly, our method relies on a bare-earth high-resolution DEM derived from LiDAR to measure the properties of river geometry. The near-infrared wavelengths are most frequently used for topographic LiDAR DEM F I G U R E 1 2 A modified figure from Whipple et al. (2022), overlain by data from the Caonillas watershed. Here, we chose to only include locations with modeled river width > 2 m (n = 68). These results suggest that channel width of bedrock rivers in the Caonillas watershed increase with drainage area is steeper than alluvial rivers and other mountain stream measurements collected across the world [Color figure can be viewed at wileyonlinelibrary.com] applications, however, they generally cannot penetrate through water and produce accurate measurements of the channel bottom (McKean et al., 2011). As a result, our method routes a given flow over the water surface of the river at the time that LiDAR was conducted. With this in mind, we need to (1) assume that LiDAR was not collected during high flows that can obscure channel geometry, (2) limit the analysis and interpretation of water depth in our modeling calculations, and (3) recognize that channel width measurements likely become less accurate for larger rivers, where bed roughness is much less than water depth at the time of acquisition. Although this method can still predict a reference channel width above the discharge during LiDAR acquisition, we cannot use it to accurately determine rating curves (such as shown in Figure 3) for these larger rivers. In addition, we recognize that the LiDAR DEM we used for our method has a resolution of $1 m. As a result, we only considered modeled river width measurements of ≥ 2 m, limiting our ability to capture width measurements of narrower, low drainage-area (c. 1 km 2 ) bedrock channels that may have width measurements close to this resolution. Although our preliminary analysis suggests that the method does a reasonable job at these small drainage areas (Figures 4 and 6), selecting only channels ≥ 2 m wide may bias the channel width measurements of small (approximately < 1 km 2 ) drainage areas. However, with higher resolution DEMs and higher magnitude reference flow conditions, this method can extend to lower drainage areas, thereby reducing this bias. Moreover, because we use pixel size (e.g., resolution) to calculate properties such as upstream drainage area and derive the river network, this method requires the DEM to have the same resolution throughout the study area. Lastly, we applied the MAP to estimate discharge for ungauged areas, but recognize that data results can be improved by using field discharge measurement methods (i.e., installing dataloggers to establish a stage-discharge rating curve, utilizing current meters, or employing artificial tracers) (Tazioli, 2011) or implementing water routing algorithms (Corato et al., 2011;Perumal et al., 2007).

| CONCLUSIONS
Through validating this workflow against both USGS gauging station data and field data, we have demonstrated the possibility of efficiently acquiring comparable bedrock channel width measurements across a large area without the need for expensive, often dangerous, and timeconsuming field campaigns. By the implementation of similar flow conditions (at mean annual discharge) across an area of interest, we show that this method further offers a way to reduce the subjectivity of field measurements and produce comparable results across different reaches and locations. As a result, we present an avenue for researchers to efficiently garner landscape-and watershed-scale bedrock channel width measurements that allow for comparisons among locations across the world that span a breadth of climates, vegetation abundance, river types, and flow regimes. Our preliminary analysis of the Caonillas watershed demonstrates the breadth of data that can remotely be acquired and analyzed through this process. By presenting a simple workflow and user guide in light of the increasing availability of high-resolution LiDAR data, we hope that this method can be used in the future to push forward the understanding in the literature of bedrock channel width, as well as improve the modeling capabilities of models, such as the bedrock incision model, that rely on channel width to produce reliable results.