Remote sensing of laboratory rivers

Remote sensing enables us to measure fluvial systems without disrupting their dynamics. Small‐scale physical models of rivers allow us to observe their geomorphic evolution, but we need remote sensing methods to monitor these laboratory landscapes without altering their flow or topography, just as with field‐scale rivers. In this paper, we review how experimental geomorphologists have adapted remote sensing for the laboratory. We consider how remote methods to monitor model topography, flow depth, velocity and planform have been employed, enabling uninterrupted experimental evolution. We also explore the transfer of techniques between field‐scale and experimental remote sensing; the controlled conditions in the lab aided the development of some methods, while others benefited from airborne deployment. We consider recent developments offered by laboratory remote sensing, including through‐water laser scanning and adaptations of structure‐from‐motion photogrammetry; we also consider new challenges associated with these developments, such as computational power. Finally, we discuss new research problems that laboratory remote sensing is opening up to geomorphology. We hope this review will be useful for experimentalists seeking to collect data remotely, continuously and/or cost‐effectively.


| INTRODUCTION
The remote sensing of rivers has brought fluvial geomorphology into the age of 'big data'.The advent of airborne LiDAR (light detection and ranging) generated datasets with millions of topographic points at sub-metre accuracy.By permitting spatially continuous topographic surveying, LiDAR opened new questions, heralding the discovery of new fault systems and the fine-scale surveying of rivers and tidal channels (Bowen & Waltermire, 2002;Haugerud et al., 2003;Lohani & Mason, 2001).More recently, the development of structurefrom-motion (SfM) photogrammetry has enabled cheap, rapid and centimetre-scale data acquisition that lends itself to the repeat monitoring of river topography (Fonstad et al., 2013;James & Robson, 2012;Javernick et al., 2014;Westoby et al., 2012).Such repeat surveys have led to major advances in the understanding of upland and lowland river morphodynamics (Cook, 2017;Cucchiaro et al., 2018).At larger spatial extents, a multi-decadal record of satellite data is beginning to allow the tracking of river morphology, migration, avulsion and seasonality at regional or even global scales (e.g., Allen & Pavelsky, 2015, 2018;Dewan et al., 2017;Donchyts et al., 2016;Pekel et al., 2016;Pickens et al., 2020).Remote sensing has transformed the questions we can ask, as well as the resolution and extent at which we can address them.
At the other end of the spatial-extent spectrum, experiments using physical models of fluvial landforms have allowed us to observe geomorphic evolution, and to test hypotheses.Although they are conveniently small, experiments are not limited to grain-scale or reachscale problems.A wide array of modelling approaches allows us to investigate the evolution of entire landforms with analogue modelling (Paola et al., 2009), or to monitor bedload transport or vegetation interactions using 1:1 scaled experiments (Wilcock et al., 2008;Zedel et al., 2021).Small-scale and analogue models have the advantage of allowing us to compress geological time, in order to observe the effects of processes that operate over durations much longer than human lifetimes (Paola et al., 2009).Moreover, the ability to control the experimental environment means we can test hypotheses in ways that would be impossible in the field.After many decades of experimental fluvial geomorphology, physical models remain a valuable tool to understand how rivers work, allowing us to explore problems that would be inaccessible using data from field-scale rivers in the natural world.
Laboratory landscapes require monitoring just as field-scale rivers do; the effort and costs involved in designing and conducting an experiment make it imperative to monitor as many variables as possible.Depending on the research question, one might monitor topography, flow depths and velocities, inundated areas, or channel patterns and geometry.Repeated data collection at regular intervals allows us to quantify rates of change in all of the above, thereby capturing erosion and deposition, lateral migration, bedform growth and migration, and larger-scale change such as delta progradation.We need methods and equipment to monitor all these aspects of experimental rivers either remotely or directly, just as we do for field-scale rivers.
Experimental models of rivers present some challenges when it comes to monitoring and data acquisition, as well as numerous opportunities that are not available with field-scale rivers.For instance, the small scale of physical experiments means that ground control points (GCPs) with good coverage of the landscape can be surveyed with high accuracy; the same GCPs can be used for multiple experimental surveys to reduce uncertainty in change detection (Kasprak et al., 2015).Moreover, environmental challenges are negligible; data acquisition is not dependent on flow rates, river crossings, accessibility or wireless connectivity as it might be for field data collection.Experiments also have the added advantage that flows can be stopped and the water drained from the flume to enable detailed surveys of streambed topography.Despite these benefits, the small scale of physical models means that data need to be captured with higher accuracy and precision (in absolute terms) than in field settings.
The rapid rates of change in physical models in geomorphology mean we can monitor the outcome of processes that might act too slowly to observe in human lifetimes or scientific funding cycles.However, this dynamic character of physical experiments also generates some challenges for experimentalists.Most importantly, we need to collect data more frequently; ideally, we need to monitor topography or river planform at a higher frequency than the process of interest in our study.Similarly, we need to collect data rapidly, so that the duration of data collection is instantaneous relative to the processes under investigation.These constraints present a challenge that makes the remote sensing of laboratory rivers an area for ongoing innovation.
In this paper, we review how experimental studies in fluvial geomorphology have borrowed from the remote sensing of real-world fluvial landscapes.First, we consider techniques to monitor experimental topography and change thereof, both in and out of the wetted channel.Second, we evaluate methods to monitor flow depth, and then velocity.Following this, we compare planform methods to monitor channel pattern, inundation, landform extent and change thereof.
We then discuss ways in which the remote sensing of field-scale and experimental rivers have influenced each other.Ultimately, we explore the ongoing challenges in the remote sensing of laboratory rivers and identify key areas for innovation.
Finally, we note that experimental fluvial geomorphology is a wide field and we cannot cover all of it in this review.Consequently, we limit our discussion to the remote sensing of experimental topography at the bedform to floodplain scale, and the associated flow properties and planform channel dynamics.Our review does not cover grain-scale channel roughness and flow resistance, or the remote sensing of grain size distributions and sediment transport.A multitude of techniques exist to monitor the grain-scale dynamics of streams in the field and in the lab, and would be best served by another standalone review.In addition, we do not consider direct monitoring methods (those in which the sensor must touch or be submerged in the flow), nor do we give much attention to techniques that may be 'remote' in the lab (such as ultrasonic water surface sensors mounted above an experiment) but would not typically be airborne or spaceborne for field data collection.We are primarily concerned with remote sensing methods that have transferred from field-scale to laboratory rivers or vice versa.

| TOPOGRAPHY
In the study of both field-scale and laboratory rivers, topographic data are crucial to building geomorphic understanding.Spatially continuous data on 2.5D or 3D river topography can be used to understand how hydraulic geometry changes at different flows, both in the lab and in field-scale rivers.Similarly, spatially continuous topographic data are needed as inputs for hydrodynamic and morphodynamic models (Mandlburger et al., 2009;Tonina et al., 2020); if multiple successive topographic surveys are captured, they can be used to evaluate the performance of morphodynamic models.Successive surveys also permit geomorphic change detection, revealing the spatial patterns and volumes of erosion and deposition (e.g., Heritage & Milan, 2009;Lallias-Tacon et al., 2014;Mandlburger et al., 2015;Williams et al., 2011).From these volumes, it is possible to infer sediment transport rates between topographic surveys using the 'morphological' methods pioneered by Popov, (1962aPopov, ( , 1962b) ) and Neill, (1971Neill, ( , 1987)); see Church (2006) for an in-depth review.These methods become particularly effective as the survey frequency increases (Lindsay & Ashmore, 2002).Table 1 compares the benefits and challenges of measuring topography remotely with those of other stream properties that can be remotely monitored in the lab.
Despite this high spatial resolution, there are some challenges associated with the use of laser transects to capture experimental river topography.Chief among these is that the experiment is usually stopped and the water drained before a scan of the bed can begin, to avoid geomorphic change during the scan itself, which can be relatively slow.Scan speed can vary with the desired resolution, which sets the distance between transects; Zimmermann (2009) reported that a 1Â2 mm resolution laser scan covering 0.83Â4.13m of their flume took 35 min.Some experimentalists have side-stepped this problem by scanning topography at a few cross-stream transects only, conducting frequent measurements at up to 30 s intervals (Abad & Garcia, 2009;Ganti et al., 2011;Reitz & Jerolmack, 2012;Schwindt et al., 2019;Straub et al., 2012).Nevertheless, stopping and draining the flow to scan the bed increases the time required to conduct an experiment, and can have a non-negligible effect on the experiment itself: although experimentalists take care to avoid geomorphic change while draining or restarting flow, some have reported topographic change during this process (Terwisscha van Scheltinga et al., 2020;Whiting & Dietrich, 1993).In addition, surface tension artefacts can occur if flow is restarted too rapidly, generating sediment entrainment that does not scale to the natural world.These effects can be negligible if topographic scans are far apart in time, but they become more important as scans increase in frequency, thereby placing an upper limit on scan frequency and prohibiting the continuous monitoring of bed evolution.
Another reason that flow needs to be stopped for laser scans is that certain light wavelengths are poorly reflected by water.This challenge could be surmounted by using lower-wavelength (e.g., green) T A B L E 1 The main variables that can be remotely sensed from fluvial geomorphology experiments.The major benefits and challenges of measuring each variable are compared.Monitoring river migration or avulsion lasers, which are not absorbed so strongly by water and therefore allow for through-water bathymetry (Friedl et al., 2018).This approach has been trialled by numerous field studies, using both airborne LiDAR (e.g., Hilldale & Raff, 2008;Kinzel & Legleiter, 2019;Kinzel et al., 2007Kinzel et al., , 2013Kinzel et al., , 2021;;Mandlburger et al., 2015;Pan et al., 2015;Richardson & Moskal, 2014;Tonina et al., 2019;Schwarz et al., 2019) and TLS (e.g., Miura & Asano, 2016;Panagou et al., 2020;Smith & Vericat, 2014;Smith et al., 2012).The TLS approach lends itself well to laboratory landscapes, and has been evaluated by Friedl et al. (2018) and Smith and Vericat (2014); they noted that steep angles of incidence, deep or highly turbid water and rough water surfaces limit the accuracy of laser returns or even the penetration of the laser pulse.A few experimentalists have developed cart-mounted lasers or laser sheet systems for through-water bathymetry, using visible-wavelength lasers at or near nadir (de Ruijsscher et al., 2018;González et al., 2007;Huang et al., 2010;Moteki et al., 2022;Seizilles et al., 2013;Stradiotti et al., 2020;Visconti et al., 2012).With either TLS or cart-mounted lasers, the refraction of the laser beam at the air-water interface means that underwater elevations are overestimated (i.e., depths are underestimated) and a refraction correction must be applied to submerged data points.Applying the refraction correction requires one to know the depth at which each laser beam is refracted-that is, the water surface elevation.Together, these constraints mean that rough (and difficult-to-measure) water surface conditions limit the use of this method, as do deeper or higher-turbidity flows through which the laser cannot penetrate; these problems are explored further in Section 2. For very shallow flows, a refraction correction may not be needed; Huang et al. (2010) discuss this in more detail.This has allowed some experimentalists to monitor their experiments continuously with laser scanners or sheets, without stopping flow (e.g., Baynes et al., 2020Baynes et al., , 2022;;Malverti et al., 2008).
Selecting a laser with a visible wavelength can raise the cost of a laser scanner, which is one additional challenge of this technique.For studies that stop the flow and drain the experiment before surveying bed topography, lasers with near-infrared (NIR) wavelengths can be used.At the time of writing, a TLS that emits NIR wavelengths can retail for as little as US $20,000 and as much as US $120,000.If through-water bathymetry is desired, a laser with visible light (such as those described above) must be selected.The newer model of the visible-light scanners used by Baynes et al. (2020, 2022) and Friedl et al. (2018) for through-water topographic surveying is currently advertised for US $27,000-$88,000.For many flume and streamtable experiments, cart-mounted laser scanners are custom built, and are therefore priced similarly to a high-grade TLS-of the order of US $100,000, in our experience.Much of the cost associated with these custom systems stems from their 'geographic' positioning system (the rails, stepping motor, instrument cart and the time to create the program that converts each laser transect to a topographic profile).
Where the goal is to scan a single transect with a laser sheet (or pair, as in Seizilles et al., 2013), costs are considerably lower: on the order of US $1000 per laser sheet.The set-up costs of the different data acquisition methods we discuss are compared in Table 2.
An alternative method for the remote sensing of experimental topography is the Moiré pattern deformation method developed by Limare et al. (2011), which is analogous to SAR interferometry methods used in the field.The technique involves projecting a fringe pattern onto the experiment and analysing its deformation by the topography.The approach has been applied in experimental models of structural geology (Santolaria et al., 2015;Soto et al., 2020), submarine gravity currents (Weill et al., 2014), Martian valley formation (Marra et al., 2014), braided channels (Leduc et al., 2014;Métivier et al., 2016;Reitz et al., 2014) and alluvial and debris-flow fans (de Haas et al., 2015(de Haas et al., , 2016;;Delorme et al., 2018).As with TLS methods, depths are underestimated (channel bed elevations are overestimated) due to refraction of light at the air-water interface, so a refraction correction must be applied.Flow depths have been measured from this set-up by draining the experiment to capture a 'dry' DEM, which is then compared to its 'wet' predecessor.While draining the experiment adds time to the acquisition process, the data capture itself is rapid and so the method has been applied with a periodicity of up to 5 min (Reitz et al., 2014).
Another effective technique for the remote sensing of experimental topography is close-range stereo-pair photogrammetry.Stereo photogrammetry has been in use for many decades as a method to survey river topography in the field, with data capture from river terraces (Chandler et al., 2002), gantry cranes (Carbonneau et al., 2003), blimps and kites (Marzolff & Poesen, 2009) through to aircraft (Lane, 2000;Lewin & Manton, 1975;Westaway et al., 2003) or even satellites (Hodúl et al., 2018;Wu et al., 2016).For flume and stream table experiments, close-range photogrammetry is used and typically involves one or two cameras mounted on an instrument cart that moves down the experiment, capturing images at a fixed distance downstream.The approach has been deployed to monitor experiments with braided rivers (Ashmore et al., 2011;Gardner & Ashmore, 2011;Gardner et al., 2018;Leduc et al., 2015;Lindsay & Ashmore, 2002;Stojic et al., 1998), alluvial fans and deltas (van Dijk et al., 2009;van Dijk, Kleinhans, Postma, & Kraal, 2012;Kraal et al., 2008), river meanders (Chandler et al., 2001;Lane et al., 2001), bed structures (Bertin et al., 2015;Butler et al., 2002;Groom & Friedrich, 2018), soil erosion (Heng et al.2010), and landscape evolution (Babault et al., 2005;Brasington & Smart, 2003;Hancock & Willgoose, 2001a, 2001b;Niemann & Hasbargen, 2005;Rohais et al., 2012;Turowski et al.2006).Since its first application to experimental fluvial geomorphology by Stojic et al. (1998), important developments have included its through-water application by Butler et al. (2002), who developed a refraction-correction method, and the work of Han and Endreny (2014), who applied close-range photogrammetry to water surface topography.Spatial resolution for close-range photogrammetry can reach a few millimetres (Gardner & Ashmore, 2011) and data capture is rapid; as few as eight photos have been used down the length of a stream table.Consequently, even though experiments are usually drained to collect data, the method has been applied at intervals as short as 5 min (Stojic et al., 1998).Because the method can be applied with one or two cameras on an instrument cart, costs can be as low as around US $500 (per camera).However, care must be taken to set up a network of ground control points with accurately known positions, requiring the use of an additional survey apparatus such as a total station (e.g., Stojic et al., 1998).This is true for most of the survey methods described in this paper.
T A B L E 2 Comparative costs for the various remotely sensed laboratory data collection methods summarised in this paper.Laser methods can be orders of magnitude more expensive, particularly when aiming for coverage of the whole streambed.For methods which do not utilise lasers or NIR wavelength reflectance, the main cost is for quality digital single-lens reflex cameras.
In the lab, close-range stereo photogrammetry has begun to be replaced by adaptations of SfM photogrammetry.Instead of using stereographic image pairs, SfM photogrammetry reconstruct a scene's geometry from multiple photos captured from different angles ('multiview stereo') and is therefore sometimes referred to as SfM-MVS.
Since its first deployments in the lab by Marra et al. (2014) and Kasprak et al. (2015), important developments have included its application to water surface topography by (Ferreira et al.2017), and to through-water surveying by Terwisscha van Scheltinga et al. (2020).
Helpful contributions were also made by Mali and Kuiry (2018) and Morgan et al. (2017), who compared the utility of free and proprietary software for laboratory SfM and found that proprietary options performed slightly better (in terms of point density, and of quality in some settings).In addition, important tests were conducted by Karmacharya et al. (2021), Leduc et al. (2019) and Morgan et al. (2017), who discussed the deployment of SfM in the lab and compared its accuracy to that of laser scanners and manual measurements and found similar millimetre-scale accuracy.
Adapting SfM photogrammetry to the remote sensing of laboratory topography overcomes some of the drawbacks of laser-based acquisition.For instance, Leduc et al. (2019) collected photographs for SfM from cameras on a cart that rolled down the length of the stream table, with a total acquisition time of just 15 min; a similar set-up used by Javernick et al. (2018) collected data in 12-14 min.Where an array of simultaneously triggered cameras is mounted above the experiment, data capture can even be instantaneous (as in Leenman & Eaton, 2021;Leenman et al., 2022;Vincent et al., 2022).Because SfM data collection can be so rapid with a camera array, the method allows us to monitor experimental topography without stopping flow (although a refraction correction is still often needed); the frequency of topographic data capture can be increased, allowing one to monitor topographic evolution quasi-continuously.Even with a synchronous camera array capturing topography instantaneously, costs are low relative to a laser scanner; at approximately US $500 per camera, our synchronous nine-camera array came to about US $4500.Finally, because SfM photogrammetry does not necessarily need to be conducted with fixed camera locations, a wide variety of camera positions can be used.The addition of lower-angle photos allows us to capture the fully 3D topography of an object; this improves on previous 'top-down' data collection, which can be thought of as 2.5D, as overhangs (e.g., an undercut stream bank) cannot be captured.This advantage of SfM photogrammetry has been fully utilised in applications to sediment core (Seitz et al., 2018) and log-jam (Spreitzer et al., 2020a) quantification in the lab.
Despite these advantages, there are some challenges associated with adapting SfM photogrammetry to the lab.As with field applications, a 'doming' effect can arise (Smith et al., 2016;Woodget et al., 2015).The effect can be avoided with careful lens calibration (Buechel et al., 2022;Kasprak et al., 2015;Leduc et al., 2019) or the use of distributed ground control points and convergent camera angles (Mali & Kuiry, 2018;Morgan et al., 2017;Smith & Vericat, 2015).Alternatively, the doming effect can be removed in post-processing by simply applying a transformation to the digitised topography (Adams & Zampiron, 2020), or by modelling and subtracting the error (James et al., 2020); however, a best practice might be to avoid the doming effect altogether through careful experimentation with camera and GCP placement before collecting data.In addition, as with laser scanning and close-range photogrammetry, the flow must be stopped to survey bed topography unless a correction is applied to adjust the elevations of inundated areas for light refraction at the air-water interface.This correction requires the water surface topography to be known across the channel, as an uneven water surface will mean refraction occurs at different elevations across or along the channel.
Laboratories present a controlled environment to test solutions for the refraction problem in SfM and close-range photogrammetry; multiple methods to adjust below-water elevations have been explored, using both lab and field data.Where cameras are directly down-looking (easily achievable in lab settings), a constant refraction correction factor can be applied (Butler et al., 2002;Shintani & Fonstad, 2017;Tamminga et al., 2015;Westaway et al., 2000;Woodget et al., 2015).When a wider range of camera angles is used, the refraction correction becomes more computationally intensive; methods that utilise camera and topographic point locations (Dietrich, 2017;Feurer et al., 2008;Kasvi et al., 2019;Westaway et al., 2001) and machine learning (Agrafiotis et al., 2019) have been proposed to deal with this problem.In laboratory settings, the possibility to finely control camera locations and angles means that the more simple solution of a constant refraction correction can be implemented.
Although lab studies have tested refraction correction methods, field evidence suggests a refraction correction is minimally effective for flow depths <40 cm (Westaway et al., 2000) or <20 cm (Westaway et al., 2001).Where experimental flow depths are very shallow, topographic inaccuracy due to refraction may be less than the error threshold applied to DEMs of difference, removing the possibility of spurious topographic change detection due to the presence or absence of water (e.g., Leenman & Eaton, 2021).Consequently, while refraction corrections are simpler in experiments than with field data, they may not even be necessary for very shallow experimental channels.
Despite these challenges, SfM photogrammetry is rapidly becoming a standard technique for topographic monitoring of laboratory rivers.Through-water photogrammetry is ideally suited to the laboratory 'environment', given the typically shallow flow depths and the clear flow that results from the removal of fine grains to avoid non-scaled cohesion and hydraulically smooth experimental channel beds.Moreover, the small size of many laboratory models allows for synchronous camera arrays that enable instantaneous topographic surveys without stopping the experiment (e.g., Leenman & Eaton, 2021;Leenman et al., 2022;Vincent et al., 2022).The technique therefore allows continuous topographic monitoring, which has so far been applied only to the planform (two-dimensional) evolution of experimental and fieldscale rivers through the use of time-lapse imagery.
To summarise, the early remote sensing of experimental riverbed topography was enabled through lab applications of laser sensors and close-range photogrammetry.Major developments have included the through-water deployment of both techniques.Recently, cheaper SfM photogrammetry has begun to be widely used, but laser-based measurements remain highly useful, as we will discuss in the following section on flow depth measurements.However, the ability to rig up simultaneously triggered camera arrays for SfM photogrammetry allows instantaneous data capture without stopping flow.Even a large camera array (e.g., 20 cameras installed above a longer stream table ) would cost an order of magnitude less than a custom-built laser line scanner.The main trade-off in switching to SfM photogrammetry is in computational time; each set of photographs requires post-processing to generate a digital elevation model.Fortunately, the process can be automated, particularly for lab data where each set of photos uses the same camera positions and ground control network.We therefore view SfM photogrammetry as an important tool for elevation measurements in geomorphic experiments, provided an automated dataprocessing workflow is carefully set up.The main outstanding challenge for both through-water photogrammetry and through-water lasers is the resolving of the water surface at time and space scales that match the resolution of the bed topography.Important progress has been made on this problem in recent decades, and is discussed further in the next section.

| FLOW DEPTH
Remote sensing can also be used to monitor flow depths at high reso- One approach to spatially distributed depth monitoring is to model flow depth based on the spectral properties of the water.
Numerous authors have estimated flow depth in field-scale rivers using different band ratios or algorithms to predict depth from the available surface reflectance data.Source imagery has been captured from unpiloted aerial vehicles (UAVs) (Flener et al., 2013;Tamminga et al., 2015), piloted aircraft (Legleiter & Fosness, 2019, Legleiter et al., 2004, 2011, 2018;Williams et al., 2011Williams et al., , 2014;;Winterbottom & Gilvear, 1997), or satellites (Legleiter & Overstreet, 2012;Niroumand-Jadidi et al., 2018, 2020); Legleiter and Harrison (2019) compared different acquisition approaches for one site.In the lab, optical depth methods have been applied to experimental alluvial fans (Leenman & Eaton, 2021), deltas (Kim & Jerolmack, 2008), vegetated streams (Gran & Paola, 2001;Tal & Paola, 2007, 2010; van Dijk, Teske, Van de Lageweg, & Kleinhans, 2013), meandering rivers ( van Dijk, van de Lageweg, & Kleinhans, 2013) and estuaries (Leuven & Kleinhans, 2019;Leuven et al., 2018).After its initial deployment in the lab by Gran and Paola (2001), an important further development was its combination with through-water laser scanning by Huang et al. (2010), allowing capture of the co-evolution of flow depth and channel-bed topography as their experiment evolved.Optical depth measurement is well suited to deployment in the lab: dye can be added to the experimental fluid so that colour intensity varies appreciably over a narrow range of flow depths, and lighting conditions can be held reasonably constant.Moreover, depth-colour calibration pans or prisms can be placed in the experiment (e.g., Gran & Paola, 2001;Huang et al., 2010;Tal & Paola, 2007), and data can be captured by a time-lapse camera so that depth can be monitored at quasicontinuously throughout an experiment.
Despite the advantages of this near-continuous depth monitoring, there are some challenges associated with applying optical depth measurement methods in the lab.As with field studies using this method, a calibration dataset is required in order to link measured flow depths to surface reflectance.Acquiring a spatially distributed depth calibration dataset can be challenging: while it is possible to drain the flow and measure the experimental bathymetry with submillimetre accuracy, the water surface topography needs to be known in order to calculate depths, which remains a challenge.Alternative methods to calibrate optical depth measurements have been piloted by Gran and Paola (2001) and Huang et al. (2010), who used pans or prisms of known water depth, respectively; these methods hold promise for widespread lab application.In addition to the problem of depth calibration data, the spectral properties of the water at a given flow depth need to remain constant over the duration of the experiment in order for the depth calibration to be applied.When dye has been added to the flow, care needs to be taken to ensure that the dye concentration remains constant over time so that the relationship between colour intensity and depth is constant throughout the experiment.Evaporation or water addition to compensate for leakages may mean the depth-colour relationship needs to be recalibrated during or between experiments.A final and related challenge with this method is to optimise the dye concentration; too little dye means that water colour does not vary appreciably over the small range of experimental depths, while too much dye reduces the transmission of light through the water column, limiting the visibility of the bed for concurrent bed topography measurements (Smith & Vericat, 2014).Huang et al.
(2010) also found that there was a maximum depth above which water colour intensity ceases to increase with depth, and this depth decreases as dye concentration increases, again meaning that too much dye makes the method less effective.
If dye concentration cannot be held constant, an alternative method used in both field and laboratory remote sensing is to subtract the bathymetry from the water surface elevation in order to estimate flow depths.In field-scale rivers and coastal waters, depths have been estimated from through-water bathymetry captured by TLS (Panagou et al., 2020) or by SfM photogrammetry (Dietrich, 2017;Woodget et al., 2015); Kasvi et al. (2019) conducted an in-depth comparison of optical and SfM methods to measure flow depths and bed topography.This surface-to-bathymetry topographic differencing method has also been applied in the lab; flow depths have been estimated by comparing the water surface to through-water TLS data (Friedl et al., 2018), to through-water bathymetry from cart-mounted point lasers (Visconti et al., 2012) or laser sheets (Seizilles et al., 2013) or to SfMderived channel topography (Javernick et al., 2018;Leduc et al., 2019;Nelson & Morgan, 2018;Terwisscha van Scheltinga et al., 2020;Wang et al., 2021).The surface-to-bed differencing method can provide detailed spatially distributed depth data.However, the method requires the water surface elevation (and cross-stream variation thereof) to be well constrained.Moreover, the water surface topography is also necessary for through-water applications of laser or photogrammetry topographic surveys, as we need to know the height of the air-water interface at which light is refracted.
Despite the importance of water surface topography to both flow (2012) (in 3D) successfully used lasers to scan the water surface, they had to add fluorescent dye to the water, so the experiment had to be drained before bed topography could be scanned.These cases highlight the difficulty of obtaining simultaneous spatially distributed data on both the water surface and bed elevations.
Simultaneously imaging the topography of both the water surface and the bed has been a major challenge for the remote sensing of experimental rivers.While important developments were made by González et al. (2007), their through-water laser scanning method depended on a 'regular, stable water surface'.This condition made their technique suitable for many experiments but not for those with rapidly evolving or laterally variable water surface topography, as they collected water surface elevations down a single central profile (using an ultrasonic sensor).For experiments with a laterally variable water surface, we need spatially distributed water surface data rather than a single profile.
Considerable progress on this problem was made by Huang et al.As with depth data, high-resolution flow velocity data can also aid in calibrating hydraulic models.
Remote sensing methods to monitor surface flow velocity in the field have capitalised on the pervasiveness of floating particles and features on field-scale rivers, which can be tracked using largescale particle image velocimetry (LSPIV) techniques.For instance, naturally introduced particles such as leaf litter, foam, ice floes and even features such as boil vortices can be tracked using LSPIV; sometimes particles are artificially introduced, although care must be taken to use tracers that are not ecologically harmful in their depositional environment and can decompose naturally.The technique has been applied using repeat imagery or video captured from helicopters (Fujita & Hino, 2003;Legleiter & Kinzel, 2020), drones (Bandini et al., 2021;Kinzel & Legleiter, 2019;Koutalakis et al., 2019;Strelnikova et al., 2020) and stationary cameras (Bradley et al., 2002;Creutin et al., 2003;Fujita et al., 1998;Hauet et al., 2008); the method has been trialled more recently with highresolution satellite data (Kääb et al., 2019;Legleiter & Kinzel, 2021), which might be particularly useful for monitoring surface velocities across large areas, large samples of rivers, or in sites that are difficult to access.
LSPIV was an adaptation of smaller-scale 'PIV', developed in the lab; experimental rivers are well suited to velocity measurement using particle tracking.The water can easily be seeded with particles, without the ecological concerns that come with releasing tracers in natural rivers.Moreover, the time-lapse camera(s) needed to monitor particle movement are often in use for other applications in the lab, for instance to monitor changes in channel position or dye colour intensity (and flow depth).Problems with camera stability and inconsistent lighting conditions due to weather and day length are also easily surmounted in the lab.PIV and LSPIV are therefore widespread in geomorphic experiments.Various iterations of the technique have been deployed in models of estuary dynamics (Braat et al., 2019;Kleinhans et al., 2017;Leuven & Kleinhans, 2019;Leuven et al., 2018), side-channel flows (Nezu & Onitsuka, 2002), flow separation at bends (Blanckaert et al., 2013), bifurcation dynamics (Marra et al., 2014), steep unconfined flows (Piton et al., 2018), scour hole turbulence (Ferreira, 2011), bedload transport (Lajeunesse et al., 2010), surface roughness effects (Sambrook Smith & Nicholas, 2005), laminar flows (Malverti et al., 2008), flow-vegetation interactions (Bennett et al., 2002;Bouma et al., 2013;Gran & Paola, 2001;Nezu & Onitsuka, 2001), flow-animal interactions (Stamhuis et al.2002), to estimate volumetric discharge remotely (Johnson & Cowen, 2016), and to test numerical models (González et al., 2008).Although PIV has a long history of use in fluid mechanics research (e.g., Westerweel et al., 2013), early applications in geomorphology include Gran and Paola (2001), who traced the movement of soap bubbles to measure surface velocity.PIV methods have even been adapted to characterise the transport of bedload grains (rather than tracers seeded in the flow) (Best, 1988;Hergault et al., 2010).These diverse examples highlight the utility of PIV for monitoring both 2D (water surface) and 3D flow structures and velocities.
Despite their utility in a range of different environments and experimental designs, there are some challenges associated with using LSPIV and PIV in geomorphic experiments.One of these is the difficulty in choosing a suitable tracer particle; while ecological concerns are minimised in the lab, tracers must still be non-toxic to humans, and the experiment should be designed so that non-decomposing tracers (such as plastic spheres) do not enter wastewater leaving the lab.In addition, tracers must be large enough that their motion is detectable by overhead cameras, but small enough to follow the instantaneous motions of the fluid being monitored (Hadad & Gurka, 2013).For experiments with shallow, braided flow, this latter requirement becomes a challenge as tracers may interact with the bed, or with bedload transport, in shallower parts of the channel.
Finally, traditional PIV requires that the flow be illuminated with a laser sheet at the plane of interest; combined with data collection hardware and software, this raises the initial cost of the method to approximately US $15,000 at a minimum.Although relatively expensive, the laser sheet can be projected through any planar cross-section of the flow, which presents a significant advantage over LSPIV.While LSPIV can be set up cheaply with an overhead camera, the technique is best suited to water surface velocity.
A related method for velocity tracking, which circumvents some of these issues, is that of injecting the flow with pulses of dye (or, historically, paint).Movement of the dye front can be recorded with overhead video or time-lapse photography to measure velocity, although this method has also been used without measurements to visualise flow structures.For instance, Mosley (1976) injected dye into his confluence experiments to illustrate the presence of helical flow cells.This visualisation method has been successfully applied in other experiments on confluences (Ashmore, 1982;Ashmore & Parker, 1983;Biron et al., 1996;Mosley, 1976) and meander bends (Whiting & Dietrich, 1993).Later experiments combined dye pulses with photography to measure (rather than view) flow velocity.After early application to braided channels by Metivier and Meunier (2003), the method was applied in experiments on sediment transport (Prancevic et al., 2014;Recking et al., 2008), meandering channels (Braudrick et al., 2009), deltas (Ganti et al., 2016) and offshore plumes (Chatanantavet & Lamb, 2014).Dye tracers are useful in that they are often non-toxic (e.g., swimming pool dyes or food colouring), and they are suitable for shallow flows due to their soluble nature (whereas tracer particles may impinge on the bed).In the field, dye tracers have a long history in studies of river flow (e.g., Smart & Laidlaw, 1977).
However, it is only recently that dye releases have been combined with remotely sensed imagery to capture velocity fields (Legleiter et al., 2019).
One further method often used to monitor flow velocity in the lab is laser Doppler anemometry (LDA), which has been in use for several decades (e.g., by Statzner & Holm, 1982).A key distinction from the dye or particle methods mentioned above is that, while measurements are highly accurate and temporally continuous, they are typically taken at a single point or cross-section, whereas dye-based or PIV methods are spatially distributed (R. M. L. Ferreira, 2011;Nezu, 2005).Finally, the method is expensive to deploy in the field (R. M. L. Ferreira, 2011) and has not gained widespread traction in the remote sensing of field-scale rivers.We therefore do not explore this technique in more detail in this review.

| PLANFORM RIVER GEOMETRY
The planform geometry of rivers (i.e., the planview extent of water) is possibly the easiest and yet most versatile type of data to collect.For instance, the remote sensing of river planform can aid in quantifying discharge (Gleason & Smith, 2014;Smith & Pavelsky, 2008) and in identifying overbank flooding (Akiva et al., 2021;Mateo-Garcia et al., 2021;Sunkara et al., 2020).Time series of river planforms can also be used to monitor seasonal changes in rivers and lakes (Pekel et al., 2016;Vulis et al., 2020).From planform data, we can learn about spatial and temporal variation in river widths (Allen & Pavelsky, 2018;Feng et al., 2022;Schwenk et al., 2017;Wang et al., 2020;Yang et al., 2020), as well as network structure (Isikdogan et al., 2017a(Isikdogan et al., , 2018;;Tejedor et al., 2017).Time series of river planform can be used to track channel migration rates (Langhorst & Pavelsky, 2023;Schwenk et al., 2017), or to identify sudden changes in channel pattern or location, such as avulsions (Brooke et al., 2022).These diverse applications highlight the large amount of information that can be gleaned from monitoring river planform over time.
The remote sensing of field-scale river planform is typically carried out using satellite data.Scientists capitalise on the poor reflectance of near-infrared radiation (NIR) by water, which allows water to be classified in satellite data.In satellite sensors that capture shortwave infrared (SWIR) reflectance in addition to NIR, the normalised difference water index (NDWI) can be used to identify water (McFeeters, 1996).More sophisticated supervised classifications are also used, and have been applied at a global scale to the Landsat data archive (Pekel et al., 2016;Pickens et al., 2020).A suite of machine learning tools is being developed to 'segment' or extract water from different satellite datasets, some of which take on the challenges posed by cloud cover and shadows (Garcia et al., 2020;Isikdogan et al., 2017bIsikdogan et al., , 2020;;Tambe et al., 2021).These various surface water datasets and tools have been applied to track the pace of lateral channel migration (Langhorst & Pavelsky, 2023;Schwenk et al., 2017) and to monitor river deltas (Jarriel et al., 2020(Jarriel et al., , 2021)), to name a few examples.
Methods that capitalise on the poor NIR reflectance of water do not translate well to the lab, given that NIR sensors are expensive (prices start at around US $2500 for a single camera).Instead, consumer-grade digital cameras can be used, especially if dye (or, occasionally, paint, e.g., Warburton & Davies, 1994;Zarn & Davies, 1994) is added to the experimental fluid so that its spectral properties differ from the surrounding sediment.Dye addition allows the channel pattern to be tracked more easily using time-lapse photography; the channel can be digitised automatically by thresholding normalised band ratios or colour values to produce binary maps of 'wet' and 'dry' areas.Care must be taken to ensure that dye concentration and lighting conditions remain constant during an experiment so that a single threshold can be used throughout.Although channel mapping was done manually for some earlier studies (e.g., Ashworth et al., 2007;Egozi & Ashmore, 2008;Whipple et al., 1998;Zarn & Davies, 1994), automated channel identification algorithms have been applied widely, to quantify lateral mobility and avulsion (Bufe et al., 2016(Bufe et al., , 2019;;Cazanacli et al., 2002;Hoffimann et al., 2019;Sapozhnikov & Foufoula-Georgiou, 1997;Tal & Paola, 2007, 2010;Wickert et al.2013), to monitor channel and shoreline dynamics on deltas and fans (Barefoot et al., 2021;Carlson et al., 2018;Chadwick et al., 2022;Esposito et al., 2018;Jarriel et al., 2019;Leenman & Eaton, 2021;Leenman et al., 2022;Lentsch et al., 2018;Miller et al., 2019;Martin et al., 2009;Nicholas et al., 2009;Piliouras et al., 2017;Piliouras & Kim, 2019a, 2019b;Reitz & Jerolmack, 2012;Reitz et al., 2010), to monitor bifurcation dynamics (Daniller-Varghese et al., 2020), to track simulated Martian valley evolution (Marra et al., 2014) and to highlight the construction of stratigraphy (Sheets et al., 2002;Terwisscha van Scheltinga et al., 2020).Since its initial application to experimental channels by Sapozhnikov and Foufoula-Georgiou (1997), the method has changed little in essence, although recent variations include the use of principal component analysis (rather than raw RGB imagery) to map the channel (Chadwick et al., 2022) or image conversion to other colour spaces such as YCbCr (Jarriel et al., 2019).The relative simplicity of automatic channel extraction methods allows large volumes of input data to be processed, so that channel patterns and lateral migration rates can be tracked almost continuously from timelapse photography.Methods for measuring channel planform and change thereof, developed and tested using lab data, have been redeployed to monitor rivers and deltas in numerical models (Liang et al., 2016;Lauzon & Murray, 2018;Piliouras et al., 2021) and in nature (Jarriel et al., 2020).
To summarise, river planform geometry is perhaps the most versatile property we can observe through remote sensing of our experiments.Time series of channel extents give information about river character, migration rates and geometry changes.Moreover, the ability to add dye to experiments means water extents can be sensed using colour imagery.The simplicity of this method permits the automatic processing of large data volumes, allowing us to quantify river planform change at high frequency.Experiments have also suggested that planimetric and volumetric change can scale (Middleton et al., 2019), so the method could be used to monitor erosion and deposition if no other data are available.Given the large volumes of data we can collect using this method, we can now move beyond simple raster math and apply methods such as PCA to identify common channel patterns, or PIV to track channel bankline movements.

| TRANSFERS BETWEEN REMOTE SENSING OF FIELD-SCALE AND EXPERIMENTAL LANDSCAPES
Experimental geomorphologists have a long history of borrowing data acquisition methods from the remote sensing of field-scale rivers.The examples cited above show the diverse ways in which the community has borrowed from field-scale remote sensing.This borrowing has been most prevalent in topographic remote sensing.For instance, airborne lasers (Huising & Gomes Pereira, 1998;Krabill et al., 1983;Lohr, 1998;Gomes Pereira & Wicherson, 1999;Petzold et al., 1999) and TLS (Bauer et al., 2003;Lichti et al., 2000) were well developed as field surveying methods before TLS was deployed in the lab by Marra et al. (2014) to monitor geomorphic experiments.Similarly, major developments in stereo photogrammetry for terrain reconstruction had been made before 1925 (Hallert, 1960); close-range photogrammetry with non-metric cameras was developed in the 1970s (Brandow et al., 1976;Welch & Dikkers, 1978) and finally deployed in the lab to monitor experimental geomorphology by Stojic et al. (1998).
SfM photogrammetry was similarly first developed for use on fieldscale topography (Fonstad et al., 2013;Westoby et al., 2012)  The borrowing of fluvial remote sensing methods has been bidirectional: while field methods have been adapted for experiments, laboratory methods have also been adapted for natural landscapes.
The small scale of laboratory rivers, ease of access and controlled environment allow for development, testing and refinement of remote sensing methods, which can then be transferred for deployment 'in the wild'.For instance, work on the through-water applications of TLS and close-range photogrammetry conducted lab tests before deploying the methods in field-scale river flows (Butler et al., 2002;Smith et al., 2012).PIV was developed to visualise and quantify flow dynamics in flumes as early as 1893 (Marey, 1983), and was adapted to field-scale rivers with the advent of LSPIV around 100 years later (Fujita et al., 1998).Similarly, photographing the passage of dye pulses is a technique developed and used widely in the lab, and has only recently been applied to the remote sensing of field-scale rivers through the airborne photography tests of Legleiter et al. (2019).
While laboratory experiments provide useful testing grounds for remote sensing apparatus, the ease of instrumenting and monitoring experiments means they also generate plentiful datasets with minimal noise in comparison to real-world data.For this reason, experimental datasets can be useful to test and develop new data processing and analysis techniques that can later be applied to field or synthetic data at different scales.For instance, methods developed by Wickert et al. (2013) to quantify experimental channel mobility from binary flow maps have been adapted to numerical model outputs (Lauzon & Murray, 2018).New methods to characterise complex multi-threaded flows were trialled on experimental and synthetic data by Jarriel et al.
(2019) and later applied to Landsat imagery (Jarriel et al., 2020).The plentiful, cloud-, snow-and dust-free data we can gather in experiments allow these methods to be tested and improved before we apply them to noisier field data.
The fact that some remote sensing techniques have developed in the laboratory and later been transferred to field-scale rivers, while others have been developed for airborne deployment and later downscaled for the lab, may reflect both the physical differences between these environments and the different perspectives we gain from landscape-or channel-scale investigations.The controllable environment of the lab is simply more suitable than the field, for the initial calibration and testing required for some methods such as throughwater photogrammetry or laser scanning.The ability to control water depth, turbidity, channel geometry, substrate type and lighting in the lab makes this environment ideal for the initial tests of new remote sensing technology (e.g., those conducted by Smith et al., 2012, andButler et al., 2002).In addition, we posit that because the scale of problems investigated in the field and lab often differ, we geomorphologists tend to have different perspectives of the field and laboratory landscape, which may have shaped the type of data we sought in each environment.A key advantage of field-scale studies over many laboratory investigations is that one can gain a wider perspective of a river, examining a catchment as a whole, or the interactions between different sub-basins or between hillslopes, rivers and floodplains.For studies at this broader scale, it can be important to gain an 'overview' of river dynamics, generating a need for spatially continuous data on When close-range photogrammetry was first deployed to monitor fluvial landscapes in the lab by Stojic et al. (1998), the improvements to efficiency and spatial resolution allowed Lindsay and Ashmore (2002) to conduct a controlled exploration of the morphological method and the influence of the survey time window, which has been integral for those seeking to apply this method in the field (e.g., Williams et al., 2011).Similarly, when Gran and Paola (2001) adapted the remote sensing methods of Winterbottom and Gilvear (1997) for optical depth measurement in the lab, they laid the groundwork for the simultaneous and rapid monitoring of bathymetry and water depth by Huang et al. (2010), albeit at a small scale.And finally, it is not just topographic surveying that has transferred well between the lab and field; the adaptation of lab-scale PIV into LSPIV in field-scale rivers by The possibility of monitoring through-water topography at high resolution and frequency revolutionises the way we can capture data during experiments.Rather than needing to stop the flow and drain the experiment to conduct topographic scans, we can allow experimental topography to evolve continuously.This approach is faster, and overcomes issues with sediment transport and surface deformation while ramping flow down or up.Moreover, if we are able to capture data on the water surface topography, or implement methods to model depth from colour intensity, we can gain simultaneous, quasicontinuous datasets on the co-evolution of channel topography and flow depth.For instance, we were able to collect just such a dataset at 1 min intervals and 1 mm resolution over a 2.4mÂ2.4mrapidly evolving alluvial fan experiment, by using an array of nine synchronous cameras for SfM and optical depth measurement methods (Leenman & Eaton, 2021).While our methods can be further improved upon, the camera array shows promise for continuously monitoring evolving topography, as do other methods for rapid and high-frequency topographic measurements (such as the 1 min periodicity laser scans of Moteki et al., 2022).
Spatial data from experiments in fluvial geomorphology can now be collected at resolutions equalling the grain scale and frequencies that capture the minimum detectable particle motion (which is set by the pixel size).This high volume of high-resolution data allows us to trial and develop new analysis methods.For instance, instead of generating binary flow maps from colour imagery, which can be ambiguous in areas of sheetflow, Leuven and Kleinhans (2019) produced binary maps separating channels and bars by thresholding their high-resolution DEMs at the median bed level.Another alternative to thresholding single bands or band ratios to map flow is instead to conduct a principal component analysis (PCA) on the imagery bands, and then map channel patterns from one of the principal components (Chadwick et al., 2022).Existing analysis methods can also be repurposed to ingest high-frequency experimental data.For instance, LSPIV techniques can be applied to track the movement of landforms, as opposed to tracer particles, to characterise geomorphic change.
This approach has been applied to track delta channel migration in experiments (Chadwick et al., 2022) and on field-scale deltas (Jarriel et al., 2021).The high volume of data also legitimises the application of machine learning methods, for instance to classify features or recognise patterns (e.g., Carbonneau et al., 2020;Donadio et al., 2021;Du et al., 2019;Guillon et al., 2020;Isikdogan et al., 2020;Li et al., 2020;and numerous others).
In addition to new analysis methods, the high volume and resolu- One workaround to this problem may be the application of radar to the lab, which has been used to monitor sediment-laden flows (e.g., debris flows) in the field (Delannay et al., 2017).Experiments where suspended sediment transport is geomorphically importantfor instance, in estuarine systems-can also be challenging.Moreover, the challenges of 'seeing' through a rough water surface limits the range of flow conditions experimentalists can use.Monitoring experimental streambed topography with a 'diagonal' perspective through perspex walls could present an option to surmount this last problem, but the difficulties of using optical or laser methods to monitor bathymetric change beneath turbid flows is an outstanding challenge in experimental remote sensing.
Another outstanding challenge is that of resolving experimental river topography in the presence of vegetation.While quantifying vegetation size and density itself is important, geomorphologists tend to be more interested in how the experimental riverbed beneath the vegetation is affected by its presence or growth.Field studies using SfM and close-range photogrammetry found that surface heights were erroneously increased or less accurately measured in the presence of vegetation (Groom & Friedrich, 2018;Woodget et al., 2015).
Given the influence of vegetation on the accuracy of SfM and closerange photogrammetry, laser scanning methods may seem more suited to vegetated experiments.However, numerous authors have noted the poor performance of laser scanners in densely vegetated areas of their experiments (Gran & Paola, 2001;Li et al., 2018;Piliouras et al., 2017;Tal & Paola, 2010).These reports suggest that we need to think carefully about monitoring topography in vegetated experiments.In situations where experimental vegetation is sparse or does not completely obscure the bed, laser scanning may still be feasible; scanners which emit a laser beam at a range of oblique angles (instead of one scan line that is perpendicular to the bed) could surmount this problem (e.g., Kyuka et al., 2021;Vargas-Luna et al., 2019), as could photogrammetry with a wider range of camera angles.Nevertheless, unless simple vegetation proxies such as rods are used (e.g., Kim et al., 2015Kim et al., , 2019)), vegetation will still need to be filtered out from topographic data to generate 'bare earth' models.In fieldscale remote sensing, scientists have built robust algorithms to filter vegetation from LiDAR (e.g., Meng et al., 2010) or radar (e.g., Cloude & Williams, 2005;Li et al., 2015) data and reveal the underlying topography.Some experimentalists have adapted these algorithms for deployment in the lab; a recent example is the work of of data-such as manually mapping channels (Ashworth et al., 2007;Egozi & Ashmore, 2008;Zarn & Davies, 1994;Whipple et al., 1998) or analysing individual video frames (Venditti et al., 2005a(Venditti et al., , 2005b)-are no longer feasible.
This last problem presents one of the most exciting challenges of experimental remote sensing: developing new ways to analyse and extract meaning from large volumes of high-resolution data.First and foremost, these methods must be automated as much as possible, to enable ingestion of large volumes of data.However, we also need methods that transfer easily between experimental scenarios and setups; for instance, we need to avoid hard-coding thresholds for certain dye colours and concentrations, or channel dimensions.As the frequency of data collection loses its upper limit, we also need new methods to reduce and make sense of the large volumes of data we collect.For instance, in previous experiments or field studies, one high-resolution DEM could be analysed in detail, with much inferred about the erosional and depositional processes at play.However, when hundreds or even thousands of DEMs are collected in sequence in a single experiment, it no longer makes sense to look at every time step in detail.Instead, we need to move forward with new methods of reducing data to a small number of variables for each time step, or finding new ways to summarise a DEM; we need to evaluate which summary variables are most useful for characterising the experimental state at a single point in time, and then track those variables over the course of the experiment.These summary variables could be traditional metrics; for instance, Leuven and Kleinhans (2019) summarised a sequence of DEMs in an estuary experiment by recording estuary width and cross-sectional area, bar dimensions and braiding index.
New metrics are continually being developed to characterise geomorphic systems, and these could be applied to experimental data.Tejedor et al. (2015aTejedor et al. ( , 2015b) ) adapted methods from graph theory to characterise topology and connectivity in delta channel networks; these methods could be applied to experimental deltas to examine how delta structure and connectivity evolve under different scenarios.
Other tools to extract and characterise channel patterns from overhead imagery (Isikdogan et al., 2015(Isikdogan et al., , 2017a) ) and to skeletonise and characterise network topology (Schwenk & Hariharan, 2021;Schwenk et al., 2020) offer considerable promise for analysing high-frequency time series of experimental photos.
The combination of high-resolution, high-frequency observations of topography, flow depths and velocities, and channel planform with new methods for data analysis and reduction allows us to test old hypotheses or reinvestigate old questions that were previously addressed with low-resolution data.For instance, through coupling high-resolution, high-frequency measurements of flow around bedforms with simultaneous bathymetry data, we could develop a deeper understanding of the influence of bed structures, roughness and flow resistance, revisiting the ideas of Aberle and Smart (2003) and Aberle and Nikora (2006); their data on the standard deviations of bed elevations were cutting-edge at the time and their findings could now be further extended by gathering high-frequency, highresolution paired data on flow dynamics and bed topography.
Another example is the influence of suspended sediment concentration on flow dynamics, which could be monitored at high resolution with a combination of PIV and/or dye injections, perhaps alongside optical depth methods calibrated for different sediment concentrations; this set-up could allow extension of the profile-based methods historically used (e.g., Coleman, 1986).Yet another is the question of topographic survey frequency explored by Lindsay and Ashmore

F
I G U R E 1 The basic set-up, typical data products and key examples for the techniques we detail.[Color figure can be viewed at wileyonlinelibrary.com] lution and frequency.Spatially distributed observations of flow depth give insight into hydraulics and flow resistance, and when coupled with topographic data can give insight into the spatial pattern of shear stress and of potential sediment transport.In addition, these data are useful for calibrating and testing hydraulic and morphodynamic models.When subtracted from the water surface elevation, spatially distributed depth data can be used to model river bathymetry.In the field, we need remote methods to monitor flow depth because direct measurements may be dangerous or expensive.Moreover, direct measurements tend to be spatially concentrated, whereas remotely sensed methods give a spatially distributed understanding of how flow depth varies within the channel.In the lab, challenges associated with safety and cost are minimised, but with shallow and slow flows it becomes necessary to have non-intrusive methods that do not alter the flow field.Moreover, in the lab we still need spatially distributed data-capture methods that show how flow depth varies down-and across-stream, in order to model the distribution of shear stress within the channel.
Photogrammetry may present one solution to capturing instantaneous, spatially distributed water surface topography.For instance,Han and Endreny (2014) andFerreira et al. (2017) applied stereo-pair and SfM photogrammetry, respectively, to estimate 3D water surface topography in the lab.However, both studies had to seed the water surface with floating particles (wax powder and cork particles, respectively), thereby limiting the potential for through-water photogrammetry.Similarly, althoughInoue et al. (2020)  (in 2D) andLegout et al.

(
2010), who used through-water laser scanning to collect bathymetry, coupled with optical depth methods.By estimating the water depth from its luminosity, they were able to apply a spatially varying refraction correction to their laser scanner data.Further developments have been made byMoteki et al. (2022), who projected a laser sheet vertically into the water column and used computer vision techniques to extract both the water surface and bed topography; scanning a series of cross-stream profiles down their entire flume took 27 s.Despite these advances, an outstanding issue is that of turbid or roughsurfaced flow, which impede through-water scanning or photogrammetry as well as optical depth methods.For experiments requiring these flow conditions, lower-spatial-resolution methods such as ultrasonic sensors may be more suitable.To summarise, experimental geomorphologists have found creative ways to measure flow depth remotely in their experiments.Through adapting optical methods, scientists capitalised on the fact that we can add dye to our experiments, making it straightforward to relate flow depth to water colour intensity.An alternative technique of measuring the difference between the water surface and bed elevations holds promise, but resolving the water surface elevation in 3D remains a challenge.This presents a conundrum, as we need to know the water surface topography to provide a refraction correction for through-water laser scanning or photogrammetry.Important progress in the problem of simultaneously resolving flow depth, water surface and bed topography has been made byHuang et al. (2010) andMoteki et al. (2022).Approaches that combine optical depth methods with photogrammetry (e.g., that trialled byLeenman & Eaton, 2021) present another possible solution.Surmounting this challenge will be especially important for the shallow, diverging flows in braided streams and distributary networks, where channel depths are highly variable across the experiment and may be shallow relative to the accuracy of currently available depth measurement methods.Highaccuracy, high-resolution data on flow depth will be most powerful when coupled with high-resolution topographic data and time-lapse imagery documenting channel migration.Collectively, these data will provide the means to test hydraulic and morphodynamic models.Moreover, these data will help us to understand how the spatial distribution of shear stress is linked to the erosion and deposition revealed through repeat surveys of experimental topography.4| FLOW VELOCITYWhile flow depth data are important, we also require spatially distributed data on surface flow velocity to help us diagnose the hydraulics in our experiments.Data on flow velocity is useful to understand flow resistance in our experimental channels, particularly where different scales of flow resistance may self-organise during the experiment in response to the sediment calibre, supply and flow, or changes thereof.
To summarise, methods to remotely sense flow velocity in geomorphic experiments typically involve tracking the progress of tracer particles or a dye front.From initial tests for visualising flow, the method evolved as timed photographs allowed the use of this simple technology for measurement.While widely used in fluid dynamics, PIV is becoming more common in geomorphology, particularly for imaging the surface velocity field.Developments with different tracer types may make the method more useful for shallow and laterally diverging flows where tracers risk interacting with the bed.Given that geomorphic experiments are often monitored by overhead cameras for time-lapse photography, the addition of tracers for velocity measurement is a simple and cheap way to measure an additional flow property.
before being adapted to monitor laboratory experiments by Marra et al in 2014 and by Kasprak et al. in 2015.
river and floodplain topography and spurring the development of stereo photogrammetry, LiDAR and eventually SfM photogrammetry.Conversely, while it is possible to explore the reach-to bedform-scale physics of flow and sediment transport in the field, these problems are well suited to experiments; the manipulability of experimental controls enables straightforward testing of hypotheses from theory or numerical modelling.This tendency towards using experimental models to understand the physics of flow and sediment transport may explain why laboratory experiments have been a fertile ground to develop methods to visualise and record flow dynamics, most notably PIV and dye-tracing.While the different scales of the problems studied in the lab and field may have shaped the initial development of these branches of remote sensing, many transfers between the two scales have yielded fruitful advances in fluvial geomorphology.For instance, transfers of topographic surveying technology from the field to the lab have increased the resolution and efficiency of topographic data capture.

Fujita
et al. (1998) was a major advance in the remote sensing of river flow dynamics, due to the larger spatial extent that can be monitored efficiently compared to direct methods.Advances with UAVs and space-based data capture continue to improve the spatial extent and range of conditions under which this method can be applied, such as the use of Planet Labs SkySat data by Legleiter and Kinzel (2021) to monitor a velocity field at 1 m resolution by tracking the motion of boil vortices.7 | DEVELOPMENTS, CHALLENGES AND OPPORTUNITIES New developments in the remote sensing of experimental rivers continue to extend the range of questions we can address.For instance, high-frame-rate cameras allow detailed reconstruction of flow fields with (LS)PIV in geomorphic experiments.Investigations with throughwater laser scanning or photogrammetry offer new ways to monitor experimental bathymetry at improved accuracy and frequency.Software and algorithms continue to improve as well, keeping pace with developments in data acquisition.For instance, Terwisscha van Scheltinga et al. (2020) benefited from commercial software updates during the course of their through-water photogrammetry experiments, which reduced the likelihood of model 'flipping' in the software.Similarly, new algorithms increase the accuracy of through-water photogrammetry, such as the machine learning methods ofAgrafiotis et al. (2019) or iterative algorithms ofDietrich (2017), which improve on previous methods to correct for refraction at the airwater interface.These data acquisition and software improvements (both commercial and scientific) broaden the range of problems we can explore with remote sensing in the lab and field, as well as the accuracy and reliability of the data we can collect.
tion of experimental data have spurred the creation of new frameworks in geomorphology.For example,Jarriel et al. (2019) used experimental data to develop a new framework to characterise temporal trends in the channelisation of multi-threaded flows.Another framework to compare surface and subsurface delta structures was developed byMiltenberger et al. (2021) from numerical and experimental data.These new frameworks allow us to constructively glean information and insight from the high volumes of experimental data we can now collect.Nevertheless, new methods to remotely monitor experimental topography and flow present some new challenges, and some challenges remain to be overcome.One ongoing challenge is that topographic data need to be rigorously georeferenced and validated, particularly as cell sizes decrease, giving the illusion of accuracy.The increasing level of detail available is only useful if the topographic data are accurately georeferenced to some local coordinate system; change detection is only meaningful if we know there is no topographic doming effect or other biases to the surface.Independent validation of topographic data (e.g., validating experimental SfM with a TLS or with a set of total-station checkpoints) is an ideal way to both quantify accuracy and identify any doming effect.For repeat topographic surveys, it is also important to assess replicability, by examining how precisely an inactive surface is represented in a sequence of topographic surveys.Measures of water surface elevation and flow depth also require careful calibration and validation, ideally with an independent measurement.Quantifying accuracy is particularly important if there are multiple sources of uncertainty in a dataset-for instance, where flow depth is calculated as the difference between the water surface and the bed surface, both of which have some uncertainty.Experimental flows may only be a few millimetres deep, making it especially important to quantify accuracy and precision in topographic data.Another outstanding challenge is that we need better methods to resolve the water surface, at high spatial and temporal resolutions to match the underlying topographic data.The water surface elevation is required for through-water applications of both TLS and SfM-we need to know where refraction of the TLS beam occurs, or what the 'uncorrected' water depth is that requires refraction correction.Even for methods to estimate flow depth from dye colour intensity, one still needs to estimate the water surface elevation in order to calculate flow depths in the calibration dataset used to relate dye intensities to depths, although calibration pans or prisms have provided a workaround to this hurdle(Gran & Paola, 2001;Tal & Paola, 2007;Huang et al., 2010).It is not very useful to acquire bed topography at millimetre-scale accuracy if the uncertainty around the water surface is several millimetres, particularly for experimental channels that are of the order of tens of millimetres deep.Existing methods to measure water surface topography-for instance, by measuring from the side of the flume or by using a row of ultrasonic sensors-work well in narrower flumes or single-threaded experiments, or in cases when topographic change is slow so that a scan time of several minutes is instantaneous relative to the pace of geomorphic change.However, for more laterally active or complex multithreaded flows in braided streams or deltas, we need further innovation to gain instantaneous, spatially continuous data on the water surface elevation.Promising progress was made on this problem byHan and Endreny (2014) andFerreira et al. (2017), who applied close-range photogrammetry to a water surface seeded with wax powder and cork particles, respectively.Of course, seeding the water surface prevents simultaneous through-water bathymetric mapping.Instead, dye-based optical depth methods coupled with through-water scanning (akin to that used byHuang et al., 2010, or Leenman & Eaton, 2021), or laser scanners that sense both the water and bed surface simultaneously(Moteki et al., 2022), might present a solution to this problem, particularly for multithreaded or rapidly evolving experimental channels in which depth and bathymetry vary widely in space and time.The outstanding challenge with the latter methods is to deploy them rapidly in larger experiments (thoughMoteki et al., 2022, had  promising results from their 10 m flume).Another ongoing challenge in the remote sensing of experimental rivers is that of turbid flows or rough water surfaces.These conditions inhibit the collection of through-water bathymetry, with either laser or photogrammetry methods.Many experimentalists have sidestepped this problem by using clear flows with minimal sediment suspension and calm water surface conditions.The ability to generate such flow conditions in the lab is one of the key advantages of experimental geomorphology over remote observations in field-scale rivers, in which geomorphically interesting flood conditions are often highly turbid.Nevertheless, these limits on the applicability of through-water bathymetry do preclude the collection of bathymetry during certain types of lab experiments.For instance, it is difficult to monitor bathymetric evolution during experiments with mudflows or debris flows.

Fernandez
et al. (2021), who adapted the methods of Brodu and Lague (2012) to filter vegetation from TLS data captured in their braided river experiments.Other experimentalists have developed their own vegetation filtering methods; for instance, van Dijk, Teske, Van de Lageweg, and Kleinhans (2013) used the 10th percentile (instead of the median) elevation from their laser profiler, while Perona et al. (2012) developed a threshold/gradient filter.Surveying laboratory topography beneath dense vegetation remains an outstanding challenge for lab-scale remote sensing; possible solutions include improved data-filtering methods or surveying technology.While high-resolution, high-frequency experimental monitoring opens new avenues for research, it also creates computational challenges.Storing raw photos for SfM photogrammetry, or even a time series of DEMs, requires more memory as data resolution and frequency increase.While instantaneous surveys (such as SfM with a camera array) decrease the time required to run an experiment by removing the need to stop flow for data collection, the time required to process the data increases due to the high volumes of data collected.Moreover, as we collect data from our experiments at higher frequencies, automated processing and analysis techniques become indispensable.Techniques that have worked well for smaller volumes

(
2002), who conducted surveys of bed topography every 10 min during a braided river experiment and examined how decreasing the survey frequency affected their estimates of scour and fill.Revisiting these experiments with 1 min or even sub-minute intervals between SfM data capture would allow us to investigate the optimal survey frequency for quantifying the processes in a given experiment, and to ascertain how that optimum might vary between processes in the same experiment or vary with imposed conditions such as the discharge or sediment supply.Finally, much of our understanding of alluvial fans stems from analogue experiments bySchumm et al. (1987),Whipple et al. (1998) andBryant et al. (1995), who ascribed much of the observed fan dynamics to fan and channel gradient but were unable to monitor gradients at high-resolution due to the equipment available at the time.Revisiting these hypotheses, or even repeating their experiments, with high-resolution and high-frequency topographic data would allow us to delve further into the dynamics of these systems.Ultimately, the remotely sensed data resolution and computation methods now available to the geomorphology community open new problems we can explore experimentally.These include:1.Coupling depth and topography observations to monitor how the cross-channel shear stress field affects deposition and erosion patterns in multi-threaded systems.2.Coupling depth and topography observations to observe at high frequency the dynamics of back-water-driven avulsions on deltas.3. Coupling high-frequency channel planform and topographic change (especially deposition) data to track the preservation of channel fills and overbank deposits in stratigraphy.4. Applying quantitative metrics for channelisation, connectivity and network topology to objectively monitor autogenic cycling in fans and deltas.5. Applying machine learning methods for classification and pattern recognition with data volumes that exceed the capacity of traditional analysis methods.8 | CONCLUSIONS We have reviewed the ways in which experimental fluvial geomorphology has borrowed techniques from the remote sensing of fieldscale rivers, as well as a few instances in which the flow of ideas has reversed with remote sensing methods developed in the lab and then deployed in the field.Experimentalists have adapted remote sensing methods for quantifying topography, including laser methods, stereopair photogrammetry and, more recently, structure-from-motion photogrammetry.They have also adapted methods to monitor flow depth, both by differencing the water surface and bed topography, and by relating the optical properties of the flow to its depth.They have used PIV and LSPIV to monitor flow velocities, applying techniques originally developed in the lab and improvements made in the field to the monitoring of channel-scale patterns in deformable-bed systems.Lastly, they have both developed and adapted methods to extract river planform from overhead imagery.These adaptations of remote sensing methods have allowed experimentalists to observe and understand the behaviour of laboratory rivers.Methods to monitor topography have allowed geomorphologists to quantify experimental evolution and to constrain the effectiveness of inverse methods to monitor sediment transport.Methods to monitor flow depth and velocity have made it possible to relate hydraulics to observed topography and to calibrate morphodynamic models.Methods to monitor planform have allowed the tracking of avulsions, the isolation and measurement of lateral mobility rates, and characterisation of channel planform and topology.Ongoing challenges in experimental remote sensing come from both the experiments themselves and from the large datasets we can now collect.Physical challenges that require further work are the rapid and accurate resolution of submerged bathymetry, flow depths and, crucially, the water surface topography.A primary analysis challenge is to develop new data reductions to extract physically meaningful insights from the quasi-continuous, high-resolution, spatially explicit, simultaneously obtained datasets on topography and flow dynamics that we are becoming able to collect.The combination of quasi-continuous, spatially explicit remotely sensed data with new analysis methods opens up a range of questions for experimental geomorphologists.For instance, we can use these data to revisit old hypotheses which may have been formed with less complete datasets; we can use these new data and methods to gain new insights into the co-evolution of channel hydraulics and topography in river systems.Finally, these new data collection and analysis methods could be particularly useful for exploring the behaviour of multi-threaded and laterally active channels, in which it has been difficult to collect complete topographic and hydraulic data until now.
Visconti et al. (2012)used a similar set-up that coupled a through-water laser scanner with a centrally placed ultrasonic water-level sensor.They were able to adjust the survey resolution so that data acquisition over a 1 m 2 section of the bed took around 3-4 min, during which bed-level change was negligible.Alter- Seizilles et al. (2013)l.(2013)developed a method to monitor both the water surface and channel bed at a single cross-section, using red and green laser sheets projected at different angles.Nevertheless, a single water surface profile may not be appropriate for all experimental conditions, particularly in stream table experiments where the water surface elevation and bed topography may vary appreciably across and down the channel.