Fluorescent particle tracers for surface flow measurements: A proof of concept in a natural stream


  • F. Tauro,

    1. Department of Mechanical and Aerospace Engineering, Polytechnic Institute of New York University,Brooklyn, New York,USA
    2. Dipartimento di Ingegneria Civile, Edile e Ambientale, Sapienza University of Rome,Rome,Italy
    3. Honors Center of Italian Universities, Sapienza University of Rome,Rome,Italy
    Search for more papers by this author
  • S. Grimaldi,

    Corresponding author
    1. Department of Mechanical and Aerospace Engineering, Polytechnic Institute of New York University,Brooklyn, New York,USA
    2. Honors Center of Italian Universities, Sapienza University of Rome,Rome,Italy
    3. Dipartimento per l'Innovazione nei Sistemi Biologici, Agroalimentari e Forestali, University of Tuscia,Viterbo,Italy
      Corresponding author: S. Grimaldi, Dipartimento per l'Innovazione nei Sistemi Biologici, Agroalimentari e Forestali, University of Tuscia, Via De Lellis snc, 01100, Viterbo, Italy. (grimaldi@unitus.it)
    Search for more papers by this author
  • A. Petroselli,

    1. Dipartimento di Scienze e Tecnologie per l'Agricoltura, le Foreste, la Natura e l'Energia, University of Tuscia,Viterbo,Italy
    Search for more papers by this author
  • M. Porfiri

    1. Department of Mechanical and Aerospace Engineering, Polytechnic Institute of New York University,Brooklyn, New York,USA
    Search for more papers by this author

Corresponding author: S. Grimaldi, Dipartimento per l'Innovazione nei Sistemi Biologici, Agroalimentari e Forestali, University of Tuscia, Via De Lellis snc, 01100, Viterbo, Italy. (grimaldi@unitus.it)


[1] In this paper, a new particle tracer for surface hydrology is proposed. The approach leverages the complementary advantages offered by particle-tracking velocimetry and traditional tracing technologies, such as dyes and chemicals, toward a practically feasible and low-cost measurement system. Specifically, the proposed methodology is based on the detection and tracking of buoyant fluorescent microspheres through an experimental system that incorporates ultraviolet lamps to elicit the fluorescence response and a digital camera to record the particle transit. This low-cost measurement system can be used in a variety of natural settings ranging from small-scale streams to rills with scales on the order of a few centimeters in hillslopes. The use of insoluble buoyant particles reduces the amount of tracing material for experimental measurements. Further, particles' enhanced fluorescence allows for noninvasive flow characterization, that is, for nonintrusively detecting the tracer without deploying probes and samplers in the water. A proof of concept experiment for the proposed methodology is conducted on the Rio Cordon, a natural mountainous stream in the Italian Alps. Flow measurements at selected stream cross sections and travel time experiments on varying stream reaches are performed to ascertain the feasibility of fluorescent particle tracers. Such experimental findings demonstrate that the particles are visible in complex natural streams and are effective in estimating flow velocities and travel times.

1. Introduction

[2] Accurate estimates of flow velocities in natural environments are essential for the understanding of fundamental phenomena such as runoff formation, rill development, erosion, and infiltration mechanics. These estimates are also crucial for the development of rainfall-runoff models and rating curves that are used in the design of infrastructures and water management systems. Whereas fully developed rivers are usually equipped with gauging stations and monitoring systems, the direct measurement of flow velocities in small and mountainous watersheds is hampered by several technical and logistic challenges. Specifically, limited accessibility, high-gradient reaches, extremely shallow water depths, and heavy suspended sediment loads represent critical elements in practical implementations [Calkins and Dunne, 1970; Comiti et al., 2007; Jodeau et al., 2008; Tazioli, 2011].

[3] To address these challenges, a variety of tracing technologies have been to date implemented to estimate flow velocities in small watershed streams and shallow water flows [Lei et al., 2005; Comiti et al., 2007; Lei et al., 2010; Zhang et al., 2010; Tazioli, 2011]. A great advantage of such methods lies in the fact that, once tracers are deployed in natural streams, they flow with the water and thus provide direct access to the travel time [Leibundgut et al., 2009]. Such information can be in the form of dye and chemical breakthrough curves [Calkins and Dunne, 1970; Planchon et al., 2005; David et al., 2011] obtained from collected samples [Wilson, 1967; Hunt et al., 2010] or from probe measurements [Schuler et al., 2004; David et al., 2011]. Even simple visual inspection is possible for highly visible dyes, such as rhodamine and uranine [Dunkerley, 2001; Smith et al., 2010]. Despite their versatility and efficiency, these techniques suffer from adsorption to natural substrates [Finkner and Gilley, 1986] and are therefore applicable to stream reaches of limited extent and under moderate to low sediment loads [Tazioli, 2011]. In addition, environmental imperviousness and severe flow regimes, such as floods or extremely shallow flows, may prevent the deployment of probes in the water, physical sampling, and the presence of technical personnel to conduct basic operations [Calkins and Dunne, 1970].

[4] A promising approach to perform local flow measurements in large rivers is large scale particle image velocimetry (LSPIV), in which the velocity field in a natural stream is obtained through a nonintrusive measurement system with automated analysis procedure [Fujita et al., 1998]. While this information is only restricted to the stream surface, such data can be used to infer average properties throughout the cross section [Dunkerley, 2001]. In LSPIV, a remote camera mounted at a considerable height and inclined from the water surface acquires sequences of pictures of a large portion of the river surface [Kim et al., 2008; Hauet et al., 2009]. Images are then analyzed through correlation algorithms originally developed for laboratory PIV experiments [Raffel et al., 2007]. In particular, the flow velocity is reconstructed by performing statistical analysis of the displacement of groups of seeded or naturally occurring tracers [Muste et al., 2008]. However, LSPIV accuracy may be limited by insufficient or excessive illumination [Bechle et al., 2012], identification of ground reference points [Kim, 2006; Fujita et al., 2007], and environmental factors affecting the tracers [Fujita and Kunita, 2011; Muste et al., 2011]. LSPIV is a powerful methodology for estimating rating curves in large riverine systems such as valley rivers [Hauet et al., 2008]. Conversely, its implementation may be difficult when transferred to mountainous creeks characterized by step and pool reaches and highly irregular water surfaces [Hauet et al., 2009]. Interesting results in local discharge measurements are also achieved from the use of image analysis tools for identifying rising air patterns [Hilgersom and Luxemburg, 2012]. Yet such methodology is highly influenced by stream morphology and flow regime.

[5] In this paper, a novel particle tracer for surface flow velocity and travel time measurements in natural environments is proposed. The enhanced visibility of the fluorescent particles along with their ability to accurately trace surface flows are successfully orchestrated into an innovative technique that combines the efficiency and versatility of traditional conceptions while improving their practical feasibility. Specifically, the proposed tracing methodology is developed into a low-cost measurement system that can be applied to a variety of real-world settings, spanning from small scale streams to rills with scales on the order of a few centimeters in natural hillslopes. The technique is based on the detection and tracking of fluorescent microspheres with a digital camera for direct flow measurements and stream reach travel times similarly to particle tracking velocimetry [Tang et al., 2008]. The use of insoluble buoyant particles limits the tracer dispersion from adsorption to natural substrates and thus reduces the amount of tracing material for experimental measurements. Particles' enhanced fluorescence allows for nonintrusively detecting the tracer without deploying probes and samplers in the water. Moreover, this inexpensive and automated technology is inherently designed to provide continuous and unmanned measurements in complex flows, including shallow water streams, overland flows, and mountainous creeks.

[6] Promising results on the characterization of the fluorescent particles in static and dynamic laboratory conditions in both daylight and dark are presented by Tauro et al. [2010, 2011]. Tauro et al. [2011] ascertained the suitability of these fluorescent particles in tracing water flows through the analysis of the particle response to different flow velocities computed through the Basset-Boussinesq-Oseen model [Soo, 1967]. The geometry and weight of the particles allow for easily detecting and tracking their trajectories in water flows through an automated procedure based on image analysis. The processing scheme uses a computationally inexpensive algorithm which correlates recorded video frames and a particle template image. As demonstrated by Tauro et al. [2011], this procedure is robust to daylight reflections and turbulent flows while only using a low-cost digital camera and low-power ultraviolet (UV) lights.

[7] As a proof of concept experiment of the feasibility of this methodology in natural settings, this fluorescent particle tracer is here used in the Rio Cordon natural mountainous stream in the Italian Alps. Two classes of experiments are conducted, that is, flow measurement experiments (FMEs) at a selected stream cross section and travel time experiments (TTEs) on stream reaches as long as 30 m. Specifically, FMEs are performed to estimate the flow velocity in an artificial section of the stream by deploying the fluorescent particles in proximity of the experimental apparatus and then tracking their transit in the field of view of the camera. TTEs are conducted to estimate the travel time for varying amounts of fluorescent particles deployed in natural stream reaches of different lengths. Flow velocities are estimated by using the tracking algorithm developed by Tauro et al. [2011], whereas recorded videos for the travel time extraction are analyzed by using the index introduced by Tauro et al. [2010]. An array of traditional tracers, including a dye and several floating objects, is also considered in this experimental study to benchmark the proposed approach.

2. Materials and Methods

2.1. Tracers

[8] The fluorescent particles are purchased from Cospheric LLC (http://www.cospheric-microspheres.com) and their cost is approximately $300 for a 1 kg sample. The spheres are white under daylight and emit yellow-green light (561 nm wavelength) if excited by a UV light source (365 nm wavelength; see Figure 1). The particle material is polyethylene and the fluorophore is embedded in the polymer matrix which allows for a long luminescence lifetime. The particles are slightly buoyant and their nominal dry density is inline image. Particles with diameter in the range inline image are used for this proof of concept experiment.

Figure 1.

View of 0.8 mm off-the-shelf fluorescent particles under (left) daylight and (right) 365 nm UV light.

[9] Measurements from the proposed methodology are compared to results obtained with more traditional tracers. In particular, FMEs are validated against experiments performed with rhodamine WT (water tracer) and corks. Further, TTEs are compared to experiments conducted with rhodamine and an array of floats, that is, polystyrene beads, corks, and plastic hollow spheres. All these floats have diameter of approximately 4 cm.

2.2. Experimental Apparatus

[10] The passage of the particles is recorded as the beads transit under an experimental apparatus that hosts both the fluorescence excitation and video acquisition units (see Figure 2). The apparatus is composed of a horizontal telescopic system of aluminum bars that rests on adjustable steel tripods. The horizontal telescopic system includes three bars that allow for a total extension of the setup of approximately three meters. Another telescopic system of vertical aluminum bars is connected to the lamp case. The case contains an array of fourteen UV inline image lamps in parallel and series connections. On the upper side of the case, a heavy duty aluminum plate connects the lamp unit to the bars. On the lower side of the case, a Poly methyl methacrylate plate protects the lamps from water. An HD CMOS Canon VIXIA HG20 camera is located on a tripod head connected to a vertical bar. The distance of the camera from the surface of interest is regulated through a compensating counterbalance. A metric ruler installed on the bars allows for calibration of the acquired videos. The distance of the lamp and video units from the surface can be adjusted according to illumination and flow conditions.

Figure 2.

Experimental apparatus including the lamp and video units. The telescopic systems allow for a remarkable flexibility of the setup in different environmental conditions.

2.3. Experimental Site

[11] The Rio Cordon watershed is a inline image natural basin located in the Dolomites, northeastern Italy. The Rio Cordon is a tributary of the Fiorentina stream that in turn flows into the Rio Cordevole. The basin drainage network extends for approximately inline image at an average slope of 47.85%.

[12] The station is equipped with a coarse sediment grille and a diversion pool for water and finer material. Moreover, turbidity, water discharge, and water quality are continuously monitored and transmitted to the Agenzia Regionale per la Prevenzione e Protezione Ambientale del Veneto (ARPAV) Arabba Avalanche Center for processing. The limited extension of the watershed along with the absence of human settlements and hydraulic infrastructures make it an ideal candidate for research purposes [Fattorelli et al., 1988]. The station is located at 1763 m above sea level where the area drained by the stream is approximately 5 km2 large. In the station proximity, the stream presents a step and pool bed whose width is approximately 5–6 m wide and with a 13% mean slope (see Figure 3a). At the station, the stream flows on a concrete inlet channel where the water gauge is installed (see Figure 3b). The concrete channel is approximately 6 m long, and its width and slope are equal to 1 m and 3%, respectively (see Figure 3c).

Figure 3.

(a) Rio Cordon natural streambed characterized by step and pool reaches. (b) View of the artificial concrete inlet where the 4 and 6 m reaches and apparatus section are reported. The water gauge is located 1 m upstream the experimental setup. (c) Upstream view of the artificial inlet.

[13] In this study, the experimental apparatus enclosing the lamp and video units is located at a cross section of the artificial channel approximately 1 m downstream of the water gauge (see Figure 4). The setup rests on a tripod on the stream left bank and on an iron bracket on the right bank concrete wall.

Figure 4.

Experimental setup at the Rio Cordon station. The apparatus rests on a tripod on the hydrographic left bank and on an iron bracket on the right bank. The distance of the camera from the stream surface is adjusted through a compensating counterbalance.

2.4. Data Analysis

[14] Here the detection and tracking scheme for FMEs and the video analysis procedure for TTEs are described.

2.4.1. Flow Measurement Experiments

[15] Estimates of the flow velocities are obtained by implementing the tracking algorithm developed by Tauro et al. [2011] on a sequence of frames depicting the transit of a fluorescent particle under the experimental apparatus described in section 2.2. Analyzed pictures are first processed to extract the green channel and then orthorectified to eliminate distortions due to camera inclination. The algorithm takes advantage of the known shape of a bead as it appears in frames acquired during the initial calibration phase of the instrumentation in the field. Further, it is assumed that a particle does not travel backward as it crosses the area of interest, which is reasonable when dealing with high-gradient mountainous streams.

[16] The implementation is based on the correlation between a template image and frames converted from the videos. For reference, an artificial template that represents an ellipse against a gray tone background is generated. The ellipse major and minor axes are determined by averaging the pixel width and height of the particles as they appear in acquired frames. The pixel intensities within the ellipse are graded from white tones in the center to the background tone to account for the diffusion of the boundaries of the particles in the image.

[17] The procedure is based on using the following normalized correlation inline image between the image and a selected template [Lewis, 1995], which in its simplest instance is the artificial template,

display math

where inline image identifies the coordinate system in the correlation matrix and inline image indicate the pixel position in the image (see Figure 5). In equation (1), f and t are the image and the template intensities as functions of the pixel position and inline image and inline image are the average template intensity and the average intensity of the image portion overlapping the translating template, respectively. Frames where the particle is present display higher correlation values. The correlation analysis is further sharpened by preprocessing the images from experiments by a background subtraction procedure. This operation allows for simplifying the images by preserving their shape and eliminating irrelevant items [Haralick et al., 1987].

Figure 5.

Sketch of the detection and tracking procedure. (a) Frames containing the particles and the artificial template are depicted. (b) The frame sequence is correlated with the template, and the location of the maximum correlation coefficient is tracked. (c) Velocity is obtained from the values of inline image by taking the inverse of the camera acquisition frequency as the time interval.

[18] The tracking procedure consists of first performing the background subtraction on the acquired images and then computing inline image in the entire image through equation (1) specialized to the artificial template. After correlation with the first image of a video, the maximum correlation coefficient inline image is compared to an ad hoc threshold value inline image. If inline image, a new template image is generated by cropping out of the previous image a square of dimension equal to the artificial template centered about inline image, where inline image is attained. The updated template is then used for correlation with the successive frame. The value of inline image is compared to inline image at each step of the procedure and the coordinates of the maximum value of the correlation coefficient inline image that identify a possible location of the particle, if any is present, are collected. The procedure of template updating is verified at each step of the algorithm by computing the correlation between the updated template and the artificial template. If the correlation falls under inline image, the updating process is discarded and the artificial template is used to identify the bead location.

[19] The velocity of the potential particle in the plane is estimated by collecting the values of inline image in the frame sequence and multiplying them by a scale factor to obtain metric units. The in-plane components of the velocity are calculated by taking the inverse of the camera acquisition frequency as the time interval. If both the velocity and angle between two consecutive beads and the inline image axis in Figure 5 are within prescribed bounds, the tracked point is more likely to pertain to a particle flowing in the channel. Further postprocessing of the images is used to validate the data obtained from the tracking algorithm.

2.4.2. Travel Time Experiments

[20] In travel time experiments, particle deployment in the stream and video acquisition are synchronized and therefore the bead release time is considered to be known. The mean time that particles take to travel along a stream reach is obtained by computing the time at which the particles cross the region focused by the camera. This is accomplished by converting the recorded videos into RGB frames and then analyzing the green channel by using a modified version of the index inline image defined by [Tauro et al., 2010]

display math

with inline image. Here inline image and inline image refer to the pixel count for the background and particle images, respectively. The term ci represents the intensity classes from 0 to 255 where the power inline image is here introduced to assign a higher weight to the brighter pixels that correspond to tones pertaining to fluorescent beads. It is noted that background images are obtained from the original ones by applying a bottom-hat transformation [Haralick et al., 1987; Gonzalez et al., 2004]. Preprocessing through the index inline image allows for identifying the sequence of frames where brightness is maximized and therefore where the particles are more likely to be (see Figure 6). This procedure can significantly reduce time consuming analysis for long videos. As the brightest frame sequences are recovered, the order number of images pertaining to the particle transit is directly identified and the arrival time is calculated from the camera acquisition frequency.

Figure 6.

Analysis of an entire travel time frame sequence through the index inline image. In the experiment, a batch of particles are deployed upstream from the camera, and their transit is recorded. The processed video is decompressed into 959 frames. (a) Color frames acquired before (693) and during (705) the particle transit in the camera field of view. Beads are visible as bright green lines in picture 705. (b) The index inline image is computed on the entire frame sequence, and relative values are displayed on a semilogarithmic plot. The transit of the particles occurs at frames 694–720, and the brightness peak is found at frame 705. Note that parameter inline image is set equal to 10 in this analysis.

2.5. Procedure

[21] Experiments are conducted during the same day in September 2011 from the early afternoon to sunset to test various illumination conditions. During the preliminary calibration phase of the experimental apparatus, the lamp unit is lowered at 12 cm from the water surface and the camera, at 17 cm from the surface, is inclined at a low angle to focus a 30 cm × 15 cm area located upstream the video unit in the middle of the channel. The camera acquisition frequency is equal to 30 Hz. Its resolution is set to inline image whereas the appropriate shutter speed is automatically adjusted according to the zoom position. Two black panel screens are integrated in the apparatus, upstream of the camera, to reduce daylight reflections.

2.5.1. Flow Measurement Experiments

[22] Surface flow velocity in FMEs is measured by applying the tracking algorithm on the transits of single fluorescent particles under the apparatus in the stream center. In Table 1, input parameters to the tracking algorithm are reported. In these experiments, approximately fifty fluorescent particles of diameter equal to 1.09 mm are individually deployed within few tens of centimeters from the lamp unit.

Table 1. Input Parameters to the Tracking Algorithm for the Flow Measurement Experiments (FMEs) Procedure
Template size270 × 270 pixels
Artificial ellipse size70 × 3 pixels
Template background tone55
inline image0.3
Admissible velocity range(0.01, 2.5) m s−1
Admissible angle range inline image

[23] Because of the inclined position of the camera with respect to the stream surface, recorded images appear distorted. Distortions are corrected by rectifying the images according to a projective transformation. Specifically, image feature points are identified in the preliminary calibration video where the metric ruler lies in the center of the camera field of view. A perspective projection matrix is then built from the undistorted position of the feature points [Forsyth and Ponce, 2002]. Once dewarped, the green channel of the pictures is rotated, resized, and converted to metric units before processing through the tracking algorithm described in section 2.4.1.

[24] Flow velocity estimates are validated against results obtained using traditional tracing methodologies. Two experiments are conducted by either releasing 0.5 mL of rhodamine or 1/2 cork along the 6 m concrete inlet (see Figure 3b).

2.5.2. Travel Time Experiments

[25] In TTEs, the mean time taken by tracers to travel along a stream reach of known length is measured. The number of experiments, the length of the flow paths, the type of tracer, and the amount of used tracer material are reported in Table 2.

Table 2. Summary of the Travel Time Experiments (TTEs)
4 m0.8 mm bead1, 4 g2
 1.09 mm bead2 g12
12 m0.8 mm bead12, 18 g2
27 m0.8 mm bead84 g1
 Rhodamine2 mL1

[26] In experiments performed with fluorescent particles, the release time of particles is recorded by the operator, whereas the arrival time at the section where the experimental apparatus is located is computed by analyzing the videos captured with the camera. Camera recording is generally initiated before particle release to minimize the effect of operators' reaction time. As reported in section 2.4.2, frames are first processed by extracting values of the index inline image (see equation (2)) to isolate the brightest sequences. On the basis of trials, the exponent inline image is set equal to 10 to maximize the contrast. Then, such sequences are manually analyzed to identify the first frame relative to the appearance of the particle tracer.

[27] In experiments conducted with the traditional tracing methodologies, tracers are released in the middle of the channel. The release and arrival times are communicated between two synchronized operators in real time. It is noted that the relatively meager number of repetitions for experiments performed with traditional tracers does not affect the paper objective, which is to verify the feasibility and efficiency of the novel tracing methodology. Travel time relative to rhodamine is estimated by taking videos and selecting the first frame where the dye is observed underneath the detection apparatus. This provides conservative estimates for the travel time.

3. Results and Discussion

3.1. Flow Measurement Experiments

[28] In experiments performed with the fluorescent particles, despite the minute size of the spheres and the relatively low camera acquisition frequency, each of the fifty beads crossing the camera field of view is visible in the frame sequence. This evidence supports the feasibility of the methodology in natural settings. Out of these fifty repetitions, five sequences of frames are retained for estimating the flow velocity based on the constraint of including at least three consecutive frames where the particle is correctly detected and tracked by the automatic procedure. Such frames are identified through the times t0, t1, and t2 that are inline image apart.

[29] In Figure 7, one of these five sequences is displayed for illustration purposes. Specifically, in Figure 7a, the orthorectified green channels of three consecutive images containing the particle are depicted. Here, the particle appears as a brighter rectangle that advances from left to right from the time t0 to t2. A red box highlights the bead against the background. Brighter areas in the images are due to lamp reflections on the stream water surface. Figure 7b reports the locations inline image for the bead transit along with the values of the correlation coefficient. For each of the five sequences, the velocities estimated from the tracking algorithm at the instants t0, t1, and t2 are averaged to compute a mean particle velocity. These five mean velocities are further averaged to obtain the estimate reported in Table 3 along with the relative standard deviation.

Figure 7.

(a) Image sequence of the particle transit under the lamp unit. The red box identifies the location of the bead in the pictures. (b) Colored dots representing inline image for each frame of the sequence. The color bar indicates the values of the correlation coefficients from 0 to 1.

Table 3. Velocities for the FMEs for Each Tracing Methodology
TracerVelocity (m s−1)SD
1/2 cork1.88n/a
Fluorescent bead1.890.18
Rating curve1.89n/a

[30] Sources of uncertainties in velocity estimates using the proposed methodology may reside into possible errors in the image orthorectification phase which may affect velocity estimation. Furthermore, the correlation algorithm may fail to detect the particle if the images contain objects similar to the beads in their geometry and intensity. This issue can be easily solved by adjusting the illumination of the field of view along with camera settings, such as the shutter speed and aperture. Moreover, given the rectangular shape of the bead image (see Figure 7), the algorithm may correlate with pixels in different portions of the particle in subsequent pictures, that is, sometimes the tail of the rectangle and sometimes its head are identified as the location of the maximum correlation coefficient. This introduces some variations in the velocity values that can be reduced by averaging over the entire sequence of frames.

[31] Table 3 reports mean flow velocities along with data from water gauge measurements and rating curves as provided by the Arabba Avalanche Center. Average flow velocities for the rhodamine and cork tracers are inferred from travel times measured along the artificial inlet. The stream cross-sectional area at the gauge is calculated on the basis of the water level measurement of 0.094 m (stable during trials executed in few hours). Flow velocity prediction through rhodamine is slightly lower than the other estimates. This can be related to the fact that the dye disperses and diffuses vertically and laterally in water, thus tracing the average stream velocity rather than the actual surface flow velocity at the observed position in the stream. Such diffusion is probably due to the relatively long flow path length of 6 m used in the estimation. A shorter flow path is not considered here for synchronization issues [Dunkerley, 2001] that are amplified by the smoothness of the concrete inlet, which leads to rapid flows, and would be further emphasized in shorter reaches.

[32] The estimate of the surface velocity from the rating curve is affected by inherent uncertainties and variations in the water depth and cross section velocity profile. In addition, the water gauge is located approximately 1 m upstream the experimental apparatus on the river right bank, where the stream may be slowed down by the effects of the wall and the hydraulic jump at the beginning of the artificial inlet. Nonetheless, velocity values are in good agreement with estimates obtained by using fluorescent particles and cork.

3.2. Travel Time Experiments

[33] Table 4 reports the main results grouped according to the flow length traveled by the tracer. Empty entries refer to failed experiments where the deployed objects disperse or stop before reaching the final section of the path. These events can be attributed to the extremely irregular stream channel bed. In particular, bulky tracers such as corks tend to stop at stream pool sections where small hydraulic jumps trap them thus nullifying the measurement. Table 4 does not include results for polystyrene beads and 4 cm plastic spheres, which failed to provide any usable result in all tested conditions because of their extremely low density, which hampered their flow in the irregular stream.

Table 4. Average Velocities (m s−1) for the TTEs Performed With Different Tracing Methodologiesa
4 m12 m27 m
  • a

    Values in parentheses are the standard deviations for measurements performed with the fluorescent particle tracer.

Fluorescent beads1.7 (0.21)1.46 (0.16)0.98
Rhodamine  0.68
1 cork  0.55
1/2 cork  none

[34] In travel time experiments, accuracy of the measurements lies in the precise synchronization of the operators [Dunkerley, 2001] and the rigor of the operators in deploying the tracers at the center of the stream cross section to have the objects flow along similar paths. These parameters play a crucial role in the experiments and may be responsible for slight variations in the values reported in Table 4.

[35] Despite these sources of uncertainties, the results consistently highlight that longer flow paths corresponds to slower average velocities. This observation can be substantiated by the varying composition of the streambed that encompasses an initial smooth streambed of 4 m composed of concrete followed by natural step and pool formations. Therefore, it is expected that the average velocity rapidly decreases after inline image as natural streambed is incorporated in the flow path.

[36] The average velocity estimated with rhodamine is usually lower than the velocity of the fluorescent particles. This may be explained by the fact that rhodamine diffuses in water along the entire depth of the stream whereas the beads lie on the water surface where flow is usually faster than at deeper levels. Interestingly, the difference between rhodamine and fluorescent particles' predictions reflects the inherent variations of the cross-sectional velocity profile. Such variations can be lumped into the velocity index inline image, where inline image and inline image are the flow velocities averaged over the stream depth and at the stream surface, respectively. Therefore, the rhodamine velocity estimate can be taken as inline image, while inline image can be assimilated to the velocity predicted by using the fluorescent particles. With reference to the inline image flow path, inline image equals 0.7 which is in good agreement with standard estimations used in the literature [Dunkerley, 2001].

[37] It is commented that velocities obtained from corks are significantly lower than estimates from other tracers. In some cases, corks fail to accurately flow on the water surface and are trapped in lateral pool areas.

[38] An additional tentative travel time experiment is performed on a 120 m flow path, not reported in Table 4. A sample of approximately 170 g fluorescent particles is deployed in the reach as well as rhodamine and floating objects, such as polystyrene beads, corks, and plastic spheres. Soon after the deployment, the dye disperses in the flow, whereas the floats are trapped into stream pools. Conversely, some fluorescent particles reach the path final section, thus crossing the camera field of view against dispersion phenomena, such as those due to surface tension. Therefore, the methodology may still be implemented on long stream reaches if larger quantities of tracing material are deployed.

4. Conclusions

[39] In this paper, the proof of concept experiment of a novel tracer is conducted on the Rio Cordon natural mountainous stream in the Italian Alps. The selected experimental site is characterized by step and pool reaches along with surface water turbulence, shallow water depths, and high flow velocities that constitute serious challenges for traditional tracing technologies. The proposed tracing methodology is based on the automatic detection and tracking of buoyant fluorescent beads through an experimental apparatus that hosts UV lamps to elicit the fluorescent response and a digital camera to record the particle transit.

[40] The proposed approach seeks to reduce practical limitations of traditional tracers that affect their implementation in natural environments, such as mountainous streams and complex river beds, by integrating low-cost fluorescent particles and off-the-shelf instrumentation with nonintrusive and continuous-time image-based methods. Specifically, fluorescent particles are used in two separate classes of experiments, that is, flow measurement at a selected stream cross section and travel time estimation on varying stream reaches. A comparison between estimates obtained through the proposed method and those stemming from the implementation of traditional tracers, such as rhodamine dye and buoyant floats, is presented to assess the efficacy of the proposed approach.

[41] The main contributions of this work are: the development of a versatile experimental apparatus for field studies; tuning of previously developed algorithms for particle tracking and detection in laboratory conditions for natural environments; performing the first proof of concept field study for this novel tracer. Experimental results demonstrate that even limited quantities of fluorescent particles are detectable in such adverse environmental conditions using relatively simple and low-cost instrumentation. Specifically, few grams of microspheres can travel up to inline image and still be automatically identified to estimate the travel time. In addition, single microspheres can be automatically tracked once released in the proximity of the apparatus to allow for estimating the cross-section surface velocity of 2 m s−1. Both these measurements are consistent with results obtained with more common tracing methodologies.

[42] While the experimental results presented in this study offer a sound benchmark for the proposed particle tracers, they should be considered as a conservative assessment of its potential. Indeed, next-generation low-cost cameras are expected to be capable of acquiring at 120 Hz, thus considerably increasing the amount of frames used for the particle tracking and velocity estimate. Moreover, advances in the image resolution and filtering are also expected to positively impact the particle detection. Finally, improvements are expected to be observed when studying streams characterized by more uniform beds and slower flow velocities.

[43] Future ameliorations of the methodology will include the development of in-house fabricated biocompatible beads that will be engineered to degrade in time spans of a few days while maintaining enhanced fluorescence and limited cost. The experimental apparatus will be refined by incorporating a smaller camera to the lamp unit to have the lens parallel to the water surface and therefore avoid the image orthorectification process. An automatic seeder will also be built and synchronized to the video recording process to reduce uncertainties on the bead deployment time. Finally, a sensitivity analysis of the performance of the methodology will be conducted on different water bodies and a feasibility analysis will be performed on a natural hillslope. These phases will provide useful insight on the effects of pertinent variables, such as illumination intensity and minimum seeded particle quantities and diameters, on the tracer accuracy.


[44] This work was supported in part by the MIUR project PRIN 2009 N. 2009CA4A4A “Studio di un tracciante innovativo per le applicazioni idrologiche,” in part by the National Science Foundation under grants CMMI-0745753 and CMMI-0926791, and in part by the Honors Center of Italian Universities (H2CU Sapienza Università di Roma). The authors would like to greatly acknowledge G. Scussel, F. Sommavilla, W. Testor, and S. Palla from the Agenzia Regionale per la Prevenzione e Protezione Ambientale del Veneto Arabba Avalanche Center for their cooperation in the experimental campaign at the Rio Cordon hydrologic station. R. Rapiti, G. Cipollari, I. Capocci, and G. Mocio are also thanked for realization of the experimental apparatus and helping with the experiments. The authors would also like to express their gratitude to Alexandre Hauet, Markus Hrachowitz, and one anonymous reviewer for their careful reading of the manuscript and for giving useful suggestions that have helped improve the work and its presentation.