Towards underwater plastic monitoring using echo sounding

15 Plastics originating from land are mainly transported to the oceans by rivers. The total plastic transport 16 from land to seas remains uncertain because of difficulties in measuring and the lack of standard 17 observation techniques. A large focus in observations is on plastics floating on the water surface. 18 However, an increasing number of observations suggest that large quantities of plastics are transported 19 in suspension, below the water surface. Available underwater plastic monitoring methods use nets or 20 fish traps that need to be deployed below the surface and are labour-intensive. In this research, we 21 explore the use of echo sounding as an innovative low-cost method to quantify and identify suspended 22 macroplastics. 23 Experiments under controlled and natural conditions using a low-cost off-the-shelf echo sounding 24 device show that plastic items can be detected and identified up to 7 m below the river surface. Eight 25 different debris items (metal can, cup, bottles, food wrappers, food container) were characterized based 26 on their reflection signature. Reflectance from plastic items diverged significantly from organic 27 material and non-plastic anthropogenic debris. During a multi-day trial field expedition in the 28 Guadalete river, Spain, half of the observed plastics items were found below the surface. As most 29 plastic monitoring and removal strategies focus on the upper layer, a substantial share of the total 30 plastic transport may be neglected. With this paper we (1) demonstrate that echo sounding is a 31 promising tool for underwater plastic monitoring, and (2) emphasize the importance of an improved 32 understanding of the existing plastic loads below the surface. 33

Plastics originating from land are mainly transported to the oceans by rivers. The total plastic transport 16 from land to seas remains uncertain because of difficulties in measuring and the lack of standard 17 observation techniques. A large focus in observations is on plastics floating on the water surface. 18 However, an increasing number of observations suggest that large quantities of plastics are transported 19 in suspension, below the water surface. Available underwater plastic monitoring methods use nets or 20 fish traps that need to be deployed below the surface and are labour-intensive. In this research, we 21 explore the use of echo sounding as an innovative low-cost method to quantify and identify suspended 22 macroplastics. 23 Experiments under controlled and natural conditions using a low-cost off-the-shelf echo sounding 24 device show that plastic items can be detected and identified up to 7 m below the river surface. Eight 25 different debris items (metal can, cup, bottles, food wrappers, food container) were characterized based 26 on their reflection signature. Reflectance from plastic items diverged significantly from organic 27 material and non-plastic anthropogenic debris. During a multi-day trial field expedition in the 28 Guadalete river, Spain, half of the observed plastics items were found below the surface. As most 29 plastic monitoring and removal strategies focus on the upper layer, a substantial share of the total 30 plastic transport may be neglected. With this paper we (1) demonstrate that echo sounding is a 31 promising tool for underwater plastic monitoring, and (2) emphasize the importance of an improved 32 understanding of the existing plastic loads below the surface. 33

Introduction 35
Plastic pollution in aquatic ecosystems is of increased global concern due to its negative impact on 36 ecosystem health and human livelihood (Cózar et al., 2014;Lau et al., 2020;van Emmerik & Schwarz, 37 2020). Much of the plastic daily discarded on land is leaked into rivers, and transported into the world´s 38 oceans (Schmidt et al., 2017;van Emmerik & Schwarz, 2020). However, estimates of plastic emissions 39 from rivers into the oceans are associated with great uncertainties due to methodological difficulties to 40 accurately quantify land-based plastic fluxes into the aquatic environment. To improve the 41 understanding of plastic transport dynamics from source to sink, reliable observations are crucial. 42 43 Plastics are abundant in all components of river systems: floating at the surface, accumulated on 44 riverbanks and floodplains, deposited in the sediment, and suspended in the water column ( objects was placed at the bottom of a small water tank and forward-looking sonar images were 66 generated using an ARIS Explorer 3000 sensor. Investigating the reflections of specific items and 67 opportunities to detect plastic items in more dynamic water bodies, such as rivers, has not been done 68 to date. 69 70 The main goal of this research is to explore the potential of echo sounding for riverine macroplastic 71 (>0.5 cm) monitoring below the water surface using an off-the-self low-cost sensor. We systematically 72 investigated the use of sonar for plastic monitoring through (1) indoor controlled tests, (2) semi-73 controlled outdoor tests, and (3) uncontrolled application under natural conditions. The controlled tests, 74 to get an insight into the scanning technique and detection abilities of the echo sounder, were performed 75 in a swimming pool. During these tests, several influencing factors on the sonar signal were examined. 76 The semi-controlled tests were carried out in the Rio de San Pedro, Spain. This test aimed to investigate 77 the plastic detection of sonar for different plastic items. Lastly, the sonar was applied for macroplastic 78 monitoring under natural conditions in the Guadalete river, Spain. In this paper, we demonstrate that 79 (1) plastics can be detected below the surface using sonar, (2) specific macroplastic items have unique 80 reflections, and (3) results from the Guadalete river suggest plastic items below the surface accounts 81 for a substantial share of the total transport. 82 The Deeper CHIRP+ operates with the Deeper Smart Sonar mobile application, which can be installed 113 on a phone or tablet. In the app, the different settings, such as the scanning beam width and sensitivity 114 can be selected. Besides the sonar readings, information about the water depth and temperature are 115 provided in the app. The sonar scan data can be saved and uploaded to Lakebook, an online platform 116 where data of the scanning activities can be stored and viewed. From Lakebook, only raw bathymetry 117 data can be exported as CSV format. Exporting raw data on signal strength and intensity is not possible. 118 This sensor was chosen because of the ratio between scanning resolution/target separation and price. 119 Besides, the ability to save and store scanning data was advantageous. The downside of this sensor is 120 the limitation of raw sonar data export. 121 122 Since raw sonar data could not be exported, screenshots of the sonar signal reflections were taken and 123 processed using MATLAB. The obtained screenshots were segmented, using K-Means clustering, to 124 exclude the background pixels (Shan, 2018). Binary images were obtained from which the dimensions 125 of the sonar signal reflection could be calculated in pixels. The 'width' of the sonar reflection in the 126 backscatter imagery depicts the time the object is underneath the transducer, and is influenced by 127 velocity of the flow (and object) with respect to the sonar transducer. To correct for this, the width and 128 depth dimensions of the sonar reflection were calculated separately. The signal width was scaled for 129 the flow velocity measured by recording the time of movement over a known distance. An example 130 sonar recording including a plastic bag, bottle and fish is shown in Figure

Controlled tests in the pool 136
The controlled tests aimed to investigate influencing factors on sonar reflection, such as the orientation 137 of objects, flow velocity and object depth. We conducted three experiments to isolate the effects of (1) 138 object size, (2) object depth, (3) flow velocity. Additionally, we tested the influence of object 139 orientation on the sonar signal reflection. 140 141 The controlled tests were done in the Kerkpolder swimming pool in Delft (51°59'25.9"N 4°19'53.3"E). 142 A framework of ropes was constructed, allowing passing items underneath the sensor at different 143 depths, velocity, and orientation, see Figure 2. We minimized the influence of object orientation during 144 the first experiments by using spherical balloons filled with water as test objects. The reflected signal 145 was therefore mainly influenced by actual object size, depth and flow velocity. respectively. These different experiments were repeated ten times. We tested the influence of object 157 orientation in a separate experiment. For this, we used a filled 1.5 L plastic water bottle. The bottle was 158 fixed to a depth of 1 m and held horizontally orientated for a duration of 30 seconds. This was thereafter 159 repeated for the bottle being vertical orientated. 160 161 The used echo sounder has several options for beam width. We used a beam angle (total angle) of 7 162 degrees, which provides the highest scanning resolution (target separation of 1 cm) and lowest spatial 163 resolution (smallest scanning area). These beam settings result in a blind zone of 15 cm at the water 164 surface, for which the sensor is not able to detect objects due to surface clutter. In the end, the 165 significance of the results was determined using an independent t-test with 0.05 as significance level. 166 167

Semi-controlled tests in the Rio de San Pedro 168
Semi-controlled test were carried out in in the Rio de San Pedro, a tidal river close to the city of Puerto 169 Real (36°31'53.9"N 6°12'56.5"W). The goal was to obtain data on plastic detection with sonar for 170 different plastic items. The sensor was deployed in the Rio de San Pedro (Figure 3 (1)) to collect 171 reflection signals for specific plastic objects, and test the performance under natural river conditions. 172 The experiment was conducted by releasing a set of objects, attached to thin fishing lines, repeatedly 173 into the river, passing the scanning beam of the sensor between 0.5 and 2.5 m below the surface. As 174 the objects were released into the river, they passed the sensor driven by the river flow velocity, as 175 illustrated in Figure 4 (1). This was repeated ten times per item. To obtain a robust dataset, and apply 176 the sensor for varying conditions (turbidity and salinity), this experiment was repeated on five days (3,  177 10 Guadalete, were multiday monitoring is performed for varying tide 191

Field tests in the Rio de Guadalete 192
The objective of the third experimental campaign was to apply the sensor for monitoring macrolitter 193 in a natural river system. To test the sensor in a natural river system, the sensor was operated during 194 18 hours of monitoring in the Guadalete river in El Puerto de Santa Maria (36°35'58.6"N 6°13'17.5"W) 195 (Spain). The sensor was deployed from a pedestrian bridge (of 100 m wide) over the river. The river 196 monitoring took place on 8, 11, 17, 22, 23, 24, 26 and 28 October 2019 for varying tidal conditions. 197 Monitoring was done for one hour per testing day and tidal condition. Additionally, to investigate the 198 cross-sectional litter distribution, we monitored at three locations across the river width. The river flow 199 at the measurement location was bidirectional because of tidal influence in the Gulf of Cádiz (Atlantic 200 Ocean). We therefore investigated the difference in vertical and cross-sectional litter distribution for 201 ingoing and outgoing tide. The monitoring location and setup is shown in Figure 3 (2) and Figure 4  202 (2). 203 204 Plastic litter objects were identified based on the backscatter images obtained during the semi-205 controlled tests, using both the signal signature as the signal indicated strength (colour). Fish were 206 discarded from the sonar readings by their specific arc-shaped reflection. To correct for the shape of 207 the angled scanning beam (cone), the monitored items over the river depth were scaled to 1 m river 208 width. The depth was divided into four zones. For each zone, the total number of items per hour is 209 presented. Besides, a division is made between the two tidal flow conditions (incoming tide and 210 outgoing tide). 211 212 The sensor was deployed using the wide beam (47 degrees) which enables scanning with the highest 213 spatial resolution (largest scanning area) but the lowest scanning resolution (least detailed scanning). 214 These beam settings result in a blind zone of 80 cm depth for which objects cannot be detected by the 215 sensor. The significance of the results is determined using an independent t-test with 0.05 as 216 significance level. Based on these results, we identified some potential sources of uncertainty. We found several outliers 234 in the observations, that may be explained by the method for pulling the items through the water. These 235 outliers can be caused by pulling the objects with a rope instead of letting them naturally flow in the 236 water when passing the sensor. Pulling could induce water displacement in front of the objects and 237 possible disturbance in the sonar signal. Moreover, the filled balloons were not as spherical as 238 envisioned and deformed while pulling them through the water. This deformation (changing object 239 dimensions) could lead to a spread in the observed sonar reflections. 240 241 The results obtained from the bottle orientation test are displayed in Figure 5.

Semi-controlled tests in the Rio de San Pedro 259
From the semi-controlled experiments, in the Rio de San Pedro, we found that the average reflection 260 footprints of specific items varied substantially ( Figure 6). It seems the detected items can be 261 characterized by specific sonar reflections. When looking at the actual item size and the reflection 262 footprint, one would expect, according to the results in section 3.1 that a larger item results in a larger 263 sonar reflection footprint. This is however not the case for all items tested. 264 Figure 6: The average sonar reflection footprint (depth and width) of the different items tested. 266 Besides, a variation in the data is observed, Figure 7. The reflection depth, width and area data for the 267 different items are not consistent but spread. When comparing the reflectance depth, width and area of 268 the different items, Table 1, we see that at least one dimension is significantly different for 18 out of 269 the 28 combinations. This supports the potential for litter qualification using sonar. 270 271 Possible reasons for the inconsistency (spreading and no direct link with the actual item size) in the 277 data is the influence of the orientation and deformation of the objects. For example, a water bottle, as 278 shown in Figure 5, can result in a very different footprint when orientated differently. Moreover, items 279 such as plastic bags and packaging are likely to deform, which can lead to potentially very different 280 sonar reflections. This makes the identification of items according to their sonar footprint complex. 281 282 Besides the dimensions of the sonar signal reflection, the sonar signal intensities are also examined. 283 The metal can corresponds to the highest signal intensity and the food wrapper to the lowest signal 284 intensity. When comparing this to the material properties of the items it can be recognised that for some 285 objects the measurements fit the expectations (higher material density results in higher sonar signal 286 intensity). However, no direct link between the sonar signal intensity and the material properties of the 287 total of tested objects was observed in this study. The potential of classifying items based on their 288 material properties and sonar reflections seems although interesting to investigate further, using for 289 example Artificial Intelligence. 290

Field application in the Rio de Guadalete 291
Lastly, the sensor was applied during a multiday trial monitoring campaign in the Rio de Guadalete. 292 The number of monitored items per hour are shown in Figure 8. In total, the river was monitored for 293 18 hours over eight different days and varying location over the cross-section of the river. The results 294 showed that significantly more items are transported during river ebb tide (water flows from inland to 295 the sea), compared to the river flood tide (water flows from the sea inland). 296 297 298 Figure 8: Total monitored items during the field campaign using the wide scanning beam (47 degrees), 299 for the three different locations over the river's cross-section. Left: monitored items for river water 300 level going from low to high (water flows from the sea inland). Right: monitored items for river water 301 level going from high to low (water flows from inland towards sea). 302 On average, during ebb tide (high to low river tide), 38 items/hour were detected by the sensor. For 303 flood tide (low to high river tide), 19 items/hour were detected. Furthermore, we found a difference in 304 litter items over the river cross-section. It appears that more litter is transported at location 1 compared 305 to locations 2 and 3. In order to find an explanation, the river's cross-section was mapped using the 306 sensor, showing that the river bottom is not uniformly shaped over the width of the river. We observed 307 erosion on the outer bend, which coincides with the monitored litter transport peak. Generally, flow 308 velocities are higher in the outer bend and potentially more items could pass the sensor compared to 309 the inner bend. 310 311 Besides counting litter items, the depth at which the litter particles were present is indicated, leading 312 to the particle distribution as illustrated in Figure 9. For each zone, the total number of items per hour 313 is presented. No clear difference is observed for the two tidal flow conditions (IN-OUT). According to 314 the results presented in Figure 9, most litter items are present in Zone 1. An important remark is that 315 due to surface clutter a blind zone, for which the sensor is not able to detect objects, of 80 cm was 316 present at the water surface. In other words, items present in the top 80 cm of the water column are not 317 taken into account. Based on our findings, 50 percent of the monitored litter is present in deeper layers 318 (Zone 2, 3, and 4) of the water column. 319 320 Figure 9: Monitored litter (items/hour/m river width) distribution over the river depth (divided in 4 321 zones) for incoming and outgoing river tide. 322 Note that the counted litter items were identified as plastics according to the footprint data obtained 323 during the semi-controlled test. However, the dataset collected during the semi-controlled experiments 324 does not cover the total range of possible litter items. Therefore there is the possibility that other litter 325 items are wrongly identified as plastics, leading to a higher plastic load than actually present. To ensure 326 litter items are correctly identified as plastics, more research is needed to determine footprints of 327 different types of items such as other anthropogenic debris and organic litter. Fish resulted in a very 328 distinct signal reflection, illustrated in Figure 1, and are accordingly assumed to be filtered correctly 329 from the data. 330 331

Synthesis 332
Using echo sounding to detect plastic 333 Our findings show that echo sounding has potential for monitoring subsurface macroplastics. Plastic 334 items can be detected and possibly be classified based on their size and material properties. Being able 335 to monitor suspended plastics in rivers takes us a step closer to estimate global plastics transport rates. 336 337 The dimensions of objects in the sonar reflection imagery are related to the actual size of the passing 338 object (a larger item results in a larger reflection). However, sonar reflections are found to be sensitive 339 to object orientation and deformation. Another factor that influences the sonar reflection is flow 340 velocity. Items passing with high velocity are displayed significantly smaller than items passing with 341 low velocities. The flow velocity upper limit for the detection of objects using echo sounding was not 342 considered in this study (it was tested up to 0.25 m/s). Depending on the actual object size, flow velocity 343 could probably be a limiting factor for plastic detection using echo sounding. 344 345 For a widespread application of the echo sounding technique in riverine plastic monitoring some 346 challenges remain. More fundamental testing is needed to discard other litter types (vegetation etc.) 347 from the sensor readings, to be certain on monitoring only anthropogenic litter and plastics. 348 Furthermore, the classification of the different plastic litter objects would be beneficial for source 349 identification and targeted cleaning strategies. We did not find a direct link between object size, 350 material properties and reflected signal. However, our results showed that the potential is there. Very 351 specific and consistent testing of objects ranging in either size or material property could contribute to 352 more robust monitoring using echo sounding. 353 354 The Deeper CHIRP+ and potential of other sensors 355 For this research we used the Deeper CHIRP+ fish finder. We chose this sensor because of its 356 accessible price, size, and user-friendliness. For a proof-of-concept this sensor suited his purpose well. 357 The main disadvantage of this sensor is the limitation in raw data export. No raw sonar data could be 358 exported, therefore screenshots of the sonar signals were processed. In general, the accuracy of the 359 results could be affected due to sonar image processing, instead of using raw sonar data. 360 361 The sensor was deployed using its different scanning beam settings. For the different settings, blind 362 zones occur near the water surface at which no objects can be detected. For the narrow and wide 363 scanning beam, a blind zone of 15 cm and 80 cm, respectively, is present. During the executed tests, it 364 was assured that the items passed the sensor below the blind zone. However, for the monitoring activity 365 in the Guadalete river, it needs to be considered that the collected data does not include the full river 366 depth, due to the blind zone at the water surface. For most echo sounding devices, blind zones or 367 blanking distances are present. This leads to limited employability in shallow waters and the use for 368 near-surface objects. The impact of this is however limited since most research efforts and cleaning 369 strategies focus, due to sampling difficulties, on (near) surface plastics (approximately up to 1.5 m 370 depth), and therefore the potential of monitoring with echo sounding devices beyond this 1.5 m proves 371 its complementarity. 372 373 Different, more advanced sensors, such as an ADCP or Multibeam echo sounder could potentially lead 374 to more detailed sonar readings and allowing particle size/properties indication. ADCPs are designed 375 for velocity measurements but are currently applied for various purposes. The study of Sassi et al. 376 (2012) shows the applicability of ADCP for monitoring suspended particulate matter in rivers and 377 marine environments. Additionally, using horizontally mounted ADCPs at riverbanks, which enables 378 monitoring during high discharges (Hoitink et al., 2009), indicates also the potential for litter 379 monitoring in rivers. However, the costs of these devices are large (>20.000 Euro) compared to 380 conventional fish finders, which makes them less broadly applicable. 381 382 Monitoring in natural rivers 383 When applying the obtained knowledge from the controlled and semi-controlled tests to the field, the 384 following aspects should be considered when using echo sounding as a monitoring technique. As 385 previously stated, the actual litter size is hard to estimate from the sonar readings because of object 386 orientation, deformation, and flow velocity, implying an uncertainty when using the sensor for 387 monitoring purposes. In addition, obtained data on litter transport depends on the chosen beam width, 388 leading to the presence of a blind zone at the water surface. 389 390 From the monitoring data obtained in the Guadalete river, a distinct difference between fish and 391 anthropogenic litter could be observed. When comparing the sonar signal data to fish finding theory, 392 fish can be discarded from other objects by the specific shaped signal. However, this assumption is 393 only based on fish finding theories and has not been validated in practice. 394 395 Plastics in suspension 396 According to our results, 50 percent of the plastics are present below 1.6 m from the water surface 397 (measured from 0.8 m depth due to blind zone). This has a large impact on current monitoring projects, 398 which focus mostly on the plastics in the top layer. Taking into account the material properties of 399 (suspended) plastics, it is likely that litter items are present at different depths based on their density. 400 Moreover, turbulence, litter shape and vegetation may also influence the vertical location of the 401 particles. 402 403 The fact that, in the Guadalete river, 50 percent of the transported litter was present in deep layers of 404 the water column stresses the importance of monitoring subsurface plastics, as they likely account for 405 a large share of the total plastic transport. Recent work shows that underwater plastics make up the 406 largest portion of the plastic mass balance in the Atlantic Ocean (Pabortsava & Lampitt, 2020), this 407 might be the same in rivers. If we want to solve the plastic crisis, more effort is needed to develop 408 monitoring methodologies for underwater plastics. The river surface cannot be the carpet of the future 409 (everything beneath we don't see). 410 411

4
Conclusions 412 Echo sounding can be used for detecting suspended riverine macroplastics. Litter items can be counted, 413 while fish can be discarded from the specific signal reflections. Moreover, mean item reflection signals 414 yield unique combinations of width, depth and surface, which can potentially be used to identify 415 different litter types. Litter size was related to the sonar signature, although factors such as flow 416 velocity, object orientation and deformation need to be also considered when estimating size. This 417 remains challenging and further experiments are needed to collect more robust reflection statistics on 418 litter items. In the Guadalete river, significantly more suspended litter is transported when water flows 419 towards the sea compared to water flowing inland. Fifty percent of the counted litter items were present 420 in the deep layers (> 80 cm depth) of the water column. 421 422 Echo sounding is potentially useful to gain a better understanding of the suspended litter transport, 423 from which prevention and mitigation strategies could be optimised. For further research, it is 424 recommended to use an echo sounder for which the raw sonar data can be exported as a standard digital 425 file. Moreover, the set of test objects should be extended, including a wider range of sizes and shapes. 426 Objects of different size made of the same material and objects of the same size and different material 427 properties should be combined for testing. Side-scan or multibeam sonars might also lead to more 428 accurate characterization of litter sizes and materials. 429