Three-Dimensional Imaging Lidar for Characterizing Particle Fields and Organisms in the Mesopelagic Zone

The ocean’s mesopelagic zone is largely uncharacterized despite its vital role in sustaining ocean ecosystems. The composition, cycling, and fate of particle fields in the mesopelagic lacks an integrative multi-scale understanding of organism migration patterns, distribution, and diversity. This problem is addressed by combining complementary technologies with overlapping size spectra, including profiler mounted optical scattering sensors, profiler, and ship mounted acoustic devices, and a custom Unobtrusive Multi-Static Lidar Imager (UMSLI). This unique sensor suite can observe distributions of particles including organisms over a six order of magnitude dynamic size range, from microns to meters. Overlapping size ranges between different methods allows for cross-validation. This work focuses on the lidar imaging measurements and optical backscattering and attenuation, covering a combined particle size range of 0.1 mm to several cm. Particles at the small end of this range are sized using an existing backscattering time series inversion method after Briggs et al. (2013). Larger particles are resolved with UMSLI over an expanding volume using three-dimensional photo-realistic laser serial imaging. UMSLI’s image rectifying ability over time allows for derivation of particle concentration, size, and spatial distribution. Technical details on the development and post-processing methods for the novel UMSLI system are provided. Image resolved particle size distributions (PSDs) revealed a size shift from smaller to larger particles (>0.5 mm) as indicated by flatter slopes from dawn (slope = 2.6) to dusk (slope = 3.0). PSD trends are supported by an optical backscatter and transmissometer time series inversion analysis. Size shifts in the particle field are largely attributed to aggregation effects. Images support evidence of temporal variation between dusk and dawn stations through statistical analysis of particle concentrations for particle sizes 0.50–5.41 mm. Spatial analysis of the particle field revealed a dominantly uniform distributed marine snow background. The importance and potential of integrated approaches to studying particle and organism dynamics in ocean environments are discussed.

The ocean's mesopelagic zone is largely uncharacterized despite its vital role in sustaining ocean ecosystems. The composition, cycling, and fate of particle fields in the mesopelagic lacks an integrative multi-scale understanding of organism migration patterns, distribution, and diversity. This problem is addressed by combining complementary technologies with overlapping size spectra, including profiler mounted optical scattering sensors, profiler, and ship mounted acoustic devices, and a custom Unobtrusive Multi-Static Lidar Imager (UMSLI). This unique sensor suite can observe distributions of particles including organisms over a six order of magnitude dynamic size range, from microns to meters. Overlapping size ranges between different methods allows for cross-validation. This work focuses on the lidar imaging measurements and optical backscattering and attenuation, covering a combined particle size range of 0.1 mm to several cm. Particles at the small end of this range are sized using an existing backscattering time series inversion method after Briggs et al. (2013). Larger particles are resolved with UMSLI over an expanding volume using three-dimensional photo-realistic laser serial imaging. UMSLI's image rectifying ability over time allows for derivation of particle concentration, size, and spatial distribution. Technical details on the development and post-processing methods for the novel UMSLI system are provided. Image resolved particle size distributions (PSDs) revealed a size shift from smaller to larger particles (>0.5 mm) as indicated by flatter slopes from dawn (slope = 2.6) to dusk (slope = 3.0). PSD trends are supported by an optical backscatter and transmissometer time series inversion analysis. Size shifts in the particle field are largely attributed to aggregation effects. Images support evidence of temporal variation between dusk and dawn stations through statistical analysis of particle concentrations for particle sizes 0.50-5.41 mm. Spatial analysis of the particle field revealed a dominantly uniform distributed marine snow background. The importance and potential of integrated approaches to studying particle and organism dynamics in ocean environments are discussed.
Keywords: imaging lidar, mesopelagic, in-situ optics, image analysis, PSD, sound scattering layer INTRODUCTION The Earth's oceans are a reservoir for an estimated 31% of anthropogenically produced CO 2 . Of the 100 Gt of organic carbon taken up by ocean surface waters per year, up to 10% is transferred to the mesopelagic (similar magnitude as fossil fuel emissions) (Gardner et al., 2000;Giering et al., 2014Giering et al., , 2018. Without this carbon export mechanism, the concentration of CO 2 in the atmosphere would be an estimated 200 ppm higher than present day levels (McDonnell, 2011;Davison et al., 2013;Volk and Hoffert, 2013). This process, known as the "biological pump" is controlled by the productivity of primary producers in the surface ocean and active/passive downward transport of organic and inorganic composite material. Current estimates suggest, up to 90% of surface water detrital material is recycled, consumed, and respired before reaching 1,000 m McDonnell, 2011;Davison et al., 2013;Hansen and Visser, 2016). Understanding the role of the mesopelagic zone (bottom of euphotic zone to 1,000 m) in mediating remineralization and transport to the deep ocean is currently a subject of intense research (Giering et al., 2019;Briggs et al., 2020).
Much of our knowledge of particle dynamics and transport through the water column has been gathered using sediment traps (Martin et al., 1987;Buesseler et al., 2007). Sediment traps function much like rain gauges, accumulating sinking particles over a period of days to weeks at predetermined depths. This allows for the collection and biological analysis of a wide size range of particles (Chester, 2000;Stemmann et al., 2008;Hung et al., 2010;Giering et al., 2018). While sediment traps have been used extensively, their spatial coverage is limited, thus decreasing the probability of capturing larger particles. Small sampling areas on the order of m 2 must also be extrapolated temporally and across large regional areas and wide depth ranges. This may be particularly problematic for resolving significant episodic, localized sinking events associated with bloom crashes (Briggs et al., 2020). Sediment traps also hydrodynamically disturb suspended particles, inhibiting their efficiency and making interpretation more complex (Yu et al., 2001).
To compensate for an imbalance between carbon export and metabolic demands in the mesopelagic, many processes such as mortality, defecation, fragmentation, and export from diel vertical migrators require further investigation to accurately gauge these contributions in carbon budget models (Burd et al., 2010;Kelly et al., 2019;Briggs et al., 2020). Many studies have shifted the view of the mesopelagic to a more dynamic transit zone of the ocean that harbors organisms participating in massive diel vertical migrations (DVM) (Passow and Alldredge, 1995;Hiaka et al., 2001;Steinberg et al., 2002;Kaartvedt et al., 2012;Taucher et al., 2018;Proud et al., 2019;Boswell et al., 2020). Some of these animal's form what is called the deep scattering layer (DSL), named for its strong acoustic reflection. This layer has long been observed to migrate upward to the surface at dusk and back downward at dawn as motivated by predation and feeding (Hays, 2003;Hansen and Visser, 2016;Calleja et al., 2018;Taucher et al., 2018). Migration by zooplankton and nekton are increasingly recognized as playing a significant role in transport of carbon through the mesopelagic (Steinberg et al., 2002;Burd et al., 2010;Kaartvedt et al., 2012;Klevjer et al., 2012;Hansen and Visser, 2016;Calleja et al., 2018;Kelly et al., 2019). Mesopelagic micronekton and zooplankton directly affect the distribution and size of suspended particles through feeding, migration and defecation. Similarly, gelatinous grazers also directly affect the repackaging and vertical fluctuation of sinking particles, through ingestion and release of larger fecal pellets. These processes affect the rates of fragmentation and remineralization of sinking particles and are often excluded from carbon export budgets. Studies of DVM and other active transport mechanisms in the mesopelagic suggest that these missing export processes could explain apparent imbalances between carbon supply and metabolic demands (Steinberg et al., 2002;Burd et al., 2010;Kelly et al., 2019). New technologies which feature in situ measurements using optical instrumentation and autonomous platforms create new potential for characterizing particle fields for size, shape, distribution, and concentration (Nayak and Twardowski, 2020).
To investigate this potential and to better understand particle dynamics in the mesopelagic ocean, acoustic, and optical techniques are used to characterize distributions of particle fields (including organisms) and spatial patchiness. The focus of this work is on using optical backscattering and beam attenuation instrumentation in combination with UMSLI to resolve particle size distributions (including organisms) from about 0.1 to 50 mm. UMSLI uses three-dimensional laser serial scanning to provide traceable volumetric images of particle and organism fields down to about 1 mm. Smaller particles are resolved from optical backscattering and attenuation with a time series inversion method (Briggs et al., 2013). Details of UMSLI data processing are discussed and particle distribution results compared between the two methods.

Imaging Lidar
Lidar systems consist of two main components, a transmitter and receiver. The transmitter consists of a highly collimated laser, often directed by a mechanical mirror device. The receiver may take many forms, but generally consist of a light receptor that quantifies intensity per unit area. Just as a conventional camera measures the reflection of light off objects with dependency on source and receiver geometry, so does a lidar system. The basis of a lidar's function is to actively illuminate a volume of water with an intense pulse of optical energy from a laser transmitter and measure irradiance returned as a function of time with a receiver coupled with a high-speed digitizer. The range (R) of a target is described by McGill (2002): where c l is the speed of light. Time of arrival (TOA) is defined as the time it takes for a laser pulse to leave the transmitter, reflect from a specific volume of water, and return to the receiver. The sampling capability of the receiver defines the gate time ( t) and temporal resolution of the system. Lasers are capable of emitting light at very fast repetition rates with high energy levels and the collimated monochromatic nature of the beam makes it ideal for probing environments such as water. A complete physical characterization of laser transmission for lidar is well summarized in several studies (Mullen et al., 2009;Caimi and Dalgleish, 2013;Dalgleish F.R. et al., 2013;De Dominicis, 2013). The method of laser imaging for UMSLI uses serially scanned pulsed lasers for extending imaging range in turbid seawater Dalgleish F.R. et al., 2013). Traditional optical cameras can be effective when significant ambient or artificial light is present in low turbidity conditions. When using artificial light in imaging, marine life can show a behavior response, i.e., attraction or avoidance, to light within their visual light frequency range (Zorn et al., 2000). Artificial light in conventional imaging also introduces backscatter path radiance effects that can obscure objects. However, unlike active acoustic solutions, the primary advantage of using optical approaches is high resolution contrasted scene descriptions essential for object classification and detailed observations.
Red lasers can be configured below the wavelength range that is visible to marine wildlife allowing for unobtrusive observation. UMSLI uses a 638 nm red diode laser which is beyond the photopic response for most marine organisms in the mesopelagic (Zorn et al., 2000). The UMSLI system was originally built with the intention of observing large marine species without affecting behavior (Dalgleish et al., 2017). It features an alternating field-of-view (FOV) that outputs time-resolved images capable of extracting three-dimensional features. The quasi-monostatic approach and dynamically expanding scan field make UMSLI suitable for resolving organisms across discrete size classes due to an adaptive resolution feature based on the angular control of the laser scan field during operation. For this study, capabilities are extended to observe relatively small particles, such as copepods, krill, and flocs of marine snow.

UMSLI Components
UMSLI's design concept was focused on surveying a large volume of water for the use of observation and identification of targets on a wide range of size scales (mm to m). Transmitters serially illuminate a volume of water by scanning a grid of pulses in a bi-directional raster pattern using an analog micromirror device (AMD) and a scan angle expansion lens. The scan field of the receiver can be instantly configured to operate in three different modes, i.e., sparse, dense, or densest. Each different scan mode features a different image resolution and scan volume as controlled by the transmitter. The resolution is adjusted by concentrating a lower density pulse grid through a wider range of angles (sparse). Once an object is detected, a higher pulse density scan can be conducted through a narrower range of angles (dense/denser) using a digital Micro-Electro-Mechanical-Systems (MEMS) scanning device. The volume scan field for each channel is maximized by an overlapping FOV region between the quasi-monostatic transmitters and receivers. The system consists of three transmitter housings (each with a different total scan field) and three receiver housings, the details of which are described below.
The UMSLI transmitter (Figure 1A), features a 638 nm red laser diode with a laser driver that controls the laser pulse output energy. A separate transmitter controller uses RS-232 communication protocol with the MEMS driver that adjusts the polar and azimuthal angle of a scanning two axis (tiptilt) MEMS mirror (or "micromirror"). The MEMS mirror is mechanically responsible for the serial scan-like pattern of the system . The reflected beam from the MEMS mirror is expanded using a small angle expansion lens that outputs a full beam divergence of 2.80 mrad in water. The beam radial intensity follows an azimuthally symmetric Gaussian profile.
The total scan angle of the volume, post beam expansion, emits over a polar angle of 0.9425 radians. For this setup, the geometry of the fully illuminated plane is consistent for both polar and azimuthal angles. The transmitter has a 4 ns Full-Width-at-Half-Maximum (FWHM) pulse duration with an average power of 185 mW at a pulse rate of 80 kHz. The transmitter emits each pulse in a raster-like fashion, such that the transmitter emits 200 lines consisting of 200 pulses each. Individual lines take the transmitter 2.50 ms to complete, which allows the system to scan a full 200 × 200 plane in 500 ms for a total of 40,000 pulses (Supplementary Appendix A). The UMSLI system's adaptive viewing mode emits the same number of pulses at a smaller angle of 0.3142 radians (dense) and 0.1047 radians (densest).
The receivers consist of a pair of bi-directional red sensitive high-speed photomultiplier modules with focusing optics and a spectral bandpass filter. Receivers are designed to collect backscatter returns from the emitted laser pulses . To eliminate ambient light, the PMT is capped with a rectangular light shield that allows for a 30 • acceptance angle; Figure 1B demonstrates a ray trace of the light shield rejection pattern. A high-speed digitizer and digital signal processor are used to reconstruct the illuminated volume and output imagery waveforms. These waveforms are transformed and processed as a volumetric scan, within a x-y spatial plane and a temporal z-plane. A single element Photo-Multiplier Tube (PMT) is used as a detector. The PMT used for this system is a Hamamatsu R11265U series, type H11934-20 with a spectral response from 300 to 920 nm, a quantum efficiency of 19% and radiant efficiency of 78 mA/W. The temporal response of the PMT is the dominant restriction for the system response time. Figure 1C shows the rise time of the impulse response to be 1.3 ns and the fall time 5.8 ns. The PMT and its components described above are housed in a Vitrovex glass sphere from Nautilus which are paired as two combined hemispheres. The 187 mm outside diameter glass housing has a total glass thickness of 14 mm and is pressure rated to a depth of 12,000 m.

Field Experiment
During March 25, 2018, the NOAA Okeanos Explorer navigated the Gulf of Mexico to explore the mesopelagic region and monitor biological migration patterns. A single deployment was carried out at each station. The first deployment was in the morning from 0430 to 0900 (local) at 28.66N, 87.86W and the second deployment was in the evening at 1,600-2,100 (local) at 28.66N, 87.46W (Figure 2A). The two sites were 57.92 km  apart. For each station, an instrument cage was profiled from the surface to 300 m. Immediately after reaching 300 m sampling began and lasted 3 h. Periods for time series were aligned with dusk ascents and dawn descents of the DSL. The dusk migration was sampled at the first site, the dawn migration at the second. Deployed instrumentation included UMSLI, Seabird Scientific ac-9 and ECO-VSF, and a hull mounted and calibrated Simrad EK-80 multifrequency echosounder. Deployment protocols and post-processing for ac-9 and ECO-VSF followed Twardowski et al. (2018) and Sullivan et al. (2013), respectively. These sensors (excluding the EK-80) were mounted onto an aluminum frame with dimensions of 1.8 × 1.2 × 1.7 (m 3 ) as shown in Figure 2B. During deployment, the EK-80 measured profiles of acoustic water column backscatter at 18 and 38 kHz to monitor migration patterns from the surface to 600 m. The EK-80 was configured to transmit narrowband signals with pulse duration at 0.2 transmissions per second. Acoustic data were processed following procedures outlined in Boswell et al. (2020). UMSLI's sampling period was programmed before deployments; for the first station, it continuously sampled from 0515 to 0630 (local) and at the second station it sampled from 1,745 to 1,900 (local). UMSLI's scan modes were sequenced, following the scan order sparse, dense, densest (see Supplementary Appendix A).

Overview
Throughout this section, emphasis is on the dense and densest lidar scan modes, which have the highest resolution for particle sizes and the most overlap given their relatively close scan angles. The sparse scan mode, while still valuable, is better adapted toward identifying marine fish and larger organisms, which is beyond the size range focus of this analysis. Densest scans for the dawn station are unusable due to lack of quality.

EK-80
During deployments the EK-80 displayed the pattern of the sound scattering layers in real-time. A persistent DSL was observed between 300 and 400 m at both stations with migrating filaments linking the surface layers during the upward and downward migrations (Figure 3). Upward migration was observed as coherent vertical layers of backscatter ascending from 300 to 375 m to shallower surface layers < 100 m at dusk. The layers detected in the 38 kHz data followed a similar pattern and revealed persistent less dense backscatter return at deeper depths (∼600 m). At dawn, multiple downward migrating layers were detected beginning at 0600 (local) in both the 18 and 38 kHz data. In the 18 kHz data, the downward migration was comprised of two main layers, splitting and settling at 300 and 450 ms, respectively. Similarly, the downward migration recorded at 38 kHz displayed several coherent layers settling to form a high-density scattering layer between 350 and 450 ms.

Lidar Derived Particle Size Distributions (PSDs)
Particle size distributions (PSDs) include all particles, i.e., living organisms as well as detrital flocs and fecal pellets. Analyzing single image slices avoids alternating resolution capabilities (Supplementary Appendix A). PSDs are reported as equivalent spherical diameter D, derived from √ 4A/π where A is the extracted filled particle area (Supplementary Appendix B.7). Note that A is a measure of the image pixel count for a particle and the calculated diameter was scaled in accordance with the resolution of the image slice. All extracted particles were then subcategorized into size bins based on calculated diameter. The concentration levels of each size bin were normalized by dividing the total number of particles counted per size bin, by the sample volume, resolution of the size bin and the fractional volume lost due to masking (Supplementary Appendix B.6). PSDs were modeled as a power law also known as a Junge distribution: where D o is a reference diameter, n o is the differential particle concentration at D o (L −1 mm −1 ) and γ is the slope of the distribution (Kitchen et al., 1982;Liley, 1992;Boss et al., 2001Boss et al., , 2009Twardowski et al., 2001;Sullivan et al., 2005Sullivan et al., , 2013Buonassissi and Dierssen, 2010;Nayak et al., 2018). Due to the path attenuation of the transmitted and returned signals, an exponential loss was observed in extracted particles for slices with increasing distance from the sensor. As a result, only particle field data from the first slice, i.e., closest to the sensor, are presented. Figure 4 shows resulting PSDs averaged over the entire dusk and dawn time series. After applying a correction for blurring (see Supplementary Appendix A), reasonable agreement was observed between sparse, dense, and densest scan modes. Power law slopes in corrected data were approximately 3.0 and 2.6 for the dusk and dawn time series, respectively, slopes were flatter at dawn for all scan modes.
For the dense scans, comparison of averaged PSDs revealed elevated particle concentrations for the dusk time series in comparison to the dawn time series below 5 mm (Figure 5), while for particles > 5 mm higher concentrations were observed at dawn relative to dusk. Above 10 mm, the dawn station recorded much higher concentrations; these size bins most likely represent swimming organisms periodically entering the sample volume. A cumulative view of the particle concentrations for each size bin revealed a non-normal distribution, this led to outlier selection using nonparametric statistics, detailed in the next sub-section.

Outlier Scans With Enhanced Particle Concentrations
Outlier scans record anomalously high particle concentrations relative to the typical marine snow background and are labeled "particle enhancements." Outliers are identified as values greater than the upper quartile whisker of a box plot for a size bin. The upper whisker is calculated as Q 3 + 2.5 × (Q 3 -Q 1 ), where Q 1 is the lower quartile and Q 3 is the upper quartile (Figure 6). Particle enhancements were assessed for the first four size bins for the dense and densest scans; beyond the first four size bins particle abundance was too low to conduct statistical analysis. Images from the dusk and dawn stations revealed particle enhancements occurring for particle size bins 3.96 (Figures 6A,B) and 5.41 mm (Figures 6C,D), respectively. For the densest scans, particle enhancements were most observed for size bin 2.52 mm. This particle enhancement analysis can be used as a metric to identify images where organisms such as copepods may be entering image volumes. The increasing trend in particle size is consistent with trends found in Figure 4. The densest station had enhancement events relatively evenly distributed across all size classes. The most notable feature in many of the densest scans were long, ellipsoid shaped organisms with enhancements for size bins 0.5, 0.96, 1.43, and 2.5 mm (Figure 6F). Large organisms were imaged in the sparse, dense and densest scan modes for both dusk and dawn stations; these occurrences are too few to represent statistically but this is a notable result which indicates a population of larger organisms present among the marine snow background (Figure 7). Size and shape characteristics indicate these include siphonophores, various cnidarians, tunicates, and micronekton.

Lidar Derived Particle Patchiness
For every particle extracted from the volume, its Euclidian distance from all other particles was determined to assess relative clustering or dispersion of particles. A two-tail z-test was conducted for this distance metric (Malkiel et al., 2006). This form of hypothesis testing determines the likelihood that the sample distribution differs statistically from the population distribution mean (X p ) and standard deviation (σ p ). Statistical significance for this hypothesis test was determined by the z-score, defined as, where X s is the mean distance between particles in the scan slice and N is the combination of distances counted for each particle.  Z-score is an indicator of significance, where z < −3 corresponds to significant (p < 0.001) particle clustering and z > 3 to significant dispersion. The spatial distribution of particles in each scan was compared to a random uniform distribution of particles.
The random distribution was simulated with similar particle concentrations and the same geometric sample volume as all scans from the dusk and dawn timeseries. The simulated uniform distribution defined the population mean (X p ) and standard FIGURE 6 | Gray scale images containing anomalous particle enhancements taken in dense scan mode for dusk (A,B) and dawn (C,D) stations, similarly images taken in densest scan mode for dusk station (E,F). Box and whisker plots per size bin for dense and densest scan modes on right with percentage of particle enhancement events provided above box and whiskers (see text for details). deviation (σ p ) to allow for an unbiased description of particle proximity when calculating the z-score. Figure 8 provides a comparison of the particle spatial distribution of an image from the dawn station with a simulated uniform distribution. The z-score for the sample image ( Figure 8A) was calculated as −0.53, indicating agreement with uniformly distributed particles.
The rectangular (200 × 100) image skews the proximity of uniform particles as exemplified by the simulated data line in Figure 8B. Z-scores are calculated for every dense scan for each station to test for closer particle proximity (clustering) or more distant particle proximity (dispersion) than the average simulated distribution. Scans which feature mostly marine snows background are expected to demonstrate a particle field with a uniform distribution. The dusk station scans strongly favored clustering as indicated by the extended lower limit tail and the peak in z-scores falling below 0 ( Figure 8C). The cumulative z-scores of the dawn station were centered around 0 ( Figure 8D), indicating the majority of scans had particles uniformly distributed in the scan volume. Visual inspection of the images such as those shown in Figures 6, 7 confirms the statistical interpretation of dusk and dawn differences in PSDs and spatial patchiness. Generally, the dusk scans featured a higher concentrations of particles less than 5 mm. The increased population of small particles (<5 mm) created a higher likelihood of clustering caused by smaller particles (Figure 5). At both dusk and dawn, less than 4% of all scans featured images with particle enhancements and clustering. There were select cases of images with large organisms > 50 mm and clustering; examples of these included Figures 7A,D with z-scores of −3.65 and −12.5, respectively. Excluding scans with particle enhancements and large organisms, the spatial distribution of the particle field followed a uniform distribution. The scarcity of larger particles in the relatively small volumes sampled allows for interpretation of spiking in optical backscatter and transmissometer time series data to be classified as individual particles (Gardner et al., 2000;Briggs et al., 2013;Giering et al., 2019). Briggs et al. (2013) present an inversion method which uses the variance-to-mean ratio (VMR) of optical signal to estimate mean particle size. Briefly, Briggs et al. (2013) used the VMR of de-trended optical time series data to obtain a weighted mean signal per particle. This per particle optical proxy was then converted to a particle cross-sectional area using estimates of sample volume, scattering efficiency Q, particle residence time in the sample volume t res , and sample integration time t samp . This area was then converted to an equivalent circular diameter. Here, we follow the same method, but replace the VMR with the height of each optical parameter spike above a running median background of smaller particles, effectively deriving a diameter for each spike height, rather than a single mean diameter for the time series. The window for the running median was seven times either t res or t samp , whichever was greater. For attenuation, Q c = 2 and t samp = 0.003 s were used (Briggs et al., 2013). The sample volume was 12.5 ml, 2.5 times larger than Briggs et al. (2013) due to the 25 cm path length and t res was estimated following Briggs et al. (2013) to be 0.39 s (dusk) and 0.61 s (dawn). Logarithmically spaced bins from 0.11 to 11 mm were derived for c p and divided by bin width and the total volume sampled after correcting for water movement for each time series to yield c p -derived PSDs for each station (Briggs et al., 2013). The c p -derived PSD was calculated separately for each of the nine ac-9 wavelengths and all the resulting PSDs were averaged. As in Briggs et al. (2013), the same method was used to calculate PSDs from the b bp time series, substituting b bp for c p and Q bb for Q c . A Q bb of 0.04 was chosen following Briggs et al. (2013) based on the mean b bp /c p ratio of ∼1/50 observed during this study. The sample volume was estimated at 0.62 ml, t samp was 1 s, and t res was estimated continuously from the estimated path length of a particle through the ECO-VSF sample volume (∼1 cm) divided by the vertical velocity of the sensor package calculated from the pressure sensor over a 3 s interval. A b bp -derived PSD was estimated separately for each of the three angles of the ECO-VSF, and the three PSDs were averaged to create a final PSD. Figure 9 presents b bp measurements recorded during the time series for dusk and dawn, with the spikier dawn timeseries indicating enhanced concentrations of large particles. Figure 10 shows combined PSDs from ECO-VSF backscattering, ac-9 attenuation, and all available lidar scan modes. PSD slopes from c p and b bp were about 4.7 and 4.5 from the dusk and dawn time series, respectively, both significantly steeper than the slopes observed from lidar (also see Figure 5). PSDs were also higher for the lidar measurements for overlapping size bins.

DISCUSSION
The marine particle field and its variability in relation to the mesopelagic ecosystem is of growing interest in the scientific community, but still lacks understanding as it is deeply connected through many processes which span a wide size spectra (Stemmann et al., 2000b(Stemmann et al., , 2004Baird and Suthers, 2007;Guidi et al., 2008;Boss et al., 2009;Burd and Jackson, 2009;Stemmann and Boss, 2012;Burd, 2013;Giering et al., 2014;St. John et al., 2016;Calleja et al., 2018;Cavan et al., 2019).
Imaging lidar is presented as a new tool with a unique niche in capturing both non-living macroscopic aggregates (marine snow) and living macro-zooplankton, crustaceans, decapods, gelatinous taxon and other swimming fauna. Lidar allows in situ imaging of large delicate aggregates such as larvacean houses, mucous webs of pteropods, and other delicate organisms such as salps and siphonophores which are otherwise destroyed by nets, meshes and pumps. Figure 7 exemplifies the imaging ability of UMSLI as it serves as a tool for observing a large array of organisms but also shows the limits of identification due to resolution. Image processing expands UMSLI's abilities beyond large organisms and into particle sizes down to 0.5 mm, allowing connectivity between intra-class organisms and processes affecting mesopelagic marine snow.
Particle abundance and size are useful metrics for biogeochemical models which describe the function of the mesopelagic zone in the biological pump process. PSDs presented here are consistent with other methods which use imaging techniques to derive PSDs for living and non-living particles ranging up to several mm (Jackson et al., 1997;Checkley et al., 2008;Picheral et al., 2010;Slade et al., 2011;Stemmann and Boss, 2012;Taucher et al., 2018). The PSD slopes presented in Figure 5 were 3.0 for the dusk station and 2.6 for dawn, where a slope of 4 indicates equal particle volumes in equal logarithmic size intervals that are representative of oceanic environments without significant influence from active biogeochemical processes Sullivan et al., 2005;Stemmann et al., 2008). Flatter slopes in this study are likely due to aggregation of particles into larger flocs of marine snow which has occurred through downward flux from upper waters (Stemmann et al., 2004;Taucher et al., 2018). For large aggregates, studies suggest that aggregation through particle-particle interaction occurs through shear and differential sedimentation, which greatly depend on collision rates between particles, stickiness, and overall diffuse turbulence of the water column (Hunt, 1980;Jackson, 1995;Stemmann et al., 2004;Stemmann and Boss, 2012). The use of instrumentation which measure aggregate sequestration processes are vital to supplementing biogeochemical models. This study demonstrates the potential of using lidar imaging to record large aggregates within natural particle fields without causing particle breakup. UMSLI shows potential for making distinctions between living and non-living organisms based on target shape, size, and perspective information from imagery. UMSLI's high sampling rate captures identifiable targets across multiple images, allowing for multiple capture angles and positions. It is conceivable to use this information with machine learning algorithms for organism identification based on size and morphology if a representative library could be constructed for algorithm training. This is a current area of interest in our lab.
While our dataset includes only two stations, preventing statistical inferences, results from this study reveal differences between stations that were plausibly associated with temporal dynamics of mesopelagic activity. Further such deployments to verify patterns could therefore enable important insight. Night time activity in the mesopelagic has been reported to shift particles into larger size classes (Stemmann et al., 2000a,b). Stemmann et al. (2000a); reported size shifts of increasing particle abundance at night for particles > 0.5 mm and showed that this pattern was reversed during the day. In the northeast Atlantic, observations reported in Lampitt et al. (1993) using time-lapse camera recordings showed evidence for increasing concentrations of particulate matter at night, linked specifically to the abundance in large particles. These observations are consistent with results from the dawn station, which recorded larger concentrations of particles greater than 5 mm in comparison to the dusk station ( Figure 5). Stemmann et al. (2000a) suggest that, "diel vertical migration is the best candidate to explain the diel transformation of PM (particulate matter)." As DVMs hold a diverse array of organisms with dynamic temporal migratory patterns, it is difficult to attribute particle field changes to a single cause. There has been evidence linking zooplankton including copepods with particulate matter fluxes and migration activity (Stemmann and Boss, 2012). However, it is difficult to discriminate particle flux contribution between migratory consumers and residential nightly feeder (Stemmann et al., 2000a(Stemmann et al., ,b, 2004Guidi et al., 2008;Giering et al., 2014;Archibald et al., 2019).
Resident and migratory mesopelagic fish are responsible for up to 15% of the total export of carbon through the mesopelagic zone, and the passive and active transport of carbon is largely attributed to swimming organisms and their fecal matter production (Davison et al., 2013;Cavan et al., 2019). It is expected that if biological activity such as feeding was apparent then resulting PSDs would demonstrate a bump at specific size classes. Evidence of this type of activity was investigated through the evaluation of outlier scans and patchiness. Overall, the results showed limited cases of scans with particle concentration enhancements for specific size classes (Figure 6). While PSD slopes for all scan modes showed no major inconsistencies which would indicate biological activity, it is obvious that the dawn station featured a higher concentration of particles > 20 mm (Figure 5). This is supported by images in Figure 7 which feature evidence of nightly feeders or migratory species. During the time periods of measurement, most observations showed uniformly distributed particle fields largely composed of marine snow (Figure 8). As mentioned, clustering in the dawn time period resulted from increased particle concentrations, likely associated with the nighttime migration. Clustering may also be caused by groups of swimming organisms. Observations of both particle enhancements and spatial clustering provided an approach to identify swimming organisms moving through the scan field in smaller size bins. The dense scan data for dawn and dusk recorded limited association of particle enhancements with clustering (<4%), this was only for smaller size bins dominated by marine snow. For the dense scans, particle enhancements were most common for the 3.96 and 5.41 mm size bins. This may be evidence of macro-zooplankton such as copepods, or small fish larvae swimming through the scan volume. The combination of images with PSD calculation allows for observation across broad size scales further connecting detritus remineralization, fauna abundance, nightly migrators, and full-time residents of the mesopelagic. Further expansion with complementary instruments could supplement the goal of observing mesopelagic dynamics in greater detail. Despite significant differences in measurement approaches for lidar, b bp , and c p , they all showed similar trends in particle abundance, increasing for the dawn timeseries. Increases in backscatter spiking for the dawn station associated with large particles was consistent with PSD results from the lidar imager. For overlapping size bins, b bp and c p approaches result in lower PSD abundances, and overall steeper PSD slopes. Moreover, the steeper slopes, much steeper than 4, may not be realistic (Sullivan et al., 2005). Differences in derived PSDs may have resulted from differences in sample volumes and the way the b bp and c p inversions were formulated. While the lidar scans a large undisturbed sample volume, the ECO-VSF has a sample volume that is < 1 mL and only ∼1 cm away from the sensor face. The c p measurement has an active pump rapidly flowing samples through a cuvette. There is thus the potential for turbulent disruption of relatively large, fragile aggregates for b bp and c p measurements, shifting particle abundance from larger to smaller particles (Slade et al., 2011). The greater turbulence associated with the c p measurement may explain its lower PSD relative to the PSD obtained from b bp inversion. Additionally, the larger the particle the more rare its occurrence, which may lead to an increasing level of under sampling with increasing particles size for the relatively small sample volumes associated with b bp and c p measurements. Indeed, this likely is the cause of apparent rollovers in the log-log representation of size distributions derived from b bp and c p that were clearly observed for the dawn station.
The inversion of b bp and c p measurements is also sensitive to the assumed efficiencies Q c and Q bb . While estimations used here are expected to approximate the bulk optical signal, Q bb and Q c differ with particle size and shape (Briggs et al., 2011(Briggs et al., , 2013. Briggs et al. (2013) showed an apparent decreasing bulk Q c for clay aggregates measured using an ac-9 device leading to an estimated mean size decrease in the inversion. Despite these differences, results from all measurements showed similar trends and magnitudes were comparable, particularly for lidar and b bp inversion results, where PSDs could be nearly seamlessly linked (Figure 10). Moreover, the comparison with c p and b bp spike method provided a valuable check. Relative agreement between independent approaches demonstrates the potential for extending the range over which PSDs can be resolved using multiple techniques, a key objective of this study.
In order to progress in the study of how organisms play a role in the transport of carbon through the mesopelagic zone, new technologies must be developed to both monitor particle fields while observing organisms without effecting behavior. UMSLI, as an instrument originally designed for imaging large marine fish and mammals, has shown extended capabilities for characterization of relatively small particles including marine snow in this field experiment. UMSLI's unobtrusive red laser creates opportunity for observing larger swimming organisms that otherwise practice avoidance behaviors from tow nets and ROVs (Peña, 2019; Urmy et al., 2019;Robison et al., 2020). With demonstrated capabilities of observing nekton in situ simultaneously with particulate organic matter fields, this technology shows potential for studying interactions of aggregates with organisms and resulting transformations through remote, undisturbed observation. The potential of UMSLI can also be further enhanced. The possibility of calibrating lidar returns in absolute units of backscattering would enable another approach for quantifying particle abundance while providing a measure of backscattering for a particle population that is not included in current measurements made by the oceanographic community with very small sample volumes. The contribution of this large particle size class on ocean color remote sensing in surface waters is unknown. Coupling optical measurements with multi-frequency acoustics may also allow seamless remote monitoring of size distributions over the range of microns to meters (Gruber et al., 2016). This is the focus of future work.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

AUTHOR CONTRIBUTIONS
TM was responsible for the processing and analysis of data and manuscript preparations. MT conceived of the study, participated in the analysis of methods and results, and provided guidance in presentation and preparation of manuscript. NB assisted with the PSD inversion method from c p and b bp data with associated results and assisted in aspects of manuscript development. AN assisted with image processing methods and manuscript development. KB processed and analyzed the acoustic data presented. FD was the principal investigator for the design, fabrication, laboratory, and field testing of the UMSLI system, and provided guidance on interpretation of raw lidar waveform arrays, filtering, and noise reduction, processing protocols, and analysis. All authors contributed to the article and approved the submitted version.