DEVELOPMENT OF A PUMP-DRIVEN VERTICAL PROFILER FOR AN AUTONOMOUS SURFACE VEHICLE

A modified SCOUT autonomous kayak, owned by the Roman Lab, performs geo-referenced depth profiles at pre-defined spatial locations. The vehicle uses a single azimuth thruster to approach a goal location, and holds station while a depth profile occurs. Using an autonomous surface vehicle (ASV) to deploy a vertical profiler enables precise but adaptive sampling in typically under-sampled environments. The profiler lowers tubing into the water column with a winch, and discrete fluid samples are taken using a peristaltic pump. A conductivity, temperature, and depth (CTD) probe relates the sample depth to sensor measurements by matching their timestamps. Water quality optical sensors, housed in the vehicle hull, are plumbed in-line with the pump for continuous measurement of pumped water properties. The profiler is designed, constructed, and tested to identify relevant system characteristics, such as flush time. The system flush time is predicted based on the total profiler volume, and maximum pump flow rate. During flushing, a step response is induced in sensor measurements due to the transition between two discrete samples. Step response testing validates the predicted flush time such that a complete system flush occurs during pump operation. Closed loop tests indicate that the optical sensors are robust to large changes in the fluid sample temperature. The completed profiler is integrated with the SCOUT ASV, and field tested at Upper Pettaquamscutt Basin, North Kingston, Rhode Island. An initial field trial occurred with concurrent manual vertical profiling done by Dr. Veronica Berounsky. Automated dissolved oxygen measurements are found to correlate with the manual vertical profile. The vehicle’s stationkeeping performance is found to be satisfactory by staying within 10 meters of the target point. Subsequent field trials confirmed that automated profiling is consistent,

Final system characteristics to automate the profiling method described in Figure

Introduction
This study presents the development, integration, and testing of a pump driven vertical profiler integrated with an autonomous surface vehicle (ASV). Environmental monitoring of lakes, coastal ponds, and estuaries often calls for vertical profiling at specific locations to monitor temporal trends in water properties. Typical long term monitoring stations include moored platforms and automated winch systems with optical sensors that measure physical and chemical water properties [1]. In recent years, many studies used these platforms for monitoring the effects of short and long term ecosystem trends [2] [3]. Recent studies using remote sensing of surface water properties identified spatial vaiability that may be overlooked relative to single point measurements [4]. A system capable of autonomous depth profiling enables evaluation of spatial variability in water properties.

Motivation
Current water quality monitoring techniques establish a record of ecosystem health through periodic grab sampling at static monitoring locations. Conservation groups, such as Watershed Watch, coordinate weekly sampling efforts throughout Rhode Island. Water samples, secchi depth, and temperature readings are taken as part of bi-weekly sampling in lake basins greater than 15 feet deep [5]. The water properties act as trophic state indicators that are linked to the water body health [3,6]. These measurements form seasonal records for rivers and lakes throughout Rhode Island, and are complemented with periodic vertical profiles.
In lake and reservoir basins, periodic depth profiles can be used to monitor seasonal changes in lake strata. The Rhode Island Department of Environmental Management (RI DEM) defines a "standardized method for performing qualitative field measurements of a water column profile in lakes, ponds, and reservoirs, using a multi-probe meter." [5]. A single profile is taken at the deepest region of a lake, and typically requires several hours to complete. In a study performed on a meso-eutrophic lake, Kallio et al. indicate that sparse surface measurements can mis-represent basin scale chlorophyll trends [7]. Although moored vertical profilers reduce the expense of short term monitoring, multiple moorings are required to achieve any spatial resolution [3].
Expanding water quality monitoring coverage is typically handled using a small boat, or autonomous surface vehicle (ASV) to deploy a sensing payload. ASV systems generally include rapid measurement systems, like those deployed from small boats by Antilla in 2008, and Crawford in 2015, with flow cells to measure surface properties on a lake [8,9]. Coupling optical sensor measurements with the vehicle position enables mapping of spatial water quality trends. Other systems, such as the one deployed by Birgand in 2016, increased the spatial coverage of a single sensor using a multi-port valve and peristaltic pump [10]. A key component of all three studies was a flow cell that established a closed volume between a sampled environment and optical sensor. Using a flow cell system also reduces the likelihood of sensor damage they aren't exposed in-situ. Similar systems used an ASV to deploy a water quality monitoring payload to great effect.
In recent years, ASVs have proven to be suitable platforms for deploying monitoring payloads. The Lake Wivenhoe ASV successfully deployed a profiling arm capable of reaching "up to 20m when the vehicle is stationary" [11]. Additional work examined diurnal variation in methane effusion at the Little Nerang Dam reservoir using the Wivenhoe and an Inference ASV [12]. The Woods Hole Oceanographic Institute (WHOI) Jetyak deployed a CTD sensor using a small winch and A-frame similar to traditional research vessels [13]. In 2015 the Hy-dronet ASV was deployed with the sensors necessary to optically and chemically track hydrocarbon trails. Using a custom built CTD rosette it can extract up to 7 samples from the water column during a profile [14]. Water sampling payloads in these vehicles were custom built, and required experimental validation to establish their operating parameters.

Purpose of Study
This document presents the design for a pump driven vertical profiler that will be used to examine spatial variability of water properties. The primary criteria for the profiler is that sensors are housed in the hull of the vehicle, rather than deployed in-situ. Fluid will be extracted from depth using a peristaltic pump, and pumped to a series of optical sensors in a closed system created by flow cells and Tygon rubber tubing. The following list describes the major design goals: 1. A closed system will be established between a point at depth, and the sensors in the vehicle.
2. The effects of system operation on a fluid sample will be characterized to determine if it is representative of the in-situ water properties.
3. The vertical profiler will be integrated with an existing SCOUT ASV, and field tested to evaluate its performance.
4. Measurements will be compared with other routine profiles.
Flow cells create an independent, closed volume around each sensor. When the system is flushed the sensors take real time measurements of the water passing through each cell. A conceptual overview of the final system design is presented in Figure 1. will discuss the vertical profiler design, the proposed method of operation, the optical sensors used, and the basic system characteristics. Chapter 3 discusses the measurement response testing that validates the system's performance, and its integration with the ASV. Finally, testing results for the profiler and ASV are presented in Chapter 4.
[2] P. Bierman, M. Lewis, B. Ostendorf, and J. Tanner, "A review of methods for analysing spatial and temporal patterns in coastal water quality," Ecological Indicators, vol. 11, no.   During profiling operations each sensor logs continuously so that the complete system response is measured. A prescribed flush time, described in Section 2.3, is used to time the pump operation such that the sensor readings relate to the fluid sampled at depth. Figure 3 summarizes the operation for a profile to several prescribed depths. An estimate of winch payout, described in Section 2.2, is used to control the winch and deploy the sampling tube to the desired depth.

Optical sensor integration
Optical sensors are ubiquitous tools for estimating in-situ water properties through various physical proxies. Utilizing a photo diode and detector these sensors emit light into a fluid volume and measure the response. The operating principles for each sensor are unique and affect their sensitivity to a closed volume. Each sensor is housed in an independent flow cell designed to suit the unique fluid volumes required. The sensors deployed for this study are seen in Figure 4. The following subsections detail the operating principles of each optical sensor and discuss how their measurements are related to the water properties of interest.

Ecopuck Fluorometer
The fluorometer is a tri-parameter sensor that measures chlorophyll, turbidity, and fluorescent dissolved organic matter (FDOM). It excites the water volume in front of the sensor head with 460 and 690 nm light and measures the return at 695 nm. Chlorophyll fluoresces when excited by blue light and a strong correlation exists between the emission and real chlorophyll concentrations [1]. Sensor measurements are made in raw sensor "counts" that are mapped to their relevant properties through a linear fit. The following equation describes the calculation applied to all measurements made by the fluorometer.
Where the coefficients SF and darkcounts are unique for each property and are specified by a manufacturer calibration. Validation of the sensor's response is performed through a combination of dark count tests and testing with turbidity standard. Dark counts are determined by placing tape over the sensor head so that no light is able to reach the detector. Comparison of the reported dark counts with the calibration dark counts indicates that no linear offset exists at the zero point of the sensor.

Aanderaa Oxygen Optode
An Aanderaa Oxygen Optode 4835 is deployed to measure dissolved oxygen and air saturation. Measurements are conducted using a process called Dynamic Luminescent Quenching. The sensor excites an oxygen permeable foil with blue light which is quenched by the release of oxygen from the foil. Figure 5 represents the physical relationship between the light emission and oxygen concentration in its sensing foil. Using the Stern-Volmer equation the oxygen optode relates the delay in the luminescent response to an oxygen concetration [2]. Measurements are sensitive to changes in salinity and temperature, but for this study only temperature compensation is applied to measurements taken. A multi-point factory calibration provides the necessary coefficients for relating the sensor readings to oxygen concentration.
Basic functionality tests of the optode can be made by placing the sensor in an open container and blowing bubbles into the water. The sensor responds by showing an increase in the O2 concentration to the point of air saturation, verifying its operation.

s::can spectrolyser spectrometer
An s::can spectrolyser is a relatively new sensor typically used in measuring the water properties of effluent discharge from wastewater facilities. It estimates nitrate, turbidity, fluorescent dissolved organic matter (FDOM), and total organic carbon (TOC) quantities in the water volume seen in Figure 6 using a 15mm sensing channel path. The sensor is calibrated using a partial least squares (PLS) method with training data gathered from five wastewater treatment facilities. By incorporating the entire spectral response in the calibration Langergraber et al. identify the capacity for in-situ nitrate detection in concentrations of 0-5 mg/L [3]. A second, single point calibration is applied to the sensor prior to field testing using a distilled water bath. Two constraints reduced the effectiveness of this sensor during deployments.
The measurement requires one minute to process data, and typical aquatic nitrate concentrations in coastal ponds are significantly lower than the ones identified in the training data. As a result this sensor was deployed in preliminary field trials but removed from the system during the remaining tests. As a result the final Ecopuck housing, seen in Figure 7, had a greater head space volume than the oxygen optode and s::can.

Winch
A small level-wind winch is outfitted with a high torque gearbox and Animatics Smartmotor, model SM23165MT, to control the profiler depth. Typical motor operation draws 30W with transients up to 50W depending on winch tension. The winch is spooled with 50 feet of Tygon tubing and a YSI Castaway CTD is attached at the deployed end with a cable grip. The drum is geared to the level wind to manage the tubing during winding. Plumbing within the winch drum connects the spooled tubing to the pump using a fluid slip ring. An image of the completed assembly with tubing un-spooled from the drum is seen below: The motor uses an internal encoder to measure the shaft position during operation. A relationship between commanded motor shaft positions and an actual payout length is described by the winch and gearbox characteristics as payout = encoder counts * drum circumf erence motor counts per revolution * gearbox reduction ratio .
The drum and wire guide were geared together such that the tubing wind was less consistent than Equation 2 implies. Field trials were performed to evaluate the tubing wind consistency and refine the conversion between encoder counts and payout length. At Upper Pettaquamscutt Basin the winch was driven to a depth target of 8 meters with five unwind-wind cycles to generate a calibration between encoder count and CTD depth. During field trials the original coefficient is found to underestimate the quantity of deployed tubing due to the constant drum circumference assumed. Figure 9 displays this under-estimation as well as the linearity observed during multiple payout cycles. Figure 9: A graph comparing the original encoder-payout estimate to the one identified during field trials. No hysteresis is observed and the payout response is linear.
The final coefficient is used for driving the winch to a set depth, and the average speed of the winch is 10 cm/s. Further linear offsets were also used to zero the payout amount once the profiler is mounted in the ASV. This accounts for the distance between the winch and water's surface.

YSI Castaway CTD
Measurement of in-situ conductivity, temperature, and depth is handled by a YSI Castaway CTD. The device records measurements to internal memory and is timestamped relative to a GPS-synced clock. During a survey the Castaway is attached to the tubing using a cable grip and configured to log continuously.
Survey data are retrieved from the device after a deployment and matched to logs collected by the vehicle based on matching timestamps. CTD data provide a reference for the tubing depth and in-situ physical properties of the sampled fluid for validation purposes.

Peristaltic Pump
A Cole Parmer Masterflex peristaltic pump is used to draw water from depth up the sampler tube. Peristaltic pumps are typically used when a closed sample path is necessary to measure fluid properties. These pumps minimize pump-fluid interactions by squeezing tubing between several rollers to generate pulsed fluid flow and do not entrain air. Two diameter pump heads were selected for testing.
The LS-24 and LS-36 pump heads accept 1/4" and 5/16" diameter tubing respectively. A Pololu 24V motor controller drives the pump, drawing up to 60W at full speed.
The flow rate was determined using repeated 30 second trials of pump operation. Each pump head was loaded with a one meter length of tubing, and operated for ten minutes to ensure that the tubing was broken in. Next, the pump was run for four 30 second intervals. For each test a kitchen scale was used to measure the pumped fluid mass and establish baseline variability in the flow rate. As expected the larger pump head displaces 35% more fluid compared to the smaller one with a subsequently higher variance across all trials. The observed deviation is around 2.5% of the total volume pumped. Adding additional tubing induces pressure losses dependent on the length and tube fittings. Flow restrictions were characterized for individual plumbing components to identify major sources of head loss and minimize them if possible.
The system plumbing was connected to the pump relative to the numbering in Figure 10. Table 2 describes the system configurations tested with the LS-36 pump head. Figure 10: A connection diagram with system configurations marked relative to Table 2.  Table 2: A list of system configurations evaluated to identify the sources of flow rate loss in the profiler.
Flow rate tests with these setups follow a similar procedure to the one described for identifying the nominal flow rate. A given system configuration from Table 2 is connected to the pump and filled with water. Pump operation occurs for 30 seconds and again discharged fluid is collected and weighed. Three tests were conducted for each system configuration and their average flow rate was compared to a nominal flow rate described in Table 1. A clear advantage is seen between 1/4" and 5/16" tubing. The final plumbing configuration was selected using 5/16" tubing for all components except the flow cells due to their construction. The final flow rate was used to predict a minimum flush time for the profiler. Table 3 identifies the volume associated with the individual parts of the flow system.

Final Profiler Characteristics
The final profiler characteristics are listed in Table 4. A consequence of the  Table 4: Final system characteristics to automate the profiling method described in Figure 3 The time and energy required for a profile are described by the following equations.
Where N is the number of samples and D max is the maximum depth. Pump operation is the limiting factor during a profile. For example, taking ten measure-

Profiler Evaluation and Integration
Sensor measurements will exhibit a delayed step response when a prior sample is flushed from the system by a subsequent one. The completed system, seen in Figure 11, underwent an evaluation to verify the predicted flushing characteristics defined in Chapter 2.   Sensors were connected to the profiler and flushed, alternating between the standard of interest and distilled water. The predicted flush and measurement periods were used to represent the system operation in the field. Waste water from the fluorometer trials was collected and re-run through the system to test the sensor response at varying turbidity concentrations. The following subsections present the results from these trials and additional sensor-specific tests to further characterize the system operation. Figure 12 shows that a complete measurement response is observed during the predicted 45 second flush time.

Ecopuck turbidity response
(a) Ecopuck high-low concentration step response (b) Ecopuck low-high concentration step response. An additional experiment passed a small "spike" of concentrated turbidity standard through the system between distilled water flushes. The observed tur-bidity measurements seen in Figure 13 verify the response time identified previously and the extent of mixing that occurs. These tests imply that only measurements taken after pumping for 90 seconds are representative of the fluid sampled at depth. Relative Turbidity Response

Ecopuck
Step Response to a Turbidity Spike  Figure 13: Ecopuck measurement response to a 100 mL spike of turbidity standard introduced to the system between distilled water flushes. An asymmetric response is observed because mixing occurs in the chamber during flushing.

Optode oxygen response
Step response tests using the oxygen optode were conducted with anoxic standard and distilled water. In Figure 14 the complete response of the sensor and chamber occurs within the predicted flushing interval. This is consistent with the sensor specifications, of a 63% response after 25 seconds [1]. The ramp up observed in Figure 14a was caused by air introduced at the tubing when water volumes were switched.
(a) Oxygen optode step response, high concentration to low concentration.
(b) Oxygen optode step response, low concentration to high concentration. Figure 14: The oxygen optode step responses indicate a delayed response when pumping anoxic fluid. A complete sensor response is observed within the predicted flush time.
During pump operation the fluid temperature changes as it travels through the system from depth and to the sensor in the kayak, which may induce measurement error. This effect was characterized with a closed loop "radiator" test. A length of tubing was filled and connected to the pump and flow cells to create the closed loop. The tubing was submerged in a bucket of warm water to heat the fluid inside. Water was flushed past the optode until an increase in temperature was measured. Pumping was stopped and the tubing loop was placed in a bucket of ice water. Once chilled pumping occurred again until a large temperature change was observed by the optode thermistor. In this experiment a temperature change of 7 degrees C was forced to represent a more extreme range than expected in the field.
During field operation temperature changes are typically small, between 0.5-1.5 deg C, and the residence time of a sample is 2 minutes. Figure 15: Closed loop temperature response of the oxygen optode. The temperature deviation during the closed loop trial was 7 degrees C. Variability in the subsequent measurements are well below the 5% accuracy of the optode thermistor. Therefore temperature changes to the fluid do not affect the sensor measurements.
A negative trend occurs because oxygen solubility is dependent on fluid temperature. Over a ten minute period, the temperature measurement never equalized, indicating that small systemic bias exists due to pumping water into a warmer environment. Although the change was negligible in lab conditions large temperature differences between the surface and depth will affect the oxygen measurements slightly.

Vehicle overview and profiler integration
The vertical profiler was integrated into a SCOUT autonomous kayak (in Figure 16). The SCOUT autonomous kayak uses a single azimuth thruster to drive survey patterns defined in a prescribed mission plan. The water quality sensors and pump were housed within the hull, while the winch is mounted on top. Battery power for the vehicle was provided at 12 and 24V. An overall schematic for the vehicle is seen in Figure 17. All sensor communications on-board are handled digitally via RS232 or RS485 to USB converters. Software drivers read and publish data from each message pass over UDP using Lightweight Communications and Marshalling (LCM) [2]. These data are logged continuously during a survey. In post-processing message timestamps for profiler state and Castaway CTD are matched to generate a representative depth profile.
Three primary software daemons are seen in Figure 18.   Figure 19: Example mission file sent to the vehicle prior to a survey. Global parameters define critical timeouts and the local reference frame. Positions during a survey are then referenced to that origin. An array of depths, a flush time, and measurement time specify how an automated profile will be performed at the target location.
Integration of the profiler required a series of upgrades to the platform, according to the following list.
1. Four lithium iron phosphate batteries increased the vehicle power capacity available for payload operation.
2. An upgraded cockpit cover provided hard points for securing the winch on top of the kayak. 3. A sheave mount was constructed to pass the tubing to a through hull fitting.
4. An RTK GPS reduced vehicle position uncertainty from roughly 10 m to ≤10 cm.

5.
A powered USB hub supported communications between the profiler components and main vehicle computer. 6. A limit switch mounted in the sheave established a set point for resetting the winch payout after a profile.
Field tests evaluated the overall system performance, and provided a validation case for the representativeness of measurements taken. Background information on the site is presented in Section 4.1, and field trial results are discussed in Section 4.2.

Site Overview
A study published by RI DEM describes the Narrow River Estuary as a composite of a tidal inlet and back bay, an estuary, and two fjord-like ponds [1]. Upper Pettaquamscutt Basin, seen in Figure 20, stays heavily stratified year round due to tidal and riverine forcing. It is historically well monitored with studies identifying anoxic conditions at the bottom, and unique concentrations of biogenic hydrogen sulfide [2]. Subsequent profiling identified fine scale plankton layers that are on the order of centimeters thick [3]. Each deployment followed a prescribed mission similar to the outline from Figure 19. All deployments were referenced to the same local origin for consistency. In December, four depths (2, 4, 6, and 8 meters) were selected for profiling.
Two profiles were conducted at the same spatial location as a control, and two additional profiles were executed 50 and 100 meters west of the starting location.
Dr. Berounsky performed a manual depth profile, using a YSI multiparameter sonde, 50 meters west of the survey transect. A YSI sonde measures dissolved oxygen and chlorophyll fluorescence, and is specified by RI DEM as the required sensor for vertical profiling [4] [5].
The vehicle depth profiles were extracted in post processing by referencing the Castaway CTD data to mission logs relative to their timestamp. Each sample taken during a profile was assigned a unique identifier based on the profiling daemon assignments for sample, and depth index. Sensor readings taken during T measure were averaged, and referenced to their extraction depth. A 45 second offset was applied to the Castaway measurement of tubing depth, reflecting the system flush time. As mentioned earlier in Subsection 2.1.2 dissolved oxygen concentration data was not salinity corrected. A factory calibration was applied to raw fluorometer data to estimate chlorophyll density.

Field trial results
The dissolved oxygen profiles, seen in Figure 21a, indicate correlation between in-situ conditions, and sensor measurements taken by the profiler. Variation in the tubing depth during sampling was negligible relative to the CTD pressure measurement accuracy. The largest variance in dissolved oxygen measurements was observed directly in the oxycline. Vehicle motion during each profile was within a 10 meter radius circle of the goal, seen in Figure 21b.

Future work and conclusion
For this study, a pump-driven water sampler was integrated with an ASV, and field tested. Flow cells encapsulated optical sensors which were plumbed in-line with a peristaltic pump. A system flush time was predicted by characterizing the pump flow rate, and evaluating the sensor measurement response confirmed that a full flush occurred after the predicted interval. The profiler was integrated with an existing ASV, and is capable of performing between 80 and 100 discrete measurements during a profile. Finally, field trials at Upper Pettaquamscutt Basin validated that measurements taken by the profiler are representative of the in-situ fluid properties.
Field trials validated the predicted system performance, and further work will explore the impact of profiling on local fluid interfaces. Tubing movement through the water column causes stirring, and can potentially disturb fine chlorophyll layers.
Additionally, the physical processes driving local fluid properties are time varying, and may change if a region is profiled again.
Chlorophyll measurements made by the fluorometer are not physically grounded, and require a correction based on a physical determination of chlorophyll concentration. The correlation between in-situ and sampler conditions indicate that sampled fluid may be stored for future processing. Due to space constraints within the ASV sampling and sensor measurements are currently mutually exclusive. Moving forward, further vehicle upgrades will permit simultaneous measurement and sampling operations.