LIMNOLOGY and OCEANOGRAPHY: METHODS Limnol. Oceanogr.: Methods 10, 2012, 934–951 © 2012, by the American Society of Limnology and Oceanography, Inc. Autonomous detection and sampling of water types and fronts in a coastal upwelling system by an autonomous underwater vehicle Yanwu Zhang*, John P. Ryan, James G. Bellingham, Julio B. J. Harvey, and Robert S. McEwen Monterey Bay Aquarium Research Institute Abstract Coastal upwelling occurs under the combined effect of wind stress and Earth’s rotation. The nutrients carried up by upwelling have great impact on primary production and fisheries. For using autonomous underwater vehicles (AUVs) to investigate complex coastal upwelling ecosystems, we have developed algorithms for an AUV to autonomously distinguish between upwelling and stratified water columns based on the vertical temperature difference between shallow and deep depths, and to accurately detect an upwelling front based on the horizontal gradient of the vertical temperature difference in the water column. During a June 2011 experiment in Monterey Bay, California, the Dorado AUV flew on a transect from an upwelling shadow region (stratified water column), through an upwelling front, and into an upwelling water column. Running our algorithms, the AUV successfully classified the three distinct water types, accurately located the narrow front, and acquired targeted water samples from the three water types. Molecular analysis of the AUV-acquired water samples shows that mussels, calanoid copepods, and podoplean copepods were most abundant in the upwelling shadow region and nonexistent in the upwelling water column. Calanoid copepods were moderately abundant in the water samples collected from the upwelling front. These results are largely consistent with previous findings from zooplankton population surveys conducted with the Dorado AUV in Monterey Bay in 2009. The novel detecting and targeted sampling capabilities permit an AUV to autonomously conduct “surgical sampling” of a complex marine ecosystem. stratification and makes water more homogeneous over depth. Consider an alongshore equator-ward wind on the west coast as illustrated in Fig. 1. The wind stress combined with the effect of Earth’s rotation will lead to a net transport of a surface layer of water offshore, called Ekman transport (Kundu 1990). To conserve mass, deep water wells up to replace the displaced surface water – thus the name “upwelling.” The upwelling process brings cooler, saltier, and usually nutrientrich deep water upward, replacing warmer, fresher, and nutrient-depleted surface water (Kudela et al. 2008). The nutrients carried up by upwelling have great impact on primary production and fisheries. When a wind persistently blows from northwest to southeast (typically in spring and summer) along the California coastline, intense upwelling takes place at Point Año Nuevo (to the northwest of Monterey Bay labeled in Fig. 2) (Ramp et al. 2005). For instance, during the June 2011 CANON (Controlled, Agile, and Novel Observing Network) Experiment, upwelling of cold deep water to the surface was shown by low sea surface temperature (SST) at Point Año Nuevo and in the upwelling filaments spreading southward from Point Año Nuevo, as displayed in Fig. 2. However, a mountain range along the northern coast reduces wind-driven turbulent verti- When surface water is warmed by sunlight, temperature at the surface is higher than that at depth, forming a stratified water column (density increasing with depth) that is gravitationally stable (Wright 1995). Coastal upwelling (Huyer 1983) (Kudela et al. 2008) is a wind-driven ocean process that breaks *Corresponding author: E-mail: [email protected]; 7700 Sandholdt Road, Moss Landing, CA 95039, U.S.A.; Phone: (831)775-2028; Fax: (831)775-1646 Acknowledgment This work was supported by the David and Lucile Packard Foundation. The authors would like to thank the MBARI Dorado AUV team Hans Thomas, Duane Thompson, and Douglas Conlin, as well as the R/V Zephyr crew. The authors are thankful to Drs. Francisco Chavez, Christopher Scholin, and Robert Vrijenhoek for their leadership and support during the CANON experiments, as well as their helpful suggestions on improving the algorithms. Thanks also go to Raphael Kudela and Cindy Ruhsam (University of California at Santa Cruz) for processing and providing the satellite AVHRR SST data, and to Mike McCann and Thom Maughan for supporting MBARI’s Shore-Side Data System (SSDS) and Oceanographic Decision Support System (ODSS), respectively. The authors would also like to thank the anonymous reviewers for their comments and suggestions for improving the paper. DOI 10.4319/lom.2012.10.934 934 Zhang et al. AUV sampling across an upwelling front plankton and zooplankton populations, as well as physical processes that can locally enhance plankton aggregation and nutrient supply (Ryan et al. 2010a; Ryan et al. 2010b; Harvey et al. 2012), thus playing an important role in ocean ecosystems. Autonomous underwater vehicles (AUVs) are becoming increasingly useful platforms in oceanographic studies due to their mobility, efficiency, and growing autonomy. An early use of AUVs for studying ocean fronts was in the 1996 Haro Strait Frontal Dynamics Experiment (Bellingham 1997). In a coordinated effort to map the tidal fronts (Farmer et al. 1995), two Odyssey-class AUVs were deployed along with ships and drifters. The AUVs conducted high-resolution surveys of temperature, salinity, and current velocity in the frontal zone (Bellingham et al. 1997; Zhang and Willcox 1997). An acoustic source was also installed on one vehicle for frontal mapping by acoustic tomography (Schmidt et al. 1996). The Harvard Ocean Prediction System (HOPS) ocean model (Robinson et al. 1996) was run to predict the front’s location and guide the AUVs’ deployment. However, the AUVs were not capable of autonomously detecting the front. A prior effort aimed at detecting and tracking fronts in Monterey Bay involved an approach called “mixed initiative” (Rajan et al. 2009). This approach begins with an AUV conducting a reconnaissance survey and sending a compressed data set back to a scientist onshore via satellite. The scientist examines the data for the presence of a temperature front and derives parameters that will permit the AUV to locate the center of the front: (1) the depth at which to transit during the search, and (2) the water temperature at the center of the front cal mixing in the northern Monterey Bay. The northern bay is also sheltered from the prevailing southward flow of upwelling filaments (from Point Año Nuevo) by its coastal recess. Due to such atmospheric and oceanic sheltering, the northern bay is referred to as the Monterey Bay “upwelling shadow” (Breaker and Broenkow 1994; Graham and Largier 1997), featuring stratified water columns with relatively high SST (as shown in Fig. 2) due to enhanced residence time and local heating. The boundary between the stratified, biologically enriched water of the upwelling shadow, and the unstratified, biologically impoverished water transported southward from the Point Año Nuevo upwelling center, is called the “upwelling front.” Upwelling fronts support enriched phyto- Fig. 1. A wind blowing from north to south drives upwelling along a north-south coastline in the northern hemisphere. Fig. 2. SST in Monterey Bay during the June 2011 CANON Experiment. The line along 36.9°N (from 121.9°W to 122.25°W) marks the Dorado AUV’s 31-km transect in Mission 2011.164.05 (see Assessment). Time is in Pacific Daylight Time (PDT). Advanced Very High Resolution Radiometer (AVHRR) SST data courtesy of Kudela Lab (University of California at Santa Cruz) and National Oceanographic and Atmospheric Administration (NOAA) CoastWatch. 935 Zhang et al. AUV sampling across an upwelling front samplers (called “gulpers”) (Bird et al. 2007) for taking water samples back to shore for laboratory analysis, as shown in the lower panel in Fig. 3. Once triggered, each gulper acquires a 1.8-L water sample in less than 2 s. In a series of field programs in Monterey Bay from October 2009 to April 2011, the Dorado AUV ran our autonomous peak-capture algorithm (Zhang et al. 2010b) to trigger the gulpers for acquiring water samples with peak chlorophyll fluorescence signals from phytoplankton patches, or water samples with peak optical backscatter signals from intermediate nepheloid layers (Zhang et al. 2010a). Such targeted water sampling allowed biologists to successfully monitor fluctuations in harmful microalgae (Psuedo-nitzschia spp.), the toxin they produce (domoic acid) and co-occurring zooplankton (invertebrate larvae and copepods) over space and time. To advance ecological studies of a coastal upwelling system, we at the search depth. This data-informed human initiative is relayed back to the AUV and thereafter integrated with machine initiative for the AUV to track the front using a repeated sequence of localization and mapping. An additional parameter sent by the scientist to the AUV—the vertical scale over which to map the front—permits optimization of horizontal resolution through the frontal zone. These capabilities successfully demonstrated adaptive control of an AUV using onboard planning and execution. However, the requirement for a human in the loop imposes limitations and introduces potential failure points. The Dorado AUV (Bellingham et al. 2000; Sibenac et al. 2002) developed at the Monterey Bay Aquarium Research Institute (MBARI) is shown in Fig. 3. It is propeller driven, with a typical speed of about 1.5 m/s. In addition to a suite of sensors, the AUV is equipped with ten 1.8-L syringe-like water Fig. 3. Upper panel: the Dorado AUV. Lower panel: the ten 1.8-L water samplers (“gulpers”) installed in the mid-section of the vehicle (courtesy of Larry Bird and Alana Sherman of MBARI). 936 Zhang et al. AUV sampling across an upwelling front The different vertical structures of temperature in stratified and upwelling water columns, and in the narrow front between them, are illustrated in the upper panel in Fig. 4. In upwelling water, the vertical temperature difference between shallow and deep depths is small due to the upwelling process; but in stratified water, the vertical temperature difference is large (warm at the surface and cold at depth). In the narrow upwelling front, the vertical temperature difference has a sharp horizontal gradient. Phytoplankton patches, featuring high chlorophyll fluorescence, can form in the stratified water column, as illustrated in the lower panel in Fig. 4. Some phytoplankton patches appear in the form of thin layers, with thickness ranging from < 1 m to a few meters, and can horizontally extend for kilometers (Cowles et al. 1998; Ryan et al. 2008, 2010b). AUV algorithm for distinguishing between upwelling and stratified water columns The key to classification is finding the feature that maximizes the separability between classes. In an upwelling water column, temperature, salinity, and other properties are much more homogeneous over depth than in stratified water. Drawing on this difference, we formulated a simple yet effective classifier for differentiating upwelling and stratified water columns—the vertical temperature difference in the water column (Zhang et al. 2011a): set a more challenging goal: to enable the AUV to autonomously recognize and acquire targeted water samples from the three distinct water types—the stratified water column, the upwelling water column, and their frontal boundary (i.e., the upwelling front). The work presented in this paper represents the first fully autonomous AUV mission that achieved the above goal, by synergetically running the following three algorithms: An algorithm for an AUV to autonomously distinguish between upwelling water and stratified water based on the vertical temperature difference in the water column (Zhang et al. 2011a, 2012a). We will summarize the algorithm in Materials and procedures as the basis for developing the front-detection algorithm (see below). An algorithm for an AUV to autonomously and accurately detect an upwelling front based on the horizontal gradient of the vertical temperature difference in the water column. The algorithm was succinctly reported in (Zhang et al. 2011b). We will fully present the algorithm in Materials and procedures. An improved autonomous peak-capture algorithm for phytoplankton patch sampling in the stratified water column. An introduction of the original algorithm and highlights of the improvements will be given in Materials and procedures. The first time the set of three algorithms were integratively applied on an AUV was in the CANON Experiment in June 2011. In a front-crossing mission, the Dorado AUV flew on a 31 km transect from an upwelling shadow region (i.e., a stratified water column), through an upwelling front, and into an upwelling water column (the transect is marked by the eastwest line in Fig. 2). Running the set of three algorithms along with our gulper management program, the AUV autonomously classified stratified water and upwelling water, accurately detected the upwelling front, and acquired targeted water samples from the three distinct water types (stratified water, upwelling water, and the narrow front between them). This unprecedented AUV mission will be presented in detail in Assessment. DTempvert = |Tempshallow – Tempdeep| (1) where Tempshallow and Tempdeep are temperatures at shallow and deep depths, respectively. DTempvert is significantly smaller in upwelling water than in stratified water. If DTempvert falls below a threshold threshDTemp, the water column is classified as an upwelling water column: DTempvert ≤ threshDTemp (2) Suppose an AUV flies from stratified water to upwelling water on a sawtooth (i.e., yo-yo) trajectory (in the vertical dimension). The algorithm for the AUV to determine that it has departed from the stratified water column and entered the upwelling water column is illustrated in Fig. 5. Previous temperature measurements in Monterey Bay indicated that DTempvert between 5 m and 20 m depths provided a strong contrast between upwelling and stratified water columns. Hence we set the shallow depth to 5 m and the deep depth to 20 m in Eq. 1. For classification, we set the threshold threshDTemp = 1°C (again based on previous AUV temperature measurements in Monterey Bay) in Eq. 2. On each yo-yo profile (descent or ascent), the AUV records Temp5m and Temp20m to calculate DTempvert = |Temp5m – Temp20m|. When DTempvert falls below threshDTemp for 5 consecutive yo-yo profiles, the AUV determines that it has entered an upwelling water column. The requirement of DTempvert on 5 consecutive yo-yo profiles falling below threshDTemp is for robust detection of Materials and procedures The Dorado AUV, as shown in Fig. 3, has a length of 4.2 m and a diameter of 0.53 m at the midsection. The vehicle’s sensor suite (Thompson 2007) includes Sea-Bird SBE3 temperature and SBE4 conductivity sensors, a Paroscientific 8CB4000-I pressure sensor, a HOBI Labs HydroScat-2 sensor for measuring chlorophyll fluorescence at 700 nm wavelength and optical backscatter at 420 nm and 700 nm wavelengths, a Sea-Bird SBE43 oxygen sensor, and a Satlantic ISUS (In Situ Ultraviolet Spectrophotometry) sensor for measuring the concentration of nitrate anions. In addition, the AUV has ten gulpers, as shown in the lower panel of Fig. 3. Each gulper has a one-way valve opening, which extends through the vehicle’s hull. When triggered, an electromagnetic pin-puller releases a dual spring array under tension, which causes a plunger to suck in the 1.8L water sample in less than 2 s (Bird et al. 2007). 937 Zhang et al. AUV sampling across an upwelling front Fig. 4. Different vertical structures of temperature (upper panel) and chlorophyll fluorescence (lower panel) in stratified and upwelling water columns, and in the narrow front between them. Temperature from high to low is represented by color ranging from orange to blue (i.e., the temperature decreases with depth) in the upper panel. A phytoplankton layer in the stratified water column is marked red (indicating high chlorophyll fluorescence) in the lower panel. Fig. 5. Illustration of the algorithm for an AUV to determine that it has departed from a stratified water column and entered an upwelling water column. The first four yo-yo profiles that satisfy DTempvert ≤ threshDTemp are marked blue. The fifth such yo-yo profile is marked green. 938 Zhang et al. AUV sampling across an upwelling front an upwelling water column (i.e., small patches of water with low DTempvert will be ignored). Conversely, suppose an AUV flies from upwelling water to stratified water. When DTempvert rises above threshDTemp for 5 consecutive yo-yo profiles, the AUV determines that it has entered a stratified water column. The horizontal distance covered by one yo-yo profile depends on the yo-yo depth range as well as the AUV’s flight attitude. In Dorado AUV Mission 2011.164.05 (to be presented in Assessment), the horizontal distance separated by 5 yo-yo profiles was about 650 m. AUV algorithm for detecting an upwelling front The vertical temperature difference DTempvert is large in stratified water but small in upwelling water. As an AUV flies on a yo-yo trajectory (in the vertical dimension) on a transect from stratified water to upwelling water, DTempvert drops sharply (along the horizontal direction) in the narrow upwelling front (conversely, DTempvert rises sharply in the narrow upwelling front on an AUV transect from upwelling water to stratified water). Hence we have developed an AUV algorithm that uses the horizontal gradient of DTempvert to detect an upwelling front. The algorithm was succinctly reported in (Zhang et al. 2011b) and is fully presented herein. The algorithm is illustrated in Fig. 6 for an AUV flying from stratified water to upwelling water. When the vehicle flies through the narrow upwelling front, DTempvert drops sharply. The horizontal gradient of DTempvert is calculated as follows GradDTemp_vert(n) = DTempvert(n) – avg{DTempvert(n – 5 : n – 9)} where n is the index of the current yo-yo profile; avg{·} stands for the average. The AUV calculates the horizontal gradient of DTempvert by finding the difference between DTempvert of the current yo-yo profile (index n) and DTempvert of several yo-yo profiles ago—the reference level. The reference level is represented by the average DTempvert from profile n-5 to n-9—the reference profiles. Thus the interval between the current profile and the center of the reference profiles is 7. The rationale for leaving a 7-profile interval is as follows. The frontal zone has fine-scale variabilities. Placing the reference profiles in the preceding water type (e.g., the stratified water as in Fig. 6) at some distance away from the current profile, and finding the difference between the reference level (i.e., the average DTempvert of the reference profiles) and DTempvert of the current profile, provides a robust contrast that reflects the horizontal change of water types. Yo-yo profile n qualifies as an “upwelling-front yo-yo profile” if both of the following conditions are met (Eq. 4 and Eq. 5): The key signature of an upwelling front is high horizontal gradient of DTempvert: |GradDTemp_vert(n)| ≥ threshTempGrad (4) where threshTempGrad is a horizontal gradient threshold. In Dorado AUV Mission 2011.164.05, we set threshTempGrad = 0.2°C over a horizontal distance of about 900 m separated by 7 yoyo profiles. (3) Fig. 6. Illustration of the AUV algorithm for detecting an upwelling front when flying from stratified water to upwelling water. The first two yo-yo profiles that satisfy |GradDTemp_vert| ≥ threshTempGrad and DTempvert ≤ threshDTemp_front are marked purple. The third such yo-yo profile is marked red. 939 Zhang et al. AUV sampling across an upwelling front Development of the peak-capture algorithm was motivated by the goal of optimally using AUVs for studying phytoplankton thin layers (Cowles et al. 1998; Sullivan et al. 2010; McManus et al. 2008; Ryan et al. 2008, 2010b). Subsurface layers of phytoplankton across a wide range of thickness are common in coastal waters, and the algorithm is effective for capturing the peak signal regardless of the layer’s thickness. In this section, we first introduce the original algorithm in the context of a subsurface phytoplankton layer. Then we highlight the subsequent improvements we have made to the algorithm. In Assessment, we demonstrate how the Dorado AUV ran this algorithm to effectively sample plankton populations in subsurface layers, some of which qualified as “thin layers.” Within a phytoplankton layer, a high biomass concentration leads to high levels of chlorophyll fluorescence and optical backscatter signals (Ryan et al. 2010b; Sullivan et al. 2010). When an AUV traverses the layer, in theory it could detect the fluorescence peak and trigger the gulper to acquire a water sample. However, a detection delay is unavoidable for any real-time peak-detection algorithm—the peak is detected only when it has just passed. Such a delay is particularly a problem for a thin layer: even a small delay will result in water-sample acquisition occurring past the fluorescence peak target. To overcome the peak-detection delay problem, our algorithm takes advantage of the AUV’s yo-yo trajectory (in the vertical dimension), as illustrated in Fig. 7. In one yo-yo cycle, e.g., an ascent profile followed by a descent profile, the vehicle crosses the layer twice, measuring a fluorescence peak at each cross- Due to the upwelling front’s proximity to the upwelling water, the vertical temperature difference in the upwelling front is low: DTempvert(n) ≤ threshDTemp_front (5) where threshDTemp_front is a vertical temperature difference threshold. We set threshDTemp_front = 1.5°C, which is 0.5°C higher than the upwelling water threshold threshDTemp = 1°C, due to the consideration that in the upwelling front temperature is vertically more homogeneous than in the stratified water, but not yet as vertically homogeneous as in the upwelling water column. On detection of 3 consecutive upwelling-front yo-yo profiles, the AUV determines that it has robustly detected the upwelling front. AUV algorithm for capturing chlorophyll fluorescence peaks We have previously developed a peak-capture algorithm for an AUV to autonomously detect chlorophyll fluorescence peaks for triggering water sample acquisition in phytoplankton patches (Zhang et al. 2010b). The algorithm cross-checks for concurrent high values of the optical backscatter signal to ensure that the sampled fluorescence peaks are true biomass maxima. Using sliding temporal windows, the algorithm keeps track of the background levels of chlorophyll fluorescence and optical backscatter signals, as well as the fluorescence signal’s average peak level on successive yo-yo profiles, thereby adapting sensitivity to ambient conditions. With signal processing in the time domain, the algorithm can localize phytoplankton patches at any depth in the water column. Fig. 7. On an AUV’s yo-yo trajectory traversing a phytoplankton layer, on the first crossing, the vehicle detects the chlorophyll fluorescence peak with a delay but saves the peak signal value in a sliding window. On the second crossing, the AUV triggers a gulper when the fluorescence measurement reaches the saved peak signal value, thus capturing the peak with no delay. If no triggering occurs on the second crossing, the AUV can trigger a gulper at the third crossing if the signal level condition is met, labeled as the “backup triggering.” 940 Zhang et al. AUV sampling across an upwelling front detected peaks’ baseline is calculated as the average within a sliding window (of a pre-set duration), rather than a running average from the mission start. 2. When low-pass filtering the raw fluorescence or backscatter measurements, the largest raw measurement value inside the low-pass filter window is excluded to prevent any singular spurious measurement from affecting the low-pass filtered output. 3. For peak detection on the AUV’s first crossing of the layer, within the sliding window (see Fig. 7), we first use a sliding median filter (shorter than the sliding window) to exclude spurious fluorescence peaks. The maximum output from the sliding median filter is saved as the peak fluorescence signal value. On the AUV’s second crossing of the layer, for triggering a gulper, it is required that two consecutive fluorescence measurements reach (or exceed) the saved peak signal value to prevent triggering by a single anomalous fluorescence measurement. 4. When a layer lies near the shallow limit of the AUV’s yoyo trajectory, due to the delay in detection of the vehicle’s attitude change, the AUV may not trigger a gulper on the second crossing of the layer. Even for a deep layer, signal variation in the layer may also render no triggering on the AUV’s second crossing. To make up for this possibly missed opportunity, we have added an option of allowing the AUV to trigger a gulper at the third crossing (i.e., the “backup triggering” labeled in Fig. 7) as long as the signal level condition is met. ing. We assume that at two consecutive crossings that are no more than several hundred meters apart, the two peaks will have similar signal levels. On the vehicle’s first crossing (on the ascent profile), the peak detection (by tracking the fluorescence signal’s slope) comes with a delay, but the vehicle uses a sliding window (which saves a number of past samples) to save the signal value of the true peak (which has just passed). On the second crossing (on the descent profile), as soon as the fluorescence measurement reaches the saved peak value, a sampling is triggered. This way, the sampling is effectively targeted on the peak. To reduce noise, the raw fluorescence measurements are low-pass filtered by taking the moving-average of a few data samples, generating Fllp. The low-pass filtered backscatter signal bblp is generated in the same way. On the AUV’s first crossing of the phytoplankton layer, a peak is detected when the slope of Fllp changes from positive to negative. The background fluorescence signal level Flbkgd is calculated as the running average from the mission start. The background backscatter signal level bbbkgd is calculated in the same way. On detection of fluorescence peaks, the algorithm cross-checks for concurrent high values of optical backscatter to ensure that water-sample acquisition targets true particle peaks and not simply physiologically controlled fluorescence peaks (Ryan et al. 2010b). Out of the detected fluorescence peaks, only those that have both high fluorescence and high backscatter signals are considered “real peaks” in a phytoplankton layer, the criteria being (1) Fllp Flbkgd ≥ a Fl AND (2) bblp bbbkgd Assessment ≥ abb where aFl The Monterey Bay primary upwelling source, the Point Año Nuevo upwelling center (see Fig. 2), and the upwelling shadow in the northern recess of the bay together form a system that hosts recurring creation and dynamic evolution of upwelling fronts. The objective of the June 2011 CANON Experiment was to advance the interdisciplinary study of the dynamic frontal zone. A key goal was to employ the Dorado AUV to acquire water samples precisely in the three different types of water columns across the upwelling front (i.e., in the upwelling shadow (stratified water), in the upwelling water, and in the narrow front between them) for investigating contrasting plankton communities in the distinct water masses. The Dorado AUV was deployed on a 31-km transect along 36.9°N (from 121.9°W to 122.25°W), as marked in Fig. 2. Water depth along this transect ranges from 24 m at the east end to 208 m at the west end. This transect was selected based on multi-year satellite SST and chlorophyll fluorescence line height (FLH) data for the month of June that showed high horizontal gradients between the upwelling filament and the upwelling shadow toward the northern end of the bay, thus providing a relatively high probability of encountering strongly contrasting water types across an upwelling front. Evolutions of upwelling fronts in Monterey Bay are highly dynamic, making it impossible to accurately predict the front’s location for preprogramming the AUV’s water sam- and abb are factors larger than one. The peaks’ baseline Fl_PEAKbaseline is calculated as the running average of the “real peaks” from the mission start. Only those peaks that are higher than Fl_PEAKbaseline can lead to the triggering of a gulper on the AUV’s second crossing of the phytoplankton layer. Note that on the second crossing of the layer, the AUV triggers a gulper when the raw fluorescence measurement (rather than Fllp) reaches the saved peak value. Hence no delay due to lowpass filtering is introduced. The algorithm has been used in many AUV missions for acquiring water samples from phytoplankton patches of various thicknesses and at various depths. For Dorado AUV Mission 2011.164.05 to be presented in Assessment, we will give the parameter settings as well as the detected layer’s thickness and peak-to-background signal ratio. Based on the original algorithm, we have made the following improvements. The Dorado AUV ran the improved peakcapture algorithm in Mission 2011.164.05. 1. The background level of fluorescence (or backscatter) is calculated as the average on the last N yo-yo profiles (each yoyo profile typically contains hundreds of data samples. N is a pre-set number), rather than a running average from the mission start. This way, the background level calculation is not affected by signal levels in the distant past. Likewise, the 941 Zhang et al. AUV sampling across an upwelling front the Earth-referenced velocity of the AUV when the seabed was in range) and the measured heading and attitude. The AUV periodically (about every 47 minutes) ascended to the surface for global positioning system (GPS) fixes to correct for the vehicle’s underwater navigation errors. The flow diagram of the AUV’s classification of different water columns and targeted sampling in the respective water columns is given in Fig. 8. Our gulper management program [originally devised by the fifth and second coauthors (Zhang et al. 2010b) and then extended by the first and second coauthors] allocates gulpers to respective water types, and applies triggering constraints (lockout time interval, max-wait time, and lockout depth, defined below) to optimize sampling performance. In the stratified water column, when the AUV traverses a phytoplankton patch (which can be close to the surface or the bottom, or in the interior of the water column), it detects the chlorophyll fluorescence peak and saves the peak fluorescence signal value on the first crossing. (On detection of fluorescence peaks, the algorithm cross-checks for concurrent high values of optical backscatter). On the second (or third) crossing, when the fluorescence measurement reaches the saved peak signal level, the AUV triggers a gulper on condition that the elapsed time since the last triggering exceeds the lockout time interval and the depth exceeds the lockout depth. The lockout time intervals between triggerings are set to prevent “dense triggerings” that would use up gulpers over a short distance. The lockout depth is set to prevent gulping air bubbles. pling locations. The AUV must make decisions by itself. During the June 2011 experiment, by running the presented algorithms, the Dorado AUV autonomously distinguished the water types, accurately located the front, and triggered water samplings in the three distinct water columns, as demonstrated in Mission 2011.164.05. Overview of the Dorado AUV’s autonomous detection and targeted sampling in Mission 2011.164.05 The Dorado AUV started Mission 2011.164.05 at 13:35 (PDT) on 13 June 2011 from the east end of the 31-km transect line, flying westward, on a yo-yo trajectory between the surface and 25-m depth (except for a small portion near the east end of the transect where the water depth is smaller than 33 m). Note that the AUV maintained a minimum altitude of about 8 m for safety. When the vehicle reached the west end of the transect line, it turned around to fly eastward. The AUV’s average horizontal speed was about 1 m/s. Its average vertical speed was about 0.15 m/s on descent profiles and 0.29 m/s on ascent profiles, respectively. Thus the flight-path angle of the yo-yo trajectory (the angle between the trajectory and ⎛ 0.15 m / s ⎞ the horizontal) was β flight _ path = atan ⎜ ⎟ = 9° on descent ⎝ 1m/s ⎠ ⎛ 0.29 m / s ⎞ profiles and β flight _ path = atan ⎜ ⎟ = 16° on ascent profiles. ⎝ 1m/s ⎠ The AUV’s underwater navigation was by dead reckoning based on the vehicle’s estimated speed (aided by a Teledyne RD Instruments Doppler velocity log (DVL) which provided Fig. 8. Flow diagram of the AUV’s classification of different water columns and targeted sampling in the respective water columns. 942 Zhang et al. AUV sampling across an upwelling front No. 3, 4, 5, 6 were triggered in the narrow upwelling front; gulpers No. 7, 8, 9 were triggered in the upwelling water column. The AUV reached the west end of the transect line in 9.6 h, and then turned around to fly eastward. We only show the westbound transect in Fig. 9 and the ensuing figures, because all water samplings had been completed on the westbound flight. Peak-capture sampling of the phytoplankton layer in the stratified water column In the upwelling shadow region (near the east end of the transect), there existed a phytoplankton layer with an intense chlorophyll fluorescence signal, as shown in the upper panel of Fig. 9. The Dorado AUV ran our peak-capture algorithm to trigger three gulpers (No. 0, 1, 2) at chlorophyll fluorescence peaks in the layer, as marked by the black triangles in the upper panel in Fig. 10. The fluorescence signal and the AUV’s depth associated with the three triggerings are shown in the three columns in the middle and lower panels in Fig. 10. In the middle panel, the raw fluorescence measurement is shown by the black solid line. The 8-sample (at 4 Hz sampling rate) low-pass filtered fluorescence signal Fllp is shown by the magenta line (the lowpass filtered backscatter signal bblp was calculated in the same way but not shown due to limit of space). Detection of the peaks (marked by circles) was by tracking the slope of Fllp. The background fluorescence signal level Flbkgd, calculated as the average on the last 10 yo-yo profiles (or the last n profiles if the profile index n < 10), is shown by the black dotted line (the background backscatter signal level bbbkgd was calculated in the same way but not shown). Out of the detected fluorescence peaks, only those that had both high fluorescence and high backscatter signals were considered “real peaks” in a phytoplankton layer, the criteria being (1) Fllp ≥ 1.5 Flbkgd; (2) bblp ≥ 1.2 bbbkgd. The peaks’ baseline Fl_PEAKbaseline was calculated as a 20-minute moving-average of the “real peaks,” as shown by the blue dashed line. Only those peaks that were higher than Fl_PEAKbaseline were eligible for leading to the triggering of a gulper. For triggering gulper No. 0, the AUV detected the fluorescence peak on ascent (i.e., the first crossing of the phytoplankton layer) and used the 8-sample sliding window to save the peak signal value as marked by the solid triangle (a 5-sample median filter was used to exclude spurious peaks). On the AUV’s descent (i.e., the second crossing of the layer), when two consecutive fluorescence measurements reached (or exceeded) the saved peak signal level, the vehicle triggered gulper No. 0, as marked by the hollow triangle. Gulpers No. 1 and 2 were triggered on the AUV’s third crossings of the phytoplankton layer. The lockout time interval was set to 2700 s, and the lockout depth was set to 1.5 m. The fluorescence peakto-background ratios at the triggerings of gulpers No. 0, 1, and 2, were 4.3, 5.4, and 7.2, respectively. The vertical full-widthat-half-maximum (FWHM) of the layer captured by gulpers No. 0, 1, and 2 was 6.8 m, 1.9 m, and 1.5 m, respectively. All On the completion of a yo-yo profile, the AUV checks whether GradDTemp_vert and DTempvert meet the conditions for an upwelling front (see Fig. 6). If so, a “front flag” is set. The depth of the maximum low-pass filtered fluorescence on the preceding yo-yo profile has been recorded. On the ensuing yoyo profile, when the AUV reaches that depth, a gulper is triggered (as long as the lockout time interval and lockout depth conditions are also satisfied). The purpose is to acquire water samples with high fluorescence signals in the front. After a triggering, the front flag is reset. If the conditions for an upwelling front are not met, the AUV further checks whether DTempvert meets the condition for an upwelling water column (see Fig. 5). If the upwelling condition is met and the lockout time interval condition is also satisfied, the AUV triggers a gulper at either a pre-set depth or at the lower turning point of the yo-yo profile (for acquiring a water sample at a relatively deep spot in the upwelling water column). A “max-wait time” is set for each gulper. In each type of water column, if the max-wait time has elapsed but the detection conditions have not been met, the AUV will trigger a gulper under the only condition of lockout depth, so that the AUV will return to shore with all gulpers filled even if some water samples are not target samples. In Dorado AUV Mission 2011.164.05, the ten gulpers were pre-allocated to the three types of water columns as follows: three (No. 0, 1, 2) in the stratified water, four (No. 3, 4, 5, 6) in the upwelling front, and the remaining three (No. 7, 8, 9) in the upwelling water. We allocated more gulpers for the upwelling front because of high interest in studying plankton populations in the front, and also because it was very hard to acquire water samples from the narrow front using traditional methods. No gulper was triggered on max-wait time, i.e., all gulpers detected the respective water columns and acquired targeted water samples. An overview of the triggerings of the ten gulpers is shown in Fig. 9, with longitude and elapsed time (since the start of the mission) both marked on the x-axis. The elapsed time was proportional to the AUV’s traveled distance, but their relationship was not exactly linear because the AUV’s surfacing frequency and duration (for GPS fixes) varied over the mission. Chlorophyll fluorescence and temperature measured by the AUV are displayed in the upper and lower panels, respectively. The measured temperature shows that the AUV started from stratified water, flew through a narrow front, and into upwelling water. The measured chlorophyll fluorescence shows that in the stratified water column (in the upwelling shadow region), there existed a phytoplankton layer with an intense chlorophyll fluorescence signal. The AUV autonomously distinguished among the three water types and triggered the gulpers in the respective water columns, as marked by the corresponding symbols in Fig. 9: gulpers No. 0, 1, 2 were triggered at chlorophyll fluorescence peaks in the phytoplankton layer in the stratified water column; gulpers 943 Zhang et al. AUV sampling across an upwelling front Fig. 9. In Mission 2011.164.05, the Dorado AUV flew westward on a 31-km transect along 36.9°N, on a yo-yo trajectory from the surface to 25 m depth (except for a shallow-water portion near the east end of the transect). Chlorophyll fluorescence and temperature measured by the AUV are shown in the upper and lower panels, respectively. Triggerings of the AUV’s ten gulpers in the three distinct water columns are marked by the corresponding symbols. three triggerings, particularly gulpers No. 1 and 2, captured high peaks of chlorophyll fluorescence. By the criteria for phytoplankton thin layers (in terms of fluorescence peak-to-background ratio and layer thickness) as given in (Ryan et al. 2010b), the layer from which gulpers No. 1 and 2 acquired peak-fluorescence water samples was a phytoplankton thin layer. Front detection and sampling in the narrow upwelling front At about 6 h after the start of the AUV mission, the vehicle encountered the upwelling front, signified by a sharp horizontal gradient of DTempvert as shown in the middle panel in Fig. 11: to the left of the front DTempvert was large, but to the right of the front DTempvert was small. Running our front detec- tion algorithm, the Dorado AUV achieved the primary goal of accurately detecting the front and consequently triggering the four gulpers (No. 3, 4, 5, 6) within the narrow front, as marked by the red diamonds in the lower panel of Fig. 11. The lockout time interval was set to 300 s, and the lockout depth was set to 1.5 m. Due to a coding error, the secondary goal of triggering at the depth corresponding to the maximum fluorescence on the preceding yo-yo profile was not achieved. The error has since been corrected. The AUV detected the front by calculating DTempvert(n) (by Eq. 1 with the shallow and deep depths set to 5 m and 20 m respectively) and GradDTemp_vert(n) (by Eq. 3) in real time. These two variables were not logged, although the Sea-Bird CTD 944 Zhang et al. AUV sampling across an upwelling front Fig. 10. Triggerings of gulpers No. 0, 1, and 2 at chlorophyll fluorescence peaks in a phytoplankton layer (upper panel). Details of the three gulpers’ triggerings are shown in the middle and lower panels and explained in the text. marked red. The upward arrow on the threshTempGrad = 0.2°C red dashed line (in middle panel) indicates the “ ≥ ” condition in Eq. 4. The downward arrow on the threshDTemp_front = 1.5°C red dashed line (in upper panel) indicates the “ ≤ ” condition in Eq. 5. The four gulpers’ triggerings in the narrow upwelling front at about 6 h (after start of mission) are consistent with the “red profiles” around 6 h in the middle panel. The “red profile” at 2.6 h was less than 300 s (the lockout time interval) away from the last peak-fluorescence triggering (i.e., the third black triangle in the lower panel of Fig. 11). Therefore, no triggering occurred at that time. For the “red profile” at 3.5 h, there was no corresponding gulper triggering. A probable cause is as follows. On the yo-yo profile immediately preced- driver did log temperature. The logging process introduced a small delay (well within the temperature data sample interval of 0.2 s), so that the temperature values used by the AUV’s onboard calculation were slightly different from the logged temperature values. Nonetheless, in Fig. 11 we use the Sea-Bird CTD temperature log to verify that the AUV’s onboard detection and triggerings in the upwelling front performed as designed, being aware that the two temperature time series had slight differences. Using the Sea-Bird CTD temperature log, we calculate off-line the vertical temperature difference DTempvert(n) and its horizontal gradient |GradDTemp_vert(n)| on each yo-yo profile, as plotted in the upper and middle panels of Fig. 11, respectively. In the middle panel, those yo-yo profiles that meet the conditions for an upwelling front (see Fig. 6) are 945 Zhang et al. AUV sampling across an upwelling front Fig. 11. In the lower panel, the onboard triggerings of the four gulpers (No. 3, 4, 5, and 6) in the upwelling front are marked by the red diamonds. The vertical temperature difference DTempvert and its horizontal gradient |GradDTemp_vert| on each yo-yo profile, calculated off-line using the Sea-Bird CTD temperature log, are plotted in the upper and middle panels, respectively (the horizontal distance separated by 7 yo-yo profiles was about 900 m). Those yo-yo profiles that meet the conditions for an upwelling front are marked red (middle panel). Those yo-yo profiles that meet the condition for an upwelling water column are marked green (upper panel). facilitate off-line verification, we will log the two onboard-calculated variables DTempvert(n) and GradDTemp_vert(n). Detection and sampling in the upwelling water column After flying through the upwelling front, the AUV entered the upwelling water column. Running our classification algorithm, the vehicle recognized the upwelling water and triggered the remaining three gulpers (No. 7, 8, 9) in the upwelling water column, as marked by the green squares in the lower panel of Fig. 11. The AUV triggered each of the three gulpers at the lower turning point of the ing that “red profile,” DTempvert = 1.43°C, very close to the 1.5°C threshold (the second condition for qualifying as an “upwelling front yo-yo profile”—Eq. 5). It is probable that the onboard-calculated counterpart was slightly higher than the 1.5°C threshold (i.e., not meeting that condition), and consequently there was no qualified “upwelling front yo-yo profile” at 3.5 h, hence no gulper was triggered. Note that the four gulpers allocated for the upwelling front had all been triggered before 6.5 h. Therefore, the “red profiles” after that time point do not have corresponding gulper triggerings. In the future, to 946 Zhang et al. AUV sampling across an upwelling front Physical and biological analysis results of the AUVacquired water samples Physical water properties at the triggerings of the 10 gulpers are shown in the left column of Fig. 12. The 10 water samples’ zooplankton (mussels, barnacles, calanoid copepods, and podoplean copepods) signals were detected using the sandwich hybridization assay (SHA) method in a shore laboratory (Scholin et al. 1999; Harvey et al. 2012), as shown in the right column. The fluorescence and optical backscatter signals at the three triggerings in the phytoplankton layer (in the upwelling shadow region) were much higher than those in the upwelling front and in the upwelling water column. The three triggerings in the upwelling water column featured low temperature, high yo-yo profile to acquire a water sample at a relatively deep spot in the upwelling water column. The lockout time interval was set to 500 s. Again, we use the Sea-Bird CTD temperature log to verify that the AUV’s onboard detection and triggerings in the upwelling water column performed as designed. In the upper panel of Fig. 11, which shows DTempvert(n) calculated off-line using the Sea-Bird CTD temperature log, those yo-yo profiles that meet the condition for an upwelling water column (see Fig. 5) are marked green. The downward arrow on the threshDTemp = 1°C green dashed line (in upper panel) indicates the “ ≤ ” condition in Eq. 2. The three gulpers’ triggerings in the upwelling water column are consistent with the “green profiles” counterparts in the upper panel. Fig. 12. Physical water properties (measured in situ by the AUV, left column) and zooplankton signals (by using molecular probes in a shore laboratory, right column) of the water samples from the 10 gulpers. Unit of the zooplankton signals in the right column is optical density (OD) at 450 nm wavelength. (Note: gulper No. 5′s water sample volume was 1/3 smaller than the other gulpers’ water samples.) 947 Zhang et al. AUV sampling across an upwelling front Comments and recommendations salinity, high nitrate (NO3–) concentration, and low dissolved oxygen, all consistent with expectations for upwelling water. Note that the water samples from the three water types were taken from different depths. Mussels, calanoid copepods, and podoplean copepods were most abundant in the three water samples collected from the phytoplankton layer. Conversely, these organisms were nonexistent in the three water samples collected from the upwelling water column. Calanoid copepods were of moderate to high abundance in the water samples collected from the upwelling front (except for gulper No. 5’s water sample, which was smaller in volume than the other gulpers’ water samples by roughly 1/3). These results are largely consistent with previous findings from zooplankton population surveys using the Dorado AUV in Monterey Bay in 2009 (where the gulpers were triggered at prescribed geographic locations and depths) (Harvey et al. 2012). One notable difference is that while calanoid copepods were moderately abundant in the frontal water samples in 2011, none of those samples contained exceptionally high calanoid abundance as was present in the frontal water samples in 2009. However, we also note a difference of their locations: the frontal water samples in 2011 were taken slightly on the upwelling water column side, but the frontal water samples in 2009 were taken slightly on the stratified water column side. Another difference is that barnacles were present in two of the three water samples (abundant in one sample) collected from the upwelling water column in 2011, while the water samples collected from upwelling water columns in 2009 showed very low to no barnacle signal. These observations call for further studies involving iterative sampling of upwelling frontal environments and their bounding water masses. We have set up a key metric of the vertical temperature homogeneity in the water column—the vertical temperature difference between shallow and deep depths (Zhang et al. 2011a). This metric was our basis for differentiating upwelling and stratified water columns, and detecting the upwelling front. Based on this metric, in this section we propose a generic metric that includes more depths, as a better measure of the vertical temperature homogeneity in the water column. By post-processing the Dorado AUV data, we demonstrate that the generic metric improves the robustness of upwelling front detection. In the middle panel in Fig. 11, we notice 2 groups of “red profiles’’ (that meet the conditions of an upwelling front) around 7.5 h and 8 h, although they did not trigger any gulper because all the four gulpers allocated for the upwelling front had been triggered before 6.5 h. Those 2 groups of “red profiles” were actually not due to real upwelling fronts, but were due to variations of temperature at 5 m and 20 m depths. For instance, across 7.5 h Temp5m remained largely the same (and was higher than Temp20m), but Temp20m was higher after 7.5 h than before 7.5 h. As a consequence, DTempvert = |Temp5m – Temp20m| was smaller after 7.5 h than before 7.5 h. This caused a drop of DTempvert across 7.5 h, and accordingly led to false alarms of upwelling front detection at 7.5 h. For a more robust evaluation of the vertical temperature homogeneity in the water column, we propose a generic metric that includes more depths: DTempvert = 1 N 1 N ∑ Tempdepth_i − N ∑i=1 Tempdepth_i N i=1 (6) where i is the depth index and N is the total number of depths included for calculating DTempvert. Tempdepth_i is the temperature Discussion 1 N ∑ Tempdepth_i is the average temperature of N i =1 1 N those depths. Tempdepth_i − ∑i=1 Tempdepth_i measures the differN The integration of multiple signal processing algorithms worked well in the very first deployment on an AUV for autonomous detection and targeted water sampling in three water types. This was due to algorithm designs based on oceanographic insights, having robustly tested elements upon which to build and improve, data-informed algorithm parameters, and appropriate environmental conditions. We can apply the presented front detection method to other oceanographic properties, e.g., salinity fronts and chlorophyll fronts. The Environmental Sample Processor (ESP) is an underwater robotic laboratory that conducts molecular biology analysis on discrete water samples in situ (Scholin et al. 2009). There is an ongoing effort of incorporating the third generation ESP into the Tethys long-range AUV (Bellingham et al. 2010). The combination of an AUVborne ESP and the AUV’s capability of autonomously detecting ocean features provides an exciting prospect of conducting molecular biology analysis on targeted water samples in situ, with no human in the loop. at the ith depth. ence (absolute value) between the temperature at each depth and the depth-averaged temperature. The average difference DTempvert (averaged over all participating depths) is a measure of the vertical homogeneity of temperature in the water column, which we call the “vertical temperature homogeneity index.” The formulation of Eq. 6 draws upon the idea in the definition of variance in statistics, just replacing the square of each difference with the absolute value of each difference. By including multiple depths, the impact of temperature variation at a single depth on DTempvert is mitigated. Thus DTempvert defined in Eq. 6 is supposed to be a more robust measure of the water column’s vertical temperature homogeneity than DTempvert defined in Eq. 1. In the special case of N = 2 (i.e., only two participating depths), if depth1 denotes the shallow depth and depth2 948 Zhang et al. AUV sampling across an upwelling front denotes the deep depth, and suppose Tempdepth_1 > Tempdepth_2, then Eq. 6 reduces to DTempvert = 1 2 {[Temp +[ = depth_1 − AUVs during the April and June 2011 CANON Experiments) is actually a simplified case of the generic vertical temperature homogeneity index defined in Eq. 6. Now we test the upwelling front detection performance with the newly defined vertical temperature homogeneity index, using the Dorado AUV Mission 2011.164.05 data set. Four depths are included for calculating DTempvert by Eq. 6: 5 m, 10 m, 15 m, and 20 m. GradDTemp_vert (the horizontal gradient of DTempvert) is still calculated by Eq. 3. DTempvert(n) and |Grad(n)| on each yo-yo profile are plotted in the upper and DTemp_vert middle panels of Fig. 13, respectively. The two conditions for detection of an upwelling front are still expressed in Eqs. 4 and 5. For the first condition, we still 1 (Tempdepth_1 + Tempdepth_ 2 )] 2 1 (Tempdepth_1 + Tempdepth_ 2 ) − Tempdepth_ 2 ] 2 } (7) 1 (Tempdepth_1 − Tempdepth_ 2 ) 2 which is the same as Eq. 1 except for a factor of 1/2. Thus DTempvert defined in Eq. 1 (used by the Tethys and Dorado Fig. 13. Using Dorado AUV Mission 2011.164.05 data set, the vertical temperature homogeneity index DTempvert calculated by Eq. 6 is shown in the upper panel. Its horizontal gradient |GradDTemp_vert| is shown in the middle panel. Those yo-yo profiles that meet the conditions of an upwelling front are marked red (middle panel). 949 Zhang et al. AUV sampling across an upwelling front quences. Oceanography 11(1):4-9 [doi:10.5670/oceanog. 1998.08]. Farmer, D. M., E. A. D’Asaro, M. V. Trevorrow, and G. T. Dairiki. 1995. Three-dimensional structure in a tidal convergence front. Cont. Shelf Res. 15:1649-1673 [doi:10.1016/ 0278-4343(94)00084-Z]. Graham, W. M., and J. L. Largier. 1997. Upwelling shadows as nearshore retention sites: the example of Northern Monterey Bay. Cont. Shelf Res. 17:509-532 [doi:10.1016/S02784343(96)00045-3]. Harvey, J. B. J., and others. 2012. Robotic sampling, in situ monitoring and molecular detection of marine zooplankton. J. Exp. Mar. Biol. Ecol. 413:60-70 [doi:10.1016/j.jembe. 2011.11.022]. Huyer, A. 1983. Coastal upwelling in the California Current system. Progr. Oceanogr. 12(3):259-284 [doi:10.1016/00796611(83)90010-1]. Kudela, R. M., and others. 2008. New insights into the controls and mechanisms of plankton productivity in coastal upwelling waters of the Northern California current system. Oceanography 21(4):46-59 [doi:10.5670/oceanog. 2008.04]. Kundu, P. K. 1990. Fluid mechanics. Academic Press. McManus, M. A., and others. 2008. Cryptic blooms: are thin layers the missing connection? Estuar. Coasts 31:396-401 [doi:10.1007/s12237-007-9025-4]. Rajan, K., and others. 2009. Onboard adaptive control of AUVs using automated planning and execution. In Proc. 16th international symposium on unmanned untethered submersible technology, Durham, NH. AUSI. Ramp, S. R., and others. 2005. Observations of upwelling and relaxation events in the Northern Monterey Bay during August 2000. J. Geophys. Res. 110:C07013 [doi:10.1029/ 2004JC002538]. Robinson, A. R., and others. 1996. Real-time regional forecasting. In Modern approaches to data assimilation in ocean modeling. Elsevier Science [doi:10.1016/S0422-9894(96) 80017-1]. Ryan, J. P., M. A. McManus, J. D. Paduan, and F. P. Chavez. 2008. Phytoplankton thin layers caused by shear in frontal zones of a coastal upwelling system. Mar. Ecol. Progr. Ser. 354:21-34 [doi:10.3354/meps07222]. ———, and others. 2010a. Recurrent frontal slicks of a coastal ocean upwelling shadow. J. Geophys. Res. 115:C12070 [doi:10.1029/2010JC006398]. ———, M. A. McManus, and J. M. Sullivan. 2010b. Interacting physical, chemical and biological forcing of phytoplankton thin-layer variability in Monterey Bay, California. Cont. Shelf Res. 30:7-16 [doi:10.1016/j.csr.2009.10.017]. Schmidt, H., and others. 1996. Real-time frontal mapping with AUVs in a coastal environment, p. 1094-1098. In Proc. IEEE/MTS Oceans’96, Fort Lauderdale, FL. IEEE. Scholin, C., and others. 1999. Application of DNA probes and a receptor binding assay for detection of Pseudo-nitzschia set threshTempGrad = 0.2°C, the same as in Materials and procedures. For the second condition, because DTempvert with four participating depths should be smaller than that involving only two depths (as in Materials and procedures) due to more averaging, we set threshDTemp_front = 0.5°C, lower than the 1.5°C threshold used in Materials and procedures. In the middle panel of Fig. 13, those yo-yo profiles that meet the conditions of an upwelling front are marked red. The upward arrow on the threshTempGrad = 0.2°C dashed line (in middle panel) indicates the “ ≥ ” condition in Eq. 4. The downward arrow on the threshDTemp_front = 0.5°C dashed line (in upper panel) indicates the “ ≤ ” condition in Eq. 5. The upwelling front is detected around 6 h, the same as in Fig. 11. Using DTempvert involving multiple depths (calculated by Eq. 6), the impact of temperature variation at a single depth on DTempvert is mitigated and is less likely to cause false alarms of upwelling front detection. As a result, the “red profiles” around 7.5 h and 8 h are much fewer than in Fig. 11. Thus robustness of upwelling front detection is improved using the generic vertical temperature homogeneity index. During the May/June 2012 CANON Experiment in Monterey Bay, the vertical temperature homogeneity index (as expressed in Eq. 6, using four depths) was used in the improved upwelling front detection/tracking algorithms run on the Dorado and Tethys AUVs. In three missions over three consecutive days, the Dorado AUV accurately detected and acquired water samples in the upwelling front (as well as in the upwelling and stratified water columns away from the front). The Tethys AUV closely tracked the evolution of an upwelling front over 4 d (Zhang et al. 2012b). Details of those AUV missions will be presented in future publications. References Bellingham, J. G. 1997. New oceanographic uses of autonomous underwater vehicles. Mar. Tech. Soc. J. 31: 34-47. ———, and others. 1997. Haro Strait experiment 1996 MIT sea grant component, Cambridge, CA: MIT Sea Grant AUV Lab. ———, and others. 2000. An arctic basin observational capability using AUVs. Oceanography. 13:64-71 [doi:10.5670/ oceanog.2000.36]. ———, and others. 2010. Efficient propulsion for the tethys long-range autonomous underwater vehicle. Monterey, CA, Proc. IEEE AUV’10. <http://www.ieee.org/organizations/pubs/ newsletters/oes/html/spring11/UnderwaterVehicle.pdf>. Bird, L. E., A. D. Sherman, and J. P. Ryan. 2007. Development of an active, large volume, discrete seawater sampler for autonomous underwater vehicles. Proc. Oceans MTS/IEEE Conference, Vancouver, Canada. Breaker, L. C., and W. W. Broenkow. 1994. The circulation of Monterey Bay and related processes. Oceanogr. Mar. Biol. 32:1-64. Cowles, T. J., R. A. Desiderio, and M.-E. Carr. 1998. Small-scale planktonic structure: persistence and trophic conse950 Zhang et al. AUV sampling across an upwelling front ———, R. S. McEwen, J. P. Ryan, and J. G. Bellingham. 2010b. Design and tests of an adaptive triggering method for capturing peak samples in a thin phytoplankton layer by an autonomous underwater vehicle. IEEE J. Ocean. Eng. 35(4):785-796 [doi:10.1109/JOE.2010.2081031]. ———, M. Godin, J. G. Bellingham, and J. P. Ryan. 2011a. Ocean front detection and tracking by an autonomous underwater vehicle. In MTS/IEEE Oceans’11, Kona, HI. IEEE. ———, and others. 2011b. Classification of water masses and targeted sampling of ocean plankton populations by an autonomous underwater vehicle. In AGU Fall Meeting Abstract OS21A-1609, San Francisco, CA. AGU. ———, M. A. Godin, J. G. Bellingham, and J. P. Ryan. 2012a. Using an autonomous underwater vehicle to track a coastal upwelling front. IEEE J. Ocean. Eng. 37(3):338-347 [doi:10.1109/JOE.2012.2197272]. ———, J. P. Ryan, M. A. Godin, and J. G. Bellingham. 2012b. Observing coastal upwelling front dynamics by AUV tracking, remote sensing, and mooring measurements. In AGU Fall Meeting Abstract OS31H-04, San Francisco, CA. AGU. (Bacillariophyceae) species and domoic acid activity in cultured and natural samples. J. Phycol. 35:1356-1367 [doi:10.1046/j.1529-8817.1999.3561356.x]. ———, and others. 2009. Remote detection of marine microbes, small invertebrates, harmful algae and biotoxins using the Environmental Sample Processor (ESP). Oceanography 22:158-167 [doi:10.5670/oceanog.2009.46]. Sibenac, M., and others. 2002. Modular AUV for routine deep water science operations. In Proc. MTS/IEEE Oceans’02, Biloxi, MS. IEEE. Sullivan, J. M., and others. 2010. Layered organization in the coastal ocean: an introduction to planktonic thin layers and the LOCO project. Cont. Shelf Res. 30(1):1-6 [doi:10.1016/j.csr.2009.09.001]. Thompson, D. R. 2007. AUV operations at MBARI. In Proc. MTS/IEEE Oceans’07, Vancouver, Canada. IEEE. Wright, J. 1995. Seawater: its composition, properties and behavior, 2nd ed. Butterworth-Heinemann. Zhang, Y., and J. S. Willcox. 1997. Current velocity mapping using an AUV-Borne Acoustic Doppler current profiler, p. 31-40. In 10th international symposium on unmanned untethered submersible technology, Durham, NH. AUSI. ———, and others. 2010a. Acquiring peak samples from phytoplankton thin layers and intermediate nepheloid layers by an autonomous underwater vehicle with adaptive triggering. In AGU Fall Meeting Abstract OS51C-1334, San Francisco, CA. AGU. Submitted 3 April 2012 Revised 9 October 2012 Accepted 24 October 2012 951
© Copyright 2026 Paperzz