1. Introduction
The increased availability of commercial off-the-shelf radiometers during the past 20 years has produced a significant increase in the number of marine optical measurements conducted during oceanographic cruises. Spectral radiometric measurements of the marine light field are now routinely collected in numerous studies related to primary productivity, bio-optical modeling, and ocean color remote sensing. The Sea-viewing Wide Field-of-view Sensor (SeaWiFS) calibration and validation plan (Hooker and McClain 2000), for example, relies on radiometric measurements made at sea by a diverse community of investigators. One of the long-standing objectives of the SeaWiFS Project is to produce spectral water-leaving radiances LW(λ) within an uncertainty of 5% (Hooker and Esaias 1993), and the sea-truth measurements are the reference data to which the satellite observations are compared (McClain et al. 1998). The accuracy of the field measurements are, therefore, of crucial importance.
The uncertainties associated with in situ optical measurements have various sources, such as the deployment protocols used in the field, the absolute calibration of the radiometers, the environmental conditions encountered during data collection, the conversion of the light signals to geophysical units in a data processing scheme, and the stability of the radiometers in the harsh environment they are subjected to during transport and use. In recent years, progress has been made in estimating the magnitude of some of these uncertainties and in defining procedures for minimizing them. For the SeaWiFS Project, the first step in the process of controlling sources of uncertainty was to convene a workshop to draft the SeaWiFS Ocean Optics Protocols (SOOP). The SOOP adhere to the Joint Global Ocean Flux Study (JGOFS) sampling procedures (Joint Global Ocean Flux Study 1991) and define the standards for optical measurements to be used in SeaWiFS radiometric validation and algorithm development (Mueller and Austin 1992). The SOOP are periodically updated as deficiencies are identified and outstanding issues are resolved (Mueller and Austin 1995). Examples of incomplete (but converging consensus) protocols are those for turbid water and above-surface measurements (including those made from aircraft).
The follow-on inquiries into controlling uncertainty sources investigated a variety of topics. The SeaWiFS Intercalibration Round-Robin Experiment (SIRREX) activity demonstrated that the uncertainties in the traceability between the spectral irradiance of calibration lamps were approximately 1%, and the intercomparisons of sphere radiance were approximately 1.5% in absolute spectral radiance and 0.3% in stability (Mueller et al. 1996). The SeaWiFS Data Analysis Round-Robin (DARR) showed differences in commonly used data processing methods for determining upwelling radiance and downwelling irradiance immediately below and above the sea surface—Lu(0−, λ) and Ed(0+, λ), respectively—were about 3%–4% of the aggregate mean estimate (Siegel et al. 1995). Hooker and Aiken (1998) made estimates of radiometer stability using the SeaWiFS Quality Monitor (SQM), a portable and stable light source, and showed that the stability of their radiance and irradiance sensors in the field during a 36-day deployment was on average to within 1% (although some channels occasionally performed much worse). More recently, Hooker et al. (1999) conducted an extensive field program to intercompare a variety of above- and in-water methods to quantify differences in the methods and techniques employed for making radiometric measurements in support of SeaWiFS validation.
Each of the activities described above focused on one particular aspect associated with radiometric measurements. The way uncertainties combine or cancel has never been fully assessed, although SIRREX-5 made an initial inquiry into this important topic and defined an experimental plan for addressing the issues involved (Johnson et al. 1999). Similarly, the recommended protocols allow the same radiometric quantities to be measured using different in-water deployment methodologies (above-surface measurements are not considered here), but the way they compare in the field is poorly documented.
As part of the SeaWiFS Project calibration and validation activities, the SeaWiFS field program conducted specific experiments to investigate these issues. The experiments took place during several Atlantic Meridional Transect (AMT) cruises on board the Royal Research Ship James Clark Ross (JCR) between England and the Falkland Islands. The odd-numbered, southbound cruises sampled the boreal autumn and austral spring, while the even-numbered, northbound cruises sampled the boreal spring and austral autumn (Aiken and Hooker 1997). Because of the geographic extent of the transects (more than 100° of latitude and 50° of longitude), the repetitive scheduling of the cruises (two per year, lasting more than 30 days each), the diversity of the environments encountered (oligotrophic gyres to upwelling zones and eutrophic coastal regions), and the use of state-of-the-art radiometers (including calibration monitoring in the field with the SQM), the AMT Program has no equivalent in the oceanographic community.
The AMT optical experiments were designed to compare a variety of deployment techniques used to measure Lu(z, λ) and Ed(z, λ). Mounting the needed light sensors on a frame and deploying the frame with a winch and crane is the most frequently used technique, although tethered, free-fall systems are becoming an increasingly popular method. The irradiance incident at the sea surface is usually collected with a sensor installed on a mast situated above the ship’s deck (often called the reference or deck cell measurement). More recently, floating references, either above or below the air–sea interface, are also frequently being used. Although there are many differences between optical instruments deployed with crane and winch systems versus free-fall units, the primary differences are related to ship-induced perturbations, wave motion, and the time required to perform a cast. Cranes have limited reach, so ship shadow can be a problem, whereas for a free-fall system, the profiler (and possibly the reference) is deployed far from the ship to avoid any ship-induced perturbations to the light field (shadows, reflections, bubbles, etc.).
Because ships are not decoupled from the ocean surface, pitch and roll can increase measurement uncertainties in winch and crane deployments, particularly when a long boom is used. Drifting references are also influenced by surface motion, but engineering solutions and deployment practices can be adopted to reduce this effect. Free-fall profilers are not subject to wave action, but they must be properly trimmed to ensure minimal tilts during descent. Winches and cranes require crew preparation time before operations can commence, and winches have relatively low descent and ascent rates, both of which result in increased time needed to complete a cast and thus increased opportunities for adverse environmental perturbations (e.g., clouds moving in front of the sun). In addition, winches and cranes involve the use of sophisticated machinery that can take a long time to fix if there are any problems. Conversely, the deployment or recovery of a free-fall system generally takes a few minutes and can be conducted with only two people.
Assessing the strengths and weaknesses of these deployment methods in the field using radiometers of different types is not a trivial task. Preliminary to any attempt of identifying the actual source of a possible difference between two sensors, it is necessary to know how the sensors compare with one another under controlled circumstances. In other words, the difference in the absolute response of the sensors when they are exposed to a known light source is required, because this establishes a baseline response for each system that permits the other sources of discrepancies during at-sea experiments to be discerned and quantified. The SQM is designed to perform these comparisons and to monitor the stability of the sensors in the field during the length of a cruise. The latter is particularly important, since an actual difference between two deployment schemes must be resolved from a difference caused by a degradation in the performance of a particular sensor. Of course, an independent evaluation of the stability of the SQM itself is needed, so changes in the SQM emitted flux is not incorrectly ascribed to a sensor’s performance.
The study presented here deals with all the above-mentioned sources of uncertainty. In particular, the following objectives are considered.
Quantify the level of uncertainty of the SQM during AMT cruises.
Measure the stability level of the AMT instruments, so differences in the deployment methodologies can be discerned.
Compare the stability-monitoring capabilities of the SQM (an expensive device) with a time series of in-water and in-air intercomparisons with a second set of radiometers (a potentially inexpensive alternative).
Establish which deployment configuration for a reference buoy that can be floated away from a ship produces the smallest uncertainties.
Compare reference measurements made far away from a ship to measurements made on a mast mounted above the ship’s main deck.
Ascertain if a modular, low-cost (16 bit) profiler is as capable as an integral, high-cost (24 bit) profiler.
Compare the quality of the data from free-fall systems and from a winch and crane system.
2. Instruments and methods
Only the AMT instruments supplied by the SeaWiFS Project are considered here, because their calibration and deployment have always adhered to the recommended protocols, and the SQM has regularly been used to monitor their performance in the field. All of the radiometric systems, including any spares, were manufactured by Satlantic, Inc. (Halifax, Canada). This commonality in equipment was considered the simplest and most cost-effective way to ensure redundacy (exchanging components during failures), intercalibration (very similar center wavelengths), and intercomparison (all of the instruments report at the same rate, are of the same size, and have similar response functions).
The SeaWiFS Project instruments used during the AMT Program included the SeaWiFS Optical Profiling System (SeaOPS), the Low-Cost NASA [National Aeronautics and Space Administration] Environmental Sampling System (LoCNESS), and the SeaWiFS Free-falling Advanced Light Level Sensors (SeaFALLS). SeaOPS is deployed from a winch and crane, whereas the LoCNESS and SeaFALLS are floated away from the ship and deployed by hand. Both the SeaOPS and LoCNESS profilers are modular seven-channel systems;that is, they are built up from (relatively) inexpensive subcomponents that are externally cabled together. SeaFALLS and its associated incident irradiance sensors are integral 13-channel designs that cannot be easily disassembled.
The incident solar irradiance data are provided by three instruments. SeaOPS and LoCNESS are deployed in parallel with a seven-channel in-air irradiance sensor mounted on a mast (a deck cell). SeaFALLS is coupled with the SeaWiFS Square Underwater Reference Frame (SeaSURF), which is composed of an in-water irradiance sensor suspended below a tethered, square floating frame, and with the SeaWiFS Buoyant Optical Surface Sensor (SeaBOSS), which is an in-air irradiance sensor fitted inside a removable buoyant collar, so it can be deployed on a mast (as a deck cell) or as a tethered buoy. Figure 1 shows the at-sea use of the SeaWiFS optical instruments within a simplified radiometric schematic. [See Hooker and McClain (2000) for a more complete discussion.] A brief description of the instruments and the way they are deployed is presented herein.
a. SQM
The engineering design and characteristics of the SQM are described by Shaw et al. (1997) and Johnson et al. (1998), so only a brief description is given here. The SQM is a portable and highly stable light source capable of monitoring the stability of radiometers to within 1% in the field (Hooker and Aiken 1998). Although the results presented here are from one lamp set, the SQM has two lamp sets with different wattage bulbs resulting in three possible flux-level settings. The SQM produces a diffuse and uniform light field and is designed to be flush-mounted to radiance or irradiance sensors with a spectral range from 380 to 900 nm. A kinematically designed D-shaped collar is used on all devices being tested to ensure that they view the same part of the SQM aperture each time they are used. The nonuniformity of the source is less than 2% over a circular area 15 cm in diameter. An internal heater provides operational stability and decreased warm-up intervals.
To account for changes in the emitted flux, three temperature-controlled photodiodes measure the exit aperture light level: one has a responsivity in the blue part of the spectrum, another in the red part of the spectrum, and the third has a broadband or white response. It is important to note that the blue detector is the most sensitive to illumination fluctuations, because the minimum flux is in the blue part of the spectrum [i.e., the signal-to-noise ratio (SNR) is the lowest]. The internal monitor data are used to normalize the flux of the source, so the actual change in the responsivity of the field sensor can be determined.
A change in the responsivity of the field sensor is distinguished from a change in the internal detectors through the use of fiducials. Fiducials are nonoperational devices with the same size and shape as the in situ radiometers. Three fiducials are used: a white one, a black one, and a black one with a glass face. The last mimics the reflectivity of the optical surface of a radiance sensor (the glass is the same as that used with the field radiometers), but the other two are designed to be significantly different. Together, all three provide a wide range of reflectivities. The time series of a fiducial, as measured by the SQM internal monitors, gives an independent measure of the temporal stability of the SQM light field. The reflective surface of a fiducial is carefully maintained, both during its use and when it is being stored. Consequently, the reflective surface remains basically constant over the time period of a field expedition. A field radiometer, by comparison, has a reflective surface that changes episodically as a result of the wear and tear of daily use. This change in reflectivity alters the loading of the radiometer on the SQM and is a source of variance for the monitors inside the SQM, which measure the exit aperture flux level, or the radiometer itself when it is viewing the exit aperture.
The original design usage of the SQM included its use in absolute radiance and irradiance measurements (Johnson et al. 1998). This capability has only recently been evaluated (Hooker et al. 2000), so none of the results presented here include any initial response values when each sensor was calibrated; all of the data are relative, in the sense that the daily deviations in a sensor’s performance are with respect to the mean response for a particular cruise (Hooker and Aiken 1998). The primary reasons for not using the SQM as part of an absolute protocol were the following.
During the time period of this study, there was only one SQM and it was being used extensively in the field or refined in the laboratory, so there was insufficient time to include it in other activities.
The experimental procedures for verifying an absolute use of the SQM were agreed to in principle some time ago, but all of the equipment needed was not available until recently (Hooker et al. 2000).
b. SeaOPS
SeaOPS is composed of an above-water and in-water set of seven-channel light sensors organized within modular subsystems. A conductivity and temperature (CT) probe plus a minifluorometer are also part of the SeaOPS system. The in-water optical sensors are a downward-viewing OCR-200 radiance sensor, which measures Lu(z, λ), and an upward-viewing OCI-200 irradiance sensor, which measures Ed(z, λ). The above-water unit (also an OCI-200 sensor) measures Ed(0+, λ). An in-water and in-air data unit (both Satlantic DATA-100 modules) receives the analog signals from the light sensors and converts them, through a 16-bit, analog-to-digital (A/D) converter, to RS-485 serial communications. The SeaOPS sensors are capable of detecting light over a four-decade range.
A custom-built profiling frame is used to carry SeaOPS. The positioning of the equipment on the frame was developed with a geometry that ensured that the light sensors were in close proximity to one another while preventing the radiance sensor from viewing any part of the support. The narrow geometry of the frame was designed to provide a minimal optical cross section. The field of view of the irradiance sensor was influenced only by the 7-mm winch wire, and careful attention was paid to the balance of the frame, even though SeaOPS had tilt and roll sensors. At the start of each cruise, the frame was trimmed with lead weights in air, accounting for the in-water weights of the sensors. Final trim checks were carried out in situ during the first (test) station. The typical lowering and raising speed of the winch was approximately 20–25 cm s−1. For most stations, the sun was kept on the starboard side except during adverse weather conditions. The crane used had about a 10-m reach over the starboard side of the ship. During AMT cruises, the SeaOPS reference was always mounted at the top of a mast sited on top of the port stern gantry mast. The mast was high enough to ensure that none of the ship’s superstructure shaded the irradiance sensor under almost all illumination conditions.
c. SeaFALLS
SeaFALLS, SeaSURF, and SeaBOSS are integral designs all equipped with (relatively expensive) 13-channel OCI-1000 and OCR-1000 radiometers, which employ 24-bit, A/D converters and gain switching. Each system is capable of detecting light over a seven-decade range. SeaFALLS measures Ed(z, λ) and Lu(z, λ) as it falls through the water column. It is based on a SeaWiFS Profiling Multichannel Radiometer (SPMR), which was modified so that it could be used with the SQM without any disassembly. SeaSURF is based on a SeaWiFS Multichannel Surface Reference (SMSR), an in-water irradiance sensor designed to be deployed a fixed distance (z0 ≈ 30 cm) below the surface, and it measures Ed(z0, λ). SeaBOSS is an in-air version of the SMSR and measures Ed(0+, λ). All of the instruments receive power and send data via a tethered cable. Like SeaOPS, SeaFALLS is fitted with a CT probe and a minifluorometer.
SeaFALLS is deployed from the stern of the vessel, and whenever possible, the ship maintains a headway speed of approximately 0.5 kt. The profiling instrument is carefully lowered into the water and slowly released at the surface until it has drifted clear of any possible shadowing effect. When the profiler reaches the desired distance from the stern (usually 30–50 m), it is ready for deployment. A continuing effort is made to prevent the telemetry cable from ever coming under tension, because even brief periods of tension on the cable can adversely affect the vertical orientation (tilt) and velocity of the profiler. To ensure this does not occur, the operator always leaves a few coils of the neutrally buoyant cable at the surface. A tangle-free and continuous feed of cable into the water is also needed, so all of the cable (approximately 300 m) is laid out or flaked on deck prior to each deployment in such a manner as to minimize any entanglements. The profiler descends at approximately 1 m s−1, so a relatively deep cast can be acquired very quickly (less than 3 min for a 150-m cast).
SeaBOSS can be mounted on a mast or deployed as a drifting buoy, whereas SeaSURF is always floated away from the boat using a buoyant frame. Any floating references to be deployed at the same time as SeaFALLS are deployed carefully to prevent damage against the ship and, in the case of the above-water irradiance sensors, to keep their cosine collectors dry. In addition, they are held at no less than 30 m behind the vessel until SeaFALLS is in the correct position for deployment. When the drop command is given, both instruments are released in unison. The reference cable is spooled out freely so that minimal tension is placed on the cable, which in turn minimizes reference tilts. When SeaFALLS reaches the point of maximum descent (usually the 1% light level), both instruments are pulled back to their original positions and are ready to be redeployed. The profiler and both references can be deployed quickly with only two people, so the ship can be stopped when light conditions are optimal. More importantly, because SeaFALLS casts do not last long, they can be timed to coincide with clouds moving clear of the sun.
d. LoCNESS
LoCNESS is not a new instrument per se, but instead is built up from the SeaOPS modular components or from spares (Fig. 2). Once assembled, LoCNESS is a free-falling unit that looks and functions very similarly to SeaFALLS, and it is deployed in the same fashion. The deck cell data were usually the same as for SeaOPS, although a separate (spare) reference sensor was used on some occasions. The principal advantage of LoCNESS is its cost and flexibility. It can be assembled from (relatively) inexpensive components (in comparison to SeaFALLS), and it can be quickly reconfigured, because the radiometers used are not integral to the design. For example, rather than measure Ed and Lu, a spare OCI-200 can be used in place of the OCR-200 radiometer, and LoCNESS can measure Ed and Eu. In addition, a special adapter plate can be used on the nose of the profiler that allows for the mounting of two sensors rather than one: the usual Lu sensor plus an Eu sensor, for example. The additional sensor does not disturb the stability of the profiler during descent. In fact, the THOR (Three-Headed Optical Recorder) option has produced the lowest and most stable tilts (less than 1° on average) of all of the profilers used on AMT cruises.
As shown in Table 1, all of the series 200 radiometers (SeaOPS and LoCNESS) took measurements in the same seven spectral bands that were selected to support SeaWiFS calibration and validation activities (McClain et al. 1992). In comparison, the series 1000 radiometers (SeaFALLS, SeaSURF, and SeaBOSS) cover the SeaWiFS bands plus other parts of the spectrum in greater detail (Table 2). The spectral overlap of the sensors and the simultaneous profiles facilitate at-sea intercomparisons of the instruments. For the analyses presented here, only the SeaWiFS bands are considered, because they are common to all of the instruments.
e. Data acquisition
For all of the profilers, the RS-485 signals from the in-air and in-water subsystems are combined in a Satlantic deck box and converted to RS-232 communications for computer logging. Although the instrument manufacturer provides data acquisition software, it is not suitable for controlling several instruments simultaneously with only one operator, nor is it well suited for calibration monitoring requirements. Consequently, a joint effort was undertaken by the University of Miami Rosenstiel School for Marine and Atmospheric Science (RSMAS) and the SeaWiFS Project to produce a new data acquisition package for all of the AMT optical instruments.
The acquisition software for SeaOPS and LoCNESS is called Combined Operations (C-OPS), and it controls both the in-air and in-water data streams. The primary task of the software is to integrate the RS-232 output from the deck box that handles the power and telemetry to the underwater instruments and to control the logging and display of these data streams as a function of the data collection activity being undertaken: dark data (caps on the radiometers), up cast, down cast, constant depth or soak events, calibration monitoring, etc. All of the telemetry channels are displayed in real time, and the operator can select from a variety of plotting options to visualize the data being collected. Many of the data streams have automated error checking protocols, and visual and audible alerts are used to notify the operator in the event of unacceptable performance (e.g., high noise during dark data collection, high internal temperature during deployment, etc.).
File naming is handled automatically, so all the operator has to do is choose which data streams to record and then select the execution mode of the data collection activity. The files are written in ASCII, and each tab-delimited file has a header structure that identifies what is recorded in each column. All data records are time stamped. The acquisition software for SeaFALLS functions are very similar to C-OPS and is called C-FALLS. The primary difference is that C-FALLS can record two reference data streams (SeaSURF and SeaBOSS) simultaneously.
Data logging for the SQM involves two computer systems: one for the device under test (DUT) and one for the SQM. With the SQM control software, one computer controls two highly regulated power supplies and acquires five other signals from a multiplexed digital voltmeter (DVM): the three photodiode voltages from inside the SQM and the voltages across two precision 1-Ω shunt resistors. All of this information is time stamped and written into a tab-delimited ASCII file. The power supplies and the DVM are controlled over a general purpose interface bus (GPIB), and the program acquiring the DVM signals converts the voltage values across the shunts to current, and adjusts the output of the power supplies to ensure a constant current supply to the lamps.
3. Results
To exploit the passage of the JCR, the AMT Program employs three sampling strategies: (a) continuous near-surface (7 m) measurements from pumped seawater, (b) towed measurements (5–80 m) using the Undulating Oceanographic Recorder (UOR), and (c) daily station measurements of the upper ocean (200 m) made close to local solar noon and then again in the midafternoon. Although bio-optical measurements are conducted during all three sampling opportunities, the ones of interest for this study are the in situ optical measurements made during the latter. In particular, this includes in-water measurements of Ed(z, λ) and Lu(z, λ), plus in-air and in-water measurements of solar irradiance, Ed(0+, λ) and Ed(z0, λ), respectively. The experiments described in the following sections were conducted during several cruises and, generally, in case-1 waters (Morel and Prieur 1977), with chlorophyll a concentrations ranging between 0.03 and 8.00 mg m−3.
Starting with AMT-5 (Aiken et al. 1998), and in parallel with the acquisition of data for SeaWiFS validation, specific experiments were executed to intercompare the data collected by the different radiometers. A concerted effort was made to deploy two or more systems simultaneously as frequently as possible. The beginning of the cast of all instruments was synchronized using handheld radios, so all sensors experienced very similar illumination conditions in the first 10–20 m of their casts. The SeaOPS (winched) deployments took the longest to complete (typically 15–20 min for the downcasts and upcasts), so it was usually possible to make several casts with the free-fall instruments during the course of a single SeaOPS cast. These experiments permit the comparison of winched (SeaOPS) versus free-fall (SeaFALLS and LoCNESS) systems and the intercomparison of the two free-fall profilers. The former is a way to assess the influence of ship shadow during the winched measurements, and the latter determines if a lower-cost profiler (LoCNESS) is an acceptable alternative to a higher-cost profiler (SeaFALLS).
Another set of deployments using SeaOPS was also conducted taking advantage that the SeaOPS frame was modified so it could collect data from three light sensors. Two of the ports were occupied by a constant pair of Lu and Ed sensors, while another Lu or Ed sensor was connected to the third port to intercompare with the routine ones. With these deployments, the sensors to be compared were almost exactly at the same depth and very close to one another (Fig. 2), so they experienced virtually identical environmental conditions. The downcast was used to produce a continuous profile for the SeaWiFS validation dataset. During the upcast, the profile was stopped at discrete depths (usually five) for soak periods 1–3 min long. The soak measurements compose the intercomparison dataset. They were averaged to limit the influence of experimental perturbations caused by waves and ship motion. In addition, sea- and sky-state digital photographs were taken at the bottom of the SeaOPS downcast or in the middle of a sequence of free-fall profiler deployments.
a. Calibration monitoring
Although the SQM lamp ring was changed after it was commissioned during AMT-3, no lamp changes were made after the start of AMT-4, and up through and including AMT-7. Figure 3 shows the percent deviation of the internal SQM blue detector measurements of the glass fiducial with respect to their individual (cruise) mean values during the AMT-4 through AMT-7 cruises (Figs. 3a–d, respectively). The data in Fig. 3 were from the first lamp set of the lamp ring (which contains two lamp sets). The SQM lamp ring was aged for 48 h before it was used at the start of AMT-4, but Fig. 3a clearly shows an exponential decay in the emitted light during the first few days of AMT-4. Nonetheless, the AMT-4 data indicate that the SQM light flux had a stability to within ±0.79%, and the data after the first week show a stability to within ±0.28%.
The stability and behavior of the SQM during AMT-5 were very similar to the SQM’s performance on AMT-3 when it was first commissioned for field use (Hooker and Aiken 1998): the data indicate a stepwise change in the SQM flux level halfway through the cruise. All three detectors show the change, and if the three detector signals are averaged together, the emitted flux of the SQM decreased by approximately 0.87%. The change in flux was due to a partial short in one of the bulbs, which resulted in a 1.2% decrease in the operating voltage of the lamp. The stability of the SQM during the periods before and after the change in light output, as estimated by one standard deviation (1σ) in the average of the three internal monitor signals, was to within 0.60% and 0.53%, respectively.
During AMT-6, the 1σ values of the red, blue, and white detectors while measuring the glass fiducial were 0.36%, 0.46%, and 0.39%, respectively. The performance of the SQM during AMT-6 was the best out of all the cruises. No lamp anomalies were experienced, and the standard deviation in the emitted flux was the lowest ever recorded in the field. The AMT-7 data show a stepwise change halfway through the cruise, as was seen during AMT-3 and AMT-5. Although the stability for the entire cruise was very good, to within ±0.43% as measured by the blue detector, the stability improves to ±0.38% and ±0.28% if the cruise is split into a first and second half, respectively.
The internal analysis of SQM stability is corroborated by considering the data from R035, one of the field radiometers used during all of the SQM deployments. The peak in the responsivity of the blue internal detector is very close to 443 nm, so a time series of this channel, shown as the plus signs in Fig. 3, is an independent measure of the stability of the SQM during AMT-4 through AMT-7. The average percent deviation with respect to the mean for this channel was 0.77%, 0.97%, 0.71%, and 0.92% for AMT-4 through AMT-7, respectively. The same values derived from the internal blue detector analysis were 0.79%, 0.80%, 0.46%, and 0.43%, respectively. The field radiometer data also show the stepwise changes observed during AMT-3, AMT-5, and AMT-7.
b. Instrument stability
Although the agreement of the 443-nm R035 channel with the SQM internal detector establishes the stability of the R035 radiometer, it is but one radiometric channel out of many. A summary of the stability of the in situ radiometers during SQM field sessions, as measured by the percent deviation with respect to the mean behavior of each channel over the individual cruise time periods, is shown in Table 3. The data confirm the general behavior originally reported during the AMT-3 cruise (Hooker and Aiken 1998): the radiance sensors are more stable than the irradiance sensors, and the least stable channels are usually the blue irradiance channels. The averages of the most and least stable OCR-200 radiance channels (i.e., the averages of the smallest and largest percent deviations) are 0.11% and 0.51%, respectively. The same values for the OCI-200 channels (which include the OCI-200 references) are 0.39% and 1.90%, respectively. The OCR-1000 and OCI-1000 instruments show similar performance, although, the OCR-1000 sensors perform worse than the OCR-200 instruments. The averages of the most and least stable OCR-1000 radiance channels are 0.26% and 0.79%, respectively. The same values for the OCI-1000 channels (including the references) are 0.34% and 1.93%, respectively.
The greater instability of the irradiance sensors is expected because (a) the cosine collectors reduce the emitted flux, so the blue portion of the spectrum is comparatively lower; (b) irradiance sensors have inherently higher noise (from the higher gain resistors in the detector circuits); and (c) the greater sensitivity of irradiance sensors to positioning errors. The greater sensitivity of the OCI-1000 sensors, which have 24-bit A/D units, does not change this basic conclusion, although the in-water (higher gain) OCI-1000 sensors do not show as much sensitivity to the blue part of the spectrum.
c. Sensor intercomparisons
In the following comparisons, only the first five SeaWiFS wavelengths (412, 443, 490, 510, and 555 nm, respectively) are considered for in situ intercomparison. The rationale for this is as follows: (a) the blue and green channels are the most important to the algorithm validation process; (b) the SNR for wavelengths above 600 nm is relatively low, so data at these wavelengths are noisier; and (c) high absorption in the red part of the spectrum means that derivation of surface values—for example, Lu(0−, λ)—need very specific extrapolation ranges (shallower and narrower than for shorter wavelengths), which are not accounted for in the data processing scheme.
1) In-water comparisons
For the experiments when SeaOPS was equipped with three sensors (section 2f), R037 (Lu), and I095 (Ed) were always used for routine profile measurements, while the remaining (spare) port was occupied by either R068 (Lu) or I100 (Ed). Figure 4a shows the overall relationship for the first five SeaWiFS wavelengths between the Lu sensors (R037 versus R068) for more than 80 simultaneous collection events at various depths. The level of agreement, as determined by the slope of the reduced major axis linear regression (Ricker 1973; Press and Teukolsky 1992) line (m) and the coefficient of determination (R2), is very good, with a difference of approximately 1.1% (m = 0.989; R2 = 0.990; and N = 451).
The individual Lu channels show a wider range of disagreement: 4.4%, 2.5%, 5.6%, 24.9%, and 3.3% for 412, 443, 490, 510, and 555 nm, respectively. It is important to note, however, that with the exception of the 510-nm channel, the individual channel data are well distributed with respect to the 1:1 line; that is, there is little evidence of an overall deterministic difference or bias between the two sensors. The overall Ed(z) comparison (Fig. 4b) is also very good, with a difference of approximately 2.7% (m = 1.027, R2 = 0.999, and N = 404), but the significant clustering of the data above the 1:1 line indicates that the two instruments collected deterministically different data. A comparison of the absolute responses of these two irradiance sensors, as determined by the SQM data, shows I095 measured on average about 2.1% lower than I100 (for the first five SeaWiFS wavelengths), which accounts for almost all of the difference between these two sensors. The Ed values show better agreement at individual wavelengths than the Lu sensors: within 1% in the best case (443 nm) and within 3.4% in the worst (665 nm).
The results presented above were derived from data collected during the upcasts when the SeaOPS frame was stopped for 1–3 min at discrete depths. The data from the continuous downcasts of the same experiments can also be used to assess how data processing, particularly extrapolations of Lu(z, λ) to the surface, affect the comparisons of surface quantities like LW(λ) or remote sensing reflectance, Rrs(λ). The term LW(λ) is defined as the upwelling light flux just above the sea surface in the zenith direction. This parameter can be obtained by extrapolating the Lu(z, λ) profile to the surface and then accounting for the internal reflection during transmission through the air–sea interface. Such extrapolations are performed over homogeneous shallow portions of the water column, generally between 1 and 2 optical depths thick, chosen (by visual examination of the radiometric, hydrographic, and fluorescence profiles) not to include noisy data close to the surface.
The LW(λ) values derived from the R037 and R068 profiles show trends similar to those observed from the soak measurements, and the slopes agree to within 1% (Fig. 4c). It is worth noting that because these two sets of comparisons are not exactly of the same kind [compared to the soak comparisons, the LW(λ) comparisons are for a unique depth at a high energy level and have a lower dynamic range], the way errors propagate with data processing cannot be quantified exactly. It is reasonable, however, to assume that in the example presented above, the uncertainties in LW(λ) introduced by data processing are on the order of 2%.
2) In-air comparisons
During AMT-7, two OCI-200 sensors (M030 and M035) were deployed as deck cells for SeaOPS. Both were sited on the starboard, stern gantry mast. The duplicate reference data allows for an intercomparison of two in-air sensors under almost identical illumination conditions. Figure 5 shows the intercomparison of the two references for more than 82 simultaneous collection events, which occurred during the in-water intercomparison profiles summarized in Fig. 4. The agreement between the sensors, as determined by m and R2, is very good, with an overall difference of approximately 1.1% (m = 0.989, R2 = 0.979, and N = 410). This is better agreement than the SQM analysis would predict, because the latter showed that irradiance sensors have an average agreement of approximately 1.9%.
The explanation lies with the results for each channel, which show that the individual agreement is more variable but fortuitously grouped around the 1:1 line. The disagreement for the 412-, 443-, 490-, 510-, and 555-nm channels are 7.1%, 2.7%, 3.8%, 0.0%, and 2.6%, respectively. A recurring bias is the M035 sensor returned higher values than the M030 sensor for all channels except the 490-nm channel. More importantly, all of the R2 values for the regression analyses are equal to or greater than 0.995. The large percentage of variance explained implies the differences are deterministic:that is, the result of a systematic uncertainty or bias. The two sensors were mounted on the same mast, so there is no reason to believe the sensors did not experience identical illumination conditions. The SQM data shows the M035 measurements were, on average, 2.7% higher than the M030 measurements, with individual channel differences of 5.1%, 1.9%, 3.9%, 0.6%, and 1.9%, respectively, which indicates that the most likely explanation for the differences in the two sensors was a difference in calibration.
In order to compare in-air reference data from OCI-200 and OCI-1000 sensors, simultaneous data from the LoCNESS deck cell (M030) and SeaBOSS (N046) mounted on a mast were collected during AMT-5 and AMT-6 free-fall profiles. Figure 6a shows that the two sensors agreed to within 2.6% (m = 1.026; R2 = 0.987;and N = 405). Again, compared with the 1.9% agreement in the SQM for irradiance sensors, this is very good agreement. Once again, individual channel differences are apparent and are well explained by the percent differences in the SQM data. Overall, the M030 sensor measured 2.1% greater than N046, but some channels, like the 510- and 555-nm channels, were as much as 5.0% greater and 0.9% lower, respectively. These differences in calibration account for the majority of the individual variances in the in situ intercomparisons.
Some of the reference discrepancies resulted from N046 measuring systematically lower irradiances (compared to M030) for some of the experiments at the end of the AMT-5 cruise. The reason for these random lower measurements is unclear. During SQM sessions, the absolute response of many of the sensors displayed sudden stepwise increases or decreases of as much as a few percent with respect to the running mean. These anomalies took place during the power-up process and were seen to remain for the duration of the session involved. That is, if the sensor powered up with a higher or lower system response, the system remained anomalously higher or lower for the duration of the SQM session. Several possible causes for the steps were investigated (e.g., excessive heat in the radiometers, illumination geometry, and gain switch), but none were correlated with the anomalies.
A detailed analysis of this problem was conducted with the AMT-5 dataset in an effort to provide the instrument manufacturer with a quantification of the problem. For the sensors used during AMT-5, the absolute value of the startup steps between two successive SQM sessions were estimated to typically range from 0.1% to 2.7%, with an occasional step as large as 7.2%. The average of the absolute value of all the steps was approximately 0.9%. Although all instruments exhibited some form of this behavior, the largest steps were associated with the 24-bit systems. Interestingly, the average steps for the 24-bit systems was the same as the 16-bit systems (a little less than 1%), because of the greater sensitivity in the blue part of the spectrum.
Two aspects of the phenomenon that should not be overlooked are the following: (a) the steps are directional—that is, the step can be above or below the baseline behavior—and (b) the steps usually occur randomly, so that they have a tendency to cancel out during the course of a cruise under most circumstances. This was not true for all sensors, and the 24-bit systems in particular were seen to go into modes wherein the steps were in a preferred direction with respect to the baseline for extended periods of time, thereby producing a bias in sensor performance. In such cases, the instrument was worked on by the manufacturer after the cruise, and the replacement of electrical subcomponents returned the instrument to a satisfactory level of performance. In the final analysis, the manufacturer has provided no explanation for the startup steps, and they were seen in all of the cruise data.
3) Reference deployments analysis
In ocean color studies, the incident irradiance data are most frequently collected from in-air sensors mounted on a deck mast, but well clear of the ship’s superstructure, or from in-water measurements extrapolated to above the surface (Mueller and Austin 1995). Getting the reference data from an instrument located a few tens of centimeters below the surface and oriented toward zenith is a recent alternative (Waters et al. 1990). This kind of deployment allows possible ship influence to be eliminated from the measurements, because the sensor can be installed on a frame and floated far away from the ship (Fig. 1), but several new factors must be accounted for: increased sensor tilts from wave motion, light focusing effects from waves, and the thickness and content of water above the sensor.
In addition to the deck cell comparisons (which are discussed in the preceding subsection), several experiments were conducted during AMT-5 to compare the benefits and weaknesses of different types of reference deployment methods. SeaBOSS was used to determine how best to float an in-air sensor away from a ship while keeping it dry and minimizing tilts. Four kinds of reference data are considered here:
an in-air deck cell reference mounted on a mast and sited as clear of the ship’s superstructure as possible (SeaOPS),
an in-air reference equipped with a flotation collar (SeaBOSS),
an in-air reference equipped with a flotation collar and a buoyant stabilization frame with elastic isolation cords fitted between the frame and the body of the irradiance sensor (SeaBOSS modified), and
an in-water reference equipped with a buoyant stabilization frame (SeaSURF).
SeaBOSS in-air buoy measurements (N046), with no stabilization frame, were compared to simultaneous SeaOPS deck cell (M030) measurements for 42 collection events (Fig. 6b). The overall difference between the references was approximately 6.4% (m = 0.936, R2 = 0.983, and N = 210). Although the variability of the measurements was higher for the buoy references because of wave motion, the mean measurements from the buoy were consistent with the deck cell measurements. When SeaBOSS was equipped with a stabilization frame, the buoy tilts were reduced, resulting in a lower standard deviation even though the sea state was rougher (average, minimum, and maximum angles measured with the two configurations cannot be directly compared because the instruments were deployed on different days with different sea states). No significant change in agreement between SeaBOSS and the SeaOPS reference, however, was observed.
A detailed example of the differences in reference deployments is available from a case study involving three different references. Figures 7a–c show simultaneous time series from the SeaOPS deck cell (M030), SeaBOSS as a buoy with no stabilization frame (N046), and SeaSURF (H045), respectively. Figure 7d shows the mean spectrum for each type of sensor and the mean extraterrestrial solar irradiance F0(λ) given by Neckel and Labs (1984). The amount of noise or variability in the data increases from the deck cell to the buoy and then to the underwater reference, with coefficients of variation (standard deviation divided by the mean) equal to 1.0%, 2.4%, and 8.2%, respectively. The increase in variance between the deck cell and the buoy is due to the enhanced wave motion experienced by the buoy. The large increase in variance between SeaBOSS and SeaSURF is due to the light-focusing effect created by surface waves, which occasionally results in Ed(z0, λ) values above F0(λ).
A channel-by-channel comparison between the in-water and in-air references (Fig. 7d) reveals that (a) the differences between the two in-air sensors (M030 and N046) are consistent with the observed differences when these two instruments were mounted on a mast together (Fig. 6), and (b) the submerged instrument (H045) tended to measure higher levels of light at the shorter wavelengths and lower values at the longer wavelengths. This latter observation is explained by the strong absorption above 560 nm caused by the (approximately) 30 cm of water between the surface and the cosine collectors of the radiometer. The influence of the intervening water between the sensor and the surface on the spectral shape of the reference spectrum is also dependent on the concentration of optically active components (i.e., phytoplankton, dissolved organic matter, nonliving particulates, etc.). Because normalization of radiometric quantities generally requires the use of incident irradiance above the surface, this implies that sophisticated corrections or filtering procedures need to be applied to the submerged reference measurements.
High-frequency light flashes caused by surface effects may be in part responsible for the higher values observed at short wavelengths for in-water reference. The discrepancy at 412 nm between the in-water reference unit (H045) and the in-air sensors (M030 and N046), however, is too large (approximately 14%) to be explained as a wavelength-dependent biasing from wave-focusing effects alone. If that were the case, it seems highly likely that the 443-nm deviation would be similar in magnitude, even assuming slightly greater absorption in the chlorophyll peak wavelengths. An error in the responsivity calibration of the H045 412-nm channel is the most likely explanation, but a recalibration of the instrument after AMT-5 showed only a 2.9% change in the 412-nm channel (the change would have lowered the irradiance values and reduced the difference in Fig. 7d). A comparison of the three sensors during SQM sessions, however, showed the flux levels measured by M030 and N046 agreed to within 2% (on average), whereas H045 measured the same flux level approximately 8% higher. If this finding is combined with the postcruise calibration change, almost all of the difference at 412 nm can be explained, although an inexplicable instrument problem remains: the responsivity of the instrument in the field was significantly different than its established responsivity during calibration.
The spectral shapes of the three references departed somewhat from that of the mean extraterrestrial solar irradiance spectrum. All three reference sensors showed a maximum at 510 nm, but such a maximum would be expected at 490 nm in clear-sky conditions, or at least the irradiance levels at both wavelengths should be closer in value. The SQM data showed that the 490-nm channel of the N046 sensor measured anomalously low during the cruise, but for this instrument the relative values of the wavelengths above 500 nm agreed reasonably well with the mean extraterrestrial solar spectrum; the difference between 510 and 555 nm in M030 and H045 seemed high. Again, these are important issues, because reference measurements are used to normalize a variety of quantities in ocean color studies.
4) Profiler data analysis
While the preceding subsection describes results from in-water sensor comparisons with several sensors mounted on the same frame and using the same DATA-100, this section documents comparisons in which several instruments were deployed separately, but at the same time. During AMT-5 and subsequent cruises, a concerted effort was made to collect simultaneous casts with two, and sometimes three, profiling radiometers. The main reason for this redundancy was twofold: (a) to be able to resolve whether differences between the in situ validation data and the remote sensing data were genuine (if both in situ measurements agreed, the remote sensing data were considered suspect), and (b) to assess the performance of the different sensor systems with respect to one another.
The in-water radiance and irradiance measurements are closely dependent on the incident solar irradiance. So it is better for the comparisons to be based on (a) quantities that are minimally affected by variations in the incident light field, or (b) measurements that are made very close together in time. Given these criteria and the emphasis in the AMT Program on synchronized sampling, the comparisons can be made on water-leaving radiances and diffuse attenuation coefficients.
The diffuse attenuation coefficient results from absorption and backscattering within the water column, and it quantifies the spectral decrease in light energy as a function of depth. For the results presented here, it was derived for the first five SeaWiFS wavelengths by computing the slope of the regression between depth and log-transformed profiles (natural logs) of Lu(z, λ) or Ed(z, λ) for the depth range used to extrapolate Lu(z, λ) values to the surface. Because the derivation of the best possible LW(λ) values is one of the priorities of the data processing scheme, the depth range used to extrapolate radiometric data is mostly based on the shape of the Lu(z, λ) profiles. This depth range is not always adequate for extrapolating Ed(z, λ) values or for computing Kd(λ) values, because irradiance measurements are generally subject to deeper and more intense perturbations (particularly from wave focusing effects) than upwelled radiance data. Consequently, Ku(λ) is used here to compare measurements from the various radiometers.
Under most circumstances, and in the absence of experimental or environmental perturbations, Kd(λ) and Ku(λ) are very similar, because the downwelled and upwelled light fluxes are influenced by the same optically active components. This no longer holds true when one set of measurements is influenced by photons that can be considered as coming from a secondary source (i.e., from a source other than the sun): for example, fluorescence or Raman scattering. These perturbations are more frequently seen in Lu(λ) measurements, because they represent a significant contribution to the measured signal, while they are masked by the (comparatively) higher ambient light level in shallow Ed(λ) measurements. Fluorescence and Raman scattering mostly influence long wavelengths (greater than 555 nm), so the data presented here are minimally affected by them.
(i) Free-fall intercomparisons
Figure 8 shows the intercomparison of LW(λ) (Fig. 8a) and Ku(λ) values (Fig. 8b) between the two free-fall instruments LoCNESS (R036) and SeaFALLS (Q016). The data are from 81 simultaneous profiles collected during the AMT-5 and AMT-6 cruises. The overall agreement between the two profilers is very good, with approximately a 3% difference in LW(λ) values (m = 1.032, R2 = 0.980, and N = 405). The LoCNESS LW(λ) estimates are almost always above the 1:1 line, which implies that there is a deterministic difference between the two instruments. A comparison of the LoCNESS and SeaFALLS radiometers with the other radiometers used with the SQM showed that the SeaFALLS values were on average 1.8% lower than the other radiance sensors. An important point to remember is that because of the relatively low SQM flux level, the SQM analysis is valid only for the high gain setting of SeaFALLS, while the measurements involved in the derivation of the LW(λ) values often include data acquired at the low gain setting.
Examination of the individual results for each channel shows the agreement between both free-fall instruments for the first three SeaWiFS channels is 1.8%, 4.6%, and 3.8%, at 412, 443, and 490 nm, respectively. The slope of the regression for the 510-nm channel comparison is considerably worse, m = 1.260. The weak agreement between the two sets is primarily derived from the AMT-6 data, where SeaFALLS yielded LW(510) values lower than LoCNESS. For the 555-nm channel regression, m = 1.136. This is mostly due to six casts from two different stations conducted in waters dominated by Synecchococcus. The reason why these stations are not sensed the same way by the two instruments is not clear. Small-scale variability (within a few meters horizontally and vertically) may be one explanation, and there is some evidence for this in the fluorometer data. If these stations are removed from the analysis, the agreement for 555 nm is closer to 7%.
The overall agreement for Ku(λ) values is within approximately 1% (m = 1.012, R2 = 0.980, and N = 405). All of the first three SeaWiFS wavelengths agree to within less than 2%, while the agreement is about 3% and 6% for the 510- and 555-nm channels, respectively. (If the Synecchococcus stations are removed, the overall agreement is within 3.5%, and all wavelengths agree to within 3% except 412 nm, which is about 6%.)
(ii) Winched and free-fall comparisons
Figure 9 shows the intercomparison of LW(λ) (Fig. 9a) and Ku(λ) values (Fig. 9b) between the winched system (SeaOPS) and the two free-fall instruments (LoCNESS and SeaFALLS) from a total of 48 simultaneous casts collected during AMT-5 and AMT-7. Differences between SeaOPS and the free-fall instruments are higher than between the two free-fall sensors, with an 8.7% difference for LW(λ) and a 2.8% difference for Ku(λ). (The R2 values are slightly higher than in the free-fall comparison, but this is primarily due to the higher dynamic range in light levels observed in the winched versus free-fall comparisons.)
The overall LW(λ) values derived from SeaOPS data were persistently lower (as compared to the 1:1 line) than those from the free-fall instruments (m = 1.087, R2 = 0.986, and N = 240). This observation holds true at all wavelengths except 510 nm, where the free-fall systems measured lower LW values. Inconsistencies between the 510 channel of various systems were recurrently noted in the AMT experiments. In most cases, the sensors agreed at the illumination level used during calibration. The 510-nm regressions usually crossed the 1:1 line very close to the calibration point, and the largest differences were at the highest illumination levels (above the flux level from the single lamp set used with the SQM). Numerous possible causes for these differences were investigated, but no satisfying answer was discerned; the manufacturer continues to investigate this issue.
Although several factors influence the comparison (calibrations, environmental variability, data processing, etc.), the consistently lower values observed for SeaOPS were most probably caused by ship shadow (Gordon 1985; Voss et al. 1986, and Helliwell et al. 1990). The fact that the difference between SeaOPS and the free-fall systems decrease with increasing wavelengths (slopes decrease from 1.107 at 412 nm to 1.040 at 555 nm) reinforces this assumption (Weir et al. 1995). Even though the SeaOPS frame was deployed using a 10-m-long crane, the data suggest that the measurements were affected by ship shadow, probably because of the large size of the ship (99.04 m long and 18.85 m wide). As illustrated by Weir et al. (1995), upwelling radiances, and consequently LW(λ), are more affected by ship shadow than the other apparent optical properties (e.g., irradiance). As explained earlier, the SQM analyses cannot fully confirm these trends, because the flux level in the SQM and in the subsurface measurements are different. However, calibration differences should not account for more than 1%–3% of the variability observed in the SeaOPS versus free-falls comparisons.
The comparison of the Ku(λ) data (Fig. 9b) shows little differences between the two sets of measurements, with an overall slope equal to 0.978, while among wavelengths the slope ranges between 1.001 and 0.970 (except, again, for the 510-nm channel). In earlier ship shadow experiments presented by Voss et al. (1986) and Weir et al. (1995), only small differences were also observed in Kd. The fact that the Ku comparison does not show obvious ship shadow effect suggests that the shadow of the ship has the same impact over the depth range used to calculate the diffuse attenuation coefficient (typically, 3–15 m); otherwise, the Ku values for SeaOPS would be lower than those derived from the free-fall instruments.
4. Discussion and conclusions
The primary objective of this study was to use the AMT dataset to quantify the uncertainties associated with many aspects of optical data collection that have not been completely addressed by previous investigations. In particular, a concerted effort was made to resolve differences between deployment schemes and those caused by a degraded performance or calibration of a particular sensor. The latter involved an extensive time series of calibration monitoring using the SQM. Seven different inquiries were considered.
1) Quantify the level of uncertainty of the SQM during AMT cruises. An analysis of the complete AMT dataset for the SQM (AMT-3 through AMT-7, inclusive) confirms the behavior originally reported for AMT-3 (Hooker and Aiken 1998): the stability of the emitted flux is to within approximately 1% during the course of a 30-day deployment. The SQM is probably the most unique optical instrument deployed on the AMT cruises. In the absence of a portable illumination source, such as the SQM, scientists deploying radiometers to the field rely on calibration data obtained, at best, before and after field campaigns, and under most circumstances, they rely on annual or biannual calibrations. With this kind of calibration scenario, changes in instrument performance are not monitored during a deployment; a linear fit is applied to the calibrations before and after the cruise to estimate any changes in the field. As shown by Hooker and Aiken (1998), this is frequently an inadequate assumption. More importantly, as was shown with the SeaSURF instrument, the responsivity of an instrument in the field can be significantly different than its responsivity as established during calibration in the laboratory. Without a portable source, this difference could not be detected or quantified.
2) Measure the stability level of the AMT instruments, so differences in the deployment methods can be discerned. The OCR-200 series instruments have an overall maximum measurement uncertainty of approximately 0.5%. The OCR-1000 instruments are less stable, with an average maximum uncertainty of approximately 0.8%. Both types of irradiance sensors, OCI-200 and OCI-1000, have a similar overall maximum uncertainty of approximately 1.9%. The former are more stable, but the low amount of blue light emitted by the SQM results in higher uncertainties; the latter are less stable, but the higher sensitivity (24-bit A/D and gain switching) compensates for the low flux in the blue part of the spectrum. A channel-by-channel comparison of the two types of irradiance sensors shows that the OCI-200 instruments are more stable if the 412-nm channel is ignored.
3) Compare the stability monitoring capabilities of the SQM (an expensive device) with a time series of in-water and in-air intercomparisons with a second set of radiometers (a potentially inexpensive alternative). The in-water sensor intercomparisons conducted during AMT-7 showed an overall difference between the Lu and Ed sensors of about 1.1% and 2.7%, respectively. These values are only about 0.6%–0.8% higher than the differences determined with the SQM, but it is important to note that the variance in the individual channels was usually higher in the intercomparison dataset than in the SQM dataset (which makes error detection more difficult). For the OCR-200 and OCI-200 radiometers, for example, the differences in the intercomparison dataset ranged from 2.5% to 5.6% and 1.0% to 3.4%, respectively. The same ranges for the SQM dataset were 0.0%–1.1% and 0.2%–3.7%, respectively. The comparatively poor performance for the SQM irradiance measurements is due to the low flux in the blue part of the spectrum.
Although the degradation in sensitivity from an intercomparison approach may be acceptable for some applications, the biggest problem is that two sensors exposed to an unknown illumination level do not permit a unique determination as to which one is wrong in the event of a (nonextreme) difference between the two: this can be solved only if three radiometers are used, but this is no longer an inexpensive alternative. The SQM provides a unique determination because of its independent set of internal monitors and because the response of all of the sensors can be used to determine the difference between normal and anomalous behavior. In addition, the constant output of the SQM permits additional types of testing that cannot be executed in a variable light field. For example, the SQM data were used repeatedly in this study to establish if the in situ differences were consistent with differences in calibration (in magnitude and sign).
4) Establish which deployment configuration for a reference buoy that can be floated away from a ship produces the smallest uncertainties. The intercomparisons between SeaSURF and SeaBOSS conducted during AMT-5 show clearly that the submerged reference produces the lowest quality data, primarily because of the negative effects of wave focusing: the two agreed to within no better than 8%. This was one of the largest differences derived from any of the AMT intercomparison experiments. Although more sophisticated filtering or data processing might produce better results, no extra effort is needed with the other reference methods, so the practice of collecting Ed(z0, λ) data was discontinued after AMT-5.
5) Determine whether or not reference measurements made far away from a ship are superior to measurements made on a mast mounted on a ship. The experiments with the SeaOPS reference and SeaBOSS show an agreement to within about 6.4%. Although adding a stabilization frame to SeaBOSS reduced sensor tilts, longer data sequences than are normally collected were needed to exploit this advantage. The intercomparisons indicate that the location and design of the deck cell measurements on the JCR significantly limit or cancel the influence of the ship’s superstructure on the data. It must be kept in mind that the potential influence of the ship on the in-air measurements is dependent on the design (and color) of the ship itself. In the case of the JCR, high superstructures are relatively remote from the stern, which constitutes a favorable condition.
6) Ascertain if a modular, low-cost (16 bit) profiler is as capable as an integral, high-cost (24 bit) profiler. The SQM analysis established the superior stability of the 16-bit radiometric systems. This, plus the advantage of being able to easily change subsystems in the event of reconfiguration requirements or component failures, makes LoCNESS an appealing alternative to SeaFALLS. (The benefits associated with SeaFALLS having 13 channels compared to LoCNESS having only 7 channels are not considered here.) The intercomparison of the two showed a deterministic difference, with the LoCNESS LW(λ) values almost always about 3% higher than the SeaFALLS values. The SQM data confirmed that the SeaFALLS data were on average 1.8% lower at low gain than the other radiance sensors used with the SQM.
7) Determine whether a free-fall system produces better optical data than a winch and crane system. The intercomparison of the LoCNESS and SeaFALLS profilers established that SeaFALLS returned (about 3%) lower data values than LoCNESS. Because the SQM analysis showed that the LoCNESS radiometers had a similar response to the SeaOPS radiometers, the higher values from the free-fall units with respect to the SeaOPS measurements means that the SeaOPS instrument measured anomalously low (by about 8%). The lower SeaOPS LW values were probably caused by the influence of the ship’s shadow. Although the ship was oriented to minimize this effect, the length of the boom (10 m) used to deploy SeaOPS was probably insufficient to completely eliminate the problem. Another advantage of the free-fall systems is their ease of deployment and the relatively shorter time they require to perform a cast. This is particularly important, because it means casts can be executed in between cloud passage, and more casts can be done in a particular unit of time. It also means station scheduling can be kept informal, with the ship being stopped only when illumination conditions are optimal.
The datasets analyzed here addressed a number of specific issues regarding uncertainties arising from in situ optical data collection. Table 4 presents a summary of the quantification of these uncertainties as a function of the various deployment systems. All of the instruments have been used on more than one AMT cruise, most of them were modified in between cruises, and all of them involve multiple subsystems, so the entries represent averages biased toward the maximum (average) uncertainties obtained for each source of uncertainty and each subsystem.
The main differences in the levels of uncertainty for each source are in calibration and data collection. The 24-bit systems (SeaFALLS) have demonstrably higher noise, so the calibration entries for these instruments are higher than for the 16-bit systems (SeaOPS and LoCNESS). The data collection uncertainties show the widest range of values with different explanations for each system: SeaFALLS equipped with SeaSURF is the largest because of the problem with wave focusing; SeaOPS and SeaFALLS equipped with SeaBOSS are the next largest because of ship shadow contamination for the former and wave motion variance for the latter;LoCNESS and SeaFALLS with a deck cell are the best with minimal uncertainties.
The SeaBOSS (buoy) measurements deserve additional comment, because the values given in Table 4 represent a best-case scenario and assume that the diffusers are kept dry during deployment. During many relatively moderate sea states, and depending on ship parameters (hull shape, freeboard, orientation with respect to the wind and swell, etc.), it can be very difficult to keep the diffusers dry at the moment of deployment and during data collection. For many of the AMT stations, SeaBOSS was inadvertently splashed, in which case, it was recovered, dryed off with compressed air, and then redeployed.
In a worst-case scenario—all of the uncertainties adding up—the only system to meet the 5% radiometric objective of the SeaWiFS Project is LoCNESS, although SeaFALLS with a deck cell is very close and only 1% higher. SeaOPS is also close, but the ship shadow contamination problem increases the uncertainty another 1.5%. It is worth noting that despite the great deal of care taken in all facets of the in situ optical data measurements (calibration, shipping, handling, deployment, radiometric monitoring, data processing, etc.), the present analysis reveals that in the worst case, the limit of the acceptable level of uncertainty (5%) has been reached. With slightly less attention to any step of the process, it is likely that the total uncertainty would increase beyond an acceptable level.
If the usual practice of considering a quadrature sum is considered, all of the deployment systems have an uncertainty at or below the 5% level. The latter is particularly important, because if the field instruments do not achieve an uncertainty level below 5%, there is no margin of uncertainty for the spaceborne sensor (i.e., SeaWiFS itself) if a total 5% uncertainty level is to be maintained for a vicarious calibration exercise (remote plus in situ instrumentation). Based on this more realistic set of criteria for the field instruments, LoCNESS, SeaOPS, and SeaFALLS, with a deck cell, all perform within acceptable limits—that is, there is still a margin of uncertainty for the spaceborne instrument—with LoCNESS performing the best and only using a little more than half of the 5% uncertainty budget (2.7%).
The Table 4 uncertainty estimates are the result of several kinds of experiments and an extensive set of trials, and although the values are close to the hoped for performance, they are averages, and as averages, they mask an important aspect of the study: individual channels (e.g., the 510-nm channel) may perform significantly worse, but the poor performance was only detected and quantified, because of the substantial effort placed on monitoring and intercomparing the sensors in the field. An individiual investigator deploying to the field with only one profiler and a reference would not be able to duplicate this study and thus could not determine the overall or individual performance of the radiometers: the problems with the 510-nm channel, for example, would remain undetected. A portable source would provide a significant improvement, and many instrument problems in this study were detected and quantified only because an SQM was available (e.g., the anomalous performance of the H045 412-nm channel).
The study clearly shows that a portable source is needed if a thorough understanding of instrument performance is to be acquired during field campaigns, but more importantly (a) the source should be a part of the absolute calibration process and (b) multiple flux levels outside the calibration point are recommended. The capability of the SQM as an absolute source is under investigation (Hooker et al. 2000), and measurements at multiple flux levels with the large number of instruments and small number of optical scientists used in the AMT cruises for this study were too time consuming, so they were not attempted. A larger team or an individual investigator with a smaller number of sensors would not have this problem. More recently, a larger number of optical scientists participated in AMT-8, and multiple flux levels were recorded (Hooker and Lazin 2000).
Acknowledgments
Many individuals have contributed to the success of various components of the SeaWiFS Field Team activities within the AMT Program, including J.-F. Berthon, J. Brown, and C. Dempsey. Their dedicated contributions are gratefully acknowledged. The stewardship of the program and the collection of the optical data have been a high priority for J. Aiken. His diligence and commitment have been essential to the high quality of the optical data collected. The participation of S. Maritorena was supported under USRA Contract NAS5-32484. The final preparation of the manuscript benefitted from the editorial and logistical assistance of E. Firestone.
REFERENCES
Aiken, J., and S. B. Hooker, 1997: The Atlantic Meridional Transect:Spatially extensive calibration and validation of optical properties and remotely-sensed measurements of ocean color. Backscatter,8, 8–11.
——, and Coauthors, 1998: AMT-5 cruise report. NASA Tech. Memo. 1998-206892, Vol. 2, S. B. Hooker and E. R. Firestone, Eds., 113 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Gordon, H. R., 1985: Ship perturbation of irradiance measurements at sea. Part 1: Monte Carlo simulations. Appl. Opt.,24, 4172–4182.
Helliwell, W. S., G. N. Sullivan, B. MacDonald, and K. J. Voss, 1990:Ship shadowing: Model and data comparisons. Ocean Optics X, SPIE,1302, 55–71.
Hooker, S. B., and W. E. Esaias, 1993: An overview of the SeaWiFS project. Eos, Trans. Amer. Geophys. Union,74, 241–246.
——, and J. Aiken, 1998: Calibration evaluation and radiometric testing of field radiometers with the SeaWiFS Quality Monitor (SQM). J. Atmos. Oceanic Technol.,15, 995–1007.
——, and G. Lazin, 2000: The SeaBOARR-99 Field Campaign: SeaWiFS Postlaunch Tech. Rep. Series, NASA Tech. Memo. 2000-206892, S. B. Hooker and E. R. Firestone, Eds. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
——, and C. R. McClain, 2000: A comprehensive plan for the calibration and validation of SeaWiFS data. Progress in Oceanography, in press.
——, G. Zibordi, G. Lazin, and S. McLean, 1999: The SeaBOARR-98 Field Campaign. NASA Tech. Memo. 1999-206892, Vol. 3, S. B. Hooker and E. R. Firestone, Eds., 40 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
——, S. McLean, J. Sherman, M. Small, and G. Zibordi, 2000: The Seventh SeaWiFS Intercalibration Round-Robin Experiment (SIRREX-7), March 1999: SeaWiFS Postlaunch Tech. Rep. Series, NASA Tech. Memo. 2000-206892, S. B. Hooker and E. R. Firestone, Eds., Vol. 13, in press. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Johnson, B. C., P.-S. Shaw, S. B. Hooker, and D. Lynch, 1998: Radiometric and engineering performance of the SeaWiFS Quality Monitor (SQM): A portable light source for field radiometers. J. Atmos. Oceanic Technol.,15, 1008–1022.
——, and Coauthors, 1999: The Fifth SeaWiFS Intercalibration Round-Robin Experiment (SIRREX-5), July 1996. NASA Tech. Memo. 2000-206892, Vol. 7, S. B. Hooker and E. R. Firestone, Eds. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Joint Global Ocean Flux Study, 1991: JGOFS core measurements protocols. JGOFS Report No. 6, Scientific Committee on Oceanic Research, 40 pp. [Available from U.S. JGOFS Planning Institution, Woods Hole, MA 02543.].
McClain, C. R., W. E. Esaias, W. Barnes, B. Guenther, D. Endres, S. B. Hooker, G. Mitchell, and R. Barnes, 1992: Calibration and validation plan for SeaWiFS. NASA Tech. Memo. 104566, Vol. 3, S. B. Hooker and E. R. Firestone, Eds., 41 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
——, M. L. Cleave, G. C. Feldman, W. W. Gregg, and S. B. Hooker, 1998: Science quality SeaWiFS data for global biosphere research. Sea Technol.,39, 10–15.
Morel, A., and L. Prieur, 1977: Analysis of variations in ocean color. Limnol. Oceanogr.,22, 709–722.
Mueller, J. L., and R. W. Austin, 1992: Ocean optics protocols for SeaWiFS validation. NASA Tech. Memo. 104566, Vol. 5, S. B. Hooker and E. R. Firestone, Eds., 45 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
——, and ——, 1995: Ocean optics protocols for SeaWiFS validation, revision 1. NASA Tech. Memo. 104566, Vol. 25, S. B. Hooker, E. R. Firestone, and J. G. Acker, Eds., 66 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
——, B. C. Johnson, C. L. Cromer, S. B. Hooker, J. T. McLean, and S. F. Biggar, 1996: The Third SeaWiFS Intercalibration Round-Robin Experiment, SIRREX-3, September 1994. NASA Tech. Memo. 104566, Vol. 34, S. B. Hooker, E. R. Firestone, and J. G. Acker, Eds., 78 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Neckel, H., and D. Labs, 1984: The solar radiation between 3300 and 12 500. Solar Phys.,90, 205–258.
Press, W. H., and S. A. Teukolsky, 1992: Fitting straight line data with errors in both coordinates. Comput. Phys.,6, 274–276.
Ricker, W. E., 1973: Linear regressions in fishery research. J. Fish. Res. Board Canada,30, 409–434.
Shaw, P.-S., B. C. Johnson, S. B. Hooker, and D. Lynch, 1997: The SeaWiFS Quality Monitor—A portable field calibrator light source. Proc. SPIE,2963, 772–776.
Siegel, D. A., M. C. O’Brien, J. C. Sorensen, D. A. Konnoff, E. A. Brody, J. L. Mueller, C. O. Davis, W. J. Rhea, and S. B. Hooker, 1995: Results of the SeaWiFS Data Analysis Round-Robin (DARR-94), July 1994. NASA Tech. Memo. 104566, Vol. 26, S. B. Hooker and E. R. Firestone, Eds., 58 pp. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
Voss, K. J., J. W. Nolten, and G. D. Edwards, 1986: Ship shadow effects on apparent optical properties. Ocean Optics VIII, SPIE,637, 186–190.
Waters, K. J., R. C. Smith, and M. R. Lewis, 1990: Avoiding ship induced light-field perturbation in the determination of oceanic optical properties. Oceanography,3, 18–21.
Weir, C. T., D. A. Siegel, D. W. Menzies, and A. F. Michaels, 1995:In situ evaluation of a ship’s shadow. Case Studies for SeaWIFS Calibration and Validation, Part 3, NASA Tech. Memo. 104566, Vol. 27, S. B. Hooker, E. R. Firestone, and J. G. Acker, Eds., 25–33. [Available from NASA Goddard Space Flight Center, Greenbelt, MD 20771.].
The optical instruments supplied by the SeaWiFS Project for ocean color remote sensing calibration and validation activities within the AMT Program. The two most important negative environmental effects are clouds and foam. Clouds are the most troublesome because they can intermittently obstruct the direct solar illumination, thereby producing a complicated light field (as shown in the figure for the SeaBOSS instrument). The typical deployment distances to depth and away from the ship are shown in parentheses for the in-water instruments. The LoCNESS instrument is shown in two configurations: the original two-sensor design and the subsequent Three-Headed Optical Recorder (THOR) configuration.
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
A schematic showing the conversion of SeaOPS into LoCNESS (top, inset panel) with the THOR configuration (right, bottom panel). The deck box provides the (computer controlled) DC power for all of the sensors and is designed to avoid instrument damage due to improper power-up sequences over varying cable lengths. The SeaOPS frame was modified before AMT-6, so three sensors could be clustered together (the third sensor connects to a third port on the DATA-100). LoCNESS was modified at the same time, so two sensors could be mounted on the nose.
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
The percent deviation of the internal SQM blue detector measurements of the glass fiducial with respect to their individual (cruise) mean values during (a) AMT-4, (b) AMT-5, (c) AMT-6, and (d) AMT-7. The SQM internal detector values are shown as the open symbol, and the 443-nm channel for radiometer R035 is also shown as the plus signs. The latter are an independent measure of the stability of the SQM during AMT-4 through AMT-7.
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
The results of intercomparing individual sensors and data processing on the SeaOPS frame during AMT-7: (a) Lu(z, λ) (R068 and R037), (b) Ed(z, λ) (I100 and I095), and (c) LW(λ) (R068 and R037). The diagonal line is the 1:1 line. (If the two sensors were calibrated with no deterministic differences and they measured exactly the same illumination field, all of the data would be clustered randomly around the 1:1 line.)
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
The intercomparison of two deck cell references (M035 and M030) during the in-water intercomparison profiles presented in Fig. 4. The AMT sampling is biased toward collecting data during cloud-free conditions, so there is very little data below 50 μW cm−2 nm−1 (which can be collected reliably only during overcast conditions).
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
An intercomparison of reference measurements collected during simultaneous profiles: (a) the LoCNESS deck cell (M030) and SeaBOSS mounted on a mast as a deck cell (N046), and (b) the SeaOPS deck cell (M030) and SeaBOSS deployed as an in-air buoy with no stabilization frame (N046).
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
A case study involving three different references: (a) SeaOPS deck cell (M030), (b) SeaBOSS as a buoy with no stabilization frame (N046), and (c) SeaSURF (H045). The mean spectrum for each sensor (from the deployment time period) and the mean solar irradiance (Neckel and Labs 1984) corrected for transmission through a standard (clear) atmosphere are shown in (d).
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
An intercomparison of derived values from simultaneous deployments of LoCNESS (R036) and SeaFALLS (QO16): (a) LW(λ) and (b) Ku(λ).
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
An intercomparison of derived values from simultaneous deployments of the winched system (SeaOPS) and the two free-fall instruments (LoCNESS and SeaFALLS): (a) LW(λ) and (b) Ku(λ).
Citation: Journal of Atmospheric and Oceanic Technology 17, 6; 10.1175/1520-0426(2000)017<0811:AEOORA>2.0.CO;2
Channel numbers and center wavelengths (in nanometers) for the in-water OCR-200 and OCI-200 radiometers used during AMT cruises. All of the channels have 10-nm bandwidths. The instrument codes are composed of a single letter representing the type of instrument and a two-digit serial number. The instrument types are as follows: I for in-water OCI-200, and R for in-water OCR-200.
Channel numbers and center wavelengths (in nanometers) for the OCI-200 references and the OCR-100 and OCI-100 radiometers used during AMT cruises. All of the channels have 10-nm bandwidths. The instrument codes are composed of a single letter representing the type of instrument and a two-digit serial number. The instrument types are as follows: H for in-water OCI-1000, M for in-air OCI-200, N for in-air OCI-1000, and Q for in-water OCR-10000. Some of the channels (1, 11, and 13) in the SeaFALLS radiometers were changed after AMT-5. The original set (AMT-3, AMT-4, and AMT-5) is denoted by Q0161 and H0231, and the final set (AMT-6 and AMT-7) by Q0162 and H0232. Two sensors are shown for SeaOPS and LoCNESS, as well as for SeaSURF, because a spare sensor (M035) was used during AMT-7 with the former, and the H045 sensor replaced H024 before AMT-5 for the latter.
A summary of the stability of the radiance and irradiance field radiometers used with each optical system (categorized in part by the number of bits used in the A/D converters) during AMT cruises. For each radiometer, the most stable (leftmost entry in range) and the least stable (rightmost entry in range) channels are shown as percent deviations with respect to the mean behavior of the channel during the cruise time period. Only the SeaWiFS channels are considered for the analysis, and the nominal center wavelengths for the most and least stable channels are shown in parentheses. Although most of the in situ analyses presented in this study are based on AMT-5 through AMT-7, AMT-3 and AMT-4 are shown for completeness (to verify trends and average properties).
A summary of the quantification of total measurement uncertainties as a function of the various deployment systems used in the AMT Program. The systems are shown with their reference configurations. Only SeaFALLS was used with multiple references: the SeaBOSS configuration is SeaBOSS deployed as a buoy, and the deck cell configuration is SeaBOSS on a mast. The entries are average values corrected for deterministic problems identified in the study. For example, if the SQM analysis showed that a particular sensor had an incorrect calibration, the data collection uncertainty for the sensor was recalculated, assuming the corrected calibration. Although uncertainties can fortuitously cancel under some circumstances, and in a worst-case scenario can simply sum together, a more realistic procedure is to sum the squares of the uncertainties and report the square root—the so-called quadrature sum.
