Modern Observing Systems Working Group Summary

Development and Use of Sensor Systems

The distributions of inorganic and organic carbon pools in the ocean are driven by temporally and spatially varying process that cannot be efficiently examined solely with expeditionary ship based observations. It is well recognized that programs based solely on expeditionary observations will undersample most biogeochemical processes. Autonomous observing systems, including moored, drifting and profiling platforms, and Volunteer Observing Ships, provide an important means of describing ocean variability on a range of spatial and temporal scales that are not accessible to observing programs based solely on research ships. These scales may extend from much finer resolution than is available from shipboard sampling to much longer term observations. However, autonomous observation platforms must use suites of sensor systems capable of monitoring many of the parameters that influence and are influenced by carbon-system processes and fluxes. The availability of these sensors (or lack thereof) is a major hindrance in the use of autonomous observing systems.

At the time the JGOFS program was designed and implemented there were, with the exception of basic bio-optical instrumentation, relatively few chemical or biological sensor systems with the precision, accuracy and endurance for useful in-situ analysis. One of the early goals of the JGOFS study was development of additional sensor systems for biogeochemical processes. The long developmental period required to bring new observing systems to an operational state (often 10 or more years) prevented these new sensors, including the SeaWIFS ocean color imager, from playing the role intended. Autonomous observing systems with bio-optical sensors were used extensively in the Arabian Sea and Southern Ocean process studies (e.g., Marra et al., 1999) and they have become a key component of the BATS and HOT time series stations (e.g. Dickey et al., 1998), but they did not play as large a role in earlier JGOFS studies. Nevertheless, development of diverse autonomous sensor systems has been an important legacy of JGOFS. These new capabilities include enhanced satellite-borne views of the global ocean (e.g., SeaWIFS; Chavez et al., 1999), moored determinations of the partial pressure of carbon dioxide (pCO₂; both absolute and air-sea difference; DeGrandpre et al., 1997; Chavez et al., 1999), in-situ observation of nutrient cycles (e.g., observations of eddy induced changes in nitrate flux; Johnson and Jannasch, 1994; McGillicuddy et al., 1998) and greatly improved estimates of carbon flux through the water column (neutrally buoyant sediment traps in the water column and long time-series observations of benthic respiration; Buesseler et al., 2000; Sayles et al., 1994; Smith et al., 1997).

Autonomous observing systems are now well poised for use in the next generation of global carbon studies, including OCTET. They will complement shipboard measurements in a variety of ways. Autonomous in-situ systems can provide efficient monitoring of temporal changes on scales that are cost prohibitive via ship board measurements. Mobile platforms, such as gliders, drifters and VOS, will give sensor systems access to broad geographic regions. In addition to acquiring data on improved spatial and temporal scales, in-situ analysis can potentially improve data quality by minimizing sample contamination as well as transformations of metastable seawater constituents. Autonomous observations cannot currently supply the breadth of specialized measurements that can be made by shipboard scientists, and problems associated with power requirements, endurance, biofouling and sensor response times also weigh against the strengths of in-situ observations. Nevertheless, a sufficient diversity of autonomous sensors is now available to make in-situ biogeochemical observations a major part of global studies. Regions of special interest that are identified via the wide diversity of shipboard analyses can be targeted for longer term observations with autonomous observing systems capable of monitoring major components of the carbon system.

Development of autonomous sensor systems typically progresses from bench top prototypes, to research systems operated in situ and finally to production of operational systems that are accessible to the broad community. The long time frames that have been required to bring new sensor systems to an operational status suggests that next generation observational programs, such as OCTET, will be influenced primarily by those sensor systems where significant developmental progress has already been made. Sensor systems will be useful not only in OCTET, but also a variety of other programs such as the Global Ocean Observing System (GOOS), Climate Variability and Predictability (CLIVAR), ECOHAB (Ecology of Harmful Algal Blooms) and RIDGE (Ridge Interdisciplinary Global Experiment). The following text and tables represent a summary of many sensor systems that are now operational or well along in development, as well as a list of sensors judged to be important but not yet under development. It would be useful in the long term to obtain community perspectives on the significance of various measurement capabilities in furthering OCTET goals. Which measurements are essential? Which are highly recommended, useful, or only of marginal importance?

Current Sensor and Sensor Platform Capabilities

Many biogeochemical sensor systems are under development or currently in use. The capability of several of these are described in Tables 1-5 with respect to deployment mode (moored, drifting), profiling capability, power requirements and response time. In each table, column 1 identifies the sensor type and column 2 indicates the current status of development: (I) if the sensor is currently operational and available commercially; (II) if it has been successfully deployed in the marine environment for extended periods (>1 month) with oceanographically consistent results but not commercially available; (III) if it is in an early stage of development with successful short-term deployments in the marine environment; or (IV) if bench top prototypes are in operation. Column 3 provides an estimated range for the sensor power consumption with a reasonable duty cycle (e.g., 1-10 measurements per day on a mooring or drifter, or one complete vertical profile per day on glider or float). Column 4 provides each sensor's processing time per sample (e.g., seconds, minutes, hours). Column 5 indicates each sensor's capabilities for moored operations (M) and/or deployment on drifters (D). Finally, column 6 indicates the capability of each sensor for profiling measurements (P), autonomous shipboard measurements (S) and manual shipboard measurements (s). Although sensor cost is a very important issue regulating the number of sensors that can be deployed and the types of integrated sensor platforms that can be constructed, sensor costs are likely to be strongly influenced by demand and therefore are highly transitory. Therefore no attempt has been made to assess this important issue. Since some sensors currently under development for seawater analysis will also find applications in freshwater and drinking water investigations, it is anticipated that sensor commercialization will decrease capital costs per unit.

Sensor Platforms

The sensor systems shown in Table 1 to 5 are appropriate to some, but generally not all, types of sensor platforms. In addition to oceanographic research vessels, potential platforms for sensor deployments include the following:

• Volunteer Observing Ships. VOS observations are generally confined to measurements of sea surface parameters and/or measurements of atmospheric chemistry along major commercial shipping routes. Repeated passages of VOS can provide observations of surface ocean chemical properties over many seasons and years.
  
  
• Moorings and Surface Drifters. Observations from moorings can provide high vertical and temporal resolution of processes at a single point via subsurface instruments, vertically profiling packages, or fiber optic links to sensors at depth. The high cost of maintaining moorings generally limits them to a few in number. However, the TAO/TRITON array of some 70 moorings in the Equatorial Pacific (http://www.pmel.noaa.gov:80/toga-tao/home.html) is an example of the effort that can be sustained if observations receive sufficient societal support. Surface drifters are proximate to the sea surface, but they can also be expected to access the mixed layer in general. The very large number of drifters deployed in the WOCE program (http://www.aoml.noaa.gov/phod/dac/gdc.html) illustrates the potential of sampling with low cost packages distributed throughout the ocean.
  
  
• Floats and Gliders. Observations from gliders and neutrally buoyant floats are distinct from VOS, mooring and surface drifter observations in allowing access to both surface and mesopelagic zones over broad geographic ranges. A principal advantage of neutrally buoyant floats is low cost and, therefore, an extremely large geographic distribution with reasonable sampling density (http://wfdac.whoi.edu). This is an important and occasionally essential advantage in assessment of biogenic particle fluxes. Gliders share with floats a capacity for substantial observational endurance and resolution on a vertical scale. They are distinct in providing a capacity for observations of parameter gradients along defined horizontal sections. Although gliders are still in development, they hold significant promise.
  
  
• CTDs and Towed Vehicles. A principal advantage of CTD based measurements is rapid profiling of selectable depths. Towed, undulating vehicles constitute an important platform for integrating chemical and physical measurements with biological measurements and sample collections over mesoscales. Both platforms require the presence of a research ship.
  
  
• Autonomous Underwater Vehicles. AUV operations provide for biogeochemical observations with somewhat reduced constraints on sensor power consumption. Deployment of larger sensor arrays is therefore possible. AUVs potentially have access to all ocean depths, and fuel-cell technology provides a capacity for great measurement endurance. The endurance and spatial coverage of AUV operations could be greatly extended via mid-ocean nodes for AUV deployment and docking. AUVs should, therefore, be capable of multiple autonomous docking and deployment cycles. Finally, it is important to recognize the potential for extending shipboard measurement campaigns via AUV-assisted operations.
  
  
• Satellites. Observations obtained via satellite include ocean color (from which chlorophyll concentration is obtained), sea surface height, wind field, temperature, dust and aerosol optical thickness. Satellite observations constitute the most potent capability for integrating global observations of the upper ocean. Although they generally provide little information on the ocean interior, the global extent of coverage greatly outweighs this limitation. The global coverage that is unique to satellite-borne sensors provides a powerful capability for integrating and constraining other observations and model results.
  

Sensors

Table 1 lists sensors for measurement of inorganic carbon system parameters. Characterization of the carbon system requires measurement of two independent carbon dioxide parameters. The partial pressure of CO₂ (pCO₂) is currently measured in situ (e.g., DeGrandpre et al., 1997) and as a mole fraction difference between the atmosphere and seawater (from which delta pCO₂ is calculated). Measurements of delta pCO₂ have been made continuously for more than 1 year on several of the TAO/TRITON moorings (Chavez et al., 1999). Seawater pH has been measured in situ using spectrophotometric procedures similar to those that have been used for in-situ pCO₂ measurements (Kaltenbacher et al., in press). Total dissolved inorganic CO₂ (C_T) measurements and total alkalinity (A_T) measurements are currently obtained only through shipboard procedures. In-situ C_T measurements are under development.

Dissolved gas measurements are essential parameters for assessments of net primary production. In-situ oxygen measurements are currently obtained through amperometry with Clark-type membrane covered electrodes (Wallace and Wirick, 1992; DeGrandpre et al., 1997). A variety of commercially available sensors are available, although long-term stability (>1 month) at the level required for oceanographic studies remains problematic (S. Emerson, personal communication). Fluorometric measurements are also possible and should be amenable to some types of in-situ work, although these sensors are also subject to stability problems (Demas et al., 1999). Commercially available gas tension devices provide N₂ fugacity estimates as a difference between total gas fugacities and the fugacities of O₂ and Ar. These measurements are needed to assess whether oxygen concentration changes are due to physical or biological processes. A Membrane Introduction Mass Spectrometery developed for measurements of volatile species has been deployed in-situ and shows promise for observation of a variety of gases in-situ (Short et al., 1999).

Table 2 summarizes the current measurement capabilities of nutrient and micronutrient sensors. A more extensive review can be found in Johnson et al. (2000). Nitrate and nitrite can currently be measured throughout the water column using several procedures. A variety of adaptations of the standard colorimetric method for nitrate (reduction on cadmium to nitrite and determination as an azo dye) have been developed, and several are available commercially. Colorimetric nitrate analysis systems have been deployed in the open ocean for time periods up to 4 months (e.g., McGillicuddy et al., 1998). Sensitive fluorometric procedures are available for all forms of inorganic nitrogen in AUV and ship based operations, but have not undergone extensive field testing. Autonomous PO₄ and Si measurements are currently available only as shipboard measurements.

Trace metals are key regulators of the ocean carbon cycle. Iron can be measured using both chemiluminescence and colorimetric procedures. These procedures are potentially robust and amenable to all forms of in-situ analysis. Detection limits must be less than 0.1 nM for open ocean applications. Autonomous Mn measurements have been obtained in situ via fluorescence based measurements (Klinkhammer, 1994).

Table 3 lists several developments that are underway on biological sensor systems. The response of the ecosystem to chemical and physical forcing is a key factor in understanding global biogeochemical cycles, and sensors that report ecosystem structure and rates of ecosystem processes, such as primary production, are critical. For example, changes in iron flux that may stimulate nitrogen fixation may dramatically alter the potential capability of the biological carbon pump. Biooptical instrumentation such as spectral radiometers, fluorometers and transmissometers make it possible to assess the stock of particles and their nature by measuring changes in the optical properties of seawater. These instruments are generally well developed and commercially available. Key improvements regarding resistance to biofouling are being made. DNA based analyses, driven by the biomedical industry, may shortly make it feasible to determine phytoplankton speciation at the genus and species level with autonomous observations (e.g., Scholin et al., 2000). Developments of instruments such as the Fast Repetition Rate Fluorometer now make it possible to remotely study the physiological state of phytoplankton (e.g., Behrenfeld et al., 1996).

Table 4 lists a variety of instruments that operate by collecting samples for later laboratory analysis. These include instruments that measure vertical particle flux, both from moored or neutrally buoyant platforms, benthic landers that measure chemical flux across the sediment-water interface and in-situ filtration devices that allow particles or dissolved chemicals present at ultra-trace concentrations to be collected from large volumes of seawater. Recent developments include submersible incubation devices that may be used to perform in-situ ¹⁴C primary production measurements and water sampling devices that can collect uncontaminated samples for trace metal analyses, obviating the need for complex, trace metal clean, sampling gear. A novel mooring that allows marine aerosols to be collected and analyzed is currently being tested.

Finally, Table 5 summarizes some of the satellite sensors that are now operational or which are in development. Satellite sensors provide an unparalleled view of the spatial and temporal variability of processes that leave a detectable imprint on the ocean surface. Such processes include the development of phytoplankton blooms which alter ocean color, and the formation of eddies which may alter seasurface height and temperature and therefore the spectrum of outgoing infrared radiation. Sensors for these systems are operational and relatively well understood. Recent developments include algorithms to determine concentrations and fluxes of air-borne aerosol including transport of micro-nutrient iron to phytoplankton. Remote sensors for salinity are in the planning stages. Satellite sensors will play an integral role in the next generation of global studies.

Potential Applications of Autonomous Sensor Systems to OCTET Objectives

Autonomous sensor systems could be applied directly to test a number of OCTET hypotheses. For example, imagine a profiling glider equipped with CO₂, oxygen and nitrate sensors, as well as a small suite of bio-optical instrumentation and a CTD. An array of 100 such instruments could be used to directly answer questions such as: How do carbon fluxes in the equatorial Pacific respond to physical variability on ENSO and PDO time scales? What is the magnitude of the carbon sink (natural and perturbed) in the North Atlantic? What is the natural variability of carbon uptake in the Southern Ocean? How do ecosystem structure and carbon fluxes respond to changes in stratification and mixed layer depth?

Another example includes an autonomous observing system that is built around a moored aerosol sampler (Sholkovitz et al., 1998). Deposition of aerosol iron on the sea surface is an episodic event that may stimulate short lived, but important plankton blooms. Such episodic events are difficult to sample from a ship, but may be well resolved by a moored system that measures aerosol concentration, nutrient and trace metal concentrations in the upper ocean and production of biomass (determined from biooptical paramters).

Development of such instruments is a reasonable programmatic goal. All of the individual sensors have now been operated for extended periods of time in the ocean. The glider platform is now operational for at least limited deployments and versions capable of making ocean basin transects are in development. Moored buoys capable of sampling aerosols have been demonstrated.

 Satellite-borne sensors in particular will be useful in measuring or inferring biogeochemical variables of interest to OCTET goals on a basin scale. Several oceanographic processes (and compartments of the carbon cycle) can be addressed with remote sensing observations. For example, estimates of air-sea gas exchange and oceanic pCO₂ can be improved using remote sensing. The air-sea exchange coefficient is usually parameterized using wind speed. Scatterometers provide global observations of wind speed and direction on a daily basis. Similarly, measurements of surface roughness (from which capillary wave height is estimated) from the TOPEX/Poseidon altimeter (Frew et al., 2000) or from scatterometry may give a more direct value of the exchange rate than wind-based parameterizations.

Relationships between sea surface temperature and pCO₂ are not globally applicable and change in space and time (Lee et al.,1998). Salinity will be measured remotely on European and US salinity missions in the mid-2000s; this will enable the application of local relationships obtained from shipboard and moored or drifting platforms. Although chlorophyll concentration (obtained from ocean color) is often invoked as a factor determining pCO₂, the algorithms that would incorporate it are still under development.

Ocean color measurements have fueled the development of a suite of primary production (PP) models that use chlorophyll concentration, irradiance, and SST (all measured remotely). There are several types of PP models with varying degrees of complexity (Behrenfeld and Falkowski,1997). Although models exist to estimate primary production, a significant level of uncertainty follows from our estimates of new or export production. Most of these estimates use a relationship between f-ratio and SST or primary production, nitrate or chlorophyll concentration (Sathyendranath et al.,1991). These approaches can be used with satellite-derived measurements, but will only be as good as the primary production estimate and as the inferred f-ratio, which may vary regionally and with time.

Satellites also provide an unprecedented opportunity to quantify variability and processes that are unresolved by coarse models and necessarily inadequate sampling campaigns. Quasi-synoptic coverage is an amazing benefit, even considering the loss of data due to cloud coverage for electromagnetic sensors. The TOPEX/Poseidon altimeter enables improved quantification of eddies for biogeochemical applications (Siegel et al., 1999) and for ocean circulation models. Coastal processes, which require higher spatial and temporal resolution than is usually possible from sun-synchronous sensors, can benefit from geostationary platforms and multispectral reflectance measurements. The NASA-NOAA Special Events Imager (SEI) will instrument a GOESS satellite in the early 2000s and will be capable of 300m (few-minute) resolution.

Sensor Wish List

As mentioned above, the development time for sensors is long and somewhat unpredictable. Efforts to continue sensor development must be an important adjunct to global biogeochemical studies. There is a clear need for a variety of additional sensor systems, including continued development of moored sampling equipment for trace metals and for gases, including inorganic carbon and freons. Sensors capable of monitoring important components of the carbon system on relatively high speed platforms (towed, undulating fish or CTDs) are also needed. Extensive development is needed before measurement of major components of the organic matter pool, particularly dissolved organic carbon, nitrogen and phosphorus, is possible. Particulate inorganic carbon sensors would be an important tool to monitor changes in the ratio of organic to inorganic carbon export, a potential control on the oceans capacity for fossil fuel carbon dioxide. Sensors capable of ultratrace measurements of metals, particularly iron, and metal speciation would also be valuable.

One of the most important advances in sensor technology is likely to follow from the development of micro electromechanical systems (MEMS). MEMS devices will provide chemical analyses on a chip. Colorimetric microfluidic MEMS devices are expected to be very versatile and amenable to mass manufacture. Since many important organic measurements are based on absorbance spectroscopy, low cost MEMS devices may allow a large number of measurement types within one small, inexpensive, robust and widely deployed system.

Refinement of existing sensors is also an important activity that must be continued. This includes important efforts to reduce size, power requirements, cost and failure rates of current sensors. Relatively modest improvements will greatly extend the usability of instruments.

Conclusions

A variety of sensor systems and platforms are under development. Although the development of measurement protocols is generally a long-term process, this is not always the case. Some sensor systems will be amenable to a wide variety of analyses. Colorimetric analyses have many attributes in common, with differences based principally on the timing and sequence of reagent additions to aqueous samples. As such, a single spectrophotometric system can be used to access a variety of measurements solely via software and reagent selections. Critical perspectives on sensor/platform efficacy will be gained through long-term in-situ testing and rigorous intercomparisons with ship-based sampling. Initial deployments of in-situ devices constitute the final stage of sensor development in which field data are obtained while sensor performance is still under evaluation. Since sensor capabilities are multidimensional with respect to power requirements, endurance, detection limits, size, cost, etc., deployment of suites of sensors for measurement of a single chemical parameter can provide important insights about not only sensor accuracy, but also the deployment modes most appropriate for each sensor design. In some cases, simultaneous deployment of a variety of sensor types can be used to assess sensor accuracy/efficacy. The suite of inorganic carbon-system measurements, total dissolved inorganic carbon, alkalinity, pCO₂ and pH, are linked via a robust physical-chemical model. While only two of these four measurements are required for calculation of all system parameters, different types of measurements may be appropriate for different depths and locations.

References

Behrenfeld M. J., A. J. Bale, Z. S. Kolber, J. Aiken and P. G. Falkowski. 1996. Confirmation of iron limitation of phytoplankton photosynthesis in the equatorial Pacific Ocean. Nature 383: 508-511.

Behrenfeld, M.J. and P.G. Falkowski. 1997. A consumer's guide to phytoplankton primary productivity model. Limnol. Oceanogr. 42:1479-1491.

Buesseler, K.O., D.K. Steinberg, A.F. Michaels, R.J. Johnson, J.E. Andrews, J. R. Valdes and J.F. Price. 2000. A comparison of the quantity and composition of material caught in a neutrally buoyant versus surface-tethered sediment trap. Deep-Sea Res. I, 47: 277-294.

Chavez F. P., P. G. Strutton, G. E. Friederich, R. A. Feely, G. C. Feldman, D. G. Foley and M. J. McPhaden. 1999. Biological and chemical response of the equatorial Pacific to the 197-98 El Nino. Science 286: 2126-2131.

DeGrandpre, M. D., T. R. Hammar, D. W. R. Wallace and C. D. Wirick. 1997. Simultaneous mooring-based measurements of seawater CO₂ and O₂ off Cape Hatteras, North Carolina. Limnol. Oceanogr. 42: 21-28.

Demas J. N., B. A. DeGraff and P. B. Coleman. 1999. Oxygen sensors based on luminescence quenching. Anal. Chem. 71: 793A-800A.

Dickey, T., D. Frye, H. Jannasch, E. Boyle, D. Manov, D. Sigurdson, J. McNeil, M. Stramska, A. Michaels, N. Nelson, D. Siegel, G. Chang, J. Wu, and A. Knap. 1998. Initial results from the Bermuda Testbed Mooring Program, Deep-Sea Res. I, 45: 771-794.

Frew, N.M. et al. 2000. AGU Ocean Sciences Meeting Abstracts, San Antonio, TX.

Johnson, K. S., and H. W. Jannasch. 1994. Analytical chemistry under the sea surface: monitoring ocean chemistry in situ. Naval Research Reviews, XLVI-3: 4-12.

Johnson, K. S., V. A. Elrod, J. L. Nowicki, K. H. Coale and H. Zamzow. 2000. Continuous flow techniques for on site and in situ measurements of metals and nutrients in sea water. In, In situ monitoring of aquatic systems: chemical analysis and speciation, J. Buffle and G. Horvai, eds. John Wiley.

Kaltenbacher, E., E. T. Steimle and R. H. Byrne. In press. A compact, in situ spectrophotometric sensor for aqueous environments: Design and applications. IEEE Oceanic Engineering Society. Underwater Technology 2000.

Klinkhmammer G. P. 1994. Fiber optic spectrometers for in situ measurements in the oceans: the ZAPS probe. Mar. Chem. 47: 13-20.

Lee, K, R. Wanninkhof, T. Takahashi, S.C. Doney and R.A. Feely. 1998. Low interannual variability in recent oceanic uptake of atmospheric carbon dioxide. Nature 396:155-159.

Marra, J., T. D. Dickey, C. Ho., C. S. Kinkade, D. E. Sigurdson, R. Weller, and R. T. Barber. 1999. Variability in primary production as observed from moored observations in the central Arabian Sea in 1995, Deep-Sea Res. II, 45: 2253-2267.

McGillicuddy D. J. Jr., A. R. Robinson, D. A. Siegel, H. W. Jannasch, R. Johnson, T. D. Dickey, J. McNeil, A. F. Michaels, A. H. Knapp. 1998. Influence of mesoscale eddies on new production in the Sargasso Sea. Nature 394: 263-266.

Sathyendranath, S. T. Platt, E.P.W. Horne, W.G. Harrison, O. Ulloa, R. Outerbridge and N. Hoepffner. 1991. Estimation of new production in the ocean by compound remote-sensing. Nature 353: 129-133.

Sayles F. L., W. R. Martin and W. G. Deuser. 1994. Response of benthic oxygen demand to particulate organic carbon supply in the deep sea near Bermuda. Nature 371: 686-689.

Scholin C. A. and 25 others. 2000. Mortality of sea lions along the central California coast linked to a toxic diatom bloom. Nature 403: 80-84.

Sholkovitz E., G. Allsup, R. Arthur and D. Hosom. 1998. Aerosol sampling from ocean buoys shows promise. EOS Trans. Am. Geophys. Union 79: 29.

Short R. T., D. P. Fries, S. K. Toler, C. E. Lembke and R. H. Byrne. 1999. Development of an underwater mass-spectrometry system for in-situ chemical analysis. Meas. Sci. Technol. 10: 1195-1201.

Siegel DA, McGillicuddy DJ, Fields EA. 1999. Mesoscale eddies, satellite altimetry, and new production in the Sargasso Sea. J. Geophys. Res.-Oceans 104: 13359-13379.

Smith, K. L. Jr., R. C. Glatts, R. J. Baldwin, S. E. Beaulieu, A. H. Uhlman, R. C. Horn, and C. E. Reimers. 1997. An autonomous, bottom-transecting vehicle for making long time-series measurements of sediment community oxygen consumption to abyssal depth. Limnol. Oceanogr. 42: 1601-1612.

Wallace, D. W. R. and C. D. Wirick. 1992. Large air-sea gas fluxes associated with breaking waves. Nature 356: 694-696.