Final Report of the First Workshop

Salinity Sea Ice Working Group (SSIWG)

La Jolla, CA, USA, 7-8 February 1998

Preliminary Assessment of the Scientific and Technical Merits for Salinity Remote Sensing from Satellite

Revised 16 December 1998
Address comments and queries to Corresponding Author:

Dr. Gary S.E. Lagerloef
Earth and Space Research
1910 Fairview Ave E, Suite 102
Seattle WA 98102 USA
phone: 206-726-0501 xt 11
fax: 206-726-0524

Introduction: The principles, as well as technical challenges, of satellite remote sensing of sea surface salinity (SSS) using low microwave frequencies (1-3 GHz) have been recognized for over two decades (Lagerloef et al, 1995; Swift and McIntosh, 1983). A Salinity Sea Ice Working Group (SSIWG) has been organized ad hoc to evaluate the technical and scientific merit of SSS remote sensing, as well as the added benefit to sea ice measurements in this frequency band. The initial workshop was held in La Jolla, CA, USA, 7-8 February 1998, to begin a critical evaluation of the scientific impact of the salinity remote sensing data at the anticipated resolution and accuracy, to consider various technical approaches and challenges, and to identify issues requiring further study. It was a gathering of ocean, climate and hydrologic scientists, along with microwave remote sensing technical and engineering specialists, totaling 24 participants (see list below). The agenda focused on salinity. Sea ice issues were deferred to a subgroup to be considered at another time.

The workshop was timely for a number of reasons. First, there has been a growing awareness among researchers of the important role that surface salinity variability plays in ocean and climate dynamics. This has prompted a general call to increase salinity observations within the evolving climate research programs. Such proposals are aimed primarily at in situ observations. The SSIWG will address the role of satellite SSS observations in such future expanded observing networks. Second, in April 1998, NASA intends to announce a call for satellite mission proposals for the Earth System Science Pathfinder (ESSP) program. ESSP supports the development and flight of low cost missions which address new and emerging scientific objectives that are not met with sensors on presently approved missions, and can be accomplished within prescribed cost and schedule guidelines ("smaller, cheaper, faster"). A serious issue, therefore, is whether the gap between scientific requirements and technical capabilities is sufficiently narrow to merit a salinity satellite mission now. Figure 1 illustrates the results from a recent simulation of global SSS and SST retrieval taking into account a variety of sensor and geophysical error sources for a particular satellite sensor being considered (see below). It is indicated that, as long as no significant error sources have been overlooked in this simulation, the large scale SSS features of the world ocean can be resolved. The accuracy and resolution of shorter scales and various time scales is a matter of most concern to understanding the scientific potential of the data.

Figure 1: Simulation of SSS and SST retrieval for a conically scanning multi-channel radiometer and radar, taking into account instrument and geophysical errors, and averaging over 2 days and 100 km grid. (S. Yueh, NASA/JPL).

The scientific merit of such a sensor is not a stand-alone issue because the same (1-3 GHz) microwave frequencies are used to measure terrestrial soil moisture (Jackson et al, 1982.). There are compelling scientific and practical reasons to make soil moisture measurements, and the radiometric signal is much stronger than for SSS (Swift, 1993). Two soil moisture missions were proposed during the first ESSP solicitation in 1996. These were given high marks for science, but were downgraded on issues of technical merit and not selected. Each noted SSS as a secondary, more exploratory, scientific objective, because the SSS accuracy requirements are more stringent than could be met by these proposed systems. Consideration is being given now among competing groups to unify behind a single mission proposal. This leads to the third point, which is that technological innovations continue to improve the outlook for reducing cost and improving accuracy. One of the major technical obstacles for working in this microwave band is the need for very large antenna structures in order to obtain the desired degree of spatial resolution. Footprint size is inversely proportional to antenna aperture, and 10-20 m apertures are required for a ~10 km footprint size, depending on orbit altitude. We may be at a time when such systems are becoming affordable, and the scientific need is becoming sufficiently compelling that a union of scientific interests; soil moisture, SSS and sea ice, can support a mission to address the global surface hydrology.

Accordingly, the SSIWG assumes the task of evaluating whether such systems could provide data of sufficient quality to make SSS and sea ice more equal partners with soil moisture in this undertaking than was the case two years ago, and if so, what the key technical requirements are. With this workshop's primary focus on salinity, the SSIWG approached this task by asking the following questions:

Will measuring global SSS from satellite "well enough" make fundamental or highly significant scientific contributions?

What can we demonstrate to be the most significant scientific benefits?

What does "well enough" mean quantitatively in terms of spatio-temporal resolution and accuracy?

Can we achieve that? What are the technological challenges?

Is there an order of preference for satellite concepts, considering cost vs performance tradeoff?

What additional studies are needed (scientific and engineering), and how soon?

What is the synergism with in situ SSS observations?

Discussions: The workshop provided a forum for the SSIWG to begin seriously addressing these questions. E. Lindstrom recounted the early discoveries of shallow low-salt surface layers within deeper isothermal layers in the tropical Pacific (e.g. Lukas and Lindstrom, 1991), and how these identify the need for salinity data to fully understand tropical mixed layer processes. G. Lagerloef reviewed the principles of salinity remote sensing, and presented the most recent SSS fields from the NOAA/NODC Oceans-98 atlas (furnished by S.Levitus and colleagues, who were not able to attend). These included the seasonal mean and standard deviation maps. The monthly sampling bins remain very sparse in the southern hemisphere, indicating that a large uncertainty remains in estimating a monthly climatology from the historical data. This underscores the intrinsic value of systematic sampling by satellite.

Figure 2: SSS image derived from the Scanning Low Frequency Microwave Radiometer (SLFMR) airborne salinity mapper. (J. Miller, NRL and J. Zaitzeff, NOAA/NESDIS).

The state-of-the-art of airborne SSS remote sensing was illustrated by J. Miller and D. LeVine as they presented the results of recent aircraft SSS remote sensing experiments in the coastal zone, where variations over a range of several psu (practical salinity unit; equivalent to 1 g salt per kg seawater) were clearly resolved (Figure 2), and available ground truth indicated a retrieval accuracy of ~1 psu (Miller et al, 1998; Levine et al, 1998). C. Koblinsky presented preliminary plans for airborne experiments over the Gulf Stream front in summer 1998. A sequence of presentations by W. Wilson and C. Swift addressed new airborne high-precision sensor development, and introduced conically scanning and axial scanning satellite radiometer concepts. D. Burrage, S. Yueh and P. Gloersen discussed SSS retrieval algorithms and simulations, sea ice and sun glitter effects. The retrieval simulation by S. Yueh modeled the major sensor and geophysical error sources for a conically scanning multi-channel radiometer and radar, and 1x1 degree area averages to estimate the SSS and SST retrieval accuracy over a 2-day global snapshot (Figure 1). The projected error for every 2 days is about 0.2 psu in equatorial areas and degrades to about 0.6 psu at high latitude due to a weaker signal at cool sea temperatures (Figure 3). The rms error within +/- 40 deg lat is less than 0.4 psu. These errors would be reduced by about a factor of two for 7 to 8-day averages, and by about a factor of 4 for monthly averages, assuming uncorrelated error. With this, the projected precision may be as small as 0.05 psu for 2x2 degree monthly averages in the tropics.

Figure 3: RMS sea surface salinity retrieval error as a function of latitude, simulated for a conically scanning multi-channel radiometer and radar, taking into account instrument and geophysical errors, and averaging over 2 days and 100 km grid cells, and extrapolating to 8-day averages. (S. Yueh, NASA/JPL).

There were then a number of presentations addressing the scientific impact and measurement requirements. X. Huang reviewed the present difficulties with upper ocean salinity boundary conditions used in models. He concluded that assimilation of SSS fields would be very useful in improving mixed layer and surface flux parameterizations. He suggested that, for the subtropical basins, SSS accuracies of 0.05-0.1 psu and 1x1 degree resolution will provide essential information for monitoring the wind-driven circulation. Uncertainties of 0.1 psu could lead to a 2-3 deg latitude displacement of the major subduction density outcrops. He concluded that higher accuracy (0.01-0.02 psu) might be necessary for monitoring the intermediate and deep water formation in high latitudes, based on the magnitude of interannual salinity signals at deep and intermediate levels. It was noted, however, that the surface expression in the Labrador Sea during the "great salinity anomaly" (Dickson et al, 1988) was much larger (~0.5-1 psu).

R. Tokmakian showed results of the parallel ocean circulation model (POCM) simulations with and without surface hydrologic forcing. The differences in the SSS rms variability, both seasonal and non-seasonal, were several tenths of a psu, particularly under the tropical zones of heavy rainfall (Figure 4). This indicates the degree of SSS variability in response to realistic surface fluxes. By inference, global SSS observations will serve to improve the mixed layer freshwater budget parameters and constrain freshwater fluxes used in GCMs. D. Stammer addressed the potential of SSS observations for ocean state estimation using the M.I.T. adjoint model. He concluded that SSS observations will provide useful information about the time-varying mixed layer salinity field and possibly constrain mixed layer models. The data would also constrain the surface freshwater flux and enhance the information about precipitation over the ocean. For example, 0.1 psu accuracy in the mixed layer salinity would constrain the freshwater flux to within 10 cm/year uncertainty assuming a nominal 35 m mixed layer depth.

Figure 4: Comparison of Parallel Ocean Circulation Model (POCM) rms SSS variability (excluding annual cycle) without surface freshwater forcing (POCM-4B, upper panel) versus with surface hydrological forcing (POCM-4C, lower panel). (R. Tokmakian, NPGS).

Next were a series of presentations relating to the tropical oceans. R. Lukas reviewed the dynamics of the tropical Pacific and warm pool. The impact of the net freshwater buoyancy flux into the ocean due to excess rainfall in the warm pool is on the same scale as the net thermal buoyancy flux (4-5 cm/year steric height equivalent for each). However, the freshwater buoyancy flux is quite important in regulating warm pool SST. Without it, the heat loss to the mixed layer by entrainment would increase and result in an estimated 0.5-0.8 C decrease in warm pool mean SST. Recent model studies by Vialard and Delecluse (1998) indicate a general deepening of the tropical mixed layer and increase in surface currents in model runs without salinity variations versus control runs with salinity variations. During TOGA COARE, using near-surface salinity and velocity data, the "ocean rain gauge" concept was demonstrated as the accumulated rainfall was estimated with the same uncertainty as obtained from the various conventional and radar rain measurements. The Hawaii Ocean Time series has interannual SSS variations with a dynamic range of 0.5 psu; very likely a detectable signal with satellite data and appropriate signal processing. R. Reynolds showed that the upper ocean salinity field has an important interannual signal in the western Tropical Pacific which effects coupled ENSO forecast. Assimilation of altimeter sea level does not reproduce the correct thermal field unless salinity effects are properly accounted for, but surface salinity alone was not sufficient to adjust the Tropical Pacific ENSO forecast model; vertical salinity profiles were required (Reynolds et al, 1998). He presented a novel approach to estimate salinity profiles from SSS, temperature profiles, historical T-S relations and altimeter sea level data (Vossepoel et al, 1998), indicating that SSS combined with other parameters in this manner can have a positive impact on coupled forecasts. R. Murtugudde reported on recent model studies showing that including salinity effects in the tropical Pacific leads to an improved cold tongue simulation and enhanced Indonesian throughflow (Murtugudde and Busalacchi, 1998). He emphasized that prediction model initialization and forecast may be severely degraded if salinity data are not available, but it is not clear to what degree. Precise statements on the need for salinity data are needed, probably from coupled model experiments demonstrating the role of surface salinity.

R. Dickson concluded the scientific talks by explaining the high correlation between the North Atlantic Oscillation (NAO) index and oceanic circulation and temperature variations in the Nordic Seas. The extreme NAO+ anomaly of winter 1995-96 appears to have produced a major outflow of sea ice and meltwater along the East Greenland Current. He speculated that the accompanying SSS signal may become as large as the great salinity anomaly of the 1960s, but there are not sufficient observations to track the feature nor estimate its magnitude. He voiced a clear interest in the potential for satellite remote sensing to track this or future large North Atlantic SSS anomalies.

Scientific issues: The group then proposed three broad primary scientific objectives for SSS remote sensing:

1. Improving seasonal to interannual [ENSO] climate predictions:

This involves the effective use of SSS data to initialize and improve the coupled climate forecast models, and to study and model the role of freshwater flux in the formation and maintenance of barrier layers and mixed layer heat budget in the tropics.

2. Improving ocean rainfall estimates and global hydrologic budgets:

Precipitation over the ocean is still poorly known and relates to both the hydrologic budget and to latent heating of the overlying atmosphere. The "ocean rain gauge" concept shows considerable promise in reducing uncertainties on the surface freshwater flux on climate time scales, given SSS observations, surface velocities and adequate mixed layer modeling.

3. Monitoring large scale salinity events:

This may include ice melt, major river runoff events, or monsoons. In particular, tracking interannual SSS variations in the Nordic Seas is vital to long time scale climate prediction and modeling. High latitude SSS variations influence the rate of oceanic convection and poleward heat transport. These measurements will also be the most technically challenging because of the SSS accuracy needed, and the relatively weaker radiometric signature at low sea temperature. Salinity signals are much stronger in the coastal ocean and marginal seas than in the open ocean in general, but the large footprint size (~10-40 km) will limit near shore applications of the data. Many of the larger marginal seas which have strong SSS signals might be adequately resolved, such as the East China Sea, Bay of Bengal, Gulf of Mexico and Coral Sea/Gulf of Papua.

Figure 5: A schematic of the principal SSS phenomena of scientific interest according to length and time scales.

Figure 5 illustrates a schematic frequency-wavenumber diagram of the scientific phenomena and the space and time scales to be resolved. The group began deliberating specific accuracy requirements for these cases, as discussed at the workshop and by E-mail dialog since. It is apparent that no single requirement is appropriate for the range of scientific problems to be addressed. For Seasonal to Interannual phenomena (Objective 1), especially relevant to ENSO prediction, the discussion at the meeting suggested that the mission could provide useful data in "near real time" to the forecast models, as well as a more precise "delayed research mode" product. The signal to address for the ENSO problem would have length >100km, weekly time scales, and signal strength of 0.05-1 psu. If the error simulation estimate of 0.2 psu for 100 km and 2 day resolution in the tropics is dominated by uncorrelated noise, the sensor could provide a resolution of, say, 200 x 500 km over 4 day averages (divide 0.2 psu error by factor of square root of N=20 samples), leading to 0.05 psu SSS precision for climate forecast models within 5 days of measurement. The research mode data might be more smoothed in time (30 days), but better resolved in space (200 x 200 or 100 x 400) and merge in situ data. These averages might have a lower bound of 0.02 psu error. Studies will be needed over the next few months to clarify the error estimate of various smoothing scenarios. The near real time mode data may be dominated by the satellite data, whereas the delayed research mode data could be made more precise by merger with in situ data.

Indications are that meeting the requirement for the surface freshwater flux problem (Objective 2) could be approached by ~0.05 psu, 2x2 degree and monthlhy resolution in low to mid latitudes. This would reflect a monthly freshwater flux of ~5 cm, which would help resolve the heaviest precipitation zones (~3 to 5 m/year). Larger area averages would reduce uncertainty even further. Schmitt (1998) has suggested a more demanding 0.01 psu requirement, which would resolve a freshwater buoyancy flux equivalent to about 10 Wm-2 heat flux. Near-real time products might be useful at tropical scales described above. Numerical modeling impact studies may be the most appropriate method to address this.

Adressing Objecive 3 in high latitudes will be much more restricted to the "research

mode" data product for maximum usefulness. Both X. Huang and R. Dickson suggested that local signals of ice melt or convective overturning are order 100 km and 0.01 psu. Since Yueh's error was 100 km, 2 day, 0.6 psu at high latitudes, useful near-real time measurements of the important sub-polar phenomena are unlikely. However, for long time scale climatic events, research quality data will require co-analysis with in situ data, and averaging satellite data to further reduce some of these errors. Averaging the 100 km, 2 day resolution samples over 5x5 degree cells and 100 days at high latitude yields N~900 samples. Dividing sample error estimate (0.6 psu) by square root of N leads to 0.02 psu potential error level if noise is all uncorrelated. This would be adequate for examining large scale changes on seasonal to decadal scales related to ice-water conversions (e.g Great Salinity Anomaly or breakup of Antarctic ice shelves).

Technical issues: The above science requirements are near the limits of what can be achieved technically, but appear to be within reach. The limitations are dictated in part by the principles of salinity remote sensing. The relation between radiometer brightness temperature (Tb) and surface temperature(SST) is given by the emissivity, which in turn, is governed in part by electrical conductivity of sea water, hence salinity. The effect is large enough at 1-3 GHz that SSS is detectable. A near linear relation exists between SSS and Tb, with other factors held constant, so that the retrieval can be expressed:

for a given SST, radio frequency(f), incidence angle(sigma) and polarization state(p = H or V). Fig. 5 shows the cases for f=1.43 and 2.65 GHz, sigma=0, and either H or V polarization (same for zero incidence). The signals are stronger at 1.43 GHz, which would be the primary channel for SSS. A1.5 to A2 in temperate and tropical open ocean conditions and A3 to A4 in polar regions (Figure 6). Thus every 0.1 K Tb uncertainty yields .15 to .2 psu uncertainty in low to mid latitudes, and 0.3 to 0.4 psu in high latitudes (the reason for the poleward error trend in Fig. 2). Sensor precision and ambient corrections will need to be at the 0.1 K magnitude or less, yet the highest resolution data will remain noisy. An assumption made repeatedly above, and one that requires verification, is that residual errors will be sufficiently uncorrelated that lower frequency and wavenumber features will not be contaminated. Accordingly, we will need to rely heavily on spatio-temporal averaging, EOF filtering, etc., to resolve the larger scales and lower frequencies of SSS variability.

Figure 6: The variation of brightness temperature due to salinity at 1.4GHz and 2.65 GHz, nadir incidence, with SST as a parameter, according to the formulas given by Klein and Swift, 1977 (Lagerloef et al, 1995)

An engineering subgroup evaluated three candidate satellite sensor concepts: An electronically scanning thinned array, a conically scanning large mesh antenna, and a cross-track scanning axially rotating antenna. Each reported similar radiometer precision and calibration stability, but they differed in spatial resolution, scanning geometry, ancillary channels and polarizations, such that it was not feasible to compare their SSS retrieval accuracies without further simulations of each such as discussed above (for the conically scanning approach).

Key error sources were identified as follows:

Major effects requiring correction:

SST: This will require either ancillary data or multiple radiometer channels for simultaneous SSS and SST retrieval.

Surface roughness and foam (wind stress): This is not well known at these microwave frequencies and there is a need for further experimental studies. The effect is expected to be ~0.1 K per m/s wind speed. Ancillary wind or sea roughness measurement or multiple radiometer and radar channels for simultaneous SSS and roughness retrieval will be required.

Ionospheric Faraday rotation: This influences the polarization vector and the effect increases with lower frequency. It can be mitigated by choice of orbit avoiding solar mid-day and making vertical and horizontal polarized measurements, and a measurement of the third Stokes parameter.

Rainfall effect: This is generally small, but can be significant during heavy rain. The effect (K) is ~.01RH, where R=rain rate (mm/hr) and H is rain layer depth (km), so the effect is ~0.1 K for a 10 mm/hr, 1 km thick, rain layer. The SSIWG recognized the importance of correcting this effect, rather than flagging and rejecting rain-contaminated data, because of the importance of retrieving SSS within the tropical precipitation zones.

Minor effects requiring correction:

Atmospheric vapor and clouds: These are nearly negligible at these frequencies except in rain (see above).

Atmospheric absorption: This is is nearly a constant offset with minor variations that can be corrected with sea level pressure data.

Major effects to be flagged and rejected:

Sun glint: This will be mitigated by choosing an orbit avoiding high sun angles.

Galactic noise: It will be possible to use sky maps to flag when the galactic core radiation reflected off the sea surface will be in the radiometer view.

Radio interference: Observations will be made in the 1.413 GHz hydrogen absorption band protected for radio astronomy. Some isolated noise may still be encountered.

Future activities:

1. During the next few months, the SSIWG will provide guidance on the scientific and technical merit of SSS remote sensing and background material for the preparation of an ESSP mission proposal.

2. Several scientific and technical issues require further study. In particular, the SSIWG must address the effects of surface roughness and rainfall on retrieval errors. Scientific impact studies using numerical ocean models can clarify the importance of SSS data to the primary scientific issues of ENSO prediction, ocean precipitation and climatic freshwater events in high latitudes. These will aid to further refine the measurement requirements. The SSIWG will provide initiative and guidance to get the necessary studies started.

3. The SSIWG will plan to complete a report within about 12 to 18 months.

4. The next meeting is tentatively scheduled to coincide with the Fall AGU in December 1988.

Participant List:

Adamec, David <>
Burrage, Derek <>
Cayan, Dan <>
Chao,Yi <>
Dickson, Bob <>
Gloersen, Per <>
Howden, Stephen <>
Huang, Xin <>
Koblinsky, Chet <>
Lagerloef, Gary <>
LeVine, David <>
Lindstrom, Eric <>
Liu, Tim <>
Lukas, Roger <>
Miller, Jerry <>
Murtugudde, Ragu <>
Njoku, Eni <>
Reynolds, Dick <>
Stammer, Detlef <detlef@lagoon.MIT.EDU>
Swift, Cal <>
Tokmakian, Robin <robint@ucar.EDU>
Weller, Bob <>
Wilson, Bill <>
Yueh, Simon <simon@stokes2.Jpl.Nasa.Gov>


Dickson, R.R., R. Meincke, S.-A. Malmberg and J.J. Lee, 1988: The 'Great Salinity Anomaly' in the Northern North Atlantic, 1968-1982, Prog. Oceanogr. 20, 103-151.

Jackson, T.J., T.J. Schmugge and J.R. Wang, 1982: Passive microwave remote sensing of soil moisture under vegetation canopies, Water Resources Res., 18, 1137-1142.

Klein, L.A. and C.T. Swift, 1977: An improved model for the dielectric constant of sea Water at microwave frequencies, IEEE Trans. Antennas and Prop., AP-25(#1), 104-111.

Lagerloef, G., C. Swift and D. LeVine, 1995: Sea surface salinity: The next remote sensing challenge. Oceanography, 8, 44-50.

Le Vine, D., M. Kao, R. Garvine and T. Sanders, 1998: Remote sensing of ocean salinity: Results from the Delaware coastal current experiment, J. Tech., in press.

Lukas, R. and E. Lindstrom, 1991: The mixed layer of the western equatorial Pacific Ocean, J. Geophys. Res. 96, 3343-3357.

Miller, J., M. Goodberlet and J. Zaitzeff, 1998: Remote sensing of salinity in the coastal zone, EOS Trans. Am. Geophys. Union, 79, 173 & 176-177.

Murtugudde, R. and A. Busalacchi, 1998: Salinity effects in a tropical ocean model, J. Geophys. Res., 103, 3283-3300.

Reynolds, R., M. Ji and A. Leetmaa, 1998: Use of salinity to improve ocean modeling, Physics and Chemistry of the Earth.

Schmitt, R.W. 1998: GOSAMOR, A program for Global Ocean SAlinity MonitORing, a part of ARGO (Array for Real-time Geostrophic Oceanography) and a contribution to CLIVAR, Unpublished Draft, Feb. 1998.

Swift. C.T. and R.E. McIntosh, 1983: Considerations for microwave remote sensing of ocean-surface salinity, IEEE Trans. Geosci. Rem. Sens., GE-21, 480-491.

Swift, C.T. (Chairman), 1993: ESTAR - The Electronically Scanned Thinned Array Radiometer for remote sensing measurement of soil moisture and ocean salinity, NASA Technical Memorandum 4523, 40pp.

Vialard, J. and P. Delecluse, 1998: Role of salinity in the dynamics and thermodynamics of the western Pacific fresh pool: A model study for the TOGA decade, J. Phys. Oceanogr., in press.

Vossepoel, F., R. Reynolds and L. Miller, 1998: Use of sea level observations to estimate salinity variability in the tropical Pacific, J. Tech., submittted.

FTP link to download figures in this document as *.gif or *.ps images.

Recent news and developments regarding the SSIWG and satellite remote sensing can be found at the SSIWG HOME PAGE.

Email Gary Lagerloef if you would like to be added to the SSIWG mailing list.