Introduction
Data dealing with the materials of the seabed has significant levels of uncertainty. This comes from the fact of sampling underwater, in different weathers, using diverse vessels and equipment, problems of sample transport and also spatial scales of observation, amongst other factors.

Integrating seabed properties data requires that uncertainties should be measured and allowed for in software, map compiling, and statistics and should be acknowledged in downsteam applications of the data. This web site summarises some of the findings of an investigation of uncertainties for the dbSEABED system.


Gauging the Uncertainties
To gauge the scales of uncertainty and investigate good ways of dealing with uncertainty issues, raw datasets that involved some degree of replication were collected from dbSEABED. That replication could be in sampling, observation, and analysis, parameter intercomparison, or spatial variability within sites and between sites.

The character of the seabed is a multi-parameter problem, and many parameters are cross-related in measurement (e.g., porosity and density) or in their final application (e.g., grainsize and erodability). A method was sought to bring the uncertainties of the sets of characters into a common basis. The problem has 3 dimensions: (a) comparing between parameters, (b) consistency of measure across environments (e.g., from shallow marine sands to abyssal muds), and (c) ability to use with the sampling, analysis and reporting stages of data.

The Coefficient of Variation (Vc) was investigated for suitability, but was found wanting. For example, at the same absolute +uncertainty, say in phi, Vc changes from sand to mud only because the average grainsize is changed. Vc is also unstable around zero, important with the phi grainsize scale. Thirdly, Vc is not comparable between parameters, and is affected by offsets such as in P-wave velocities (1500m/s up). These features made Vc an inappropriate choice for compiling uncertainties across a multi-parameter, multi-environment database like dbSEABED.

To solve the problem the concept of Full Scale Deflection and Dynamic Range was adopted from engineering. Range Scaled Variation (RSV) is a scaling of a measure of variability or deviation (usually SD or RMS - Standard Deviation, Root Mean Square) divided by the half the range of plausible data values. That range is for instance, all values encountered in sediments. Thus for p-wave velocities the range approximates to 2000m/s (1500-3500) and an uncertainty of 50m/s scales to 5% RSV. Obviously, when using RSV, each parameter range also needs to be stated.


Magnitudes of Uncertainty
Scaled to terms of RSV, the variabilities observed in diverse sets of seabed observations make sense (Figure 1). Figure 1 shows some values, separated into classes of uncertainty:

• sampling: including disruption of a sample, representativeness, etc
  
• measurement: including with laboratory instrumentation, visual description, electronic probes in situ
• validity: which refers to the relevance of a measurement to a property, for example of settling rates to grainsizes
• local spatial: referring to the patchiness observed at a site
• spatial-temporal: results variability between overlapping subsequent surveys
• wide spatial: the unpredictability of results that are geographically separated by >5km
  

An example of the sampling issues is the probability effect with small samplers of representing larger clast sizes (Ferguson and Paola ). The magnitude depends on the match between sampler footprint and clast sizes. Sampling effects also include differences of probe and corer design and type (Mulhearn 2001; Laban 2000).

Amongst measurements, well constrained analyses with the same machinery (tests of precision) have RSV of order 2% whereas inter-instrumental sets have RSV of 4-5% (tests of accuracy) (e.g., data in Syvitski et al. 1991). By employing samples that have both a detailed analysis and a textural description it is possible to gauge the uncertainties involved in word-based sediment description. Many datasets exist where this can be done, for example Hollister (1973). The result is that for a wide suite of gravel-mud types the RSV based on RMS deviation is of order 3%.

Included in validity uncertainties are effects of calibration, such as the temperature/pressure state for determinations of geomaterial p-wave velocities (Shumway 1958). Another example lies in the relationship between the different reported central grainsizes: moment. median, and Folk and Inman 'Mean' grainsizes.

Spatial variabilities can be scaled in the same way, using the semivariance (actually [2*semivariance]^0.5) in place of SD. A range of measured values for the variation at one site - usual dimensions about 100m - is shown in Figure 1, (yellow) covering variables VsVo, acoustic attenuation, phi grain size, porosity, density, silt fraction. Measured values for spatial variance over larger distances are shown (red, Figure 1) also. They are the source of the major uncertainties on maps of seabed properties.


References

Hollister, C.D., 1973. Atlantic continental shelf and slope of the United States - texture of surface sediments from New Jersey to southern Florida. U.S. Geological Survey Professional Paper 529-M, 23 p.

Jenkins, C. J. (Subm.). Quantifying the uncertainty in marine substrate mappings. [Continental Shelf Research.]

Laban, C., 2000. Comparison of sampling and grain-size analysis methods. In: Seabed News, July 2000, 3-6. Southhampton Oceanography Centre, Southhampton, UK. [WWW page, http://www.eu-seased.net/services/issue1/_pages/page6.html].

Mulhearn P.J., 2001. Influences of Penetrometer Probe Tip Geometry on Bearing Strength Estimates for Mine Burial Prediction. DSTO Technical Report, DSTO-TR-1285, 22p. Maritime Operations Division, Aeronautical and Maritime Research Laboratory, Defence Science and Technology Organization, Melbourne, Australia.

Shumway, G., 1958. Sound Velocity vs. Temperature in Water Saturated Sediments. Geophysics, 23, 494-505.

Syvitski, J.P.M., LeBlanc, K.W.G., Asprey, K.W., 1991. Interlaboratory, interinstrument calibration experiments, in: Syvitski, J.P.M. (Ed.), 1991. Principles, methods, and application of particle size analysis. Cambridge Univ. Press, Cambridge, UK, pp. 174-193.