Contents

General Statement Variability of Seabed Measurements Observed Measurement Accuracies Implications for predictions

dbSEABED aims to produce as accurate as possible outputs from highly varied input data to do with the seabed. The quantitative benchmark which is set for this is "that outputs shall have an accuracy and reliability that is better than the variability which is observed on repeat measurements from individual seabed sites (ie. within ~20m radius)".

In this page we review what that benchmark is for a range of parameters and show some of the results for calibrations of the information processing done by dbSEABED.

Back to top

The uncertainties inherent in reports of measurements of the seabed arise from :
(i) sampling technique followed by sample transport, treatment and preparation
(ii) selectivity in sampling and subsampling
(iii) techniques of measurement, whether observational or instrumental
(iv) calibrations involved in transformations from raw to final data outputs
(v) limitations imposed by precisions in reporting, averaging, etc.

Back to top

Grainsize measurements
For grainsize analysis, essentially 4 methods are in use:
(i) sievving which is a shape/size sorting process in 2 dimensions (sieve mesh plane);
(ii) settling velocity which is a size/density/shape/roughness differentiating process (equivalent hydraulic radius);
(iii) image analysis which is a shape/size sorting process in 2 or 3 dimensions
(iv) human visual observations (eg, deck descriptions).

Equivalent machinery is: (i) sievving; (ii) settling column, pipette analysis and Sedigraph; (iii) TRACOR and GALAI image analysis systems,  Malvern Laser Interferometer; (iv) hand lens.

The results of grainsize analyses performed by these different techniques on the same sediment, can be very different (Syvitski 1991)
- especially where grain densities depart from quartz density, grains are hollow or flat/elongate (as with heavy minerals, forams, shells).

Acoustic properties
Compressional wave velocities are commonly quoted to a precision of 1 m/s (e.g., Hamilton 1980) - far finer than accuracies achieved even under laboratory controlled conditions. Using advanced and uniform equipment, Schaftenaar & Carson (1984) and Carson & Christensen (1977) observed measurement errors of +1% (+15-20m/s) for Vp in sediments (better accuracies of +0.5% may be obtained at higher confining pressures). Bachmann (1985) estimated errors on his determinations of Vp in sediments as +0.2-0.33% (+3-5m/s).

Estimates of the accuracy on porosity measurements are usually about 4-7% (absolute; Boyce 1976, p.696) especially since techniques vary so widely from Gamma Ray attenuation to weighing water contents.

Dispersion. Because of acoustic dispersion in sediments, acoustic velocity depends on frequency. Dispersion is a consequence of attenuation and is a feature of sound interaction with sediments (Wingham 1985; but some dissent, see Hamilton, 1972). It may arise from the medium (sediment) and the environment (structure). According to Stoll (1989) the frequency at which the change in velocity is most rapid rises as the sediment permeability lessens, ie, is higher for muds. According to Biot Theory the ramping of velocity due to dispersion may set in in coarse sediment at frequencies as low as 100Hz. For all sediment types, the rise in velocity as frequency increases is of order 50-100m/s (Stoll 1989, p.25). Unfortunately, the majority of laboratory determinations of Vp have been made around frequencies where dispersion occurs.

Strain rate. As pointed out by Stoll (1980), the dynamic shear modulus is reduced by >40% between strains of 10^-6 to 10^-4. The modulus is approximately constant for strains < 10^-6 - the range for acoustics. The strains applied for an experiment or measurement can have significant effect on results for velocity, attenuation and shear moduli (Stoll 1980, 1989; Winkler & Nur 1982). The following are examples of strain rates achieved with different instruments:
(i) Large air gun, 25m range, 40Hz, 5 x 10 ^-6; (ii) 1 lb dynamite, 25m range, 260 Hz, 2 x 10^-9; (iii) Resonant column tests,  10^-6 to 5 x 10^-3; (iv) Torsional shear tests,  5 x 10^-4 to 10^-2.

Sample disturbance. Measurements of physical properties on samples collected from the seafloor, transported to a laboratory and arranged into a measuring device suffer considerable inaccuracy relative to in situ values.
A. In geotechnics differences of 10-20% in the unconfined shear strengths of sediments are regularly observed between laboratory and in situ measurements (Young & others 1988). Collection and handling sediments changes the sediment fabrics, leading to geotechnical 'remoulding'. Whether this loosens or tightens the sediment fabrics depends on whether the materials are dissipative or dilative (e.g., Kayen & others 1989). In any case strong and repeated disruptive strains are able to alter the strengths of sediments by Chaney & Richardson (1988).
B. Richardson & others (1988) documented the measurement variability of in situ and lab Vs results on loose surface sediments. They found an average difference of about +3.5-5.5 m/s implying (1*) variances of about +10% at Vs of 13-80 m/s. The lab measured Vs were consistently lower. They caution against applying this difference to data from different settings and techniques.
C. Sample collection inevitably involves the loss of acoustically significant natural gas bubbles (see Boyle & Chotiros 1995), changes to temperature-pressure conditions and saturation. The issue of 'porosity rebound' for sediments taken from deep sedimentary burial has been extensively investigated (e.g., Hamilton 1976d). It is now not thought to cause significant alterations of porosities or velocities (Urmos & Wilkens 1993).

Natural spatial/temporal variance. The properties of a seabed can change considerably over short spatial scales.
A. Statistics from Richardson & Briggs (1993) indicate that the variances - expressed as % of average values - Coefficients of Variation (CV) - for acoustic properties measured within 100m over a wide range of environments average:
 

These values are a combination of measurement errors, plus seafloor lateral and vertical variations.
The CV for grainsizes is wide - as presented - but this is an artefact of the fact that the phi scale includes 0 and that the average grainsize is close to zero. Similarly with Vb/Vw, but in general (with the same variance) the closer a parameter’s mean value is to 0, the higher will be the CV. If the CV’s are recalculated as a percentage of half the plausible range (PR) instead of the average, of values for sediments, then they become:
 

Observe that grainsizes actually show no greater dispersion than the acoustic properties and that most parameters have a dispersion of ~10% of their plausible ranges.

Sediment bioturbation. Bioturbation operates mainly in the top 0.1m of sediments then rapidly declining in effects to about 0.3m. Burrows of metres length and depth are rarely encountered.

At the mesoscopic scale it is obvious that bioturbation will change the acoustic responses of sediments, by introducing burrow voids,  surface roughness and buried grainsize inhomogeneities. The commonness and scales of these structures are well illustrated in the geological literature (Leeder 1982; Harris & others 1991). Their effect will be to enhance the backscatter and volume scattering losses in propagation experiments. On the other hand, bioturbation can also work to homogenize sediments which have a pre-existing layered structure (Richardson & others 1983). An appropriate way to view the processes is as a decay process for both strong layering and strong homogeneity, tending with time to an equilibrium degree of inhomogeneity for the environment.
Bioturbate churning of sediments appears also to modify their structure at a granular level. Richardson & Young (1978) and Richardson & others (1983) conducted some preliminary studies of the problem, with both studies indicating that bioturbation in silty muds increases porosity. The first study found a 16% porosity increase due to reworking by small bivalves, attended by parcelling of the sediment into fibril-bound fecal pellets that created clustering of grains. The second study revealed porosity increases of up to 25%. A profile of effects was also observed, with: (i) a surficial coarser grainsize, higher Vp section; (ii) a fine grainsize low Vp, high porosity section to ~15cm; and, (iii) a grainsize between the former, and raised Vp.
Meadows & Tait (1989) found that bioturbation by annelid worms could either increase or decrease (from initial 2.5*10-2 to 1*10-2 and 6*10-2) the permeability of estuarine muddy sands, but that in their experiments the shear strength was increased slightly, which correlated with a decrease of water contents.

In situ measurement. In situ measurements do not suffer from the problems of sample disturbance, but have their own set of random and systematic errors.
In situ experiments are carried out over a range of scales. Small scale in situ acoustic experiments (<0.1m3) are subject to extreme fluctuations depending on the local environment such as presence of: burrows, shells or gas voids; concretions; organic debris; sedimentary layering or ripples; or isolated gravel clasts and shells. (Laboratory measurements tend to avoid these extremes by selection of specimens). Large scale experiments (>10m3) are subject to difficulties from layer stratigraphy, seabed roughness, buried bedrock and lateral sediment changes that are almost impossible to determine sufficiently to understand the experimental setup. As an example: in most ocean environments 15m of stratigraphy (1 wavelength of a 100Hz in situ experiment) represents >300,000 years of geological history and multiple sealevel and glacial climate cycles (e.g., Collins & others 1998; Carter & Johnson 1986; Kennett 1982).

Examples of these factors include:
(i) influence of fractures which are present at a range of scales (e.g., Carlson & Gangi 1985);
(ii) difference between intrinsic (material) and apparent (structural, environmental) acoustic attenuations; apparent attenuations at low frequencies (<100Hz) - and hence large scales (>15m) - probably account for the attenuations greater than expected from the materials alone in plots of attenuation data (e.g., Stoll 1989, Fig. 5.6; LeBlanc & others 1992); an analytical example of the effect of sedimentary layering on low frequency acoustic attenuation is given by Kerner & Harris (1993);
(iii) differences between laboratory, vertical seismic profiling and sonobuoy measurements of the variation of sonic velocity with depth in marine sediments;
(iv) results can be strongly influenced on site-specific conditions, especially the local geology including the stratigraphy and bedrock.
Many in situ experimental techniques assume values for unmeasured or poorly known input variables; the experimentally derived outputs may be highly sensitive to these inputs.

Temperature effects. As a sample is retrieved its temperature will alter from usually 0.5 to 10C to 20*C or higher. As shown comprehensively by Shumway (1958) temperature changes produce changes of acoustic velocity. His data on a range of natural sand-silt, silt and clay sediments in their natural seawater shows a change of ~70 m/s between 0-20*C.
If finer resolution of velocity relationships to other properties is sought then only data with documented laboratory / in situ temperature and pressure would be used. This is not a practical option; too few data points would be available for analysis.

Back to top

The confidence on a single measurement from a population with significant variance is not high. Imagine an application of the current study, where one sediment sample is taken from the seafloor and measured for an acoustic experiment. The probability that the sample provides an estimation of the acoustic conditions within 10% of Vp and * (Porosity) if the measurement errors are *Vp= +5m/s and *Po = +5% (2*) can be calculated. In certain seafloor types where ripples, dunes or mud-weed patches are present the variability implies an even worse level of confidence.

The test may be put another way: How many samples would be required at a site to obtain a mean result which can be compared at the 95% (2*) level against the acoustic prediction which has confidences *PredVp = +30 m/s and *Pred* = +15% (2*) ?
Assuming that only a small number of measurements is to be made and with accuracies *Vp= +5m/s and *Po = +5% (2*) the confidence interval *CI about the expected mean of a small sample is:

The conclusion is that in the face of known levels of error with velocities and porosities, multiple samplings/ measurements will have to be made at experimental sites in order to accurately test geoacoustic predictions.
The statistics discussed here do not account for systematic errors.

Back to top

Chris Jenkins (Email)
INSTAAR, University of Colorado
5-Feb-2002