Territory Stories

Sediment Quality Sampling Design for Darwin Harbour



Sediment Quality Sampling Design for Darwin Harbour


Brinkman, Richard; Logan, Murray; Northern Territory. Department of Environment and Natural Resources; Australian Institute of Marine Science


E-Publications; E-Books; PublicationNT; Report Number 44/2019




Darwin Harbour


In the context of increasing development and associated pressures, this project aims to inform the development of a first systematic, long-term, sediment monitoring program for Darwin Harbour which takes into consideration the physicochemical nature of Darwin Harbour sediment and the oceanographic processes which will influence the movement of contaminated sediment in the Harbour. The rationale for the program is that seabed and estuarine sediments are both an extensive habitat and the ultimate repository for many contaminants that enter waterways. In addition, monitoring of contaminants in sediment may facilitate the identification of increasing contaminant loads in the Harbour which may not be detected by water monitoring programs due to the high flushing rate within Darwin Harbour and infrequent water sample collection.


Made available by via Publications (Legal Deposit) Act 2004 (NT).

Table of contents

1. Executive summary -- 2. Introduction -- 3. Methods -- 3.1 Overview of methodology -- 3.2 Tidal Hydrodynamics -- 3.3 Wave Dynamics -- 3.4 Sediment Modelling -- 3.5 Sediment sampling design analysis -- 4. Results -- 4.1 Tidal and wave driven hydrodynamic processes -- 4.2 Sediment transport modelling -- 4.3 Sediment characteristics from previous sampling programs -- 4.4 Sediment sampling design analysis -- 4.4.1 Existing chemical sediment data, Outer Harbour sediment monitoring data, and designated sampling sites -- 4.4.2 Hydrodynamic modelling layers -- 4.4.3 Exclusion zone masks -- 4.5 Spatial Model fitting -- 4.5.1 Background on statistical techniques for designing sediment sampling program -- 4.5.2 Results from statistically derived sampling design - East Arm -- 4.5.3 Results from statistically derived sampling design - Outer Harbour -- 4.5.4 Representation of sampling sites mapped with hydrodynamic and sediment modelling parameters -- 4.6 Harbour Sediment Zonation, and conceptual representation: -- 5 Conclusions -- 6 References -- 7 Appendix 1




Sediment quality; Tidal hydrodynamics; sediment sampling; design analysis

Publisher name

Northern Territory Government

Place of publication



Report Number 44/2019


43, 74 pages : colour maps ; 30 cm

File type





Attribution International 4.0 (CC BY 4.0)

Copyright owner

Northern Territory Government



Parent handle


Citation address


Page content

-22 3. Spatial Model fitting A target or set of targets is required against which the effectiveness and accuracy of candidate sampling designs can be tuned or gauged. This target should represent the full underlying conditions and in essence represent a saturated sampling design - a sampling design in which all possible locations/sites are sampled. Whilst this is logistically not possible, given an adequate set of baseline data, statistical spatial models can be generated to estimate the underlying patterns. The resulting predicted layers can be used to represent the targets. Spatial models are complex statistical models that attempt to recreate the full feature space from which a limited set of samples were collected. In so doing, they attempt to incorporate two-dimensional patterns and correlations to allow prediction to areas in between samples. In the simplest cases, simple surfaces can be derived by linear interpolation between all the sampling points. However, when samples are distributed unevenly, there are strong spatial dependencies and/or the bounding domain is not a simple rectangle, more complex methodologies are required. Ecological and environmental processes are often correlated through space. To account for these spatial dependencies within a spatial model it is useful to incorporate a Gaussian Random Field (GRF) which specifies a spatially dependent covariance structure in which locations that are closer to one another in space will in general be more highly correlated to one another than locations that are further apart. Large or complex spatial models soon become intractable using a traditional frequentist modelling framework. By contrast, equivalent Bayesian models are typically very computationally expensive. Integrated Nested Laplace Approximation (INLA: Rue, Martino, and Chopin 2009) is a Bayesian approximation framework that offers the philosophical advantages of a full Bayesian approach, yet with the computational efficiency of frequentist approaches. We can consider a GRF to be stationary if the degree of correlation between two points is dependent only on the distance between the points, or non-stationary if we allow the correlation function to vary over the landscape. An extreme form of non-stationary model occurs when there are physical barriers that disrupt or block the flow of contagious processes. In such cases, just because two locations are in close proximity, does not necessarily mean that they will be highly correlated. Consider a simple example of the diffusion of a dye in water throughout a tank. The dye will spread out from the source and gradually disperse throughout the tank. Consequently, the correlation between the concentration of dye at any two locations during dispersion will be dependent on the distance between the two locations. If however, the tank had a partial barrier that restricted the flow of dye molecules, then two locations either side of the barrier might have very different dye concentrations despite being in close proximity. Barrier models are able to account for these obstructions. yi (i ,) log(i) = 0 + u(si) + i 0 N(0, 1000) u(si) GRF (r ,u) i N(0,2) (e) eeee (u) 0e0u ( 1 r ) 1e1 1 r where yi is the ith observation of the target chemical variable, i is the independent and individually variable random effect used to model the very short range erratic dependencies and u(si) is the Gaussian Random Field and is used to model the long-range structural (autocorrelation) dependencies. A diffuse prior is applied to the intercept (0) and i a vector of independent Gaussians. The Matern family spatial random

Aboriginal and Torres Strait Islander people are advised that this website may contain the names, voices and images of people who have died, as well as other culturally sensitive content. Please be aware that some collection items may use outdated phrases or words which reflect the attitude of the creator at the time, and are now considered offensive.

We use temporary cookies on this site to provide functionality.
By continuing to use this site without changing your settings, you consent to our use of cookies.