Graphical methods for the exploration of ecological databases
Ben Raymond, Lee Belbin, Jonathan Stark


Scientific data repositories are becoming increasingly characterised by large data sets stored in web-accessible, structured databases. Exploratory analysis is an important part of the process of converting such data into scientific knowledge.

We describe an exploratory technique that uses structured graphs to represent data. A graph consists of a set of nodes connected by edges, where the nodes represent entities of interest. Each node has a set of attribute data, and a graph is constructed by forming edges between those nodes that have common attributes. The resultant graph provides insight into the structure of the data and the relationships between the entities of interest. The nodes and their attributes can be defined in terms of any of the variables in the database. In a classical ecological data set, the nodes might be sampling sites and the attributes might be species.

We have developed a web browser-based application to assist with this exploratory procedure. The user can specify the node and attribute definitions using data from a local database, a web-enabled database using web services, or from a local file. A dynamic interactive viewer enables graphs to be explored. Graphs can be analysed using graph-theoretic tools, including analyses related to connectivity, simplification, node clustering, and subgraph analyses. The ability to compare a graph to one obtained using the same entities but different attribute data is particularly useful in explaining the original graph structure.

The method is applicable where entities can be linked by potentially shared properties, and is well suited to many types of ecological data. Graph-based techniques are commonly used in some types of ecological analyses, including trophic webs and community ecology. We demonstrate the technique using benthic data from marine sediments near Australia's Casey station in Antarctica. Samples were collected from Brown Bay, which is adjacent to a disused tip site and is known to have high levels of many contaminants. Control samples were collected from nearby, uncontaminated O'Brien Bay. The resultant graph structure indicates a difference in the benthic communities of the two bays that is related to sediment metal concentrations.