Coupling Bayesian networks and geospatial software for environmental risk assessment

Abstract

The Gulf of Finland, the easternmost part of the Baltic Sea, is a fairly narrow and shallow gulf that has witnessed a significant increase in maritime traffic, especially in oil transportation in the last years. While several measures have been taken to improve the safety of sea traffic in the area, a risk of a major oil spill accident exists. Although oil spills can be combated at sea at least to some extent, substantial amounts of spilled oil can reach the shoreline, especially in small and narrow sea areas like the Gulf of Finland. This may cause significant environmental damage. The risks that oil transportation poses to environmental values are spatially distributed because some locations within the transport network, are more prone for accidents, and because ecological values are not evenly distributed along the shores. In this paper we describe a method and tool that was developed to assess the risks that oil transportation poses to environmental values of the Finnish shore of the Gulf of Finland. In a recent research project (SAFGOF) a Bayesian network (BN) describing oil tanker accidents in the Gulf of Finland was developed. The network encodes a set of initial conditions that was considered to cover all realistic situations. A large number of oil spill simulations were carried out using the initial conditions. The environmental values for the assessment were available from a previous study, where information about shoreline habitats was collected and sensitivity indicators for oil spill combating were developed. The HUGIN software for BNs was used in SAFGOF and it was also chosen for this study. Geoinformatica was chosen as the geospatial software for the study. A Perl foreign function interface was developed for the HUGIN C library so that it could be easily coupled with Geoinformatica. The internal plug-in technology of Perl (Perl modules) along with some graphical widgets was used to couple the risk assessment tool to the Geoinformatica graphical user interface application. The risk assessment tool was implemented as a Perl module, which presents a dialog-box to the user and contains callback-functions for the interaction with the BN and for the computations. The dialog-box is also used for specifying which layer to use as the environmental values for the computations. Use of the plug-in requires some preliminary coupling work for linking a BN with the set of geospatial environmental scenarios as raster files. When the user opens a BN into the tool, the tool looks for the coupling file. Additionally, the tool looks for an image file of the network. The coupling is defined in a text file, which contains a file name template for the scenario files and tuples, which link a specific state of a specific variable in the BN to a specific match in the file name template. The file name of each scenario must therefore contain codes that reveal the state that each scenario variable had when the scenario was computed. The plug-in contains several checks for the correctness of the coupling of the scenario rasters with the BN. The risk assessment tool supports an interactive workflow, where a value layer is created from, e.g., a vector layer created by valuation tool, a scenario is defined by a BN, and an overlay computation is initiated to produce a risk map. Generic visualization tools can then be used to prepare the map.