Keywords: environmental impact, Bayesian networks, GIS, machine learning, mining, slate.

An environmental impact assessment (EIA) procedure is very useful for protecting the environment and natural resources. Current EIA methods, however, are insufficient to support systems and decisions related to real impact conditions; with the exception of parameters for specific environmental factors with threshold values that are accepted internationally and for which evaluation can be made on the basis of measureable units, current methods are simple, subjective and imprecise. Popular methods include the Leopold matrix, and the Batelle-Columbus method, which, respectively, implement a qualitative and a quantitative evaluation. These methods fail, however, to take into a account, issues such as ecosystem spatial heterogeneity and the intrinsic imprecision and ambiguity associated with these systems and with the judgments of experts and stakeholders.

The aim of our study is, using Bayesian networks in combination with a GIS, to improve on existing methods for the assessment of environmental impact of a slate mining operation, by taking into account ecosystem spatial heterogeneity, uncertainty in the definition of variables, and the interrelationships between variables.

Bayesian networks (BN, also Bayesian belief networks, probabilistic networks, causal networks [1]) are graphical models of the causal relationships between variables which also specify their joint probability distribution.

Our Bayesian network was constructed from expert information based on field data for environmental elements affected by the mine and reflecting the susceptibility of these environmental elements to primary, secondary or synergic impacts. For the construction of the BN, we used the GeNIe system facilitated by the Decision Systems Laboratory of the University of Pittsburgh [2].

Once created the Bayesian network enabled us to do the following: 1) to draw inferences in relation to the environmental impact for points in a grid covering the area studied, for incorporation in a geographical information system and creating an environmental impact map; 2) to better understand the structure of relationships between mining tasks and different environmental variables, as also the true repercussions of these mining tasks on the environment; 3) to adequately reflect the heterogeneity of the ecosystem in terms of its diversity and the complexity of its response to mining activities; and 4) to draw on a knowledge system that is enriched each time new data is added.

As far as we are aware, no references exist in relation to the application of this type of tool to the evaluation of the impact of mining activities, despite the fact that these activities have substantial repercussions on the environment. The quantity of affected environmental factors, the synergy between them, and above all, the complexity of impacts in terms of magnitude, certainty, duration and type, means that traditional methods used for the evaluation of the risk associated with mining do not reflect the real consequences of the activity.

BN results combined with the analysis elements in a geographical information system (GIS) provide a graphic representation of the work area classified as zones that are homogeneous in terms of environmental impact.

Vulnerable environmental features were grouped into seven categories: flora, fauna, ecosystems, soils, atmosphere, water, and landscape. In the specific case of the slate mine studied for the purpose of this research, the network indicated that the most significant actions were those that affect soil and water. If the impact of any environmental variable included in prior expert knowledge were changed to critical in subsequent layers (third and fourth layers, for example), the nodes most affected would be hydric erosion and soil degradation, water flow and morphology, and to a lesser extent, the landscape node.

Thus, the actions with most impact were associated with earth movements, the creation of openings, the accumulation of mining waste, and the pumping of mine drainage water directly into water courses. These actions implied a modification in surface runoff dynamics (fundamentally in terms of drag and downstream sedimentation) and also affected the canopy and consequently landscape.

Future activities planned for this project include applying the network to other mines (and improving its functioning), configuring the network for prediction purposes (classification), and incorporating decision nodes (influence diagrams) that are representative of different mine operation activities.

1

S.L. Lauritzen, "Some modern applications of graphical models", in Highly Structured Stochastic Systems. eds. P.J. Green, N.L. Hjort, S. Richardson, S., Oxford University Press, 13-32, 2003.

2

Decision Systems Laboratory, University of Pittsburg, http://genie.sis.pitt.edu/about.html, accessed on: March, 2006.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £105 +P&P)