The traditional approaches of risk analysis are based on a coarse scale and limited influential factors. Non-significant qualitative factors in the models could result in loss of main information that in turn affects the prediction accuracy. Further, the traditional methods seldom consider uncertainty of the datasets in the models that lead probably to instability of their applications.

Given multiplicity of the factors involved in natural disaster system, Dr. LI Lianfa and Prof. WANG Jinfeng from Institute of Geographical Sciences and Natural Resources Research(IGSNRR), Chinese Academy of Sciences, proposed an approach for vulnerability assessment of human beings based on Bayesian network.

In the test of partial regions in Wen Chuan’s 2008 catastrophic earthquake, the researchers incorporated multiple aspects related to the vulnerability in their model, e.g. hazard intensity, natural environment, traffic, land-use and socioeconomic factors. The optimal model output the receiver operating characteristic curve (ROC) area of beyond 0.9 for every risk level. The predictive result reached an accuracy of 0.85 in comparison with practical survey of the disaster and the relevant loss.

The test demonstrated that the method had higher detection probability and identification accuracy. This approach will help to the identification of high risk regions and mitigation of natural disasters. Moreover, it is also used in the other disasters involving in risk analysis of public health such as air pollution.

The study has been published in Risk Analysis（Li LF, Wang JF, Leung H, Zhao SS. 2012. A Bayesian method to mine spatial datasets to evaluate the vulnerability of human beings to catastrophic risk. Risk Analysis. DOI: 10.1111/j.1539-6924.2012.01790.x）.