Selection of representative indicators for flood risk assessment using marginal entropy and mutual information

Floods are the most frequent types of natural disasters. From the perspective of disaster management, indicators associated with floods are important for accurate flood risk assessment. However, the application of all indicators related to flood risk assessment decreases the evaluation efficiency, because the definitions of the indicators may overlap. Moreover, the volume of data required for collection and evaluation is significantly large, making the evaluation practically impossible. Thus, a scientific and objective method to select indicators for flood risk assessment based on the entropy theory was developed herein. First, the existing 28 assessment indicators were analyzed and probability-based data were constructed for each indicator considering 28 districts in a midwestern region of Korea. The information quantity for each indicator was then obtained using marginal entropy and mutual information generated in the entropy theory. Next, the total information quantity based on the numbers of combination of indicators was derived by considering the information quantity for each indicator and the overlapping mutual information between the indicators. The maximum amount of information (161.55) was obtained by combining 18 out of the 28 flood risk indicators. The selected 18 indicators reflected regional characteristics better than those used in the existing method, demonstrating that the flood risk of the target area could be adequately assessed.