Incorporating interaction terms in multivariate linear regression for post-event flood waste estimation

Multivariate linear regression methodology has been conceived as a viable technique in flood waste estimation. The fundamental assumption of the conventional flood waste model, independence between input variables, may not work in reality. As an alternative, we evaluated the effectiveness of including interaction terms in flood waste modeling. The secondary objectives include to suggest the strategy in flood waste mitigation and to explore a plausible explanation to the modeling results. In the scheme of model development and assessment, ninety flood cases in South Korea were statistically analyzed. Input variables for regression analysis were selected from available datasets in the national disaster information system and the selected variables were flood damage variables used to quantify the amount of flood waste. According to the results, incorporating the interaction terms improved the estimation accuracy of the model. The single-variable sensitivity analysis revealed that mitigating damage to rivers and croplands would most efficiently reduce potential flood waste generation. The interaction terms appeared to compensate for the over/underestimated waste amounts by single terms, and they explained the nonlinear response of waste generation. Observations made throughout the field survey revealed that the nonlinear and interactive pattern of flood waste generation corresponded to the results from the regression analysis. In a practical aspect, incorporating the interaction terms would be an effective method to enhance the flood waste estimation model without costly works for further variables exploration.