A data?driven method for predicting debris?flow runout zones by integrating multivariate adaptive regression splines and Akaike information criterion

摘要

Debris-flow runout zones are important parameters for delineation of endangered areas and design of mitigation works. It is necessary to properly determine the debris-flow runout for hazard assessment and control measures design. Many empirical-statistical correlations have related runout characteristics (e.g., runout distance) to debris-flow volume or topographical parameters. However, the predictability of empirical-statistical models is always questionable due to the great randomness of debris flows and assumption of inputs and functional relationships prior to analysis. This paper develops a data-driven method to predict the debris-flow runout by integrating multivariate adaptive regression splines (MARS) and Akaike information criterion (AIC) without assumption of input parameters and specific function relationships. The debris-flow volume and topographical parameters (e.g., catchment internal relief) are investigated as candidate inputs to train MARS models for estimating the runout characteristics of debris flows. Then, the most appropriate inputs and MARS models for debris-flow runout are identified by AIC. The proposed approaches are illustrated using channelized debris-flow data in Wenchuan earthquake zone. Results showed that the developed method can select the most appropriate MARS models for the debris-flow runout zones in Wenchuan earthquake-stricken area. The developed MARS models portray the regression relationships entirely "driven" from the training data and provide objective and explicit relationships of debris-flow runout zones. Compared with the previous empirical correlations in Wenchuan area, the proposed MARS models have higher prediction accuracy. Even for extreme debris-flow events, the MARS models also show satisfactory accuracy.